JMIR Publications

ABSTRACT

Background:

Backgroung: Accurately predicting the survival outcomes of lung cancer patients receiving chemotherapy remains challenging.

Objective:

Objective:

Aiming to improve patient management, this study constructed a multivariate model based on machine learning methods to assess the all-cause mortality risk in these patients.

Methods:

Methods:

This study retrospectively recruited 1278 post-chemotherapy lung cancer patients from Guangzhou Chest Hospital between 2007 and 2019. Candidate features such as demographic characteristics, environmental exposures, clinical information, and patient-reported symptoms were collected by questionnaires and the electronic medical record system. The survival status and deceased date were investigated twice a year. On the training set, a total of 84 predicted models were built with five machine-learning algorithms individually or binately. On the testing set, concordance indexes (C-indexes) were calculated to select the optimal model, and the model performance was internally validated through receiver operating characteristic curve (ROC), calibration curves, and decision curve analysis (DCA). Furthermore, Shapley additive explanation (SHAP) and Restricted cubic spline (RCS) were utilized to feature attribution analysis.

Results:

Results:

The optimal model ultimately retained 21 prognosis-association features, including age, gender, BMI, smoking status, environmental smoke, MDASI-LC total score trajectories, CD56, TNM stage, histology, and pre-chemotherapy blood biomarkers (CYFRA21-1, CA12-5, CA19-9, WBC, D-dimer, CEA, HGB, CRP, NSE, PLT). On the testing set, the model acquired a C-index of 0.702 (95% CI: 0.652-0.753). The decision curves demonstrated positive clinical benefit when the risk thresholds were 0.40-0.69, 0.62-0.99, and 0.72-0.99 for 1-, 3-, and 5-year mortality predictions. The calibration curves showed that the predicted mortality probabilities fluctuated around the observed probabilities, and the Brier scores for 1-, 3-, and 5-year predictions were 0.20, 0.18, and 0.11 for 1-, 3-, and 5-year mortality predictions. The AUC of the model was 0.740, 0.777, and 0.915 for 1-, 3-, and 5-year mortality predictions. Interpretability feature attribution analysis revealed that the significant features could predict all-cause mortality risk in chemotherapy-treated lung cancer patients.

Conclusions:

Conclusions:

Our machine learning models exhibited excellent discrimination, calibration, and clinical benefit in predicting the mortality risk of chemotherapy-treated lung cancer patients, which could help clinicians in personalized prognostic management. Clinical Trial: None