JMIR Publications

ABSTRACT

Background:

.

Objective:

This study was designed to use machine learning methods to determine demographic and clinical characteristics of patients with advanced or metastatic NSCLC that may predict likelihood of receiving NGS-based testing (ever versus never NGS-tested) as well as likelihood of timing of testing (early versus late NGS-tested).

Methods:

De-identified patient-level data were analyzed in this study from a real-world cohort of patients with advanced or metastatic non-small cell lung cancer (NSCLC) in the U.S. Patients with non-squamous disease, who received systemic therapy for NSCLC, and had at least three months of follow up data for analysis were included in this study. Three strategies, logistic regression (LR) model(s), penalized logistic regression using lasso penalty (PLR) and eXtreme Gradient Boosting (XGboost) with classification trees as base learners, were used to identify predictors of ever versus never as well as early versus late NGS testing from an a priori defined set of variables. Data were split into D1 (training + validation) (80%) and D2 testing (20%) sets, and the three strategies were evaluated by comparing their performance on multiple m=1000 splits in the training (70%) and validation data (30%) within the D1 set. The final model was selected by evaluating performance from validation data while taking into account considerations of simplicity and clinical interpretability. Performance was re-estimated using the test data D2.

Results:

A total of 13,425 met criteria for the ever NGS-tested group and 17,982 were included in the never NGS-tested group. Performance metrics showed the Area under ROC (AUC) evaluated from validation data was similar across all models (77%-84%). Among those in the ever NGS-tested group, 84.1% (n=11,289) were early NGS-tested, and 15.9% (n=2,136) late NGS-tested. Factors associated with both ever having NGS testing as well as early NGS testing included later year of NSCLC diagnosis, no history of smoking, and evidence of PD-L1 testing (all p<0.05). Factors associated with a greater chance of never receiving NGS testing included older age, lower ECOG performance status, Black race, higher number of single-gene tests, public insurance, and treatment in a geography associated with Molecular Diagnostics Services (MoIDX) Program adoption (all p<0.05).

Conclusions:

Predictors of “ever” versus “never” as well as “early” versus “late NGS testing” in the setting of advanced or metastatic NSCLC were consistent across machine learning methods in this study demonstrating the ability of these models to identify factors that may predict those most and least likely to receive testing in accordance with clinical practice guidelines. There is a need to ensure that all patients, regardless of age, race, insurance status and geography, all factors that were associated with lower odds of receiving NGS testing in this study, are provided with equitable access to NGS-based testing.