Biden Administration’s First Attempt to Use Social Risk to Adjust Health Care Spending Leaves Marginalized Communities Behind
This post was jointly written by Juliette Price, Chief Solutions Officer, and Ken Robin, Chief Data Analyst
The ACO REACH offering was billed as the first value-based offering from CMMI to explicitly support health equity–but our analysis shows that the tool used to measure social risk would shift resources away from dense, urban cities with high Black and Latino populations to largely rural areas in the South. As an example, we dive into how the Bronx in New York ends up being labeled as “low need” when the methodology is applied.
ACO REACH is the Biden Administration’s first major value-based offering from the Centers for Medicare and Medicaid Services (CMS). Debuting in Spring 2022, it fully replaced the Direct Contracting Model for Medicare that was introduced under the Trump Administration. The new model places equity at its center, while retaining most of the risk-taking mechanisms of Direct Contracting. We’ve written before about this model’s introduction and the aspects that stayed the same and changed from Direct Contracting.
Central to the model and CMS’ messaging around this offering has been health equity–it’s in the name (REACH stands for Realizing Equity, Access, and Community Health), in every press release about the program, and in several portions of the program’s infrastructure: health equity played a role in the application and scoring methodology; new benefit enhancement options were added to help reduce barriers to care access; participating ACOs will have to collect demographic and social data that have not been required before by CMS to help monitor and evaluate the model’s impact on underserved populations; and all participating ACO will need to submit and implement a Health Equity Plan to measurably reduce health disparities.
But arguably the largest programmatic support for equity in the model is the introduction of social risk adjustment to the benchmarking methodology, impacting the total budget that ACOs have to work within to find savings. The idea was to acknowledge that beneficiaries in more vulnerable communities would need more support and the social risk adjustment would help create room in the target budget for ACOs to provide that support.
Social Risk Adjustment to Benchmarking Methodology
In this model, CMS introduced a Health Equity Benchmark Adjustment that adjusts beneficiary-level benchmarks (target budget) based on a composite measure of neighborhood and individual need. We previously wrote about the introduction of this Benchmark Adjustment and how it was crafted in an attempt to infuse equity into the REACH model.
This approach of modifying a benchmark or target budget by using social–not clinical–data is part of a growing movement in health care, as evidence mounts that social risk factors such as poverty, structural racism, social isolation, and limited community resources impact health in ways that the traditional health care delivery system can no longer ignore, but is also unprepared to address.
The 2014 Improving Medicare Post-Acute Care Transformation Act required the HHS Secretary to conduct research into how Medicare’s value-based payment programs (such as REACH) could address this challenge. Two sets of studies were conducted and delivered to Congress in 2016 and 2020, and in the latter, specific recommendations suggested “CMS should continue to support providers and plans addressing social risk factors through models, supplemental benefits, and VBP payment adjustments.”
How the Health Equity Benchmark Adjustment Works
The ACO REACH Health Equity Benchmark Adjustment is mathematically straightforward - a composite score for each participating beneficiary derived from just two components: the Area Deprivation Index (ADI) (a percentile score from 1-100), and Dual Medicaid Status (Medicare only vs. full or partial dual eligibility). Because Dual Status is a yes/no condition, it functions simply as an upward adjustment to the composite score, with 25 points added if a beneficiary is partially or fully eligible.
The two measures that feed the adjustment are not equally weighted - the ADI is a continuous variable based on a percentile and Dual Medicaid Status is binary, so CMS developed an approach to combine them. A composite score will be calculated starting with the ADI value of a beneficiary’s Census Block Group of residence, then adding 25 points if the beneficiary is dually eligible. For example, an individual living in a highly deprived neighborhood (90th percentile ADI), who is also dually eligible (+25), would receive a composite score of 115 (90+25). This approach acknowledges the multi-level and cumulative nature of vulnerability - that the impacts of poverty are generally more acute for individuals living in deprived neighborhoods compared to those living in relatively affluent neighborhoods where resources such as quality schools, child care, grocery stores and safe streets promote stability.
Based on this composite score, CMS will then take all aligned beneficiaries and organize them into deciles (10 equal size groups) ranked from low to high with the top decile theoretically representing the highest need population. Benchmark adjustments will ultimately be made based on the decile into which each beneficiary is placed.
Both upward and downward adjustments will be made. An upward adjustment of $30 per beneficiary per month (PBPM) will be applied to beneficiaries in the top decile, and a smaller downward adjustment of $6 PBPM will be made for beneficiaries in the bottom five deciles (representing the lower risk 50% of the population). No adjustment will be applied to individuals in deciles 6-9.
The (Technical) Trouble with the Area Deprivation Index
While the concept of using a standardized measure to calculate relational vulnerability of beneficiaries is in theory a good idea, we now turn to the tool that CMMI has chosen to use in the ACO REACH model. Since Dual Status is an additive condition in the formula, the ADI is thus the foundational measure upon which the ACO REACH conception of equity is built, so it is critical that it effectively capture deprivation across a diverse array of neighborhoods and populations.
The ADI was developed at the University of Wisconsin-Madison and is built entirely using American Community Survey (ACS) 5-year estimates. The index includes 17 socioeconomic indicators from the ACS, grouped into six domains.
