Machine Learning to Identify Biomarkers in AD

Dermatology Times, Dermatology Times, April 2022 (Vol. 43. No. 4), Volume 43, Issue 4
Pages: 33

A recent study aimed to identify whether using specific serum biomarkers could predict therapeutic response in patients with atopic dermatitis.

A recently published study that analyzed data from a previous clinical trial using a statistical machine learning approach was unable to identify specific serum biomarkers, including cytokines and chemokines, predictive of immunosuppressive treatment outcomes in patients with atopic dermatitis (AD). Findings from the study were published online in January 2022 by Reiko Tanaka, PhD, who a part of the engineering faculty in the Department of Bioengineering at Imperial College London, United Kingdom, and colleagues, in Skin Health and Disease.

Biomarkers may help predict therapeutic responses and thereby stratify patients and improve treatment selection—outcomes particularly helpful for decision-making in AD, according to Mio Nakamura, MD, a clinical assistant professor of dermatology at the University of Michigan Health, in Ann Arbor. 

“When initiating a new treatment for AD, it can take weeks to a few months to see benefits of a medication,and most treatments for AD need [an] at least a 3- to 6-month trial before declaring it effective or ineffective,” Nakamura told Dermatology Times®. “If we can predict treatment outcomes, we can potentially save the patients a lot of wasted time.”

Nakamura, who wasn’t involved in the study, added that a significant challenge in AD treatment is insurance coverage for certain medications, but the identification of predictive biomarkers could prove useful in overcoming this hurdle.

“We typically need to go through insurance-mandated step therapies,” she said. “We could potentially convince insurance companies to do away with step therapy requirements if we can prove that a specific medication will likely work best for the patient.”

In the study by Tanaka et al, the investigators produced a statistical machine learning model that used data from a published longitudinal study comprising 42 patients with AD who were treated with either azathioprine (Azasan, Alcami Corporation) or methotrexate for a 24-week period. The study provided data on 26 serum cytokines and chemokines which were measured prior to the initiation of treatment. In addition to cytokines and chemokines, the investigators also included filaggrin gene mutation, sex, and age in the model.

According to the investigators, the machine learning model “described the dynamic evolution of the latent disease severity and measurement errors to predict AD severity scores,” approximately 2 weeks ahead. 

The AD severity scores included those from the Eczema Area and Severity Index (EASI), SCORing AD (SCORAD), and Patient Oriented Eczema Measure (POEM). They conducted feature selection to identify important serum biomarkers that could predict the AD severity scores. 

The developed model outperformed reference models for time-series forecasting. When used for further testing the predictive ability of specific biomarkers,the researchers found that the predictive performance of the model was not significantly improved by including some biomarkers as covariates. According to the investigators, this findingsuggests the biomarkers measured prior to treatment initiation were unable to predict future AD severity scores.

The investigators wrote that while a lack of evidence for predictive biomarkers of the studied treatments “should not be interpreted as evidence of an absence, our results suggest that the effect of biomarkers on the prediction of severity scores, if any, is likely to be small or too subtle to be captured by our linear model,” given that “the prediction errors of future scores by our model was mostly attributed to errors in the score measurement process.” The investigators added that data from a larger cohort could better inform future research of the effect of biomarkers on severity score prediction in AD.

Study limitations included the small patient population; however, the investigators noted that the AD severity scores were assessed at 6 different time points, which suggested a methodological strength.

Nakamura noted that while biomarkers may potentially identify whether certain treatments will be successful for specific patients, there is a need to balance this information with the individual patient’s treatment goals and preferences.

“For example, even if science shows that the patient would do best on dupilumab [Dupixent,Sanofi and Regeneron Pharmaceuticals Inc), if the patient is needle-phobic, dupilumab wouldn’t be an option, and the same would be true if the patient has comorbidities that are contraindications to certain therapies,” she said. “Even if we can develop scientific ways to stratify AD patients and predict treatment outcomes, we will still need to see how to best utilize these in the real-life setting.”

“What we are increasingly finding for AD specifically is that it's a heterogeneous disease, and there are a lot of different factors that can affect the specific and individual immune profile,” she said. “Because of that, what we're seeing is that there's no one-size-fits-all approach to treatment for the condition.” 

Benjamin Ungar, MD, assistant professor of dermatology and director of the Rosacea & Seborrheic Dermatitis Clinic at Mount Sinai, New York, New York, explained in an interview with Dermatology Times® that certain predictive biomarkers could help direct selection of treatment and therefore be an important aspect of personalized care in AD. 

While the new study did not find the examined biomarkers predictive of treatment response in AD, Ungar, who wasn’t involved in the biomarker study, stated that the findings do not necessarily close the door on its use in this disease.

“One thing that we are seeing with the research—and this is work that we have done at Mount Sinai—is that biomarker associations with disease severity and treatment response are really the best when they combine both blood[and] skin biomarkers,” Ungar explained. “I think moving forward,continued efforts to identify these biomarkers from multiple different areas and multiple different approaches are going to be needed to improve that kind of capability.”

Disclosures:

Tanaka and colleagues report no relevant conflicts of interest. Nakamura and Ungar also report no relevant conflicts. The study was funded by the British Skin Foundation.

Reference:

Hurault G, Roekevisch E, Schram ME, et al. Can serum biomarkers predict the outcome of systemic immunosuppressive therapy in adult atopic dermatitis patients? Skin Health and Disease. 2022;2(1). Doi: 10.1002/ski2.77