Our Research

Most Medicare–Medicaid duals are labelled “high risk” based on demographics, diagnoses, prescriptions, and other individual-level characteristics. Using 100% of national fee-for-service claims for 3.9 million duals, our study found that adding delivery system context—provider networks, facility characteristics, market structure, and access barriers—raises prospective spending prediction from R² 0.45 to 0.62 and improves acute care prediction models' sensitivity from 25.0% to 33.8% while keeping specificity above 97%.

Population health management programs coordinate care for over 80 million Medicaid beneficiaries but lack systematic clinical decision support for determining when to intervene and which interventions to select for patients with complex conditions. Our objective was to develop and validate a clinical decision support system integrating acuity prediction and intervention selection models for population health management programs.

Waymark developed an AI system that continuously analyzes unstructured notes from community-based care teams to flag urgent safety risks, surface preventive care opportunities, and generate learning feedback, improving the accuracy of actionable safety alerts from 68% to over 95% over 12 weeks. By synthesizing data streams across community health worker observations, pharmacy records, and hospital feeds, the system shifts care management from reactive to proactive for Medicaid patients with complex needs.

A study of nearly 191,000 Medicaid-enrolled women across 26 states found that Waymark Signal for Maternity, integrating clinical and social determinants of health data, can identify opportunities for prenatal care coordination among patients receiving Medicaid 55 days before traditional clinical indicators emerge. Incorporating social data also eliminated the model's accuracy gap between Black and white patients, pointing to a more equitable approach to early risk identification within Medicaid maternity care.

This commentary notes the superiority of targeting rising-risk patients rather than high-cost claimants for Medicaid cost containment based on analysis of 13.1 million beneficiaries across 15 states. Early identification of and intervention for rising-risk patients is a more effective way to prevent the progression of chronic conditions and manage associated costs than attempting to reduce extreme utilization, which tends to decrease naturally over time.

A study of 3,175 Medicaid patients found that Waymark's reinforcement learning approach to care management recommendations reduced acute care events by 20.7% compared to standard experience-based practices, while also narrowing disparities in outcomes across racial, ethnic, and gender groups. The findings suggest that reinforcement learning can improve both the effectiveness and equity of care management decisions for patients with complex medical and social needs.

A study of 14.2 million Medicaid recipients found that Waymark's Signal for Quality Improvement predicted gaps in nine quality measures, including preventive care and chronic disease management, significantly more accurately than non-predictive outreach methods, with the addition of social determinants of health data further improving performance and reducing racial disparities in prediction accuracy. The results suggest that incorporating both clinical and social data into risk models can help close care gaps more equitably among low-income populations.

Our new peer-reviewed research examines what health plans, providers, and state policymakers can expect as recent Medicaid policy changes in H.R. 1 take effect. The findings paint a stark picture: excess mortality and preventable hospitalizations are projected to surge, community health centers could lose up to 26% of their revenue, rural hospitals face heightened risk of closure, unemployment will rise, and local economic output will fall.

Artificial intelligence (AI) has primarily enhanced individual primary care visits, yet its potential for population health management remains untapped. Effective AI should integrate longitudinal patient data, automate proactive outreach, and mitigate disparities by addressing barriers such as transportation and language. Properly deployed, AI can significantly reduce administrative burden, facilitate early intervention, and improve equity in primary care.

We conducted a natural experiment study using difference-in-differences analysis of propensity score–matched US counties from 2011 through 2019. Loss of CHC sites was associated with an increase in age-adjusted all-cause mortality of 3.54 deaths per 100 000 population (95% CI: 1.19, 5.90; p = 0.003) in the year following the loss. The largest increase was observed for cancer mortality (2.61 per 100 000; 95% CI: 0.59, 4.62; p = 0.011). Primary care physician density and patient volume loss both mediated the relationship.

We found that the benefits of expansion generally balanced out the costs while redistributing health from higher to lower income groups. In probabilistic sensitivity analysis, we found—using a health opportunity cost threshold of $150,000—that Medicaid expansion was cost-effective in reducing CVD outcomes 53% of the time and both cost-effective (efficient) and equity enhancing 26% to 29% of the time.

In this observational study, we used a target trial emulation framework to estimate the potential of addressing transportation barriers to improve T2DM outcomes. Specifically, we sought to estimate whether not experiencing transportation barriers, compared with experiencing them, would be associated with better glycemic, blood pressure, and cholesterol outcomes among adults with T2DM in a network of community-based health centers.

In this cross-sectional study of 19,725 Medicaid enrollees across 2 states and 2 independent health plans, those experiencing specific procedural prescription denials had a higher risk of physiologically related emergency department visits and hospitalizations compared with those without a denial in the subsequent 60 days after matching and further adjustment for risk of acute care. These findings suggest that although procedural prescription denials may aim to curb immediate drug costs, some denials may prompt heightened acute care utilization and expenses that outweigh the short-term prescription budget savings.

Sadiq Patel, Waymark's VP of Data Science and AI, examines barriers to AI/ML adoption in Medicaid and lays out potential solutions to address these challenges — including actively seeking input from patients and the people who serve them: PCPs, community health workers, pharmacists, therapists, care coordinators, and community organizations.

Using microsimulations, we compared both the statistical power and false positive rates of A/B testing and Sequential Multiple Assignment Randomized Trials (SMART) for developing personalized communications across multiple effect sizes and sample sizes. SMART showed better cost-effectiveness and net benefit across all scenarios, but superior power for detecting heterogeneous treatment effects (HTEs) only in later randomization stages, when populations were more homogeneous and subtle differences drove engagement differences.

