Models of care for the delivery of secondary fracture prevention after hip fracture: a health service cost,clinical outcomes and cost-effectiveness study within a region of England

Current guidelines in fracture prevention

The onset of osteoporosis is asymptomatic and it is often recognised only after an older person falls and sustains a fracture. There have been widespread calls to improve the identification and treatment of hip fracture patients to reduce the risk of further falls, fractures and mortality. 1,4,10 The risk of further fracture can be reduced by up to half with bone protection therapy. 1,11–13 As most fractures result from a fall, interventions to reduce the risk of falls may be effective in preventing further such events; however, direct evidence is lacking. Over the past decade, guidance from a number of professional bodies has been published for the management of hip fracture patients [British Orthopaedic Association (BOA) Blue Book,1 Scottish Intercollegiate Guidelines Network (SIGN),14 National Institute for Health and Care Excellence (NICE)13,15]. NICE technology appraisal (TA) guidelines TA 160/16113 are related to the effectiveness of bone protection therapy and clinical guideline (CG) 21 relates to falls prevention. 15 In the UK, secondary prevention of fracture is underutilised and widely neglected. 1 As a consequence, compliance with NICE publications TA 16113 and CG 2115 is low. Audits by the National Hip Fracture Database (NHFD)16,17 and the Royal College of Physicians18 suggest that the situation is improving but still inadequate, such that prior to discharge only 66% of hip fracture patients were on bone protection medication and 81% received a falls assessment. 17

As almost half of all hip fracture patients have had a prior fracture,4 responding to the first fracture provides a golden opportunity to prevent the second. The BOA Blue Book1 provided guidance on secondary prevention of fragility fractures. A comprehensive service should consist of osteoporosis assessment, including a dual-energy X-ray absorptiometry (DXA) scan to measure bone density, if appropriate, treatment with bone protection therapy in osteoporosis patients, falls risk assessment and systems to improve adherence and persistence with therapy. Organising such services is challenging owing to the multidisciplinary care that patients require. 3 The 2011 NICE hip fracture CGs make specific recommendations regarding the treatment and multidisciplinary management of patients including liaison and integration of services. A fracture liaison service (FLS) is the recommended model proposed by the Department of Health to organise secondary fracture prevention services1 in a ‘one-stop shop’ setting delivered by a nurse specialist supported by a lead clinician (‘champion’) in osteoporosis. 19 However, currently only 30% of hospitals in England have established a FLS. 17 A single model incorporating all components of secondary fracture prevention has not been mandated. Current practice is for various combinations of these components to be used within a hospital (and in some cases no components are used).

Current knowledge

The clinical effectiveness of co-ordinator-based models of care has been demonstrated, in terms of improving the uptake of appropriate osteoporosis management such as measuring bone density and the use of antiresorptive drug therapy. 4,10 There is growing evidence of these models’ cost-effectiveness20,21 and that they can provide cost savings to the NHS. 4,10,22,23 Evidence is emerging on the ability of co-ordinator-based systems to reduce the incidence of hip fractures. A review of the Glasgow Osteoporosis and Falls Strategy reported that hip fracture rates in the city had reduced by 7.3% over the decade, compared with a 17% increase in fracture rates for the entire population of England over the same period. 19,24 These findings are consistent with observational data from the USA by Dell, who reported a 37.2% reduction in hip fracture rates. 25 However, the strongest evidence on effectiveness has recently been provided by an Australian study that was designed as a prospective observational trial with a concurrent control group in which, compared with standard care, targeted identification and management significantly reduced the risk of refracture by > 80%. 26

Across the UK there is variation in the care pathway of the treatment and management of hip fracture patients and in the way secondary fracture prevention services are structured and organised. Even with a co-ordinator-based system in place, the structure of services can vary between hospitals. For example, hospitals use different models of orthogeriatric care, in which some hospitals now have specialised orthogeriatric wards and in others patients are seen on the trauma ward. Some hospitals may co-ordinate the care of hip fracture patients only while such patients are admitted as inpatients, whereas others have ensured that their osteoporosis service is integrated across primary care to monitor patients’ adherence to bone protection therapy.

Objective 1: characterise secondary prevention of hip fracture across hospitals in a region of England (work stream 1)

The first phase of this project was to comprehensively describe and explore the variation and disparity in secondary fracture prevention services offered to hip fracture patients across hospitals in a region of England, and to identify the dates of key changes made to service delivery over the past decade.

Objective 2: identify the reasons why hospitals chose their specific model of service delivery and assess barriers to change (work stream 2)

The aims of the qualitative study were to (1) ascertain the reasons why each hospital has adopted their current and most recent models of care; (2) establish factors that facilitate or act as barriers to changes in service delivery; and (3) identify the elements of care of hip fracture patients that health professionals think are most effective.

Objective 3: evaluate the impact that changes to the delivery of secondary fracture prevention has had on health outcomes by altering trends in hip refracture rates, NHS costs and life expectancy (work stream 4)

A natural experimental study design27 evaluates the impact that changes hospitals in the region have made to the way in which they deliver secondary fracture prevention services for hip fracture patients has had on a range of health outcomes (mortality, second hip fracture and other fragility fractures).

Objective 4: establish the cost-effectiveness of different hospital models for delivery of secondary fracture prevention (work stream 3)

The health economics work calculates the hospital and non-hospital costs associated with hip fracture in the year of fracture and subsequent years and evaluate the costs (quality-adjusted) life expectancy and cost-effectiveness of the different hospital models of care.

Characterise secondary prevention of hip fracture across hospitals in a region of England (work stream 1)

In order to collect information on models of care for secondary fracture prevention at each hospital, we conducted a service evaluation which comprised:

• developing a questionnaire to collect information on staffing levels and new appointments over the past decade, procedures for case finding, osteoporosis assessment (including location of DXA scanner), falls assessments, treatment initiation and follow-up, as well as integration across primary, secondary and community care

• identifying health professionals at each hospital to complete the questionnaire through a regional network of clinicians working in osteoporosis management.

Identify the reasons why hospitals chose their specific model of service delivery and assess barriers to change (work stream 2)

This qualitative research component involved the following stages:

• identifying a local collaborator at each hospital through a regional network of clinicians and, with their input, identifying other health-care professionals and service mangers working at the hospital in secondary fracture prevention

• conducting qualitative interviews with 3–5 health-care professionals or service managers at each hospital

• thematic analysis conducted using codes to identify themes and subthemes.

Evaluate the impact that changes to the delivery of secondary fracture prevention has had on health outcomes by altering trends in hip refracture rates, NHS costs and life expectancy (work stream 3)

A natural experiment study design was used to assess the clinical effectiveness of changes in service delivery in terms of reducing mortality and secondary fracture rates using the following procedure:

• trends in rates of 30-day mortality, 1-year mortality and secondary hip fracture established from Hospital Episodes Statistics (HES) data for hospitals in the region

• interventions, or changes in service delivery, and their corresponding dates were identified from the service evaluation conducted in work stream 1

• interrupted time series analysis and Cox proportional hazards regression modelling used to analyse impact of each intervention on outcomes of interest.

Establish the cost-effectiveness of different hospital models for delivery of secondary fracture prevention (work stream 4)

An economic analysis of the costs and cost-effectiveness of different services will be undertaken as described below.

• Data from HES and the Clinical Practice Research Datalink (CPRD) will be used to identify the hospital and non-hospital costs of a hip fracture.

• A Markov model will be developed to evaluate costs and cost-effectiveness of each model of care (using a measure of clinical effectiveness from work stream 3).

• Utility scores (from literature search) will be used to provide the weights required to calculate the quality-adjusted life-years (QALYs) of the different models of care under evaluation. 28

Current knowledge of variations in delivery of fracture prevention services

Studies have shown a major variation in how hospitals organise their services for the treatment of hip fracture patients and management of secondary fracture prevention. The NHFD is a national audit from the Royal College of Physicians that includes information on the number of orthogeriatric sessions and specialist nurses per centre to deliver secondary fracture prevention. 17 Although the number of orthogeriatric hours per week is modestly related to reported hip fracture numbers, there was a wide variation in FLS service delivery. In 2013, 62% of hospitals reported that they had no fracture liaison nurse, 27% reported up to one whole-time equivalent (WTE) post and 11% reported more than one WTE. Furthermore, in those centres with a specialist nurse, there was no relationship between volume of fragility fractures and number of staff (Figure 1).

FIGURE 1.

The relationship between reported number of WTE specialist nurses for secondary fracture prevention and estimated number of fragility fracture patients seen in that hospital per year. Each data point represents a hospital that returned > 0 WTE of specialist nurse in the NHFD 2014 report. The number of fragility fractures per hospital was estimated using five times the number of proximal femoral fractures. The line shows a lowess plot with a bandwidth of 0.9.

These variations in services using even crude metrics such as staff resource are too great to be explained by local case mix and volume of hip fracture patients across hospitals. Establishing the differences between service models and levels of care in more details will allow clinicians and commissioners to identify existing gaps in care, and could help to reduce unwarranted variation across hospitals.

Current guidelines

A number of professional bodies have provided national guidance for the care of hip fracture patients reflecting two key aspects. The first type of guidance relates to optimising initial recovery after hip fracture and focuses around the provision of patient-centred care29 within three broad categories:

1. optimising surgical procedure – including early timing of surgery, preoperative appropriate correction of comorbidities and type of implant

2. early mobilisation

3. multidisciplinary management from admission to discharge.

The second type of guideline relates to secondary fracture prevention and there are many more guidelines available, including guidance from Canada22,30,31 the USA32 and the UK. 1,14,15,33 The ‘Capture the Fracture’ initiative from the International Osteoporosis Foundation (IOF) provides additional international guidance. 34 Guidelines describe a comprehensive fracture prevention service consisting of four main components:

1. case finding those at risk of further fractures

2. undertaking an evidence-based osteoporosis assessment, including a DXA scan to measure bone density, if appropriate

3. treatment initiation in accordance with guidelines for both bone health and falls risk reduction

4. monitoring and improving adherence to recommended therapies.

Such services require multidisciplinary care from orthogeriatrics, rheumatology, falls services and primary care. 3 Running such a service can be challenging and, as a result, such services often require a dedicated co-ordinator to provide a link between the different multidisciplinary teams involved in the care pathway. This is known as a co-ordinator-based system of care. 10 This co-ordinated, multidisciplinary approach to patient care is known as a FLS. 19 FLSs have now been introduced internationally. 35,36 In the UK, the Department of Health has proposed a model of best practice which is delivered by a nurse specialist supported by a lead clinician (‘champion’) in osteoporosis. 1 In addition, The Care of Patients with Fragility Fractures, published by the BOA, recommends fully integrating orthogeriatricians into the way the fracture service works in order to best meet the complex needs of hip fracture patients. 1

However, despite the guidance that is in place, there is no overall agreement on the best way to organise these services, and a single model comprising all elements of a FLS has not been mandated.

Changes in hip fracture care over the past decade

Guidance from a number of professional bodies for the management of hip fracture patients (BOA Blue Book,1 SIGN14 and NICE29,33) has led to key changes in hip fracture care across the UK. The NHFD collects data on staffing levels as well as on many other aspects of care, and has shown a significant increase in the number of consultant grade orthogeriatric hours per week since reporting began in 2009, and a particularly sharp upwards trend in the number of fracture liaison nurse hours per week since 2012. 16,17,37,38

Comparison of services across hospitals

To allow direct comparison of service levels between hospitals, staffing levels were calculated as WTEs and as ratios in terms of the number of WTEs per 1000 hip fracture patients, using data on the number of hip fractures reported in the 2013 NHFD report. 17 For simplicity, we grouped all nurses working in fracture prevention (e.g. trauma nurse, osteoporosis nurse specialist) together under the title ‘fracture liaison nurse’. In addition, we identified and described variations in how the four main elements of a fracture prevention service (identification, investigation, initiation and monitoring) are co-ordinated and conducted.

Identification of key changes in service delivery

Information on changes in service delivery, such as the appointment of a new member of staff, the opening of a new ward or changes in procedures for referring patients for DXA scans, was assembled and presented to the study team, which included a number of clinicians. Together, the team identified the dates of the key changes to feed into the analysis in work stream 3. Once a list of changes was assembled for each hospital, the list was sent to a health-care professional at that hospital to ensure that they agreed with the changes that had been identified.

Current disparity in fracture prevention across hospitals

Changes in service delivery at hospitals in the past decade

The changes in service delivery at each hospital are summarised in Table 5. These were identified through the service evaluation questionnaire, with additional information supplemented from the qualitative work. Members of the project working group agreed on the key changes that had occurred based on the data collection, with significant input from clinicians around what factored as a key change.

Looking specifically at how WTE of specialist staff changed between 2003 and 2013 (Table 6), we see a general increase in the number of consultant orthogeriatricians and specialist nurses across most hospitals in the region, reflecting new guidelines for both optimal recovery after hip fracture and secondary fracture prevention introduced during this period.

Description of current services

In conducting a service evaluation designed to look at changes in service delivery at a set of hospitals over the past 10 years, a number of observations were noted. Although there has been an overall increase in service provision in terms of both orthogeriatric and fracture liaison nurse appointments, a considerable variation remains in how fracture prevention services at 11 NHS hospitals in one region of England are organised. The largest disparities were in how the service was run (an inpatient service vs. an outpatient service) and in the levels of staffing provided for the FLS service. Some hospitals had only a consultant-led service in order to improve perioperative care of fracture patients, whereas other hospitals had a nurse-led service to tackle fracture prevention. Few hospitals have both a fracture liaison and orthogeriatric service, as is recommended by CGs. 1,13,15,29

Details on staffing levels across hospitals were already provided by the NHFD; however, this evaluation showed that the role of certain staff members such as orthogeriatricians also varied between hospitals in terms of their role in case finding, osteoporosis assessment, treatment initiation and monitoring. In some cases, these roles were actually performed by more junior members of staff. Although some hospitals reported having an orthogeriatrician in post, their role in fracture prevention may be extremely limited.

Changes in service models

The previous decade has demonstrated marked increases in the provision of both orthogeriatric and/or FLS nurse staff in each of the hospitals. Only in one hospital was a service discontinued. However, the increase in service provision is variable, with little relationship to the number of hip fracture patients. By the end of the period of observation, the number of orthogeriatric staff per 1000 hip fracture patients varied from 0 to 4.3. The equivalent number of FLS nurses to numbers of hip fractures varied from 0 to 9.3. This variability is greater than one would expect for local differences in NHS structure or patient case mix and again demonstrates the need to relate service investment with clinical effectiveness in order to demonstrate the relative value of such services.

