BMI Z-Score Imputation

The analysis did not conduct sensitivity analyses for missing-not-at-random BMI mechanisms. If BMI missingness is not missing at random, the article says more sophisticated missing-data models would be needed. Missing time-varying BMI z-scores were imputed using a separate linear mixed effects model under a missing-at-random assumption. Predicted BMI values were generated at creatinine measurement times and transformed into BMI z-scores with the UK90 growth reference through the childsds R package. BMI z-scores were calculated using the UK 1990 growth reference because paediatric BMI varies with age and sex.