Predicting first trimester weight using multiple imputation and a random effects model to evaluate gestational weight gain among Latinas

Background: Self-reported pre-pregnancy weight serves as a proxy for actual pre-pregnancy weight in most studies. However, we observed that self-reported pre-pregnancy weight and estimated pre-pregnancy weight (measured first trimester weight [FTW] minus expected FTW gain) differed by ≥15 pounds (lbs) in 15% of a cohort of immigrant Hispanic women. This discrepancy may result in inaccurate gestational weight gain estimates.

Methods: FTW provides an alternative starting weight because FTW gain is small. A mixed model with a random intercept and random slope for gestational weeks was used to predict FTW in Latinas receiving prenatal care at a community clinic (n=1,197) because many women were missing a measured FTW. Alternatively, the FTW was estimated by subtracting the overall fixed effect slope times gestational weeks from the second measured weight. Results from both methods were compared to measured FTW in a subset of women with both self-reported and measured FTW (n=187). Missing values for covariates thought to influence pre-pregnancy weight were imputed using five replications.

Results: Applying the average rate of weight gain, predicted FTWs were on average 1.3lbs heavier than the measured weight (range: -13.4 to 11.3lbs), with 17.6% of weights differing by ≥5lbs. Using a random coefficients model, predicted FTWs were on average 0.7lbs heavier than the measured weight (range: -5.6 to 5.2lbs), with only 1.1% of weights differing by ≥5lbs and 43.9% of weights differing by <1lb.

Conclusions: A random coefficients model produces more accurate estimates of FTW, suggesting an improved method for evaluating gestational weight gain in Latinas.