Kaiser Permanente Advanced Alert Monitor prevents 500+ deaths annually with ML-powered deterioration prediction

Hospitalized patients face risk of rapid deterioration without sufficient early warning. Kaiser Permanente needed a scalable way to identify inpatients at high risk for adverse events—such as imminent ICU transfer or unexpected death—before deterioration became critical.

Kaiser Permanente developed the Advanced Alert Monitor (AAM), a machine learning algorithm trained on hundreds of millions of data points from hospitalized patients. The model uses granular EHR data—lab values, vital signs, and other clinical indicators—to predict deterioration risk within the next 12 hours. The algorithm was paired with a clinical workflow that accounts for individual patient goals of care.

The AAM prevented more than 500 deaths per year across the health system while reducing high-risk readmissions by 10%, as published in The Joint Commission Journal on Quality and Patient Safety.

• Pairing a predictive algorithm with a well-designed clinical workflow—not just the model—was essential to achieving impact.
• Respecting patient goals of care in the alert response protocol was critical for ethical and effective deployment.
• Large-scale EHR data (hundreds of millions of data points) enabled high-accuracy predictions at a 12-hour horizon.