Is Risk Adjustment Enough in Population Analytics?

Practical Analytic Approaches to Healthcare Challenges

Most analysts conducting population health data analysis understand that no two populations are completely the same. Even within a population with similar attributes - people with congestive heart failure, for example – there will be variances in the population subgroups. Many factors contribute to these variances, including age and gender, comorbidities and geographic differences in treatment patterns as well as provider unit pricing and efficiency.

Analysis of cost and utilization data, even with the use of “per member per month” and “per 1,000” calculations, does not produce a fully accurate comparison because of variations in risk or severity of the population. For a more “apples to apples” comparison, a common approach is the application of risk or severity adjustment methodologies to the data to normalize utilization and cost across the populations.

Risk adjustment methodologies, such as the Milliman Advanced Risk Adjuster (MARA), Optum’s Episode Risk Groups, or Verisk’s DxCGs, are common tools used to risk adjust results to account for the variations in populations’ risk and severity.

Consider the example below. Four PCP groups are each treating a population of patients who have Diabetes without Coronary Artery Disease as identified using Milliman’s Chronic Condition Hierarchical Groups. Each medical groups’ population is of varying size and consists of a mix of commercial and Medicare patients. In Figure 1 we see that the total medical and pharmacy Allowed PMPM cost for all services received by the identified patients is highest for Group 3 at $1,063 PMPM. (All services, regardless of diagnosis.)

Next, using MARA, we take the risk of each population into consideration. As shown in Figure 2, the population attributed to Group 2 has the highest average MARA concurrent risk score at 3.11, and the Group 3 population’s average concurrent risk is slightly lower at 3.06. A higher overall risk score means that the population’s medical conditions are more severe. Adjusting the cost experience for the population severity allows for a more comparable analysis of the cost of the two populations. We see that, even with the risk adjustment, Group 3’s allowed PMPM is well above the other three groups.

Now, if we want to compare the efficiency of resource utilization in the treatment of these similar populations, an additional adjustment is needed to separate out unit cost versus resource use. Milliman’s GlobalRVUs are a unit value system that covers the entire range of healthcare services, including physician, hospital, DME, and pharmacy.

In Figure 3 we see that additional calculations have been added to our analysis. The Risk Adjusted RVUs PMPM is the total RVUs (RVUs) for each group’s diabetic population divided by the population’s average MARA concurrent risk score, giving us an overall measure of utilization. Allowed per RVU is the total allowed amount (not adjusted) divided by the total number of RVUs associated with the delivery of services. Utilization Efficiency is calculated as the ratio between the PCP group’s risk adjusted RVUs to the overall average.

Using these data, we see that the main driver of the higher risk-adjusted allowed PMPM is that Group 3’s Utilization Relativity Ratio (based on Risk Adjusted RVU PMPM) is 1.15 or 15% higher than the aggregate for the groups and 19% higher than the next highest group (0.97). We also see that relative price, as measured by Allowed Per RVU, is a relatively small contributor to the difference in risk-adjusted allowed PMPM as the Allowed Per RVU for Group 3 ($45.14) is only 2% higher than the aggregate for the group ($44.39).

Applying risk adjustment methodologies to cost of care analysis is the approach many analysts use. But is risk adjustment enough? As shown in the example above, leveraging an RVU methodology in conjunction with risk adjustment can provide additional critical insights into variances due to both population severity and provider efficiency.