Dynamic serum tumor-marker and immune-inflammatory biomarker changes for risk stratification of immune checkpoint inhibitor-associated myasthenia gravis in non-small cell lung cancer.

Background: Immune checkpoint inhibitors (ICIs) have reshaped the treatment landscape of non-small cell lung cancer (NSCLC), but immune-related neuromuscular toxicities remain difficult to anticipate. ICI-associated myasthenia gravis is uncommon, frequently occurs early after treatment initiation, and may overlap with myositis or myocarditis. Clinically accessible blood-based indicators that help identify patients entering this high-risk period remain insufficiently defined. Objective: This study evaluated whether early dynamic changes in serum tumor markers and immune-inflammatory biomarkers could serve as noninvasive indicators of subsequent ICI-associated myasthenia gravis in patients with NSCLC. Methods: Consecutive patients with NSCLC who received their first ICI-based therapy between 1 January 2018 and 31 January 2025 were retrospectively screened at a single center. Baseline, Day-21, and Day-42 laboratory windows were prespecified before endpoint modeling. Serum tumor markers included carcinoembryonic antigen, cytokeratin 19 fragment antigen 21-1, neuron-specific enolase, squamous cell carcinoma antigen, and carbohydrate antigen 125. Immune-inflammatory biomarkers included neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, systemic immune-inflammation index, systemic inflammation response index, C-reactive protein, lactate dehydrogenase, and albumin. The primary endpoint was definite or probable ICI-associated myasthenia gravis occurring within 180 days after first ICI exposure. Landmark models were developed using Firth bias-reduced logistic regression and internally validated by bootstrap resampling. The Day-21 dynamic score used clinically pragmatic thresholds defined before endpoint modeling. Results: The final baseline cohort included 826 patients, of whom 27 developed definite or probable ICI-associated myasthenia gravis within 180 days. The Day-21 landmark cohort included 782 patients and 21 events occurring after Day 21. Compared with non-event patients, patients who later developed ICI-associated myasthenia gravis showed greater Day-21 increases in cytokeratin 19 fragment antigen 21-1, carcinoembryonic antigen, neutrophil-to-lymphocyte ratio, C-reactive protein, and lactate dehydrogenase, together with a larger decline in albumin. A four-component Day-21 dynamic score combining tumor-marker instability, neutrophil-to-lymphocyte ratio activation, C-reactive protein/lactate dehydrogenase flare, and albumin decline stratified event rates after Day 21 from 1.2% in the low-risk group to 6.7% in the high-risk group. The integrated Day-21 model achieved an apparent area under the receiver operating characteristic curve of 0.759 and an optimism-corrected area under the receiver operating characteristic curve of 0.731, indicating moderate internal discrimination. Conclusion: Early dynamic changes in routine serum tumor markers and immune-inflammatory biomarkers were associated with increased risk of ICI-associated myasthenia gravis in patients with NSCLC. The Day-21 integrated biomarker score may help identify patients requiring closer neuromuscular and cardiac toxicity surveillance during the first few weeks of ICI therapy, but external validation is required before clinical implementation.