KI-basierte Vorhersage beeinflussbarer Gesundheitsrisiken mit Leistungsdaten

As Prof of Data Science at bbw University Berlin Dr. Zahn will present 3 practical Examples of claims- based risk Prediction implemented by his private company DCC Risk Analytics. DCC was founded in 2008 to implement Morbidity Risk Adjustment (MRSA) in the BAS German GKV Health Fund and supports the new German Health Claims Research Center (FDZ) at BfArM since 2024. DCC collaborates with PWC on several consulting projects i.e. for RKI, vfa, GKV and PKV clients. The 1. example is the Geriatric Risk Coaching “smartCasaPlus” (sCP) operated since 2016 for 4 Sickness Funds. It combines continuously learning intervention specific hospitalization risk Prediction Models with evaluation-based Impact and Signin Models. sCP has proven to achieve a 5-year RoI of 3.1 by focusing multi-channel health coaching to elderly patients with impactable hospitalization risks. The 2. example results from the German-Canadian collaboration Project “AI based Risk Prediction and Treatment Effect estimation” (AIR_PTE). It has been developed with McGill University Montreal using Regression Trees to predict pharmacological impact on patients with deep Vene Thrombosis (VTE). It employs Risk Adjusted Propensity Score Matching (RAPSM) to evaluate treatment impact on medical outcomes and treatment costs. The AIR_PTE Rapid Evidence Generator (REG) has so far been reused by the German Center for Orthopedics (DZO) at Schiller University Jena to predict and prevent Spinal Surgery Revision Risk and the Berlin Institute of Health (BIH) at Charite to predict morbidity Costs with distributed deep learning on 10 years of German GKV and UK Biobank claims with European OMOP Ontologies. The 3. most modern example employs Long-Short-Term-Memory (LSTM) Deep Leaning on 8 quarters spatio-temporal diagnostic and treatment pattern from 6 sector claims to predict Nursing Risks according to §25b of the new German Health Data Use Act. It is currently piloted to directly impact treatment costs associated with escalating dementia and consequences of fall in elderly patients with non-recognized Nursing Grade. Content: Experiences to predict individual Health Risks on German Health Claims; Impact options in German Health Insurance and focusing with Impact Models; Results of German-Canadian Project AIR_PTE - AI based Risk Prediction and Treatment Effect estimation