As the federal government moves toward policies that base reimbursement on hospital readmission rates, a new study shows that many risk-prediction models do a poor job of forecasting which patients are at high or low risk for repeat hospital stays, says a Medscape Medical News article.
From a review of 7,843 citations, Devan Kansagara, MD, and colleagues from the Veterans Affairs Medical Center in Portland, Oregon, identified 30 studies and 26 models for predicting hospital readmission risk. The authors identified few models capable of reliably identifying preventable readmissions. Their findings, published in the October 19 issue of JAMA, suggest that the models generally fail to properly consider the social and behavioral variables that combine with clinical factors to determine readmission risk, the article says.
According to the study, a special calculation, called a c statistic, was especially useful for assessing the relative ability of various models to discriminate between patients who are at high and low risk for readmission. A c statistic of 0.5 indicated the model was no better than a coin flip for predicting readmission. A c statistic of 0.7 to 0.8 correlated with modest or acceptable discrimination ability. Good results are more likely when the c statistic is greater than 0.8.
Of 286 potentially relevant articles identified from the medical literature, 21 studies reported c statistics with values from 0.55 to 0.83. Only 6 of the studies reported a c statistic above 0.7, indicating a modest ability to discriminate risk. Even among these top performers, the predicted 30-day readmission rates for patients differed by from 20.4% to 34.5% from the actual readmission rates, Medscape says.