#3815 A DEEP LEARNING APPROACH TO PERSONALISED ANTI-HYPERTENSIVE MEDICATION TITRATION

Abstract Background and Aims Hypertension is the number one risk factor for premature death worldwide. Artificial Intelligence (AI) Clinical Decision Support Systems are an important next step hypertension management but require rigorous evaluation before assimilation into routine clinical practice. Our aim to develop evaluate AI decision support tool trained on randomised trial data. Method The Systolic Blood Pressure Intervention Trial (SPRINT) showed that intensive BP control SBP <120 mm Hg results in significant cardiovascular benefit high-risk patients with compared <140 Hg. We a feed-forward neural network using Keras Tensorflow R data from 9361 persons SPRINT randomized predict probability of increase, reduce or no-change total anti-hypertensive medications at each visit. designed model investigator deviations protocol. Six baseline patient variables (age, sex, race, aspirin use, eGFR, group assignment (intensive standard)), two visit (Systolic Diastolic Pressure), previous Pressure) were inputted after normalisation centering. Hyperparameters tuned grid search method. conducted internal validation 20% set. was performed unseen test set typical scenarios (n = 50). An Shiny app developed enter new information display sensitivity analysis probabilities calculated changing variable by plus minus 10 (Figure 1). To avoid poor performance out distribution cases we tested five methods Out Distribution (OOD) detector, 1. Autoencoders, 2. Normalizing flow, 3. Local Outlier Factor (LOF), 4. Probabilistic Principal Component Analysis (PPCA) Kernel Density Estimation (KDE). Results accuracy 76.7% dataset 1000 Epochs. revealed suboptimal (e.g. recommending reduction medication when >160 zero). This OOD samples prompted creation detector use series System Hypertension. AUROCs 0.77 AE, 0.96 Flow, 0.82 LOF, 0.88 PPCA Estimation. Conclusion feasible generalisability safety