Births and their outcomes by time, day and year: a retrospective birth cohort data linkage study

Confidential enquiries

Quality health care has been defined as care that is safe and effective and results in a positive experience. 12

There has been a longstanding concern about the safety of maternity care, in terms of maternal mortality and of stillbirths and neonatal mortality. The UK has the longest-running programmes of confidential enquiries into pregnancy-related deaths in the world. After earlier developments in Scotland, Confidential Enquiries into Maternal Deaths have been run in England and Wales since 1928. 13 Following confidential enquiries into perinatal deaths at regional and local levels, the Confidential Enquiry into Stillbirth and Deaths in Infancy was set up nationally in 1993. 14,15 Learning from the enquiries regarding ways to reduce preventable factors associated with deaths has led to considerable improvements in service quality. Following a retendering exercise, the MBRRACE-UK (Mothers and Babies: Reducing Risk through Audits and Confidential Enquiries across the UK) consortium is now responsible for these enquiries and has a key role in monitoring current policies. 15

Choice of settings for birth

At the beginning of the 20th century, nearly all women gave birth at home; by the end of the century, hospital birth was almost universal. Many factors contributed to this major change, but from 1970 onwards there was an explicit policy that all births in England should take place in hospital. In response to this, the proportion of births occurring at a mother’s home in England and Wales fell from 13% in 1970 to < 1% of all births in 1987, before starting to rise again. The rise was already occurring when the House of Commons Health Committee reviewed available evidence in its Inquiry into Maternity Services. 16 The Inquiry’s conclusions were wide-ranging, but it is perhaps best remembered for its conclusion that:

The policy of encouraging all women to give birth in hospital cannot be justified on grounds of safety.

Reproduced with permission from the House of Commons. 16 Contains public sector information licensed under the Open Government Licence v3.0

Although limited, the evidence, coupled with demand from women and support from maternity charities and the Royal College of Midwives (RCM), led to changes in government policy in both England and Wales during the 1990s. 16,17 The Welsh Office issued the Protocol for Investment in Health Gain, which emphasised the need for women to have more control and an informed choice about the care given in pregnancy and childbirth. 18 In England, the Department of Health and Social Care responded to the report by setting up the Expert Maternity Group. Its report, Changing Childbirth,19 published in 1993, endorsed many of the policy changes set out in the Health Committee’s report, moving towards woman-centred care and a greater role for midwives, with at least 30% of women having a midwife as the lead professional.

These policies continued after the change of government in 1997. They were set out in detail in 2007 in Maternity Matters,20 after the aims had been outlined earlier in the National Service Framework for Children, Young People and Maternity Services. 21

The policy called explicitly for ‘an evidence-based culture’, which learned from adverse events and highlighted the value of routine data collection and analysis. 20

There was a strong emphasis on choice and this included a ‘choice guarantee’ on place of birth and other aspects of maternity care. 20 The Maternity Matters document stated that a ‘guiding principle’ for the modern maternity services was that:

All women will need a midwife and some need doctors too.

© Crown Copyright. Reproduced with permission from the Department of Health and Social Care. 20 Contains public sector information licensed under the Open Government Licence v3.0

Although the proportion of births taking place at the mother’s home rose to well over 2% in the decade after the Changing Childbirth report was published, this was followed by a small decrease from 2008 to 2012. The number of alongside midwifery units (AMUs) rose from 26 in 2007 to over 100 in 2016, whereas the number of freestanding midwifery units (FMUs) rose slightly from 56 to 62 over the same time period (Miranda Dodwell, BirthChoiceUK, 2016, personal communication). In 2012, an estimated 9% of women gave birth in an AMU and 2% of women gave birth in a FMU. 22 This represented some change but fell well short of the 30% proposed in Changing Childbirth. 19

In 2002, the Welsh Assembly Government set a target of 10% for home births by 2007 as part of its strategy to develop midwifery-led care and access to services that were unfamiliar to many communities. 23 The National Service Framework for Children, Young People and Maternity Services in Wales24 had similar ambitions to those for England. This was followed by Strategic Vision for Maternity Services in Wales, the cornerstone of which continued to be tackling health inequalities. 25 Choice of place of birth and access to midwifery care pathways were part of the plan.

In response to the limited nature of the previous evidence and a recommendation by the National Institute for Health and Care Excellence (NICE) for further research, the Birthplace Programme of Research, a large-scale prospective cohort study, was funded. It aimed to evaluate and compare the risks and benefits for healthy women with straightforward pregnancies, who are at ‘low risk’ of complications, of giving birth at home or in FMUs, AMUs or obstetric units (OUs), focusing in particular on birth outcomes. 26 The Birthplace research found that outcomes for babies were similar for women planning birth in all settings but compared with women planning a hospital birth, women planning a home birth or birth in a midwifery unit had a lower risk of having a caesarean section, an assisted delivery or a haemorrhage.

The research also found that women planning a home birth or midwifery unit birth were more likely than women planning for birth in an OU to have a ‘normal’ birth. 26

This research was used as a basis for the 2014 revision to the NICE Intrapartum Care Guideline. 27 Birthplace26 also confirmed that the policy of choice of place of birth was safe, and that planning care in midwife-led care settings at home and in FMUs or AMUs for women at low risk of developing complications during labour and birth resulted in fewer invasive and costly medical procedures. 27

Health inequalities

One of the themes in the House of Commons Health Committee’s report that received less attention was health inequalities and maternity care for disadvantaged women. This reflects the extent to which the degree of emphasis placed on preventing health inequalities, investing in public health and improving access to maternity care has varied over time and between governments.

From 1997 to 2010, during the 13 years of Labour government, there was a clear emphasis on inequalities in health outcomes and the need to know where specific health ‘gains’ could be achieved. The National Service Framework for Children, Young People and Maternity Services, published in 2004,21 followed frameworks for other key areas of health care, including cancer, heart disease and diabetes. It looked at perinatal mortality and morbidity, and other aspects of child health, in relation to structural inequalities, as well as at the impact of poverty and disadvantage succeeding from one generation to the next. 28 There was an emphasis on cross-departmental policies to reduce the number of children living in poverty. 29 A key strategy for improving maternity experiences included improving maternity rights and pay.

The strategic review of health inequalities in England, chaired by Michael Marmot, emphasised the social determinants of health and health inequality. 30 The review team concluded that, in addition to the social justice argument for addressing health inequalities, there was also a pressing economic imperative for change as annual costs to the exchequer attributable to inequality in illness were estimated to be £31–33 billion in productivity losses, £20–32 billion in lost taxes and higher welfare payments and over £5.5 billion in additional NHS costs. 30 It was therefore also necessary to engage members of the public in protecting and promoting their own health.

Reducing the costs of litigation

Broader sociopolitical changes were also influential, including the regulation of hours by the European Working Time Directive,31 reductions in the numbers of junior doctors and the increasing cost of maternity litigation. A report by the NHS Litigation Authority showed that between 2000 and 2009, 5087 claims for injury were made against the NHS, for which it paid out £3.1B in damages. 32 The three most frequent categories of claim were those relating to the management of labour, caesarean section and cerebral palsy. 32 Improving the overall quality of maternity care and reducing deaths and injury are important, as is the need to make the best use of limited resources.

Recent policy developments

After the change in government in 2010, it was several years before new maternity policies were developed. The NHS Five Year Forward View,33 published in October 2014, set out a vision for how the NHS in England would have to adapt to meet the needs of a changing population and committed further funding for the NHS. It stated that a ‘one size fits all’ approach to health care was no longer applicable to a diverse population and proposed new models of care, including in relation to maternity care. Speaking of the need to move away from a factory-like model of ‘care and repair’, the NHS Five Year Forward View set out a greater focus on community engagement and prevention of disease through improved public health, although its view of this was based more on individual than on structural measures.

It also spelled out the considerable growing demands on the service as a result of people living longer, opportunities and pressures to provide ever more complex science- and technology-led interventions and the continuing impact of smoking, alcohol use, lack of exercise and rising levels of obesity on levels of diabetes, stroke, heart disease and cancer. A national maternity review was subsequently commissioned to investigate the changes to maternity services needed to meet the needs of the population. 33

It was also asked to ensure that the recommendations of the Morecambe Bay investigation34 of ‘serious failures of clinical care’, which led to avoidable deaths of mothers and babies in one maternity unit, could be embedded throughout the NHS in England. 35

The review’s findings were incorporated in a Five Year Forward View for maternity, Better Births,35 a strategic framework with many broad-ranging aims. It aimed to make a significant culture change, based on two fundamental principles, described as being:

The importance of women being able to make choices about their care, and the safety of the mother and baby being paramount.

Reproduced with permission from NHS England. 35 Contains public sector information licensed under the Open Government Licence v3.0

This safer, more personalised care, in which maternity services are responsive to the needs and circumstances of individual mothers and babies, would be focused on reducing avoidable deaths and damage through a focus on acute services delivered by multidisciplinary teams. In addition to the distress caused to families by negligence during maternity care, compensation costs the NHS an average of £560M per year. 35 This care would also encompass more community-based and midwife-led services for women with a straightforward pregnancy who are at low risk of complications. Better Births35 also preserved the commitment to the choice of place of birth given in Maternity Matters. 20

Meanwhile, the Welsh Government set out its strategic vision for maternity services in 201125 and set up the Maternity Network Wales in 2015. 36 This has three main priority areas: (1) to lead on a number of interventions to reduce stillbirth rates in Wales, (2) to improve the quality of maternity services by implementing the recommendations made by the Quality and Safety subgroup of the All Wales Maternity Service Implementation Group in its report published in June 201337 and (3) to work in partnership with women and families. 38 This is to be achieved through a number of subgroups set up by the Network.

NHS care at weekends

From 2001 onwards, a considerable number of analyses have shown higher mortality rates among patients admitted to hospital at weekends than on weekdays and there has been considerable debate about how to interpret these rates. 39,40

We have not attempted to summarise them here, as we are aware that a systematic review is under way. 41 Instead, we have given examples that highlight some of the arguments. For example, a study using routinely collected data for over 4 million emergency admissions found that the overall adjusted odds of death was 10% higher for weekend admissions than for weekday admissions. 42 Another study using patient records for 14 million hospital admissions found a higher risk of in-hospital death within 30 days of admission for weekend admissions than for weekday admissions, but the chance of actually dying at the weekend was lower than the chance of dying mid-week. 43 In contrast, a prospective cohort study with 74,000 stroke patients showed no difference between 30-day survival at weekends and on weekdays. 44

Even if there are weekend effects, it is unclear if they reflect differences in quality of care or differences in case mix. For example, one study found that patients admitted at the weekend were more seriously ill than those admitted during the week. 45 Another suggested that the weekend effect arises from patient-level differences at admission rather than reduced hospital staffing or services. 46 Even where it is suspected that staffing makes a difference, there are questions about whether this reflects numbers or type of staff or their seniority. One study examining nearly 4 million elective procedures concluded that consultant seniority has no impact on predicting 30-day mortality. 47

These uncertainties have not impeded calls from politicians and others for a 7-day health service. NHS England established a Seven Days a Week Forum and its report, published in November 2013,48 was cited prominently in the report of the Review Body on Doctors’ and Dentists’ Remuneration published in 2015. 49

Despite the conflicting evidence, in the NHS Mandate published in 2015, the Secretary of State for Health for England, called for ‘a truly seven-day health service’:

The NHS should be there for people when they need it; this means providing equally good care seven days of the week, not just Monday to Friday.

Reproduced with permission from the Department of Health and Social Care. 35 Contains public sector information licensed under the Open Government Licence v3.0

Although these policies have been labelled ‘24/7’, the analyses have been largely restricted to differences between days. As we show in the following section, this is not the case for analyses of births, and times of day have been taken into account when discussing the staffing of maternity services. Although it is self-evident that maternity services need to operate 24 hours a day, 7 days a week, there are still many questions about the type and seniority of staff needed.

Staffing of maternity hospitals

Although adequate staffing is central to delivering high-quality, efficient health care, most of the debates have been about medical staffing and have mainly related to consultants, although they featured in negotiations about revisions to junior doctors’ contracts49 and their decision to take industrial action in response. Although the RCM repeatedly expresses concern about shortfalls in midwifery staffing, it has not been in this context. This is a matter of concern to the College and it has supported our project at each stage, stating that the information we aimed to produce would have been useful to it.

The National Patient Safety Agency ‘Hospital at Night’ Programme,50,51 set up in 2005, proposed that the way to achieve safe clinical care was to have out-of-hours care delivered by a multiprofessional team with the full range of skills and competencies across a whole range of disciplines in a hospital to meet the immediate needs of patients. This did not apply to obstetrics as, according to the Royal College of Obstetricians and Gynaecologists (RCOG),51 women in labour needing medical care would be managed by obstetricians within the department and not by the Hospital at Night team. Thus, the emergency obstetric rota would be staffed for 168 hours a week (‘24/7’) from within the discipline of obstetrics and gynaecology.

For a number of years, the RCOG advocated increasing the hours of consultant presence in maternity units. 52,53 Initial recommendations were for units to have a minimum of 40 hours of consultant obstetrician cover a week, with exceptions for units with fewer than 1000 births a year and those with low complication rates. In addition, units with > 5000 births were to have 60 hours of consultant cover a week. The Healthcare Commission’s maternity services report, Towards Better Births, reported that only 68% of trusts reported meeting this standard in 2007, with some trusts reporting a very minimal presence, as low as 10 hours per week. 54

In 2007, new recommendations were made for a minimum of 60 hours of consultant presence in all OUs, with a specific recommendation of 98 hours for units with 4000–5000 births. For units with over 5000 births per annum, the RCOG recommended consultant presence on the labour ward both day and night, for 168 hours per week, referred to as ‘24/7’ cover, by 2010. For the largest units with over 6000 births, the suggested time frame for implementation was earlier, by 2008. 55

This drive for increased consultant presence came at a time when there were also increasing pressures on the workforce. The RCOG report High Quality Women’s Health Care56 identified the impact of the European Working Time Directive on women’s health care, and specifically the introduction of the 48-hour week, which led to a need to increase the numbers of consultants to provide cover. The RCOG report, The Future Workforce in Obstetrics and Gynaecology,57 calculated that 168 hours’ cover for a delivery suite, along with other clinical activity, would require the employment of 18–21 consultants per OU. This led to proposals for obstetric services to be centralised, with the closure of small OUs, as these might be uneconomic to run with the recommended level of consultant presence.

In terms of obstetric staffing, the Strategic Vision for Maternity Services in Wales policy referred to compliance with the European Working Time Directive, the need for services to comply with the RCOG guidance for 40 hours of consultant labour ward presence per week, depending on the number of deliveries per year, and for a middle-grade doctor at level Specialty Trainee 3 or higher or at Staff Grade, Associate Specialist and Specialty Doctor grades to cover the labour ward at all times at which a consultant is not present. 25

By 2014, only two units in the UK had achieved 168 hours per week of consultant presence58 even though numbers of consultant obstetricians and gynaecologists in England had risen substantially. Over the years between 2002 and 2017, numbers of full-time equivalent staff had almost doubled, with an increase of 20% in the last 5 years of this period. 59,60

Despite the decisive strategic direction taken by the RCOG, the evidence that these levels of consultant presence would increase safety for mothers and babies was not clear, particularly in relation to the increase in size of OUs resulting from this policy. 61,62 A systematic review and related case studies from England, France and Sweden concluded that ‘A model considered equitable by consultants and which includes prospective cover for holidays, thus truly providing continuous consultant labour ward presence, requires 26 consultants’. 63 This would be possible only within current budgets for England only in very large, tertiary level units providing care for a disproportionately large number of women with complex care, and so attracting the higher tariffs available for paying for their care.

Based on several factors, including limited available evidence on the impact of 24-hour consultant presence, with little to show that this improves the quality of care, together with competing arguments for deploying any additional staffing, such as employment of more midwives or having greater consultant obstetrician presence at weekends, the authors concluded that the case for 24-hour consultant obstetrician presence on the labour ward had so far not been demonstrated. 63

Given the lack of evidence that 24-hour resident consultant presence on the labour ward has any impact on women’s outcomes, the RCOG subsequently produced a report to update its guidance on obstetric staffing. 64 It stated that:

It is recognised that there is huge variability of service provision around the country in terms of workload complexity, geography and current middle grade staffing. For this reason, there is no single staffing model which is suitable for all UK units . . . Within obstetrics it is no longer possible to make recommendations about hours of consultant presence on the labour ward based on number of deliveries because of the diversity of consultant contracts and working practices.

Royal College of Obstetricians and Gynaecologists. Providing Quality Care for Women: Obstetrics and Gynaecology Workforce. London: RCOG; 2016. 64 Royal College of Obstetricians and Gynaecologists © 2018

Royal College of Obstetricians and Gynaecologists. Providing Quality Care for Women: Obstetrics and Gynaecology Workforce. London: RCOG; 2016

It did, however, strongly recommend that all OUs should have a minimum labour ward consultant presence during working hours between Monday and Friday, with the aim of extending this to every day of the week. 64

Reducing stillbirths and neonatal deaths

As well as the focus on keeping care for healthy women straightforward and on avoiding iatrogenic effects and rising costs, there has been a concurrent focus on a traditional cornerstone of quality maternity care, that is, improving safety by reducing avoidable maternity and perinatal injuries and deaths. This comes at a time of rising birth rates, rising ages at childbirth and increasing rates of obesity and other clinical and social complexity. These factors have led to greater demands on the service, and an increased proportion of women needing medical care and more intensive midwifery support. The Morecambe Bay Inquiry34 highlighted the importance of careful needs assessment and close interdisciplinary working to ensure consistent and improving levels of safety in maternity care. These are all drivers affecting commissioning priorities and strategic decision-making in service delivery.

