3M Health Information Systems
Integrating Sociodemographic Factors into Risk Adjustment: Important Considerations for NQF’s “Robust Trial Period”
Why is it necessary that risk adjustment incorporate sociodemographic factors for my diabetic schizophrenic patients who have unstable housing?
Healthcare is fundamentally about people. That’s why, at the end of the day, it is the differences and disparities among individuals that are at the heart of the challenge facing the National Quality Forum (NQF) as it debates incorporating sociodemographic factors into risk adjustment.
Here’s a real-life example of the importance of SES factors to risk adjustment: Robert is a diabetic patient of mine who is schizophrenic with episodes of psychosis. He has difficulty with his meds in part because his housing situation is not stable. From time to time he is homeless. If there is any possibility of stabilizing his diabetes, he will need additional case management time over and above a diabetic schizophrenic who does not have the added SES burden. The case manager would not just deal with “medical” issues like making sure that Robert is taking his meds every day but also working with Robert to address conflict with neighbors that in turn are making him extremely anxious. In this case, the neighbors were extremely rowdy with loud music. The case manager was able to defuse the situation – when the neighbors were told by the housing authority to move. The same challenge applies to my asthmatic patients who live in substandard housing and are exposed to different allergens than those impacting middle-class asthmatics. In this situation, the case manager might help with making sure that insects exacerbating the asthma attacks are eliminated from the apartment.
As the health system moves toward value-based payment, which is directly linked to more effective population health management, the incorporation of SES factors as part of risk adjustment is critical. This was clear in the comments NQF received in response to its technical advisory board’s final report, “Risk Adjustment for Socioeconomic Status or Other Sociodemographic Factors”:
“NQF recommends that outcome measures be adjusted for clinical severity because it affects outcomes, but up until now we have not recommended adjustment for sociodemographic factors, in part because of their link to disparities…NQF received more public comments on this topic than any other project to date. (emphasis added) At the heart of it though, people want performance measures to provide fair comparisons across those being measured, but also agree that we cannot lose sight of disparities in healthcare and health faced by disadvantaged patients or ignore the challenges of the providers and health plans that care for them…In its deliberations on the report’s policy implications…the Board emphasized the need for a time-limited, robust trial period and strongly urged the field to develop and use sociodemographic-adjusted measures.”
With any policy decision, there are both political and scientific aspects to the NQF board decision. Despite the fact that over 95% of the comments supported incorporation of SES factors, the NQF board opted for a time-limited robust trial period. While many of us will try to address this issue politically, this blog will focus on key considerations in defining the meaning of “robust trial period”.
Risk-adjustment of patient populations is critically important to ensuring there are appropriate incentives to improve patient outcomes and guard against risk-selection practices seeking to “cherry pick” those that are easiest to treat at the least cost. CMS has used Diagnosis Related Groups (DRGs) for over 30 years as the basis for the dual tasks of paying for and managing hospital-based services. However, when DRGs were implemented for payment purposes, policymakers recognized the need to adjust the DRG payment system for unmeasured factors that could not be sufficiently explained by the clinical risk-adjustment alone. Additional adjustment factors were incorporated for indirect medical education (IME) and disproportionate share (DSH) to reflect differences in average hospital cost resulting from different patient populations and different hospital missions. (This was outlined in a recent blog by my colleague, Rich Fuller.)
The DSH adjustment reflected the impact of serving a disproportionate share of low-income populations on the relative costs faced by hospitals. The adjustment for DSH was made outside of the clinical classification model and was quantified from the residual variation after applying the clinical risk-adjustment model, (i.e. from what has not been measured directly but observed to exist). Over time, the empirical magnitude of the adjustments has been seen to decrease most likely as a result of refinements to the DRG classification and coding improvement by large inner city hospitals that better reflect the illness severity of their patients.
We can learn several things from this experience. It is not only reasonable to test for effects that SES may have on a risk-adjusted measure, but it should form part of a prudent approach to health service financing. By making the adjustment exogenous to the clinical risk-adjustment we permit transparency and review. As we improve and data provided improves we have to revisit any SES adjustment and see if our assumptions hold. It is likely that in many situations, particularly those under direct provider control of limited duration, SES adjustments will not be warranted. In situations where outcomes depend upon interaction with the wider community or the diversity of patient types subject to measurement is greater, it is more likely that SES adjustments will be required for fair evaluation and to avoid risk-selection. The need for an SES adjustment should be tested and, if found necessary, should be applied transparently with ongoing evaluation as the system evolves.
