Gestational Weight Gain Growth Charts Adapted to Japanese Pregnancies Using a Bayesian Approach in a Longitudinal Study: The Japan Environment and Children’s Study

Background Tracking gestational weight gain (GWG) during pregnancy makes it possible to optimize pregnancy outcomes, and GWG growth curves are well suitable for this purpose. The GWG guidelines for Japanese were revised in 2021. However, currently, there are no GWG growth curves to guide women on how to gain weight to meet these guidelines. Methods Using data on 96,631 live births from the Japan Environment and Children’s Study (JECS), we created descriptive GWG percentile curves estimating the trajectory of GWG required to meet the GWG guidelines stratified by pre-pregnancy body mass index (BMI). For both analyses, Bayesian mixed models with restricted cubic splines adjusted for maternal characteristics were used. Results GWG curves substantially differed by pre-pregnancy BMI and were higher among multiparas and those with lower maternal age and with no previous disease. We estimated that underweight, normal weight, overweight, and obese women who gain 8.4 to 11.1 kg, 6.4 to 9.1 kg, 3.8 to 6.5 kg, and <1.9 kg at 30 weeks of gestation are on the trajectory to reach the new guidelines at 40 weeks of gestation. Conclusion We provide GWG percentiles curves for Japanese women, as well as GWG trajectory curves to meet the new GWG recommendations. These results may help pregnant women monitor weight during pregnancy.

A c c e p t e d V e r s i o n Background 21 Gestational weight gain (GWG) is known to influence birth outcomes including birth weight, 22 and risk of preterm and cesarean delivery [1][2][3] . As both inadequate and excessive weight gain 23 are known to increase risk of adverse outcomes, guidelines and recommendations have been 24 created to inform women and practitioners on how much weight should be gained during 25 pregnancy (1,4) . Recently, in March 2021, the Japanese Society of Obstetrics and Gynecology 26 (JSOG) revised its GWG guidelines, which was adapted also by the Japanese Ministry of 27 Health, Labor and Welfare. In these new guidelines, they advise to reach 12 to 15 kg at 40 28 weeks (versus 9 to 12 kg previously), 10 to 13 kg (versus 7 to 12 kg), 7 to 10 kg (versus no 29 official recommendation) and ≤ 5 kg (versus no official recommendation) in pre-pregnancy 30 underweight, normal weight, overweight and obese populations respectively. However, these 31 ranges do not suggest whether the woman is on track or gaining too fast or too slow, and may 32 not be very informative to monitor or to conduct interventions on weight gain during 33 pregnancy. 34 Fetal and child growth is usually monitored via growth charts and there have been attempts to 35 create gestational weight gain growth curves in the same way. Unfortunately, the 36 methodological qualities of studies, older than 2014, have been questioned in a systematic 37 review by Ohadike in 2016. 9 More recent studies published this last decade have used 38 frequentist multilevel models accounting for repeated measurements, with either a restricted 39 cubic spline model or second degree fractional polynomial model for gestational age 8 . The 40 largest study up to date was conducted by Santos and al 5 from multiple cohorts in Europe, the 41 United States and Oceania in which they used generalized additive models for location, scale 42 and shape (GAMLSS) with a Box-Cox t (BCT) distribution 6,7 , a method which is nowadays 43 considered the main method for making growth curves. However, its limitation is it ignores 44 dependence between observations. 45 A c c e p t e d V e r s i o n Compared to these models, Bayesian mixed models may be able to create more 46 accurate curves as it allows the number of measurements and the timing of measurements to 47 vary by individual, allows inclusion of covariates that affect growth in the model, and can 48 overcome power issues due to small sample size 6 in addition to incorporating non-49 independence of measures of each individual as in other multilevel models. While Bayesian 50 mixed models have such strengths, its use remains rare; this has only been applied to 51 construct one fetal growth chart 7 and one gestational weight gain growth charts 12 in our 52 knowledge. 53 Thus, in this study, we aimed to create GWG curves showing the distribution of weight gain 54 among Japanese women using a recent national birth cohort, and from these curves, to 55 suggest GWG trajectory curves to meet the new GWG recommendations.

57
Study population 58 The Japan Environment and Children's Study (JECS) is a nationwide birth cohort. Detailed 59 methodology has been previously reported 13,14 . In brief, pregnant women were recruited 60 through the first antenatal visit at participating co-operating health care providers as well as

96
As the late-pregnancy visit may be close to delivery, some mothers had a missing or same 97 value for either value, in which case only one of the measurements were kept for analysis.

98
Implausible values of weight gain (over 30kg or under -10kg) were removed. The outcome, 99 gestational weight gain, was calculated as the measurement at each visit/delivery minus pre-    In the Bayesian framework, all parameters are random and require prior distribution.

119
The population-level effects (equivalent to fixed effects in Frequentist framework) 120 the overall intercept is the expected value of the first women enter in the study. This 121 parameter follows a normal distribution 0 ~ (3,1) with a mean at 3 corresponding to zero 122 weight gain on (exponential scale -20) with a standard deviation of 1 (2.7kg). In this study, 123 data were available from the week 1 to the 4 th varying according to BMI. Therefore, we 124 specified a realist prior for this parameter to avoid impossible values. The group-level effects (equivalent to random effects in Frequentist framework) 133 represent the deviation in the intercept for the i th woman and follow a cauchy distribution. The error term 138 | is the variance of the i th woman at the j th week that is not explained by the model. We   accurate highest density interval (HDI) (15) . We successed to reach this condition with the 180 normal weight group but we faced to memory issue to compute predictive posteriors. Hence, 181 we reduced the sample size at 12,000 impacting ESS but not estimators and HDI.    raw data available) shows that Chinese women seems to have a 50th percentile slightly higher 244 after 4 months of pregnancy but it seems correspond to other studies at the end of delivery.

245
These studies suggest that Asians in general have a lower GWG compared to Caucasians and 246 suggest that more research in this population is needed.

247
The largest study on GWG curves up to date was conducted by Santos and al 5