We analyzed data for 48.4 million patients receiving Medicaid across 34 states and Washington DC, and found that nearly 40% are for conditions that coule be prevented or managed through timely access to primary care. As many Medicaid programs struggle to manage rising costs, these findings demonstrate that early interventions can meaningfully improve outcomes and reduce costs for their Medicaid populations.

A large population of incarcerated people may be eligible for prerelease and transition services under the new Medicaid Reentry Section 1115 Demonstration Opportunity. We found that several disease prevalence rates were sufficiently high among incarcerated populations to likely skew overall Medicaid population prevalence of these diseases when prerelease and transition services are expanded, implying the need for planning of additional data exchange and service delivery infrastructure by state Medicaid plans.

A study of 1,652 rising-risk Medicaid patients across Washington and Virginia found that Waymark's community-based teams, guided by Waymark Signal's machine learning models and supported by AI-powered administrative tools, reduced all-cause acute care events by 22.9% compared to a matched control group. The findings demonstrate that pairing early community intervention with technology-enabled outreach and streamlined operations can meaningfully improve health outcomes for Medicaid patients before they become high utilizers of emergency and hospital care.

We examined how competing risks, baseline health care costs, and indirect costs can differentially affect cost-effectiveness analyses for racial and ethnic minority populations. We illustrate that these structural factors can reduce estimated quality-adjusted life-years and cost savings for disadvantaged groups, making interventions focused on disadvantaged populations appear less cost-effective.

We argue that persistent data quality issues and insufficient integration of clinical and social risk factor data have contributed to the growth of value veneers by limiting the ability of Medicaid programs to proactively identify and deliver targeted interventions to at-risk patients. We also outline how recent advances in data science can enable Medicaid programs to deliver higher-value care to beneficiaries.

Findings report that Medicaid expansion was associated with a reduction in cardiovascular disease (CVD) mortality overall and in particular among minority and female subpopulations.

We sought to estimate minimum threshold Medicaid payment rates to enable community health worker (CHW) program sustainability, and found that higher Medicaid fee-for-service and capitated rates than currently used may be needed to support financial viability of CHW programs. We also present a revised payment estimation approach that may help state officials, health systems and plans discussing CHW program sustainability.

A study of 10 million Medicaid patients across 26 states found that Waymark Signal identified at-risk patients at triple the rate of standard risk models, while also correcting a racial bias in which standard cost-based models were less accurate at identifying Black patients as high-risk compared to white patients. The findings point to a more accurate and equitable approach to proactively identifying Medicaid patients who would benefit from early outreach and care.

Numerous studies have shown that the failure to address individuals’ health-related social needs (HRSNs) can result in poorer health outcomes and increased health care costs. Here, we propose an alternative value-based arrangement for Medicaid managed care that addresses social needs by placing primarily non-clinical staff at the center of care to maximize impact.

Access to quality primary care is associated with improved health outcomes and lower medical costs, but social determinants of health (SDOH) can impact access to care. Patients with the greatest risk for poor health outcomes may be disconnected from primary care. Connecting these patients to primary care improves health outcomes reduces emergency department utilization and reduces total cost of care.

Here, we argue that adoption of value-based care (VBC) in Medicaid has been limited due to a lack of revenue optimization opportunities via risk adjustment, the complexity of implementing VBC models across state Medicaid programs, high member churn, and the growth of “value veneers,” or modest value-based arrangements that nominally pass as VBC but do not meaningfully alter care delivery.

In this decision analytical model, the cost of providing evidence-based interventions for social needs averaged $60 per member per month. The findings of this study suggest that a substantial increase in resources would be needed to implement a comprehensive approach to addressing social needs that falls largely outside of existing federal financing mechanisms.

Food insecurity is associated with many aspects of poor health but trials of food insecurity interventions typically focus on outcomes of interest to funders rather than quality of life outcomes that may be prioritized by individuals who experience food insecurity. Our findings show that food insecurity elimination may improve important, but understudied, aspects of health.

Social determinants of health affect the distribution of individual-level social risks to health, such as food and housing instability, and inadequate transportation. For these reasons, those in the health care system are increasingly seeking to help with social risks, often by working with agencies and community-based organizations for multi sector collaboration.

PCP density is a meaningful mediator of the relationship between urbanity and life expectancy. The mediation effect observed was higher in rural counties compared to all counties. Understanding how PCP density may be increased in rural areas may be of critical benefit to rural life expectancy.

For observational studies, we study the sensitivity of causal inference when treatment assignments may depend on unobserved confounders and develop a loss minimization approach. Our approach is scalable and allows flexible use of model classes in estimation, including nonparametric and black-box machine learning methods. Based on these bounds for the conditional average treatment effect, we propose a sensitivity analysis for the average treatment effect.

We studied the gender wage gap between male and female physicians using a large national practice survey. We observed that the gap varied by whether physicians were compensated by fee-for-service or value-based capitated payments. Additionally, we found that other future models might better align with primary care effort and outcomes.

Insulin is considered an essential medicine for people with diabetes, but its price has doubled during the past decade, posing substantial financial barriers to patients in the US. We studied out-of-pocket spending on insulin, considering possible risk factors impacting the likelihood of someone experiencing catastrophic spending (spending more than 40 percent of post-subsistence family income on insulin).