Ranking hospital performance based on individual indicators: can we increase reliability by creating composite...

Area of Science:

• Health Services Research
• Quality Improvement
• Biostatistics

Background:

• Healthcare report cards increasingly use provider-specific performance indicators.
• Hospital performance data is often presented in 'league tables' for comparison.
• Many individual indicators have low to moderate rankability, limiting accurate hospital comparisons.

Purpose of the Study:

• To define conditions for improving hospital rankability by combining multiple binary indicators.
• To assess the rankability of composite indicators formed from pooling less rankable binary indicators.

Main Methods:

• Utilized Monte Carlo simulations to evaluate the rankability of composite ordinal indicators.
• Examined scenarios with varying prevalences (0.05, 0.10, 0.25) of three binary indicators.
• Assessed the impact of within-hospital correlation (ranging from -0.25 to 0.90) on rankability.

Main Results:

• High rankability of a composite indicator was achieved when component binary indicators were strongly correlated (≥0.5).
• When correlations were weak (<0.5), composite rankability often decreased compared to individual indicators.
• Composite rankability was significantly influenced by the prevalence of the most common indicator and the degree of within-hospital correlation.

Conclusions:

• Pooling highly correlated binary indicators can yield a composite indicator with enhanced rankability.
• Combining weakly correlated or independent indicators may result in lower rankability than constituent components.
• Combine binary indicators to improve rankability only if they represent the same underlying quality of care concept.