Predicting Readmissions

CMS’ hospital readmissions rate reporting project and reimbursement penalties program are underway.  Many other payers are following suit.  A new study in the Journal of the American Medical Association gives further reason why these programs need to be  stopped until they can be redesigned in light of the issues that research has revealed.   (JAMA Article)   The JAMA study is a review of all the research on all the readmission risk prediction models.  Risk prediction is important for two reasons.  One is that it should guide the use of risk adjusters in the reporting and reimbursement programs.  The second is that if it can be done in real or near-real time, it can help hospitals identify those patients whose discharge needs special attention.

None of the models had any truly strong power to discriminate between patients likely to be readmitted and those not likely to be readmitted.  Of the models based on administrative data, which could only be used retrospectively for hospital comparison purposes, none had any real predictive value, including three that CMS developed for its program.  Models with a real-time use of administrative data for patient intervention had a somewhat better record, but only in a limited population.  Models that used clinical, chart or survey data performed better.  Some of the types of data that seemed to be most helpful were prior use of medical services and functional status.  Only one model explicitly tried to identify potentially preventable readmissions, but with poor results.

None of these models had a level of performance acceptable for their intended uses.  The inability to develop risk prediction models tells us that in fact most readmissions may not be predictable or preventable.  Therefore, reporting even “risk-adjusted” readmission rates probably isn’t giving people any useful information about the quality of a hospital.  And even worse, using those rates to reduce reimbursement is unfair and potentially endangers patient care by reducing hospital resources.  We have made this point several times in the last few weeks and we can only assume that experts have pointed it out to CMS as well, but no change appears to be imminent.  A bad outcome is surely coming.