As part of the Five Year Forward View, the government’s ambition is to halve the rates of stillbirth and neonatal death in England by 2030;65 the Saving Babies’ Lives Care Bundle is designed to guide commissioners and clinicians on how to develop services to make this happen. 66 It brings together four elements of care: (1) reducing smoking in pregnancy, (2) risk assessment and surveillance for fetal growth restriction, (3) raising awareness of reduced fetal movement and (4) effective fetal monitoring during labour.

In 1928, the stillbirth rate for England and Wales was 40.1 per 1000 total births and this fell steadily from the 1950s to the turn of the century, but then levelled off and thus became a cause for concern. In 2012, the stillbirth rate started to fall again, reaching 4.5 per 1000 total births in 2015. Even before the implementation of the government’s strategy, over the period 2013–15, the stillbirth rate in the UK as a whole, published by MBRRACE-UK, fell by almost 8%, largely because of a reduction in antepartum stillbirths at gestational ages of 32 weeks and over. 67

The neonatal mortality rate for England and Wales (i.e. deaths in the first month of life) fell more steadily from 30.9 per 1000 live births in 1928 to 2.7 per 1000 live births in 2013 and 2.5 per 1000 live births in 2014, although the decline had become less marked and the rate increased slightly to 2.6 in 2015. Data for the UK as a whole, published by MBRRACE-UK, showed that neonatal death rates fell marginally from 1.84 per 1000 live births in 2013 to 1.74 per 1000 live births in 2015. 67 In 2015, nearly two-thirds of these neonatal deaths were of babies who were born preterm and two-thirds of the stillbirths occurred before the pregnancy reached full term. 67 Rates published by MBRRACE-UK differ from those from civil registration as MBRRACE-UK systematically excludes events following the termination of pregnancy at ≥ 24 weeks of gestation and the deaths of babies born before 24 weeks of gestation. 67

Preterm birth and congenital anomalies account for a significant proportion of neonatal deaths and stillbirths, but the causes of preterm birth are poorly understood. Some countries, notably the Nordic countries, whose stillbirth and neonatal mortality rates our politicians aspire to emulate, have much lower rates of preterm birth and also lower rates of obstetric intervention than the countries of the UK. 68 A better understanding of the causes of preterm birth and the development of interventions to prevent them are urgently needed. This requires research programmes to evaluate the efficacy of both preventative and ‘rescue’ interventions.

The MBRRACE-UK consortium has led a coalition of stakeholders in developing a national, standardised Perinatal Mortality Review Tool to support local review processes. This is expected to improve assessments of the quality of care and postmortem investigations. The quality and incompleteness of data are also issues, preventing a full picture from being known, and also preventing accurate monitoring of changes.

In 2014, the RCOG launched a long-term quality improvement programme, Each Baby Counts, with the aim of halving the number of babies who die or are left severely disabled as a result of preventable incidents occurring during term labour, by 2020. Each Baby Counts is a measure that brings together the number of babies who are stillborn at term, who die within the first week of life or who are diagnosed with severe brain injuries. 69

In Wales, the National Assembly’s Health and Social Care Committee held an inquiry into stillbirths in 2012, which was reported in February 2013. It made a number of recommendations to reduce stillbirth rates,70 including the setting up of the Maternity Network Wales to take this work forward. 38

Variations in the patterns of the time when babies are born may affect stillbirth and neonatal mortality rates and can have implications for the staffing needs of maternity services. This can affect the care that women receive and their experience of that care. Investigations into patterns of birth timing and subsequent outcomes may inform changes in service provision that can improve care to make it safer and more personalised.

In particular, the strategy to reduce stillbirths may increase rates of induction of labour, which would have implications for the proportion of women eligible for midwifery-led care and for the pattern and timing of births in different settings. 66,69

Time of day of birth

Up to the middle of the 20th century, there was a popular assumption that times of birth and death were by and large predetermined, although it was recognised that unexpected, possibly untoward, events could change them. The second half of the century saw increasing use of medical intervention to influence the time of birth and to prolong the lives of babies born very preterm or with conditions previously thought to be incompatible with life.

It was established in the 19th century that numbers of spontaneous births varied by time of day and were higher at night than during the day. 71–73 For example, in his Essai de physique sociale,74 published in 1835, Adolf Quetelet cited data from Brussels and Hamburg, which showed that the numbers of live births in the 6-hour periods before and after midnight were higher than those in the corresponding periods before and after noon. An exception was found in the Hamburg data, which were subdivided by season and which showed that numbers of live births were high on winter mornings.

In the 20th century, data from a special birth certificate in use in Buffalo City, New York, showed that numbers of spontaneous births in 1937 were highest in the early hours of the morning from 03.00 to 05.00, whereas the much smaller numbers of births involving obstetric intervention peaked from 09.00 to 11.00. 75 A study of over 16,000 births in Birmingham, England, in the early 1950s focused on the time of onset of spontaneous labour and found that it peaked in the middle of the night, around 02.00, especially for primiparous women, with numbers of births peaking from 03.00 to 05.00. 76 Intervention rates were low in this period but caesarean sections were more likely to occur during the day.

Reviews published in the 1960s and 1970s found that, in previous studies, births were most likely to occur between 01.00 and 06.00. 71,73 Further studies in the 1960s pointed out that this peak was apparent only in predominantly normal births, which did not involve invasive clinical procedures. Some of the rationale for the rising rates of induction, acceleration and caesarean section from the 1970s onwards was an intention to concentrate births into conventional working hours in order to optimise access to clinical care. 77–79

The impact of these rising rates was to move the peak in overall numbers of births into daytime hours. For example, studies in Switzerland of births from 1979 to 1981 showed that the highest numbers of births occurred on weekdays between 08.00 and 16.00,80 but this analysis did not subdivide births by mode of onset and birth. An analysis of all births in the state of Massachusetts, USA from 1989 to 1995 showed a peak between 07.00 and 09.00, the times associated with planned caesarean section. 81 When the analysis was restricted to singleton vaginal births of spontaneous onset without stimulation, the variations were less marked but numbers of births were slightly higher in the daytime between 09.00 and 15.00, with a peak from 11.00 to 13.00.

Other US states now record times of birth79 and an analysis of births in 2013 in 41 states plus the District of Columbia showed that overall numbers of hospital births were concentrated into daytime and early evening hours. 82 Elective caesarean sections were concentrated into mornings with two peaks at 08.00 and 12.00, whereas numbers of emergency caesarean sections were highest during late afternoons and early evenings. Numbers of both induced and non-induced vaginal births were highest between 12.00 and 18.00, although the numbers of non-induced births varied considerably less.

In the USA, only out-of-hospital births showed a pattern similar to that in earlier analyses, with numbers of births being highest between 03.00 and 05.00. 82 In contrast, a study in a hospital in Spain with a high caesarean section rate showed that overall numbers peaked during morning hours, but unlike in the USA, numbers of spontaneous births were still highest in the early hours of the morning. 83 Similarly to this, an earlier study in Norway found that numbers of spontaneous births peaked in the early hours of the morning, especially for births to multiparous women, whereas numbers of births to primiparous women peaked later in the day. 84

A similar effect was seen in the Netherlands in a study that compared vertex singleton term births without oxytocic drugs to women having care from midwives with those births to women having care from obstetricians. 85 Among women cared for by midwives, numbers of births to primiparous women peaked between 08.00 and 09.00 and births to multiparous women peaked at 05.00. For women cared for by obstetricians, the corresponding peaks were several hours later, between 14.00 and 15.00 for primiparous women and between 08.00 and 09.00 for multiparous women.

An unpublished study of births in seven hospitals in northern Portugal after spontaneous onset of labour showed that neither spontaneous nor operative births peaked in the early hours of the morning (Cristina Teixeira, Polytechnic Institute of Bragança, 2015, personal communication).

These comparisons raise questions about styles of practice and settings for care. A particular question is why the timing of spontaneous births has retained its traditional early morning peak in some settings, but not in others. Another question is about the extent to which births following induction are concentrated into daytime hours.

Day of birth

In England and Wales, a study of daily variations in numbers of births in the 1970s found a pronounced weekly cycle, with numbers of births being lowest on Sundays, followed by Saturdays, and the highest numbers of births occurring from Tuesday to Friday. Numbers of births were low on public holidays, with Christmas Day and Boxing Day having the lowest numbers of births in each year. 1,86 Similar patterns were observed in a number of other countries87–95 and an analysis of births in Israel showed a deficit on Saturdays. 96

The patterns of birth in England and Wales changed over the 7-year period of 1970–6. The ratio of the average numbers of births on Saturdays to the overall daily average for the year decreased from 1.00 in 1970 to 0.93 in 1976, while the corresponding ratio for births on Sundays fell from 0.88 in 1970 to 0.77 in 1976. The ratio for births on Mondays rose from 0.92 in 1970 to 0.97 in 1976. It was not possible to tabulate these births by method of onset of labour or delivery, but data from a different source, the Hospital In-Patient Enquiry (HIPE), showed that induction rates rose steeply in the early 1970s, from 23% in 1970 to 39% in 1974, and then fell slightly, reaching 35% in 1976. 97

Data from other countries,87,90,93 and a subsequent special analysis of HIPE data for England and Wales,98 suggested that day of the week variations in both elective caesarean section and induction made a major contribution to the daily variations observed in births. In particular, a study in France compared data for 1946–50, when there was little day of the week variation, with data for 1968–75, by which time a deficit of births on Sunday had become well established. 95

Further analyses of patterns of birth in England and Wales showed that over the period 1979–96, weekly patterns of birth in England and Wales were more stable than they had been in the 1970s. The ratio of the average numbers of births on Saturdays to the overall daily average for the year ranged between 0.92 and 0.94 over the period, and the corresponding ratio for births on Sundays rose from 0.79 in 1979 to 0.83 in 1996. The ratio for births on Mondays rose slightly from 0.97 to 0.99 over the period. 2 A second, more detailed, analysis was undertaken of births over this period but the birth registration data again took the form of counts and were not linked to data about methods of delivery. 2 Separately from this, tabulations of data from the Hospital Episode Statistics (HES) for England from the mid-2000s showed the extent to which elective caesarean sections and, to a lesser extent, births following induction are more likely to occur on weekdays than at weekends. 99

Births in midwife-led settings

There has been very limited recent work on births in out-of-hospital settings. Births in midwife-led settings, including home births, are now low in number compared with births in OUs and, therefore, are more difficult to analyse.

The local study of births in Birmingham, England, in 1950 and 1951 referred to previously (see Time of day of birth) showed that numbers of domiciliary births, which accounted for a much larger proportion of all births than they do now, peaked from 02.00 to 07.00 and spontaneous births, which formed a majority of births in this analysis, followed the same pattern. 76

The analysis of day of the week variation in births in the 1970s found some day of the week variation in births occurring at home, but it was very much less marked than in hospital births. 100

More recently, additional analyses of the Birthplace data looked at timing of births planned in midwife-led settings in alongside and freestanding midwifery units and at home and compared them with hospital births in terms of likelihood of having a caesarean section, an instrumental delivery, a ‘straightforward’ birth or a ‘normal birth’. Times of births were subdivided into weekdays during ‘office hours’ from 09.00 to 16.59, weekday nights and weekends. 101

Women planning OU birth were less likely to have a birth without obstetric intervention if their birth occurred on a weekday during ‘office hours’ than if they gave birth at night. In births planned in AMUs and at home, there was no difference between time of day or day of the week and any of the main outcome measures.

Nulliparous women who planned to give birth in FMUs and who gave birth during weekday ‘office hours’ were less likely to give birth by caesarean section than nulliparous women who gave birth at night. Multiparous women who gave birth on weekdays during ‘office hours’ or at weekends were less likely to have a normal birth than those who gave birth on a weekday at night.

Overall, the authors suggested that their findings could be interpreted as consistent with a possible effect of non-clinical factors on intervention, but that there were other possible explanations and that further research was warranted. 101

Birth outcomes in relation to time and day of birth

Analyses of stillbirth and neonatal mortality by time of day and day of birth has a long history. Some early studies analysed outcomes by time of day of birth. The 19th century Brussels data included numbers of stillbirths, which were higher in the afternoons. 74 Stillbirth and neonatal mortality rates were the focus of an analysis in the early 20th century of births in New York State in 1929102 and 1936,75,103 and of an analysis of data from an unspecified location in the UK,104 which also found higher rates in the afternoon.

Subsequent analyses of mortality by time of day later in the 20th century found different patterns. Studies in Switzerland found that perinatal mortality rates were higher at night and that rates rose in births occurring through the late afternoon and the early evening and then began to fall in births occurring after midnight. 80,105,106 An analysis of births in Sweden from 1973 to 1995 found higher mortality rates among babies born at night, from 17.00 to 01.00, followed by a minimum around 04.00 and a second peak around 09.00, especially for preterm births. 107 In contrast, an analysis of neonatal deaths in California found that neonatal mortality rates were slightly higher in babies born at night, but were higher among babies born in the early hours of the morning than among those born before midnight. 106 A further study subdivided births in Sweden from 1991 to 1997 by risk of death and suggested that babies born at night were at higher risk of early neonatal death but not of intrapartum death. 108

A number of studies from the 1970s onwards analysed the outcome of pregnancy by day of the week of birth. An analysis of perinatal mortality in England and Wales among births in the years 1970–61,86,100 showed a weekly cycle, with higher rates among babies born at weekends, but warned that these were crude rates, so no conclusions could be drawn. A subsequent analysis of data for England and Wales for the years 1979–962 used more advanced statistical modelling techniques and found that stillbirth and early neonatal mortality rates tended to be higher among births on Saturdays and Sundays, respectively, than among births on other days, but this was also restricted to crude rates.

The findings of other studies were inconsistent. A study in Canada found slightly higher crude rates of stillbirths and neonatal deaths among births at weekends, but the difference disappeared after adjustment for gestational age,109 and a study in Australia found no difference after adjustment for birthweight. 2,91

An analysis by day of the week published while this project was under way used unlinked HES data for England and concluded that perinatal mortality was highest at the weekend,62 although the authors’ analysis and interpretation of their results was highly criticised in subsequent rapid responses. 110–112 Like past analyses of rates of stillbirth and perinatal mortality in England and Wales, this analysis made no distinction between intrapartum and antepartum stillbirths.

Data about the exact timing of death, subdivided into before labour, during labour or not known, are recorded on the medical certificate of stillbirth. The completeness of these data has improved in recent years but the quality has not been reassessed. Better quality data are available, however, from confidential enquiries. 113–115

An analysis of data for 1993–5 from the All Wales Perinatal Survey had pointed to higher rates of mortality related to intrapartum asphyxia in the holiday months of July and August and suggested a possible association with the rotation of junior doctors to new posts in August, but this was based on relatively small numbers of deaths. 116 Similar effects were reported in intrapartum-related deaths in Scotland. 117 The analysis of data for England and Wales for 1979–96 showed that all perinatal mortality rates varied depending on the time of year. 2

Like previous analyses for England and Wales, this did not take account of the time of day of birth as it was not included in routine data systems at the time. Analyses by day of birth and time of day by method of onset of labour and gestational age were not possible, as all the relevant data items were not recorded at a national level for England and Wales, even though they have long been recorded locally.

More recent analyses of the outcome of births by day of week and time of day from Scotland118 and the Netherlands119 that have dichotomised time of day and day of the week as ‘in hours’ and ‘out of hours’ have attracted considerable media attention and raised questions about the safety of ‘out-of-hours’ care. Both studies reported raised mortality rates among babies born outside usual weekday working hours. The analysis of Scottish singleton birth and neonatal mortality data for the years 1985–2004 showed raised neonatal mortality attributed to asphyxia at weekends and on weekdays outside the hours of 09.00 to 17.00. 118 The authors found no changes over time and suggested that the findings were also true in other countries of the UK.

A ‘rapid response’ based on data from the West Midlands from 1995 to 2009 showed, however, that similar patterns were found in the region in the late 1990s but had subsequently disappeared, and there was no evidence of a raised mortality rate ‘out of hours’ in the years 2005–9. 120

The inconsistent nature of previous findings could reflect the underlying complexity of contributory factors and the challenges inherent in examining cause of death questions. The provision of care, in particular the provision of facilities to allow rapid delivery and prompt resuscitation, would be expected to affect different types of death in different ways. These provisions might be crucial in the response to an acute event, such as uterine rupture. They would not, however, alter the course of a baby affected by, for example, renal agenesis, which is invariably fatal. Furthermore, there is considerable potential for bias within any such analysis. For example, both planned caesarean delivery at term, which carries a low risk of stillbirth,91,100,109 and high-risk births, such as elective preterm births, tend to take place during the normal working week and could contribute to either an overestimate or an underestimate of the relative risk of giving birth out of hours. Finally, stillbirths and neonatal deaths are rare events, especially intrapartum stillbirth, which accounted for only around 9% of stillbirths reported to MBRRACE-UK. Therefore, a large dataset is needed to analyse births and their outcome by day of the week and hour of the day. Recent developments in data recording and linkage in England and Wales mean that these analyses are now possible for large numbers of births from 2005 onwards, as are analyses of data about any subsequent hospital care for mothers and babies. As a result of the novel data linkage work described here, the analysis planned for the project and completed in part is the first to be undertaken.

