Data Scientist

The Why Behind Wellvana:

The healthcare system isn’t designed for health. We’re designed to change that. We’re Wellvana, and we help doctors deliver life-changing healthcare.

Through our elevated value-based care programs, we’re revitalizing an antiquated system that’s far too long relied on misaligned incentives that reward quantity of care not the quality of it.

Our enlightened approach—covering everything from care coordination to clinical documentation education to marketing— ties the healthy outcomes of patients directly to shared savings for primary care providers, health systems and payors.

Providers in our curated network keep their independence, reduce their administrative headaches, and spend more time with patients. Patients, in turn, get an elevated experience with coordinated care between appointments that is nothing short of life-changing.

Named 2024 "Best in Business" and 2023 "Best Place to Work" by Nashville Business Journal, we’re one of the fastest-growing healthcare companies in America because what we do works. This is the way medicine is meant to be.

Clarity on the Role:

Wellvana's Medical Economics team measures the causal impact of value based clinical care programs designed to optimize healthcare utilization in the Medicare population. These programs are evaluated based on health, utilization and cost outcomes for the one million+ Medicare beneficiaries aligned to our Accountable Care Organizations. This role focuses on that work: building evaluation cohorts, running matching and estimation pipelines, and helping maintain a shared analytical framework used across multiple program domains. Done right, this work naturally leads to consulting with clinical and operations teams on the optimization of Wellvana’s care management program portfolio, including identification and optimization of new programs.

What's Expected:

• ETL & cohort construction – Build treatment/comparison cohorts from claims and enrollment data (SQL templates + pandas), correctly handling enrollment windows, mortality exclusions, and null-sentinel edge cases in categorical covariates.
• Matching & estimation – Build and maintain causal inference model pipelines (coarsened exact matching (CEM), propensity scoring and outcome models, etc.), estimate ATT, IRR, etc, and validate balance diagnostics before results move downstream.
• Diagnostics & QA – Check positivity assumptions, flag sign-flip anomalies in savings estimates, and escalate governance-level judgment calls rather than resolving them unilaterally.
• Reporting & visualization – Generate result tables, diagnostic plots, and HTML exports summarizing program impact for internal stakeholders.
• Cross-domain consistency – Help maintain parity across evaluation domains (e.g., SNF admission vs. readmission modules), surfacing undocumented asymmetries rather than silently propagating them.
• Productionize Models – Prepare models for production, move to production and own periodic runs in order to monitor results.
• Consult – Engage with business partners on business case, methodology decisions and results, as well as Provider Partner and Patient targeting for program expansion.
• AI Assisted Development - The team uses AI coding tools (e.g., Claude Code, CoCo) as a standard part of the development workflow – for accelerating ETL/pipeline scaffolding, refactoring existing notebooks, drafting diagnostic checks, and reviewing code for the kind of latent issues this codebase has surfaced before (silent null-handling, dead branches, undocumented asymmetries between domains). This is a tool-use expectation, not a replacement for judgment:
• You're expected to use AI tools to move faster on execution — writing boilerplate, drafting first-pass matching or ETL code, generating test cases — while still owning correctness of the output.
• You are not expected to accept AI-generated code without understanding and verification of architecture and results, especially around causal methodology (matching specs, estimator choice, null-handling logic).
• Comfort working conversationally with an AI coding assistant to review, explain, or refactor an existing module is treated as a baseline skill for this role, similar to Git fluency.

What's Required:

• Integrity: The right way is the only way.
• Dependability: You do what you say you’re going to do.
• Advocacy: You fight for the best possible outcome for providers and their patients.
• Clarity: You make it all understandable.

Education:

• Bachelor's degree in statistics, biostatistics, economics, data science, epidemiology, or related quantitative field
• Coursework or applied exposure to causal inference concepts — matching methods, potential outcomes framework, confounding — even if not yet production experience.
• Master's degree in a quantitative field, or equivalent applied experience preferred

Years of Related Experience:

• 5+ years of experience in medical economics, healthcare analytics, or population health analytics
• Git-based version control workflow experience
• Healthcare claims data experience — familiarity with MS-DRG and ICD-based classification
• Experience with Snowflake or similar cloud data warehouses

Skills/Competencies/Behaviors:

• Python (pandas) + SQL — working proficiency, including ability to write/modify SQL templates against claims and enrollment data
• Comfort reading and modifying existing analytical codebases (Jupyter notebooks, modular Python scripts) rather than only building greenfield analyses
• Experience using AI coding assistants (e.g., Claude Code, Copilot, Cursor) as part of a real development workflow, with the judgment to verify rather than blindly accept output — particularly on statistical/methodological logic
• Direct exposure to CEM, propensity score matching, AIPW, or TMLE (academic or applied)
• Medicare program structures (MSSP, ACO REACH) is a plus but not required
• QA/diagnostic mindset — validating balance diagnostics, spotting positivity violations and sign-flip anomalies, escalating rather than unilaterally resolving
• Reporting/visualization — result tables, diagnostic plots, HTML exports for non-technical stakeholders
• Judgment calibration — knowing what to escalate vs. fix independently (explicitly called out as a 6-month success marker)
• Cross-functional consulting — comfort engaging clinical/ops/finance partners on methodology and business case, not just producing output