One about leveraging the risk stratification model to improve equity

California Department of Healthcare Services wants to transition its Medicaid program from traditional clinical care focus on the sick members to population health management (PHM) with a whole-person care approach to address the SDOH in its new risk stratification and segmentation (RSS) guideline. These requirements comply with NCQA PHM standards and incorporate utilization data integrated with other data sources, such as geographic and social needs data. To prevent the exacerbation of health disparities and algorithm biases. Additionally, the effectiveness of the RSS methodologies is continuously reassessed to optimize care and improve health equity.

At IEHP, efforts are being made to mitigate biases, including conducting an audit of several algorithms with independent guidance and oversight by a vendor, analyzing the impact of risk stratification and quality care provision by race/ethnicity, and launching a new initiative to close data gaps and improve organizational governance of algorithm development and use.

IEHP's RSS approach to optimize quality and equity assesses outcome disparity and care access and program provision disparity based on identification, enrollment, and outcome across all ethnic groups of members for any care management programs. Social and environmental driver data, such as the Healthy Places Index, Area Deprivation Index, and Social Vulnerability Index, are added to go beyond clinical and behavioral data and capture social needs and community risks. The approach also involves test and learn experiments to make adjustments that can reduce disparity and improve equity with data validation, as well as exploring and planning new data sources to further help reduce data gaps and bias.

Two areas of potential bias are being focused on: representation (data) bias and label choice bias. Representation bias occurs when algorithms fail to hit the right target for underserved groups due to being trained or evaluated in non-diverse populations. Label choice bias arises from a mismatch between the ideal target the algorithm should be predicting and a biased proxy variable the algorithm is predicting. During this audit, bias was found in predicted risk, such as patients of similar risk scores with different disease burdens. Statistical test such as false negative ratio was used to detect bias.

“IEHP's RSS approach to optimize quality and equity assesses outcome disparity and care access and program provision disparity based on identification, enrollment, and outcome across all ethnic groups of members for any care management programs”

The modified RSS model has demonstrated a reduction of gap-in-care and equity optimizations, with significant improvements in the top two tiers of needy members, covering more members with a gap in care based on HEDIS measures and including more high-cost members who account for 80% of IEHP's total claims cost. Improved health equity has been achieved through fairer and more even access and eligibility for minority groups to enhanced or extensive care management.

In conclusion, while RSS is needed for PHM, data and algorithm bias is common. Rigorous RSS requirements for PHM aim to support proper care by identifying the right person and their needs, in the right setting and at the right time, including preventive care for low-risk members. Health plans and care management teams need to be aware of and review their methodology and explore and test alternative models to optimize care and improve health equity.