The utility of imputed matched sets. Analyzing probabilistically linked databases in a low information setting

Area of Science:

• Data science
• Epidemiology
• Biostatistics

Background:

• Accurate data linkage is crucial for analyzing motor vehicle crash (MVC) and hospital data.
• Different linkage methods, such as high probability and imputed matched sets, may yield varying results depending on data availability.

Purpose of the Study:

• To compare the performance of high probability matched sets versus imputed matched sets.
• To evaluate how varying levels of linkage information affect the accuracy of matched datasets derived from motor vehicle crash (MVC) and hospital databases.

Main Methods:

• Simulated datasets from multiyear motor vehicle crash (MVC) and hospital databases were used, with known true matches.
• Linkages were performed with different amounts of available information.
• Distributions of high probability and imputed matched sets were compared to true matches for variables like occupant age, MVC county, and MVC hour. Regression models analyzed simulated hospital charges and hospitalization status.

Main Results:

• In high information settings, both methods showed no significant differences compared to true matches for demographic and crash variables.
• In low information settings, high probability matched sets differed significantly for occupant age and MVC county, while imputed matched sets did not.
• Imputed matched sets demonstrated greater robustness in low information settings for analyzing hospital charges and hospitalization status.

Conclusions:

• The amount of information available for data linkage is critical.
• High probability matched sets are suitable for high-to-moderate information settings and case-specific analyses.
• Imputed matched sets are preferable for low information settings, especially for population-based analyses in motor vehicle crash (MVC) research.