Caesarean delivery and risk of stillbirth in subsequent pregnancy: a retrospective cohort study in an English population
Ron Gray, National Perinatal Epidemiology Unit, Department of Public Health, Richard Doll Building, University of Oxford, Old Road Campus, Headington, Oxford OX3 7LF, UK. Email firstname.lastname@example.org
Objective Two recent studies indicate an increased risk of stillbirth in the pregnancy that follows a pregnancy delivered by caesarean section. In this study, we report an analysis designed to test the hypothesis that delivery by caesarean section is a risk factor for explained or unexplained stillbirth in any subsequent pregnancy. We also report on the proportion of stillbirths in our study population, which may have been attributable to previous delivery by caesarean section.
Design Retrospective cohort study.
Population Linked statistical data set of 81 784 singleton deliveries registered in Oxfordshire and West Berkshire between 1968 and 1989.
Methods The crude and adjusted hazard ratios for stillbirth in deliveries following a previous delivery by caesarean section, compared with no previous caesarean, were estimated using Cox regression.
Main outcome measure Stillbirth.
Results The unadjusted hazard ratios for all, explained, and unexplained stillbirths were 1.54 (95% CI 1.04–2.29); 2.13 (1.22–3.72); and 1.19 (0.68–2.09), respectively. After adjustment for maternal age, parity, social class, previous adverse outcome of pregnancy, body mass indexand smoking the hazard ratios were 1.58 (0.95–2.63), 2.08 (1.00–4.31) and 1.24 (0.60–2.56).
Conclusions Pregnancies in women following a pregnancy delivered by caesarean section are at an increased risk of stillbirth. In our study, the risk appears to be mainly concentrated in the subgroup of explained stillbirths. However, there are sufficient inconsistencies in the developing literature about stillbirth risk that further research is needed.
Provide feedback or get help You are viewing our new enhanced HTML article.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
For reasons which remain largely unclear, rates of delivery by caesarean section in the UK, and elsewhere, have risen over the past 20–30 years.1,2 In addition to an increased risk of immediate complications for both the mother and baby, this major operative procedure has also been associated with an increased risk of specific adverse events in subsequent pregnancies, which include placenta praevia, abruption and uterine rupture during labour.3,4 A recent analysis of routinely collected Scottish maternity data found a doubling in the risk of unexplained antepartum stillbirth5 in the pregnancy following the one delivered by caesarean section. A study using the Missouri linked maternity data set found a borderline significant increase in the risk of stillbirth following caesarean section overall, with no significant association in white mothers but a significant elevation of risk of about 40% in black mothers.6 These studies are, as yet, the only reports on this association and it therefore merits investigation in other large data sets.7 We report here an analysis designed to test the hypothesis that delivery by caesarean section in any previous pregnancy is a risk factor for (a) any stillbirth, (b) explained stillbirth and (c) unexplained stillbirth in a subsequent pregnancy. We also report on the proportion of stillbirths in our study population, which may have been attributable to previous delivery by caesarean section.
Source of data
The Oxford record linkage study (ORLS) is a database of linked anonymised birth registrations, death certificates, stillbirth certificates and statistical abstracts of NHS maternity and other hospital information for part of the south of England.8 The data set used in the present analysis includes detailed information collected between 1968 and 1989 on approximately 300 000 births to women who lived in the NHS health authority areas of Oxfordshire and West Berkshire at the time of delivery. Collection of detailed maternity data in ORLS ceased in 1989 as a result of Department of Health reforms to NHS data sets.
All deliveries recorded in the ORLS between January 1968 and December 1989 to women who had two or more completed pregnancies were included. Two groups of deliveries were defined. The first, the ‘exposed’ population, consisted of each delivery that occurred in women who had been delivered by caesarean section in any of their previous pregnancies. The second, the ‘unexposed’ population, consisted of each delivery for each woman who had not had a delivery by caesarean section. When a woman had a delivery by caesarean section, all her subsequent pregnancies were classified as exposed. Exclusion criteria were deliveries at less than 28 or more than 43 completed weeks of gestation, perinatal death due to congenital anomaly or rhesus incompatibility, and records with an implausible interpregnancy interval of less than 189 days. Deliveries in which data were missing about exposure (i.e. mode of delivery in a previous pregnancy) or gestational age were also excluded from the analysis. Finally, cases where the cause of stillbirth seemed implausible were also excluded: most of these were deaths where the cause of death indicated a neonatal death following a live birth rather than a stillbirth.
