How do I compare healthcare analytics vendors for risk stratification?

Compare healthcare analytics vendors for risk stratification across five criteria: methodology disclosure (HCC vs. claims-based vs. clinical signal vs. combined — and disclosed, not black-box), calibration to your population, recalibration cadence, workflow routing (do high-risk flags actually go to care managers?), and SDOH integration. Disqualify any vendor that uses a "proprietary risk score" without methodology disclosure — you cannot defend a network or care management decision to a board or auditor using a model you can't explain.

Risk stratification is the foundational analytic in population health and value-based care. Every downstream workflow — care management enrollment, AWV scheduling priority, post-discharge follow-up routing, network steerage, value-based contract HCC capture — depends on knowing which patients are highest risk. Vendors differ substantially in how they compute risk. Some use HCC RAF (CMS-published methodology, fully disclosed). Some use a proprietary claims-based score. Some layer clinical signals (lab trends, medication changes, vitals). Some incorporate SDOH (PRAPARE, Z-codes, external SDOH feed). The methodology choice affects which patients are flagged and which are missed.

The risk score that gets the most marketing attention is the proprietary "AI-driven" composite score. The risk score that has the most operational value is usually HCC RAF combined with a few clinical signals — because HCC RAF is what CMS uses to set the benchmark, and benchmark is what determines settlement. A patient flagged as "high risk" by a proprietary score but with low HCC RAF doesn't drive contract economics; a patient flagged "high risk" by HCC plus rising clinical signals does. The other under-disclosed issue is recalibration cadence. A risk score calibrated on national data three years ago performs worse on your population today than a score recalibrated quarterly on your own historical outcomes. Vendors rarely disclose recalibration frequency unless asked specifically.

Use this decision framework. First, require the vendor to disclose risk methodology in writing — exact data inputs, score components, recalibration cadence. Second, require a test deployment on your historical data: how well does the score predict the outcomes (hospitalisation, ED visit, total cost) that actually mattered in your population last year? Third, require the score to expose its inputs so a clinician or care manager can see why a patient was flagged. Fourth, confirm the score routes to operational workflow (care management task list, PCP gap roster, network steerage). Vendors that fail any of these four should be disqualified regardless of brand strength.