In a previous post on predictive analytics, we discussed the variables that determined predictive accuracy. Today’s topic is usability of predictive results.
Old news is only good for wrapping fish. The point of prediction in health care is to head off bad outcomes before they happen. So for predictive usability, data must be timely. In predictive analytics, old data is only useful to assess the predictive ability of a specific model. But if you’re using prediction to, for example, develop care management cohorts, data should be extracted automatically and continuously. Some elements of clinical data can be incorporated into predictive data sets in as little as two days from occurrence. Claims data, however, is anywhere from 2–6 months old by the time it is reported.
Prediction is about action. And action should be taken in areas where it can have the greatest impact, such as improving quality or minimizing costs. For example, if a model does a great job at predicting which patients will develop CHF, the information may not be actionable, since it’s unclear whether providers can prevent such an outcome. On the other hand, many interventions exist to help keep for CHF patients from being hospitalized, making predictive models that predict CHF hospitalizations more valuable.
Predicted results should be easy to act on. Once outcomes are predicted, predictive modeling solutions should make it easy for providers to know what actions to take, and for which patients. Providers should see a list of patients that are stratified based on their level of risk. This risk stratification should also link to actionable lists that highlight each patient’s clinical status and specific care gaps. Then, the provider can prioritize resources accordingly.
For more on usability within predictive analytics, download Optum white paper “Predictive Analytics: Propel Your Data to the Next Level.”