Evaluation of two-fold fully conditional specification multiple imputation for longitudinal electronic health record...

Area of Science:

• Biostatistics
• Data Science
• Health Informatics

Background:

• Multiple imputation (MI) methods often fail with longitudinal data due to ignoring temporal structures.
• Standard MI approaches struggle with intermittent missingness in time-series data.
• Existing methods face collinearity and over-fitting issues when handling multiple time blocks.

Purpose of the Study:

• To introduce and evaluate a novel two-fold fully conditional specification (FCS) MI algorithm.
• To assess the algorithm's suitability for imputing missing electronic health records (EHRs) in longitudinal studies.
• To compare the performance of the two-fold FCS algorithm against traditional MI methods.

Main Methods:

• A simulation study was conducted using a generated dataset with missing completely at random data across ten time blocks.
• The two-fold FCS algorithm was applied, conditioning only on temporally local measurements.
• Performance was compared against complete case analysis and baseline-only MI using a time-to-event model.

Main Results:

• The two-fold FCS algorithm demonstrated superior data utilization compared to baseline MI.
• Efficiency gains were dependent on the strength of within- and between-variable correlations.
• The algorithm enhanced the plausibility of the missing at random assumption by incorporating repeated measures.

Conclusions:

• The two-fold FCS MI algorithm is a robust method for handling missing data in longitudinal EHRs.
• This approach offers significant advantages over traditional MI techniques for time-series data.
• The method improves analytical validity by leveraging temporal dependencies and repeated measures.