Predictive models help identify best biologic treatments for psoriasis

Gradient-boosted decision trees and logistic regression were effective in identifying the optimal biologic therapy for psoriasis patients, according to a study.

“Different patient types, which can be identified through statistical and machine learning methods applied to registry data, respond to specific biologic drugs in different ways,” Mia-Louise Nielsen, MSc, of the department of dermatology at Copenhagen University Hospital - Bispebjerg and the study’s lead author, told Healio. “Information about previous treatment with biologic drugs, patient age, BMI, baseline PASI/DLQI scores and status of concomitant [psoriatic arthritis] are important when predicting the outcome of a biologic therapy.”

Researchers conducted a population-based cohort study of all patients treated for moderate to severe psoriasis with biologics using data from Danish nationwide registries.

Seven patient types were identified using variables including sex, age at presentation, age at diagnosis, previous biologic treatments and baseline PASI/DLQI scores. Success rates for each drug were estimated and predictive prognostic models were used to identify the optimal treatment for each cluster.

A total of 2,034 patients with 3,452 treatment series were included.

A specific cytokine target was identified in in 63.6% of those in which gradient-boosted decision trees were used, and in 59.2% of those in which logistic regression was used.

Rates of success depended on both the specific drug used and the patient characteristics, according to the study.

In predicting the most successful drugs, gradient-boosted decision trees predicted the top choice in 48.5%, the top two choices in 77.6% and the top three accuracies in 89.9%. Logistic regression models had accuracies of 44.4%, 75.9% and 89% for the top one, two and three drugs, respectively.

“We showed with our analyses that to some extent it is possible to predict which biologic drug the individual patient will benefit the most from,” Nielsen said. “By predicting the optimal biologic drug for treatment of moderate to severe psoriasis, the number of failed treatment attempts can be minimized which could lead not only to better patient care, but also reduced health care cost.”