Randomly and Non-Randomly Missing Renal Function Data in the Strong Heart Study: A Comparison of Imputation Methods

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Diabetes is prevalent in American Indians, often co-occurring with kidney and cardiovascular diseases.
• Longitudinal studies face challenges with missing data due to high morbidity and mortality.
• Missing data in such studies may not be missing at random, potentially compromising findings.

Purpose of the Study:

• To evaluate five data imputation methods for renal function in the Strong Heart Study.
• To compare imputation method performance under different missing data mechanisms (random vs. non-random).

Main Methods:

• Utilized a subset of 2264 participants from the Strong Heart Study with complete renal function data across three exams.
• Assessed listwise deletion, mean of serial measures, adjacent value, multiple imputation, and pattern-mixture methods.
• Employed three missing at random (MAR) models and one non-missing at random (NMAR) model.

Main Results:

• The pattern-mixture method demonstrated superior performance for imputing renal function data.
• This was particularly evident when data were not missing at random (NMAR).

Conclusions:

• The pattern-mixture method is recommended for imputing renal function data in longitudinal studies with NMAR data.
• Identifying the missing data mechanism is crucial for selecting the most accurate imputation technique.