Aims

To build on work done in a previous project to link data from civil registration, notification of birth to allocate NHS numbers to babies and data about care during delivery in order to analyse linked maternity and neonatal data for England and Wales about births in the years 2005–14 to compare daily, weekly and yearly variations in numbers of spontaneous and other births by time, day and season of birth and to compare variations in rates of adverse outcome.

Objectives

To achieve these aims, the objectives were to answer the following questions:

1. How do numbers of births vary depending on time of day, day of the week and time of year of birth and how does this relate to methods of onset of labour and delivery and multiplicity?

2. Subject to the availability of data, how do patterns of birth vary between maternity services in relation to variations in medical and midwifery staffing, patterns of intervention and size of unit?

3. How does the outcome of pregnancy in terms of rates of cause-specific intrapartum stillbirth and neonatal and infant mortality rates and rates of morbidity recorded at birth and at hospital admission in the first year of life vary depending on the time of day, day of the week and time of year of delivery in relation to gestational age and intervention in the onset of labour and delivery?

4. Have the patterns observed changed over the years 2005–14?

The project aimed to inform decision-making by providing information about how the numbers of births in England and Wales vary by time of day, day of week and day of the year. It separately analysed births before term and post-term births occurring after 42 weeks. In addition, it took account of whether labour and birth occurred spontaneously or whether their timing was affected by inducing labour or by undertaking either a planned caesarean section or an unplanned or ‘emergency’ caesarean section during labour, or before the onset of labour.

It was intended that the rates of death and severe problems in babies or their mothers would be analysed in relation to these factors and that the results would be related to information about midwifery and obstetric staffing in maternity units and NHS trusts and to their overall rates of induction and caesarean section. This information could have been used by the NHS in planning both levels of staffing in terms of midwives and obstetricians and in detailed rostering, with the aim of trying to match the numbers of women in labour and giving birth with the numbers of midwives available and the availability of specialist obstetric care should complications arise.

Data for England and Wales

Data for England only

Data for Wales only

Additional data

Birth registration and birth notification (‘ONS birth records’)

As a result of a project led by City, University of London, to link birth registration data with the birth notification dataset,5 the ONS now routinely links these two datasets and uses the linked data in its annual publications. 138 Linked data from this source for the years 2005–14 were provided to the project (see Figure 1). Linked birth registration/birth notification records are referred to in this report as ‘ONS birth records’.

Linked-in mortality data

Data from death registration had already been linked to ONS birth records (see Figure 1).

Data about deaths of babies under 1 year of age, linked to the corresponding ONS birth record, were provided to the project by the ONS for births in 2005 to 2014. The ONS also provided data for deaths at 1 year or over of age of children born from 2005 to 2014 occurring up to the age of 9 years.

The ONS provided records of deaths of mothers within a year of giving birth linked to the corresponding ONS birth records for births from 2008 onwards.

Data for linkage

Data to be used in future analyses

Linkage of ONS birth records for England with Maternity HES delivery records

Linkage of ONS birth records for England with Maternity HES birth records

Quality assurance of the linkage to delivery records

This section describes the extensive quality assurance process for the linked HES delivery records and ONS birth records only. Similar quality assurance of the linked HES birth records and ONS birth records could not be achieved within the project time scale. The aim was to assess the linked records to efficiently identify and preserve one correctly linked HES delivery record with the maximum amount of delivery information from all possible links to create a file to be used in analysis.

Only the ONS birth records that linked to a mother’s [HESID] and HES delivery record could be quality assured. There were instances where the linkage algorithm did not link an ONS registered birth to a HES delivery record. This could be because no delivery record existed. For example, the majority of home births and births in private health facilities are not usually captured by HES. In addition, some trusts may not have submitted HES data at particular times, possibly if they were experiencing problems with their IT systems. There may also have been a linkage error. Overall, 1.7% of singleton and 4.3% of multiple ONS birth records without a link to a HES delivery record did have a link to the mother’s [HESID]. There has not been time to investigate these problems further.

The quality assurance work was carried out in the secure environment of the ONS VML. Microsoft Access^® (Microsoft Corporation, Redmond, WA, USA) database software was used as it was the only database software available in the VML at that time, and Structured Query Language (SQL) code was used for replicability. The relatively large size of the data files and restrictions on database size and processing speed constrained the design of the methods used. These are described further in Appendix 1.

Data preparation and data quality checks are described in Appendix 2.

A pilot was initially carried out on data for 2005 to help formulate the methodology. The procedure described below in this section is the final version that was used for the linked ONS births and HES delivery data for births in all of the 10 years. Clerical checks were carried out throughout the development phase using relevant variables and identifiers to inform and improve the procedure.

To determine which links to preserve or discard, the links between the ONS birth records and HES delivery records were categorised either as ‘same mother same baby’ or ‘same mother different baby’. The latter category accounts for instances in which the same mother has given birth to another baby within the 10-year study period. The category ‘wrong link’ was used when neither of the previous categories applied, suggesting incorrect maternal linkage or data quality problems. These scenarios are summarised in Table 8. They exclude occasions when a birth registration record did not link to any HES delivery records.

An ONS birth record may link to one correct or incorrect HES delivery record, or to more than one HES delivery record. Reasons for one, none or some of these records being correct are shown in Table 8.

The quality assurance exercise aimed to keep ‘same mother same baby’ linked records and move all others to the unlinked file. If more than one correct link was available, only one linked delivery record containing the maximum amount of delivery information was chosen. This was to create a one-to-one relationship and support ease of analysis.

Baby data items were compared to ascertain that the linked records related to the same baby. These were the baby’s date of birth, sex, gestational age, birthweight and location of birth.

The codes used for locations of birth differed between the datasets and also varied over time as a result of organisational changes within the NHS. These were translated into a consistent common location code, the ‘derived location code’, and allocated to an ‘assigned trust’ designed specifically for the study, as described in Derivation of variables and in Appendix 2. Lookup tables were created to map all the location of birth codes ([CESTRSS] from birth registration records, [ORGCODEZ] from birth notification records and [PROCODE]/[SITETRET] combinations from HES delivery records) to ‘assigned trusts’. If ‘assigned trusts’ matched, this indicated a linkage match. This proved to be a test that was too sensitive, resulting in too many false negatives where the assigned trust of birth did not match when all the other baby items matched. This may have resulted from women transferring between nearby units or coding conventions between trusts in a local neighbourhood. A further lookup table was therefore created, which resulted in a ‘geographic match’ where the main OU in a trust was within 25 miles of an OU within another trust. This distance was reduced to 10 miles for London, and greater distances were used in rural areas such as Wales and South West England. This was developed using latitude and longitude co-ordinates from the BirthChoiceUK maternity unit database (used with the permission of Rod Gibson Associates Ltd) to calculate a matrix of distances between each trust’s OU. Home births and births taking place elsewhere were designated geographic matches by default as the codings in the birth notification and HES datasets did not distinguish these birth locations well.

Four combinations of the baby data items were used to check whether or not a linked ONS birth record and HES delivery record related to the same baby, in order of decreasing certainty from 1 to 4, as shown in Table 9. Combination 1 required all the data items to match. Combination 2 required geographic location and date of birth to match plus either birthweight or gestational age or sex. Combination 3 required only geographic location and date of birth to match and birthweight, gestational age and sex will not match in these cases. Combinations 2 and 3 took into account the high occurrence of missing values and poor data quality in HES. Combination 4 accounted for a degree of data entry error in the date of birth by allowing it to differ by up to 4 days, but only if backed up by a matching geographic location of birth, and with birthweight, gestational age and sex not differing very much between ONS and HES if the values were present. If they were different and the date of birth was not an exact match, it was considered to be a different baby. The four combinations were applied in order so that the most reliable matches were identified first.

If the linked records referred to the same baby and there was more than one linked HES delivery record to choose from, then the record with the most onset of delivery and delivery method information was chosen, as these variables were of key importance for the analysis. The presence of valid values in these and other fields were used to inform this.

The procedure was implemented using deterministic stepwise rules to be carried out in order, with any linked records dealt with at each stage excluded from subsequent stages. A summary is given in Box 1. Fuller details are provided in Appendix 2 (see Table 29). It differed slightly between singleton and multiple births to account for the greater complexity of linkage of multiple births.

The first two rules cleaned the HES delivery data by eliminating invalid linked records and duplicates. It is known that some trusts generate multiple copies of the same delivery episode record, usually six or nine copies of the same record. In these cases, only one was kept to be quality assured and the others were discarded for efficiency.

Once all rules in the procedure had been carried out, it was confirmed that all linkages had been assigned a decision, and that there were no more records to check.

Further checks were made on the following:

1. Private hospital births should not have a link to a HES delivery record unless the mother or baby were later transferred to a NHS hospital for any reason.

2. Each HES delivery record should link to the correct number of ONS birth records depending on whether it is for a singleton or multiple delivery.

3. The baby’s date of birth in the ONS birth record should fall within the HES delivery admission period. This was not always the case even when it was a confident link. The check could not be made for home births as there were no admission dates.

4. If there was more than one correct and relevant linked HES delivery record for an ONS birth record, any missing onset of delivery and method of delivery information was filled in from these.

The quality assurance procedure took the equivalent of 55 working days for singleton births and 16 days for multiple births. Microsoft database software and processing limitations of the VML at that time slowed the procedure down. Microsoft SQL Server has since been made available in VML and would speed up this process considerably.

Seventeen per cent of linked records for singleton births were dealt with in the first two cleaning stages, and a further 80% were dealt with by rules 3 and 4. All the remaining rules were required to deal with the residual 3% of more complex cases.

Five per cent of linked records for multiple births were dealt with in the first two cleaning stages, and a further 60% were dealt with by rules 3 and 4. This left 35% as residual records, reflecting the greater complexity inherent in records of multiple births.

For each year of birth, the complex residual cases were clustered within particular trusts. This was probably attributable to IT problems and changes to systems in those trusts.

A summary of the results for singleton and multiple births by calendar year is given in Appendix 2 (see Table 30).

Of the total 6,468,586 singleton ONS birth records, 97% had been linked to one or more HES delivery records and could be quality assured. Of these, 98% remained linked to one HES delivery record and 2% had all HES links discarded after quality assurance. Ninety-five per cent of all ONS singleton birth records overall were left with a link to a HES delivery record for analysis.

Of the 208,326 multiple births, 95% were linked to one or more HES delivery records and could be quality assured. Of these, 98% remained linked to one HES delivery record and 2% had all HES links discarded after quality assurance. Ninety-three per cent of all ONS multiple birth records overall were left with a link to a HES delivery record for analysis.

For singleton births, the percentage of all birth records left linked to one HES delivery record after quality assurance increased fairly consistently over time, from 93% in 2005 to 95% in 2014. For multiple births, the rate increased from 94% in 2005 to 97% in 2010, but then started to decrease again to 87% in 2014. This reflects the same patterns as the percentage of ONS birth records that were originally linked to a HES delivery record and are biased by the number of records that linked to a [HESID] but not that delivery record, were excluded from quality assurance.

The majority (76%) of the discarded linked records for singleton births were attributable to duplicate copies of HES delivery episodes and were dealt with in the preliminary cleaning stage of the quality assurance procedure. The remaining reasons for discarding links are discussed in detail in Appendix 2. Of interest are the 4,854 links categorised as ‘no-matches’. Although they are relatively small in number, they give insight into linkage error and quality assurance error. Some missed links were also discovered, raising further questions about the quality of the linkage, discussed in Appendix 2. General insights into data quality issues that were revealed by the quality assurance are also given.

Certain features were more common in incorrectly linked records than in correctly linked ONS birth records. These included births linked using a less reliable stage of the linkage algorithm that excludes the NHS number, births at home and in the London region, and births where the birthweight or gestational age was missing in HES. This suggests some bias in the linkage and often relates to data quality issues. These patterns differed between singleton and multiple births. Fuller details are given in Appendix 2.

Quality assurance of HES birth records linkage

The quality of linkage of the baby file has yet to be assessed.

Linkage of ONS birth records for Wales to Patient Episode Database for Wales/National Community Child Health Database

Linkage to CMACE stillbirth and neonatal death records

In this project, the CMACE stillbirth and neonatal death data for England and Wales were linked to stillbirth registration records in order to distinguish between antepartum and intrapartum stillbirths, as at registration this information is incomplete and, where coded, the quality has not been assessed. For neonatal deaths, CMACE used a different classification to record the cause of death information and it was considered that this would provide a useful comparison to the information recorded on the death certificate.

Database construction

In order to hold the data for analysis, database structures were needed. For much of the duration of the project, the only available database was Microsoft Access^®. As Microsoft Access^® databases have limited capacity, they are unsuitable for holding large datasets, so subsets of data had to be physically stored in individual annual files. This led to a network of 50 or more interlinked databases, making processing very slow. It was not until relatively late in the project, in the summer of 2016, that Microsoft SQL Server was made available to us. This allowed all the ONS data for England and Wales combined and the HES data for England, including readmission data, to be stored in a single database and be processed at speeds that were previously unimaginable.

The implications of this in terms of delay to the project, notably the slow processing when using Microsoft Access^® and the need to spend time re-programming for Microsoft SQL Server, are described more fully in Appendix 1. As the initial importing of much of the data was performed in Microsoft Access^®, this is explained in some detail in Appendix 2 to facilitate the audit trail back to the original data. Appendix 2 also describes how the linked data received from the HSCIC were prepared for storage in the database.

The following datasets were stored in the database environment:

1. Birth registration records linked to birth notification records provided by the ONS.

2. Stillbirth, infant and child mortality and maternal mortality records provided by the ONS.

3. Mothers’ HES delivery records.

4. Mothers’ HES non-delivery admission records. These consisted of antenatal admissions, postnatal admissions and admissions unrelated to maternity.

5. Babies’ HES birth records.

6. Babies’ HES subsequent admission records. These included postnatal admissions and subsequent admissions unrelated to the child’s birth, for example as a result of accidents.

The structure of these records has considerable numbers of many-to-one relationships, for example successive births to the same mother, more than one birth record in a multiple delivery and successive subsequent admissions of mothers and babies. This structure is supported in the relational database environment, summarised in Figure 2. In this, each box refers to a separate table in the SQL database. This is essentially identical to the Microsoft Access^® data structure except it now includes readmissions and data for births in all the years are in one table.

FIGURE 2.

Relational database structure for HES delivery and birth records linked to ONS linked birth notification/birth registration records.

In the following text, the names of database fields are enclosed in square brackets. The word linkage is taken to mean the link between database tables using unique keys and the phrase ‘record linkage’ to mean the matching of ONS to HES records.

In Figure 2, arrows represent possibilities for linking tables using keys. Black lines represent links that must exist, for instance every delivery ICD-10 code record must be associated with a particular delivery record. Green lines represent links that may exist. For example, a delivery record may link to an ONS record. The primary key (PK) is unique. An example of a foreign key (FK) would be [Delivery ID] in the Deliveries Baby Tails table. For twins, this table had two records each with the same [Delivery ID] pointing to the same delivery record. Although labelled FK in this diagram as this is their role, these are not declared as FKs in the database. FKs impose referential integrity, which is not a function that was required in this project but these fields have been indexed. Appendix 2 lists all the unique PKs in each database table.

In HES, a mother or baby may be diagnosed with a variety of conditions and undergo a number of procedures. There is space for up to 20 diagnostic codes and 24 procedure codes. To efficiently analyse these records, it is necessary to normalise the data so that questions can be asked of them without having to consider the multiplicity of the many-to-one relationship. For example, to determine if a mother had an episiotomy it is tedious to work through all 24 fields for operation codes in her HES record, but if the data are normalised in a relational database this is reduced to one question on a single field.

Derivation of variables

New variables were derived in the ONS and HES delivery datasets based on standardising or combining the original values in the data fields to create better quality or more useful variables. These are listed in detail in Appendix 2. Some derived variables involved fairly minor manipulation, but much more work was required to impute the two key variables of parity and location of birth codes. Parity is very incompletely recorded in HES and was not recorded for all births at birth registration until May 2012. Location codes for maternity units were derived to overcome the many changes in organisation codes resulting from NHS reorganisations over the period and the different ways of coding for site of birth in the datasets used.

Database documentation

All database tables, fields and descriptions are listed in Appendix 2. This includes the metadata obtained from the ONS about birth registration, birth notification and death registration, as well as metadata about variables obtained from the HSCIC in HES delivery and birth records and in records of subsequent admissions of mothers and babies after birth.

The list of preferred variables for use in analysis is given in Appendix 2, Preferred variable list and the means of accessing these from standard statistical packages are given in Appendix 2, ODBC link.