Strengths and limitations

A major strength of this evaluation is the heterogeneity of NHS hospitals examined, from smaller district general hospitals to large tertiary major trauma centres, adding to the generalisability of the report’s findings. A limitation is the reliance on clinicians’ recall and understandings of events over the last decade. Although human resources departments can confirm the appointment of new staff, changes may often come about by alterations in the role and responsibilities of existing staff members.

Implementation of complex interventions

The implementation of complex interventions is increasingly being studied using implementation theory. 41 Implementation research comprises a number of approaches and theoretical stances in order to help understand something of the complexity of change within health services. One such theory, extended NPT42 – which builds on two previous theoretical models, the Normalisation Process Model43 and NPT44 – describes how the collective actions of agents drives the implementation of new services. This provides a counterpoint to network perspectives whereby the implementation of innovations is viewed as a process of transmission and stabilisation through networks,45,46 and psychological perspectives that prioritise the role of individuals in instigating change. 47,48 The theory builds on previous iterations of the theory by combining the notion of implementation as a social process with psychological and network approaches to increase our understanding of the phenomena.

Extended Normalisation Process Theory

According to the theory, the successful implementation of an intervention is based on the ability of agents to fulfil four criteria, described using constructs. 42 These are outlined in Table 7.

Making business cases for a fracture liaison service

A potential barrier to the implementation of FLSs is the challenge of making effective cases to obtain funding. 32 Funding for the commissioning of new services is obtained by writing business cases and presenting them to managerial bodies in the trust. These cases may be developed by a range of professionals. Central to this process are the clinicians and service managers working within the department. Finance managers are also involved in costing developments and calculating their potential income generation. Clinicians and service managers from other departments may also have a role, along with patient representatives. If supported, these may be referred to local Clinical Commissioning Groups (CCGs) for final funding approval, especially in the case of larger-scale service developments.

Commissioning (purchasing) processes within the NHS are complex. In April 2013, responsibility for commissioning services changed from the Department of Health and local primary care trusts (PCTs) led by managers, to NHS England, a new organisation overseeing 211 GP-led local CCGs. These currently control 65% of the NHS budget and operate within defined geographical regions. The rationale of this restructure is that GPs are best placed to understand the clinical needs of the local population and are therefore able to utilise resources most effectively. 49 The introduction of CCGs has coincided with a rise in the number of foundation trusts in England. Foundation trusts are independent organisations responsible for providing over half of NHS hospital, mental health and ambulance services operating in local regions. They are contracted to deliver services that have been commissioned by local CCGs and NHS England and are regulated by an independent body called Monitor. A board of directors is responsible for setting budgets for the financial year and establishing targets and health-care priorities. 50

Commissioning has been targeted by the Department of Health as a means of raising standards in the NHS and ensuring that resources are evenly distributed. 1 The NHS Commissioning Board states that funding decisions should be systematic and transparent and that considerations should include cost-effectiveness, improvements in care quality and the strategic plans of NHS trusts. 51 Core public health priorities, prioritising the distribution of resources in key disease areas, have also been identified. 52

There is some support available for clinicians and service managers on how to develop business cases for a FLS. This includes written guidance,32,53 training courses54 and business case templates. 55 According to this support, business cases should describe the health profile of the population, including the numbers of fractures, outline service design with reference to established fracture prevention guidelines, outline costs, particularly how the service will generate savings,55 and utilise a number of different stakeholder groups. 32 More general guidance on the processes of designing services and making business cases also exists and some CCGs have developed business case templates to assist in this process. 56

Several studies have explored the experiences of commissioners of making purchasing decisions in the UK, including those operating under the now abolished PCTs57,58 and the CCGs. 59,60 These have examined relational aspects of commissioning57,61,62 and identified factors that impact on decision-making. 58,63 Recent research has also explored the new roles of GPs in the commissioning process. 64,65 However, only one study has explored the experiences of providers in acute care of making business cases. 66

Sample

The sample comprised 43 professionals from all 11 hospitals in one region in England who were involved in the delivery or organisation of secondary fracture prevention after hip fracture. These included orthogeriatricians, fracture prevention nurses, trauma nurses, hospital practitioners in osteoporosis, surgeons and service managers. Participants were purposively sampled to include a range of characteristics such as profession and years in role and to ensure that they were adequately drawn from each of the 11 hospitals. 67

Potential participants were identified by the clinical lead/champion in osteoporosis and an operational service manager in trauma. In three waves of recruitment, the study team approached potential participants by e-mail to provide them with a brief overview of the study, including a participant information booklet for further information. If no response was received within 2 weeks, the e-mail was followed up with a telephone call. Snowball sampling68 was used such that participants recommended other professionals involved in fracture prevention. In total, 82 health-care professionals were contacted to take part in the study, of whom 43 agreed to take part. The remainder either declined or were unavailable. Rather than us aiming to achieve saturation,69 criterion sampling was applied to ensure an adequate range of professionals to enable us to address the issues under study. 70

Ethics approval

Ethics approval was provided by the University of Oxford’s Central University Research Ethics Committee in 2012, reference number MSD-IDREC-C1-2012-147. Each NHS trust involved provided research and development approval. Participants all provided their written consent prior to interview. This consent process emphasised the aims of the research and that participation was voluntary. Participants were also asked to provide their consent to audio-recording and to the inclusion of anonymous quotations in study publications.

Interview procedure

A qualitative researcher (Sarah Drew) undertook interviews in 2013. All interviews except one were conducted face to face at the participants’ workplaces; one interview took place by telephone. Interviews were between 30 and 50 minutes long. A topic guide (see Appendix 2) was used to help to ensure that similar questions were asked of all participants. The topic guide was structured into three areas. The first topic area was based on the four core elements of a fracture prevention service outlined by the IOF as part of the Capture the Fracture initiative. 34 This helped the researcher to explore participants’ views on the best models of care for the secondary fracture across the four main components of a fracture prevention service and co-ordination of care. The second area covered in interviews was structured around the four constructs of extended NPT to enable exploration of participants’ experiences of implementing fracture prevention services. The third area explored in interview was participants’ experiences of making business cases for a FLS, factors that they felt informed the decision of commissioners to approve the new service and what they considered to be the best strategies for making them. The interviewer used standard qualitative interview methods such as ‘probing’ to help achieve depth in the interviews. 71 After the first four interviews we revisited the topic guide and made amendments to it to ensure that it enabled us to successfully explore the issues under study. The topic guide shown in this report is the final topic guide used in the study. All interviews, including the first four pilot interviews, are included in the final data set.

Data analysis

Interviews were audio-recorded, transcribed, anonymised and imported into the qualitative data analysis software NVivo (QSR International, Warrington, UK). Analysis took place in two parts. Part 1 used coding and an abductive approach to address how and why secondary fracture prevention services can be successfully implemented in secondary care using extended NPT to provide theoretical basis to the analysis. Part 2 was an inductive thematic analysis, albeit that it was focused on exploring participants’ experiences of making business cases for a FLS, factors that they felt informed the decision of commissioners to approve the new service and what they considered to be the best strategies for making them.

Part 1: using extended Normalisation Process Theory to understand how and why secondary fracture prevention services can be successfully implemented, barriers and enablers to change and elements of care seen as most effective

Part 2: exploring the experiences of clinicians and service managers of developing and making business cases for a fracture liaison service

In the second part of the qualitative research we explored the experience of business case development for a FLS. This was addressed through the latter part of the interviews, as described in Methods. Of the 43 participants in the study as a whole, 33 had experience of making cases (Table 10), and we analysed the material from these interviews in relation to this topic. In the findings presented here, we describe health-care professionals’ views and experiences of making business cases for a FLS; identify factors they think impact on the decision of managerial bodies to approve the service; and explore what they feel are the most successful strategies for making an effective case.

General practitioners and the NHS: why is the Clinical Practice Research Datalink a valuable source of data?

Primary care health services play an essential role in the UK NHS. GPs working in NHS primary care play the role of gatekeepers to the system: access to secondary care is possible only after referral from a GP practice, with the exception of urgent attention in hospital emergency departments.

In addition, GPs are a key part of NHS health-care continuity: they receive clinical letters from secondary care (consultants and other health professionals) to inform or suggest shared therapeutic plans, and they are responsible for most repeat prescriptions for long-term conditions. As part of this, they also receive hospital discharge letters regarding all patients registered with them; the information contained, including inpatient treatments and diagnoses, among others, is then used by GPs to co-ordinate outpatient care (Figure 3). All of this adds value to the information gathered in primary care medical records.

FIGURE 3.

The role of GPs in the NHS, and the flow of information into primary care GP records. A&E accident and emergency.

The CPRD gathers information from computerised primary care records, which can be enriched with other digital linked data from a number of data sets, providing a unique source of information for epidemiological research.

The Clinical Practice Research Datalink

Jointly funded by the National Institute for Health Research (NIHR) and the Medicines and Healthcare products Regulatory Agency, the CPRD (formerly General Practice Research Datalink) is a unique data source, used extensively by researchers worldwide for observational and interventional research. Such studies have resulted in > 900 published articles and a huge number of conference presentations.

The core CPRD Gp OnLine Data (GOLD) version of the database was used as the source from which patient-level bespoke data sets were extracted. It comprises anonymised clinical records for a representative > 10 million people living in the UK. For this sample, the data set contains information on any diagnoses (primary and secondary care based) recorded by GPs in primary care records using Read/OXMIS codes, clinical measurements taken in primary care (including body mass index, smoking, alcohol drinking and others), routine laboratory results, referrals, GP prescriptions [identified using British National Formulary (BNF) codes90] and administrative information (date of registration in the primary care practice, date of death, etc.). The encoding of patient identifiers ensures anonymisation of the data. A full list of medical codes used to identify hip fractures can be found in Appendix 3.

This variety of data was provided in a number of text files, which have been checked, imported and linked in a database management system (MySQL, Oracle Corporation, Cupertino, CA, USA) using programming (Python; www.python.org). The required variables have then been extracted from the database in the right format to be analysed using conventional statistical packages.

Linked hospital and mortality data

About 60% of the primary care practices contributing to CPRD have agreed to linkage to the HES and Office for National Statistics (ONS) data. 84,91 These data sources provide information on hospital admission episodes in NHS hospitals and mortality (date and cause of death), respectively.

Hospital Episode Statistics and ONS information was also combined with the core CPRD GOLD data set to improve the completeness and accuracy (of diagnoses, inpatient procedures and mortality) of the data analysed.

Introduction

The HES database gathers clinical information on the records for hospital admission episodes in England, including both those in NHS hospitals and those in the independent sector that are commissioned by the NHS. Data on such episodes have been recorded from 1987, and only a sample (around 10%) of the episodes were included before that time.

Inpatient data are available from 1989–90 onwards; these were initially delivered every financial year but more recently have been delivered on a monthly basis. Since 2006, data collection and preparation have been undertaken by the Secondary Uses Service, which is part of the Health and Social Care Information Centre.

The data collected in HES are submitted by participating hospitals in all NHS trusts in England for reimbursement purposes. Although it is primarily an administrative database, HES is designed to allow also for secondary use for research.

Data available in Hospital Episode Statistics

For all NHS hospital admission episodes, outpatient and accident and emergency (A&E) appointments in England, each HES record contains the following information:

• hospital diagnoses and procedures

• administrative data – date of admission/discharge, etc.

Hospital inpatient data are available from 1989 onwards, outpatient appointments data are available from 2003 and A&E information has been recorded since 2007.

A bespoke anonymous patient ID (called gen HESid) is created for each individual, and used to track patients through the data set while preserving confidentiality. The patid (allocated to each patient from his or her practice) is used to enable linkage of HES data to other relevant data sets, such as ONS mortality and CPRD data.

Hospital admissions in Hospital Episode Statistics

Of the different types of information available within the HES database, this study used a data set including hospital admission episodes. Such records contain information on a primary hospital diagnosis (reason for admission), a number of secondary/concomitant diagnoses (comorbidity) and a number of in-hospital procedures.

In addition, information on patient complexity used for reimbursement is also available in the form of Healthcare Resource Group (HRGs), which can be used to calculate the cost to the NHS of each hospital admission episode.

Data sources

A cohort of hip fracture patients treated at hospitals in the South Central region using English national HES data was linked to the national mortality data from 1999 to 2011. The data came from a linked data set of English national HES and data from death certification built by the team that developed and manages the Oxford record linkage study. The HES data set includes records of all inpatient episodes undertaken in NHS trusts in England, including acute hospitals. Information about deaths comes from death certificates held by the ONS, which includes information on the date of death. A cohort of hip fracture patients (4260 patients per year in South Central) was identified through OPCS Classification of Interventions and Procedures version 4 (OPCS-4) operation codes103 and International Classification of Diseases, Tenth Edition (ICD-10) diagnostic codes. 104 The hospital provider code allowed identification of the 11 hospitals in the region.

Information on hip fracture hospital admissions was obtained from the HES database. Each record relates to a ‘finished consultant episode’: the period of time an individual spends under the care of one NHS consultant. Private procedures are excluded from HES, as there is no requirement for private hospitals to provide routine data. Mortality data from the ONS were linked and extracted using an encrypted patient identifier.

Patients

All patient episodes containing an ICD-10 code indicating hip fracture (S72.0, S72.1, S72.2 or S72.9) as the primary diagnosis and with a start date occurring within the study period (1 April 2003–31 March 2013) were identified (Figure 4). Hip fracture episodes were excluded if they indicated the patient was < 60 years of age, admitted as a day-case, admitted as a non-emergency or admitted as a result of trauma such as a transport accident (external cause ICD-10 codes: V01–V99). Hip fracture episodes were excluded if it was indicated that the patient had been admitted for any previous hip fracture (regardless of eligibility criteria) any time between 1 April 1999 and 31 March 2013. Provider codes and treatment site codes were used to identify episodes occurring within the 11 hospitals in the region of interest.

FIGURE 4.

Population flow diagram.

