The application of machine learning to longitudinal gene-expression profiles has demonstrated potential to decrease the assessment gap, between biochemical determination and clinical manifestation, of a patient's response to treatment. Although psoriasis is a proven testing ground for treatment-response prediction using transcriptomic data from clinically accessible skin biopsies, these biopsies are expensive, invasive, and challenging to obtain from certain body areas. Response prediction from blood biochemical measurements could be a cheaper, less invasive predictive platform. Longitudinal profiles for 92 inflammatory and 65 cardiovascular disease proteins were measured from the blood of psoriasis patients at baseline, and 4-weeks, following tofacitinib (janus kinase-signal transducer and activator of transcription-inhibitor) or etanercept (tumor necrosis factor-inhibitor) treatment, and predictive models were developed by applying machine-learning techniques such as bagging and ensembles. This data driven approach developed predictive models able to accurately predict the 12-week clinical endpoint for psoriasis following tofacitinib (area under the receiver operating characteristic curve [auROC] = 78%), or etanercept (auROC = 71%) treatment in a validation dataset, revealing a robust predictive protein signature including well-established psoriasis markers such as IL-17A and IL-17C, highlighting potential for biologically meaningful and clinically useful response predictions using blood protein data. Although most blood classifiers were outperformed by simple models trained using Psoriasis Area Severity Index scores, performance might be enhanced in future studies by measuring a wider variety of proteins.
Copyright © 2019 The Authors. Published by Elsevier Inc. All rights reserved.