Clinical Evidence and Machine Learning

Base was not using machine learning at the time described because its subscriber sample size was insufficient. Real-world uncontrolled health data requires 10,000 to 100,000 data points before machine learning becomes reliable. Base data-mined thousands of clinical trials over approximately one year to extract intervention-to-biomarker relationships. Biomarker cases are reviewed every two weeks with functional medicine doctors, and users receive their top four recommended actions at a time. Health companies claiming AI-based personalization should be treated cautiously unless they can demonstrate adequate data volume.