Looking Upstream to Improve Healthcare Risks and Value

For those of us who have spent our careers trying to improve individual and population health, the concept of looking upstream has probably not been our first impulse. 

Imagine three friends standing by a river. Suddenly they see a person being swept by them in the rushing water, about to hurtle over the waterfall downstream. Before the friends can act, more people appear, rushing downstream and headed toward the waterfall, all at risk of drowning. One of the friends jumps into the river to try to help, but can only save one person at a time. The second friend grabs a raft that is close by in order to save as many people as possible, but the water is moving too quickly and many more are lost. The third friend, however, is nowhere to be seen. This intrepid individual has run upstream to try to keep any more people from falling into the river.

This common parable, which I heard during a recent talk by David Faldmo, a physician assistant from the Siouxland Community Health Center in Sioux City, Iowa, is used to illustrate the challenge of addressing public health. For those of us who have spent our careers trying to improve individual and population health, the concept of looking upstream has probably not been our first impulse. Like the first two friends in our story, our focus has tended to be on those in front of us who are at immediate risk.

That focus is borne out by familiar statistics that show almost half of healthcare spending is used to treat just 5% of the population (Kaiser Family Foundation, 2012)— typically those with the highest disease burden. But are we actually driving value and impact with our jump-into-the-river approach? Our upstream parable provides the answer, and is proven by a statistical concept.

Regression to the mean—and why heroics aren’t enough

Each year, a cohort of individuals in any population falls into a high-risk health category. In that first year, this high-risk cohort typically receives more (and more expensive) care, which helps to lower the cohort’s risk and healthcare spend in the following year—statisticians call that regression to the mean. (In statistics, regression to the mean is the phenomenon that if a variable is extreme on its first measurement, it will tend to be closer to the average on its second measurement.) But this reduction in healthcare spending the following year is likely to happen whether or not we do anything to address the population cohort as a whole. And by focusing our efforts on those already identified as high-risk, we’re no closer to helping the cohort of those who will be at high risk next year—those who are still upstream but who are just as inexorably being swept toward the waterfall.

One of the key tenets of population health management is to identify and address care gaps within cohorts. Certainly, part of that is caring for those currently at risk. But by looking upstream, we can begin to identify the cohort of individuals who will be at high risk in the future and develop plans to help them get out of the water before the current becomes too strong.

Hierarchical Condition Categories (HCCs) help identify those upstream

Fortunately, there are multiple tools at our disposal to identify future risk. Many existing solutions are valuable and accurate, but are narrow and specific in scope (such as looking at the ten-year risk of cardiovascular disease), or are retrospective in nature (i.e., based on the number of hospital/emergency room visits). There are also a number of global algorithms, including proprietary systems as well as those that are publicly available. One of these is the Hierarchical Condition Categories (HCC), an algorithmic model that was developed by the Centers for Medicare and Medicaid Services (CMS) to adjust payments to private healthcare plans for the health expenditure risks of their members.

While HCCs were originally developed to adjust payment models, they provide an excellent approach for predicting future health risks—in particular, Health and Human Services’ recent refinement to the HCC model. The algorithm takes into account age, gender, and a subset of HCC codes that have been filed in claims during the current calendar year that have been demonstrated to predict healthcare utilization. Each individual is assigned a calculated score, with higher scores representing higher anticipated utilization of healthcare services in the immediate future. Individuals in the general population or in subpopulations (such as those with diabetes) can be ranked to prioritize intervention type and intensity.

HCCs differ from other risk indicators such as hospital or emergency room visits, service and treatment gaps, health risk assessments, and medication counts in several ways. First, HCCs predict future utilization of healthcare spending, and they take into account a broad set of ICD codes. The algorithm is publicly available at no cost and there is good evidence that its predictive capability is not exceeded by other, proprietary calculators. All of these factors make HCCs an ideal holistic resource for identifying cohorts that have the highest likelihood of moving into a high-spend state in the near future.

While HCCs take a broad approach to risk, other approaches can be used to identify cohorts at risk for specific, high-cost conditions such as cardiovascular disease. The American Heart Association offers a pooled cohort risk calculator that takes into account factors that include age, gender, cholesterol levels, and lifestyle factors such as smoking to calculate the likelihood of an individual contracting cardiovascular disease within the next 10 years.

Healthcare delivery systems are continuing to add new data sources to their risk calculations, including clinical data from the EHR as well as claims data and other sources such as patient-generated data and socio-economic data, which can contribute more than 60% of health outcomes. That will enhance providers’ ability to filter populations and identify specific upstream risk cohorts. Additional analysis will proactively identify gaps in care, which care teams can use to engage at-risk patients—at the same time, improving outcomes and reducing costs. 

J. Siemienczuk, MD