Readmission Risk Models Display Poor Predictive Ability
Current hospital readmission risk models for comparative or clinical purposes perform poorly
TUESDAY, Oct. 18 (HealthDay News) -- Most hospital readmission risk models have poor predictive ability, according to a review published in the Oct. 19 issue of the Journal of the American Medical Association.
Devan Kansagara, M.D., M.C.R., from the Portland Veterans Affairs Medical Center in Oregon, and colleagues investigated the performance of validated readmission risk prediction models, and evaluated their suitability for clinical or administrative use. Studies describing models tested in both validation and derivation cohorts were identified, of which 30 studies of 26 unique models met the inclusion criteria. Information on population, setting, sample size, follow-up interval, readmission rate, model discrimination and calibration, the type of data used, and the timing of data collection was extracted. The most common outcome used was 30-day readmission, with only one model addressing preventable readmissions.
The investigators found that, of the 14 models that could potentially be used to risk-adjust readmission rates for hospital comparison based on retrospective administrative data, nine were tested in large U.S. populations and had poor discriminative ability (c-statistic, 0.55 to 0.65). Seven models could potentially identify high-risk patients requiring interventions early during hospitalization (c-statistic, 0.56 to 0.72) and five models identified patients at hospital discharge (c-statistic, 0.68 to 0.83). Of the six studies comparing different model performance in the same population, model discrimination was improved with functional and social variables.
"Most current readmission risk prediction models that were designed for either comparative or clinical purposes perform poorly," the authors write.