To account for SES factors, the methodologies used for payment and improving outcomes should be required to take full advantage of currently available information. This should require taking full advantage of claims data to account for as many SES factors as possible, which means relying on severity adjustment and detailed clinical categories. Chronic illnesses not only need to be stratified by severity of illness, but there should be clinical distinctions that specify the clinical detail that is available. For example, a patient with schizophrenia needs to be stratified by severity of illness. However, if the individual also has diabetes, he or she should be identified in a separate risk category. Without this detailed approach to risk categorization, it is likely that managed care organizations will assiduously pursue adverse risk selection—in other words, create incentives to provide treatment for individuals who are least likely to generate high medical expenses and to limit services to high-utilization populations. One large Medicaid HMO enrolling dually eligible individuals (Medicare and Medicaid) had a specific program for hemophilia case management that attracted a large number of these patients. Not surprisingly, they stopped the program as their reimbursement rates did not recognize this specific category of highly complex individuals. The same applies to quadriplegic or paraplegic patients – virtually all of whom are dually entitled – on both Medicare and Medicaid. Unfortunately, current risk adjustment methodologies do not take into account chronic illness burden, resulting in inaccurate capitation risk adjustment and inability to track and incentivize better outcomes.
In addition, there are research initiatives that document the impact of data elements that are not routinely available, but the research literature indicates that they are important SES factors and should be taken into account. Homelessness, for which there are good accepted definitions, is such a factor. The Housing and Urban Development definition of homelessness should be considered for adoption. Functional and mental health status is collected in several CMS reimbursement programs such as home health. The current proposed and bipartisan legislation The Better Care, Lower Cost Act (S. 1932/H.R. 3890) requires the collection of health status. In addition, the following data elements should be considered: Zip code or standard metropolitan region; Developmental disability status; Living in group home; Home Health Status (e.g. is there support at home).
From a statistical point of view, exogenous adjustment for SES factors is the recommended approach (as is done for DSH in the inpatient PPS). Calculating a single, explicit, and exogenous factor to be used in conjunction with the risk-adjustment model retains separation of performance measure relative to the individual patient from the community/provider level SES adjustment. Unlike the exogenous IME or DSH adjustments in PPS, which were calculated from the unmeasured (residual) effects at the community/provider level, the advent of more comprehensive data, and a willingness to expand this data further, permits calculation of an adjustment factor across classes of individuals. An empirically calculated SES adjustment factor based upon observation of variation of individuals can be summed to provide an aggregated adjustment for SES impacts that may be applied to communities and providers with a disproportionate level of patients in lower SES statuses. The SES adjustment is explicit and empirically derived, rather than derived as part of a “black box” peer grouping process. This way it can be assessed to determine whether its magnitude makes sense or if it varies across measures or locations. No single provider could claim to be treated unfairly because the method upon which the adjustment is based would correctly be subject to public scrutiny and comment. Put simply, it would be transparent. It would also focus efforts to reduce the need for such an adjustment on the community services needed to address the issue.
In summary, based on the discussion above, the following should be key considerations during the NQF “robust trial period.”
- The risk adjustment needs to be clinically detailed and categorical in nature
- The risk adjustment must have face validity—that is, have credibility beyond statistical validity. For example, not including significant mental health disorders as a factor in the risk adjustment for preventable readmissions may not have statistical validity. In other words, the risk adjustment should include significant chronic mental health disorders whether or not it adds statistical validity to the readmission risk adjustment measure. Without it, health professionals in the field know that it is clinically invalid.
- Risk adjustment methodologies that are clinically detailed (i.e. have face validity that includes detailed attention to severity of illness) should be researched in conjunction with available data elements. Additional data elements should also be considered simultaneously.
- SES adjustment should be exogenous.
- It’s important to note that the exogenous risk adjustment does not need to be present in perpetuity. For example, over time the IME exogenous factor has become less important as the DRG classification system has improved. Such an approach mitigates the concern of some that our suggested approach will lead to lower quality care for the more vulnerable SES individuals.
At 3M, we are currently working with several state Medicaid programs, local health authorities, and other researchers to implement the above approach to incorporating SES into risk-adjustment. We welcome conversations with other researchers and leaders of health care organizations that are interested in pursuing a “robust trial” of risk adjustment that incorporates sociodemographic factors.
Norbert Goldfield, MD, is medical director, Clinical and Economic Research, 3M Health Information Systems.