Predictive Modeling: The Application of a Customer-Specific Avoidable Cost Model in a Commercial Population

Published in Outcomes and Insights in Health Management

Author(s): Adam Hobgood, MS; Karen Hamlet, MA; Chastity Bradley, PhD; Elizabeth Y. Rula, PhD; Carter Coberley, PhD; and James E. Pope, MD

Although the rate of increase in health care expenditures continues to be a major challenge for the United States, a significant portion of this burden could be avoided or at least limited with timely interventions. In an effort to improve quality of life and limit unnecessary medical expenses, a total population health (TPH) approach is becoming increasingly popular among health care service providers. Unlike traditional, disease-specific approaches, TPH is a holistic approach that monitors the entire population (both diseased and non-diseased members) to deliver impactful, personalized interventions to members in greatest need. To better identify members with emerging health risks, Healthways Center for Health Research developed an avoidable cost predictive model that specifically identifies the high-risk segment of the population likely to have near-term, costly yet avoidable inpatient events. This allows prioritization of these patients for personalized programs aimed at mitigating or managing their risk.

The custom-built model was constructed using neural networks based on historical member claims data from a specific customer. A comparative analysis was conducted to assess model performance within the realm of total avoidable inpatient costs. In the comparative assessment, a superior capture rate of avoidable inpatient costs was observed with the avoidable cost model compared with other models. Compared with a model developed to predict high-cost members in a diseased population, the avoidable cost model captured an additional $15 million dollars in total avoidable inpatient costs. Optimal model performance was attributed both to customization of the model specific to the study population of interest and to the target variable of avoidable inpatient costs. Overall, these results demonstrate the success of the newly-developed avoidable cost model in identifying members for cost-effective interventions aimed at identifying and mitigating the factors likely to lead to a hospital admission.

Key Takeaways:

• Special predictive models are needed to support total population health, an approach that does not define individuals by disease, but provides the level and type of intervention appropriate to each person’s individual needs.
• Prediction of future avoidable inpatient events allows programs to maximize value by focusing efforts where there is opportunity for impact.
• Compared to other high-cost predictive models for diseased populations, the avoidable cost model captured an additional $15 million in avoidable inpatient costs.