Definitions of maternal and obstetric characteristics
Exposure to previous delivery by caesarean section in a pregnancy was classed as ‘yes’ or ‘no’. We also created a binary variable called ‘previous adverse pregnancy outcome’, which were those pregnancies preceded by any previous pregnancy resulting in either a baby weighing 2500 g or less, or a pregnancy delivered at less than 37 completed weeks of gestation, or a pregnancy resulting in stillbirth, were coded as ‘yes’; the remainder were classified as ‘no’. Parity was defined as the number of previous births recorded in ORLS, maternal age was defined as age in years at the time of delivery and body mass index (BMI) at booking was grouped as obese (≥30), overweight (25–29), normal (18.5–24.9) and underweight (<18.5).9 Smoking was defined as current smoker or nonsmoker at the time of antenatal booking. Social class was grouped, as recorded in the ORLS, using the Registrar General’s occupational classification.10
Information on cause of stillbirths was coded in the ORLS from extracts of the stillbirth certificates. The underlying cause of death was coded using the International Classification of Diseases (ICD) eighth revision from 1968 to 1979 and the ninth revision from 1979 to 1989.11,12 For the period covered by this study, stillbirths were registered at or above 28 weeks of completed gestation.13 On the basis of the information available, it was not possible to differentiate antepartum from intrapartum stillbirths. After exclusion of stillbirths due to congenital anomalies and rhesus isoimmunisation, we classified stillbirths as due to (a) maternal diseases and complications of pregnancy, (b) abruption/haemorrhage, (c) mechanical causes, (d) fetal haematological disorders (nonrhesus), and (e) the remainder of stillbirths. Groups (a) to (d) collectively were termed ‘explained’, and group (e) was termed ‘unexplained’.
Data were analysed by comparing deliveries exposed to a previous delivery by caesarean section with those deliveries not so exposed. The unadjusted risk ratio of stillbirth in the exposed group of deliveries compared with the unexposed group was calculated.
The stillbirth rate was compared between exposed and unexposed deliveries for each grouping of stillbirth separately (all stillbirths, ‘explained’ stillbirths, and ‘unexplained’ stillbirths) and hazard ratios were estimated using Cox proportional hazards models.14,15
Missing values, confounders and statistical power
Some of the data items included in our analyses, such as social class, smoking and BMI, were only collected for part of the study period. We therefore had to consider two options. One option was to use all records. This would maximise statistical power but limit our scope for control of confounding. A second option was to use only those records with complete information about confounders. This would maximise our ability to control for confounding but reduce statistical power. Therefore, we analysed data for the whole population with no adjustment for confounders, for a reduced population with control for some confounders, and for a further reduced population with control for a wider range of confounders.
The sample size was fixed as the total number of eligible singleton deliveries in the ORLS. Power calculations based on the observed stillbirth rate of 0.35%, the observed caesarean section rate of 7.1%, the total sample size of 81 707, and 5839 exposed pregnancies indicated 80% power to detect a relative risk (RR) of 1.76 (e.g. to compare 0.34 with 0.60%) but only 47% power to detect an RR of 1.5 (e.g. to compare 0.34 with 0.50%).
For explained stillbirths, we undertook separate analyses for causes of death grouped as (a) stillbirths caused by maternal disease and complications of pregnancy, and (b) stillbirths caused by abruption/haemorrhage. Since each woman could contribute more than one delivery to the analysis, we used the Huber16 method in all statistical models to estimate standard errors, which allowed for the nonindependence of these deliveries. Finally, we calculated the proportion of stillbirths in the whole ORLS population, which may have been attributable to delivery by caesarean section assuming causality. In this calculation, we used the fully adjusted hazard ratio for all stillbirths as an approximation for the adjusted RR and the following equation:
As a check on the completeness of the ORLS data we compared the annual rates of stillbirth from the ORLS data with corresponding figures for the county of Oxfordshire, which we obtained from the Office for National Statistics. The rates were similar.