It is important to establish the laterality of the second versus index hip fracture to avoid double counting the index fracture. Given that the ICD-10 codes do not indicate laterality, we used the OPCS codes from NICE CG124 2011:29 W191, W241, Z762 + W192, Z762 + W193, Z762 + W242, Z763 + W192, Z763 + W193, Z763 + W242, W371, W381, W391, W461, W462, W482, W471, W472, W481, W931, W941 and W951. Eighty-six per cent of primary fractures and 79% of second hip fractures had relevant procedure codes.

Outcomes

The primary outcome of interest was time to second hip fracture within 2 years of a primary hip fracture. To ensure that this was a separate fracture and not the first fracture recoded, we counted second hip fractures only if admitted in a separate ‘continuous inpatient spell’ (CIPS) and at least 30 days after admission for the primary fracture. Secondary outcomes of interest were (1) time to death within 30 days; (2) time to death within 1 year following a primary hip fracture admission; and (3) non-hip fractures.

Interventions

The primary exposure (‘intervention’) was the implementation within individual hospitals of specific change to the model of post-hip fracture care. Information on the nature and timing of such changes had been obtained through a detailed evaluation of hip fracture services within the 11 hospitals of interest over the last decade, and was carried out prior to data being obtained. 39 Dates for the introduction or expansion of either an orthogeriatric or a FLS model of post-hip fracture care occurring throughout the study period were identified a priori (Table 11 and see Appendix 4).

Confounders

Confounding factors controlled for were age, sex, Index of Multiple Deprivation (IMD) score and Charlson Comorbidity Index (none, mild, moderate and severe).

Sample size calculation

Statistical analysis

Survival analysis

A total of 33,152 hospital episodes were identified as pertaining to a primary hip fracture. The total number over the study period per hospital of interest ranged from 1030 to 5895. The proportion of female admissions significantly changed over time, decreasing from 78.2% (2003–4) to 72.0% (2012–13). Mean age increased slightly over the study period from 82.7 years (2003–4) to 83.1 years (2012–13).

Two of the 13 interventions could not be analysed in relation to time to second hip fracture owing to insufficient pre- or post-intervention follow-up time once a 12-month lag period was introduced. Interventions that were preceded by a significant pre-intervention trend in health outcome were not evaluated in relation to that particular health outcome owing to the bias of a pre-existing secular trend likely to be incorporated in such estimates of intervention impact. For this reason two interventions were not evaluated in relation to 1-year post-fracture mortality.

Second hip fracture

There were 1288 patients identified as sustaining a second hip fracture, at an average annual directly age- and sex-standardised proportion of 4.2%. This proportion remained stable throughout the study period (p-trend = 0.11), with no significant change in annual rate within any year of follow-up. Of the patients identified as having sustained a second hip fracture, 883 (69%) had both procedure and laterality codes for both index and second fracture. For 96% of this subset of patients with known procedure and laterality, the second fracture occurred at the contralateral side to the index fracture; that is only 4.4% of these patients sustained a second hip fracture on the same side as the index fracture.

The SHRs from the survival models showed no evidence for an impact on time to second hip fracture (Figure 5) following any of the interventions when analysed separately or when pooled by type of intervention: orthogeriatrician [SHR 0.95, 95% confidence interval (CI) 0.79 to 1.15] or FLS (SHR 1.03, 95% CI 0.85 to 1.26). Analyses of intervention impact on time to second hip fracture remained unchanged when stratified by sex, age (< 75 years or ≥ 75 years) or Charlson Comorbidity Index score (0 or ≥ 1).

FIGURE 5.

Forest plot of SHRs for 2-year second hip fracture, by type of change in service delivery. a, Prior orthogeriatrician; b, prior FLS; and c, prior FLS and orthogeriatrician.

Thirty-day and 1-year mortality

The average annualised age- and sex-standardised proportions for 30-day and 1-year mortality were 9.5% (n = 3033) and 29.8% (n = 9663), respectively. Overall, age- and sex-standardised 30-day mortality declined from 11.8% in 2003–4 to 7.1% in 2011–12 (p-trend < 0.001) (Figure 6). When compared with the year 2003, 30-day mortality remained stable until 2005–6 but significantly decreased thereafter. Similarly, the age- and sex-standardised 1-year mortality of 33.1% in 2003–4 remained stable until 2006–7 and thereafter decreased markedly to 26.0% in 2011–12 (overall p-trend < 0.001).

FIGURE 6.

Annual and quarterly regional trends in mortality (30 days and 1 year) and second hip fracture (2 years) after primary hip fracture during the study period. Reproduced with permission from Hawley SJ, Javaid MK, Prieto-Alhambra D, Arden N, Lippett J, Sheard S, et al. Clinical effectiveness of orthogeriatric and fracture liaison service models of care for hip fracture patients: population-based longitudinal study. Age Ageing 2016;45:236–42, under a Creative Commons licence (https://creativecommons.org/licenses/by-nc/4.0/legalcode).

The pooled estimated impact of introducing an orthogeriatrician on 30-day and 1-year mortality was HR 0.73 (95% CI 0.65 to 0.82) and HR 0.81 (95% CI 0.75 to 0.87), respectively (Figures 7 and 8). Thirty-day and 1-year mortality were likewise reduced following the introduction of a FLS: HR 0.80 (95% CI 0.71 to 0.91) and HR 0.84 (95% CI 0.77 to 0.93), respectively.

FIGURE 7.

Forest plot of HRs for 30-day mortality, by type of change in service delivery. a, Prior orthogeriatrician; b, prior FLS; and c, prior FLS and orthogeriatrician. Reproduced with permission from Hawley SJ, Javaid MK, Prieto-Alhambra D, Arden N, Lippett J, Sheard S, et al. Clinical effectiveness of orthogeriatric and fracture liaison service models of care for hip fracture patients: population-based longitudinal study. Age Ageing 2016;45:236–42, under a Creative Commons licence (https://creativecommons.org/licenses/by-nc/4.0/legalcode).

FIGURE 8.

Forest plot of HRs for 1-year mortality, by type of change in service delivery. a, Prior orthogeriatrician; b, prior FLS. Reproduced with permission from Hawley SJ, Javaid MK, Prieto-Alhambra D, Arden N, Lippett J, Sheard S, et al. Clinical effectiveness of orthogeriatric and fracture liaison service models of care for hip fracture patients: population-based longitudinal study. Age Ageing 2016;45:236–42, under a Creative Commons licence (https://creativecommons.org/licenses/by-nc/4.0/legalcode).

The reductions in mortality were seen whether introducing an orthogeriatric or FLS model of care for the first time or as part of an expansion of an existing service. For example, 1-year mortality was reduced following the appointment of a second orthogeriatrician within Hospital 2 in August 2007, compared with service delivery during the time with one orthogeriatrician already in post (HR 0.78, 95% CI 0.67 to 0.91). Likewise, adding two extra nurses and a consultant ‘champion’ for osteoporosis to the FLS model of care in Hospital 10 in May 2008 was associated with a further reduction in 1-year mortality (HR 0.76, 95% CI 0.63 to 0.92), compared with the relatively less intensive FLS in place before this intervention (see Figure 7).

Non-hip fractures

Overall, 2.8% of index hip fracture patients went on within 2 years to have a major non-hip fragility fracture requiring hospital admission. At a regional level, the rate of subsequent major non-hip fracture after primary hip fracture increased from 1.9% in 2003 to 2.5% in 2009 (p-trend = 0.03) (see Figure 6). Pooled SHRs indicated no significant impact of either orthogeriatrician or FLS (Figure 9).

FIGURE 9.

Forest plot of SHRs for 2-year major non-hip fracture, by type of change in service delivery interrupted time-series analysis. a, Prior orthogeriatrician; b, prior FLS. Reproduced with permission from Hawley SJ, Javaid MK, Prieto-Alhambra D, Arden N, Lippett J, Sheard S, et al. Clinical effectiveness of orthogeriatric and fracture liaison service models of care for hip fracture patients: population-based longitudinal study. Age Ageing 2016;45:236–42, under a Creative Commons licence (https://creativecommons.org/licenses/by-nc/4.0/legalcode).

The results of the interrupted time series analysis are presented for the five biggest hospitals in the region only because of the limited number of observations available at each time point. Figures 10 and 11 present a visual representation of the quarterly trends in rates of mortality and second hip fracture at each of these individual hospitals. Using a segmented linear regression approach, no interventions were associated with either a step or a slope change in 2-year secondary hip fracture (Table 12). For 30-day and 1-year mortality, large effect sizes were observed for the estimated step changes (Tables 13 and 14), but none reached conventional levels of statistical significance owing to limited statistical power.

FIGURE 10.

Quarterly regional trends in second hip fracture (2 years) after primary hip fracture during the study period, by hospital. The grey bars represent the date of the interventions in each hospital. The dotted line represents a specialist nurse intervention at the hospital that did not treat hip fracture. (a) Hospital 2; (b) Hospital 4; (c) Hospital 7; (d) Hospital 8; and (e) Hospital 10.

FIGURE 11.

Quarterly regional trends in mortality (30 days and 1 year) after primary hip fracture during the study period, by hospital. The grey bars represent the date of the interventions in each hospital. The dotted line represents a specialist nurse intervention at the hospital that did not treat hip fracture. (a) Hospital 2; (b) Hospital 4; (c) Hospital 7; (d) Hospital 8; (e) Hospital 10.

TABLE 12 - Results of segmented linear regression models for second hip fracture outcome for each hospital

Hospital	Baseline trend		Step change		Slope change		Step change		Slope change		Step change		Slope change
	β1	p-value	β2	p-value	β3	p-value	β4	p-value	β5	p-value	β6	p-value	β7	p-value
Hospital 10
Full model	–0.23	0.25	2.12	0.430	2.89	0.590	–2.61	0.470	0.35	0.530
Parsimonious model	< 0.01	0.96
Hospital 8
Full model	0.04	0.67	–0.53	0.81	0.16	0.78
Final model	0.04	0.27
Hospital 2
Full model	–0.71	0.81	2.99	0.39	–0.45	0.55	2.08	0.63	0.5	0.47
Parsimonious model	–0.02	0.69
Hospital 7
Full model	0.11	0.8	0.20	0.95	–0.37	0.47	1.77	0.39	0.24	0.38
Parsimonious model	< 0.01	0.89
Hospital 4
Full model	–0.33	0.51	–2.56	0.57	0.44	0.49	–2.53	0.51	0.68	0.5
Parsimonious model	–0.02	0.75

TABLE 13 - Results of segmented linear regression models for 30-day mortality for each hospital

Hospital	Baseline		Intervention 1				Intervention 2				Intervention 3
	Baseline trend		Step change		Slope change		Step change		Slope change		Step change		Slope change
	β1	p-value	β2	p-value	β3	p-value	β4	p-value	β5	p-value	β6	p-value	β7	p-value
Hospital 10
Full model	0.00	1.00	–3.93	0.23	0.42	0.55	–0.67	0.89	–1.52	0.28	3.11	0.54	1.21	0.36
Parsimonious model	–0.14	< 0.001
Hospital 8
Full model	< 0.01	0.98	–3.75	0.088	< 0.01	0.99
Parsimonious model	< 0.01	0.96	–3.75	0.073
Hospital 2
Full model	0.40	0.42	–4.84	0.24	–0.09	0.91	–2.79	0.45	–0.49	0.44
Parsimonious model	–0.16	0.001
Hospital 7
Full model	0.7	0.45	–4.1	0.28	–0.73	0.45	2.28	0.35	–0.33	0.34
Parsimonious model	–0.18	< 0.001
Hospital 4
Full model	–0.49	0.56	–1.57	0.74	0.73	0.45	–4.51	0.25	–0.11	0.86
Parsimonious model	–0.19	0.011

TABLE 14 - Results of segmented linear regression models for 1-year mortality for each hospital

Hospital	Baseline		Intervention 1				Intervention 2				Intervention 3
	Baseline trend		Step change		Slope change		Step change		Slope change		Step change		Slope change
	β1	p-value	β2	p-value	β3	p-value	β4	p-value	β5	p-value	β6	p-value	β7	p-value
Hospital 10
Full model	0.43	0.22	–8.07	0.10	0.61	0.56	–4.89	0.48	–2.76	0.19	6.89	0.36	2.08	0.29
Parsimonious model	–0.13	0.063
Hospital 8
Full model	–0.08	0.72	–1.64	0.66	–0.44	0.36
Parsimonious model	–0.31	0.003
Hospital 2
Full model	0.74	0.33	–6.99	0.26	–0.55	0.65	–3.05	0.58	–0.43	0.65
Parsimonious model	–0.26	0.001
Hospital 7
Full model	–0.29	0.8	–1.51	0.75	0.12	0.92	–1.51	0.62	–0.06	0.89
Parsimonious model	–0.32	< 0.001
Hospital 4
Full model	–0.36	0.79	–3.02	0.7	0.22	0.89	2.01	0.74	–0.25	0.79
Parsimonious model	–0.3	0.01

Main findings

In this study we observed an overall decline in mortality after primary hip fracture across the region as a whole from financial year 2003 to financial year 2011. Despite this increased survivorship following primary hip fracture, rates of second hip fracture remained relatively stable across the region. It was demonstrated from pooling estimates by intervention type that the introduction and/or expansion of orthogeriatric and FLS models of care was significantly associated with reduced post-hip fracture mortality. An alternative way of expressing this beneficial impact is using the number needed to treat,111 which is the number of hip fracture patients needing to be treated following an intervention in order to prevent one excess death, compared with service delivery as it was before the intervention. Assuming a pre-intervention survival of 90% at 30 days and using the pooled estimates of intervention impact, the numbers of patients needed to treat to avoid one excess death at 30 days after a service model intervention are 12 and 17 for orthogeriatric and FLS type interventions, respectively. However, neither orthogeriatric nor FLS interventions evaluated here, either individually or pooled by type, had any significant impact on time to second hip fracture.