A strength of the ADI is that it is calculated at the Census block group level, which is a more granular geography compared to Census tracts or counties. A block group is a subdivision of a tract, defined to contain between 600 and 3,000 residents. For more than 220,000 block groups in the nation, the ADI includes two measures of education (percent with less than 9 years of education; percent with at least a high school diploma), two measures of employment (percent employed in a white-collar occupation; unemployment rate), four measures of income (median family income; income disparity; percent below poverty level; percent below 150 percent of poverty level), four measures of housing (median home value; median gross rent; median monthly mortgage; home ownership rate), one measure of household composition (percent of single parent households), and four measures of household resources (percent without a car; percent without internet; percent without complete plumbing; percent of housing units with more than one person per room).
An initial assessment of ADI national percentiles suggests that not all types of community need are equally captured, with rural and southern areas generally portrayed as more acutely deprived compared to deeply impoverished urban neighborhoods.
One example of this shortcoming is Bronx County in New York State. Profound deprivation is well established in many areas of the Bronx, particularly in the South Bronx which includes some of the nation’s poorest neighborhoods. ADI national percentiles contrast with this portrayal of the county. In fact, based on 2019 ACS data only 24 out of 1,090 ranked Census Block Groups in the Bronx score at or above the 90th percentile, which is the initial cut-off for the upwards adjustment in ACO REACH. Weighted by population, the mean block group ADI for Bronx County is 28, suggesting that on average, the Bronx is a less deprived region than 72% of the country. In a practical sense, 50,000 Bronx residents live in one of the Block Groups above the 90th percentile, implying that only 3.5% of the population would qualify as “high need.”
Multiple existing community assessments would suggest a profile of extreme need in the Bronx. For example, the Centers for Disease Control and Prevention’s Social Vulnerability Index also uses ACS data to score tracts and counties from 0 (least vulnerable) to 1 (most vulnerable). The 2018 overall SVI score for the Bronx is .9927, indicating nearly the highest possible level of vulnerability. The SVI actually includes many of the same indicators as the ADI, but of course they are weighted differently.
A review of 2019 ACS 5-year estimates further confirms resident vulnerability in the Bronx. Out of 62 counties in New York State, the Bronx ranks last (worst) in percent of population below 150% of poverty, at 40%. Montgomery County is a distant second at 31%. The Bronx is also last by a considerable margin in unemployment rate, at 10%. In the Education domain, the Bronx is again last in the state with just 72.8% of the adult population holding a high school diploma. Some housing indicators are exceptions, with the Bronx ranked 8th in median home value among New York Counties.
The methodology used to calculate base scores for the ADI yields a county profile that places the Bronx as the 10th least deprived county in New York. This comports closely with housing value data but contradicts county status in education, income, and employment. Given a weighted average ADI percentile of 28, even after the Dual Eligibility Status adjustment is factored in, application of the Health Equity Benchmark Adjustment to the Bronx will likely not financially reward providers who serve these county residents. In fact, since the ACO REACH approach to promoting equity includes a downward adjustment of $6 PBPM for beneficiaries in the five lower risk deciles, being in this REACH model and serving many Bronx residents may actually be financially disadvantageous for providers.
The Implications of a Flawed Social Risk Mechanism
The implications of the current ACO REACH equity adjustment structure could be both far-reaching and acute.
While the use of a social risk adjustment tool is not currently widely adopted in health care, programs like ACO REACH serve as testing grounds which could lead other policy makers to adopt the practice. A flawed tool used in this early adoption could easily be carbon-copied to other programs, leading to replication of a flawed method of social risk adjustment. Additionally, while the dollar amount of upward and downward adjustment to the benchmark (+$30/-$6) are not fiscally significant to the program/participating ACOs, there is no reason why, in replicative models, these dollar amounts wouldn’t expand dramatically, potentially causing more harm.
The ADI does not include metrics such as percent immigrant population, English language proficiency, or percent urban population. While exclusion of these Census variables may have been determined based on legitimate factor analysis, the result may be that urban Hispanic populations are left on the outside looking in when ACO REACH equity adjustments are calculated.
If the ADI is weighted such that certain types of deprivation are not captured, then a metric expressly deployed to promote equity may ironically and misleadingly suggest lack of need in ethnic enclaves within some of the nation’s most deprived neighborhoods. Most acutely, Performance Year 1 of the ACO REACH program begins in January 2023, meaning time is running out for CMS to issue a change in the methodology they’ve laid out to-date.
While the authors of this piece acknowledge that all CMMI models are “test” models to further illuminate the path towards high-quality, lower-cost, patient-centered care, we believe there are ways in which the Health Equity Benchmark Adjustment can be modeled before its implementation to ensure no harm is done to disadvantaged communities that have been purposefully marginalized in the past. Even if this means suspending the adjustment for PY1 (2023) until more review, modeling, and conversation can occur, we support getting health equity right over rushing to implementation.
Lastly, we cannot write on this topic without acknowledging the deep harm that has been done to many marginalized communities in the past by federal policies; technical aspects of federal programs have, for many centuries in this country, led to the systemic underinvestment in many communities across the country, and while we do not at all suppose that the Biden Administration has done this intentionally, a closer look at the evidence before us would suggest the implementation of the Health Equity Benchmark Adjustment in its current form runs afoul of the Administration’s goals on Health Equity.
What are your thoughts on the Health Equity Benchmark Adjustment? We’d love to hear from you! Drop us a line or share your thoughts with us on social media.
About the Author: Juliette Price is the Chief Solutions Officer at HSG. Follow her on Twitter and connect with her on LinkedIn. Dr. Ken Robin is the Chief Data Analyst at HSG.
