Recurrence risk of stillbirth in a second pregnancy
Dr. S Bhattacharya, Dugald Baird Centre for Research on Women’s Health, Aberdeen Maternity Hospital, Aberdeen, AB25 2ZL, UK. Email email@example.com
Please cite this paper as: Bhattacharya S, Prescott G, Black M, Shetty A. Recurrence risk of stillbirth in a second pregnancy. BJOG 2010;117:1243–1247.
Objective To examine the risk of recurrence of stillbirth in a second pregnancy.
Design Retrospective cohort study.
Setting Scotland, UK.
Population All women who delivered their first and second pregnancies in Scotland between 1981 and 2005.
Methods All women delivering for the first time between 1981 and 2000 were linked to records of their second pregnancy using routinely collected data from the Scottish Morbidity Returns. Women who had an intrauterine death in their first pregnancy formed the exposed cohort, whereas those who had a live birth formed the unexposed cohort.
Main outcome measure Stillbirth in a second pregnancy.
Results After adjusting for confounding factors, the odds of recurrence of stillbirth in a second pregnancy were found to be 1.94 (99% CI 1.29–2.92) compared with women who had had a live birth in their first pregnancy. Other factors associated with recurrence of stillbirth in a second pregnancy included placental abruption (adjusted OR 1.96; 99% CI 1.60–2.41), preterm delivery (adjusted OR 7.45; 99% CI 5.91–9.39) and low birthweight (adjusted OR 6.69; 99% CI 5.31–8.42). A Bayesian analysis using minimally informative normal priors found the risk of recurrence of stillbirth in a second pregnancy to be 1.59 (99% CI 1.10–2.33).
Conclusions Women who have stillbirth in their first pregnancy have a higher risk of recurrence in their next pregnancy.
Stillbirth continues to be a major concern for obstetricians in both the developing and developed world. In approximately 60% of cases attribution of cause of death is possible,1 but unexplained stillbirths continue to baffle obstetricians and neonatologists alike. From the point of view of the couple hoping to start a family, there are few events more distressing than the loss of a first pregnancy. Although advances in antenatal care have meant that stillbirth is now rare, anxiety regarding future pregnancies remains, especially in cases following unexplained stillbirth. Based on sparse and conflicting evidence, clinicians generally counsel parents about recurrence and other adverse outcomes following stillbirth, depending on whether or not the cause of stillbirth is recurrent.2–4 There is, however, much controversy around recurrence in a subsequent pregnancy if the initial stillbirth was unexplained.5–7
In a previous study using local, routinely collected obstetric data,8 we showed that after adjusting for confounding and mediating factors, the odds of recurrence of stillbirth were no higher in women who had had an intrauterine fetal death in their first pregnancy. However, as the sample size in that study was small and the 95% confidence intervals large, we attempted to repeat the study using national data from Scottish Morbidity Records (SMR) to determine the risk of recurrence of stillbirth in a second pregnancy.
Approval was obtained from the Privacy Advisory Committee of the Information and Services Division (ISD) of the National Health Service, Scotland. A retrospective cohort study design was employed, using routinely collected data from the SMR databases. Data were extracted on all women who had a first delivery beyond 24 weeks of gestation between 1981 and 2000 from the ISD databases (SMR01 and 02), and who had been followed up for at least 5 years to identify a second pregnancy event. The exposed cohort comprised all women who had had a stillbirth in their first pregnancy, whereas all those who had had an initial live birth comprised the unexposed cohort. The principal outcome assessed was the occurrence of stillbirth in a second pregnancy. Sociodemographic and pregnancy characteristics in the second pregnancy were assessed as possible confounding factors.
Sociodemographic data available from the database included: age at delivery; smoking status, both during and before pregnancy; and Carstairs’ social deprivation category.9 The interpregnancy interval was also assessed for any difference associated with exposure status. These variables were compared between exposed and unexposed cohorts, and any variable found to be statistically significant upon univariate analysis was entered as a covariate in the final logistic regression model.