Mortality

We found the average annual proportion of primary hip fracture patients dying within 30 days and 1 year to be 9.5% and 29.8%, respectively, which is consistent with previous reports in a similar population. 112,113 The overall downwards trend in mortality after hip fracture identified here between 2003 and 2011 is also consistent with previous findings. 114

The beneficial impact on mortality following the introduction and/or expansion of an orthogeriatric model of care, as found here, is a plausible finding given our knowledge of specific details of changes to models of care identified from the pre-analysis service evaluation. For example, it was reported in Hospital 8 that the appointment of an orthogeriatric clinical lead for hip fracture care meant that 90% of patients were seen preoperatively for optimisation for surgery. Likewise, the service evaluation identified that the appointment of a second orthogeriatrician in Hospital 2 was reported to have led to hip fracture patients being taken to theatre quicker and in better condition. These are likely to be important factors, given previous evidence that trauma-related complications play a key role in post-hip fracture mortality,115 and that earlier surgery is associated with lower risk of death. 116 A recent study similarly reported that hospitals with orthogeriatric services were found to have significantly lower mortality after hip fracture than hospitals without such services,94 and findings from RCTs also demonstrate the benefits of geriatric hip fracture care on delirium117 and mortality. 118 It should also be noted that later on in Chapter 6 of this report, using national-level data from the CPRD database, there was a reduction in 30-day mortality (but not in 1-year mortality) between October 2007 and September 2008 (the publication of the BOA Blue Book and the NICE TA 161). There are hospitals (4, 5, 8 and 10) in which interventions occur at around the time of, or after publication of, these guidelines. It is possible that, in these hospitals, some of the changes made to the service models may be a reflection of the guidance and adherence to them. Our findings are in keeping with the UK NICE CGs, which recommend orthogeriatric assessment and continued review of patients admitted with hip fracture,3 and the BOA’s publication The Care of Patients with Fragility Fractures, which states that ‘senior medical input – from a consultant orthogeriatrician . . . is now essential in the good care of fragility fracture patients’. 1

The reasons for a significant decrease in mortality following the introduction and/or expansion of a FLS model of care are not as clear as for the orthogeriatric model. An environment of better co-ordination of multidisciplinary care with better communication between staff following the introduction and/or expansion of a FLS is likely to play a role. In addition, the comprehensive assessment and inspection of routine bloods and medical history that an osteoporosis nurse specialist carries out may contribute to the identification of secondary diseases and comorbidities. Furthermore, patients receiving zoledronic acid have previously been shown to be at a reduced risk of death,119 with a much greater reduction than could be attributed solely to lower fracture incidence on treatment. 120 The mechanisms that mediate the remainder of the drugs effect on mortality are not known. 120 Given that treatment recommendation is within the remit of the FLS, it may be that this is a contributory factor in the process in reduced long-term mortality. It is also likely that introducing or expanding a FLS entails wider underlying changes to service delivery that could also contribute to the reduction in morality following the implementation of change. Similar to our findings here, implementation of osteoporosis guidelines by a fracture nurse has previously been shown to be associated with a 33%101,121 reduction in post-fracture mortality following any index fragility fracture, thereby prompting the conclusion that measures to prevent fractures also reduce mortality. 101

Second hip fracture

We report here an average annual direct age- and sex-standardised proportion of 4.2% for second hip fracture within 2 years of a primary hip fracture. This is in accord with previous studies,122–124 although direct comparisons are difficult owing to different lengths of follow-up time used. We found that the regional incidence of refracture remained stable during the study period (see Figure 6), consistent with a recent analysis of trend in second hip fracture in Olmsted County, MN, USA. 8

In this context, the BOA’s publication The Care of Patients with Fragility Fractures states that the most effective health-care solution for fracture patients is the FLS, routinely delivered by a nurse specialist supported by a lead clinician in osteoporosis. 1 In a review of co-ordinator-based systems for secondary fracture prevention, Marsh et al. 10 reported that such systems appeared to be able to overcome barriers to osteoporosis intervention and treatment in fragility fracture patients. It is surprising, therefore, to find a null effect on hip refracture risk following the introduction or expansion of an orthogeriatric or a FLS model of care. One reason may be because of patients’ inadequate adherence to prescribed therapies. During the period of the study, the FLSs that were analysed focused on identification, investigation and initiation of bone therapy, leaving monitoring to primary care services. However, the adherence of patients within NHS primary care services is poor, at < 40% by 12 months. 125 This highlights the importance of incorporating monitoring within the FLS scope to reliably close the secondary prevention care gap. Compounding this issue is that the earliest demonstrated time to antiosteoporosis treatment efficacy in terms of hip fracture risk reduction is 6–36 months. 126 Specifically, zoledronic acid has been shown to reduce clinical fracture risk after hip fracture only after 12 months. 119 In the present analysis, 35% and 63% of second hip fractures within 2 years occurred within the first 6 and 12 months, respectively. This is consistent with the clinical audit of the FLS programme at Glasgow,127 and an extraordinarily high risk of early second hip fracture in a large Danish study. 128 The high rate of early refracture combined with the delayed onset of fracture risk reduction associated with therapies recommended or dispensed within the orthogeriatric and FLS models of care studied here may have contributed to our finding of no impact on refracture risk. In order to detect whether or not the interventions had an impact on longer-term refracture risk, we carried out a post-hoc analysis of second hip fracture risk, in which extended follow-up (when possible) and a time-varying hazards model were used to focus only on risk in the second and third years after a primary hip fracture. However, no significant changes in pooled estimates were found. Data on treatment initiation and adherence were not available in the HES data set and, therefore, this issue could not be further investigated here.

It has been noted that the risk of subsequent fracture after hip fracture can be offset by increased post-fracture mortality,129 which has been used elsewhere to explain increasing refracture trends during 2000–10. 130 It is possible that this dependency of fracture risk on mortality may be a contributory factor to the lack of beneficial impact on second fracture rate here observed. Given the overall ensuing improvement in survival after the introduction and/or expansion of orthogeriatric and FLS models of care, refracture risk would be expected to increase over the period of time from which excess mortality was reduced.

To our knowledge, this is the first study to evaluate orthogeriatric or FLS models of care in terms of impact on hip refracture rate after hip fracture. Our finding of no impact may seem surprising given the positive results from the Kaiser Southern California Healthy Bones Program (Kaiser SCAL) study. 96 Kaiser SCAL reported an overall 37% reduction in hip fracture rate associated with the introduction of intensive champion-led multidisciplinary osteoporosis disease management programmes across 11 medical centres. Key differences in study design between the Kaiser SCAL study and our own may account for the dissimilar findings, notably their consideration of all hip fractures rather than only refracture and that the authors estimated the exponential rise in hip fractures for 2006 from observed rates in 1997–9. Furthermore, the Kaiser SCAL programme took place in a highly integrated setting in which members received > 95% of all medical care at the centres involved in the programme.

Implementation of osteoporosis and fall prevention guidelines has also been reported to be associated with a significant reduction in subsequent fracture risk after a non-vertebral index fragility fracture. 101 Similarly, fracture patients receiving care within the intervention programme at Concord Repatriation General Hospital (Sydney, NSW, Australia) were found to be at 80% reduced risk of subsequent fracture, compared with patients electing standard care. 26 It is likely, however, that a proportion of the difference reported reflects selection bias with worse comorbidities in those who did not attend and better health-seeking behaviour in those who did (95% remained on initial treatment throughout the study period). It is also important to note that these previous analyses considered refracture after any index fragility non-vertebral fracture as opposed to our focus on hip fracture.

Strengths and limitations

A major strength of the present analysis is the use of a natural experimental design for which each hospital acted as its own control in a before-and-after impact analysis. The comprehensive nature of the HES database allowed for case-mix adjustment to estimates of intervention impact, thereby aiding transparency of such comparisons over the time course through the controlling for age, sex and Charlson Comorbidity Index. Clinical audit findings provided detailed descriptions of all major changes to service delivery of post-hip fracture care and secondary fracture prevention during the study period in each relevant hospital, enabling the timing and nature of each intervention to be described a priori. Our definition of primary hip fracture was robust and addressed the possibility of erroneously identifying readmissions as second hip fractures by only counting second hip fractures sustained in a separate CIPS and after 30 days after the primary admission.

A main limitation of our analysis is that confounding events coinciding with the interventions of interest here evaluated cannot be ruled out. We consider this is unlikely given that the estimated impact on each health outcome was consistent across hospitals and for interventions that occurred at different time points over the study period. Secular trend may also have introduced bias into impact estimates; however, this issue was addressed by excluding from analyses those estimates of impact for interventions that were preceded by a significant trend in the respective health outcome (two estimates of impact on 1-year mortality were hence excluded). Although we have pooled service model interventions according to a broad definition of either orthogeriatric or FLS models of care, further aspects of these interventions were subject to variation and were not able to be included, such as processes used to case find, DXA scanning and the undertaking of falls prevention. Although such details were not controlled for in analyses, details of each intervention are provided (see Appendices 5–8). We were also unable to compare, at the patient level, rates of diagnostic testing and medication use. It also needs to be noted that given our concentration on health outcomes after hip fracture in this evaluation, our findings may not reflect the effectiveness of models of care among patients having sustained a non-hip fragility fracture. A limitation is that routine hospital admissions data are collected for administrative rather than research purposes, and concerns have been raised over the completeness and accuracy of such data. Last, HES data underestimate the total number of admissions by excluding a minority of privately funded procedures, but this unlikely to bias the observed results.

Data sources

The data set used for this analysis was extracted from the CPRD, which comprises computerised medical records for > 8% of the UK population chosen to be representative of the wider UK population. We used linked ONS data to obtain information on mortality.

Patients

From the CPRD we identified index hip fracture cases occurring between 1 April 1999 and 31 March 2013, defined as patients with no previous hip fracture in the preceding 3 years. Patients with < 3 years from registration into a participating GP practice were, therefore, not included. A list of the Read codes used is included in Appendix 3. Patients < 50 years of age were excluded, as were patients registered in a GP practice outside England or Wales, as we had ONS data for these countries only and the NICE guidelines here evaluated were not implemented in Scotland or Northern Ireland.

Selection of control patients

Each hip fracture case was matched to two control patients without a Read code for a hip fracture on age (within 1 year), sex and GP practice and who were registered at the practice at the index date of the matched patient with hip fracture.

Interventions

The interventions here analysed were key national guidelines published within the study period and were identified prior to statistical analysis. Five specific guidelines were evaluated: NICE CG 21 (November 2004),15 NICE TA 87 (January 2005),33 BOA Blue Book (September 2007),1 NICE TA 161 (October 2008)13 and Best Practice Tariff for inpatient hip fracture care (April 2010). 76 The time points of these interventions are displayed in Figure 12.

FIGURE 12.

Time points of interest over study period.

The interventions can be divided into two broad groups: those with a sole pharmacological content and those with service delivery/falls interventions.

National Institute for Health and Care Excellence CG 2115 focused on assessment and prevention of falls in older people including after fragility fractures of the hip. A number of interventions were recommended, including individualised multifactorial interventions including strength and balance training, home hazard assessment and intervention, vision assessment and referral, and medication review with modifications/withdrawal, in addition to specific interventions, when appropriate, such as cardiac pacing, education and information giving. NICE TA 8733 focused on the clinical thresholds for the use of different pharmacological agents in postmenopausal women who have sustained a clinically apparent osteoporotic fracture. The BOA Blue Book in 20071 outlined the key standards for the care of hip fracture patients in the acute setting as well as secondary fracture prevention from both the bone health and the falls perspectives. NICE TA 16113 updated the guidance for TA 87 and included strontium ranelate. The Best Practice Tariff76 formalised the standards from the Blue Book care during the acute phase and then linked them to changes in reimbursement for trusts at the patient level.

Outcomes

Statistical analysis

Data were aggregated, separately for cases and controls, in the form of age- and sex-standardised biannual proportions of each outcome of interest. The national guidelines were evaluated using an interrupted time series analysis. 108 A segmented linear regression model was specified for each outcome. A full model was specified which included regression terms for all interventions to be analysed and a final parsimonious model was derived by way of removing non-significant regression terms (p ≥ 0.1) in a process of backward elimination. The presence of autocorrelation was tested using the Durbin–Watson test. All Durbin–Watson statistics were close to the value of two and above the lower bound; therefore, we accepted the null hypothesis of no autocorrelation.

Owing to insufficient data points between NICE CG 21 (November 2004)15 and NICE TA 87 (January 2005),33 these two interventions were evaluated as one, as were the BOA Blue Book (September 2007)1 and NICE TA 161 (October 2008)13 (see Figure 12). Aggregated data points between interventions to be evaluated as one were, therefore, excluded from analyses. In order to minimise the number of data points removed from the time series, interventions occurring within the first or last month of a 6-month period were considered to have occurred at either the beginning or the end, respectively. Aggregated data points were not included in analyses of a particular outcome if follow-up time before the end of the study period was insufficient.

Segmented linear regression results

For simplicity, the figures referred to below are from the final parsimonious segmented linear regression analyses, with the coefficients reported in Appendix 7. Results are expressed for each intervention in terms of immediate ‘step’ and/or ‘trend’ change in each aggregated outcome measure under study. Estimates from full models with forced inclusion of each intervention are also reported in Appendix 8.

Index hip fracture and refracture

The number of primary hip fractures occurring per 6 months between 1999 and 2013 are shown in Figure 18a. When the time series was compared with publication of the guidelines under evaluation, there was a significant trend-change in the number of index hip fractures occurring following the October 2004–March 2005 period wherein the NICE CG 21 and NICE TA 87 were published. Prior to this period the biannual proportions were stable, whereas a post-intervention trend of –7.17 (95% CI –8.75 to –5.6; p < 0.001) index hip fractures per 6 months was detected (see Figure 18a). This association was unchanged for aggregated proportions of overall hip fractures (see Figure 18b).

FIGURE 18.

(a) Number (1999/2000–2012/13) of primary hip fractures; and (b) overall number of primary and secondary hip fractures.

Considering refracture, an initial stable rate of 2.49% (95% CI 2.12% to 2.87%) was estimated for subsequent hip fracture, although a step-change reduction of –0.95% (95% CI –1.67 to –0.23; p = 0.012) between October 2007 and September 2008 (publication of the BOA Blue Book1 and NICE TA 16113) was found (Figure 19a). This reduction in refracture at the hip was not observed for subsequent major non-hip fracture (see Figure 19b). There was no evidence associated with any of the other guidelines evaluated for a change in level or trend in biannual proportions (see Figure 19).

FIGURE 19.

(a) Post-index date second hip fracture: cases (black circle); and (b) post-index date major non-hip fracture: cases (black circle) and controls (green triangle).

