Empirical Bayes methods for estimating hospital-specific mortality rates

Area of Science:

• Health Services Research
• Biostatistics
• Health Economics

Background:

• Current methods for estimating hospital mortality rates, like those used by the Health Care Finance Administration (HCFA), may not fully account for patient severity.
• Identifying exceptional hospitals based on unadjusted mortality rates can lead to the problem of multiple comparisons and inaccurate conclusions.

Purpose of the Study:

• To present alternative, more accurate methods for estimating hospital-level mortality rates for Medicare patients.
• To develop a model that accounts for patient severity and provides reliable standard errors for mortality rate estimation.
• To address the issue of multiple comparisons when identifying high-performing or low-performing hospitals.

Main Methods:

• Utilized an empirical Bayes model to capture variations in observed hospital-specific mortality rates.
• Employed a logistic regression model to adjust for patient mix and severity differences across hospitals.
• Analyzed national Medicare mortality data and national samples with patient severity descriptors for four disease conditions.

Main Results:

• Observed substantial variation in unadjusted death rates across hospitals using national data.
• Found significant differences in mortality rates based on patient severity in the developed models.
• Limited sample sizes in national samples restricted reliable estimation of between-hospital variation in adjusted mortality rates.

Conclusions:

• The proposed model-based approach offers more accurate hospital mortality rate estimates compared to traditional methods.
• The empirical Bayes and logistic regression models provide a principled way to derive standard errors and resolve multiple comparison issues.
• Further research with larger sample sizes is needed to reliably estimate between-hospital variations in adjusted mortality rates.