Case Study: Kaiser Permanente predicts critical patient deterioration and reduces unplanned ICU transfers with H2O.ai

Kaiser Permanente - Customer Case Study

Kaiser Permanente, a large integrated healthcare system serving about 10 million members, faced a critical challenge: a small share of patients who experience sudden deterioration and unplanned ICU transfers account for disproportionate ICU admissions, deaths and longer hospital stays. To predict and prevent these events, Kaiser built an Advanced Alert Monitoring (AAM) system that analyzes EMR data, vital signs, labs, comorbidities and bed history—using open-source H2O tools from H2O.ai to develop the underlying predictive models.

Using H2O.ai’s open-source H2O platform, Kaiser engineered features, trained and rigorously validated models and configured the AAM to score risk and notify clinicians when thresholds are exceeded. The H2O.ai-enabled system produces results every six hours (configurable hourly) and can identify deterioration signals roughly 12 hours before clinical decline, allowing teams to intervene earlier—by moving patients to ICU sooner or taking other actions—to improve patient outcomes.

Kaiser Permanente

Patricia Kipnis

Principal Statistician