Authors: Wells, A. R., Guo, X., Coberley, C. R., & Pope, J. E.

Decades of research exist focusing on the utility of self-reported health risk and status data in health care cost predictive models. However, in many of these studies a limited number of self-reported measures were considered. Compounding this issue, prior research evaluated models specified with a single covariate vector and distribution. In this study, the authors incorporate well-being data into the Multidimensional Adaptive Prediction Process (MAPP) and then use a simulation analysis to highlight the value of these findings for future cost mitigation. Data were collected on employees and dependents of a nationally based employer over 36 months beginning in January 2010. The first 2 years of data (2010, 2011) were utilized in model development and selection; 51239 and 54085 members were included in 2010 and 2011, respectively. The final results were based on prospective prediction of 2012 cost levels using 2011 data. The well-being–augmented MAPP results showed a 5.7% and 13% improvement in accurate cost capture relative to a reference modeling approach and the first study of MAPP, respectively. The simulation analysis results demonstrated that reduced well-being risk across a population can help mitigate the expected upward cost trend. This research advances health care cost predictive modeling by incorporating well-being information within MAPP and then leveraging the results in a simulation analysis of well-being improvement.

Key Takeaways:

  • This study applies Multidimensional Adaptive Prediction Process (MAPP) to administrative claims and self-assessment data to evaluate the improvement in predicting future healthcare cost. MAPP is a methodology in which multiple econometric models are estimated for cohorts of the population as opposed to applying the same model across the entire population.
  • Covariates that asses an individual’s evaluative and experienced well-being captured in the Well-Being Assessment (WBA) were more significant leading indicators of future cost acceleration for cohorts with lower initial cost. This finding indicates that this type of data is useful in predicting future health care cost among individuals that rarely interact with the health care system.
  • Results show that the use of innovative predictive modeling techniques and the inclusion of well-being data resulted in improved model accuracy. The ability to predict future healthcare costs across an entire population is critical in identifying subsets of the population with escalating cost potential and more efficiently delivering population health improvement programs.