Estimating the risk of adverse birth outcomes in pregnant women undergoing non-obstetric surgery using routinely collected NHS data: an observational study

Article history

The research reported in this issue of the journal was funded by the HS&DR programme or one of its preceding programmes as project number 12/209/59. The contractual start date was in April 2014. The final report began editorial review in April 2015 and was accepted for publication in January 2016. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HS&DR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Paul Aylin reports grants and other from Dr Foster (company has included an analysis of weekend mortality in previous ‘Hospital Guide’ and have paid travel expenses to biannual conference on Global Comparators in Healthcare), outside the submitted work. Alex Bottle reports that The Dr Foster Unit at Imperial College London, where he is based, is part-funded by Dr Foster Intelligence, a private health-care information company (wholly owned by Telstra Corporation Ltd).

Copyright statement

© Queen’s Printer and Controller of HMSO 2016. This work was produced by Aylin et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Patient concerns

Our study arose as a direct result of questions posed in the high-risk obstetric anaesthetic clinic by patients who were scheduled to undergo cancer surgery while pregnant. Patients wanted to know the statistical risks of an adverse outcome to the pregnancy, including miscarriage, stillbirth, premature delivery or problems following birth such as admission to neonatal intensive care. Unfortunately, the current evidence base is not sufficient to answer these questions and, therefore, this constitutes a ‘knowledge gap’.

A postal survey of women who had undergone non-obstetric surgery during pregnancy in our maternity unit over a 5-year period was then conducted. All respondents (75% response rate) expressed concern regarding the lack of availability of statistical data that could guide their decision. Although they felt that they were adequately counselled, they all agreed that, if there had been more information available, they would have been more confident in their decision-making and less anxious regarding the pregnancy outcome.

Previous literature

Previous literature suggested that non-obstetric surgery is carried out in approximately 1–2% of pregnancies,1 with common operations being appendectomy, cancer surgery and orthopaedic procedures. In this situation, women and their doctors are understandably anxious about the risk of harm to the fetus. However, there is limited available evidence quantifying the risks of miscarriage (fetal loss before 24 weeks’ gestation), stillbirth (fetal loss after 24 weeks’ gestation), premature labour or infant death post delivery.

A Canadian study investigated data from 2656 women between 1971 and 1978. 2 Patients were matched to controls by age and geographical area. There was no statistically increased risk of fetal loss among the group as a whole. However, there was an increased risk of fetal loss in women undergoing a general anaesthetic, which was most marked for women undergoing general anaesthesia for obstetric or gynaecological procedures. Some of these obstetric procedures were cervical cerclages, procedures to prevent recurrent fetal loss, and some bias will therefore have resulted. The study did not differentiate between fetal loss at different stages of pregnancy, did not look at prematurity and did not control for coexisting illness, parity or smoking.

Mazze and Källén et al. 3 analysed outcomes of 5405 Swedish women who had undergone surgery during pregnancy between 1973 and 1981 (during which period there were a total of 720,000 births in Sweden). There was no increase in rates of congenital malformations or stillbirth; however, there were significant increases in death within 7 days of delivery and in prematurity.

The other Swedish studies involved subsets of the original data – specifically investigating appendicectomy4 and laparoscopic surgery (here the data were expanded to include 2,015,000 deliveries from 1973 to 1993). 5 Sixteen per cent of women undergoing appendectomy after 24 weeks’ gestation delivered on the day of their operation, with 22% delivering within 1 week. This resulted in a significant increase in prematurity and death within 7 days of delivery, but not in stillbirth.

A systematic review of the literature from 1966 to 2002 identified 54 papers, involving a total of 12,452 patients. 6 The miscarriage rate among patients undergoing surgery during pregnancy was 5.8% (10.5% if surgery took place in the first trimester); stillbirth occurred in approximately 2% of surgeries and premature delivery in 8.2%. There were, however, no controls for comparison. The clearest data (although still poorly controlled) exist for appendicitis, with surgery-induced delivery occurring in 4.6% of women undergoing appendectomy and stillbirth in 2.6%, compared with 1.2% for other surgical procedures (p < 0.001). Fetal loss in the presence of peritonitis was 10.9%, which suggests that the condition itself rather than the operation may lead to fetal harm.

