Evaluating the validity of multiple imputation for missing physiological data in the national trauma data bank

Area of Science:

• Trauma research
• Data science
• Biostatistics

Background:

• The National Trauma Data Bank (NTDB) frequently lacks crucial physiological data (Glasgow Coma Scale, Respiratory Rate, Systolic Blood Pressure).
• This data deficiency hinders accurate trauma system evaluation and clinical research by compromising comparisons between patient groups.
• Missing data can impact the validity and feasibility of risk adjustment strategies in trauma research.

Purpose of the Study:

• To assess the validity of Multiple Imputation (MI) for imputing missing physiological data within the NTDB.
• To determine the impact of MI on frequency distributions of key variables.
• To evaluate how MI affects the association between physiological data and mortality, and its influence on risk adjustment models.

Main Methods:

• Utilized a dataset of 170,956 NTDB observations with complete physiological data.
• Artificially introduced missing data into the complete dataset to simulate real-world scenarios.
• Applied MI techniques to impute the missing data and compared the results with the original complete dataset.
• Assessed risk adjustment validity by comparing adjusted Odds Ratios (OR) of mortality between observed and MI-generated datasets for 100 hospital pairs.

Main Results:

• Multiple Imputation (MI) successfully preserved the original frequency distributions of physiological variables.
• Associations between physiological data and mortality were maintained after data imputation using MI.
• The median absolute difference in adjusted Odds Ratios for mortality between the observed and MI datasets was 3.6%, indicating minimal impact on risk adjustment.

Conclusions:

• Multiple Imputation (MI) is a valid method for handling missing physiological data in the NTDB.
• When carefully implemented, MI ensures reliable frequency distributions and preserves mortality associations.
• MI does not compromise the accuracy of risk adjustment models used for inter-hospital mortality comparisons.