Abstract
Abstract
Background
Immune checkpoint inhibitors (ICIs) are a standard treatment for advanced non-small cell lung cancer (NSCLC), but clinically practical predictors of immune-related adverse events (irAEs) and treatment continuity remain limited.
Methods
We conducted a retrospective cohort study using two nationwide Japanese administrative claims databases: Medical Data Vision (MDV) and the Japan Medical Data Center (JMDC). Patients with advanced NSCLC treated with ICIs were identified. Those with available thyroid function data were included in analyses of ICI-induced hypothyroidism (MDV, n = 1786; JMDC, n = 1083), and patients receiving first-line ICIs were analyzed for treatment continuity (MDV, n = 1088; JMDC, n = 1007). Machine learning models including Elastic Net were developed and externally validated to predict hypothyroidism. Treatment durability was assessed using time to next treatment or death (TTNT-D).
Results
Elastic Net showed the most consistent performance across cohorts (AUC: 0.72 training, 0.73 validation, 0.71 test). Baseline thyroid-stimulating hormone (TSH) was the strongest predictor of hypothyroidism. Serum albumin consistently emerged as an important predictor across all machine learning models. Higher albumin levels were associated with an increased incidence of hypothyroidism and significantly prolonged TTNT-D (MDV: HR 0.75; JMDC: HR 0.79; both
p
p
Conclusions
Baseline TSH was the strongest predictor of hypothyroidism, while serum albumin was associated with both hypothyroidism and treatment continuity in patients receiving first-line ICI therapy for NSCLC, suggesting its potential as a practical biomarker for clinical risk stratification.