Tutorial on Data Balancing: Application to Benchmarking Clinicians.

Area of Science:

• Health Services Research
• Biostatistics
• Health Informatics

Background:

• Clinician performance evaluation is crucial but challenging due to varying patient populations.
• Existing methods for performance comparison can be statistically complex and computationally intensive.

Purpose of the Study:

• To introduce data balancing and stratified covariate balancing as accessible methods for benchmarking clinicians.
• To provide a tutorial for readers without extensive statistical backgrounds on comparing clinician performance against peer groups.

Main Methods:

• Patients are categorized by conditions/comorbidities, defining distinct patient types.
• Frequencies of patient types treated by a clinician are measured.
• Expected outcomes are calculated by simulating peer group performance on the clinician's patient distribution, using synthetic cases when necessary.

Main Results:

• Stratified covariate balancing enables direct comparison of clinician performance on a standardized patient cohort.
• This approach bypasses the need for estimating propensity or covariate balancing weights.
• The use of synthetic cases ensures a match for all clinician patient types, facilitating robust benchmarking.

Conclusions:

• Data balancing offers a practical and understandable framework for evaluating clinician performance.
• The described methods are applicable to electronic health records data and can be implemented using Standard Query Language.