Machine-learning models help predict remission in Crohn's

Reuters Health Information: Machine-learning models help predict remission in Crohn's

Machine-learning models help predict remission in Crohn's

Last Updated: 2019-05-23

By Will Boggs MD

NEW YORK (Reuters Health) - Machine learning models can help identify patients with Crohn's disease likely to be durable responders to ustekinumab before committing to long-term treatment, researchers report.

"Commonly available labs and demographics at week 8 of treatment can allow us to identify those who will not respond and prevent our patients from suffering for long periods of time to see if it eventually works," said Dr. Akbar K. Waljee from Michigan Medicine and Ann Arbor Veterans Affairs Medical Center.

"It also can help prevent unnecessary cost of medications and medication monitoring with drug levels," he told Reuters Health by email.

Anti-tumor necrosis factor (TNF) drugs like ustekinumab fail to induce remission in 20% to 30% of patients and lose effectiveness within a year in another 30% to 40% of patients. Machine-learning methods have been used to try to match patients with inflammatory bowel disease (IBD) to the treatment most likely to work for them, but this approach has not yet been applied to ustekinumab.

Dr. Waljee's team used data from three randomized clinical trials of ustekinumab in patients with active Crohn's disease (UNITI-1, UNITI-2, and IM-UNITI) to create models that predict long-term remission on ustekinumab therapy using C-reactive protein (CRP) level as a biomarker of disease activity.

The researchers defined CRP levels below 5 mg/L as evidence of Crohn's disease remission and CRP levels of 5 mg/L or higher as evidence of continued Crohn's disease activity or treatment failure.

The baseline model (at week 0) included five demographic predictors and 10 laboratory predictors, and the week-8 model included the same five demographic predictors along with CRP levels at weeks 0, 3, 6 and 8 and the ratio of serum ustekinumab level to CRP level at week 0, 3 and 6 after the first ustekinumab dose.

The accuracy (mean AUROC) for predicting remission for the baseline model was 0.56 in the testing set and 0.59 in the representative testing set and for the week-8 model was 0.78 for both testing sets.

Sensitivity and specificity were 63% and 64%, respectively, for the baseline model and 79% and 67%, respectively, for the week-8 model, the researchers report in JAMA Network Open, online May 10.

The likelihood of treatment success increased sharply with lower baseline CRP measurements (with baseline CRP 14.65 mg/L or lower predicting treatment success) and with lower CRP levels at weeks 3, 6 and 8.

Measures of ustekinumab were not associated with improvements in the performance of any of the models tested, and they were associated with worse performance of a pragmatic model that used the week-6 albumin to CRP ratio (AUROC, 0.75 without and 0.71 with ustekinumab-to-CRP ratio).

"Knowing this finding might help clinicians decide to switch drug classes or to augment ustekinumab with an anti-tumor necrosis factor or a Janus kinase inhibitor if a substantial objective improvement in inflammation is not seen before the second dose rather than commit additional time and resources to costly monotherapy intensification, which appears unlikely to be effective," the researchers note. "This finding allows clinicians to limit futile care with expensive biological agents such as ustekinumab 90 mg, which has a list price of approximately $22,000 per dose."

"In clinical practice, either the week-8 random forest model or the pragmatic model using the week-6 albumin to CRP ratio could predict outcomes, although the pragmatic model appears more likely to be used at the point of care," they conclude.

The study did not have commercial funding. One of the nine authors reports financial ties to Janssen Biotech, which sells ustekinumab.

SOURCE: https://bit.ly/2wdUPTB

JAMA Netw Open 2019.

© Copyright 2013-2019 GI Health Foundation. All rights reserved.
This site is maintained as an educational resource for US healthcare providers only. Use of this website is governed by the GIHF terms of use and privacy statement.