Infrastructure for Case Management Risk Stratification

Risk stratification requires documented high-risk patient identification protocols—comorbidity thresholds, prior utilization criteria, social determinant flags—that define when a patient triggers intensive case management. The baseline confirms these protocols are established. Without findable, current documentation of what constitutes high-risk (e.g., ≥2 admissions in 90 days OR HbA1c >9 with social barriers), the ML model's intervention intensity recommendations lack a defensible clinical basis and cannot be audited.

The ML model requires systematic capture of comorbidities, prior ED visits and admissions, medication adherence indicators, functional status assessments, and social determinants. Template-driven workflows ensure these fields are consistently populated at admission. High-risk screening results and care coordination activities are logged per baseline context. Without structured capture of these inputs at every patient encounter, the model trains on incomplete datasets and stratifies unreliably.

Risk stratification ML requires formal schema mapping patient features to risk predictors: comorbidity codes, utilization frequency fields, medication adherence scores, frailty indices, and social determinant categories. The model needs entities (Patient, Comorbidity, SocialBarrier), relationships (Patient.has.Comorbidity, Patient.lacks.SupportSystem), and validation constraints to compute composite risk scores. Consistent field definitions across all records enable feature engineering that drives model accuracy.

Risk stratification must access EHR clinical data (comorbidities, vitals, labs), HRIS-equivalent for care team assignment, payer data for utilization history, and pharmacy data for medication adherence—in near-real-time at admission. The baseline confirms EHR integration and UM software access. For autonomous high-risk flagging at admission, the model must query multiple data sources via API without manual extraction steps. A unified access layer ensures the model always operates on complete patient context.

Risk stratification models degrade as patient population characteristics shift and clinical criteria evolve. InterQual and Milliman update high-risk thresholds periodically; the model must recalibrate when these change. Readmission risk model recalibration is confirmed in baseline context. Event-triggered updates—when new payer contracts change utilization criteria or when model performance metrics drop below threshold—keep stratification clinically valid without requiring quarterly manual review cycles.

Case management risk stratification requires integration between EHR (clinical data), UM software (case manager worklists), pharmacy systems (medication adherence), and payer data (prior utilization). The baseline confirms risk scores flow to care manager worklists. API-based connections to these systems enable the model to assemble complete patient risk profiles and route high-risk flags to the right case manager without manual handoffs. Full cross-organizational integration is beyond current capability given HIE limitations.