Adverse pregnancy and perinatal outcome in patients with recurrent pregnancy loss: Multiple imputation analyses with propensity score adjustment applied to a large‐scale birth cohort of the Japan Environment and Children’s Study

Problem Several studies have reported the increased risk of preterm birth, premature rupture of membranes, and low birth weight in patients with recurrent pregnancy loss (RPL). There have been a limited number of large population‐based studies examining adverse pregnancy and perinatal outcome after RPL. Multiple‐imputed analyses (MIA) adjusting for biases due to missing data is also lacking. Method of study A nationwide birth cohort study known as the “Japan Environment and Children’s Study (JECS)” was conducted by the Ministry of the Environment. The subjects consisted of 104 102 registered children (including fetuses or embryos). Results No increased risk of a congenital anomaly, aneuploidy, neonatal asphyxia, or a small for date infant was observed among the children from women with a history of RPL. A novel increased risk of placental adhesion and uterine infection was found. The adjusted ORs using MIA in women with three or more PL were 1.76 (95% CI, 1.04‐2.96) for a stillbirth, 1.68 (1.12‐2.52) for a pregnancy loss, 2.53 (1.17‐5.47) for placental adhesion, 1.87 (1.37‐2.55) and 1.60 (.99‐2.57) for mild and severe hypertensive disorders of pregnancy, respectively, 1.94 (1.06‐3.55) for uterine infection, 1.28 (1.11‐1.47) for caesarean section and .86 (.76‐.98) for a male infant. Conclusion MIA better quantified the risk, which could encourage women who might hesitate to attempt a subsequent pregnancy.


| INTRODUC TI ON
Miscarriage is the most common pregnancy complication with a frequency of 15%. 1,2 Recurrent pregnancy loss (RPL) is defined as two or more losses at any time during pregnancy. 2,3 Most of these occur before 12 weeks of gestation.
RPL is a heterogenous reproductive problem with multiple etiologies and contributing factors include age, body mass index (BMI), parity, and smoking habit. 1,2 Identifiable causes of RPL include antiphospholipid syndrome (APS), uterine anomalies and parental and embryonic chromosomal abnormalities. [1][2][3][4][5][6] Endocrine, infectious, and immune inflammatory conditions have been reported to be associated with RPL. Of these, APS is the treatable etiology. 1,2 There has been no randomized control trial to compare the live birth rate between with and without surgery for a uterine anomaly or preimplantation genetic diagnosis for a translocation.
However, the cumulative live birth rate was 84% in non-genetic carriers and 85.5% in couples with no explanation in the previous studies. 5,7 The live birth rate decreased significantly according to the number of previous miscarriages in both groups. 8 Information on the adverse pregnancy and perinatal outcome is available but limited. 9,10 The risk of a preterm birth (PTB) is most frequently examined. [9][10][11][12][13][14][15] Other previous studies have shown a significantly increased risk of very PTB. [13][14][15][16] Premature rupture of membrane (PROM), 12,14,17 low birth weight (LBW), 11,13,15 caesarean section, placenta abruptio, and hypertensive disorders of pregnancy (HDP). 17 However, there has been controversy regarding whether RPL increases the risk of a congenital anomaly or neonatal asphyxia. A casecontrol study with 18 534 malformed and 17 544 non-malformed babies indicated that multiple malformations, Down's syndrome, anencephaly, spinabifida, talipes equinovarus, congenital dislocation of the hip and LBW were associated with previous miscarriage and stillbirth. 11 Another study with 638 recurrent miscarriages (RM) patients and 3099 non-RM patients also reported a risk of congenital anomalies with RM. 12 A recent study found no association between RM and congenital anomaly or aneuploidy. 10 These studies did not employ a population-based cohort, but rather case-control retrospective approaches. The influence of covariates and medical histories were not considered. 11,12 Furthermore, recent concern regarding the treatment of missing data has been raised because the generalizability of findings might be limited by the extent of the missing values.
We have conducted the nationwide population-based birth cohort study known as the "Japan Environment and Children's Study (JECS)" planned by the Ministry of the Environment, Government of Japan. [18][19][20][21][22] The study subjects consisted of 104 102 registered pregnancies recruited during the first 3 years of the JECS, and their babies are now being followed up for 13 years mainly to examine the influence of the uterine environment on the fetus.
We determined the adverse pregnancy and perinatal outcome according to the number of previous pregnancy losses (PL) reported by the JECS with the use of multiple imputation analyses (MIA). Eligibility criteria for expectant mothers were as follows: that they (i) resided at the time of recruitment in any of the study areas selected by 15 Regional JECS Centers located countrywide, (ii) had an expected delivery date after August 1, 2011, and (iii) were capable of comprehending the Japanese language and completing the selfadministered questionnaire. [18][19][20][21][22] The sample size has been calculated in the JECS protocol by the Ministry of the Environment. 23 In principle, pregnant women completed the questionnaire during the second (MT1) and third trimester (MT2). Their medical records were transcribed by doctors or research coordinators at registration (DrT1), just after delivery (Dr0m) and at 1 month after delivery (Dr1m).

