Risk-adjusting for pediatric populations
Successfully managing the health of pediatric populations requires health systems to develop the ability to risk-adjust populations to inform value-based contracting strategies, program development, resource allocation and outcomes measurement. The good news: risk-adjustment methodologies, software interfaces, and management programs have been in place for decades – largely driven by shifts to managed care in adult populations. The challenge: these methodologies, interfaces and programs are weakest in their ability to predict risk for kids.
How do we define “risk?”
Risk-adjustment in healthcare is essentially a predictor of utilization of medical services over a defined period of time. Inputs to risk scoring models are either exclusively or predominantly generated by past healthcare utilization. This information is rolled up in a set taxonomy (e.g., Diagnostic Risk Groupers, or DRGs) and weighted by additional factors including severity or co-morbidities. The resulting risk score reflects projected utilization relative to the average (1.0). The primary data source for this calculation is demographic data and claims data.
Why does it matter?
Risk-adjustment methodologies are only modestly successful in accurately predicting utilization. The most prevalent risk adjustment methodologies explain anywhere from 10 to 30 percent of future utilization in a given population. Or, stated another way, the data used in the model (demographic and claims data) only explain 10 to 30 percent of future healthcare utilization. What factors predict the remaining 70 to 90 percent of utilization? Intuitively, we can guess that the answer is some combination of social and economic determinants, health literacy, patient activation and compliance, and others.
Another important limitation of risk-adjustment methodologies is the tendency to significantly under-predict utilization for more complex, chronic, or acute patients. This matters because these individuals are often frequent users of medical care, with high annual spend. And yet, because of this, these populations tend to lend themselves well to population health management approaches.
While some information is better than no information, it is important to proceed with caution when using risk-adjustment data to design population health programs and contracts. This caution is particularly important when designing pediatric population health programs and contracts.
Why are methodologies so limited in their applicability to pediatric populations? The answer is threefold.
The perennial caution that kids are not small adults holds true for risk-adjustment analysis. Recent studies estimate that children represent less than ten percent of annual health care spending. This disproportion carries through risk-adjustment analyses, which focuses on adult conditions, health experience, and progression of disease or injury that are incongruous or absent in pediatric populations. It is the rare child who requires knee or hip replacements, or who receives a diagnosis of congestive heart failure. As a result, the sensitivity of recognizing risk, and the accuracy of predicting utilization for pediatric populations is significantly limited.
Thankfully, most kids are well, most of the time. Most children utilize medical care for minor acute episodes, or management of pediatric chronic disease. Prospective risk methodologies are particularly challenged for children, because of a natural regression to the mean. If a child breaks his arm last year, it is not correlated with the likelihood of breaking another bone this year. Where pediatric-risk adjustment finds more applicability is in populations of children with special health care needs, who frequently require medical intervention and management.
Claims data is necessary but insufficient in predicting kids’ medical utilization. Population health research tells us that social factors, physical environments, and behaviors – our own and of those around us – have significantly more impact on our health outcomes than factors like medical utilization. We find this to be particularly true for pediatric populations, whose development makes them particularly sensitive to social, environmental and behavioral influences. Claims data is a powerful means of understanding medical utilization, but does not yet meaningfully extend to measure social determinants of health. And, by extension, no risk-adjustment methodologies have found a way to meaningfully incorporate and weight social determinants of health into their calculations.