The Apixio Blog

A Key for Effective Population Management: Use of Proper Risk Adjustment Data

The way that America pays for healthcare is changing completely. We are moving away from a “fee-for-service” healthcare system, where clinicians are paid for every service provided, to a “fee for value” one where teams of clinicians are paid to keep individuals healthy in a cost-effective way. Studies have shown that “fee-for-service” payment incentivizes too much care, which is often inappropriate or simply wasteful. In a value-based payment world, hospitals and health systems will profit by caring for populations to enable better health outcomes at lower costs.

The first step to cost-effective care is to identify individuals and group them (“stratify”) based upon their likelihood of incurring higher costs over a defined period. Higher costs may result from deterioration of their health (e.g. worsening heart failure, poorly treated diabetes). Studies have demonstrated that over half of all healthcare costs result from caring for the sickest 5% of a population.

If a health system could identify these high-end users of healthcare and target the appropriate resources to them, we could significantly reduce costs to the entire system. For some patients, consistently high costs might be unavoidable, such as someone with end-stage kidney disease on dialysis. For others, costs may be avoidable, such as monitoring and proactive treatment of a person suffering from heart failure to avoid a costly hospitalization.

Still, health systems and payers have historically risk stratified patients in a flawed way. They have used diagnosis, prescription, and procedure codes in medical billing data (“claims”) to target individuals, given its availability. However, using claims tends to either under- or over-represent diseases given the many problems inherent in applying the diagnosis codes.

First, since physicians and their staff place codes on claims primarily to get paid (more than 75% of all health plan contracts remain “fee for service”), there is little attention paid to the specificity of the diagnosis.  For example, a patient may have “pre-diabetes” (which could likely be metabolic disease), but “diabetes” is coded.

Second, claims data lacks particular clinical context for disease, which is an important predictor of future costs (e.g. colon cancer stage 4 has a much different prognosis than stage 1) given the limitation of the diagnosis code set.

Third, physicians are often not familiar with all of the code types available—there are 50 different codes for diabetes—and therefore mis-apply them to claims.

Lastly, diagnosis codes on claims may no longer be relevant for the individual to predict near-term cost.  An example is breast cancer in remission for fifteen years.

Wrong stratification leads to misallocation of resources for population management.  Outcomes suffer, costs continue to creep upwards.  Population health is called into question.  And this is a GIGO problem—garbage in, garbage out.

Rather than relying upon poorly coded or mis-coded information for stratification purposes, health systems should use data abstracted specifically for risk adjustment purposes. In the case of global payment setting on an individual basis, applicable for Managed Medicare or Medicaid, or for transfer payments applicable for non-grandfathered Affordable Care Act (ACA) plan products, there is a yearly process of reviewing patient clinical records to determine whether the encounter visit documentation supports one or more conditions being actively treated.  Each applicable chronic condition and its severity applies towards a “risk score”, a score of how sick a patient is compared to others.  Since these codes are the result of a close read of the medical chart, and attention is paid to the details of conditions which have near-term cost implications, they are well suited for analytics to stratify populations and target the individuals for particular intervention.

What comes out of the risk adjustment process is probably the closest thing that we will have to true understanding of disease and its severity across large populations. It can augment our understanding of populations to enable the right care at lower costs for each individual.

© Apixio 2015