There are a number of problems with the currently available evidence. It all dates back 20–40 years and is therefore unlikely to be representative of current outcomes given the improvements in anaesthetic drugs, surgical techniques and neonatal care. Furthermore, the Swedish data4,5 were collected with the aim of studying births and, therefore, patients who miscarry, the largest group of adverse birth outcomes, are unrecorded in these studies. The studies are also, in general, poorly controlled and have conflicting results regarding the risk of surgery. Duncan et al. 2 suggest that surgery is associated with an increased risk of fetal loss (including miscarriage), and the Swedish studies4,5 suggest that there is no increase in stillbirth but that there is an increase in prematurity and early neonatal death, particularly in the case of appendectomy. Furthermore, although it is clear from the data on appendectomy that the risk to the fetus when a pregnant woman undergoes surgery is not uniform, there have been few attempts to quantify the risk by other types of surgery.

Nonetheless, of the evidence that is available, none relates directly to NHS outcomes, and there is no current NHS policy regarding carrying out non-obstetric surgery in pregnant women.

The project had three main objectives:

1. to carry out a descriptive analysis of the data, describing counts of each adverse outcome by year, maternal age, procedure type, socioeconomic status and trimester of pregnancy

2. to calculate the absolute risk and the relative odds of each adverse outcome in those women who have undergone surgery compared with those who have not

3. to independently analyse broad groups, such as elective and emergency operations, as well as common procedures such as appendectomy, cholecystectomy, specific cancer surgeries and orthopaedic surgery.

Hospital Episode Statistics database

Hospital Episode Statistics (HES) is an administrative database that includes records of all patient admissions and day cases in all English NHS hospital trusts (www.hscic.gov.uk/hes). In HES, each record contains data on patient demographics (e.g. age, ethnicity and socioeconomic deprivation based on postcode of residence), the episode of care (e.g. hospital name, date of admission and discharge) and clinical information. 7,8 Diagnoses for each patient are recorded using the International Classification of Diseases, Tenth Edition (ICD-10). Procedures performed during an episode are coded using the Office of Population, Censuses and Surveys’s Classification of Surgical Operations and Procedures (OPCS), version 4 (OPCS-4). In addition, each episode relating to the delivery of a baby contains details about the labour and delivery (e.g. parity, mode of delivery, gestational age, birthweight) in supplementary data fields known as the HES ‘maternity tail’.

In HES, each patient is assigned a unique identifier. This makes it possible to link historical medical records.

Each record represents the continuous period of time during which patient is under the care of a consultant and is called an ‘episode’. Episodes can be linked into admissions (also called a ‘spell’) to one hospital provider.

Selection of the cohort

We examined 10 years of HES maternity data from 2002–3 to 2011–12. Pregnancy records were defined as those that contained information about a delivery in either the OPCS procedure fields or the maternity tail9 (Table 1), or information about a spontaneous abortion in the ICD-10 diagnosis fields (Table 2).

Duplicate records were identified on the basis of the date of admission, the date on which the episode started, provider code, HES identification number and consultant code. 11,12 The resulting sample was restricted to women aged between 15 and 49 years. 11,13–15

Definition of outcomes and risk factors

In our original proposal we aimed to examine six outcomes: miscarriage, stillbirth, preterm labour, low birthweight, prolonged length of neonatal stay and neonatal death prior to discharge from hospital. Following receipt of data, it became clear that the linkage process on which the maternity tails were derived did not allow prolonged length of neonatal stay and neonatal death prior to discharge from hospital to be determined. We therefore substituted three new outcomes related to delivery: caesarean section, maternal death in hospital and a long inpatient stay (maternal). Our final list of adverse outcomes, based on mothers’ records, was spontaneous abortion (associated with hospitalisation), preterm delivery, caesarean section, maternal death and long maternal inpatient stays. Our adverse outcomes based on infants’ records were low birthweight and stillbirth. The definitions of these outcomes are listed in Table 2.

Adjustment for potential confounders is important to account for any differences between those women who received non-obstetric surgery during pregnancy and those who did not. We examined risk factors identified from previous research literature16–23 and these are defined in Table 3.

Risk factors such as maternal age, multiple pregnancy, socioeconomic deprivation, gestational diabetes, comorbidities, hypertension/pre-eclampsia and cardiac diseases were identified using information recorded within the hospital episode relating to the delivery or spontaneous abortion. Operations on the amniotic cavity and obstetric surgery were included in the analyses if they occurred during pregnancy (not during delivery or spontaneous abortion admission). Comorbidities (prior to pregnancy) were identified using women’s historical admission records for 3 years prior to pregnancy. The list of non-obstetric procedures, based on operating theatre procedures defined in previous research,26 was derived from the OPCS procedure fields. A non-obstetric procedure was included in the analysis if it occurred during pregnancy.

