Strategies for handling missing data in electronic health record derived data

Area of Science:

• Health Informatics
• Biostatistics
• Epidemiology

Background:

• Electronic health records (EHRs) contain rich data for patient outcomes but pose statistical challenges.
• Missing data in EHRs is a significant issue, potentially invalidating study conclusions.
• Differentiating missing data from negative values (e.g., undocumented conditions) complicates analysis.

Purpose of the Study:

• To focus on analytical approaches for handling missing data in EHRs, particularly multiple imputation.
• To leverage the wealth of variables in EHRs to mitigate bias caused by missing information.
• To highlight the importance of including health status and utilization predictors in imputation models.

Main Methods:

• Focus on analytical strategies for missing data in EHRs, emphasizing multiple imputation.
• Utilize broad EHR variables, including those assessing overall health status (e.g., Charlson Comorbidity Index) and healthcare utilization (e.g., number of encounters).
• Integrate external data sources (e.g., National Death Index, census data) for less biased imputation variables.

Main Results:

• The probability of missing data in EHRs may correlate with disease severity and healthcare utilization.
• Including health status and utilization predictors, even if seemingly unrelated, can improve imputation.
• Linking EHR data with external sources can provide less biased variables for imputation.

Conclusions:

• Multiple imputation, using comprehensive predictor variables from EHRs and external data, is a key strategy for handling missing data.
• Addressing missing data in EHRs is crucial for enhancing the validity of patient-centered outcome research.
• Further methodological research and improved epidemiological training are needed for clinical investigators using EHR data.