Imputation strategies in the trauma registration

Area of Science:

• Trauma care research
• Biostatistics
• Data analysis

Background:

• Trauma databases frequently exhibit substantial missing physiologic data.
• Multiple imputation (MI) presents a potential solution for handling these missing values.
• Understanding the impact of imputation models on performance metrics is crucial.

Purpose of the Study:

• To assess the influence of simplified multiple imputation (MI) models on the standardized W statistic (Ws) in trauma databases.
• To evaluate the effectiveness of different imputation strategies in managing missing physiologic data.
• To determine if simplified models are adequate for imputing missing data in trauma registries.

Main Methods:

• Utilized data from three Dutch trauma care networks to examine local missing data patterns.
• Developed five multiple imputation (MI) models (MI 1-5) based on existing literature and expert consensus.
• Compared MI models against maximal single imputation and complete case analysis (CCA) using the standardized W statistic (Ws).

Main Results:

• Calculated Ws values across three regions, revealing variations based on imputation methods and missing data prevalence.
• Observed that while significant differences between imputation and CCA were not consistently found, substantial variations emerged in the region with the highest proportion of missing data.
• Found no additional benefit from supplementary variables in the imputation process.

Conclusions:

• Different imputation strategies demonstrably influence Ws values in trauma data analysis.
• Simplified imputation models can adequately impute missing data without compromising results.
• The choice of imputation method is important, particularly in datasets with significant missingness.