The relational database brings together the three datasets – birth registration, birth notification and HES – for England for the 10 years from 2005 to 2014 together with linkages to mortality data and readmission records. Because of the structure of the database and the new facilities in the VML, it is now possible to perform complex analyses with relatively little processing time. Using Open Database Connectivity (ODBC), data can be extracted from the database directly into other statistical software applications. SQL tables can be linked to using Microsoft Access^®, or applications such as Statistical Analysis System [SAS^® (SAS Institute Inc., Cary, NC, USA)], Stata^® (StataCorp LP, College Station, TX, USA) and R (The R Foundation for Statistical Computing, Vienna, Austria) can import entire datasets.

Number of observations used in the analysis

There were 7,013,804 births registered in England and Wales that occurred from 1 January 2005 to 31 December 2014. Of these, there were 6,676,912 babies whose mothers were resident in England and 336,892 babies whose mothers were resident in Wales. Multiple and singleton births were identified using information on multiplicity recorded at birth registration. There were 6,468,586 singleton and 208,326 multiple birth records in England that were linked to HES delivery records by HSCIC. After quality assurance, which was described in detail in Chapter 2, 6,138,487 singleton births and 193,270 multiple birth records remained linked to HES delivery records (Figure 3).

FIGURE 3.

Births registered in England and Wales by multiplicity and linked to the hospital delivery records: 2005–14. QA, quality assurance.

Of the 327,263 singleton babies either resident or born in Wales, 300,385 birth records were linked to the NCCHD and PEDW by NWIS. There were no duplicates and the quality of linkage was good; therefore, no records were removed after quality assurance. Similarly, 9014 multiple birth records were linked to the NCCHD and PEDW by NWIS and all records remained linked after quality assurance (see Figure 3).

Variations in overall numbers of births in England and Wales by time of day and day of the week

Over the period from 2005 to 2014, the overall numbers of births in England and Wales showed the same pattern as in preceding years, with a regular weekly cycle with the numbers of births each day increasing from Mondays to Fridays, with lower numbers of births on Saturdays and the lowest numbers of births on Sundays. In addition, numbers of births varied seasonally, with a peak in late September. As shown in Figure 4, this pattern was interrupted by public holidays, with the lowest numbers of births in most years being on Christmas Day and Boxing Day.

FIGURE 4.

Daily live births in England and Wales: 2005–14.

To illustrate this, data for 1 year, 2011, are shown in Figure 5, in which it can be seen that numbers were relatively low at New Year, at Easter and on other public holidays, including in early May when the usual pattern of public holidays was disrupted by an extra public holiday related to a royal wedding. Visual inspection suggests that these dips were preceded by short-term increases in numbers of births. This was investigated in the analyses described below.

FIGURE 5.

Number of live births by day, England and Wales: 2011.

The overall numbers of births each year between 2005 and 2012, apart from a slight decrease in 2009, and then fell slightly in 2013 and 2014. To investigate whether or not the daily patterns changed over time, the average numbers of births on each type of day were compared with the daily average for the corresponding year.

The daily pattern changed very little over the period 2005–14 (see Appendix 3, Table 73). This is illustrated in Figure 6. For clarity, it shows only 2005, 2009 and 2014, namely the first, middle and final years of the project period.

FIGURE 6.

Ratios of average numbers of daily live births to overall average for the year, England and Wales: 2005, 2009 and 2014.

The data in Figures 5 and 6 come from a special analysis by the ONS of all live births. In what follows, we have used the data acquired for the project to describe variations in singleton births, as described earlier.

Variations in numbers of singleton births in England and Wales by day of birth and by time of day of birth

Variations over time

It has already been shown above that the daily variations in numbers of live births were stable over the period 2005–14. The overall hourly pattern for all singleton births is shown in Appendix 3, Table 74, which shows the ratio of average numbers of births per hour to the overall average for the year for 2005–14. As with data for daily variations in numbers of births, data about hourly patterns are shown for births in 2005, 2009 and 2014 in Figure 7. Again, because the overall numbers of births each year rose and fell over the study period, we have expressed the average numbers of births each hour as a percentage of the overall hourly average. There was relatively little change over the years studied.

FIGURE 7.

Average numbers of singleton births per hour in relation to overall hourly average for each year, England and Wales: 2005, 2009 and 2014.

The exception was that in 2005 and other early years of the period, there was a sharp peak in numbers of births between 10.00 and 10.59. This changed slightly over the 10-year period and by 2014 and immediately preceding years, this had broadened out to cover the time period from 09.00 to 11.59, and the smaller peaks early and late in the afternoon had become less prominent.

Variations by time of day and month

Singleton births were tabulated by month and time of day. The average numbers of births per hour were then calculated. As there is known seasonality in numbers of births, the ratios of the hourly averages to the overall monthly averages were then calculated for the years 2005–14 combined. This was done separately for term births, late preterm births and very preterm births (Figures 8–10). Although ratios were calculated for all 12 months, only 4 months’ ratios are shown. They were chosen to be December, containing the shortest day, June, containing the longest day, and March and September, containing the spring and autumn equinoxes, respectively. The time of day of births in October 2008 were known to be missing, so these births were omitted from analyses of time of day but not from analyses of day of birth.

FIGURE 8.

Average numbers of singleton births at 37–42 weeks of gestation or more per hour in relation to overall hourly average for each month, England and Wales: 2005–14.

FIGURE 9.

Average numbers of singleton births at 32–36 weeks of gestation per hour in relation to overall hourly average for each month, England and Wales: 2005–14.

FIGURE 10.

Average numbers of singleton births at 23–31 weeks of gestation per hour in relation to overall hourly average for each month, England and Wales: 2005–14.

Numbers of term births showed a pronounced peak in the morning and a smaller peak in the afternoon (see Figure 8), as with all singleton births (see Figure 7). There was little sign of any differences between months of birth.

Numbers of late preterm births also peaked in the afternoon, but to a much greater extent than term births (see Figure 9). Although there appeared to be greater differences between the months of birth, these did not show a recognisably consistent pattern and further analysis is needed to assess whether or not the differences were bigger than would be expected by chance.

The very much smaller numbers of early preterm births showed a more erratic pattern, with signs of three peaks, in the morning, the early afternoon and the early evening, in some months and four peaks in others (see Figure 10). Without further analysis, it is impossible to assess the extent to which these were compatible with random variation. It was clear, however, that hourly numbers of births were below average from midnight to 09.00 and tended to be above average between 09.00 and midnight. This pattern was discernible but much less marked for late preterm births.

Characteristics of mothers and babies

Mothers’ ages

When numbers of singleton births were analysed by day and by mothers’ ages grouped into 5-year age groups, marked differences were apparent. Initially, mothers aged under 15 years and those aged 45 years and over were tabulated separately, but as the numbers were small they were combined with adjoining age groups.

Variations by day of birth are shown in Table 14 and Figure 11. These show the same overall patterns as for all births, but there is a considerable age gradient, with relatively small daily differences for women aged under 20 years to a much larger concentration of births to women aged 35 years and over onto weekdays.

TABLE 14 - Average numbers of singleton births by day and age of mother, England and Wales: 2005–14

Day of the week	Age group (years)
	Under 20	20–24	25–29	30–34	35–39	40 and over	All
Monday	102	344	513	551	317	75	1,902
Tuesday	106	356	526	559	319	76	1,942
Wednesday	108	358	530	561	319	76	1,952
Thursday	109	363	538	568	324	77	1,979
Friday	110	365	537	566	322	78	1,978
Saturday	105	333	474	477	253	57	1,699
Sunday	100	317	450	450	238	52	1,607
Public holiday	107	336	474	477	253	57	1,705
All	106	348	509	531	297	70	1,861

FIGURE 11.

Ratio of average daily numbers of births to overall daily average by age of mother, England and Wales: 2005–14.

The same applies to variations by time of day (Figure 12). Numbers of births show a morning peak and a smaller afternoon peak, but the variation is much less marked in births to younger women than in births to older women.

FIGURE 12.

Ratio of average hourly numbers of births to overall hourly average by age of mother, England and Wales: 2005–14.

Gestational age and birthweight

Singleton births are analysed by gestational age and birthweight by day in Figure 13. As 91% of babies born at 22–31 weeks’ gestation weighed less than 2,500 g, they are not subdivided by birthweight (see Figure 13c), although it may be useful to subdivide this group in future analyses to gain an insight by the timing of the most preterm births.

FIGURE 13.

Ratio of average numbers of singleton births to overall daily average by birthweight, England and Wales: 2005–14.

These analyses suggest that, for term and late preterm births, babies weighing under 2,500 g are more likely to be born on Thursdays and Fridays and less likely to be born on Sundays and Mondays than heavier babies. Variations by time of day, shown in Figure 14, show differences within both late preterm and term births between low birthweight babies and those weighing 2,500 g or more. They were more likely to be born between 09.00 and 12.00, in the afternoons and in the hours around midnight.

FIGURE 14.

Ratio of average hourly numbers of singleton births to overall daily average by birthweight, England and Wales: 2005–14.

Settings for birth

Using the derived location codes, all singleton births were analysed by time of day and day of the week for each type of birth setting. As described in Chapter 2 (see also Appendix 2), maternity units with a consultant OU were subdivided according to whether or not they also had an AMU within the same hospital. Unfortunately, it was not possible to distinguish between births occurring in an OU or an AMU on the same site. FMUs at premises without a consultant OU are shown separately. There were relatively small numbers of births in private and military hospitals and even smaller numbers of births in hospitals with no maternity facilities (Table 15). There were also a few births recorded as taking place ‘elsewhere’, that is, in a range of other places, including in transit to maternity facilities. There is a marked day of the week pattern, with numbers of births in hospitals with OUs increasing from Monday to Friday (Figure 15a). The numbers are considerably lower on Saturdays and public holidays and much lower on Sundays. Although there is a weekly cycle in numbers of births at home and in FMUs, it is much less marked and there is no deficit in numbers of births on public holidays. Daily variations in other types of birth setting (see Figure 15b) are much wider, but are based on much smaller numbers and so would have had much wider confidence intervals. Births in private and military hospitals are much more heavily concentrated into weekdays than those in NHS hospitals and show much bigger deficits at weekends. Numbers of births in the other settings show weekly cycles that are less pronounced.

FIGURE 15.

Variations in singleton births by day of the week, England and Wales: 2005–14.

When numbers of births are plotted by time of day, they also show different patterns. In Figure 16, they are also shown as ratios. Numbers of births in hospitals with OUs peak in the mornings from 09.00 onwards, then, after a much smaller afternoon peak, they decrease after 17.00 and increase again up to midnight and then fall off after 05.00–06.00 (see Figure 16a). Numbers of births in FMUs and at home were highest from midnight to 10.00, peaking from 04.00 to 06.00. They were lowest during the day and during the hours up to midnight (see Figure 16a). The patterns in the other settings are much less distinct, as they are based on small numbers, but there is a sharp peak at 08.00 in births in military hospitals and morning peaks followed by early and late afternoon peaks in private hospitals (see Figure 16b). Births that occur either in hospitals without maternity units or elsewhere are unlikely to have occurred there intentionally and it is notable that those in non-maternity units tend to occur in the morning, whereas those ‘elsewhere’ follow similar patterns to births at home and in FMUs, although they have a flatter peak from 04.00 to 08.00.

FIGURE 16.

Variation in singleton births by time of day, England and Wales: 2005–14.

It is clear that the patterns observed for hospital births are related to obstetric intervention and that the hourly patterns vary by day of the week. Singleton births in hospital were analysed for England and Wales separately in what follows, as information on onset of labour and method of delivery was coded differently in HES and the PEDW.

Methods

Data for analysis

The steps taken to derive a subset of data for analysis are illustrated in Figure 17. The source dataset consisted of all 6,468,586 registered singleton births occurring from 1 January 2005 to 31 December 2014 in England. After linkage and quality assurance, 6,138,487 singleton births were judged to be linked to the correct HES delivery record. A total of 542 additional records were excluded because, after a careful comparison of the records, it was ascertained that the linkage between birth registration and birth notification records was incorrect. As shown in Figure 17 and Table 77, the resulting subset contained 6,137,945 records, that is, 94.9% of the source dataset. Further investigation was carried out to assess whether or not the births in this subset differed from the source dataset with respect to region of residence, year, month, type of day and hour of birth, age of mother, sex and gestational age of baby. These analyses are reported in detail in Appendix 3, Table 78. For each of these variables, there was a statistically significant difference between the distributions of births included in and excluded from the derived subset. Births in the years from 2009 onwards were more likely to be linked than those in the years up to 2008. However, the differences between the source dataset and the derived subset were small: below 1 percentage point for any category.

FIGURE 17.

How the subset was derived for analysis.

The subset used for these analyses included 6,054,536 births recorded as having taken place in NHS maternity units, excluding 83,409 births that took place at home, in private hospitals, in NHS hospitals without maternity facilities, or elsewhere outside a hospital. Births outside hospitals and those occurring at home have low rates of inclusion in Maternity HES and, as a consequence, low rates of linkage, so overall data quality would have suffered had they been included in the analysis.

Finally, 960,921 birth records with missing or inconsistent data were excluded from the subset. These records included 898,733 with missing data about the onset of labour, a key variable, 11,042 with mode of birth missing, 37,114 with gestational age missing and 14,032 with inconsistent combinations of data items. The excluded births made up 15.9% of the subset, leaving data for 5,093,615 births, as shown in Figure 17 and Table 16. The records excluded because of missing information were compared with those retained to see if this changed the distribution of births with respect to region of residence, year, month, type of day and hour of birth, age of mother, sex and gestational age of baby.

These comparisons are reported in detail in Appendix 3, Table 80, and show statistically significant differences between included and excluded births for all variables, except the baby’s sex. Nonetheless, differences between all linked NHS hospital births and the subset were small, the biggest difference being 1.02 percentage points for a category. Relatively fewer births in later years had to be excluded, compared with births in earlier years, as the completeness of data had improved.

Results

Daily and weekly patterns of births by onset and delivery

Table 16 shows the number of births by onset of labour, mode of birth and gestational age group.

TABLE 16 - Numbers of singleton births in NHS maternity units by onset of labour, mode of birth and gestational age group: 2005–14

Among the elective caesarean births, 445,543 were recorded as caesarean onset and the remaining 24,450 were recorded as ‘spontaneous onset’. We combined these two categories because the latter group was too small for meaningful estimation.

Onset of labour	Mode of birth	Number of births				Percentage
		Gestational age			Total
		Term	Preterm	Post-term
Spontaneous	Spontaneous	2,401,010	121,915	45,487	2,568,412	50.4
	Emergency caesarean section	308,248	29,440	11,121	348,809	6.8
	Instrumental	425,955	21,542	12,472	459,969	9.0
Induced	Spontaneous	577,544	38,249	60,075	675,868	13.3
	Emergency caesarean section	188,413	11,913	38,335	238,661	4.7
	Instrumental	162,882	7,168	28,006	198,056	3.9
No laboura	Elective caesarean sectiona	441,500	23,879	4,614	469,993	9.2
	Emergency caesarean section	84,384	44,465	4,998	133,847	2.6
Total		4,589,936	298,571	205,108	5,093,615	100.0
Percentage		90.1	5.9	4.0	100.0

Figure 18 shows the average number of births per hour in NHS maternity units in England over the course of the week and on public holidays. The pattern is the same, whether we look at all singleton births in NHS maternity units in England (green line), or the somewhat smaller subset of births for which we have complete data and that we could include in our analyses (black line). This suggests that our subset is representative of the timing of births in NHS maternity units in England from 2005 to 2014.

FIGURE 18.

Average number of singleton births in NHS maternity units per hour over the week, England: 2005–14.

The peaks in mean frequencies occur on weekdays between 09.00 and 12.59, with a smaller secondary peak between 14.00 and 14.59 on weekdays. A second period of relatively high numbers of births, although with a much smaller peak, occurs between 22.00 and 06.59, regardless of the type of day. Low numbers of births can be seen between 08.00 and 08.59 on weekdays and also at weekends and on public holidays.

It is important to consider the proportion of births that occur outside ‘usual’ working hours. Overall, 71.4% of all singleton births in NHS maternity units and 71.5% of those in the analysis subset occurred ‘out of hours’, that is at weekends, on public holidays or on weekdays between 17.00 and 08.59. Conversely, 28.6% of all births in NHS hospitals and 28.5% of births in the analysis subset occurred within ‘usual’ working hours, that is on weekdays between 09.00 and 16.59. As in a ‘normal’ 40-hour week, ‘usual’ working hours account for 23.8% of the total of 168 hours in a week, the extent to which births have been concentrated into ‘usual’ working hours is limited.

Figure 18 shows the overall daily and weekly cycles of births, whereas Figure 19 demonstrates how these cycles differ by onset of labour and mode of birth. There are clear differences in circadian and weekly patterns. For a more detailed view of average numbers of births per hour in each of the eight combinations of mode of onset and births, for a typical weekday (Thursday), see Appendix 3, Figure 27. Graphical data exploration, which is not shown here, suggested that the circadian patterns do not vary appreciably by gestational age group. We therefore concentrated on circadian patterns in term births in this analysis but we also compared weekly patterns of births in preterm, term and post-term gestational age groups (see Figure 20).

FIGURE 19.

Average number of term births per hour by type of day, onset of labour and mode of birth, England: 2005–14.

Spontaneous onset

Figure 19a shows the average numbers of births per hour with spontaneous onset over the course of a week, by mode of birth. Figure 19d then shows instrumental births and emergency caesarean sections on a larger scale to bring out their patterns more clearly. Just over half of all births in our dataset are spontaneous births following spontaneous onset of labour. These have a roughly sinusoidal pattern and are most likely to occur between 01.00 and 07.00, with a peak around 04.00, and a trough in the afternoon. In addition, there is a small but noticeable interruption of the sinusoidal pattern at 08.00.

