{"stable_id":"998bcea29b46d14e","slug":"epic-sepsis-model-negative","company":"Michigan Medicine / Epic Sepsis Model v1","workflow_name":"Real-time proprietary sepsis alerts","function_code":"clinical_ops","pattern_codes":["threshold_as_control","continuous_decisioning","creator_to_judge"],"changed_assumption":"A deployed alert can add workload while missing most cases.","evidence_strength":"verified","publication_tier":"showcase","freshness":"current","reviewed_at":"2026-08-23","updated_at":"2026-08-22","source_quality_summary":"1 peer reviewed; publication outcomes are verified.","caveat_summary":"One health system; v1 only, not later v2.","freshness_caveat":null,"workflow_summary":{"before":[{"actor":"Hospital clinicians","action":"Monitor admitted patients and recognize sepsis through contemporary clinical practice.","actor_type":"control"},{"actor":"Clinical team","action":"Diagnoses and treats patients using ordinary EHR information.","actor_type":"human"}],"hinge":"A deployed alert can add workload while missing most cases.","after":[{"actor":"Epic Sepsis Model v1","action":"Calculates a proprietary risk score every 15 minutes and generates alerts at the selected threshold.","actor_type":"ai"},{"actor":"Clinician","action":"Reviews or disregards the alert and retains all treatment decisions.","actor_type":"human"}],"decision_mode":"shared","decision_marker":"Human sets the threshold; AI decides each instance"},"before":[{"order":1,"actor":"Hospital clinicians","action":"Monitor admitted patients and recognize sepsis through contemporary clinical practice.","handoff_to":"Antibiotic-order workflow","control":"Timely antibiotics are the study proxy for clinical recognition; no historical manual queue is reconstructed."},{"order":2,"actor":"Clinical team","action":"Diagnoses and treats patients using ordinary EHR information.","handoff_to":"Patient care","control":"Human diagnosis and treatment authority."}],"after":[{"order":1,"actor":"Epic Sepsis Model v1","action":"Calculates a proprietary risk score every 15 minutes and generates alerts at the selected threshold.","handoff_to":"Clinician alert queue","control":"Hospital-selected threshold within the recommended range; model missed 67% and alerted on 18% of stays in the validation."},{"order":2,"actor":"Clinician","action":"Reviews or disregards the alert and retains all treatment decisions.","handoff_to":"Patient care","control":"Ordinary surveillance remains the failure path for missed cases; alert fatigue is an explicit control concern."}],"decision_rights":"Clinicians own treatment; hospitals choose thresholds.","exception_path":"Clinical judgment can override alerts; missed cases rely on ordinary surveillance.","removed_work":["No work removed; alert-review work added"],"outcomes":[{"metric":"Missed sepsis and alert burden","baseline":"Contemporary practice","result":"Missed 67% of sepsis while alerting on 18% of hospitalizations; AUROC .63","period":"2018-12-06 to 2019-10-20","scale":"38,455 hospitalizations; 2,552 sepsis cases","attribution_caveat":"One health system; v1 only, not later v2.","evidence_label":"verified"}],"executive_lesson":"Local validation is a precondition, not a refinement.","anti_pattern":"Do not generalize v1 results to v2.","questions_for_leaders":["Where is the operating threshold set and who can override it?","What measured result would trigger rollback or retraining?","Which residual decisions must remain human-owned?"],"collections":["regulated-autonomy","negative-results"],"bundle_version":"1.0.0","bundle_fingerprint":"sha256:c23c6cc2b88153d008ea8fda928f632ce0011fc2d4c5036672a16e5d895bab93","canonical_url":"https://brianletort.ai/transformations/epic-sepsis-model-negative"}