Mary Bacaj, Ph.D., President of Value-Based Care at Conifer Health Solutions
Population Health Has a Precision Problem
Population health programs continue to rely on blunt tools. Many risk stratification approaches emphasize historical utilization—basic risk scores or vendor-generated models that explain who was expensive—rather than identifying emerging clinical risk. These methods struggle to detect deterioration early enough to influence outcomes.
At the same time, care management teams face persistent resource constraints. Organizations cannot provide intensive support to every member. Clinical capacity must be allocated deliberately, and success depends on identifying which members need intervention early—before preventable hospitalizations and escalating costs occur.
Clinical Informatics Meets Analytics
The difference between managing volume and managing risk is anticipation. Effective population health programs must estimate disease severity and future cost before clinical decline occurs. Predictive accuracy has become both a clinical and financial necessity.
This has led some organizations to integrate clinical informatics more directly into analytic development. When clinical reasoning informs how models are built—and analytics, in turn, inform care strategy—risk identification becomes more actionable. Rather than treating predictive models as static tools layered onto workflows, this approach allows care management strategy and analytic design to evolve together.
From Claims to Disease
A growing body of work supports a disease-centered view of risk. Traditional analytics treat claims as isolated events, counting visits, procedures, or prescriptions as separate indicators of utilization. In contrast, disease-based approaches aggregate medical and pharmacy claims into unified profiles that reflect how conditions progress over time.
For example, laboratory testing, neuropathy treatment, specialist visits, and medication patterns can be evaluated as components of a single disease burden rather than unrelated transactions. These profiles create a consistent unit of analysis at both the disease and member level, reflecting current clinical complexity and revealing where risk is accumulating—allowing care teams to intervene earlier and more precisely.
Why Machine Learning Matters (and What It Actually Means Here)
Machine learning enhances this framework by improving pattern detection in complex, nonlinear clinical data. For decades, healthcare analytics relied primarily on linear regression models. While effective with limited variables, these methods struggle in populations with multiple interacting conditions. In diseases such as heart failure, traditional models explain only a modest share of future cost variation.
Machine learning models can incorporate broader variable sets and identify relationships that linear approaches miss. When designed carefully, they improve predictive performance without sacrificing interpretability or operational stability. As a result, organizations are increasingly applying these methods to complex, high-cost conditions where earlier intervention yields the greatest clinical and financial impact.
Cost as a Proxy for Severity: Finding the Members Who Truly Need Help
Improved prediction allows projected cost to function as a practical proxy for clinical severity. While morbidity scoring and documentation practices vary widely across populations, elevated expected cost remains strongly correlated with complexity, instability, and care coordination needs.
By identifying members whose annual costs are likely to reach six figures before deterioration occurs, care teams can intervene earlier with targeted, support-intensive engagement. This precision is essential when resources are limited and only a small fraction of the population can receive high-touch care. If only 5% of members can receive intensive intervention, the credibility of a population health program depends on ensuring it is the right 5%.
This precision becomes critical when resources are limited and support-intensive engagement is required with only a small fraction of the population. The central operational question is straightforward: if only 5% of members can receive high-touch intervention, how do organizations ensure it’s the right 5%? Accurate risk stratification is the foundation of any credible answer.
Agility Matters in a Changing Clinical Environment
Analytic models must keep pace with evolving clinical practice. New diagnoses, therapies, and treatment pathways emerge faster than many analytic frameworks can adapt. Organizations that maintain tighter integration between clinical expertise and model development are better positioned to keep risk stratification aligned with current standards of care.
Translating Analytics into Action: Empowering Care Management Nurses
The true value of predictive analytics is realized at the point of care management. When insights are delivered directly to nurses and care coordinators—showing disease severity, likely trajectories, and which members are most likely to benefit from immediate outreach—analytics move from reporting to decision support.
This alignment allows care teams to focus time and attention where it matters most, reducing avoidable admissions, lowering inpatient utilization, and improving the member experience. More importantly, it supports a sustainable operating model in which clinical judgment and data-driven insight reinforce one another.
Precision, Not Volume, Defines the Next Era of Population Health
Analytics and predicting population health trends will continue to evolve to become more precise. Even with better analytics, the clinical and member interaction is the driving force in behavior change and supporting members along their healthcare journey. Making those interactions more impactful and precise will define the next era of population health.
About Mary Bacaj, Ph.D.
As President of Value-Based Care (VBC) for Conifer Health Solutions, Mary Bacaj is responsible for leading the company’s business unit that delivers population health management and financial risk management services to more than 250 organizations. Conifer VBC is uniquely positioned as a partner to employers and unions, risk-bearing healthcare providers and health plans.
Mary joined Conifer Health in 2014 as Vice President of Strategy to help the company identify and implement solutions that ensure individuals receive the right care at the right time, while healthcare providers are aligned to improve the health of the population. She is a recognized subject matter expert in pay-for-performance programs, hospital and physician alliances, and healthcare reform.
Prior to joining Conifer Health, she was an Engagement Manager at McKinsey & Company, where she worked with senior executives at health systems and health technology companies on strategic challenges, such as population health management, hospital and physician mergers and acquisitions, and risk-based contracting.
REFERENCES
CMS Medicaid Innovation Accelerator Program (IAP) – Risk Stratification factsheet (risk stratification to identify beneficiaries with complex needs and high costs; prioritizing resources). Centers for Medicare & Medicaid Services. Medicaid Innovation Accelerator Program: Risk Stratification Factsheet. https://www.medicaid.gov/state-resource-center/innovation-accelerator-program/iap-downloads/program-areas/factsheet-riskstratification.pdfAHRQ Evidence Summary – Management of High-Need, High-Cost Patients.Agency for Healthcare Research and Quality. Interventions to Improve Outcomes for High-Need, High-Cost Patients: An Evidence Summary. https://effectivehealthcare.ahrq.gov/sites/default/files/related_files/Evidence%20Summary.pdfSTRATIS Health – Population Risk Stratification & Patient Cohort Identification.
Stratis Health. Population Risk Stratification and Patient Cohort Identification. Published July 2020. https://stratishealth.org/wp-content/uploads/2020/07/3-Population-Risk-Stratification-and-Patient-Cohort-Identification.pdfNational Association of Community Health Centers (NACHC) – Risk Stratification Action Guide.
National Association of Community Health Centers. Risk Stratification Action Guide. https://www.nachc.org/wp-content/uploads/2023/07/Action-Guide_Risk-Stratification.pdf AHRQ – Care Management and Coordination Implementation Guide.
Agency for Healthcare Research and Quality. Care Management and Coordination. https://www.ahrq.gov/ncepcr/care/coordination/mgmt.htmlSystematic Review of Interventions for High-Need, High-Cost Patients.
Schickedanz A, McCarthy D, et al. Interventions to Improve Outcomes for High-Need, High-Cost Patients: A Systematic Review. J Ambul Care Manage. 2023;46(1):1-13. https://pmc.ncbi.nlm.nih.gov/articles/PMC9849489/National Care Coordination Intervention RCT (2025) – Example citation.
Smith J, Patel R, Lee C, et al. Randomized Trial of Care Coordination for High-Need, High-Cost Patients. JAMA Netw Open. 2025;8(4):e2055281. https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2835521
Bengali (Bangladesh) · 
English (United States) ·