Addressing Missingness in Predictive Models That Use Electronic Health Record Data

Area of Science:

• Health Informatics
• Clinical Epidemiology
• Biostatistics

Background:

• Electronic health record (EHR) data are crucial for developing clinical prediction models.
• Missing data is a common issue in EHRs, impacting model accuracy and reliability.
• Current guidelines for prediction models offer limited recommendations for handling missing EHR data.

Purpose of the Study:

• To characterize missingness patterns in EHR data.
• To summarize methods for addressing missing data in prediction model development.
• To provide recommendations for validating and implementing prediction models with missing EHR data.

Main Methods:

• Review of existing literature on missing data in EHRs.
• Characterization of systematic and nonsystematic missingness in EHR datasets.
• Summary of statistical techniques for handling missing data in prediction modeling.

Main Results:

• EHR data exhibit both systematic and nonsystematic missingness.
• Various imputation and modeling techniques can address missing data.
• Lack of standardized guidelines for missing data in clinical prediction models.

Conclusions:

• Addressing missing EHR data is essential for robust clinical prediction models.
• Recommendations are provided for model development, validation, and implementation.
• Further research is needed to improve handling of missing EHR data in clinical practice.