Validity of different algorithmic methods to identify hospital readmissions from routinely coded medical data.

Area of Science:

• Health Services Research
• Healthcare Quality Improvement
• Health Informatics

Background:

• Hospital readmission rates are key metrics for healthcare quality and pay-for-performance programs.
• Administrative data methods, such as Centers for Medicare & Medicaid Services (CMS) unplanned readmissions and SQLape potentially avoidable readmissions, are used to identify readmissions.
• The comparative criterion validity of these methods for identifying actually avoidable readmissions remains unclear.

Purpose of the Study:

• To compare the criterion validity of the CMS and SQLape methods in identifying hospital readmissions.
• To determine which method more accurately identifies actually avoidable readmissions through manual record review.

Main Methods:

• Seven independent reviewers manually reviewed 738 inpatient stays across seven hospitals.
• Evaluated the sensitivity, specificity, positive predictive value (PPV), and F1 score for original and adapted CMS methods and the SQLape method.
• Assessed the ability of each method to identify unplanned, potentially avoidable, and actually avoidable readmissions.

Main Results:

• Both CMS method versions showed higher sensitivity (92%/86% vs. 62%) and PPV (84%/91% vs. 71%) than SQLape for their respective outcomes.
• For identifying actually avoidable readmissions, CMS methods maintained higher sensitivity (90%/85% vs. 66%).
• Positive predictive value did not significantly differ between methods when identifying actually avoidable readmissions.

Conclusions:

• The CMS method exhibits superior criterion validity and sensitivity for identifying actually avoidable hospital readmissions compared to the SQLape method.
• The findings support the primary use of the CMS method in healthcare quality initiatives.
• Accurate identification of avoidable readmissions is crucial for effective quality improvement strategies.