Handling over-dispersion of performance indicators.

Area of Science:

• Health Services Research
• Statistical Modeling
• Quality Improvement

Background:

• Performance indicators can exhibit over-dispersion, exceeding expected random variability.
• Ignoring over-dispersion may lead to incorrect classification of healthcare institutions as abnormal.
• Accurate assessment of institutional performance is crucial for quality improvement.

Purpose of the Study:

• To investigate methods for handling over-dispersion in healthcare performance indicators.
• To evaluate the impact of different statistical approaches on institutional classification.
• To identify a robust statistical model for estimating over-dispersion.

Main Methods:

• Retrospective analysis of publicly available healthcare data.
• Examined data included coronary artery bypass graft survival, emergency readmission rates, and teenage pregnancies.
• Utilized funnel plots to visualize the effect of over-dispersion handling methods on institutional banding.

Main Results:

• The choice of method for handling over-dispersion significantly influences institutional banding.
• Both multiplicative and additive approaches are viable, yielding reasonable results.
• The additive random effects formulation demonstrates a stronger conceptual basis.

Conclusions:

• A random effects model offers a potential solution for managing over-dispersion.
• This statistical approach has been adopted by the UK Healthcare Commission for star rating derivation.
• Implementing robust statistical methods enhances the accuracy of healthcare performance evaluation.