Classifying hospitals as mortality outliers: logistic versus hierarchical logistic models

Area of Science:

• Health Services Research
• Biostatistics
• Health Outcomes

Background:

• Hierarchical logistic regression is recommended for provider profiling due to patient clustering within hospitals.
• Standard logistic regression presents challenges in accurately profiling healthcare providers.

Purpose of the Study:

• To compare hospital outlier status using standard logistic regression versus hierarchical logistic modeling for patient mortality.
• To assess the impact of shrinkage estimates on hospital outlier identification.

Main Methods:

• The study analyzed patients admitted to English acute, non-specialist hospitals (2007-2011) with specific diagnoses or procedures.
• Standardized mortality ratios (SMRs) were compared between non-hierarchical and hierarchical models, with and without shrinkage estimates.
• Correlation analysis was performed on SMRs from different modeling approaches.

Main Results:

• Standard logistic and hierarchical models showed highly statistically significant correlation in SMRs (r > 0.91, p = 0.01).
• Hierarchical modeling with shrinkage estimates (Model 2) identified fewer outliers compared to standard logistic regression.
• 21 hospitals shifted from low outlier to not-an-outlier, and 8 from high outlier to not-an-outlier with shrinkage estimates.

Conclusions:

• Both standard logistic and hierarchical modeling methods identify similar hospitals as mortality outliers.
• The choice between methods depends on the modeling aim: judgment versus improvement.
• Shrinkage estimates may be more suitable for judgment-based provider profiling.