Mortality

In relation to the guidelines evaluated here, a significant step-change reduction in 30-day mortality occurred of –2.81% (95% CI –3.73 to –1.85; p ≤ 0.001) between October 2007 and September 2008 (publication of the BOA Blue Book1 and NICE TA 16113) (Figure 20a). Although this step change was not reflected in 1-year mortality following the same publications, a significant reduction in 1-year mortality of –5.56% (95% CI –7.59 to –3.52; p < 0.001) was seen immediately following the introduction of the Best Practice Tariff in April 201076 (Figure 21a). No other step change or trends were found in 30-day or 1-year mortality. When a ‘difference in differences’ analysis was carried out for mortality between cases and controls, the significant reduction in 30-day mortality remained (see Figure 20b), as it did for 1-year mortality (see Figure 21b); however, for the difference in 1-year mortality between cases and controls, there was a trend increase following the October 2007 to September 2008 period (the publication of the BOA Blue Book1 and the NICE TA 16113) of 0.98% (95% CI 0.16 to 1.8; p = 0.022) in biannual proportions that persisted throughout the rest of the study period (see Figure 21b).

FIGURE 20.

(a) Post-index date mortality within 30 days among cases (black circle) and controls (green triangle); and (b) post-index date 30-day mortality: difference in differences (black circle) between cases (blue diamond) and controls (green triangle).

FIGURE 21.

(a) Post-index date mortality within 1 year among cases (black circle) and controls (green triangle); and (b) post-index date 1-year mortality: difference in differences (black circle) between cases (blue diamond) and controls (green triangle).

Bone-strengthening drugs

The time series of biannual proportions of treatment-naive index hip fracture patients receiving an incident prescription for a bone-strengthening drug in the first year following hip fracture is presented in Figure 20b. This shows an initial upwards trend in the proportion receiving such a prescription of 1.1% per 6 months, with a marked step change of 14.5% (95% CI 11.1% to 17.8%; p ≤ 0.001) taking place between pre-publication of NICE CG 21 (November 2004)15 and post-publication of the NICE TA 87 (January 2005). 33 These publications were also associated with a small increase (0.49%, 95% CI –0.05% to 1.03%; p = 0.073) in the prior upwards trend (see Figure 20b). Similar estimates in both a trend and a step change following these publications were also found in the time series of antiosteoporosis medication initiation in the first 4 months after index hip fracture date (Figure 22a).

FIGURE 22.

Any antiosteoporosis medication post index date among cases (black circle) and controls (green triangle) within (a) 4 months; and (b) 12 months.

In analyses of the biannual proportions of patients with at least one bisphosphonate prescription at 10–14 months following index hip fracture, an overall trend increase of 0.96% (95% CI 0.62% to 1.27%; p < 0.001) per 6 months was detected. In addition, a step-change increase of 8.71% (95% CI 5.04% to 12.4%; p < 0.001) was observed between the pre-publication of NICE CG 21 (November 2004)15 and post-publication of the NICE TA 87 (January 2005)33 (Figure 23). However, a modest step-change decrease of –3.79% (95% CI –7.4% to –0.17%; p = 0.041) was found to occur between October 2007 and September 2007 (publication of the BOA Blue Book1 and NICE TA 16113) (see Figure 23).

FIGURE 23.

Bisphosphonate prescription 10–14 months post index date among treatment-naive individuals at baseline among cases (black circle) and controls (green triangle).

When analyses of any antiosteoporosis medication within 1 year were stratified by sex, the step-change increase associated with the time period October 2004–March 2005 was seen among both men (9.07%, 95% CI 3.68% to 14.5%) and women (15.2%, 95% CI 11.1% to 19.3%) (see Appendix 9). Among men, however, this time period was also associated with a trend increase of 1.74% (95% CI 0.94% to 2.53%; p < 0.001) per 6 months that was then followed by a step decrease of –7.52% (95% CI –14.1% to –0.95%; p = 0.027) following the October 2007–September 2008 time period (see Appendix 9). Coefficients from full models stratified by sex are also reported in Appendix 10.

A summary of the associations between the guidelines and changes in outcomes from the interrupted time series models is shown in Table 16.

Setting and data sources

We used two sets of data sources to estimate the costs associated with a hip fracture. For hospital costs, we used the HES database and for primary costs we used CPRD GOLD (see Chapter 4 for description of data sources). We adopted the same incidence-based approach to identify hip fracture patients in both sets of data and estimate the costs of hip fracture.

Study participants

To be consistent, we used the same approach to identify patients with a hip fracture in the HES and CPRD data sets. We searched the hospital finished consultant episodes for patients > 60 years of age who had had an emergency hospital admission with a primary ICD-10 diagnosis code for hip fracture (S72.0–S72.2, S72.9) between April 2003 and March 2013 (HES data set) and April 2003 and March 2012 (CPRD data set). We extracted all primary (CPRD) and hospital (HES) records before and after that admission. A number of exclusion criteria were applied to minimise misclassification: (1) day cases were excluded by imposing a condition that patients had to stay at least one night in hospital, unless death occurred in the first 24 hours of admission; (2) individuals who had had a previous hip fracture between April 1999 and March 2003 were excluded to reduce duplicate coding of hip fractures that occurred before the period of analysis but led to repeat hospital admissions due to complications or unresolved sequelae; and (3) patients were also excluded if they had had a hip fracture due to trauma, such as transport accidents, identified using ICD-10 codes (V01–V99).

When estimating hospital costs in the year before and after fracture we included only patients with an index admission after 1 April 2008, to ensure that outpatient and emergency attendances costs would be included. We refer to this set of results as ‘total hospital’ costs and contacts. Conversely, we used the whole HES data set (April 2003–April 2013) to report costs due to hospitalisation, critical care and day cases, and benefit from the increased statistical power. We refer to these results as ‘hospitalisation’ costs. When estimating primary care costs, we included only patients who were registered with a GP at the time of index hip fracture admission.

Second hip fractures were identified using the same approach as for the index fracture. To ensure that they were separate fractures and not hospital readmissions resulting from adverse effects of the index fracture, we counted second hip fractures only if admitted in a separate CIPS from index admission and at least 30 days after admission for the primary fracture. A CIPS is made up by one or more hospital spells (i.e. time patient stays in one hospital) and is defined as a continuous period of care within the NHS, regardless of any transfers to another hospital. A hospital spell starts with the index admission, involves treatment by one or more consultants (i.e. finished consultant episodes) and ends when the patient dies or is discharged from hospital.

Primary care costs

Primary care contacts included GP consultations in clinic/surgery, telephone contacts and out-of-office visits. It also included nurse face-to-face and non-face-to-face contacts and contacts with other community health-care professionals (e.g. health visitor or physiotherapist). GP and nurse consultations excluded repeat prescriptions for which the patient was not seen, notes and reports and laboratory/radiology requests and results. Following previous research,141 GPs were identified using the following codes: senior partner; partner; associate; non-commercial local rota of less than 10 GPs; commercial deputising service; GP registrar; sole practitioner; and GP retainer. Nurses were identified using practice nurse; community-based nurse; hospital nurse; school nurse; and other nursing and midwifery. CPRD records listing administrative staff, such as secretaries, IT staff, practice and fund managers and receptionists, were not counted as a clinical direct contact with the patient and were, therefore, excluded from the costs. Following a previous study,142 we only counted one consultation per day if more than one was recorded per patient. Primary care contacts and tests were costed using unit costs from national cost databases (Table 17). 143 We excluded from our costs tests that are routinely performed as part of a primary care consultation, such as blood pressure measurement, to avoid double counting. Pharmaceuticals were costed by matching each prescribed medication to a BNF code, moving from the most detailed level (subparagraph) to the top level (chapter) until a match was found. 90 The number of medications per patient stratified by BNF code were then multiplied by the respective unit costs. The unit costs for each BNF code concerned the net ingredient cost per item prescribed reported in the Health and Social Care Information Centre Prescription Cost Analysis. 145 These were estimated using the average for each BNF level (from subparagraph to chapter) using the number of items prescribed as weights. Primary care costs were computed by multiplying the number of contacts/test/prescribed items by their unit costs. Costs per patient were then summed across these different resource categories and aggregated into monthly and annual amounts for the purposes of the analysis.

Hospital costs

Each finished consultant episode in a hospital spell was assigned into a 2012–13 HRG using standard software. HRGs consist of standard groups of clinically similar treatments that consume a common set of health-care resources. All resource use was valued using 2012–13 prices that were obtained from the schedule of reference costs for NHS trusts. 144 Total costs per patient were aggregated into monthly and annual amounts for the purposes of the analysis.

Statistical analysis

We determined the marginal costs attributable to a hip fracture by comparing costs in the year before and after the index hip fracture. The national total annual primary and hospital costs of hip fracture were determined by multiplying the incidence of hip fracture in the UK (79,243 in 2010) by the estimated costs per hip fracture. 135

The HES database was censored in 31 March 2013, and complete follow-up was not available for all cases. Hence we report total hospital costs for those patients with complete follow-up data at years 1 and 2 following hip fracture and for the whole sample after adjusting for censoring using the methodology developed by Lin. 146 Costs are reported as means together with their 95% CIs, obtained from 1000 bootstrap estimates.

Predictors of primary care and hospitalisation costs of hip fracture were estimated using a generalised linear model (GLM). After reviewing the literature, we examined the following predictors of costs in the year of the hip fracture: age at fracture; type of fracture (head and neck S72.0; pertrochanteric S72.1; subtrochanteric S72.2; unspecified S72.9); sex; Charlson Comorbidity Index score (complications at fracture and occurring up to 3 years before fracture);147 place of residence pre and post fracture (own home or care home: residential or nursing home); occurrence of second hip fracture; history of other osteoporatic major fragility fractures requiring hospitalisation pre and post hip fracture (spine, wrist, pelvis, rib, humerus and other identified with ICD-10104 diagnosis codes S22, S32, S42, S52.0–S52.3, S22.5 and S22.6); primary hip replacement and revision;148 complications of internal orthopaedic devices (ICD-10 code T84) and infection or haemorrhage following procedure (T81.0 and T81.4); dislocation (M24.3 and M24.4); malunion and non-union of fracture (M84.0–M84.2); periprosthetic fracture (M96.6); other/unspecified postprocedural musculoskeletal disorders (M96.8, M96.9); sequelae of fractures of the femur (T93.1); hip luxations (S73.0); ethnicity (white and non-white); year of hip fracture; and income deprivation measured by the IMD. We assumed that once a patient moved to a care home they would remain there for the rest of their life. We included only covariates that had a frequency of at least 100 patients in the samples available. Selection of covariates for the final model included the type of hip fracture and other covariates from the list specified that were found to be significant and met the inclusion criteria. For hospitalisation costs, a variable was deemed to be statistically significant if p < 0.01 to account for the large sample size and 95% CIs were reported for ease of comparison with other studies. For primary care costs, a variable was deemed to be statistically significant if p < 0.05. The choice of the GLM model family and link functions was informed by the modified Park test and the Box–Cox test, respectively. Model fit was assessed using Pregibon link test and different family and link functions were compared using Akaike’s information criterion. Univariate analyses in continuous variables were performed using Student’s t-tests. All analyses were performed using Stata version 12 (StataCorp, College Station, TX, USA).

Patient sample

Patient outcomes and hospitalisation costs (Hospital Episode Statistics data set)

The average follow-up of the cohort was 2.6 years (median 1.8 years; SD 2.5 years) from index hip fracture, during which time 6.6% of patients suffered a second hip fracture (Table 20). Mortality at 30 days and 1 year was estimated to be 9.4% and 31.2%, respectively. After index fracture, the majority of patients were recorded as being discharged to their own home (49%), or transferred to another hospital (23%) or to a care home (18%). Ten per cent of patients died in hospital during the index admission. The number of patients in a care home increased to 24% after 1 year of complete follow-up. In contrast, after a second hip fracture, 32% of patients were discharged to a care home after hospital discharge, which increased to 40% among patients with at least 1 year of follow-up. Hospital readmissions for any reason with inpatient stay in the year following the index hip fracture totalled 0.9 [median 0, SD 1.4, interquartile range (IQR) 0–19] per patient or 1.8 (median 1, SD 1.6, IQR 1–19) per patient readmitted. Of these readmissions, 57% were emergency admissions and accounted for 18.1 (median 0, SD 35.9, IQR 0–22) additional days in hospital per patient or 37.2 (median 23, SD 43.9, IQR 8–50) additional days per patient readmitted. About 50% of diagnosis codes in non-emergency inpatient admissions following index admission were hip fracture related.

The hospitalisation costs associated with index admission for primary hip fracture were £8663 (median £8049, SD £4605) compared with £8544 (median £8049, SD £4112) for the second hip fracture. Length of stays in the index admission were 20.5 (median 14, SD 20.0, IQR 9–307) and 20.8 (median 15, SD 18.8, IQR 9–139) days for primary and second hip fracture, respectively. For patients suffering a subsequent hip fracture (6.6%), the hospital admission following the second hip fracture resulted in significantly higher length of stay (2.0 days; p < 0.001; 18.8 vs. 20.8 days) and costs (£406; p < 0.001; £8138 vs. £8544) relative to the index fracture. Within the first year following primary hip fracture, the total hospitalisation costs were estimated to be £13,826 (median £10,425, SD £11,016), of which 75% were because of hip fracture-related admissions (£10,375, median £8050). The distribution of hospitalisation costs was skewed (Figure 25) with a small proportion of cases staying in hospital for a whole year resulting in very high costs (36 patients had hospitalisation costs above £100,000). Hospitalisation costs and length of stay were highly correlated (Spearman’s correlation coefficient: 0.82; p < 0.001).

FIGURE 25.

Distribution of hospitalisation costs in the year after primary hip fracture.

Figure 26 reports the hospitalisation costs in the months before and after primary hip fracture. The annual cost in the year of the fracture was estimated to be £10,860 (95% CI £10,710 to £11,011) higher than the previous year. The highest costs occur in the first 6 months post hip fracture dropping sharply afterwards to pre-fracture levels of expenditure and remaining fairly constant throughout the second year post fracture. The 2-year survivors show a similar pattern of costs relative to all patients. However, while the costs in the second year after hip fracture remain numerically higher than in the year pre-fracture (£112, 95% CI –£29 to £274) this was not statistically significant.

FIGURE 26.