| Study design and participants
The present study was based on the jecs-ag-20160424 dataset, which includes 104 102 registered children (including fetuses and embryos), and was released restrictively to all concerned in July, 2016 ( Figure 1). A total of 1994 children of mothers with multiple pregnancies were excluded because several outcomes were influenced by multiple pregnancies. Furthermore, 310 fetuses or embryos terminated by induced abortion were also excluded. In addition, a total of 5586 children (fetuses or embryos) whose mothers had participated for the second or the third time were excluded.
Finally, 96 212 participants were included in the main analysis. The mean (SD) age at registration was 30.7 (5.1). The mean (SD) gestational weeks at registration was 14.0 (5.7) weeks.
The JECS protocol was reviewed and approved by the Ministry of the Environment's Institutional Review Board on Epidemiological Studies and by the Ethics Committees of all participating institutions.
Written informed consent was obtained from all participating women.

| Outcome, exposure, and covariates
The pregnancy histories provided by the doctors in filling out the DrT1 form were given priority over the participants' answers with regard to the number of previous PL (categorized as 0, 1, 2, 3 or more).
The obstetric outcomes included stillbirth > 20 weeks' gestation, late miscarriage, early miscarriage <12 weeks' gestation, PL, PTB (<37 and <34 weeks' gestation), PROM, placenta praevia, abruptio placenta, adherent placenta, oligohydramnios, mild and severe HDP, uterine infection and caesarean section. The perinatal outcomes were small-for-date of the 10th percentile (SFD), IUFD, sex, Apgar score <7 at 5 minutes, an umbilical artery blood pH <7.1, the presence/absence of congenital anomalies of the head, eyes, ears, face, limbs, lungs, heart, intestine, urogenital organs, skin, and skeleton and chromosome aneuploidy. The presence of chromosomal aneuploidy as indicated by the doctors in filling out the Dr1m was given priority over the Dr0m.
Potential covariates were maternal age at registration, BMI, marital status, the presence/absence of IVF-ET, previous live birth, smoking, education, and income.

| Statistical analysis
The associations between the number of previous PL and covariates were tested using chi-squared tests for categorical variables.

Main Data
Associations between each outcome and the covariates were examined using Fisher's exact test.
The prevalence of each outcome including the crude odds ratio (OR) and the 95% confidence interval (CI) were calculated for 0 (reference), 1, 2, and 3 or more PL (as categorized). We conducted both complete-case analysis and MIA using logistic regression to estimate the risks of previous PL affecting the adverse pregnancy and perinatal outcome. In the complete-case analysis, the unadjusted model (Model I) was used to estimate the independent effects of the PL history, and the adjusted model (Model II) included potential confounders in the model, adjusting for the propensity score including all covariates with P values <.05. 24,25 In general, it has been found that propensity score adjustment has the advantages of being less biased, providing more robust estimates than those of the traditional logistic regression. An unlimited number of covariates can be theoretically included in propensity score estimation so that propensity score adjustment has been employed when adjusting for a large number of covariates and widely used in recent observational research.
Adjusted Model III in accord with Model II included MIA. Since the percentage of missing values in Model II across all variables ranged from 5.2% to 21.1%, prior to conducting logistic regression analyses using propensity score adjustment, MIA was applied to handle missing data.
Five imputed datasets were created for each outcome, and all variables that were used in Model IIs were included in each imputation model using the propensity score. Use of MIA is more efficient in most settings and widely recommended for improving biases within complete-case analyses. 26 Model IV additionally adjusted for covariates related to the gynecological history in Model III. It was speculated that women with PL could have a greater chance of being diagnosed with a gynecological disease as a result of frequent ultrasound sonography. Thus, we performed MIA with and without covariates for gynecological history.
All calculations were carried out using SPSS version 23 and 24 (IBM Corp., Japan).

| RE SULTS
Histories of previous PL were available for 94 658 participants out of 96 212. Among these women, 77.6% (73 413) had no history of PL, 17.5% (16 547) had experienced one PL, 3.8% (3633) experienced two losses, and 1.1% (1065) experienced three or more losses (Table 1). Age, BMI, IVF-ET and previous live births were positively associated with the number of previous PL. The rates of marital status, smoking, education background, and income were influenced by the number of previous PL. All the listed variables were significantly associated with an increasing number of previous PL (Table 1, P < .0001 for all covariates). In the following analyses, women with no PL were regarded as the reference category.

