Program

PO1-4-9

Algorithmic prediction of responder and non-responder allergic rhinitis patients recieving sublingual immunotherapy

[Speaker] Osamu Kaminuma:1,2
[Co-author] Minoru Gotoh:2,3, Akihiro Nakaya:2,4, Kazufumi Katayama:2, Nobumasa Watanabe:2, Mayumi Saeki:2, Tomoe Nishimura:2, Noriko Kitamura:2, Kimihiro Okubo:2,3, Takachika Hiroi:2
1:Center for Life Science Research, University of Yamanashi, Japan, 2:Allergy and Immunology Project, Tokyo Metropolitan Institute of Medical Science, Japan, 3:Department of Otorhinolaryngology, Nippon Medical School, Japan, 4:Department of Genome Informatics, Graduate School of Medicine, Osaka University, Japan

Background: Several sublingual immunotherapy (SLIT) drugs for the treatment of allergic rhinitis (AR) have recently been launched. Since long-term therapy is required, despite its high efficacy, the existence of ~30% patients refractory to this therapy is a major problem of SLIT.
Methods: SLIT with cedar pollen (CP) extract was performed to patients with CP-reactive patients with AR. After the 2-year therapy, patients whose disease severity was well improved (high-responder; HR) and was unchanged or exacerbated (non-responder; NR) was classified. Upon comprehensive measurement of serum cytokine levels, these parameters were processed to establish an ensemble-learning algorithm that divide HR and NR groups.
Results: The disease conditions of over 70% AR patients were improved by the SLIT, though remaining patients failed to show improvement even after 2 years. After the selection of top 33 HR and bottom 34 NR patients, their sera taken before the start of the therapy were employed for 50 cytokine measurement. Only IL-12p70 level was significantly different between HR and NR, though this parameter alone could not completely distinguish those groups. However, processing all cytokine data with Adaptive Boosting (AdaBoost) algorithm, a strong leaner that divide HR and NR groups with a high (~97%) accuracy was established. Clustering analysis for cytokines clearly showed that Th1 and Th2 cytokines were highly correlated especially in HR patients.
Conclusion: By employing AdaBoost algorithm, AR patients who would respond or not respond to SLIT could be predicted before the treatment. The dependence of Th1/Th2 imbalance in the pathogenesis of each patient may be a determinant of the responsiveness to SLIT. Further investigation for cellular population, mRNA expression, and genetic variation with ensemble algorithms will be useful for establishing accurate predictive diagnosis of SLIT responsiveness and for clarifying the mechanisms of SLIT.

Advanced Search