Studies show that patients who undergo an unplanned transfer to the ICU experience worse outcomes than patients admitted directly. These patients typically stay in the hospital 8 to 12 days longer and have significantly higher mortality rates – these patients account for only 5% of patients but represent one-fifth of all hospital deaths. The challenge is to find patients before they “crash” and need to be moved to the ICU, but these patients often don’t have symptoms that clinicians can recognize as leading to a serious change in condition.
AI models can be used to find patients who are likely to crash. The machine learning models use patient medical records, laboratory results, and vital signs from patients to find early warning signals of deteriorating condition. These models can then be used with existing patients in realtime to determine their risk of a crash and as part of an early warning system for clinicians so they can intervene before the ICU transfer is needed. The AI system can also provide reason codes for a particular patient, which can help clinicians understand where they should begin their treatment.
The mission at H2O.ai is to democratize AI for all so that more people across industries can use the power of AI to solve business and social challenges. The healthcare industry is a key focus for the company with an initiative to help develop AI healthcare solutions including dedicated, experienced resources for customers, driving healthcare AI events and meetups for healthcare professionals, and membership in Health IT Now, the leading coalition of patient groups, provider organizations, employers, insurers, and other stakeholders. H2O.ai is already working with top healthcare companies including Change Healthcare, Armada Health, Kaiser Permanente, and HCA, and its products include industry leading features for machine learning interpretability required by the healthcare industry for compliance purposes.
Related Case Studies
Director, Change Healthcare
"H2O has been the driver for building models at scale. We are talking about billions of claims. You can't do this with standard off the shelf open source techniques. "Watch the Video