| Model I (crude analysis)
Crude analysis showed a significant association with stillbirth, PL, PTB (both <37 and <34 weeks' gestation), placental adhesion, HDP, caesarean section and a male infant with three or more PL (Model I in Tables 2 and 3). There was no risk associated with PROM, placenta praevia, oligohydramnios, abruptio placenta, or uterine infection. No association with SFD, IUFD, LBW, low Apgar score, low pH, congenital anomaly, or aneuploidy was observed.

| Model II (the adjustment of multiple covariates)
After the adjustment of multiple covariates with significance for each outcome, stillbirth, PL, placental adhesion, HDP, caesarean section and a male infant remained significantly associated with three or more pregnancy losses (Model II in Tables 2 and 3). The significance of PTB (both at <37 and <34 weeks' gestation) disappeared after the adjustment.

| Model III (the adjustment with MIA)
After the adjustment with MIA, a novel increased risk of placental adhesion and uterine infection was found (Model III in Table 2 Tables 2 and   3). Regarding stillbirth, ORs tended to be increased and was significantly higher with three pregnancy losses. With regard to HDP and caesarean section, the prevalence for women with three pregnancy losses was significantly higher. Marginally increased risks were 1.18 (.96-1.46) for LBW. There was no risk of a very PTB <34 weeks, PROM, placenta praevia, oligohydramnios or abruptio placentae after adjustment. As for placenta praevia, the prevalence tended to increase, but that of women with three pregnancy losses was not significant.

| Model IV (the adjustment with MIA including gynecological histories)
Similar results were obtained with MIA including covariates for gynecological history (Model IV, Tables 2 and 3).
The risk of placental adhesion and uterine infection was analyzed according to the number of induced abortions. The results were similar to those in RPL (Table S1).

| Analysis including multiple participants
In another supplementary analysis to assess the risk as it appears in a clinical setting where patients present with a range of pregnancy

| D ISCUSS I ON
The present study revealed no risk of a congenital anomaly, aneuploidy, neonatal asphyxia, or SFD related to RPL. We found a novel significant association between RPL and uterine infection and placental adhesion. The risk might be due not to RPL pathology but to surgery because the results of analysis according to the number of induced abortions were similar to those in RPL (Table S1) and 79.4% of facilities use curettage at induced abortion in Japan. 28 Surgical management of the miscarriage using curettage is mainly selected. The present study found no association with a PTB <34 weeks or PROM after MIA in contrast with previous studies. [12][13][14][15][16][17] Further study with the use of MIA is necessary to confirm this association.
HDP and caesarean section were both associated with RPL. This is in line with the results of a population-based study with 154 294 women, which indicated an increased risk of caesarean section, placenta abruptio, and hypertensive disorder after two or more miscarriages. 17 Recently, 472 variants in 187 genes have been reported to be associated with RPL. A meta-analysis revealed a significant association between RM and 21 genetic variants with ORs .51-2.37. 29 Common risk alleles such as annexin A5 might influence both HDP and RPL. 30,31 Finally, women with three or more PL had a lower tendency to have a male infant on the index pregnancy A previous study proved that boys were significantly more common than girls among births prior to a secondary RM and the chance of a live birth after RM is lower in those with a firstborn boy compared with a firstborn girl. 32 The study also revealed that birth of a girl was a significantly more common outcome of a live birth after a secondary RM, and that the maternal carriage of male-specific H-Yrestricting HLA class II alleles was associated with the reduced birth rate of boys.
The major limitation was that there was no distinction among different etiologies for RPL, nor whether interventions were performed. The prevalence of early miscarriage was only .36% because many of the participants were recruited after 10 weeks' gestation.
Thus, it was one of the limitations that the early miscarriage and pregnancy loss results might not be reliable.
The present study represents the largest nationwide birth cohort study in Japan. The results are reliable because pregnancy and delivery information was drawn from medical records by doctors and research coordinators. The comparisons between completecase analyses and MIA allowed for a relevant sensitivity analysis to quantify the risk of a response bias. MIA allowed for a reliable estimation of the results and helped to minimize the risk of bias.
ORs tended to be lower after the adjustment with MIA when they were compared using Model II and III. Many of covariates included in propensity score increased the rate of missing data, which might lead to over-adjustment in Model II. Thus, MIA might insure the stability of the results in Model III, compared with those in Model II. On the other hand, there were no remarkable differences were observed in the results of Model III and Model IV, suggesting few effects from covariates related to the gynecological history. We, therefore, assumed that the results of Model III as the most important and reliable estimation minimizing the risk of bias.
This information, especially the finding that there was no increased risk of a live birth with a congenital anomaly or aneuploidy in women with a history of RPL as compared to women with no history of pregnancy loss, could encourage women who might hesitate to attempt a subsequent pregnancy. Many patients with RPL are afraid that their baby will have an anomaly because an abnormal embryonic karyotype is the most common cause of RPL.