Putting Predictions to the Test

Abstract

The Centers for Medicare andMedicaid Services have implemented financial penalties for hospitals with above-average readmission rates, andhospitals aredirectingquality improvementefforts atpredictingand preventing readmissions after surgery. Although clinicians are often aware of the risk factors for readmission, predictive models provide objective data, and their use is preferable over clinical judgment.1 Nomograms are a specific type of tool that offers individualized predictions. Rather than enumerating risk factors or creating risk groupings, nomograms incorporate weights for variables based on their prognostic significance. The quantification of risk factors allows nomograms to give continuous probability estimates of a given outcome. By reducing heterogeneity and taking multiple variables into account simultaneously, nomogram predictions yield a precise prognostication for each patient. In their study, Tevis et al2 develop a nomogram that predicts postoperative readmissions with high accuracy. Ideally, predictivemodels canbeprospectively applied to larger patient populations, and external validation is valuable in this regard. Tevis and colleagues2 use prospective data to successfully validate their nomogram. However, the interpretationofseeminglyobjectivestatisticalmodels isoftenquite subjective.1 Predictions change with time and with the availability of additional data; secular changes in the health care environmentmust be accounted for as well. In fact, there appears to be adecreasing readmission rate over time in this retrospective study,2 and whether this is attributable to formal or informal interventions is unknown. Surgeons will be interested in this nomogram if it provides accurate predictions for their patient populations. Going forward, surgeons will need to focus on gaining a better understandingofwhy readmissionsoccur. The important and oftenunaddressed limitation to this tool andmanyother tools like it lies in what to do with the predictions rendered. What mustwedo toprevent readmissions?When consideringpostoperative readmissions, we must also consider whether an ounce of prevention may be worth a pound of prediction.