We estimated the beginning of pregnancy as:

• for delivery: admission day minus gestational age (if available) or admission day minus 36 weeks (if gestational age not available, but diagnosis field indicates preterm delivery) or admission day minus 40 weeks (otherwise)

• for spontaneous abortions: 3 months prior to admission day.

We assumed that the procedure was related to cancer or acute appendicitis if the ICD-10 code in any diagnosis field in the procedure admission was C00–C97 or K35, respectively.

From this point onwards, when we describe pregnancies in which surgery was or was not carried out we are referring specifically to non-obstetric surgery.

Statistical methods

We carried out a descriptive analysis of the data, describing total number and rates of risk factors, outcomes and missing data, and counts of each adverse outcome by year. We described counts of outcomes by common surgery groups and trimester of surgery.

We calculated the crude risk, odds ratio (OR), relative risk (RR), attributable risk, numbers needed to harm (NNHs) and associated 95% confidence intervals (CIs) of adverse birth outcomes in pregnancies in which surgery was carried out compared with pregnancies in which surgery was not carried out. 27 All pregnancy records were included in the analysis when spontaneous abortion associated with hospitalisation and maternal death were the outcomes of interest. For the remaining outcomes, only pregnancy records that ended in delivery were included. In addition, we made some comparisons of risk between surgical groups. We independently analysed broad groups of procedure (see detailed definition of groups in Appendix 1) and estimated RR and associated 95% CIs, attributable risk and NNHs of adverse birth outcome in pregnancies where surgery occurred compared with pregnancies where surgery did not occur. We also repeated these analyses by trimester of procedure.

We calculated adjusted ORs, RR and attributable risk using two approaches:

• We calculated the adjusted RR using a simple relationship:28

where P₀ denotes the proportion of untreated subjects who experience the outcome of interest and OR_adj denotes the OR obtained from a logistic regression model. The 95% CIs for the RR were estimated by substituting the upper and lower CIs for the OR from the multivariate logistic regression model. 29 It is a simple method that approximates a risk ratio from the adjusted OR and derives an estimate of an association or treatment effect that better represents the true RR. We used backwards elimination regression, and only explanatory variables with a p-value < 0.05 were considered significant. The overall model evaluation was presented by the Akaike information criterion, the R² and c-statistic.

• We calculated the adjusted RR, attributable risk and NNH using Austin’s method. 30 We assumed that a dichotomous outcome variable Y is observed for each subject (with Y = 1 denoting success and Y = 0 denoting failure). Furthermore, let T_i denote treatment status of the ith subject (with T = 1 denoting treatment and T = 0 denoting no treatment), whereas X_1i, X_2i . . . X_ki denotes the value of k confounding variables. The following logistic regression model relates the odds of the outcome to treatment status and baseline confounding variables:

where β denotes the log-OR and e^β denotes the OR. Using this formula, one can determine the probability of the outcome if a given subject were treated and the same subject were untreated. The probability of the outcome if a subject were treated is:

If a subject were not treated, then the probability of the outcome is:

The mean probability (p¯T=1) of success in the cohort if all patients were treated and the mean probability (p¯T=0) of success in the cohort if all patients were untreated is then calculated. These are also referred to as the marginal probabilities of success for treated and untreated subjects. The risk difference and the RR can be estimated as p¯T=0−p¯T=1 and p¯T=1/p¯T=0, respectively. The NNH is defined as the inverse of the risk difference. 31,32

The CIs for each measure of effect were estimated using a non-parametric bootstrap method. 33 A bootstrap sample is a random sample drawn with replacement from the original sample such that the random sample has the same size as the original sample. We created 1000 bootstrap samples and estimated the quantity of interest in each of them. The end points of the non-parametric 95% CIs would be the 2.5th and 97.5th percentiles of that quantity across the bootstrap samples. This method allows for comparison of outcomes between two populations whose only difference was the exposure (non-obstetric surgery during pregnancy). It permits an estimate of the adjusted attributable risk and NNH.

We carried out two-level logistic regression to investigate the effects of hospitals on the adverse birth outcome. 34 To test for significant differences between proportions we used chi-squared tests. Data were analysed using the SAS 9.2 software package (SAS Institute, Cary, NC, USA).