The study team had access only to fully anonymised data. Research ethics committee approval for the study was granted by Oxfordshire Local Research Ethics Committee A.
There were 111 586 second or subsequent singleton births recorded in the ORLS between 1968 and 1989. After exclusions due to missing data, implausible interpregnancy interval or rhesus incompatibility, there were 81 784 births eligible for inclusion in the analysis of which 367 were stillbirths. We excluded stillbirths in which the causes of death were congenital anomalies (n= 56, 15%) and stillbirths which were miscoded (n= 21, 6%). The remaining 81 707 deliveries included 290 stillbirths classified as explained (n= 113, 39%) or unexplained (n= 177, 61%) stillbirths.
A comparison of the characteristics of those deliveries exposed to previous delivery by caesarean section and those deliveries not exposed to a previous caesarean section is shown in Table 1. Mothers who had undergone previous delivery by caesarean section were generally older, of lower parity, of higher social class and more likely to have had previous adverse pregnancy outcomes than mothers who had not had a previous caesarean section.
Table 1. Comparison of maternal and perinatal characteristics in those pregnancies following any previous delivery by caesarean section and those not, showing numbers of records with data on each characteristic
|Maternal age||81 649|| ||<0.001|
|<20||1746 (2.3)||89 (1.5)|
|20–25||23 485 (31.0)||1500 (25.7)|
|26–30||30 343 (40.0)||2223 (38.1)|
|31–35||15 791 (20.8)||1481 (25.4)|
|36–40||3982 (5.3)||464 (8.0)|
|>40||465 (0.6)||80 (1.4)|
|Parity||80 389|| ||<0.001|
|1||47 718 (63.9)||3790 (68.0)|
|2||17 882 (24.0)||1381 (24.0)|
|3||5916 (8.0)||401 (7.0)|
|≥4||3125 (4.2)||176 (3.1)|
|Previous adverse pregnancy outcome**||81 707|| ||<0.001|
|No||69 643 (91.8)||4899 (83.9)|
|Yes||6225 (8.2)||940 (16.1)|
|BMI||48 042|| ||<0.001|
|<18.5||861 (2.0)||67 (1.5)|
|18.5–24.9||27 245 (62.4)||2518 (57.7)|
|25–29.9||11 931 (27.3)||1274 (29.2)|
|30+||3638 (8.3)||508 (11.6)|
|Smoking||51 776|| ||0.182|
|No||36 245 (76.7)||3493 (77.6)|
|Yes||11 027 (23.3)||1011 (22.5)|
|Social class||69 821|| ||0.002|
|I||6503 (10.0)||597 (11.8)|
|II||14 075 (21.7)||1130 (22.3)|
|III||30 485 (47.1)||2304 (45.5)|
|IV||9114 (14.1)||696 (13.7)|
|V||3778 (5.8)||280 (5.5)|
|Other||797 (1.2)||62 (1.2)|
The risk of stillbirth in the group of deliveries not exposed to a previous caesarean section was 263/75 868 = 3.5 per 1000 (95% CI 3.1–3.9) and in the group exposed to previous caesarean section it was 27/5839 = 4.6 per 1000 (95% CI 3.0–6.7). This yielded an RR of 1.30 (95% CI 0.91–1.87) and an unadjusted risk difference of 1.16 per 1000 births (95% CI 0.79–2.73) without adjustment for potential confounders or for women having more than two deliveries included in the study.
A classification of the cause of stillbirth by exposure or nonexposure to delivery by previous caesarean section is shown in Table 2. Stillbirths that occurred in the group of deliveries exposed to previous caesarean section were more likely to be explained than those in the unexposed group. Furthermore, in the group of deliveries exposed to previous caesarean section, stillbirths were more likely to be classified as being associated with an abruption/haemorrhage or mechanical causes than those occurring in the unexposed group.