Hospitalisation costs in the months before and after primary hip fracture. a, Complete cases, including those who died in that year.Reproduced with permission from Leal J, Gray AM, Prieto-Alhambra D, Arden NK, Cooper C, Javaid MK, et al. Impact of hip fracture on hospital care costs: a population based study. Osteoporos Int 2016;27:549–58, under a Creative Commons licence (https://creativecommons.org/licenses/by-nc/4.0/legalcode).

Patient outcomes and primary care costs (Clinical Practice Research Datalink data set)

The average follow up of the cohort was 2.5 (median 1.9, SD 2.3) years from index hip fracture, during which time 7.2% of patients suffered a second hip fracture (Table 21). Mortality at 30 days and 1 year was estimated to be 5.7% and 26.3%, respectively; these rates were lower than the mortality rates found in the hospital care data set. After index fracture, the majority of patients were recorded as being discharged to their own home (55%), or transferred to another hospital (19%) or to a care home (19%). The number of patients in a care home increased to 23% after 1 year of complete follow-up. GP contacts in the year of the hip fracture amounted to 8.3 (median 6, SD 9.3) per person, of which 77% were clinic or surgery appointments and the rest were telephone contacts and out-of-office visits. In the same year, 67 (median 43, SD 86.6) medications were prescribed per patient including non-medication items (e.g. wound management bandages). Excluding non-pharmaceutical items, the most common type of medications were for the cardiovascular system (BNF chapter 2), accounting for 33% of all medications, and the central nervous system (BNF chapter 4), accounting for 19%, followed by gastrointestinal (BNF chapter 1) and nutrition and blood systems (BNF chapter 9) accounting for 12% each, and endocrine system (BNF chapter 6) accounting for 10%. 90

The primary care costs associated with primary hip fracture were £1065 (median £660, SD £1798), of which medications and non-pharmaceuticals accounted for £614 (median £248, SD £1586) of the costs and GP contacts accounted for £358 (median £246, SD £409).

Figure 27 reports the primary care costs in the months before and after primary hip fracture. Similar to hospital care, there is an increase in costs before hip fracture, with the highest costs occurring in the first 6 months post hip fracture. Nonetheless, although the costs for all patients in the first year after hip fracture were numerically lower than the year pre fracture (£26, 95% CI –£102 to £50), this was not statistically significant. However, when we considered only the 2-year survivors, we found the costs in the first and second year post hip fracture to be significantly higher than the costs in the year prior to hip fracture. Compared with the year prior to the hip fracture, primary care costs, in patients who were alive 2 years post hip fracture, were £256 (95% CI £160 to £352) and £273 (95% CI £167 to £380) higher in the first and second year following hip fracture, respectively. This was mostly led by a significant increase in GP contacts and in the costs of medications and non-pharmaceuticals.

FIGURE 27.

Primary care costs in the months before and after primary hip fracture. Two-year survivors concerns patients who did not die within 2 years post hip fracture and follow-up data were available for this period. a, Complete cases, including those who died in that year.

Total hospital costs before and after index fracture

Patients who had a primary hip fracture after 1 April 2008 and complete 1-year follow-up (44%, n = 14,552) were used to compare hospital resource use and total costs in the years before and after the fracture (Table 22). In the year of the hip fracture, patients had 1.03 additional hospital admissions (i.e. inpatient stay, day cases and regular day/night attenders) (p < 0.001), 27.9 additional hospital inpatient days (p < 0.001), 0.54 additional A&E contacts (p < 0.001) and 0.01 additional outpatient visits (p < 0.001), compared with the previous year.

Including both outpatient and emergency contacts, the total costs were estimated at £14,264 (95% CI £14,092 to £14,436) in the year of the fracture, of which 96% (£13,635) was because of inpatient stay and critical care. Unadjusted for other covariates, men had significantly higher total hospital costs than women (£1188; p < 0.001). Having a hip fracture resulted in additional costs of £10,964 (95% CI £10,767 to £11,161), compared with the year prior to hip fracture. Adjusting for censoring, the 1-year costs were similar to the complete-case analysis at £14,163 (95% CI £14,008 to £14,317). The costs in the first 2 years following hip fracture (2 years) adjusted for censoring were £16,302 (95% CI £16,097 to £16,515), compared with £16,270 using only the complete cases (n = 12,155).

Primary care and hospital care costs 1 year and 2 years post index fracture

The primary care and hospital care costs in the first year and second year (conditional on surviving the first year) post hip fracture per sex and age group are reported in Table 23. The total primary care and hospital care costs in the year of the hip fracture were £15,329, of which £14,264 was a result of hospital care and £1065 was a result of primary care. Conditional on surviving the first year, the total primary care and hospital care costs in the second year after the hip fracture were £4242, of which £3072 was a result of hospital care and £1170 was a result of primary care.

Annual costs of hip fractures in the UK

The total annual costs associated with all incident hip fractures (in the year of hip fracture) in the UK among those aged ≥ 50 years (n = 79,243) were estimated at £1215M (Table 24).

Predictors of primary care costs in first year following hip fracture

The predictors of hospitalisation costs are shown in Table 25. A GLM with gamma family and identify link function had the best fit. Adjusting for all covariates, there were no statistically significant differences in primary care costs between the types of hip fractures. Costs were inversely associated with age (–£15 per additional year) and positively associated with multiple deprivation, with costs decreasing with more deprived patients. A higher Charlson Comorbidity Index score at index hip fracture increased primary costs by approximately £131 per additional unit of the Charlson Comorbidity Index score.

Predictors of hospitalisation costs in first year following hip fracture

The predictors of hospitalisation costs are shown in Table 26. A GLM with gamma family and identify link function had the best fit. Adjusting for all covariates, men had higher hospitalisation costs than women (£910, 95% CI £679 to £1141) and higher length of stay (4.5 days). Furthermore, costs were positively associated with age (£45 per additional year) and inversely associated with income, with costs rising with income deprivation. A higher Charlson Comorbidity Index score at index hip fracture increased hospitalisation costs by approximately £694 per additional unit of the Charlson score. Transiting to a care home for the first time after index fracture was associated with higher hospital costs (£5583, 95% CI £5197 to £5970) and a longer length of stay (22 days) relative to patients who went back to their previous accommodation, possibly indicating poorer health.

Holding all else constant, having a second hip fracture within the same year as the primary one was associated with an additional £9198 (95% CI £8059 to £10,337) of expenditure in hospitalisation costs. Major (non-hip) fragility fractures requiring hospitalisation post hip fracture were also found to be significantly associated with higher hospitalisation costs (£5705, 95% CI £4434 to £6975). Among hip fracture-related complications, surgical complications within and after index admission were the most frequently reported and were associated with higher costs and length of stay (22 and 25 days, respectively) relative to no complications. Periprosthetic fracture was also associated with significantly higher hospitalisation costs relative to patients without this, at £9569 (95% CI £6302 to £12,835). All of these cost differences remained significant after adjusting for length of stay.

Finally, patients who died within 30 days were associated with lower hospitalisation costs (–£4672, 95% CI –£4906 to –£4437), mostly as a result of lower length of stay, than those who survived the first 30 days (mean of 11.3 days vs. 41.9 days). Patients who died after 30 days had higher costs (£2549, 95% CI £2246 to £2853) than the survivors, again mostly because of a longer length of stay (mean of 53.2 days vs. 38.3 days).

Interventions under study

We estimated the cost-effectiveness of three models of secondary fracture prevention for all patients with a hip fracture admitted to a NHS hospital:

1. introduction of an orthogeriatrician model of post-hip fracture care

2. introduction of a fracture liaison nurse model of post-hip fracture care

3. usual post-hip fracture care (no orthogeriatrician or fracture liaison nurse).

In addition, the cost-effectiveness of these models of care was assessed in patient subgroups in terms of age at hip fracture (60, 70, 80 and 90 years), sex (male or female) and Charlson Comorbidity Index before hip fracture.

Model structure

A decision-analytic model was developed to evaluate the costs, (quality-adjusted) life expectancy and cost-effectiveness of the different models of secondary care hip fracture prevention under evaluation. Given the natural history of hip fracture progression with recursive events, the most appropriate type of model was judged to be a Markov model which was developed in Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA). The model was used to simulate the natural history of the hip fracture population across health states representing the history of hip fracture and major non-hip fractures and hospital discharge to the patient’s own home or to a care home. Hence the following health states were defined (Figures 28 and 29):

1. death within 30 days following first or second hip fracture admission

2. second hip fracture

3. other major fragility non-hip fractures requiring hospitalisation

4. history of primary hip fracture

5. history of major fragility non-hip fractures requiring hospitalisation

6. history of second hip fracture

7. history of second hip fracture and major fragility non-hip fractures

8. death within year.

FIGURE 28.

Model structure and health states in the first year of simulation. (a) Discharge to care home; and (b) discharge to own home. Reproduced with permission from Leal J, Gray AM, Hawley S, Prieto-Alhambra D, Delmestri A, Arden NK, et al. Cost-effectiveness of orthogeriatric and fracture liaison service models of care for hip fracture patients: a population based study. J Bone Miner Res 2016; in press.

FIGURE 29.

Model structure and health states in the years following the first hip fracture. (a) Discharge to care home; and (b) discharge to own home.

The model had three absorbing states: history of second hip fracture and major fragility non-hip fractures (progressing only to death after reaching this state), death within 30 days following hip fracture and death within year. We assumed that if a patient transited to a care home they would remain there for the rest of their lifetime.

The model structure and assumptions were informed by the hip fracture, the needs of the decision problem and discussions with clinical experts, health economists, statisticians and epidemiologists involved in the project. This meant that an iterative process was used to define the model structure when the agreed conceptual framework was revisited given the results of the data analysis and new findings in the published literature.

The time horizon of the analysis was lifetime and the population moved between health states according to defined transition probabilities. A cycle length of 1 year was considered appropriate given the natural history of hip fracture patients. Half-cycle correction was performed using the lifetable approach. 153

The model simulated the transition of a cohort of patients with an index hip fracture through the health states over time, to estimate expected costs and outcomes (see Figures 28 and 29). For example, at the start of the simulation, patients with a hip fracture could die within 30 days or be discharged home or to a care home (nursing or residential care home). In the same cycle, patients could then develop a second hip fracture, other major fragility fracture requiring hospitalisation (non-hip such as pelvic, spine, wrist, humerus and rib), have no further events or die (see Figure 28). If patients experienced a second hip fracture, they could die within 30 days or, if alive, be discharged to a care home or their own home. In the next cycles, patients were allowed to continue progressing if they had not yet reached an absorbing state (see Figure 29).

Finally, the model adopted the perspective of the NHS and Personal Social Services. All costs and effects were discounted beyond the first year of simulation using an annual discount rate of 3.5%, based on current UK government recommendations. The price year was 2012–13 and, when necessary, costs were adjusted to that year using the UK health sector pay and prices inflation factor.

Data sources

The data sources used to inform the model inputs consisted of HES database, CPRD GOLD and published literature.

Model inputs

Quality of life in patients with hip fracture

A literature search was conducted to identify studies reporting preference-based quality of life for patients with hip fracture [e.g. European Quality of Life-5 Dimensions (EQ-5D), EQ Visual Analogue Scale, Health Utilities Index 2, Health Utilities Index 3, Short Form questionnaire-6 Dimensions, standard gamble, time trade-off, etc.]. The following databases were searched for studies published between 1 January 1990 and 1 April 2014: EconLit, EMBASE, Global Health, MEDLINE, NHS EED and HTA and Web of Science. Literature reviews and reference lists of the papers found in the literature review were also searched for additional studies. Data were extracted from each study using a pro forma.

We identified 64 studies reporting preference-based quality of life data in populations with hip fracture (Figure 30). We synthesised data from studies reporting absolute utility values at different follow-up times using a metaregression approach. These corresponded to 32 populations of hip fracture patients, 187 observations and 21,085 hip fracture patients. Each observation was weighted by the respective sample size divided by the variance of the mean utility value and synthesised using a linear mixed-effects model. The following predictors were examined: follow-up time (months), follow-up time to the power of two, and the use of the EQ-5D instrument to elicit the utilities. The resulting model was used to predict the EQ-5D utility values of hip fracture patients at the following time points: onset of the fracture (0.44 at 0 months), 6 months (0.52) and 18 months (0.65) (Table 35). We assumed changes in the mean utility values between onset of hip fracture and 1 year and between 1 year and 2 years post hip fracture to be straight line transitions and hence used the predict utility values at the mid-point time periods. We assumed the utility value for the hip fracture population to remain constant after the first year post hip fracture (i.e. 0.65). Finally, we assumed second hip fracture and major non-hip fractures requiring hospitalisation to be associated with the same utility values in the year of the event as those at the onset of hip fracture (0.44). The most common non-hip fractures requiring hospitalisation were pelvic fractures.

FIGURE 30.

Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) diagram for literature review of preference-based quality of life studies in hip fracture populations.

TABLE 35 - Utility values for hip fracture patients

Linear mixed-effects model
Number of observations	187
Number of groups	32
Parameters	Mean	SE
Follow-up time (months)	0.0180	0.0021
Follow-up time (months2)	–0.0003	0.00006
EQ-5D	–0.1912	0.0881
Constant	0.6187	0.0776
Random effects (SD)	0.0000
Probability > χ2	0.0000
Utility predictions	Mean	95% CI
Utility at onset of hip fracture	0.44	0.22 to 0.66
Utility at 6 months	0.52	0.30 to 0.74
Utility at 18 months	0.65	0.40 to 0.89

Analysis

A model of care was deemed to be cost-effective if the ICER was below £30,000 per QALY gained. The ICER was estimated by dividing the difference in mean costs (ΔC) by the difference in mean effects (ΔE) (life-years and QALYs) for a given model of care compared with its next best alternative. As more than two models of care were compared, to facilitate the identification of the most cost-effective option we used the net benefit framework. 157 This consists of estimating the net monetary benefit (NMB) of each model of care:

where E and C represent the mean effects (QALYs and life-years) and costs, respectively, for a given model of care and λ represents the maximum willingness to pay per unit of effect (i.e. £30,000 per QALY gained). The most cost-effective model of care is identified as being the one with the highest NMB. Finally, the internal validity of the model was checked using sensitivity analysis (extreme values) and by comparing the model outputs with the data used to build the model.

