Data-driven personalised recommendations for eczema treatment using a Bayesian model of severity dynamics

Atopic dermatitis (AD) is a chronic inflammatory skin disease. AD has heterogeneous phenotypes, making it challenging to predict treatment effects for each patient and to generate personalised treatment recommendations. Here we aim to develop a computational model that predicts the evolution of AD severity and generates treatment recommendations for individual patients. We modelled the temporal evolution of eczema severity by applying a previously developed computational framework (EczemaPred) to the daily record of Patient-Oriented SCORing Atopic Dermatitis (PO-SCORAD) collected from 16 AD patients over 12 weeks in an observational study. We also leveraged historical data from 337 AD patients to kickstart the model training and reach more robust conclusions. We estimated the effects of topical corticosteroids and emollients on the next days PO-SCORAD, and generated personalised treatment recommendations using Bayesian decision analysis on whether treatment should be applied to improve PO-SCORAD on the next day for each of the 16 patients. We calibrated daily PO-SCORAD recorded by patients with monthly SCORAD assessed by clinical staff to improve the data quality. This study demonstrated a proof-of-concept for generating personalised treatment recommendations for AD using a Bayesian model that integrates multiple sources of information, including PO-SCORAD, SCORAD, and treatment usage.