The Social Determinants Of Health (with Justen Nestico)

This video provides an in-depth exploration of how advanced AI and machine learning are being utilized to revolutionize risk assessment in the self-funded health insurance market, particularly through the lens of Social Determinants of Health (SDOH). Justen Nestico, Director of Actuarial Solutions at Verikai, discusses their platform which aggregates traditional medical and pharmaceutical claims data (ICD-10, J codes) with thousands of behavioral, socioeconomic, and financial data points to create a holistic risk profile for individuals and groups over the next 12 to 24 months. This methodology is crucial for stop-loss carriers and MGUs seeking to underwrite groups, especially those transitioning from fully insured plans with limited claims history, providing the necessary insight to price risk accurately and confidently.

Verikai’s unique value proposition lies in its ability to leverage purchasing information (derived from credit card transactions via data aggregators) and neighborhood characteristics (e.g., pollution levels, proximity to quality hospitals/ERs) to predict future claims activity. The data is tokenized and de-identified to maintain HIPAA compliance, ensuring individual privacy while still allowing the machine learning models to tie specific behaviors and characteristics to claims risk. The platform scores risk on a continuum, with 1.0 representing the national average, allowing underwriters to quickly identify groups in the highest risk percentiles that may require additional premium loading or specific stop-loss measures (such as A-specs or lasers, though Verikai cannot directly identify individuals for lasering due to HIPAA constraints).

A significant portion of the discussion focuses on the systemic implications of SDOH, confirming strong correlations between health outcomes and race, income, and education. The data reveals expected inequalities, such as higher rates of premature births and maternal mortality among minority women. This confirmation of systemic issues shifts the focus from mere risk assessment to potential intervention. Nestico outlines a future vision for Verikai centered on AI-enabled intervention and Care Management, moving beyond quoting to proactively controlling risk. Examples include using predictive models to identify individuals at high risk for opioid addiction or premature birth, allowing health plans or captives with high clinical integration to intervene early (e.g., providing nutritional support or home environment modifications, as seen in the anecdote about addressing asthma triggers in a member’s home). The conversation also touches on the disruptive potential of GLP-1 drugs (like Ozempic) and the need for the industry to figure out how to ensure the resulting weight loss is sustainable to realize a positive ROI on the high cost of the medication.

Key Takeaways: • Holistic Risk Assessment via SDOH: Effective risk prediction in self-funding requires integrating traditional medical/pharmaceutical claims data with external behavioral and socioeconomic data, which provides a more complete picture of an individual's health trajectory than claims data alone. • Behavioral Data Sources: Key behavioral insights are sourced from purchasing information (credit card transactions) and neighborhood characteristics (e.g., pollution, distance to quality healthcare), which act as strong proxies for lifestyle and financial status. • AI Explainability is Crucial: To move beyond "black box" models, AI platforms must provide explainability by extrapolating feature importance, allowing actuaries and underwriters to understand which variables (medical vs. behavioral) are driving the risk score for a specific group. • The Five Correlates of Good Health Outcomes: Five factors strongly correlate with better health outcomes and lower claims risk: 1) Low-intensity exercise (walking, yoga, gardening); 2) High degree of religious activity (suggesting social connection and stress reduction); 3) Mentally engaging hobbies (puzzles, reading, collecting); 4) Driving a luxury car (proxy for wealth and socioeconomic status); and 5) Comfort with technology (independent of age, correlated with education and wealth). • Addressing Systemic Inequalities: AI models confirm strong correlations between socioeconomic factors (race, income, education) and poor health outcomes (e.g., premature births, maternal mortality), highlighting areas where targeted interventions can improve health equity and reduce high-cost claims. • Future of AI is Intervention: The highest value for AI in healthcare is shifting from static risk scoring (underwriting) to dynamic, AI-enabled intervention and Care Management, allowing for proactive cost control by keeping high-risk individuals out of the hospital. • Opioid Addiction Risk Prediction: AI can combine prescription data with behavioral and socioeconomic risk factors to identify individuals at high risk of opioid addiction, enabling early intervention before a crisis occurs. • GLP-1 Drug Strategy: While GLP-1s show promise in reducing comorbidities related to obesity (heart disease, diabetes), their high cost necessitates a strategy focused on ensuring the weight loss is sustainable through behavioral and lifestyle modification tools to achieve a positive long-term ROI. • Financial Incentives Drive Behavior: The push from CMS to align financial incentives in Medicare (e.g., through Accountable Care Organizations and rewarding SDOH focus) is expected to migrate to the commercial space, potentially leading to more efficient spending and better outcomes in self-funded plans. • Challenges in Data Compliance: Maintaining HIPAA compliance requires strict data de-identification (using tokens) and limits the ability of the AI platform to directly disclose individual names to carriers for specific actions like lasering.

Key Concepts:

• Social Determinants of Health (SDOH): All non-genetic factors that influence health, including personal behaviors, socioeconomic status, neighborhood characteristics (e.g., pollution, proximity to hospitals), and social connections.
• AI Risk Prediction/Scoring: Using machine learning to aggregate diverse data sets to forecast an individual or group's expected claims cost over a defined period (e.g., 12 or 24 months).
• AI-Enabled Intervention: The proactive use of predictive AI models to identify high-risk individuals and trigger specific care management or social support actions before severe, high-cost health events occur.

Examples/Case Studies:

• Asthma Intervention: An employer’s primary care model investigated a member’s frequent ER visits for asthma, identifying poor air quality at home. The intervention involved replacing the carpet and improving filtration systems, addressing the root cause and stopping the high-cost ER utilization.
• Premature Birth Risk: The data highlights premature births as a high-cost condition with a huge socioeconomic component, presenting a prime opportunity for AI-driven intervention (e.g., targeted maternity management and nutritional support) to improve outcomes and reduce prolonged NICU stays.