Descriptive analysis

We identified 6,486,280 pregnancies in the period April 2002 to March 2012. Two-thirds of the study population were aged between 20 and 34 years (73.4%) and the majority were multiparous (55.5%) (Table 4). Twenty-seven per cent of women were living in the most socioeconomically deprived area based on the Carstairs measure of socioeconomic deprivation quintile. More than 10% of our population had an emergency admission to hospital a year prior to pregnancy and nearly 10% had previously had a caesarean delivery. Table 4 also includes information about maternal complications such as gestational diabetes (1.5%), hypertension/pre-eclampsia (3.4%) and cardiac disease (0.3%).

Spontaneous abortions associated with hospitalisation accounted for 5.8% of all pregnancies (Table 5). The numbers of maternal deaths following spontaneous abortion associated with hospitalisation or delivery were very small; only 235 cases were identified, and this corresponds to a maternal death rate of 4 per 100,000 pregnancies. Among our cohort there were more than 450,000 (7.5%) preterm deliveries and nearly 1.5 million (23.9%) elective or emergency caesarean deliveries.

Table 6 presents total number and rates of missing records for gestational age, socioeconomic deprivation, birth status and birthweight. In nearly 40% of all pregnancies, no information about gestational age was available. Less than 1% of records included no information about socioeconomic deprivation.

We identified 47,628 pregnancies in which surgery had been carried out at some time (0.7%). Of these surgeries, 25,445 (53.4%) were elective procedures and 22,183 (46.6%) were non-elective. The most common type of surgical procedure was abdominal (12,493, 26.2%), followed by dental (5365, 11.3%), nail and skin (4762, 10.0%), musculoskeletal (4563, 9.6%), ear, nose and throat (ENT) (3060, 6.4%) and perianal (2977, 6.2%) (Table 7). There were 3062 cases of appendectomy and 1306 cases of cholecystectomy. The most frequent procedures were diagnostic endoscopic examination of the peritoneum (5518), drainage of a lesion of the skin (1662) and emergency excision of an abnormal appendix (1448) (Table 8).

The proportions of common procedure groups for different outcomes were similar to the overall proportions, except for abdominal procedures and spontaneous abortion associated with hospitalisation (see Table 7).

Spontaneous abortion associated with hospitalisation

Among our cohort, we identified 376,323 (5.8%) spontaneous abortions associated with hospitalisation, 3120 of which occurred in women who had undergone surgery during their pregnancy. The percentage of spontaneous abortions associated with hospitalisation has decreased from 6.4% to 5.3% since 2002–3 (Figure 1).

FIGURE 1.

Rate of spontaneous abortion associated with hospitalisation per year in England.

Preterm delivery

Of the 457,793 (7.5%) preterm deliveries between 2002–3 and 2011–12 in England, 1.1% were to women who had undergone surgery during pregnancy. The percentage of all deliveries that occurred preterm increased from 6.7% in 2002–4 to 8.0% in 2011–12 (Figure 2).

FIGURE 2.

Rate of preterm delivery per year in England.

Maternal death

The number of maternal deaths following delivery (or spontaneous abortion associated with hospitalisation) was small. Only 235 cases were identified, and this corresponded to a maternal death rate of 4 per 100,000 pregnancies (Figure 3). Moreover, only 12 of these women had undergone surgery during pregnancy.

FIGURE 3.

Number of maternal deaths per 100,000 pregnancies per year in England.

Caesarean section

We identified 1,461,707 (23.9%) pregnancies resulting in delivery by caesarean section over the 10-year period of the study; 12,836 of these caesarean sections were carried out in women who had undergone surgery during that pregnancy. Figure 4 shows that the proportion of caesarean deliveries slightly increased from 22.6% (in 2002–3) to 24.9% (in 2011–12).

FIGURE 4.

Rate of caesarean section per year in England.

Long inpatient stay

A long inpatient stay for a delivery episode was defined as the upper quartile of length of stay plus 1.5 times the interquartile range (10 days or longer for a caesarean delivery, 4 days or longer for a vaginal delivery). Figure 5 represents the trends in long inpatient stays from 2002–3 to 2011–12.

FIGURE 5.

Rate of long inpatient stay per year in England.

Stillbirth

Over the past 10 years, the stillbirth rate has remained relatively stable, at 0.54% (Figure 6). In our study population, 33,774 pregnancies resulted in stillbirth, of which only 411 occurred in women who had undergone surgery during their pregnancy. Not all records contained information about stillbirth; however, the percentage of missing data decreased from 2.8% in 2002–3 to 0.15% in 2011–12.

FIGURE 6.