Characteristics of the first and second pregnancy, and delivery, were also compared, and odds ratios with 99% confidence intervals were calculated using univariate and multivariate logistic regression. Pregnancy and delivery complications are recorded in the SMR databases using the International Statistical Classification of Diseases and Related Health Problems 9th Revision (ICD-9) or ICD-10 coding criteria. The variables specific to both the first and second pregnancies assessed were pre-eclampsia, abruptio placentae, placenta praevia, preterm delivery (defined as delivery occurring before 37 completed weeks of gestation), post-term delivery (defined as delivery occurring at or after 42 weeks gestation), induction of labour, caesarean delivery and low birthweight (defined as a birthweight of <2500 g). Statistical significance was set at 0.01 in view of the large data set and multiple comparisons. As the data spanned more than 20 years, the year of delivery was entered into the final logistic regression model as a continuous covariate (after being centred by subtracting 2000). Statistical analysis was performed using the Statistical Package for Social Scientists (spss) v17.0 (SPSS Inc., Chicago, IL, USA). Bayesian analyses were conducted using WinBUGS.10 The model presented in the previous study included body mass index (BMI), marital status and social class, which were not available in the current dataset.8 The final model, excluding BMI and marital status, and using Carstairs’ deprivation category as a surrogate for social class, was fitted to the data from the previous study using spss. A Bayesian analysis was carried out fitting the same logistic regression model in the new Scottish data using normal prior distributions, with means of 0 and variances of 1 for all log odds ratios. With these priors, odds ratios of between 0.1 and 7.1 are fairly probable. In WinBUGS 1000 iterations were used as a burn-in phase, and estimates were based on a further 5000 iterations. Ideally the Scottish data used for the analyses should have excluded any births included in the previous study of Black et al.8 However, this was not possible because identifying information, including location, was removed during the data extraction by ISD.
Data were not available on the cause distribution of stillbirth, and therefore all women with a stillbirth in the first pregnancy were included (n = 3094), whereas 368 663 women had a live birth in their first pregnancy, as recorded in SMR01 and 02, between 1981 and 2000. A total of 2677 and 306 627 women in the stillbirth and live birth groups, respectively, had a second ongoing pregnancy, and therefore formed the exposed and unexposed cohorts for this analysis.
Table 1 presents the comparison of sociodemographic characteristics between the exposed and unexposed cohorts. More women in the stillbirth group belonged to the lower socio-economic categories (expressed as Carstairs’ decile 4 or below) in both pregnancies, although there was a small overall reduction in this proportion seen in the second pregnancy. There were more smokers, both current and former, in the group of women with an initial stillbirth. This was true for both first and second pregnancies, although there was an overall reduction in the number of smokers in the second pregnancy.
Table 1. Comparison of sociodemographic and pregnancy characteristics between women with a stillbirth and those with a livebirth in their first pregnancy
|Age at second delivery*||27.67 (4.86)||26.92 (5.69)||<0.001|
|Interpregnancy interval*||3.17 (2.03)||1.89 (1.62)||<0.001|
|Carstairs’ category (≤4) in second pregnancy||21 3837 (71.6%)||2001 (77.3%)||<0.001|
|Carstairs’ category (≥3) in second pregnancy||84 982 (28.4%)||586 (22.7%)|| |
|Never||91 391 (29.8%)||587 (21.9%)||<0.001|
|Ever||64 089 (20.9%)||703 (26.3%)|
|Not known||151 147 (49.3%)||1387 (51.8%)|
|Pre-eclampsia (second pregnancy)||6643 (2.2%)||102 (3.8%)||<0.001|
|Placenta praevia (second pregnancy)||2007 (0.7%)||24 (0.9%)||0.114|
|Abruption (second pregnancy)||9367 (3.1%)||131 (4.9%)||<0.001|
|Preterm delivery (second pregnancy)||14 638 (4.8%)||413 (15.5%)||<0.001|
|Low birthweight (second pregnancy)||14 162 (4.6%)||382 (14.3%)||<0.001|
|Stillbirth in second pregnancy||1309 (0.4%)||50 (1.9%)||<0.001|
The results of univariate comparison of pregnancy complications and interventions between the exposed and unexposed cohorts show that compared with women who had an initial live birth, pre-eclampsia was less common in the stillbirth group in their first pregnancy, but more common in their second pregnancy. There was little difference in the incidence of placenta praevia between the two comparison groups, and between the two pregnancies. Placental abruption, however, was much more prevalent, especially in the first pregnancy, in the stillbirth group.