Table 2. Causes of stillbirth* in the index pregnancy classified by exposure to delivery by caesarean section in any previous pregnancy or no exposure
|Explained||99 (37.6)||14 (51.9)|
|Maternal diseases and complications of pregnancy||43 (16.3)||2 (7.7)|
|Abruption/haemorrhage||35 (13.3)||8 (30.8)|
|Mechanical||15 (5.7)||3 (11.5)|
|Fetal haematological disorders (nonrhesus)||6 (2.3)||1 (3.9)|
|Unexplained||164 (62.4)||13 (48.2)|
|Total||263 (100)||27 (100)|
The hazard ratio for stillbirth associated with exposure to previous caesarean section was 1.54 (95% CI 1.04–2.29). After adjustment for maternal age, parity, social class and previous adverse outcome of pregnancy in the deliveries with complete recording of these items, the hazard ratio was 1.53 (95% CI 1.00–2.34); and in the smaller subset with further adjustment for BMI and smoking it was 1.58 (95% CI 0.95–2.63; Table 3).
Table 3. HR for stillbirth following any previous delivery by caesarean section
|All stillbirthsc||1.54 (1.04–2.29)||1.53 (1.00–2.34)e||1.58 (0.95–2.63)j|
|Explained stillbirths||2.13 (1.22–3.72)||2.11 (1.16–3.84)f||2.08 (1.00–4.31)k|
|Maternal diseases and complications of pregnancy||0.71 (0.17–2.91)||0.75 (0.18–3.24)g||1.34 (0.29–6.34)l|
|Abruption/haemorrhage||3.20 (1.48–6.94)||3.57 (1.57–8.13)h||3.02 (1.06–8.63)m|
|Unexplained stillbirths||1.19 (0.62–2.13)||1.15 (0.62–2.13)i||1.24 (0.60–2.56)n|
Adjusted hazard ratios for the association between delivery by caesarean section and explained stillbirths and unexplained stillbirths were 2.08 (95% CI 1.00–4.31) and 1.24 (95% CI 0.60–2.56), respectively (Table 3), although estimates were much less precise because of the small number of stillbirths in these subgroups.
The increased risk of stillbirth is small in absolute terms but nevertheless it may account for a significant proportion of the stillbirths in those exposed to caesarean section. If this increased risk represents a causal process and is not simply a noncausal association, we can calculate the proportion of all stillbirths (explained and unexplained) in the whole population attributable to delivery by caesarean section as given below. Using a hazard ratio of 1.58 to approximate the adjusted RR in a population with a caesarean section rate of 7.1% and an average stillbirth rate of 0.35%, around 4% of all stillbirths in the population may be attributable to previous delivery by caesarean section.
The results of the present study show that pregnancies occurring in women previously delivered by caesarean section are associated with an increased risk of stillbirth.
The estimated hazard ratios appear robust to adjustment for a number of potential confounders which have been shown to be associated with stillbirth including socio-economic status,17 pre-pregnancy weight,18 maternal age and parity,19 smoking20 and previous adverse outcome of pregnancy, although in the adjusted models the confidence intervals were also consistent with no effect. As shown in Table 3, the hazard ratios remain remarkably similar in the adjusted and unadjusted samples we considered. Therefore, there seems to have been little confounding in our study by measured confounders although we cannot rule out residual, unmeasured confounding. If the association between caesarean section and subsequent stillbirth is causal then around 4% of all stillbirths in the population may be attributable to previous delivery by caesarean section.
The ORLS21–25 includes information on a wide range of potential confounders including BMI and individual-level social class. There are, however, some limitations of the present study. First, we were unable to distinguish between antepartum stillbirths from intrapartum stillbirths in this data set. Second, we had no information on the indication for caesarean section for the majority of records. Third, the pregnancies occurred between 27 and 47 years ago. Since that time, there have been a number of changes in prenatal and neonatal care in addition to a change in the caesarean section rate. These changes, in practice, mean that the balance of risk factors influencing an association between delivery by caesarean section and subsequent stillbirth may have changed.
We are aware of two other published studies on the association between delivery by caesarean section and subsequent stillbirth.5,6 In contrast to our study in which we considered all pregnancies, both studies limited their analyses to a woman’s first two pregnancies. We also analysed our data using this definition and the findings (data not shown) were similar to those reported here.
Smith et al.5 used a large Scottish database of births between 1992 and 1998. They were able to control for some important potential confounders and they had enough details on stillbirths to select the subset of unexplained antepartum stillbirths. Smith et al. found an increased risk of unexplained antepartum stillbirth from 34 weeks of gestation onward, an effect that was not attenuated after adjustment for confounding.
