Comparison of Methods for Interpolating Gestational Weight Gain Between Clinical Visits in Twin and Singleton Pregnancies.

Abstract Purpose Researchers are interested in studying longitudinal patterns of gestational weight gain, yet this requires daily/weekly weights, and maternal weight is measured only during prenatal visits. We evaluated the relative accuracy and precision of methods for estimating maternal weight between prenatal visits among twin and singleton pregnancies. Methods We analyzed cohorts of dichorionic twin and singleton pregnancies delivered from 1998-2013 in Pittsburgh, Pennsylvania. We mimicked a typical study by retaining pre-pregnancy, first prenatal visit, glucose screening visit, and delivery weights, using these to fit interpolation models, estimating weight throughout pregnancy using 16 different methods, and calculating the difference in kilograms between predicted and measured weight gain among remaining weights. We evaluated performance of each model by calculating root mean squared error (RMSE). Results RMSE ranged from 1.55 to 6.09 kg in twins (n=2067) and 1.45 to 4.87 kg in singletons (n=7331). The most accurate and precise methods incorporated restricted cubic splines, random intercepts and slopes for pregnancy, and internal knots demarcating trimesters/quantiles (RMSE=1.55/1.56 kg in twins, 1.45/1.45 kg in singletons), while individual-level linear interpolation between proximal measurements also performed well. Conclusions Accuracy and precision of methods for estimating maternal weight between measurements differed by method, and was best among individually-tailored and flexible models.