In contrast, instrumental births after spontaneous onset of labour are more likely to occur between 09.00 and 17.00 than in the night and early morning. The least frequent time for births in this group is 08.00. Births where spontaneous onset of labour was followed by an emergency caesarean section are more frequent between 09.00 and 00.00 than in the early hours of the morning. There are three distinct dips in the distribution, the most pronounced of which is seen at 08.00, with somewhat smaller dips at 13.00 and 20.00.

In general, the patterns of births after spontaneous onset over the course of the day do not vary much between weekdays, weekends and public holidays. Nonetheless, numbers of spontaneous births after spontaneous onset are slightly lower on Sundays and public holidays than on weekdays, as we explore in detail below.

Induced onset

Figure 19b shows the average number of induced births per hour over the course of a week, by mode of birth. Figure 19e then shows instrumental births and emergency caesarean sections within this group on a larger scale. Spontaneous births after induced onset of labour follow a sinusoidal pattern, with a peak during the hour before midnight and a trough just before noon.

The same midnight peak appears when inductions are followed by instrumental births or emergency caesarean sections, although in these groups the patterns are less regular than for induced spontaneous births. The three groups of induced births share a weekly cycle that has its highest peak on Friday night and its trough on Monday morning.

Caesarean births

Figure 19c shows the distribution of caesarean births without prior onset of labour separately for elective and emergency caesarean sections. Elective caesarean sections are mostly carried out on weekdays between 09.00 and 12.00, with a pronounced peak between 09.00 and 10.59. Very few occur between 17.00 and 07.00 on weekday evenings and nights, and even fewer at any time at weekends and on public holidays.

Emergency caesarean sections without prior onset of labour follow a somewhat similar pattern to elective caesarean sections, with a pronounced peak between 09.00 and 12.00. There is a difference, however, as average numbers of emergency caesarean sections without prior onset of labour are higher than average numbers of elective caesarean sections in the evening, at night and in the early morning, and the same applies to weekends and public holidays.

Differences between preterm, term and post-term births

Next we considered whether or not the differences by type of day within each combination of onset of labour and mode of birth for preterm and post-term births were similar to those for term births. Estimated RRs and their confidence intervals, and p-values relating to the interaction effect for all eight combinations of onset of labour and mode of birth, are shown in Figure 20. Our analysis suggests that in six of the eight combinations of onset of labour and mode of birth, differences by type of day were not the same for all gestational age groups. The exceptions were emergency caesarean sections and, to a lesser extent, instrumental births after spontaneous onset. For these, the interactions between gestational age and ‘type of day’ in predicting number of births per day were not statistically significant (see Figures 20b and 20c).

FIGURE 20.

Estimated RRs and 99% confidence intervals by type of day and gestational age group, England: 2005–14.

For spontaneous births after spontaneous onset, shown in Figure 20a, there was evidence of differences by type of day for term births, as reported above. However, there was no evidence of differences by type of day among preterm births. As numbers of post-term births were much lower, all the confidence intervals for post-term births were wider than, and included, the intervals for term births. This meant that no statistically reliable differences could be detected between these two groups.

Among induced births (see Figures 20d–f), the average numbers of births by day followed the same overall pattern regardless of gestational age group, but differences between types of days tended to be widest among preterm births and narrowest among post-term births.

For elective caesarean sections (see Figure 20g), the pattern among post-term births was similar to that among term births, although differences between types of days were somewhat smaller for post-term births. Preterm elective caesarean sections, followed a different pattern, however, with numbers being highest on Fridays and on weekdays before a public holiday, and lower on Mondays and weekdays after a public holiday than on other weekdays.

Finally, in emergency caesarean sections without prior onset of labour, again, there is little to suggest differences between post-term and term births. Preterm births in this group followed a similar pattern to preterm elective caesarean sections, however, with numbers being highest on Fridays and on days before public holidays.

Methods

Results

In Wales, 6% of births in NHS hospitals were preterm (before 37 weeks), 89% occurred at term (37–41 weeks) and 5% were post-term (42 weeks or more). Just over half of all preterm births occurred spontaneously compared with 64% of term births. For post-term births, over 51% of babies were born instrumentally and 26% were born by emergency caesarean section (Table 19).

As in England, these differences in mode of birth were associated with differences in the timing of birth, as can be seen in Table 20 and Figures 21–23. Because the numbers of births in Wales were smaller than for England, the numbers of births on public holidays were not shown separately, being extremely small.

FIGURE 21.

Singleton preterm births by time of day, day of the week and mode of birth, Wales: 2005–14.

FIGURE 22.

Singleton term births by time of day, day of the week and mode of birth, Wales: 2005–14.

FIGURE 23.

Singleton post-term births by time of day, day of the week and mode of birth, Wales: 2005–14.

Numbers of preterm births remained at fairly similar levels from Monday to Saturday, but were lower on Sundays. Numbers of both elective and emergency caesarean sections increased from Monday to Fridays and were highest on Thursdays and Fridays. The variations by time of day and day of the week are shown in Figure 21. Along with post-term births, they were aggregated into 3-hour time slots, as the numbers were small. This shows that numbers of emergency caesarean section peaked later in the day than elective caesarean section and that numbers of spontaneous preterm births peaked in the early hours of the morning, as Figure 22a shows.

Among term births, spontaneous births predominate and show similar patterns to those in England as Figure 22a shows. The much smaller numbers of instrumental and emergency caesarean births are plotted on a larger scale in Figure 22b. Numbers of both are relatively high early in the morning then reach a low at 07.00–08.00 before numbers rise again.

Post-term births, shown in Figure 23, show high rates of intervention, with most births being assisted vaginal births, involving the use of forceps or ventouse. These births exhibit similar time of day patterns. Along with spontaneous births, emergency caesarean section peaked in the early hours of the morning. In contrast, the relatively small numbers of elective caesarean section took place during daytime hours.

Method

Results

In England and Wales, just over 2% of all live births from 2005 to 2014 occurred at home and 99.7% of these were singleton births (Table 22).

There was considerable regional variation in the proportion of all live births occurring at home. This ranged from about 1% of births in the North East region to 4% in the South West region. All regions showed similar time trends, with a steady decline from 2008 to 2012, followed by a levelling off (Figure 24).

FIGURE 24.

Home births as a percentage of all live births by region, England and Wales: 2005–14.

There was a slight deficit in the average numbers of home births on Saturdays and Sundays compared with on weekdays (Table 23). Although very small, this was also observed in the 1970s. 100 This meant that the RRs of the average number of home births on each day of the week to the overall daily average for each year (Table 21) showed similar patterns as those for all live births, but the differences were much narrower. Although the RR for Friday births fluctuated from 0.99 to 1.05 for home births, it varied from 1.05 to 1.10 for all live births. On Saturday, the RR ranged from 0.93 to 0.99 for home births and from 0.91 to 0.92 for all live births. The difference was most marked on Sundays when the RR varied from 0.92 to 0.96 for home births and 0.85 to 0.88 for live births.

Approximately 97% of home births occurred at term (37–42 weeks of pregnancy), 2% were preterm (before 37 weeks) and under 0.5% were post-term (over 42 weeks).

Average numbers of term births varied by time of day and day of week (Figure 25) and were highest in the early hours of weekday mornings between 01.00 and 07.00.

FIGURE 25.

Average numbers of term births at home by hour of day and day of the week, England and Wales: 2005–14.

The much smaller numbers of preterm births fluctuated much more widely by time and day of week, so the times of birth have been aggregated into 3-hour groups (Figure 26). Although overall numbers of births did not vary by day of the week, and average numbers were highest in early mornings between 03.00 and 08.00, this varied by day of the week, with the most marked peak being on Sunday and Monday mornings, followed by Saturdays.

FIGURE 26.

Average numbers of preterm births at home by hour of the day and day of week, England and Wales: 2005–14.

Intrapartum stillbirths in England and Wales: 2005–9

To examine the question of how birth outcomes vary by time of day and day of the week, the data in the current study provided relatively detailed information on the causes of stillbirth, with information on the obstetric characteristics of all mothers, including those whose babies died, for a large number of births.

As stated earlier, the analysis was restricted to intrapartum stillbirth. It was pointed out by the project Study Advisory Group that the ONS classification of the causes of stillbirth does not adequately distinguish between intrapartum and antepartum stillbirth. This distinction was made possible using data from CMACE. Therefore, arrangements were made to acquire and link in these data. As described in Chapter 2, these data were available only for the years 2005 to 2009.

In England and Wales, there were 17,817 registered stillbirths that occurred from 1 January 2005 to 31 December 2009 and 15,240 of these were successfully linked to CMACE data (see Chapter 2). Of these, 1,255 (8.2%) were intrapartum stillbirths (Table 24).

Early neonatal, neonatal and infant mortality in England and Wales

Although descriptive analyses of neonatal and infant mortality were well under way by the time the project funding ended, they need further work, so we have not included them in this report.

It was intended that we would to go on to analyse mortality by fitting logistic regression models equivalent to a Poisson general linear model. These models are along the same lines as those we used for analysing births by onset of labour and mode of birth, as described earlier. They were to include the same categories of covariates and their interactions and follow the same model selection strategy. If overdispersion were apparent, we would use negative binomial models with a logarithmic link. These models would have produced RRs for stillbirth rates and for neonatal, infant and postneonatal mortality. We intended to exclude deaths attributable to congenital anomalies as they are not immediately preventable by care during pregnancy and birth. For other mortality outcomes, time and date of birth might operate as risk factors whose effects might potentially be mitigated.

To model survival up to 1 and 2 years of age, as well as maternal and infant morbidity, we would fit Cox proportional hazard regression models using the same framework to define the linear predictors. The assumptions of proportionality of hazards were to be tested using methods based on extensions of the Cox models. 152

Other planned outcome measures

Time did not allow us to analyse any of the other outcome measures that we had identified. We did, however, do some groundwork, following research elsewhere about the extent to which the morbidity information needed to construct both positive and negative indicators was captured in HES.

A major analysis by the NPEU153 compared maternal morbidity reported in HES with the types of morbidity that had been the subject of studies by the UK Obstetric Surveillance System (UKOSS). 154 This showed that although the quality and completeness of Maternity HES improved during the time period of the analysis, from April 2003 to March 2013, there were discrepancies in the data from HES. We planned to take this work into account in our analyses. As partners in Euro-Peristat, we also followed the progress of the work of the EUROpean NETwork on Severe Acute Maternal Morbidity, which focuses on using routine data to report severe maternal morbidity. 155

Members of our Study Advisory Group suggested neonatal follow-up studies, which had developed indicators that we could consider using. We investigated these and found that many were based on primary data collection and, were thus based on data that are not recorded in HES.

Nevertheless, there is considerable scope for using the data we have linked to derive both short-term and long-term positive indicators of the outcome of indicators that we have identified and to undertake the outstanding analyses set out in Appendix 3. We shall be applying for funding to do this.

Patient and public involvement in the project team

Miranda Dodwell and Mary Newburn have worked as co-investigators throughout the project, attending project team meetings, at which they have helped to shape the analyses, contributed to the interpretation of findings and been involved in the writing of project reports and papers for publications. In addition, Miranda has used her knowledge of maternity data gained from her maternity user work with BirthChoiceUK to contribute to the data preparation, working in the VML as an approved researcher on the location of maternity units. Although Rod Gibson, another member of the project team, was not engaged specifically for PPI, his background with data analysis for BirthChoiceUK has also meant that his contribution has a PPI element to it.

Miranda Dodwell and Mary Newburn also organised and, along with Alison Macfarlane, participated in engagement and involvement activities.

Patient and public involvement in the Study Advisory Group

A Study Advisory Group was formed for the project (see Appendix 5). Its membership included service users recruited from four different areas of England: Liverpool, Coventry, West Middlesex and Surrey. These women had different interests and backgrounds. They all had experience of involvement in maternity services planning and service user feedback in their local areas, through membership of their MSLC. One is a leading member of the Association for Improvements in the Maternity Services (AIMS); one was invited to join the Study Advisory Group as a parent representative; one who, in her employment, holds a senior position in public health, was invited to join as the public health representative; and one who has specific experience relating to breastfeeding, was particularly valued for her knowledge of infant feeding issues and how these may be related to mode of birth, time of birth and staffing.

Involvement meetings

Together with NCT staff, Miranda Dodwell and Mary Newburn organised three meetings with service users and third-sector groups. A further meeting was organised in conjunction with the Mother and Child Research group at City, University of London (@City_CMCHR). These were all used to involve women in the planned analysis and the presentation and reporting of the emerging findings.

Social media

Real-time information about the first involvement day was tweeted and retweeted by participants including Mary Newburn, BirthChoiceUK and other participants who between them have more than 10,000 Twitter followers and so have considerable reach among the maternity community. Further activity has been generated around subsequent meetings.

We have used the short, accessible title ‘Birth Timing’, and the hashtag #BirthTiming when communicating to service users about the study, and #MSLC, which draws in service users on these multidisciplinary forums. In England, MSLCs are being replaced by Maternity Voices Partnerships (MVPs). Now, the hashtag #MVP is therefore used alongside or instead of #MSLC.

Both NCT and BirthChoiceUK will continue to use their social media presence on Facebook and Twitter to alert their followers to summaries of research on the BirthChoiceUK website, papers in peer-reviewed journals and NIHR published reports.

Patient and public involvement in outputs and dissemination

Both Mary Newburn and Miranda Dodwell have been actively involved in the writing up of the paper describing patterns of births, something that is reflected in the authorship. They have contributed to two additional articles on linkage methods and data cleaning, making detailed comments on the draft. The time of birth analyses were presented to service users at the fourth involvement day and points raised have contributed to the discussion section of the paper. There have also been opportunities to contribute to the creation of research posters for the International Population Data Linkage Conference (Swansea, 24–26 August 2016) and the RCM annual conference (Harrogate, 19–20 October 2016) and the 12th International Normal Labour and Birth Research Conference, (Grange over Sands, 2–4 October 2017).

Feedback from service users will inform our dissemination in terms of messages and target audience. Learning from the patterns of planned home birth, during which women are not affected by an institutional environment, and births in all settings studied, which peak during night-time hours, service users wanted commissioners and professionals to be made aware of the impact of institutional care, and to consider critically issues such levels of lighting, privacy and intimacy. They felt that it was very important for research studies to differentiate by place of labour and birth. In terms of messages for the public, service users wanted positive messages to be the focus of the story, rather than risks and harms, which could undermine confidence and self-efficacy. Concerned that media reporting cannot be relied on to convey positive, or even balanced and accurate messages, they urged that dissemination:

• should tell ‘the story’ to women directly (so no gatekeeping effect); for example, via YouTube (YouTube, LLC, for example, San Bruno, CA, USA), Facebook, blogs and MSLC networks

• should provide bite-sized messages for women (i.e. give very clear conclusions).

Planned dissemination beyond the lifetime of the project

Now that the results have been published, BirthChoiceUK will provide a summary of the research and links to the research outputs on its website. NCT will disseminate the findings to parent representatives and to ‘patient’ advocates serving on MSLCs, MVPs160 and labour ward forums, as well as publishing in magazines and journals for parents and NCT practitioners.

BirthChoiceUK (@BirthChoiceUK), Miranda Dodwell (@mirandadodwell1) and Mary Newburn (@MaryNewburn1) have a substantial lay following on Twitter. A posting will also be made on the Facebook discussion of National Maternity Voices, and London MVP Network.

The NCT will disseminate the findings to parent representatives and ‘patient’ advocates serving on MVPs and labour ward forums. This will have the effect of highlighting research findings and service users’ views and priorities to clinical managers and commissioners.

The NCT plans to post results on its parent-facing website and on ‘babble’, its intranet for NCT practitioners and other workers. An article will be published in Perspective, NCT’s continuing professional development journal for NCT practitioners (antenatal teachers, breastfeeding counsellors and postnatal leaders). Despite there being no additional funding, it may be possible to arrange for a ‘Birth Timing’ speaker to present results and facilitate discussion at a further involvement meeting in 2018, convened by NCT for service user members of MVPs and labour ward forums. Now that the results have been published, a telephone journal club discussion is being arranged. 159 NCT may also be able to incorporate the findings into conference presentations and talks to policy-makers and professionals, such as the All Party Parliamentary Group on Maternity, which was set up with the support of NCT, and continues to be serviced by NCT. All dissemination activities will be enhanced and reach a wider audience with the use of social media such as Facebook and Twitter.

Planned activities that have not been possible

It was not possible to complete all of the planned analyses and writing up, so opportunities for dissemination to the public have been more limited. There was no further funding for work to continue beyond the end of June 2017. As a result, some planned activities were not possible, such as disseminating a short summary of findings via NCT local branch newsletters.

Unplanned dissemination activity

The NCT organised a meeting on 17 October 2017 with the University of Birmingham entitled ‘Monitoring babies during labour: What are the issues for pregnant women and their health professionals?’, a dissemination event for the INtelligent Fetal AssessmeNT Monitoring (INFANT) randomised controlled trial. 161 Two posters containing project results and lay summaries of key findings of the project were shared at the event and attracted considerable interest.

