Methodological approaches to imputing early-pregnancy weight based on weight measures collected during pregnancy

Abstract Background Early pregnancy weights are needed to quantify gestational weight gain accurately. Different methods have been used in previous studies impute early-pregnancy weights. However, no systematically compared imputed accuracy across different imputation techniques. This study aimed compare four methodological approaches imputing weight, using repeated measures of collected from two cohorts Tanzania. Methods The mean ages at enrollment were 17.8 weeks for Study I and 10.0 II. Given the age distributions enrollment, extrapolated interpolated included: (i) simple based on nearest measure, (ii) arithmetic measures, (iii) mixed-effects models, (iv) marginal models with generalized estimating equations. For model equation methods, was further varying degrees flexibility by fitting splines polynomial terms. Additional analyses included dropping third-trimester weights, adding covariate log-transforming before imputation. Mean absolute error accuracy. Results 1472 women 6272 measures; II 2131 individuals 11,775 measures. Among approaches, had highest (smallest error: 1.99 kg 1.60 Studies II, respectively), while other three showed similar Depending underlying data structure, allowing appropriate degree remote may improve performance. Conclusions Mixed-effects superior performance commonly strategies.