Sensitivity and uncertainty analysis

Model parameters and structural assumptions were evaluated in one-way and probabilistic sensitivity analysis. The key uncertainties in the model structure were identified during the discussions for the conceptual framework. The distributions for the regression coefficients informing the several models described were obtained by bootstrapping the sample and re-estimating the regression models. This ensured the correlation between coefficients to be fully captured. The choice of distributions used for the remaining parameters was made according to recommended practice. 158 Relative effectiveness measures (i.e. HRs) were modelled using a log-normal distribution. Parameters concerning proportions/probabilities were modelled using beta distributions.

A cost-effectiveness acceptability curve was constructed159 using the NMB results and analysis of covariance methods were used to determine the proportion of variance in the incremental costs and QALYs saved explained by parameter uncertainty. 158 Finally, the overall contribution of the model inputs to the decision uncertainty was explored using the expected value of perfect information (EVPI). The EVPI per patient was estimated non-parametrically. 158

The EVPI for the total population who stand to benefit from reducing the decision uncertainty was also estimated. This required information on the period over which information about the decision will be useful, T (5-, 10- and 15-year scenarios), and the number of hip fractures in England, Pt,

where Pt was 70,000 in England and the discount rate used, r, was 3.5%.

Representative patient

Two cohorts of 1000 identical men and women were used to simulate a representative patient aged 83 years at hip fracture, with an average pre-admission Charlson Comorbidity Index score of 1.2 and living in their own home before the fracture.

For our male cohort, the introduction of an orthogeriatrician and a fracture liaison nurse would result in a reduction of 26 (95% CI 17 to 34) and 19 (95% CI 9 to 28) deaths within 30 days of primary hip fracture, respectively, compared with usual care (Table 36). Within 1 year of primary hip fracture, the reduction in deaths by introducing an orthogeriatrician and a fracture liaison nurse, compared with usual care, was 58 (95% CI 42 to 71) and 46 (95% CI 27 to 63), respectively. Over the lifetime of the cohort, when compared with usual care, there would be an increase of 0.18 (95% CI 0.14 to 0.22) and 0.14 (95% CI 0.09 to 0.19) life-years (undiscounted) spent in their own home if an orthogeriatrician or a fracture liaison nurse were to be introduced, respectively.

For our female cohort, the introduction of an orthogeriatrician and a fracture liaison nurse would result in a reduction of 16 (95% CI 11 to 22) and 12 (95% CI 5 to 18) deaths within 30 days of primary hip fracture, respectively, compared with usual care (Table 37). Within 1 year of primary hip fracture, the reduction in deaths by introducing an orthogeriatrician and a fracture liaison nurse, compared with usual care, was 42 (95% CI 31 to 52) and 33 (95% CI 20 to 46), respectively. Over the lifetime of the cohort, when compared with usual care, there would be an increase of 0.17 (95% CI 0.13 to 0.22) and 0.13 (95% CI 0.08 to 0.19) life-years (undiscounted) spent in their own home if an orthogeriatrician or a fracture liaison nurse were to be introduced, respectively.

Owing to better survival, the numbers of second hip fractures and major non-hip fractures were higher with the introduction of an orthogeriatrician or a fracture liaison nurse, compared with usual care, in both the female and the male cohorts (see Tables 36 and 37).

The average discounted hospital cost per male patient was £23,825 with an orthogeriatrician, compared with £23,690 and £23,037 with a fracture liaison nurse and usual care. Mean primary care costs associated with an orthogeriatrician were also higher, £3523, than a fracture liaison nurse and usual care (£3469 and £3275, respectively). Mean care home costs were higher if an orthogeriatrician was introduced (£13,946) compared with having a fracture liaison nurse (£13,710) or usual care (£12,789). Combining all health and social care costs included in the model, mean discounted costs were £41,714 when an orthogeriatrician was introduced, £41,068 when a fracture liaison nurse was introduced and £39,101 for standard care (Table 38). The discounted average QALYs gained by male patients were 1.74 with an orthogeriatrician, 1.72 with a fracture liaison nurse and 1.62 with usual care.

For our female cohort, the average discounted hospital cost per patient was £24,463 with an orthogeriatrician, compared with £24,372 and £23,881 with a fracture liaison nurse and usual care. As with male patients, the average primary care costs associated with an orthogeriatrician were higher, £4947, than those associated with a fracture liaison nurse and usual care (£4895 and £4716, respectively). Mean care home costs were considerably higher than for male patients and introducing an orthogeriatrician was associated with highest costs (£23,274) compared with a fracture liaison nurse (£23,006) or usual care (£21,976). Combining all health and social care costs included in the model, mean discounted costs were £53,104 when an orthogeriatrician is introduced, £52,472 when a fracture liaison nurse is introduced and £50,573 if usual care is provided (see Table 38). The discounted average QALYs gained by female patients were 2.50 with an orthogeriatrician, 2.48 with a fracture liaison nurse and 2.38 with usual care.

After combining costs and outcomes in an incremental cost-effectiveness analysis, and at a £30,000 per QALY threshold, the most cost-effective model of care was introducing an orthogeriatrician (Table 39). The probability of adding an orthogeriatrician being the most cost-effective option at £30,000/QALY was estimated to about 70% across both sexes.

Figure 31 reports the cost-effectiveness acceptability curve and EVPI per female patient associated with the different models of care. The population EVPI over 5 years was estimated to be £23M at the £30,000 per QALY gained threshold. This suggests that undertaking additional major commissioned research work to further reduce decision uncertainty is likely to be of significant benefit.

FIGURE 31.

Cost-effectiveness acceptability curve and EVPI per female patient. The curves provide the probability of the models of care being the most cost-effective option at any willingness-to-pay value for an additional QALY gained. Please note that the curves of the models of care add to one at any given point. The EVPI curve provides the value of EVPI at any willingness-to-pay value for an additional QALY gained. OG, orthogeriatrician.

Figure 32 reports the parameters according to their impact on the variance of incremental costs and QALYs from the comparison of introducing an orthogeriatrician with introducing a fracture liaison nurse. Variance in the estimated incremental costs was dominated by the variation in the following variables: relative effectiveness inputs, regression models predicting hospital costs, natural history inputs (e.g. death within 30 days, time to second hip fracture), and the likelihood of being discharge to a care home following hip fracture. Variance in the incremental QALYs was dominated by the variance in effectiveness inputs and natural history inputs.

FIGURE 32.

Orthogeriatrician vs. fracture liaison nurse: analysis of covariance analysis of proportion of sum of squares for incremental QALYs saved and incremental costs explained by the uncertainty in the model. The horizontal axis represents the variation in incremental costs and QALYs that is associated with the uncertainty in the model inputs. Reproduced with permission from Leal J, Gray AM, Hawley S, Prieto-Alhambra D, Delmestri A, Arden NK, et al. Cost-effectiveness of orthogeriatric and fracture liaison service models of care for hip fracture patients: a population based study. J Bone Miner Res 2016; in press.

Sensitivity analysis

Table 40 reports the results of a range of other sensitivity analyses using the female cohort. Overall, the results were robust to changes in the majority of assumptions. Excluding social care costs from the analysis resulted in the model of care with an orthogeriatrician becoming cost-effective below the £20,000 per QALY threshold. The results were also robust to using relative effectiveness measures concerning the comparison of orthogeriatrician or fracture liaison nurse relative to hospitals that had no previous orthogeriatrician or fracture liaison nurse rather than the pooled estimates. Finally, the assumptions regarding the costs of the models of care had an impact in terms of the results. Using the number of hip fractures reported in the smallest hospital in the HES data set (220 per year) to estimate the costs per patient of the models of care resulted in introducing a fracture liaison nurse becoming the most cost-effective option.

Subgroup analysis

We further assessed the cost-effectiveness of the models of care in several subgroups of hip fracture patients defined according to their age, sex and Charlson Comorbidity Index score at index hip fracture (Table 41). For patients up to the age of 80 years, the model of care with an orthogeriatrician was the most cost-effective option. For patients aged 90 years, introducing a fracture liaison nurse became the most cost-effective option if the Charlson Comorbidity Index score at index hip fracture was 5.

Osteoporosis assessment

Only one of the hospitals studied had a DXA scanner available on site so they could perform the DXA scan while in an inpatient setting. At every other hospital patients had to travel to a different location for their DXA scan at a later date. In addition, some hospitals conducted the osteoporosis assessment at an outpatient appointment, too, so this could require patients to attend two separate outpatient appointments. There were mixed feelings about the impact of patients needing to attend extra outpatient appointments, with some feeling that this was not an excuse for missing appointments and that patients need to take more responsibility for their own care. They did agree that this would be more problematic for older and frailer people relying on public transport, and that DXA scans and consultant appointments should be offered together. They felt that mobile DXA scans were an excellent idea if the scanner was not available at the local hospital.

Treatment initiation

As discussed in Chapter 2, many hospitals with FLSs now initiate osteoporosis treatment for patients within the hospital setting, while several hospitals still send treatment recommendations to patients’ GPs. All patients within the group were initially prescribed treatment by their GP. One patient was experiencing problems with this, having been waiting 6 weeks at that point in time for his or her GP to take action following a treatment recommendation made by a hospital consultant. There was a sense that GPs often lacked knowledge and interest in osteoporosis, compared with other diseases such as cancer and heart disease. There is a lack of information available in GP surgeries about the risks of osteoporosis, and patients wished that they had had more information before they fractured and were diagnosed. Some patients have previously researched bone-strengthening treatments themselves when they found that their GP lacked knowledge about the options available, and had made their own choices about which treatment they wanted. Others felt that GPs also lacked training in diet and nutrition, with one person having taken calcium supplements for years with their GP never mentioning that it would be helpful to take vitamin D supplements alongside. The patients also felt that GPs were too busy to read all of the letters they received from hospital staff, and they sympathised with this and understood that GPs could not be experts in everything. Although hospitals with established FLSs and orthogeriatric services are taking over more of the responsibility for treatment initiation and monitoring, patients felt that it was still more convenient to visit GPs rather than attend any additional hospital appointments. Several patients were members of their local Patient Participation Group, which they thought was an excellent way of expressing their concerns to GPs.

Adherence and monitoring

All of the patients involved had been monitored after discharge by the FLS at their local hospital. They were initially sent a letter, and if they did not respond then they were contacted by telephone, and referred for DXA scans every 2–5 years post fracture (depending on the severity of their osteoporosis and age). The group were very surprised by the low rates of adherence, but could understand why many people failed to take their treatment. The asymptomatic nature of osteoporosis means that you cannot tell if the treatment is working, making people less likely to stay on it. They felt that their follow-up DXA scans helped them to monitor how their treatment was working for them. Patients agreed that this was a potential reason why changes to service delivery had no effect on rates of second fracture, and felt that more monitoring after discharge was important. Many patients are not aware that there are alternative medications available if they have side effects from their first treatment and this should be more clearly explained.

Multidisciplinary care

One particular point our patients agreed with was the importance of multidisciplinary care for hip fracture patients, as was also highlighted by many of the clinicians interviewed. Patients were aware of how other medical conditions can greatly hinder recovery following a major fragility fracture, and that multidisciplinary care, in their own experience, can help to pick up on pre-existing medical conditions.

NHS England/Public Health England

This study will be useful to analysts estimating the burden of hip fracture and osteoporosis and the long-term cost-effectiveness of interventions for the prevention and management of these conditions. Previous estimates of the health and social care costs of hip fractures in the UK range from £2B to £3B per year, but UK cost data on hip fracture were limited. The results of our study show that hip fractures are an enormous cost to the health-care system. Furthermore, the findings highlight the impact of complications following initial hospital discharge as a main driver of costs and the importance of preventing hip refractures.

This 1-year hospital cost of hip fracture was used by the multistakeholder FLS implementation group economic model which includes the National Osteoporosis Society (third sector), the Royal College of Physicians Clinical Effectiveness Evaluation Unit, Fragility Fracture Audit Programme FLS-Database Workstream, NHS England CCG gateway and Public Health England as a robust up-to-date estimate of the economic impact of hip fracture and potential benefits of a FLS at the local CCG level as well as the national level. These data will be used by NHS England to inform decisions about health service changes aimed at achieving greater efficiency and better patient care within the NHS. These findings also received significant press coverage in several daily newspapers as well as ITV news following presentation at the 2015 IOF-ESCEO (European Society for Clinical and Economic Aspects of Osteoporosis and Osteoarthritis) conference, where they will have been seen by the general public as well as clinicians, NHS commissioners and policy-makers. The articles emphasised the increasing number of hip fractures and the financial implications, and made the case for policy-makers to prioritise bone health through the universal provision of FLSs.

Fracture liaison service workshop

Although national guidelines recommend service models targeting secondary fracture prevention at patients following a fragility fracture, data on the clinical effectiveness of these services are rare. The increasing burden of hip fractures is also a concern in other countries across Europe, North America and Australasia. Data from this study showed that both the introduction and the expansion of an orthogeriatric service or a FLS were effective in reducing 30-day and 1-year mortality following a hip fracture. This information was presented at an international workshop held in Milan that was targeted at health-care professionals who were either in the process of setting up a FLS or considering it. The findings of this study support the effectiveness of fracture prevention services in reducing mortality and may persuade more hospitals worldwide to adopt this model of care and set up a FLS.

Clinical Practice Research Datalink codes for fragility fracture

As part of this study we created a robust list of CPRD Read codes for fragility fractures, osteoporosis-related medications and comorbid conditions. A rigorous procedure was used to generate the lists, with input from two clinicians who each generated two separate lists and then discussed and agreed on any discrepancies. Generating such a thorough and extensive list is a time-consuming process and organisations often do not share lists of codes with each other once these are generated, so the procedure is often repeated. We now have a thorough and extensive list of Read codes which can be used by other researchers studying osteoporosis or fragility fractures. We have shared this list with researchers from the Falls and Fragility Fracture Audit Programme being undertaken by the Royal College of Physicians. This is a national clinical audit designed to audit the care that patients with fragility fractures and inpatient falls receive in hospital and to facilitate quality improvement initiatives.

