The cohort study
We used the HES dataset for nine financial years from April 2000 to February 2009 for the cohort study. In HES, individual patients are allocated the same identifier for each episode of care and core fields contain patient demographics, clinical information, and hospital administrative data. Diagnostic information is coded using the International Classification of Diseases 10th revision (ICD10) and operative procedures are coded using the UK Office for Population Censuses and Surveys classification, 4th revision (OPCS4). For delivery episodes, the HES dataset also has an additional "maternity tail" which includes parity, birth weight, gestational age, method of delivery, and pregnancy outcome. However, not all records that describe a delivery episode have data entered into this tail.
All women who gave birth to a singleton first and second baby from 1st April 2000 to 28th February 2009 were eligible for inclusion. A delivery was defined as an episode of care that included a relevant OPCS4 code (R17-R25) or ICD10 code (O80-O84) for the mode of delivery, or a maternity tail with a valid date of birth for a baby. Deliveries were coded as a cesarean section by the relevant OPCS4 codes (R17 for primary CS or R18 for emergency CS) or if OPCS4 codes were not available, by the delivery method specified in the maternity tail. If neither of the sets of codes were available (0.5% of delivery episodes), vaginal delivery was assumed.
We confined the analysis to NHS trusts (acute hospital organizations) that had a reasonable level of data completeness on parity, defined as having parity information in the maternity tail for more than 50% of the delivery episodes in at least seven of the nine years covered in the study. NHS trusts were included if the proportion of nulliparous women in that trust was between 25% and 55% (which corresponded to the overall rate in England and Wales ± 15%) to remove hospitals with poor data quality. Women were allocated to the NHS trusts that existed in February 2009 to take account of previous organizational mergers.
We used the HES patient identifier to trace the second births of those women who had been recorded as nulliparous and who had singleton first births in identified NHS trusts. Diagnostic information for the second birth was taken from the core HES diagnosis fields. Cases of placenta previa were identified by the ICD10 code O44. Pregnancy complications were identified using ICD10 codes O10, O11, O16 for pre-existing hypertension; O13 for gestational hypertension; O14 and O15 for pre-eclampsia and eclampsia; O24.4 and O24.9 for gestational diabetes; and O40 for polyhydramnios.
The effect of previous CS delivery on the risk of placenta previa was estimated using unadjusted and adjusted odds ratios. Multiple logistic regression was used to calculate the odds ratios and the risk of placenta previa adjusted for maternal age (<20, 20-29, 30-39, ≥40), maternal ethnicity (White, Asian, Black and Other), deprivation, inter-birth interval (<1, 1-2, 2-3, 3-4, 4-5, ≥5 years), placenta previa at first birth and indicators for pregnancy complications (i.e., pre-existing hypertension, gestational hypertension including pre-eclampsia and eclampsia, gestational diabetes and polyhydramnios). Deprivation was measured using the 2004 Index of Multiple Deprivation (IMD) rank of the English Super Output Areas. The IMD combines a range of economic, social and housing indicators into a single deprivation score for each small area in England . Categories were defined by partitioning the ranks of the 32,480 areas into quintiles (0-20th percentiles, 20-40th percentiles, etc.) and labeled 1 (least deprived) to 5 (most deprived). Women were allocated a category based on their region of residence. We calculated the number needed to harm using the adjusted estimates from the logistic regression.
We examined whether the effect of CS on placenta previa rates was related to the level of other risk factors, such as whether the effect of CS on the placenta previa risk differed between younger and older women. The significance of an interaction term between previous CS and other risk factors was assessed with the likelihood ratio test. All analyses were done in Stata/SE 10.0.
The study is exempt from IRB and UK NREC approval as it involved analysis of an existing dataset comprising information on delivery episodes for women who cannot be identified directly or through the HES patient numbers linked to them.
Review of literature and meta-analysis
We searched Pubmed, Embase, Web of Science, Cinahl and the Cochrane Library for the period January 1980 to January 2011 to identify studies that examined the relationship between previous cesarean section and placenta previa. The keywords of "placenta previa"/"placenta praevia" and "cesarean"/"caesarean" were searched in MESH headings, titles and abstracts to locate relevant articles. We also checked the references of the selected articles and previous reviews. We only included articles written in English and limited the search to peer-reviewed journals.
We selected studies in which placenta previa was diagnosed or recorded at third trimester or during delivery. If two or more relevant articles used the same data source in overlapping years, we selected the study that adjusted the effect size by age or parity, and if there is still replication, the study with the larger sample size.
Estimated log odds ratios and standard errors of log odds ratios were calculated from raw data presented in the included papers. The meta-analysis was performed using a random-effects model, and summarized the degree of consistency across the study results using the I2 measure, the percentage of total variation across studies that is due to heterogeneity rather than chance. Meta-regression was used to assess whether the effect size was associated with date of publication, study design (cohort vs. case-control), source of data (population vs. hospital based), method of diagnosis of placenta previa (confirmed at delivery vs. recorded at hospital or birth registry databases), and whether the results were adjusted for age and parity. We also did a subgroup analysis of population-based cohort studies focusing on the association between first-birth cesarean delivery and second-birth placenta previa and compared the magnitude of this association with the results from our cohort.