CN93 - Quality of life scale-based machine learning approach to predict immunotherapy response in patients with advanced non-small cell lung cancer

Background

Despite the significant survival benefits offered by immune checkpoint inhibitors (ICIs) in advanced non-small cell lung cancer (NSCLC), disease progression is inevitable. A diverse patient-reported QOL scale attempts to predict outcomes for ICI-treated NSCLC patients using machine learning.

Methods

This study analyzed three multicenter trials of atezolizumab (anti-PD-L1) in NSCLC patients with completed baseline EORTC QLQ-C30 V3 and QLQ-LC13 scales. The phase III IMpower150 cohort served as the discovery set, while the phase II BIRCH and phase III OAK cohorts were used for validation. MOVICS was used to compute cluster analysis and cluster predictor indices (CPI). Predictive ability was assessed with time-dependent AUC, validated with external datasets (OAK, BIRCH) using the PAM algorithm.

Results

In the IMpower150 Discovery Set, using ten consensus ensemble clustering algorithms with QOLS data, two subtypes emerged: Cluster 1 (CS1) and Cluster 2 (CS2). CS2 patients had shorter median overall survival (13.14 vs. 21.42 months, hazard ratio [HR] 2.07 [1.64-2.62]; P < 0.0001) and progression-free survival (5.7 vs. 8.3 months, HR 1.70 [1.42-2.04]; P < 0.0001) compared to CS1. Clinical benefit rates for CS2 were 57% compared to 68% for CS1 (P=0.0027). The PAM algorithms were validated with a similar variety of outcomes in external cohorts. CS2 consistently predicted unfavorable OS (OAK, P < 0.0001; BIRCH, P < 0.0001) and PFS (OAK, P = 0.032; BIRCH, P < 0.0001) compared to CS1.

Conclusions

Our study demonstrated the promise of integrative machine learning to effectively analyze QOLS. This approach could be used to predict clinical outcomes in advanced NSCLC patients undergoing atezolizumab immunotherapy.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

This research was funded by the National Natural Science Foundation of China, Grant No. 82060475; Chunhui program of the Chinese Ministry of Education, Grant No. HZKY20220231; the Natural Science Foundation of Guizhou Province, Grant No. ZK2022-YB632; Youth Talent Project of Guizhou Provincial Department of Education, Grant No. QJJ2022-224; China Lung Cancer Immunotherapy Research Project, Excellent Young Talent Cultivation Project of Zunyi City, Zunshi Kehe HZ (2023) 142; Future Science and Technology Elite Talent Cultivation Project of Zunyi Medical University, ZYSE-2023-02; Collaborative Innovation Center of Chinese Ministry of Education, Grant No. 2020-39.

Disclosure

All authors have declared no conflicts of interest.