Article Text
PDF
Abstract
Background Immunotherapy (IO) has continued to revolutionize cancer therapy particularly in difficult-to-treat cancers such as non-small cell lung cancer (NSCLC). However, durable responses among patients treated with IO monotherapy (IO-mono) are observed in less than half of patients.1 Combining IO with chemotherapy (IO-chemo) increases response rates but also raises financial burden and toxicity.2 It is thus of paramount importance to optimize treatment selection to avoid unnecessary side effects. We demonstrate here a multi-center predictive biomarker study that leverages clinico-genomic and machine learning to identify patients who potentially benefit from adding chemotherapy in reducing the risk of progression early into intervention.
Methods A total of 996 propensity-matched metastatic NSCLC patients were enrolled; patients from MD Anderson Cancer Center, Mayo Clinic, and Standup-2-Cancer Consortium were pooled into a discovery cohort for model development (n=678) and patients from Dana-Farber Cancer Institute were used as an external validation cohort (n=318). Individualized treatment score estimating the likelihood of benefit from adding chemotherapy to the baseline regimen (IO-mono) were deduced for every patient using 28 genomic and 6 clinical predictors. We investigated 5 distinct scoring systems developed by combining 3 machine learning methods (linear-based regression, distributional-based regression and decision-tree) with 3 loss functions (poisson, squared, and logistic). Repeated sampling (n=30) with 5-fold cross-validation was performed to fine-tune models hyperparameters. A composite model was proposed by integrating different scoring systems using attention-based mechanism driven by a metaheuristic algorithm.3
Results Performance were evaluated as averaged treatment effect of weighted risk reduction for 3-months progression between subgroup of patients who were treated according to vs. against model’s recommendation. We observed that the composite scoring system had superior performance (risk reduction: -16% to -19.3%, p-interaction=0.0095) in comparison to the standalone scoring system (risk reduction: +13.6% to -18.8%, p-interaction: 0.083 to 0.81) (figure 1). Patients treated against recommendation had worse 2-year progression-free survival (hazard ratio [HR, against as reference] =0.60, p=0.016 and HR = 0.58, p=0.0076 in IO-mono and IO-chemo respectively) (figure 2). Tobacco exposure, adenocarcinoma, APC and NTRK3-mutation were among features associated with great effect from ICI-mono, while Male and FBXW7-mutation derived benefit from combo therapy.
Conclusions This proof-of-concept of a modeling approach to treatment-specific biomarkers demonstrates capability in identifying both individualized and subgroup benefit from adding chemotherapy to immunotherapy, bringing the goal to precision immunotherapy closer.
References
Saad Maliazurina B, et al. Predicting benefit from immune checkpoint inhibitors in patients with non-small-cell lung cancer by CT-based ensemble deep learning: a retrospective study. The Lancet Digital Health 2023;5(7):e404-e420.
Spencer Kristen R, et al. Biomarkers for immunotherapy: current developments and challenges. American Society of Clinical Oncology Educational Book 2016;36:e493-e503.
Mirjalili Seyedali, Seyed Mohammad Mirjalili, Andrew Lewis. Grey wolf optimizer. Advances in Engineering Software 2014;69:46–61.
Interaction plots for standalone and composite models in external validation cohort (DFCI)
Comparison of 2-year Progression Free Survival stratified by actual treatment plans
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See http://creativecommons.org/licenses/by-nc/4.0/.