1271P - Integrating machine learning prediction and causal inference analysis to unravel determinants of progression in EGFR-mutant, advanced NSCLC

Background

Understanding the associates and causes of progression after first-line treatment in EGFR-mutant advanced NSCLC remains a critical yet elusive endeavor. In this study, we employed a comprehensive approach integrating machine learning classification and causal inference analysis to unveil the determinants of progression in this patient cohort.

Methods

Retrospective data from eleven academic centers were gathered to compile a cohort of EGFR mutant patients. Classical patient features, treatment characteristics, and the neutrophil to lymphocyte ratio (NLR) at the onset of first-line treatment were analyzed. Fifteen machine learning algorithms were employed to identify the optimal model, evaluated based on F1 score and area under the curve (AUC). Additionally, Linear Non-Gaussian Acyclic Model (LINGAM) was utilized for causal inference analysis to delineate the causal relationships driving progression.

Results

Among 333 patients initially evaluated, 205 cases were deemed suitable for machine learning analysis after accounting for missing data and follow-up duration. Linear discriminant analysis emerged as the most accurate model, achieving an AUC of 0.74 and an F1 score of 0.77. The top three influential variables identified by this model were smoking status, NLR category, and type of first-line treatment with variable importance scores of 1.2, 0.8, and 0.6, respectively. Moreover, LINGAM analysis revealed NLR category, smoking status, and gender as causal factors for progression, with causal estimate scores of -0.20, -0.17, and -0.04, respectively.

Conclusions

This study represents a pioneering effort in utilizing machine learning to explore the determinants of progression after first-line treatment in EGFR-mutant, advanced NSCLC. Our findings underscore the roles of smoking status, NLR category, and type of first-line treatment in driving progression. Furthermore, the causal inference analysis sheds light on smoking status, NLR category, and gender as causally significant factors. This understanding paves the way for elucidating the pathogenesis and developing novel therapeutic strategies for EGFR-mutant advanced NSCLC.

Clinical trial identification

Editorial acknowledgement

During the preparation of this work the authors used ChatGpt 3.5 in order to enhance readibility. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.