Preterm births

Unlike the earlier analyses, we had data available to analyse preterm, term and post-term births separately. We found that, unlike those births at term, numbers of spontaneous preterm births after spontaneous onset of labour did not vary by time of day and were not lower at weekends and on public holidays. A similar result was found in a study of 2,005,096 births from 1 January 1960 to 30 September 1994 in Denmark. Lower than usual numbers of term births occurred on Christmas Eve, when the most important family gathering takes place in the evening, and the following 3 days, but there was no reduction in numbers of preterm births. 167

Data linkage for England

Linked birth registration and notification records from ONS were linked to HES delivery and birth records by the data linkage team at the HSCIC. Although this team has extensive experience of linking external datasets to HES, this project offered new challenges. We used a linkage algorithm, which was similar to that routinely used by the HSCIC, to link ONS death records to HES records, but there were issues with the quality of linkage. In the period 2005–14, 4–6% of the ONS birth records for each year were incorrectly linked to the HES delivery record and 4–7% of HES delivery records were linked to more than one ONS birth record.

The number of birth registration and notification records linked to the HES delivery records using the NHS number increased over the years from 2005 to 2014. This was not surprising, as completeness of the mother’s NHS number in the registration and notification linked records increased over time. In 2005, the mother’s NHS number was present in more than two-thirds of the records, and this increased to over 90% in 2014.

In addition, the mother’s NHS number was missing from a small proportion of HES records. A further one-quarter of the registration and notification linked records in 2005 were linked using exact date of birth, sex and postcode and this decreased to 3% in 2014. There were concerns about using the postcode in the linkage algorithm for linking data for earlier years, as the HES index may not hold all historical postcodes of residence of the mother and the postcode on registration and notification linked data was recorded at the time of registration. It is possible that the mother could have moved since having the baby and this variable is also subject to recording and reporting errors.

Overall, a linkage rate of over 90% was achieved and it improved over time, especially in 2014, when there had been a shorter time between birth and the data linkage being carried out, meaning that the [HESID] encrypted identifier derived from patient identifiers, including postcodes, would have been less likely to have changed. This suggests that [HESID] at birth should be retained as a separate field for linkage.

Although ONS birth registration data have remained of consistently high quality, there have been issues with data quality and completeness in Maternity HES. The numbers of births and deliveries in London are under-represented in Maternity HES, which could be attributable to under-reporting, or complete lack of reporting, of births by several hospitals. 171 In addition, HES currently captures few home births and none occurring in private hospitals, even though data about all births should be submitted to Maternity HES.

Quality assurance of linkage

The aim of the quality assurance was to prepare the linked dataset for analysis but the results also shed light on the quality of the data and the linkage itself. This makes it informative for other users of similarly linked data and for HSCIC staff doing the linkage.

The method adopted compares common baby data items, giving most importance to matching babies’ dates of birth and locations of birth. Altogether, 98% of births originally linked to HES were left with one preserved link. This meant that 95% of all singleton births and 93% of all multiple births in England in the period of 2005–14 were linked to one correct HES delivery record after quality assurance.

Administrative duplicate HES records were found to be the main reason for discarding links. Other reasons were other births to the same mother within the study period, the presence of multiple HES records per delivery for various reasons, and error in the linkage and in the quality assurance method itself.

Certain features were more common in incorrectly linked records than in correctly linked records. Compared with correctly linked records, incorrectly linked records were more likely to have been linked using a less reliable stage of the linkage algorithm or to relate to births occurring at home or to residents of the London region or to births for which the birthweight or gestational age was missing from HES records. This suggests some bias in the linkage and often relates to data quality issues. These patterns differed between singleton and multiple births. Missed links were also discovered, which together with the previous findings, provide additional evidence on the quality of the linkage.

The linkage error identified by quality assuring the linked dataset suggests that linkage carried out using similar algorithms and by ‘trusted third parties’ generally cannot be assumed to be error-free and may affect any analyses carried out on them. The ‘trusted third party’ model replaces identifiers with an encrypted code, and the de-identified data only are made available to researchers in a secure environment. This has implications for other research projects that are not informed of the quality of the linkage, or do not have the means to quality assure the data themselves using identifiers. In particular, no assessment of the quality of the linkage algorithm is routinely published by HSCIC. We were able to detect this through using the VML to access personally identifiable data.

Data linkage for Wales

The linkage rate was approximately 95%, which was similar to the rate achieved in linking birth registration and notification linked records to Maternity HES. However, the quality of linkage to the NCCHD and PEDW was much better than Maternity HES and the data in all the fields of interest in the NCCHD were much more complete than in the corresponding fields in Maternity HES. The PEDW data are equivalent to Maternity HES but their completeness could not be assessed and compared for the reasons stated earlier, because the key data items were not obtained from the PEDW.

Timing of birth

In England and Wales, over the years 2005–14, numbers of births varied by day of the week. Numbers were highest on weekdays and increased through the week, with the highest numbers of births on Fridays. Numbers were lower at weekends, reaching a minimum on Sundays, and low on public holidays, with the lowest numbers on Christmas Day and Boxing Day. This is in line with patterns established from the 1970s onwards and did not change over the period studied.

Numbers of births varied by time of day. On weekdays, numbers peaked sharply from 09.00 to 11.00, with much smaller peaks in the afternoon. Every day, numbers of births then reached much broader maxima at night and were highest in the early hours of the morning. This is the first time that births in England and Wales have been analysed by time of day at a national level.

The overall patterns were dominated by births in NHS hospitals with OUs, in which the majority of births took place. Numbers of births in FMUs and at home varied much less by day of the week, but were still slightly lower at weekends than on weekdays. There was no morning peak in numbers of births in these settings, which showed a sinusoidal pattern, with numbers of births per hour being highest at night, reaching a maximum in the early hours of the morning.

Singleton births at term in England were analysed by method of onset and mode of birth. Just over half of these births were spontaneous after the spontaneous onset of labour and they showed similar patterns to births in FMUs, with average numbers of births per hour being highest in the early hours of the morning. Just over one-fifth of births followed induced labours and their numbers reached were highest in the hours around midnight, regardless of the mode of birth. Nearly 12% of births were by planned caesarean section and were concentrated between 09.00 and 12.00. Analyses of term births in Wales by mode of birth showed similar patterns.

It was planned to analyse the outcome of births in terms of mortality and morbidity in mothers and babies by time of day and day of the week, but the project was severely delayed by problems in accessing data and by subsequent IT infrastructure problems in handling large files of data for more than 7 million births in the 10-year period. Although patterns of birth have been described, more rigorous statistical analysis of some topics is needed. Analyses of the outcome of pregnancy had only just started when the funding period ended and analyses of differences between trusts and their maternity units had yet to be done.

Data linkage

To prepare the data for analysis, we have successfully acquired and linked birth registration and notification data for more than 7 million births in England and Wales for a 10-year period (2005–14). We have linked these to Maternity HES records for births in England. For births in Wales, we have linked them to PEDW records, which are linked to the NCCHD records. For births in both countries, we have also linked in records of subsequent admissions of mothers and babies. This work has been done in the secure setting of the ONS VML to ensure the confidentiality of personally identifiable data.

To hold the hospital data for England and the registration and notification data for England and Wales and organise them for analysis, we have built an innovative relational database. Outside this, we have organised the hospital data for Wales for analysis. These data can now also be used for other analyses of births and their outcome.

Although the linkage was successful, the overall quality was dependent on the quality and completeness of Maternity HES data. This improved considerably up to 2010, but then declined at the end of the study period and subsequently. Although the new, more wide-ranging, Maternity Services DataSet174 is being implemented, it is still very incomplete and it would be short sighted to abandon Maternity HES unless the new data for England are complete.

We have evaluated the linkage and found it to be of good quality. The quality assurance also identified a technical problem with the ‘trusted third party’ model, which is planned to be used increasingly under the Digital Economy Act 2017. 178 This replaces identifiers with an encrypted code, and the deidentified data only are made available to researchers in a secure environment. Quality assurance of the linked dataset identified some linkage errors. This was possible because of our ability to access personal identifiers in the secure environment of the VML. This suggests that linkage carried out using similar algorithms and by ‘trusted third parties’ generally cannot be assumed to be error-free and may affect any analyses carried out on the data.

The VML is an invaluable facility and the ONS has more than 10 years’ experience of running it. The disadvantage is that even after an upgrade, its infrastructure is limited in relation to the increasing number of users. The problems we encountered may, to some extent, have been a consequence of the size and complexity of our project but work of this sort is likely to increase in the future, given plans to increase the extent of data linkage under the Digital Economy Act 2017. 178

The relational database and the linked data for Wales are valuable assets for future research both by our team and by other researchers. To ensure timeliness, resources and mechanisms are needed to routinely update the linked data on a regular basis, rather than relying on one-off applications for data access, which may be subject to delays.

Data for subsequent years are likely to be affected by ‘opt-outs’ by people who are unwilling for their data to be used for purposes other than their health care. As these opt-outs tend to be clustered in particular areas, withdrawal of their data is likely to lead to bias. 179

The wider context of health data for England

The problems we have encountered should be set in the wider context of the way in which the Health and Social Care Act 20129 has led to a fragmentation of public health data analysis in England. A major consequence has been the barriers to data access, not only for academic researchers but also for public health analysts who were moved out of the health service to local authorities and Public Health England.

Although the Health and Social Care Act 20129 designated the HSCIC as the main authority responsible for health information, in practice there has been fragmentation and duplication. For example, the HQIP is responsible for commissioning clinical audits, and many other organisations have responsibilities for the collection and analysis of data, including NHS England, the Department of Health and Social Care, NHS Improvement, Public Health England and the Care Quality Commission, whereas the ONS analyses birth and death data. It is very apparent that England has fallen well behind Wales and Scotland, and behind other countries internationally, in its linkage and analysis of data.

Research using the linked data in the relational database

Questions about the timing of birth and its outcome are as relevant now as they were 5 years ago when the project was first designed. Although an analysis of perinatal mortality by day of the week was published in 2015,62 this was widely criticised and the responses to it raised questions for further investigation. We shall, therefore, be seeking additional funding to finish our proposed analyses of the outcome of pregnancy in terms of mortality and morbidity and of the variations between maternity units.

Further questions have arisen in the course of the project. In particular, the Saving Babies’ Lives safer baby care package is based on four inter-related approaches, the second of which is likely to increase the already rising proportion of births in which labour is induced. This may have implications for the time of day and day of week on which births occur. We recommend that further research is carried out to monitor changes in induction rates in relation to changing maternity policy such as the Saving Babies’ Lives Care Bundle.

The database we have created would be useful to explore this question. More generally, it is a resource that can be used for other purposes and by other researchers. One project, TIGAR,124 has been funded by the MRC to use the database and is under way. Alison Macfarlane and Nirupa Dattani are among the co-investigators. Another researcher has now submitted a funding application that includes further linkage to housing data from the 2011 Census180 to investigate children’s respiratory problems and analyses of associations with air pollution.

So far, our analyses of timing of birth in relation to onset of labour and mode of delivery have focused on singleton births. As we have found that the data on multiple births are of much better quality than we had been led to believe, and there are major gaps in knowledge about the outcome of multiple pregnancy, there is scope for analyses of data about children from multiple births, both in relation to the timing and mode of birth and more generally.

Development of the linkage

To maintain timeliness, the database should be updated regularly, with Data Sharing Agreements that allow for the routine provision of data rather than one-off operations, which are vulnerable to delays. Strategies are also needed to deal with the consequences of opt-outs from use of data for research and planning as this will lead to bias. In the longer term, changes to give potential objectors the option to distinguish between research and other uses of their data would be useful, given the considerable extent of public support for health research.

There is also scope to build on our linkage by further linkage to other data. A strength of our linked dataset is that it starts from birth registration, which is statutory. It contains demographic data that are more robust than those in clinical datasets, which have other priorities. When the new datasets being implemented in England and in Wales are more complete, it would be beneficial to link them in as well in order to link a wider range of clinical data to complete population-based demographic information.

Research resources

Our ability to do the work described here has relied on access to the VML. After its upgrade in the summer of 2015, the numbers of users increased and its infrastructure became overloaded, leading to delays in our project, despite the measures that the ONS VML team took to help us. The size and complexity of our project gave rise to particular challenges, but we thought that it would not be surprising if there were other users with projects on a similar scale in the future. This is an important resource run by an experienced and supportive team of staff, so we considered that further enhancement and development of its IT infrastructure and geographical accessibility would be a worthwhile investment.

Since our project ended, and in response to feedback from users, ONS is improving the connectivity and capacity of its secure research service (SRS), formerly known as the VML. These developments enable accredited researchers to securely access the SRS from organisations that can demonstrate they meet the security requirements for trusted access. In addition, the new technology refresh ensures the service can accommodate multiple simultaneous complex analytical research projects without any loss of performance to other researchers.

Articles

Macfarlane A, Dodwell M, Newburn M, Gibson R, McMullen S. Re: Association Between Day of Delivery and Obstetric Outcomes: Observational Study. BMJ Rapid Response 25 November 2015. 351: www.bmj.com/content/351/bmj.h5774/rr-7.

Dattani N, Macfarlane A. Linkage of Maternity Hospital Episode Statistics data to birth registration and NHS Numbers for births in England 2005–2014: methods. A population-based birth cohort study. BMJ Open 2018;8:e017897. https://doi.org/10.1136/bmjopen-2017-017897

Harper G. Linkage of Maternity Hospital Episode Statistics data to birth registration and notification records for births in England 2005–2014: quality assurance of linkage of routine data for singleton and multiple births. BMJ Open Mar 2018;8:e017898. https://doi.org/10.1136/bmjopen-2017-017898

Martin P, Cortina-Borja M, Newburn M, Harper G, Gibson R, Dodwell M, et al. Timing of singleton births by onset of labour and mode of birth in NHS maternity units in England, 2005–2014: a study of linked birth registration, birth notification, and hospital episode data. PLOS ONE 2018;13:e0198183. https://doi.org/10.1371/journal.pone.0198183

Posters

How We Use and Protect Information from the NHS About Births. Poster sent to every maternity unit. Sent on 28 July 2014.

Dattani N, Macfarlane A. Analysis of births at home in England and Wales by time of day and day of the week using linked administrative data. Poster presented at Farr Institute International Conference, University of St Andrews, St Andrews, 26–28 August 2015.

Macfarlane A, Dattani N. Using data linkage to build a national database of data about birth and its outcome for policy analysis in England and Wales. Poster presented at Farr Institute International Conference, University of St Andrews, St Andrews, 26–28 August 2015.

Harper G, Macfarlane A, Dattani N, Dodwell M, Newburn M, Gibson R, et al. Analysing the Daily, Weekly and Yearly Cycles of Births and their Implications for the NHS Using Linked Data. Poster presented at the International Population Data Linkage Conference, Swansea, 24–26 August 2016.

Macfarlane A, Dattani N, Harper G, Cortina-Borja M, Martin P, Gibson R, Dodwell M, Newburn M. Births and their Outcome: Analysing the Daily, Weekly and Yearly Cycles and their Implications for Midwifery Care. Poster presented at the Royal College of Midwives Annual Conference, Harrogate, 19–20 October 2016.

Macfarlane A, Martin P, Cortina-Borja M, Newburn M, Harper G, Gibson R, et al. ‘A Time to be Born?’. Poster presented at the 12th International Normal Labour and Birth Research Conference, Grange Over Sands, 2–4 October 2017 (winner of the Scientific Review Team Prize).

Macfarlane A, Martin P, Cortina-Borja M, Newburn M, Harper G, Gibson R, et al. Births: the Weekly Cycle. Poster presented at the 12th International Normal Labour and Birth Research Conference, Grange Over Sands 2–4 October 2017.

Conferences and presentations

Dattani N, Macfarlane A, Datta-Nemdharry P. Using Data Linkage to Create a National Database of Birth and Maternity Data. Paper presented at the Royal Statistical Society Conference, Sheffield, September 2014.

Macfarlane AJ. ‘Care.data’: Bungled Opportunity or Unjustified Intrusion? Invited presentation to the Royal Statistical Society meeting, London, 7 October 2014.

Macfarlane A, Dattani N. Using Linkage of Administrative Data to Build a National Database for Analysis of Birth and its Outcome in England and Wales. Presentation at the Administrative Data Research Network Conference, Edinburgh, 1–2 June 2017.

Harper G. Quality Assuring Linked Birth Registration and Hospital Episode Statistics Delivery Records – Results and Lessons Learned. Presentation at the Administrative Data Research Network Conference, Edinburgh, 1–2 June 2017.

Harper G. Insights on Third Party Linkage and Data Quality from Quality Assuring Linked Birth Registrations and HES Delivery Records in England 2005 to 2014. Presentation at the Royal Statistical Society International Conference, University of Strathclyde, Glasgow, 4–7 September 2017.

Macfarlane A, Martin P, Cortina-Borja M, Dattani N, Harper G, Newburn M. Gibson R. ‘A Time to be Born’? Presentation at the Royal Statistical Society International Conference, University of Strathclyde, Glasgow, 4–7 September 2017.

Martin P, Cortina-Borja M, Dattani N, Harper G, Gibson R, Newburn M, et al. Variations in Numbers of Births by Day of the Week in Relation to Onset of Labour and Mode of Giving Birth, England 2005–2014. Presentation at the Royal Statistical Society International Conference, University of Strathclyde, Glasgow, 4–7 September 2017.