Establishing levels of fracture prevention services in hospitals in Spain

Despite national and international guidance from various professional organisations such as the BOA, NICE and the IOF, there is major variation in the care pathway for the treatment and management of hip fracture patients and in the way secondary fracture prevention services are organised across hospitals in the UK. This study began by establishing whether or not a FLS or other type of fracture prevention service (such as an orthogeriatric-led service) was in place at 11 different hospitals in one region of England. A questionnaire for clinicians was developed to gather this information and showed large variations in services across hospitals in one region of England. This work was presented to a group of clinicians in Spain. Following this presentation, the group that we met with intend to use this questionnaire to gather information about FLSs across hospitals in Spain. This will enable them to identify hospitals with the most need for improvement to target for a programme to develop FLSs.

Hospital 2

The involvement of an orthogeriatrician in the care of hip fracture patients has ensured that the majority of patients admitted with a hip fracture are now seen preoperatively and the orthogeriatrician attends the daily trauma meetings and does a daily ward round (assisted by an elderly care specialist registrar). This allows patients to reach theatre quicker and with less physical deterioration by optimising any preoperative condition, such as pre-existing medical comorbidity and acute conditions, ensuring that those taking warfarin are identified and assessing fitness for anaesthesia. Involving the orthogeriatrician in postoperative care ensures early identification and treatment of complications such as chest infections and myocardial infarction. Rehabilitation goals are set with the orthogeriatrician leading multidisciplinary team meetings.

The impact of introducing an orthogeriatrician at this hospital on 30-day and 1-year mortality was HR 0.76 (95% CI 0.58 to 0.99) and HR 0.78 (95% CI 0.67 to 0.91), respectively. In the year pre intervention the number of deaths at 30 days and 1 year was 45 (12.0%) and 121 (32.2%), respectively and, after intervention, this has now fallen to 29 (6.9%) and 108 (25.6%) deaths, respectively.

Hospital 7

The consultant geriatrician started to focus on hip fracture in July 2004. The geriatrician’s sessions have increased over time to 30.5 hours per week, and an extra session was added in January 2012. There is also a specialty doctor in geriatrics. The service has changed from providing reactive care to becoming a service that sees all patients preoperatively and postoperatively, assesses and manages falls and bone health, and provides discharge planning. Patients are now seen on a dedicated hip fracture ward. The hip fracture unit works in a collaborative fashion, and uses a care pathway and multidisciplinary paperwork. In the unit, two fast-track beds are available, and the unit operates within 36 hours to meet the Best Practice Tariff. The hip fracture unit has been crucial in building expertise and experience.

The impact of introducing an orthogeriatrician at this hospital on 30-day and 1-year mortality was HR 0.69 (95% CI 0.54 to 0.89) and HR 0.85 (95% CI 0.73 to 0.99), respectively. In the year pre intervention the number of deaths at 30 days and 1 year was 50 (8.5%) and 172 (29.4%), respectively, and, after intervention, this has now fallen to 38 (6.1%) and 162 (26.1%) deaths, respectively.

Hospital 8

On starting in post, the clinical lead orthogeriatrician reviewed the existing service. Based on this review, the orthogeriatrician changed the pathway after 6 to 8 months in post, moving the service towards an acute model of care for fragility fracture patients. There is now a joint trauma round that includes weekends run by the clinical lead orthogeriatrician and the trauma lead who started at the same time (March 2009) (this is now run jointly by the clinical lead and a consultant orthogeriatric surgeon who started in July 2012). There are six consultant ward rounds per week, resulting in a total of 7.5 Direct Clinical Care. There is a trauma meeting with orthopaedic surgeons at which they ensure that all patients are seen. All patients with fragility fractures who are admitted at the hospital are placed under the care of the orthogeriatric team. They see patients of all ages with any fragility fracture. Younger patients are identified and seen for further follow-up. Older fragility fracture patients (and all hip fracture patients whether aged over or under 70 years) are seen by day 2, and by day 3 on the ward. On the weekends there is another geriatrician who provides cover. Around 90% of patients are seen preoperatively to optimise them for surgery. This all matches the criteria given in the Best Practice Tariff. 76

The impact of introducing an orthogeriatrician at this hospital on 30-day and 1-year mortality was HR 0.57 (95% CI 0.45 to 0.73) and HR 0.81 (95% CI 0.70 to 0.95), respectively. In the year pre intervention the number of deaths at 30 days and 1 year was 68 (12.8%) and 176 (33.2%), respectively, and, after intervention, this has now fallen to 33 (6.5%) and 124 (24.3%) deaths, respectively.

The reasons for a significant decrease in mortality following the introduction and/or expansion of a FLS model of care are not as clear as for the orthogeriatric model. Implementation of osteoporosis guidelines by a fracture nurse has previously been shown to be associated with a 33% reduction in post-fracture mortality following any index fragility fracture, prompting the conclusion that measures to prevent fractures also reduce mortality. 101,121 While FLSs conceivably contribute to an environment of better co-ordination of care with better communication between staff, it may be that the appointment of such nurse specialists reflects wider underlying changes to a hospital’s service delivery that led up to the successful implementation and change in hospital care model. In addition, a RCT has indicated that patients receiving zoledronic acid were at 28% reduced risk of death, compared with those receiving placebo,119 yet only 8% of such a reduction can be attributed to lower fracture incidence on treatment. 120 The mechanisms that mediate the remainder of the drugs effect are not known, although an effect on cardiovascular events and pneumonia may play a role.

To our knowledge this is the first study to evaluate the FLS model in terms of impact on hip refracture rate after primary hip fracture. We found no evidence that FLS models of care reduce the risk of second hip fractures. The average annual age- and sex-standardised proportion of second hip fractures was just 4.2% for hospitals in this region of England, and remained unchanged and stable through the study period from 2003 to 2013. Although this might seem surprising, given the positive results reported by the Glasgow,24 Concord21 and Kaiser SCAL studies,25 there are many important key limitations associated with these studies, as described earlier, that may explain this discrepancy.

To try to understand the reason for the lack of association, it is necessary to tease apart the key elements of a FLS: case finding; osteoporosis assessment including a DXA scan to measure bone density if appropriate; treatment initiation with bone protection therapy in osteoporosis patients; and systems to improve adherence and persistence with therapy. As part of the qualitative study, we identified the elements of care of hip fracture patients that health professionals think are most effective in preventing secondary fractures after hip fracture. This included the processes for undertaking the four main components of a fracture prevention service and co-ordination of care.

Case finding

Participants felt that such patients were relatively easy to identify as they were invariably admitted and remained in hospital for a period of time and, therefore, presented a ‘captive audience’. Attendance of fracture prevention co-ordinators or other health-care professionals at daily preoperative trauma meetings was seen as effective at identifying patients at risk of further fractures. Perioperative or postoperative ward rounds were successfully used to identify patients, although participants were concerned that patients seen perioperatively could be missed if they were particularly ill or when staff were absent, as it meant that there was no one to identify cases. To mitigate this, participants emphasised the importance of a back-up such as computerised databases to identify patients retrospectively. The processes underlying case finding are done well for this element of a FLS.

Osteoporosis assessment

This was best done in an inpatient setting, when possible, to enable clinicians to initiate bone protection therapies more quickly. A risk of ‘losing’ patients after discharge was identified, as at that time they may not receive appointment letters, may be too frail or may forget to attend appointments. Participants were of the opinion that services should adhere to NICE guidelines by providing DXA scans to those aged < 75 years and treating those aged ≥ 75 without the absolute need for DXA. The location of the DXA scan, whether in an inpatient setting, an outpatient setting or a community hospital, was seen to impact on whether or not patients received it. Conducting a scan in the post-discharge outpatient setting gave patients time to recover from their operation, but this was tempered by concern about the failure of patients to attend appointments. This was seen as being particularly problematic when patients lived a long way from the scanner. The process of referral to scans was important; while some orthogeriatricians and fracture prevention nurses described how they were able to refer patients directly, others had to request an appointment via primary care. Interviewees felt that the latter approach was problematic, as it meant that patients had to wait longer, delayed the start of treatment and made it more likely that patients would ‘get lost’ in the system. In the < 75 years age group there is greater potential for patients to miss receiving osteoporosis assessment and not receive bone protection therapy if needed, although only a small minority of hip fractures patients are aged < 75 years.

Treatment initiation

Patients who were aged ≥ 75 years and who did not need a DXA scan should have their treatment initiated in an inpatient setting when possible. This meant that they could begin treatments more quickly and enabled clinicians to assess whether or not they could tolerate it. However, some clinicians were concerned that patients were sometimes too ‘shocked’ postoperatively to understand how and why they were taking the therapies, which could impact on adherence. Some participants were worried that patients were not being prescribed treatments while they were in an inpatient setting. One clinician had found it useful to put ‘checks’ in place so that his or her colleagues were able to spot if they had not been prescribed. Furthermore, treatments were not always included on discharge summaries. For those aged < 75 years who received a DXA scan or who needed treatments that could not be initiated immediately, therapies were initiated either in primary care or in an outpatient setting. However, there was a concern that this was not being done consistently in primary care and that GPs lacked ‘alertness’ about the importance of the therapies.

Further information is provided by correlating this with statistical analysis of national data from CPRD in Chapter 6. Figures 13–15 show that there has been a substantial increase in the prescribing of antiosteoporosis medication within 12 months of hip fracture, particularly since 2004. However, differences are observed stratified by age. In those aged > 85 years, prescribing rates have increased from 10% to 50%, and in those aged 75–84 years they have increased from 20% to 55%, but in those aged < 75 years the increase has been smaller, from around 20% to 40%. This would fit with qualitative data that younger patients may be missed or that in earlier years there was differential undertreatment of the older patient. An aspect not addressed in the qualitative study is a sex effect (see Figure 13), whereby the national increase in antiosteoporosis medication is much higher in women than in men. There is potential for the < 75 years age group, and for male patients, to be less likely to initiate osteoporosis therapy.

Monitoring

Participants were most concerned about the low levels of adherence to bone protection therapies and felt that monitoring had the potential to improve adherence to oral therapies. A number of participants thought that zolendronic acid could help improve adherence, as it is given at a clinic rather than at home and given once per year rather than being taken regularly. However, they also described how the provision of this therapy was often constrained by local guidelines as well as by the requirement to have good kidney function, an issue with many patients. Patients prescribed oral bisphosphonates had their monitoring delegated to primary care. However, there was a worry that this management was often ‘suboptimal’ and some participants were concerned that GPs did not always have enough knowledge to monitor patients effectively. On account of these problems, participants thought that more monitoring could be conducted by secondary care. However, there was no consensus on how this could be systematically achieved. Participants discussed the relative advantages and disadvantages of using questionnaires, telephone calls and outpatient appointments to perform monitoring.

Within this study, the service evaluation highlights how monitoring was undertaken by secondary care at seven sites and the remainder was delegated to GPs. Monitoring by secondary care included telephone calls and questionnaires. Only one hospital had a FLS that included a monitoring pathway, but, as this happened at the end of 2011, it was not possible to examine the impact of this on outcomes, as the HES data for our analysis were collected up to the end of 2013. The IOF adherence gap report162 highlights that up to 60% of patients who take a once-weekly bisphosphonate and nearly 80% who take a once-daily bisphosphonate discontinue treatment within 1 year.

Monitoring and adherence to therapy is the weakest element of FLS for hospitals within the region, and was highlighted as a clear cause for concern by health professionals in the qualitative study. This is a key element that could explain the lack of effectiveness of the FLS intervention.

The FLS model of secondary fracture prevention is centred on the efficacy of antiresorptive drugs. It is widely known that the risk of further fracture can be reduced by up to half with bone protection therapy. 1,11–13 The clinical effectiveness of these drugs is reviewed in NICE TA 161 guidance, in which it is noted that for non-vertebral fracture types, individual data on hip, leg, pelvis, wrist, hand, foot, rib and humerus fractures were sometimes provided, whereas some studies presented only data for all non-vertebral fractures grouped together. In their consideration of the evidence it is interesting that they note that ‘all these drugs have proven efficacy in reducing the incidence of vertebral fragility fractures in women with osteoporosis, but that there were differences between the drugs as to the degree of certainty that treatment results in a reduction in hip fracture’. They considered evidence for effectiveness to be robust for alendronate and risedronate, but not for etidronate, strontium and raloxifene, and the effect of teriparitide required more research. Data from 14 RCTs indicated that between 81% and 100% of patients persisted with bisphosphonates in the first year of treatment. This contrasts starkly with adherence and persistence outside the clinical trial setting, which is substantially lower.

Statistical analysis of our UK data from CPRD shows that the increase in bisphosphonate use is largely driven by the use of alendronate, with minimal use of other bisphosphonate therapies, for which the trend over the past decade has been largely flat. This is reassuring, given the evidence of effectiveness of alendronate from RCTs.

Although oral bisphosphonates have been demonstrated to be effective in a trial setting, there is a lack of generalisability in terms of real-world effectiveness, largely driven by issues in adherence and persistency in therapy. In particular, the oral bisphosphonates, although taken only weekly, have a very complicated method of administration that needs to be carefully followed to ensure reasonable bioavailability. Hence, the lack of association observed in our study may reflect the poor quality of the monitoring element of FLS in the hospitals in the region of study.

Problems relating to adherence and persistence with therapy can be overcome with intravenous (i.v.) bisphosphonates, which need administering only once per year. Data from RCTs have shown significant reductions in subsequent fracture risk among hip fracture patients on i.v. antiresorptive treatment in all patients, including those with cognitive impairment. 119 However, specifically in this trial, zoledronic acid has only been shown to significantly reduce the risk of second hip fracture after a primary hip fracture after 12 months, with the same observed for the composite outcome of all non-vertebral fractures. In addition, generalisability is primarily affected by the trial excluding patients with a life expectancy of < 6 months in the investigator’s judgement. We could not address this exclusion criterion within our study.

Our data from the natural experiment highlight that 35% and 63% of secondary hip fractures within 2 years occurred within the first 6 and 12 months, respectively. Further in our hip fracture population, the mortality rates were 9.5% and 29.8% at 30 days and 1 year, respectively, with an average life expectancy of just over 2 years.

Although i.v. medications such as zoledronic acid overcome problems with adherence to therapy monitoring, a move from oral to i.v. bisphosphonates may not be as clinically efficacious as perceived owing to (1) the high mortality within the first year of hip fracture; (2) the fact that the majority of second hip fractures occur within the first year; and (3) the effectiveness of zoledronic acid on reducing secondary fractures occurs after the first year. A further consideration is that many patients will not be eligible for treatment with zoledronic acid owing to their poor renal function.