Proportion of deliveries resulting in one or more stillborn child per year in England.

Low birthweight

A total of 342,631 (5.6%) pregnancies between 2002–3 and 2011–12 in England resulted in the delivery of one or more low-birthweight newborns, of which 3846 were to women in whom surgery had been performed. There was little change in the proportion of low-birthweight deliveries over the 10-year period (Figure 7). A notable decrease in 2007–8 can be explained by a high percentage of missing birthweight fields for that year. The proportion of birthweight values that are missing has fallen greatly over the period: from 26.0% (in 2002–3) to 10.0% (in 2011–12).

FIGURE 7.

Proportion of deliveries resulting in one or more low-birthweight newborns per year in England.

Unadjusted analyses

Table 9 gives the crude risk for each outcome by risk factor. Almost 6% of all pregnancies resulted in spontaneous abortion associated with hospitalisation. A higher risk was estimated for women aged > 40 years (14.5%) or women who had cardiac disease (11.4%). The risk of spontaneous abortion associated with hospitalisation was higher if surgery was carried out during pregnancy (6.6%). The risk of a spontaneous abortion associated with hospitalisation in pregnancies in which surgery occurred was 13% higher than in pregnancies in which surgery did not occur (RR 1.13, 95% CI 1.09 to 1.17).

A little over 6 million deliveries took place in NHS hospitals in England between 2002–3 and 2011–12, 7.5% of which were preterm. In approximately half of pregnancies with multiple gestations (i.e. more than one baby), delivery occurred preterm. Women with gestational diabetes, hypertension/pre-eclampsia or cardiac diseases were at higher risk of a preterm delivery. Pregnancies in which surgery occurred had a 48% higher risk of a preterm delivery (RR 1.48, 95% CI 1.44 to 1.52) than in pregnancies in which surgery did not occur.

Nearly 24% of all deliveries were by caesarean section. Previous caesarean delivery was associated with a higher risk of caesarean section. Moreover, two-thirds of women (69.7%) who had an emergency admission in the year prior to their pregnancy had a caesarean delivery. Furthermore, women who had gestational diabetes during pregnancy, or hypertension/pre-eclampsia, had a higher risk of caesarean delivery. Women had a 21% higher risk of a caesarean delivery (RR 1.21, 95% CI 1.19 to 1.23) if they had undergone surgery during their pregnancy.

Nearly 10% of women had a long hospital stay following delivery, and the risk was higher for women with gestational diabetes, hypertension/pre-eclampsia or cardiac diseases (19.7%, 19.3% and 13.4%, respectively). The risk of a long inpatient stay was similar in women who underwent obstetric or non-obstetric surgery (13.0% and 11.8%, respectively). Women had a 24% higher risk of a caesarean delivery (RR 1.24, 95% CI 1.21 to 1.28) if they had undergone non-obstetric surgery during their pregnancy.

The percentage of deliveries ending in a stillbirth was 0.6%. Multiple pregnancy and surgery on the amniotic cavity were associated with a higher risk of stillbirth. The risk of stillbirth was 68% higher if the woman had undergone surgery during pregnancy than if she had not (RR 1.68, 95% CI 1.53 to 1.86).

The percentage of women who gave birth to one or more low-birthweight babies was 5.6%. Moreover, among women with multiple pregnancy the risk of low-birthweight babies was much higher (51.8%). The risk of delivering a baby with a low birthweight was 57% higher for women who underwent a procedure during pregnancy than for those who did not (RR 1.57, 95% CI 1.52 to 1.62).

Tables 10 and 11 present information about the timing of surgery. In almost half of cases (42.0%), information about when in pregnancy the procedure was carried out (i.e. gestational age) was not provided. Only 4% of surgeries and 0.2% of elective operations occurred during the same admission as the delivery or spontaneous abortion (associated with hospitalisation). Of the surgeries carried during pregnancy for which gestational age was recorded, 45% were carried out in the first trimester, 26% in the second trimester and 29% in the third trimester.

A closer examination of timing between delivery (or spontaneous abortion associated with hospitalisation) and surgery suggests that < 6% of operations occurred within 1 week of the end of the pregnancy. In the case of pregnancies ending in spontaneous abortion associated with hospitalisation, 11% of operations occurred during the week prior to the abortion. Among women who underwent abdominal surgery and whose pregnancy ended in spontaneous abortion associated with hospitalisation, 16% of procedures occurred in the week prior to the abortion (see Table 11).

