087 Personalised predictions of AD severity dynamics and treatment recommendations using a Bayesian machine learning approach

People with atopic dermatitis (AD) would benefit from personalised treatment strategies. We aim to develop a computational tool that makes predictions of AD severity dynamics and generates recommendations. introduced EczemaPred, framework predict patient-dependent dynamic evolution using Bayesian state-space models. used EczemaPred the Patient-Oriented Scoring Atopic Dermatitis (PO-SCORAD) by combining for nine items PO-SCORAD (six intensity signs, extent eczema, two subjective symptoms). validated this approach longitudinal data 347 patients twice-weekly measurements over 17 weeks. further extended integrate available on another dataset 16 daily recording 12 estimated effects decision analysis generate achieved good performance predicting its weekly. Estimated responses strongly depended patient’s clinical phenotype allowed us patient-specific demonstrated use as coherent recommendations, while dealing missing values measurement errors. could be applied other scores such EASI or POEM.