Linkage of birth registration data to birth notification data

As described in Chapters 1 and 2, the data linkage aspects of the project built on two preceding projects. The first of these, which linked birth registration data to birth notification data held in the interim NN4B system, made data about the timing of birth, gestational age and other key variables available at a national level. The linkage was done within ONS, where Nirupa Dattani was head of Child Health Statistics. She was able to make arrangements for a researcher employed and funded by City, University of London, to design the project, develop the linkage methods and assess the quality of the birth notification data. 5,6 The data were used in a number of analyses both at the time and subsequently. 122,182–187

The ONS mainstreamed the linkage and used the linked data in its publications. The collaboratively written articles about the linkage contributed to the success of City, University of London, the Nursing and Midwifery Unit of Assessment of the 2008 Research Assessment Exercise.

From the start, the project was designed to include further linkage to the Maternity HES for England and maternity data from the NCCHD/PEDW for Wales. We did not have the funds needed to do this immediately but we included the full plans in our applications for ethics and other approvals. In 2005, we obtained ethics approval for this (reference number 05/Q0603/108) from East London and the City Local Research Ethics Committee and approval from the Patient Information Advisory Group under Section 60 of the Health and Social Care Act 2001 (PIAG 2–10(g) 2005). To prepare for applications for further funding, permission was also obtained from the Information Centre’s Security and Confidentiality Advisory Group to link birth registration and notification data to Maternity HES.

The Information Centre, as it was originally called, was being established at the time this project was being set up. The Department of Health and Social Care statistician responsible for analyses of Maternity HES transferred to the new organisation and continued to take an active interest in the project, as did his counterparts in Wales.

Linkage to data about maternity care in England and Wales

The opportunity for a second project to do the next stage of the linkage came in July 2008, when funding was obtained from the MRC through a successful application to the Joint Wellcome Trust/MRC Electronic Patient Data Linkage Initiative. By this time, Nirupa Dattani had left the ONS as her post had been relocated to Newport. This meant that the work could no longer be done within the ONS and permission was requested to use the ONS secure setting, the VML.

Before starting work on the second project, all the approvals obtained previously were renewed. This included successful applications for a substantial amendment to the ethics approval and for further permission to process patient-identifiable data without consent, now under Regulation 5 of the Health Service (Control of Patient Information) Regulation 2002. 188 ONS Approved Researcher Status was obtained for the project and for all the staff who would be accessing data in the VML. Approval was also obtained from the Caldecott Guardian for Wales.

The former Department of Health and Social Care statistician responsible for Maternity HES had left the Information Centre when his post was relocated to Leeds and, for many years, no externally identifiable member of staff had long-term responsibility specifically for Maternity HES. Although a senior member of the Information Centre staff joined the project as co-applicant, this was on the understanding that he would not be able to devote much time to the project. He did, however, give key help with data access problems. In contrast to the situation in England, we continued to have longstanding named contacts in Wales.

The application for HES data was submitted to the Information Centre’s Data Management Subgroup, which had superseded the Security and Confidentiality Advisory Group. The new body confirmed that permission to link civil registration and registration data to Maternity HES was still valid; therefore the path appeared clear to appoint a researcher, who started in July 2008.

Unfortunately, however, when a meeting was held with Northgate Solutions to which HES data processing had been outsourced, it transpired that a new committee, the Data Access Advisory Group (DAAG), had superseded the Data Management Subgroup. It was necessary to reapply for access to HES data, a process that delayed the project by 4 months and increased the costs of accessing the data. Although we were able to successfully complete and document the data linkages,7,8,123 losing 4 months of the 2-year funding period severely restricted our ability to make use of the linked data for analysis as planned.

These first two projects showed that it is feasible to link data from several sources to derive an enhanced database that contains a much wider range of data than any of the original datasets individually. 6–8,123,187 The second project also highlighted the danger that valuable project time could be wasted by the delays involved in applications for access to data from Maternity HES. We tried to learn from both the positive and negative experiences when designing the project that forms the subject of this report.

Application to the National Institute for Health Research for funding

The successive stages of our application to NIHR took place over the period from December 2012 to September 2013, which was a time of considerable organisational change. As a result of the implementation of the Health and Social Care Act 2012, the Information Centre for Health and Social Care, as it was then known, was taken over by Connecting for Health, which had previously been responsible for NHS IT infrastructure, to form a new organisation, the HSCIC. By and large, most of the senior management of the new organisation came from an IT infrastructure background and key people were unfamiliar with data access issues.

The National Information Governance Board was abolished and the staff responsible for overseeing ‘Section 251 approvals’ were transferred to the CAG of another new organisation, the Health Research Authority. The administration of Research Ethics Committees was centralised. The longstanding administrative staff of our local Research Ethics Committees, based in Mile End Hospital, were replaced by new staff based in Bristol. In view of all these changes, we made sure that we maintained contact with the key staff, including data providers, and discussed our proposal with them at each stage of the application process in an attempt to avoid the problems we had experienced previously. This included attending the Data Linkage Stakeholder Forum which the HSCIC held in October 2013 to encourage the use of data linkage and which gave an opportunity for informal discussions about our project.

Approvals

Our project funding started in January 2014 and NIHR encouraged us not to delay the start. Before the project started, we applied for a substantial amendment to our ethics approval and a second approval (CAG 9–08(b)/2014) under Regulation 5 of the Health Service (Control of Patient Information) Regulations 2002 to link in data from the HES for England and the PEDW about subsequent admissions of mothers and babies without patient consent. This second application for ‘Section 251’ approval took the form of an application to form a database to be held in the secure environment of the VML at the ONS along with the previously linked data. We obtained approvals for the amendment to our ethics approval and the new ‘Section 251’ application during the first month, although the latter applied only to England. We were told that this was pending the completion of new arrangements being made for accessing data for Wales, which had been disrupted by the implementation of the Health and Social Care Act.

On the advice of our Study Advisory Group, we applied to the HQIP for access to data about stillbirths and neonatal deaths from the CMACE, held by the NPEU, which had taken over responsibility for the confidential enquiries into stillbirths and neonatal deaths from CMACE. As explained in Chapter 3, this was to obtain more reliable data about intrapartum deaths than was available from ONS cause of death coding procedures. Successful applications were made to amend our ethics and ‘Section 251’ approvals to support this.

Owing to an oversight, we had not included data on deaths of children aged over one year and deaths of mothers within 1 year of giving birth in our earlier applications. These data are routinely linked by ONS to the corresponding birth records and are thus readily available, subject to permission. We therefore applied for amendments to our ethics and ‘Section 251’ approvals and ONS Microdata Release Panel for access to these data. In addition, we applied for ONS Approved Researcher Status for the project and all members of the project team.

All these applications went smoothly in a timely manner. During the course of the project, we renewed these approvals and also our first ‘Section 251’ approval annually. The barrier that delayed the project was gaining access to data from HES and, to a lesser extent, to NHS data for Wales.

Application for Hospital Episode Statistics data for England

We arranged to meet the staff from the HSCIC with whom we had discussed the project during the application process in late March 2014, to finalise arrangements and our application to the DAAG for HES data.

On 17 March 2014, the HSCIC staff contacted us to cancel the meeting. They told us that this was because of a review and suggested that we contact them 2 months later, at the end of May, to rearrange the meeting. The next day, three members of the project team attended the NIHR welcome meeting at which NIHR’s concern that projects are completed on time and within budget was stressed. We told our programme manager about the latest developments, pointed out that they were entirely outside our control and shared our concerns about the delay.

We arranged a meeting with the HSCIC manager in charge of the data applications and he offered to try to arrange access to data. He made an arrangement to telephone the principal investigator on 9 April 2014, but this call did not take place.

It became clear that other circumstances were affecting our communication with HSCIC. In previous months, NHS England had been heavily criticised in the media and more widely for the approaches it was using to inform the public about its plans to use data from general practice and link them to hospital data in its new care.data system. High-profile press reports about data leaks and the sale of personally identifiable data to private companies undermined public confidence about the use of NHS data. This led to an inquiry by the House of Commons Health Committee into the handling of NHS patient data. Kingsley Manning, Chairperson of the HSCIC, and Max Jones, Director of Information Services, were invited to appear at a hearing on 8 April 2014,189 the day before the promised telephone call.

After this, we had no further contact from the HSCIC. No explanations were given either proactively or in response to e-mail enquiries. We therefore resorted to the HSCIC’s formal complaints procedure, and made a complaint on the grounds that it was advertising services on its website and elsewhere but not providing them. There was no further news until 9 May 2014 when a letter from the Director of Data and Information Services was posted on HSCIC’s website, announcing the review of processes and governance for the sharing of data releases. In response to our complaint, we subsequently received a letter from him dated 30 May 2014 saying that the review was necessary to improve the process of application for data.

The HSCIC’s moratorium was still under way in early June 2014, so we were still unable to arrange a meeting to discuss our application. We therefore decided to submit our application without meeting HSCIC to discuss it.

At the time HSCIC stopped processing applications, in April 2014, it had already built up a backlog of applications and this continued to grow during the hiatus that followed. It then set out to implement a completely new ‘one size fits all’ application system to process the wide range of applications for data it receives.

Once the moratorium was over, we had a number of contacts with HSCIC staff, some of whom appeared to be working in parallel with each other and giving us conflicting messages on the progress of our application. Most told us that the sender had checked that our application was in the queue but was not in a position to tell us when it would be processed. This lack of information made it impossible to plan work or to keep NIHR informed about likely progress with the project. To add to our frustration, we, and our university’s IT department, received repeated requests to delete data that HSCIC’s records said we had received in 2007. We presumed that this was a consequence of our application to the Security and Confidentiality Advisory Group, which had been approved in 2007, as described above. We were unable to comply with requests to delete data we had never held and, in any case, had never intended to hold at the university.

Lobbying for access to data

The DAAG published its minutes on its website and at some points published data about numbers of applications awaiting consideration. It was very apparent that numbers of applications in the queue were growing, so we joined together with other affected researchers to lobby for access to data, initially through the Health Statistics User Group, which then made representations to other organisations.

As the HSCIC senior management came mainly from an IT infrastructure background, there was limited experience of how data are used in health research or public health analysis. In particular, there appeared to be a lack of awareness that research applications would have already been scrutinised by funders and their reviewers, by ethics committees and, in most cases, also by the Health Research Committee’s CAG. As a result, the checks made by these organisations were repeated, even though most applicants to DAAG included these approvals with their applications.

In contrast, the misuses of data that had caused concern in the press and in parliament had not been subjected to this level of scrutiny and were largely from the private sector. Alison Macfarlane contributed to letters sent from the Health Statistics User Group to the Chief Executive of the HSCIC and from the President of the Faculty of Public Health to the Secretary of State for Health proposing that the applicants who have obtained permissions should be subjected to a streamlined process that did not replicate approval processes already successfully completed. This did not appear to have any impact.

It also appeared from the DAAG minutes that applications were not being dealt with in chronological order. Through convening meetings with other affected researchers, we found that the Wellcome Trust, the research councils and the cancer research charities were taking active approaches to the HSCIC. We therefore asked if NIHR could approach HSCIC to make the case for all the researchers it funded, as we felt that this would have more impact than approaches from individual researchers. Our programme manager told us that NIHR staff had been told not to approach the HSCIC as the matter was being dealt with at the highest level.

Despite the continuing uncertainty, we were given the go-ahead to rearrange our technical meeting with the Data Linkage and Extract Service and a useful telephone meeting was held. We were unable to send a representative to the HSCIC’s stakeholder meeting on 21 July 2014 as we were told it was oversubscribed.

We did not receive any news about our application until mid-September when we were told that the HSCIC wanted to submit our application to the 24 or 29 September meetings of DAAG and wanted a telephone meeting to clarify some details. This took place on 17 September 2014 and we were told that a draft application document would arrive by the end of the week to be checked and amended before submission. The document did not arrive and we continued to ask about it for some weeks. We noted from the online minutes that on 24 September, DAAG considered other applications, while on 29 September it set aside all applications to devote the entire meeting to discussing the data needs of Genomics England Ltd.

Finally, in early November, Alison Macfarlane, who was visiting Leeds for other reasons on 10 November 2014, e-mailed the HSCIC to say that she would be calling at its office at 17.00 to discuss our application. Two members of HSCIC staff met her with a draft document. They discussed the details in it and e-mailed it to her straight away. This made it possible to get back to them quickly with further corrections.

The application was then considered at the DAAG meeting on 18 November 2014, but was not approved, as the committee wanted further information and had questions such as regarding the details of the Section 251 approvals. Although we had passed this information to the HSCIC, it seems that it was not shown to the committee. We provided answers to the questions very quickly and the application was resubmitted to the DAAG the following week and approved at its meeting on 25 November 2014, subject to information being checked with the CAG. Our application was finally approved on 16 December 2014.

Negotiating data-sharing agreements

Although we had now obtained permission to access the data we needed, the project still could not start until overarching data-sharing contracts at organisational level and project-specific contracts, originally known as Data Re-use Agreements and now known as Data Sharing Agreements, were signed.

In our case, the organisational level contract, known as a Data Sharing Framework Contract, was between the ONS, which is holding the data for the project in its secure facilities, and the HSCIC. It forms part of a wider service-level agreement, which was under renegotiation in early 2015. As part of its support for our project, the ONS negotiated with the HSCIC to deal with the Data Sharing Framework Contract in advance of the rest of the agreement and this went through smoothly.

Unfortunately, this was not the case with the Data Re-use Agreement, which arrived from the HSCIC with numerous drafting errors, which we corrected. This was followed by a long delay, after which the HSCIC told us that it had not understood our application and so had returned it to its committee, the DAAG. We then found that the version it agreed lacked one of two required technical appendices and we were not allowed access to data until the full document was approved.

There was a further wait until the HSCIC sent a revised version, now described as a Data Sharing Agreement, using a completely different form. As well as containing further new drafting errors, this new form did not cover our situation, in which the data are held at ONS but the work is done by the City, University of London, project team using the VML.

While we were waiting for these problems to be resolved, we were asked by our NIHR programme manager to document the problems we were experiencing. We hoped that this might lead to help in dealing with the problems, so were disappointed to receive an e-mail informing us that NIHR intended to discontinue the project unless we could arrange immediate access to the data.

We forwarded this e-mail to the HSCIC and also gave notice that we intended to make a Stage 2 complaint, following on from the Stage 1 complaint we had made early in 2014. The matter was escalated to a more senior member of HSCIC staff who quickly redrafted the contract in a way that made it possible for work to finally start in June 2015. Although this solved our immediate problems, we found, when we came to try to renew the Data Sharing Agreement 2 years later, that it had been incorrectly drafted.

Access to data for Wales

As was mentioned above, the Health and Social Care Act 2012 disrupted aspects of data flow between England and Wales. This meant that when our second Section 251 approval was granted in early 2014, it did not apply to Wales. When we queried this with our contacts at the NWIS, we were told that the problem was being discussed at a very senior level and that primary legislation in the Welsh Assembly might be needed to resolve it.

We explored the possibility of accessing data for Wales via the Secure Anonymised Information Linkage (SAIL) Databank, based at the University of Swansea. It transpired that the SAIL anonymisation process includes removing all dates, so data from it would not enable us to do our current project for which dates were a key data item.

Fortunately, pending primary legislation, the NWIS made an application for a ‘Section 251’ application to share data outside Wales. This was considered by the CAG in November 2014. Eventually it was decided that the two countries could sign a Memorandum of Understanding to cover the situation until legislation was enacted. Early in 2015, we were given to understand that linkage could proceed and in February 2015 we met the NWIS to take this forward and enable linkage with Welsh data.

Recruitment of staff

Our initial plan, set out in our application to NIHR, was to use the data from our previous projects for further analysis while obtaining and linking data for more recent years. At the start of the project, we therefore started to recruit the project researcher and part-time administrator. We advertised both posts in February 2014 and received over 30 applications for each. Cecile Bremont took up the appointment of part-time project administrator in April 2014. The post of project researcher was offered to an excellent applicant but she subsequently withdrew as she was offered a post elsewhere. By this time, the HSCIC moratorium was under way, and we were told by NIHR not to offer the post to another applicant at this stage. Although this was understandable, not having a researcher in post severely restricted our capacity to undertake the analyses we had identified that could be done with ONS data alone.

By the time we were in a position to start work on the data linkage in 2015, only 6 months remained on our contract, although application for a Variation to Contract was under discussion with NIHR. As recruitment processes can take at least 3 months if the successful candidate has to give notice from another post, we discussed with NIHR the possibility of employing a highly experienced applicant from the previous recruitment process, who had not been available for interview and was only available part time. As she was still available and was willing to come for 6 months in the first instance, we interviewed and appointed her, with the aim of extending her appointment and recruiting a second member of staff whose skills complemented her if our application for Variation to Contract was successful. This had the advantage that we were immediately able to appoint an experienced person, but of course the disadvantage was that being part time meant that the work she did progressed more slowly than if she had been employed full time.

Data items needed for quality assurance

We were in active contact with colleagues at ONS and with the HSCIC’s Data Linkage Service during the protracted process of gaining access to data, so that the groundwork was laid to start work on our project as soon as permissions were finally granted. The HSCIC provided the linked data for England and work started in the VML to assure the quality of the linkage. As this got under way, it became apparent that further data items that we had not requested were needed for this. The only way to obtain these was to request a re-issue of the data.