Table 12 shows the numbers of adverse birth outcomes in pregnant women undergoing laparoscopic and open abdominal surgeries. Approximately two-thirds of all abdominal operations were laparoscopic. Where gestational age was known, the number of abdominal operations in the first trimester that were laparoscopic was nearly five times (3102) the number of open procedures (643). This ratio was reversed in the second trimester (606 laparoscopic vs. 995 open), and by the third trimester 2.5 times as many procedures were open (698) as were laparoscopic (284).

Regression analysis

We used a logistic regression model to determine the relationship of surgery with adverse birth outcomes, while adjusting for other risk factors. Tables 13–20 give the ORs and total numbers of cases (n) of the final model for each outcome of interest. The choices of explanatory variables were constrained by previous studies and by their significance (with p < 0.05) in each model, and are given as follows.

• Multiple pregnancy, parity, previous emergency admission, gestational diabetes, hypertension/pre-eclampsia, cardiac disease, operation on amniotic cavity, obstetric and non-obstetric surgery – binary coded with ‘NO’ as reference group.

• Maternal age, in six categories: 15–19 years, 20–24 years, 25–29 years (reference group), 30–34 years, 35–39 years and ≥ 40 years.

• Socioeconomic deprivation (Carstairs deprivation quintile), in six categories: 1 (least deprived, reference group), 2, 3, 4, 5 (most deprived) and 6 (unknown).

• Comorbidities (Charlson Comorbidities Index score): 0 (reference group) and 1+.

• Tables 13–20 show that, after adjusting for risk factors, pregnancies in which surgery was carried out compared with pregnancies in which surgery was not carried out had significantly higher odds of a spontaneous abortion associated with hospitalisation (OR 1.14, 95% CI 1.10 to 1.18); preterm delivery (OR 1.48, 95% CI 1.44 to 1.53); caesarean section (OR 1.30, 95% CI 1.27 to 1.33); long inpatient stay (OR 1.25, 95% CI 1.21 to 1.28); maternal death (OR 4.72, 95% CI 2.61 to 8.52); stillbirth (OR 1.65, 95% CI 1.50 to 1.82); and low birthweight (OR 1.53, 95% CI 1.48 to 1.59).

Final analysis

This section presents the final results: adjusted RR and attributable risk and NNH for pregnancies in which surgery occurred compared with pregnancies in which surgery did not occur (Table 21). The risk of an adverse birth outcome for pregnancies with specific types of operations is also presented.

After adjusting for potential confounders, the RR associated with hospitalisation was 1.13 (95% CI 1.09 to 1.17) for spontaneous abortion, 1.43 (95% CI 1.39 to 1.47) for preterm delivery, 4.72 (95% CI 2.61 to 8.52) for maternal death, 1.21 (95% CI 1.19 to 1.23) for caesarean section and 1.22 (95% CI 1.19 to 1.25) for long inpatient stay in pregnancies in which surgery was carried out compared with pregnancies in which it was not. The RR for maternal death was high but was based on very small numbers with very wide CIs. The RR for stillbirth and low birthweight was 1.64 (95% CI 1.50 to 1.81) and 1.49 (95% CI 1.44 to 1.54), respectively. Using Austin’s method,30 we found little or no difference in our estimated RRs except for caesarean section (which is a relatively common outcome). NNH represents the number of operations associated with one additional adverse birth outcome. We estimated that, for every 143 pregnancies in which a surgical operation was carried out, compared with pregnancies in which a surgical operation was not carried out, there was one additional spontaneous abortion (associated with hospitalisation); for every 31 operations during pregnancy there was one additional preterm delivery; for every 25 operations there was one additional caesarean section; for every 50 operations there was one additional long inpatient stay; for every 287 operations there was one additional stillbirth; for every 39 operations there was one additional newborn with low birthweight; and for every 7692 operations there was one additional maternal death. Again, this estimate came with very wide 95% CIs (3571 to 33,333 operations); see Table 21. We repeated our analysis excluding records with missing values for preterm delivery, stillbirth and low birthweight, and found little difference in RR, but attributable risk and NNH changed more because of the differences in baseline incidence.