Labour was much more likely to be induced in both pregnancies in the exposed cohort. Both elective and emergency caesarean sections were less common in the first pregnancy in this group of women, but there was a reversal of this in the second pregnancy, when women with a previous stillbirth were more likely to have a caesarean delivery.
Almost 60% of the stillbirths were delivered preterm in comparison with 5.7% of the live births. Although the prevalence of preterm delivery was reduced in both groups in the second pregnancy, it remained higher in the stillbirth group compared with the live birth group.
Low birthweight (defined as a birthweight ≤2500 g) mirrored the effect observed in preterm delivery, producing a crude odds ratio of 3.44 (99% CI 3.08–3.84) of low birthweight in a second pregnancy if the first pregnancy was stillborn. Recurrence of stillbirth in a second pregnancy was almost 4.5 times more likely, prior to adjusting for confounding factors.
Table 2 presents the unadjusted and adjusted odds ratios obtained by traditional and Bayesian analyses. After adjusting for confounding factors (binary deprivation category, smoking status, interpregnancy interval and second pregnancy factors: year, placental abruption, pre-eclampsia, low birthweight and preterm birth), the odds ratio of recurrence of stillbirth in a second pregnancy were 1.94 (99% CI 1.29–2.92) in women with a previous stillbirth. Interpregnancy interval, placental abruption, preterm birth and low birthweight in the index pregnancy were all also found to be associated with a recurrence of stillbirth in a second pregnancy. The adjusted odds ratio for stillbirth in a second pregnancy in the presence of placental abruption was 1.96 (99% CI 1.60–2.41), whereas adjusted odds ratios for preterm delivery and low birthweight were 7.45 (99% CI 5.91–9.39) and 6.69 (99% CI 5.31–8.41), respectively. Pre-eclampsia in a second pregnancy appeared to be protective for stillbirth in a second pregnancy. No relationship was found with year of delivery.
Table 2. Unadjusted and adjusted odds ratios (99% CI) of a stillbirth in a second pregnancy from traditional analyses and adjusted odds ratios (99% credible intervals) from Bayesian analyses with normal priors
|Characteristics||Unadjusted OR (99% CI)||Adjusted OR (99% CI)*||Adjusted OR (99% CI)*|
|Age at second delivery||1.01 (1.00–1.03)|| || |
|Interpregnancy interval||1.04 (1.01–1.08)||1.04 (1.01–1.08)||1.04 (1.01–1.06)|
|Year of delivery-2000||0.99 (0.78–1.00)||0.99 (0.98–1.01)||1.00 (0.98–1.01)|
|Carstairs’ category (≥3) in second pregnancy||0.73 (0.62– 0.87)||0.93 (0.78–1.12)||0.92 (0.78–1.08)|
|Ever||0.80 (0.67–0.95)||0.94 (0.76–1.17)||0.93 (0.76–1.11)|
|Not known||1.05 (0.88–1.25)||1.08 (0.86–1.35)||1.07 (0.88–1.30)|
|Pre-eclampsia (second pregnancy)||0.91 (0.55–1.50)||0.40 (0.24–0.67)||0.41 (0.25–0.63)|
|Placenta praevia (second pregnancy)||1.81 (0.94–3.47)|| || |
|Abruption (second pregnancy)||6.71 (5.56–8.09)||1.96 (1.60–2.41)||1.96 (1.63–2.35)|
|Preterm delivery (second pregnancy)||30.62 (26.49–35.38)||7.45 (5.91–9.39)||7.29 (5.97–9.05)|
|Low birthweight (second pregnancy)||30.37 (26.30–35.07)||6.69 (5.31–8.42)||6.71 (5.46–8.30)|
|Stillbirth in first pregnancy||4.44 (3.34–5.90)||1.94 (1.29–2.92)||1.59 (1.10–2.33)|
The Bayesian analysis with zero-centred normal priors for the log odds ratios gave posterior medians for the odds ratios that were broadly similar to the frequentist analysis. The inferences drawn from the analyses identified the same factors as being influential for stillbirth in a second pregnancy. The prior distribution variances of value 1 appeared to pull some results towards the null, in particular the result for previous stillbirth. The odds ratio of recurrence of stillbirth in a second pregnancy was 1.59 (99% CI 1.10–2.33) in women with a previous stillbirth.
