Machine Learning Prediction of Financial Toxicity in Patients with Resected Lung Cancer.

Abstract

Backgrounds: Financial toxicity (FT) refers to the financial stress and detrimental impact on quality of life experienced by patients due to treatment costs. In patients with resected lung cancer (LC), we sought to identify those at risk of developing moderate or severe ("major") FT using machine learning (ML) techniques based on preoperative characteristics.

Study design: Patients who underwent LC resection at a single center between January 2016 and December 2021 were surveyed to ascertain demographic information, financial data, and presence of major FT. Clinicopathologic variables were extracted from a prospective database. Patients were randomly divided into training and test sets. First, we identified the most informative features. Then, 4 ML algorithms (decision tree, random forest [RF], gradient boosting, and extreme gradient boosting) were trained. We ensembled the 4 models' predictions to optimize the model.

Results: There were 1477 patients identified, of whom 462 (31.3%) completed the survey. 46 (10.0%) patients experienced major FT. The variables most influential in our models included age, race/ethnicity, smoking status, household income, credit score, marital and employment status, size of residence, BMI, histology, extent of resection, and preoperative forced expiratory volume in 1 second. The ensemble model yielded an accuracy of 0.86, precision of 0.93, and sensitivity of 0.86, leading to an F1 score of 0.88, indicative of a reliable algorithm.

Conclusion: ML algorithms can accurately identify patients at risk of experiencing major FT following LC surgery. Preoperatively identifying LC patients vulnerable to financial stress may allow an opportunity for intervention to address downstream cost considerations.