We were reluctant to take this course because of the earlier delays, and discussed this with the Head of Data Access at the HSCIC. We pointed out that changing our specification would mean agreeing a new Data Sharing Agreement and expressed our concern that this might add further delay. He commented that our work has indicated a need for additional fields in order to improve the linkage and hence data quality and went on to suggest that this in itself is an interesting finding. He added that this could have an impact on other research in this area and this is discussed in Chapter 2.

Although the revised Data Sharing Agreement did not have to go back to DAAG, this revision took longer than originally suggested because it co-incided with a rush of applications for data before the HSCIC was required to start deleting the records of ‘Type 2 objectors’. These were people who had objected to their records being used for purposes other than their health care, including secondary analysis, because of concerns about the data being used for commercial purposes. This short-term increase in numbers of applications introduced a further delay, although we were able to get on with other work while the revision was under way. The data linkage work was also done in parallel, so the HSCIC provided the revised extract to the project in March 2016, as soon as authorisation was given.

Using the Office for National Statistics’ secure setting

This project has involved linking and analysing personally identifiable data, including data items recorded at birth registration under the Population (Statistics) Acts. We were able to do this and our previous project only because we were able to access the data in the secure setting of the ONS VML. This is its facility for providing secure access to sensitive detailed data. Access is provided for approved researchers working on defined and approved non-commercial projects, which are judged to serve the public good. Data cannot be downloaded, but users can access the data at their desk if they are part of a government organisation, or otherwise in a ‘safe room’ at one of the ONS sites. In our case, going from the university to the VML access point in Pimlico involved a journey of around 30 minutes by public transport or bicycle. A member of the project team who lives in Gloucestershire drove to the ONS building in Newport to use the VML.

Before outputs can be released from the secure environment of the VML, they have to be scrutinised and cleared by VML staff before being e-mailed to the person who generated the output. Usually, this is dealt with promptly, typically on the next working day, but occasionally there have been longer delays, resulting in further delays if outputs have to be discussed with other members of the project team.

At the time this project started, the VML had been in operation for over 10 years without upgrading and there were frequent problems with its servers, creating major challenges for the team responsible for keeping it up and running. In the summer of 2015, it was upgraded, and projects were migrated to new facilities, but the upgrading increased the extent to which it was used, particularly within government.

The data for the project related to over 7 million babies and their mothers, with numerous ‘many-to-one’ relationships. As described in Chapter 2, the best way to hold such data is in a relational database. At the start of the project, the only relational database software available was Microsoft Access^®. Although this software has many useful features, it is limited in its file sizes. This meant that the data for each year had to be split into an ONS birth registration file, a HES delivery record file and a HES birth record file. The result was a large number of separate databases, which then had to be interlinked. Although this provided a workable solution, setting up the infrastructure was labour intensive. Because the VML’s facilities are shared between many users, the processing speeds were mostly very slow.

Unfortunately, this was exacerbated by a succession of problems. These were discussed at each stage with members of the ONS VML team, which made considerable effort to support our project and help to resolve them.

The first was a technical problem, which caused work to be lost up to 10 to 20 times a day, and which made it impossible to run programmes that needed to run for many hours. It is possible that this was caused by very short-term outages. Microsoft Access^® is particularly sensitive to this as it expects to maintain continuous access to data discs and is intolerant to outages of even a millisecond, which do not affect other software. This problem lasted several months, despite the best efforts of a project member and the VML team to identify the cause. It was possible to continue work on the project by dividing tasks into many short stages and logging progress in detail. Although progress was made during this period, it was painfully slow.

Shortly after the outage problem was solved, processing speeds in the VML plummeted, probably owing to an increase in the number of users rather than any specific technical issue. Logging of computational tasks showed that, on occasion, processing speeds sometimes fell to 100 times slower than normal. The problem was discussed with the VML team, which responded in July 2016 by providing us with a separate server box. Although this helped to some extent, our data remained on the same network drive and it was access to this drive that was causing the bottleneck.

In a further attempt to help us manage the very large database, the VML team made Microsoft SQL Server available to the project in June 2016. This had not been available earlier as there were a number of security issues to be resolved. Despite an initial reluctance on our part to switch technologies at this stage in the project, some experimentation made it clear that this was a fast and robust platform to use for storing and manipulating the data. The change entailed a considerable amount of work to design a new database, reimport the data and rewrite some of the code. Microsoft improved performance in terms of speed and reliability made the move to the new database platform worthwhile. Meanwhile, other members of the team had to continue their analysis work on the old server while the database was being transferred to the SQL system and they were not able to benefit immediately from the improvements.

Despite all of these problems, the project team produced a relational database of good-quality linked data, including data for England from the three main datasets (birth registration, birth notification and HES) for the 10-year period 2005–14, together with additional linkages relating to mortality and location of births. Non-delivery admissions of mothers and subsequent admissions of babies were also included, linked in by [HESID]. The database could be updated with data for more recent years using the same techniques. In future, this can be achieved much more efficiently given the additional software and processing power available in the VML as a result of the improvements prompted by this project.

Impact on progress with the project

In our progress report to NIHR in August 2016, we said that we hoped to be able to answer the main research questions by the end of the funded project, but would need to seek further funding to complete all the analyses. It soon became apparent, in view of the problems described above, that this was overoptimistic and we discussed the situation with our programme manager. This led to a meeting in Southampton on 24 October 2016, as a result of which we were invited to apply for a second Variation to Contract. This entailed a considerable amount of work and led to several requests for further detailed information. The net result of all this was being informed on 17 March 2017, 2 weeks before the funding ended, that we had not been successful and that no further funding would be available.

Comments

We have set out the above in considerable detail, as it is clear that problems of these types are not typical of those dealt with previously by NIHR programme managers who have had limited experience of what is involved. This reflects the fact that secondary analysis of routine data is a form of research that is not usually funded by NIHR and it has complex constraints that can be out of the researchers’ control.

This project started at a time when access to routine data was receiving considerable hostile media attention and at a time when the HSCIC was a relatively new and inexperienced organisation. In our attempts to gain access to data for analysis, we have been in touch with many other researchers who planned to use HSCIC data. As a result, we know that this project is among many others that have been adversely affected and that considerable sums of research money have been wasted as a consequence. As well as public funds, this has included money donated by the public to research charities. Members of the public have been invited to object to the use of their personal data for purposes other than their health care, without distinguishing between the purposes. It is clear from the level of donations to health research charities and also from surveys that health research enjoys considerable public support, although use of people’s data for commercial purposes does not.

At meetings and conferences that we have attended, it has become increasingly apparent that although Wales and Scotland are among world leaders in the innovative use of data linkage for analysis of health problems, and Northern Ireland is moving forward rapidly, England is well behind. In addition, data access is a problem, not only for research, but for public health analysis more generally. Public health analysts, who were moved out of the NHS to local authorities and Public Health England, have found themselves no longer able to access the data they need for their work. We have been involved in the widespread lobbying that has prompted the UK Statistics Authority to act on this and, in the wake of a series of meetings to identify the problems, an English Health Statistics Steering Group has been set up.

At the time we submitted this report in July 2017, we were aware that over the past 2 years, the HSCIC has taken steps to improve its application processes. An online application system has been introduced190 and a new committee, the Independent Group Advising on the Release of Data (IGARD),191 with wider representation and a lay chairperson has replaced DAAG. At the time we submitted this report in July 2017, we were not yet in a position to comment on the impact of this change, although we were about to apply to IGARD to renew our Data Sharing Agreement. When we did so, we experienced many problems.

If NIHR does want to encourage the researchers it funds to make greater use of routine data, we would suggest that it could include people with routine data experience among its staff. It would also be useful to make it possible to delay the starts of projects until access to routine data has been secured. We were under considerable pressure to start in January 2014, by which time we had ethics and ‘Section 251’ approvals, although we did not yet have permission to access HES and the data for Wales. Although we built a delay of 4 months into our project management plans, this was in line with our previous experience rather than the unexpectedly dysfunctional circumstances that we actually encountered.

We are extremely disappointed that, having got to the stage at which we had acquired the data, linked them, quality assured the main linkages, built the innovative relational database and started the analyses, we were unable to continue and complete them. As soon as this report was submitted, we started looking for sources of additional funding to do this, but a year had passed before we were able to identify a source for which we were eligible to apply and this opportunity arose as a result of changes in eligibility criteria.

Linkage bias (England only)

There was a statistically significant difference in distribution between ONS birth records that were linked to HES delivery records by the HSCIC and those that were not linked in terms of multiplicity, age of mother, ethnicity and region of residence, as shown in Table 27.

Data preparation and quality checks

Both mothers’ and babies’ identifiers were available for the ONS birth and HES delivery records to use in the quality assurance process. These were mother’s date of birth, NHS number and postcode of usual residence, and baby’s date of birth, NHS number and sex.

It was necessary to normalise the baby information in the maternity tail section of each HES delivery record so that the data could be stored and interrogated efficiently in the database environment. This meant that if information was held on more than one baby in the maternity tail part of a delivery record, this was transformed into separate records for each baby. For example, this meant that there were two records, one for each of the two babies in a twin delivery. Each record contained the same common episode identifier, APC fields and maternity tail fields, but with the respective baby’s information in each one.

For variables common to birth registration and birth notification, the maximum amount of reliable information was derived from the two sources. The baby’s date of birth, birthweight and sex were taken from birth registration, and gestational age was taken from birth notification, unless values were missing.

Inconsistencies were found between ONS records and HES records in the data types and formats used for variables such as dates and the baby’s sex. These variables were recorded into common formats for use in the linked database.

The extent of missing values for key items of data about the baby in the ONS birth records for 1 January 2005 to 31 December 2014 and HES normalised records for 1 April 2004 to 31 March 2015 is given in Table 28.

A value would have to be missing on both birth registration and birth notification to be considered missing overall on ONS birth records in Table 28. Although the total numbers may not be directly comparable as a result of definitional and date differences, the percentages of missing values per variable gives an indication of differences in data completeness.

The HES delivery data have a much higher proportion of records with missing values than ONS birth records, for example up to 37% for gestational age in singleton births. The proportion of records with missing information is lower for multiple births in HES, as data items for the first baby are often repeated for their siblings in the same delivery.

When a baby’s date of birth was missing on HES, it was derived from operation dates if a delivery code was present. If sex of baby, gestational age or birthweight were missing, they could not be derived.

Data about gestational age and birthweight are known to have problems, notably implausible values, truncation and rounding in HES. ONS birth records can also have implausible values for birthweight but this has been improving with the introduction of ‘warning’ edits to birth notification systems.

Full quality assurance procedure

Quality assurance procedures for singleton and multiple births to select only one correct linked HES delivery record for each ONS birth record and to discard incorrect linkages (detailed version) are shown in Table 29.

Quality assurance results

For a summary of quality assurance results for singleton and multiple births linked to HES delivery records by calendar year, see Table 30.

Reasons for discarding Hospital Episode Statistics records linked to an Office for National Statistics birth

Table 31 shows a breakdown of reasons for discarding HES delivery records linked to an ONS birth excluding duplicate copies of HES delivery episodes and invalid deliveries for singleton births in England by year.

An ONS birth record may have been linked to one or more HES delivery records for a variety of reasons given in Table 31. After quality assurance, some or all of the available linked HES delivery records will have been discarded, leaving a link to one correct HES delivery record or to none at all. Discarding all the linked HES delivery records for a birth suggests incorrect linkage and is discussed in the next section.

Overall 76% of all the linked HES delivery records that were discarded were duplicate copies of HES delivery episodes, and 0.5% were invalid delivery records, dealt with in cleaning stages 1 and 2 of the methodology. A summary of the remaining reasons by year for singleton births is given in Table 31.

Same mother different baby

This accounted for 60% of all discarded links. This is expected when the study period covers 10 years and each birth is linked by the mothers’ identifiers to all other deliveries to that mother in that period.

Multiple episodes – pre/post delivery admission

This accounted for 21% of discarded links. It arose where additional episodes relating to a delivery, usually relating to a pre- or post-delivery admission. The numbers of these generally decreased over time, suggesting that data quality is a factor.

Multiple episodes – part of spell

The third main reason for discarding links, accounting for 16%, was links to other episodes in a spell having been discarded, keeping only the link to the episode with the most amount of delivery information.

Other reasons

The remaining three reasons each accounted for less than 2% of discarded links. The sixth column shows when multiple delivery records contained the same information and links to superfluous records were discarded. This is also a data quality issue. There was a large increase in these in 2014. The eighth column relates to a small number of linked singleton birth records that were identified as ‘potential multiple births’, described below (see Data and linkage quality issues).

The seventh column shows ‘no matches’; links discarded when there are no matching variables or insufficient numbers of matching variables according to the quality assurance methodology, and the baby’s date of birth did not indicate a ‘same mother different baby’. This group is of interest because it could indicate erroneous false positive links by the linkage algorithm, or it could indicate erroneous false negatives in the quality assurance methodology. Fifty-seven percent (2,743) of this group were linked by match rank 1 of the linkage algorithm, the most reliable stage of the linkage algorithm that matches all available identifiers including the NHS number, suggesting that they should be correct maternal linkages. These were checked in greater detail clerically for explanations.

Of these, 2,323 (85%) were true no matches, whereby none of the five baby data items matched at all or only the place of birth matched, or only the sex matched or only the gestation matched. In these cases, the baby’s date of birth had not been transposed nor was it a different baby to the same mother. This suggested that incorrect maternal linkage or some other data quality issue has been involved. This accounted for 0.04% of all singleton birth records originally linked to a HES delivery record.

It appeared that 366 links (13%) were potential false negatives made in error by the quality assurance methodology. The dates of birth in these do not match owing to transposition in the date of birth elements. If this had been factored into the methodology, these links would have not been discarded.

The remaining 94 links (3%) were undeterminable, with some discriminatory baby data items matching and some not. For example, the date of birth may match or have been transposed, but birthweight and/or place of birth may be very different.

Some of the categories of problems shown in Table 31 may have been restricted to particular NHS trusts but further investigation would be needed to check this.

Data and linkage quality issues

Quality assurance of the linked dataset for such a large population revealed issues with the quality of the data within each of the datasets and also the quality of linkage that otherwise might not be detectable. Although it is already known that HES has data quality issues, issues were found in the ONS birth records that warranted further investigation.

First, the consistent common location code that was assigned to the place of birth codes was found not to match between birth registration and notification for 22% of the singleton births with a correct HES delivery link after quality assurance. This suggested that the linkage between birth registration and notification could be incorrect. A sample of these records that also had very discordant mother and baby identifiers and baby tail information was given to the ONS, which carried out this linkage, for checking.

The ONS confirmed that the baby NHS number or the sequence number192 matched for these cases and, therefore, that the linkage was correct between ONS birth registration and birth notification, regardless of the discordant variables. However, no quality assessment of the sequence number assignment process was available, so 542 of the most suspect cases were flagged. It was decided to exclude these from analyses.

Second, a small number of singleton birth records appeared to be multiple birth records (see Table 31) based on a combination of HES variables suggesting a multiple birth delivery, and other birth registrations existing to the same mother on the same day, yet these were not labelled as a multiple birth by the ONS. Again, these cases were checked by the ONS, which confirmed that they should be multiple births. The error was explained by a delay in some records being registered and received, possibly causing issues with finding the corresponding record in its system.

Third, after quality assurance, a small number of multiple births did not have all the expected number of ONS birth records, for example, a twin birth had only one ONS birth record associated with it rather than two. These were not always explained by stillbirths prior to 24 weeks’ gestation, which should not be registered as births. Investigation revealed that the missing ONS birth records could often be found in the ‘unlinked’ ONS birth records file, to which no HES records were linked. This could be the case even if the mothers’ identifiers, including NHS number, clearly matched.

This was discussed with our contact at the HSCIC, who investigated this. It was explained that these missed links were attributable to either the HES patient index having been updated since the linkage was done originally or linkage error. The project did not attempt to find any further missing links.

The advantage of having the identifiers available to the project team for both ONS birth and HES delivery records made it possible to investigate and explain the above issues, but other research projects may not. Overall, this highlights how researchers and analysts should not assume that any routinely collected administrative datasets and linkages carried out between them are error-free.

Differences between records with correct and incorrect linkage

Checks were carried out to ascertain if the ONS birth records that remained correctly linked to one HES delivery record after the quality assurance procedure, and the ONS birth records with all links to HES delivery records discarded after the quality assurance procedure, differed significantly by key variables. Discarding all links could indicate linkage error, data quality issues or false negatives in the quality assurance methodology.

Table 32 compares the proportions of values for the key variables for both groups, and chi-squared tests were used to test the null hypothesis that the distribution of a variable is the same in the two groups. Variables from the ONS birth records include the match rank of the linkage algorithm, the region of birth, the year of birth, the month of birth, the day of birth based on an 8-day categorisation (including public holidays), the day of birth based on an 11-day categorisation (including specific public holidays and the days before and after them), the hour of birth, the age of the mother, the sex of the baby, the ethnicity of the baby, gestational age, if the birth was a stillbirth and the location of birth. Missing gestational age and missing birthweight in HES were also included.