Summary of findings
On analyzing a large population-based cohort of women with information on their first two pregnancies, the recurrence risk of stillbirth was found to be slightly increased in women with a previous history of stillbirth after adjusting for possible confounding factors. Other factors found to be significantly associated with the recurrence of stillbirth in a second pregnancy were placental abruption, preterm delivery and low birthweight.
Comparison with literature
In general, for the causes of stillbirth that are recurrent, like abruption or pre-eclampsia,11,12 the risk of a further stillbirth is high. However, that still leaves up to 40% of stillbirths where the cause is unknown, and evidence in the literature is conflicting regarding the risk of recurrence of stillbirth in these women.5–7
Results from a previous publication, using a data set that partially overlapped that of the current study, indicated that after adjusting for confounding factors there was no significant increase in the odds of recurrence of stillbirth (adjusted OR 1.2; 95% CI 0.4–3.4); analysis of Scottish national data showed this odds ratio to be increased (adjusted OR 1.94; 99% CI 1.29–2.92).8 Two possible explanations for this difference come to mind. As the number of stillbirths occurring in a second pregnancy in the data set used by Black et al.8 was very small, we cannot rule out a type-II error in our inability to find an increased risk. On the other hand, our inability to adjust for maternal BMI and other possible confounding factors in the current analysis could account for the increased recurrence risk seen in the results. Moreover, whereas the previous study based on data from the Aberdeen Maternity and Neonatal Databank included all women that had a stillbirth after 20 weeks of gestation, the present study used a cut-off of 24 weeks of gestation. The odds ratios for recurrent stillbirth from the Bayesian analysis fell between the results from the previous study and the traditional statistical analysis of the Scottish data. The results of the Bayesian analysis supported an increase of the odds of stillbirth in a second pregnancy in those with a previous stillbirth, as the 99% confidence interval excluded 1.
Strengths and limitations
To our knowledge, this is the largest population-based analysis of recurrence of stillbirth in a second pregnancy. The SMR database has been more than 99% complete since its inception, and is subjected to regular quality checks.13 The internal linkage of subsequent pregnancy records of the same woman was accomplished by means of probability matching, which has been demonstrated to be more than 97% accurate.
Most of the limitations of this study rise from the lack of data in relation to factors such as smoking, BMI and cause of stillbirth – factors that have all been implicated in the literature as being associated with reproductive outcomes. Moreover, the data spanned 20 years, during which several changes have taken place in terms of antenatal care that are likely to influence the outcomes and exposure. Year of pregnancy event was included in the final logistic regression model to adjust for this. A further limitation was the overlap of data from the previous study with that of the current one. To test the inferences of the previous study, ideally both the traditional and Bayesian analyses should have examined only data entirely separate from the data of that previous study.8 However, this was not possible because all identifying information, including location of birth, had been stripped from the Scottish national data set. If the two sets of data could have been separated, the Bayesian analysis could have used informative priors based directly on the estimates and standard errors from the previous study. However, a rough estimate of the overlap with the Scottish data based on inclusion dates could have been as little as 8%, so the impact is likely to be small. Lastly, the analysis of such a large population-based data set is likely to throw up statistically significant differences, which may or may not be clinically relevant. We have taken account of this by setting our statistical significance at 1%. Moreover, it is possible to find biological explanations for all of our results, which are not different from those reported in the literature.
This research adds to the body of evidence regarding the recurrence of stillbirth. Although we found a higher risk of recurrence, this was not as high as previously reported. This information can be used to reassure women with a previous unexplained stillbirth, although obstetric vigilance should not be curtailed in a subsequent pregnancy.
After adjusting for confounding factors, the risk of recurrence was increased in pregnancies following stillbirth.
Disclosure of interests
The authors declare that they have no conflict of interest.
Contribution to authorship
SB analysed the data and wrote the first draft of the paper. MB and AS helped with the clinical interpretation of the results, and contributed to the writing of the paper. GP analysed the data and contributed to the writing of the paper.
Details of ethics approval
Formal ethical approval was not required as routinely collected, anonymised data were used. Approval was obtained from the Privacy Advisory Committee of ISD NHS Scotland.
This project was partially funded by the Chief Scientists Office in Scotland.
The authors would like to thank Ms Joanne Hattie for extracting the data.