Unfairness toward rural beneficiaries in Medicare's hierarchical conditions categories score

Ravi B. Parikh, Kristin A. Linn, Junning Liang, Sae-Hwan Park, Torrey Shirk, Deborah S. Cousins, Caleb Hearn, Matthew L. Maciejewski, Amol S. NavatheHealth 2025 · DOI

TL;DR

This study reveals that Medicare's risk adjustment system, based on Hierarchical Condition Categories (HCCs), systematically underestimates the clinical risk of rural beneficiaries. This leads to lower payments for providers serving these populations, potentially exacerbating healthcare access disparities.

What Problem Does This Paper Address?

Healthcare payment systems often use risk adjustment to ensure that providers are reimbursed appropriately for the complexity of patients they treat. The Centers for Medicare and Medicaid Services (CMS) primarily uses the Hierarchical Condition Category (HCC) score, which predicts a patient's future healthcare spending based on their diagnoses. The goal is to prevent "adverse selection," where plans might avoid sicker patients. However, this study questions whether the HCC system accurately reflects the true clinical risk of all patient populations. Specifically, it investigates whether rural beneficiaries, who often face greater barriers to healthcare access and utilization, are disadvantaged by a system that relies on spending as a proxy for risk. The core problem is that if spending patterns are influenced by factors other than pure clinical need (like access to care), a spending-based risk adjustment model might systematically misrepresent the health status of certain groups, leading to unfair financial consequences.

Methods

The researchers used a large dataset of traditional Medicare fee-for-service beneficiaries from 2018-2019. They employed the CMS-HCC v24 scoring system to predict spending. The key methodological innovation was comparing the HCC-predicted spending risk with a more objective measure of clinical risk: mortality. They analyzed the calibration of the HCC spending model by comparing predicted versus observed spending across different deciles of predicted spending. Crucially, they then compared observed mortality rates between rural and urban beneficiaries within these same deciles of predicted spending. This allowed them to assess if the HCC model, which is designed to predict spending, also accurately reflects mortality risk, and if this relationship differs between rural and urban populations.

Datasets

The study utilized a nationally representative sample of 4,170,277 traditional (fee-for-service) Medicare beneficiaries from 2018-2019. This data was derived from Medicare Part A and B medical claims, outpatient, inpatient, and skilled nursing facility claims. Hospice and durable medical equipment claims were not available. The study defined rural and urban beneficiaries based on definitions from the Health Resources and Services Administration, specifically using 2013 Office of Management and Budget (OMB) delineations and Rural-Urban Continuum Codes (RUCCs).

Key Insights and Ideas

  • Spending is a flawed proxy for risk in rural areas: The study strongly suggests that relying on healthcare spending to predict clinical risk systematically underestimates the true health needs of rural beneficiaries.
  • HCCs underpredict mortality for rural populations: While the HCC model might predict spending reasonably well for urban beneficiaries, it fails to capture the full mortality risk for rural individuals.
  • Financial implications are significant: The miscalibration of HCC scores for rural beneficiaries translates directly into lower payments for providers and Medicare Advantage plans serving these communities, potentially worsening existing healthcare access disparities.
  • Diagnosis coding limitations: Rural beneficiaries may have fewer documented diagnoses due to lower access to care, further contributing to underestimation of their clinical complexity within the HCC framework.

Key Results

  • Spending Calibration: The HCC spending model was well-calibrated for urban beneficiaries but systematically overpredicted spending for rural beneficiaries in lower spending deciles and underpredicted in higher deciles.
  • Mortality Calibration: One-year mortality rates were higher for rural beneficiaries (4.25%) than urban beneficiaries (3.80%). The HCC model underpredicted mortality for rural beneficiaries (Observed/Expected ratio of 1.17) while being close to calibrated for urban beneficiaries (O/E ratio of 0.96).
  • Excess Mortality: Rural beneficiaries, particularly dual-eligible ones, experienced significantly higher excess mortality (observed mortality exceeding predicted mortality) compared to their urban counterparts across various predicted spending levels. For dual-eligible rural beneficiaries in the top 50% of predicted spending, excess mortality was 2.34%, compared to -0.54% for urban dual-eligible beneficiaries.

Limitations and Future Work

The authors acknowledge several limitations:

  • Post-processing adjustments: They could not observe any post-processing adjustments to HCC scores, which might affect their downstream policy applications.
  • Generalizability: The findings may not generalize beyond the 2018-2019 period and traditional Medicare beneficiaries, potentially differing in Medicare Advantage or other populations.
  • Subgroup precision: While the sample size for rural dual-eligibles was substantial, precision for subgroup analyses might be limited.

Future work suggested includes:

  • Incorporating geographic or SDOH covariates: Adding geographic information or social determinants of health (SDOH) data to the HCC model could improve accuracy. However, they note that previous attempts with area-level or individual survey-based SDOH metrics have not always been successful.
  • Interaction terms: Developing models with interaction terms between rurality and clinical risk factors could better capture rural-specific nuances.
  • Gold-standard diagnosis validation: Collecting data from well-represented surveys or health-related questionnaires could help impute missing diagnoses and improve calibration, especially for populations with limited access to care. This would require significant investment but could lead to more equitable risk adjustment.

Summary updated

Abstract

Risk adjustment is used in healthcare payment to mitigate the payer incentive to select for healthier populations and to improve fairness of quality assessment. The Centers for Medicare and Medicaid Services (CMS) has used a spending-based metric, the CMS Hierarchical Condition Category (HCC) score, to determine risk. However, the HCC score is potentially confounded by access and utilization differences, which are related to income and rurality. In this study, we investigate how related HCC scores are to mortality, a more objective indicator of clinical risk state, and whether that relationship differs between rural and urban populations. We examined calibration of the HCC spending model by calculating the predicted-to-observed spending ratio within deciles of the HCC score. We then compared urban and rural beneficiaries' clinical risk by comparing observed mortality rates within deciles. Our results demonstrate that the HCC model underpredicts mortality, while overpredicting spending, for rural beneficiaries. In contrast, it is well-calibrated for urban beneficiaries. These findings suggest that risk models based on HCCs may systematically disadvantage rural beneficiaries because HCC-based risk-adjusted spending may not fully account for baseline clinical risk.
BibTeX
@inproceedings{parikh2025unfairness,
  author = {Ravi B. Parikh and Kristin A. Linn and Junning Liang and Sae-Hwan Park and Torrey Shirk and Deborah S. Cousins and Caleb Hearn and Matthew L. Maciejewski and Amol S. Navathe},
  title = {{Unfairness toward rural beneficiaries in Medicare's hierarchical conditions categories score}},
  booktitle = {Health},
  year = {2025},
  doi = {10.1093/haschl/qxaf167},
}
Figure