Accuracy of hemoglobin A1c imputation using fasting plasma glucose in diabetes research using electronic health records data

Abstract

In studies that use electronic health record data, imputation of important data elements such as Glycated hemoglobin (A1c) has become common. However, few studies have systematically examined the validity of various imputation strategies for missing A1c values. We derived a complete dataset using an incident diabetes population that has no missing values in A1c, fasting and random plasma glucose (FPG and RPG), age, and gender. We then created missing A1c values under two assumptions: missing completely at random (MCAR) and missing at random (MAR). We then imputed A1c values, compared the imputed values to the true A1c values, and used these data to assess the impact of A1c on initiation of antihyperglycemic therapy. Under MCAR, imputation of A1c based on FPG 1) estimated a continuous A1c within ± 1.88% of the true A1c 68.3% of the time; 2) estimated a categorical A1c within ± one category from the true A1c about 50% of the time. Including RPG in imputation slightly improved the precision but did not improve the accuracy. Under MAR, including gender and age in addition to FPG improved the accuracy of imputed continuous A1c but not categorical A1c. Moreover, imputation of up to 33% of missing A1c values did not change the accuracy and precision and did not alter the impact of A1c on initiation of antihyperglycemic therapy. When using A1c values as a predictor variable, a simple imputation algorithm based only on age, sex, and fasting plasma glucose gave acceptable results.