Australian EnterpriseAI Index
Back to incident database
CriticalHealthcare1 July 2021

Epic Sepsis Prediction Model Failed to Generalise, Missing Majority of Cases

Epic Systems / University of Michigan

What happened

A University of Michigan study found that Epic's commercially deployed Sepsis Prediction Model performed significantly worse than claimed when deployed in their health system. The model flagged only 7% of sepsis patients before clinical deterioration and generated so many false positives that clinicians began ignoring alerts.

Root cause

Model trained and validated on a non-representative population without external validation across diverse hospital demographics; no ongoing performance monitoring after deployment.

Architectural failure

Absent model drift monitoring and post-deployment performance benchmarking; training data governance did not account for demographic variation; no model rollback mechanism.

Outcome

Study published in JAMA Internal Medicine sparked widespread re-evaluation of AI clinical decision tools. Raised systemic concerns about commercial AI tools in healthcare.

Architectural Failure Patterns

These pattern categories on aipatterns.com.au describe the systemic failure modes this incident exhibited.

Cite this incident

https://corporateai.com.au/incidents/epic-sepsis-model-failure-2021

Quick facts

Date
1 July 2021
Organisation
Epic Systems / University of Michigan
Sector
Healthcare
Severity
Critical
Regulatory bodies
FDA AI/ML-Based Software as a Medical Device (SaMD)21st Century Cures ActHIPAA
Tags
healthcare-aimodel-driftgeneralisationclinical-decision-supportpatient-safetyalert-fatigue

Explore failure patterns

aipatterns.com.au