888-P: Therapy Recommendation Neural Network for Comorbid Cardiometabolic Diseases

Objectives: Type 2 diabetes, hypertension, and hypercholesterolemia often develop in tandem with risk factors that appear to be more than additive. This work studies early therapeutic intervention beyond the silos that are created when looking at each morbidity separately. Methods: After reproducing national guidelines for three morbidities using a neural network, we rely on transfer learning to record real world evidence outcomes from 469,496 primary care patients to optimise therapeutic outcomes from a multi-morbidity perspective in the following 2 years. Single morbidity treatment recommendations are adequate for the majority, but are optimally complemented with a comorbid therapy for a minority. Shapley values explain differences and are used with digital twin cohorts to reject the null hypothesis for no clinical benefit over guidelines with 95% confidence. Results: Applied retrospectively on a test set with 10,676 cardiometabolic therapy decisions, 807 comorbidity recommendations were identified, 575 with p