Central to a successful population health management strategy is identifying, assessing and mitigating risk, since you want to be able to implement appropriate interventions for those individuals who will incur the highest healthcare costs. But how do you determine those individuals in your population who represent future risk? There are challenges around such identification, including:
- High-risk individuals may not be participating in the formal healthcare system (through a lack of regular exams or other provider visits) which prevents insight from the physician.
- Some individuals simply don’t present as high-risk today, but because of lifestyle or other factors, they could potentially be so in a year or two.
- Because of financial reasons or a perceived lack of urgency, individuals could be delaying necessary care. This could lead to compounded risk and higher future costs.
Challenges such as these emphasize the importance of predictive modeling, which helps organizations identify and assess future risk and then use that insight to develop interventions that can proactively mitigate it. Given this valuable role, it’s important that such modeling is as robust as it can be. A study published earlier this year makes an important contribution to the field of predictive modeling, by advancing the science of such models with a more accurate framework.
Published by Population Health Management and authored by the Healthways Center for Health Research, “Predicting Health Care Cost Transitions Using a Multidimensional Adaptive Prediction Process” introduces a new method of predictive modeling called the Multidimensional Adaptive Prediction Process (MAPP). MAPP improves upon traditional predictive modeling, which uses a single model based on a limited and fixed set of variables and then applies that across an entire population.
In contrast, MAPP doesn’t rely on a single model – it instead draws from a collection of predictive modeling methods. Nor does it rely on a single set of covariates within a given model, instead allowing for a multitude of covariate combinations within each model. In this study, rather than be estimated on the entire population at one time, MAPP is estimated on four distinct cohorts of the population – low, medium, high and very high – each defined by total annual healthcare cost. By incorporating such a multi-faceted approach based on multiple models, covariate sets and cost cohorts, MAPP more explicitly accounts for the multi-dimensional, dynamic nature of a real-life population.
MAPP was tested on three years of administrative healthcare claims for approximately 25,000 people. A traditional predictive model was also estimated in order to yield a ”status quo” or reference modeling set of results. The two modeling approaches were then compared to see which one was more effective in predicting how individuals would move between cost categories from one year to another.
The study showed that, when compared to the traditional model, MAPP correctly identified 25 percent more individuals who remained in a high-cost cohort and 10 percent more individuals who transitioned from a lower-cost to a higher-cost cohort. Overall, MAPP more accurately identified $17.6 million of incurred healthcare costs.
The confirmation of MAPP as a more accurate methodology for predictive modeling has important applications. MAPP could improve traditional uses of predictive modeling, such as population risk adjustment, premium settings and targeted interventions designed to reduce costs. MAPP could also help health systems, providers and plans improve allocation and planning of their human, technology and physical resources, which could lead to lower cost of care and superior patient outcomes.