Our study differed from that of Smith et al. in a number of ways. First, we did not concentrate on antepartum stillbirths because we could not distinguish them from intrapartum stillbirths in our data set. Second, Smith et al. used a modified Wigglesworth26 classification of stillbirth, whereas the cause of the stillbirth in the ORLS is classified by ICD8 and ICD9 coding. We, therefore, created categories roughly analogous to those of Wigglesworth classification, treating unexplained stillbirth as a residual category. Third, we were able to examine a greater range of confounders than Smith et al. including BMI and individual-level social class. Fourth, the average rate of caesarean section in the Smith et al. study was 14.7%, whereas our study had an average rate of 7.1% showing the difference in obstetric practice in different study periods.
Salihu et al.6 used the Missouri maternally linked cohort data set between 1978 and 1997 and determined the average caesarean section rate as 18.1%. The rates of stillbirth were 0.44% in those women with a history of delivery by caesarean section in their previous (first) pregnancy compared with 0.41% in those not delivered by caesarean section in their previous pregnancy, a difference that was not statistically significant. Their adjusted estimate of risk of stillbirth in the two groups was borderline significant at 1.1 (95% CI 1.0–1.3). They did, however, find differences between black mothers and white mothers. There was no effect of previous caesarean section among the white women but in the black women there was an increased risk associated with caesarean section (OR 1.4, 95% CI 1.1–1.7). Thus, albeit in different subgroups, all three studies – the Scottish study, that from Missouri and ours – show an association between caesarean section and subsequent stillbirth.
The average rate of caesarean section in the ORLS between 1968 and 1989 was only 7.1%. Given that rates of caesarean section are now around 25% on average in the UK population,27 it seems possible that if this study was conducted today the fraction of stillbirths attributable to caesarean section in the whole population could be much higher than 4%, making caesarean section account for a higher percentage of all stillbirths. This might also explain why stillbirth rates have ceased to fall and have increased according to Confidential Enquiry into Maternal and Child Health report.28
However, a noncausal association is also possible. It is plausible that an association between previous caesarean section and subsequent stillbirth could be mediated through abnormal placentation and subsequent placental dysfunction. It is possible, for example, that some women may have a genetic predisposition to, or may be exposed to, an environmental factor which could lead to a recurrent tendency to placental dysfunction/growth restriction in all their pregnancies.29 In a first pregnancy this might produce problems leading to a caesarean section and in a second pregnancy to stillbirth. Therefore, caesarean section may not have caused placental dysfunction, rather the placental dysfunction may have influenced the clinical factors that increase the indications for caesarean section in the first pregnancy and the risk of stillbirth in the second. However, when the analysis of Smith et al.5 was confined to women delivered by caesarean section after being in labour for more than 10 hours, an association with subsequent stillbirth was still apparent. A proportion of these caesarean deliveries will have been performed for ‘failure to progress’, which would weaken the hypothesis that the association could be explained by abnormal placentation. The other possibility is that the operation of caesarean section itself may lead to abnormal placentation.5
Caesarean section guidelines from National Institute for Health and Clinical Excellence30 published in 2004 include the increased risk of antepartum stillbirth in future pregnancies as an adverse effect of caesarean section. These guidelines have been based only on the study of Smith et al. The evidence of an adverse effect is now mounting with data from two further studies but these studies remain inconsistent in the details of affected subgroups. Therefore, further information is required to determine whether this association is found consistently, whether the association is causal and what causal mechanisms may be involved.
We would like to express our thanks to Leicester Gill and Myfanwy Griffith for preparing the data set for our use. We also thank June Leach from the Office of National Statistics for supplying 1968–73 stillbirth data for Oxfordshire from the Registrar General’s annual review and the 1974–89 stillbirth data for Oxfordshire from KPVS.
This work was funded by a core grant to the National Perinatal Epidemiology Unit from the Department of Health. Dr Kurinczuk is partially funded by a National Public Health Career Scientist Award (PHCS022) from the department of Health and NHS R&D. The Unit of Health Care Epidemiology and its work on the Oxford Record Linkage Study is funded by the NHS National Co-ordinating Centre for Research Capacity Development. The opinions expressed are those of the authors and not necessarily those of the Department of Health.