Separating variability in healthcare practice patterns from random error.

Area of Science:

• Health Services Research
• Biostatistics
• Cardiovascular Medicine

Background:

• Improving patient care quality is crucial in cardiovascular hospitalization research.
• Estimating treatment and outcome variability across providers is challenging due to naive statistical methods.
• Existing methods like Gaussian hierarchical and Bayesian models have limitations.

Purpose of the Study:

• To evaluate moment-adjusted imputation for estimating variability in dichotomous treatment patterns and outcomes.
• To compare moment-adjusted imputation with existing statistical methods in quality improvement studies.
• To identify disparities in early physician follow-up after myocardial infarction.

Main Methods:

• A simulation study comparing Gaussian hierarchical, Bayesian hierarchical, empirical Bayes, and moment-adjusted imputation methods.
• Application of selected methods to identify disparities in early physician follow-up post-myocardial infarction using the CRUSADE registry data.
• Focus on density estimation in the presence of measurement error.

Main Results:

• Bayesian methods demonstrated sensitivity to prior and tuning parameters.
• Moment-adjusted imputation performed well, even with modest sample sizes.
• Disparities in early physician follow-up were identified across 225 hospitals.

Conclusions:

• Moment-adjusted imputation is a robust method for quality improvement studies, particularly for analyzing healthcare data with measurement error.
• This method offers an effective alternative to traditional Bayesian approaches, showing better performance and lower sample size requirements.
• The findings highlight significant variations in post-myocardial infarction care, emphasizing the need for quality improvement interventions.