Multivariate matching and bias reduction in the surgical outcomes study

Area of Science:

• Health Services Research
• Biostatistics
• Epidemiology

Background:

• Administrative data often lacks detail for outcomes studies, necessitating medical chart abstraction.
• Exact matching in case-control studies limits covariate use, impacting statistical power and cost-efficiency.
• Multivariate matching allows simultaneous consideration of numerous covariates, overcoming exact matching limitations.

Purpose of the Study:

• To develop multivariate matched case-control pairs for analyzing death after surgery in the Medicare population.
• To assess the efficacy of multivariate matching in improving the similarity between cases and controls.

Main Methods:

• Utilized 830 index cases of patients who died within 60 days of admission.
• Identified controls who survived, matched on death risk and patient characteristics using up to 173 variables.
• Included general and orthopedic Medicare surgical cases from Pennsylvania (1995-1996), with controls selected statewide or within the same hospital.

Main Results:

• Matched controls demonstrated significantly greater similarity to cases upon hospital admission compared to typical patients.
• Bias reduction typically exceeded 50% and frequently approached 100%.
• The average difference between cases and matched controls for most variables was less than 0.2 standard deviations.

Conclusions:

• Multivariate matching methods enhance the quality of matches in studies using Medicare claims data.
• These improved matches facilitate a better understanding of the causes of surgical outcomes.
• The approach offers a valuable tool for outcomes research with large administrative datasets.