We identified 47,628 pregnancies in which surgery was carried out (0.7% of all pregnancies); in 25,445 cases (53.4%) the procedure was elective and in 22,183 (46.6%) it was non-elective. In general, the risks of adverse outcomes were higher for non-elective surgery than for planned surgery, although non-elective surgery was not associated with a significantly higher risk of spontaneous abortions associated with hospitalisation. Twenty-six per cent (12,493) of all surgical operations were abdominal. The RR of adverse birth outcomes (excluding maternal death) was higher for pregnancies in which surgery was carried out than for pregnancies in which it was not. The RR of spontaneous abortion associated with hospitalisation was 1.90 (95% CI 1.81 to 1.99), with a NNH of 20 (Table 22). Laparoscopic abdominal surgery was associated with an even higher RR for spontaneous abortion associated with hospitalisation (RR 2.47, 95% CI 2.34 to 2.60) and seemed to contribute most of the risk associated with abdominal surgery, as open abdominal surgery was not associated with a raised risk (RR 0.65, 95% CI 0.56 to 0.76). Appendectomy was not associated with an increased risk of spontaneous abortion associated with hospitalisation, but was associated with a raised risk of preterm delivery (RR 1.49, 95% CI 1.34 to 1.64), caesarean section (RR 1.13, 95% CI 1.06 to 1.21), long inpatient stay (RR 1.32, 95% CI 1.20 to 1.46) and low birthweight (RR 1.43, 95% CI 1.26 to 1.62).

Approximately 2.5% (1306) of all surgeries during pregnancy were cholecystectomy. A comparison with no surgery during pregnancy showed a difference only in spontaneous abortions associated with hospitalisation (RR 1.36, 95% CI 1.12 to 1.64).

Approximately 1.5% (710) of all operations were in patients with a diagnosis of cancer. The RR of adverse birth outcomes was higher for pregnancies in which cancer-related surgery was carried out than for pregnancies in which surgery was not carried out, with the exception of spontaneous abortion associated with hospitalisation (RR 0.53, 95% CI 0.36 to 0.78). The RR was 5.31 (95% CI 4.82 to 5.82) for preterm delivery, 1.66 (95% CI 1.50 to 1.83) for caesarean section, 1.48 (95% CI 1.23 to 1.79) for long inpatient stay, 3.12 (95% CI 1.73 to 5.60) for stillbirth and 4.44 (95% CI 3.86 to 5.08) for low birthweight.

Our analysis comparing outcomes following laparoscopic and open appendectomy (RR 2.36, 95% CI 1.71 to 3.26) and all abdominal surgery (RR 3.82, 95% CI 3.29 to 4.41) further illustrates the apparent increased associated risk of spontaneous abortion with laparoscopic procedures (Table 23).

Tables 24–26 present adjusted RRs and their 95% CIs, attributable risks and NNHs for pregnancies in which surgery was or was not carried out, by trimester. The RR of preterm delivery, caesarean section, stillbirth and low birthweight was between 20% and 30% higher for operations carried out in the third trimester than for those carried out in the first trimester. There was little difference by trimester of operation in the RR of a long inpatient stay.

Table 27 presents the results of two-level logistic regression for adverse birth outcomes. The intraclass correlation coefficients were close to 0 (from 0.0939 for spontaneous abortion associated with hospitalisation to 0.008 for low birthweight), meaning that only 1.0–10.0% of the variance is attributable to trust level. Random parameters of level 2 for stillbirth and maternal death were even smaller, and these are not presented here.

Key findings

• We identified 6,486,280 pregnancies, among which non-obstetric surgery was carried out in 47,628 (0.7%), in the period April 2002 to March 2012.

• The most common surgical procedure group was abdominal (26.2%), followed by dental (11.3%), nail and skin (10.0%), musculoskeletal (9.6%), ENT (6.4%) and perianal (6.2%). There were 3062 cases of appendectomy and 1306 cases of cholecystectomy.

• Non-obstetric surgery during pregnancy was associated with a higher risk of adverse birth outcomes than if no surgery was carried out. We estimated that for every 143 pregnancies in which a surgical procedure was performed, there was one additional spontaneous abortion (with a hospital admission); for every 31 procedures there was one additional preterm delivery; for every 7692 procedures there was one additional maternal death in hospital; for every 25 procedures there was one additional caesarean section; for every 50 procedures there was one additional long inpatient stay; for every 287 procedures there was one additional stillbirth; and for every 39 procedures there was one additional low-birthweight baby.

• Dental, perianal, breast, cancer, abdominal, ENT and musculoskeletal procedures were associated with a higher risk of adverse birth outcomes than no surgery.

• For almost half of operations (42.0%), no information about when in pregnancy the procedure was carried out was recorded.

• For pregnancies in which gestational age was recorded, the RR for preterm delivery, caesarean section, stillbirth and low birthweight was between 20% and 30% higher for those operations carried out in the third trimester than for those performed in the first trimester. There was little difference by trimester of operation in the risk of a long inpatient stay.

Findings in relation to other studies

Study strengths and limitations

This is the first study to report the risk of adverse birth outcomes following non-obstetric surgery during pregnancy across NHS hospitals in England. One important strength of this research is the use of the large and rich administrative data set. We extracted data for nearly 6.5 million pregnancies between 2002–3 and 2011–12, 10 times the total in the published literature to date. Furthermore, our data are much more recent and better represent current outcomes.

Another strength is that we estimated adjusted attributable risk and the NNH. The use of Austin’s analytical method30 was a particular strength here, as this allowed us to compare outcomes between two populations whose only difference was the exposure (non-obstetric surgery during pregnancy). These measures are more useful than an OR to women who want to be informed of the risk associated with a non-obstetric procedure, above and beyond the background risk.

Recommendations for further research

Our study has demonstrated that there is a statistically significant increase in the risk of adverse birth outcomes following non-obstetric surgery during pregnancy in England. However, we have identified a number of limitations that would benefit from further research to usefully inform a variety of medical practitioners and the general public. We offer a small set of recommendations for further research:

1. Further research is required to evaluate the association of non-obstetric surgery and spontaneous abortion.

2. Further research is needed to evaluate the impact of non-obstetric surgery on the baby (e.g. neonatal intensive care unit admission, prolonged length of neonatal stay, neonatal death) and could be assessed by linking the maternal and baby records within the HES database. Use of large clinical databases, such as EuroKing Maternity systems (www.euroking.com/), linked to the HES database could be usefully exploited for this purpose.

Dissemination activity

• To date, we have disseminated findings from this project as an oral presentation at the Applied Epidemiology Scientific Meeting in Warwick University, March 2015.

• We presented some results to an international science competition FameLab (Lithuania) in April 2015 (www.famelab.lt/apie/).

• We shall give an oral presentation at the Dame Hilda Lloyd Congress Medal plenary session at the Royal College of Obstetricians and Gynaecologists (RCOG) World Congress in June 2016.

• We have written up our findings in an academic peer-reviewed journal. 54

• We have already engaged with the British Society of Endocrine and Thyroid Surgeons and the Association of Breast Surgery, both of which have guidelines. Both organisations have agreed to consider hosting guidance that we produce on their websites.

• The data will also be used by institutions such as RCOG and the American College of Obstetricians and Gynecologists (ACOG). We are in contact with the RCOG, and it is supportive of our study and would assist in the dissemination of findings to both health-care professionals and the public.

• We shall contact the ACOG directly following publication to ensure that they are aware of our study. We shall also discuss the possibility of publishing a British guideline taking into account all available evidence, with the RCOG, the Anaesthetists Association of Great Britain and Northern Ireland and the Royal College of Midwives (RCM).

• We shall contact the RCM directly to ensure that it is aware of our study, and to make sure that it is included in any joint guidelines produced.

• We shall ensure that a lay summary of our peer-reviewed findings is available online, initially on the Imperial College London website.

Contributions of authors

Paul Aylin was instrumental in securing the data.

Paul Aylin, Alex Bottle and Violeta Balinskaite carried out the analysis and wrote the first draft.

All authors contributed to the original research proposal, helped to refine the classification of outcomes used and the procedure groups for further analysis, and commented on subsequent drafts of the report.

Ethics

We have permission from the Confidentiality Advisory Group under Section 251 of the NHS Act 2006 (formerly Section 60 approval from the Patient Information Advisory Group) to hold confidential data and analyse them for research purposes [PIAG 2–05(d)/2007]. We have approval to use them for research and measuring quality of delivery of health care from the South East Ethics Research Committee (10/H1102/25).

Publications

Balinskaite V, Bottle A, Sodhi V, Rivers A, Bennett PR, Brett SJ, et al. The risk of adverse pregnancy outcomes following nonobstetric surgery during pregnancy: estimates from a retrospective cohort study of 6.5 million pregnancies [published online ahead of print September 14 2016]. Ann Surg 2016.

Data sharing statement

Under the terms of our contract with the Health and Social Care Information Centre, we are allowed to hold our data only for as long as is specified in our data release agreement (usually 3 years unless renewed) and we are unable to share these data with other organisations. At the end of our contract we are obliged to delete the data. All source data, however, are available on application to the Health and Social Care Information Centre.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HS&DR programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HS&DR programme or the Department of Health.