CT Scans Reveal Thymus Health Predicts Immunotherapy Success Across Cancers

Immune checkpoint inhibitors (ICIs) have transformed cancer treatment, yet a substantial proportion of patients derive limited benefit. Current predictive biomarkers—primarily tumor-intrinsic factors like PD-L1 expression and tumor mutation burden (TMB)—are imperfect and fail to account for the patient's host immune competence. This study addresses that gap by investigating the thymus, a central immune organ responsible for T cell maturation, as a measurable determinant of immunotherapy efficacy.

The researchers developed a self-supervised deep-learning framework trained on 5,674 independent CT scans to automatically quantify 'thymic health'—a radiographic proxy for thymic functionality—from standard-of-care chest CT images. This model was then applied to the Harvard-NSCLC cohort (n=1,218 patients with NSCLC treated with ICIs) and the Harvard-PAN cohort encompassing melanoma, renal, breast, bladder, esophageal, and other cancers, totaling 3,476 patients treated at the Dana-Farber Harvard Cancer Center. Biological validation was performed in the independent, prospectively enrolled TRACERx NSCLC cohort (n=464).

In the Harvard-NSCLC cohort, higher thymic health scores were significantly associated with improved progression-free survival and overall survival following ICI therapy. Critically, these associations remained significant across clinically stratified PD-L1 expression levels and TMB categories, suggesting thymic health provides prognostic information orthogonal to established biomarkers. In TRACERx, thymic health at diagnosis positively correlated with blood T cell receptor excision circles (TRECs)—a direct measure of thymic output—and with T cell receptor (TCR) diversity in both blood and tumor, as well as immune-signaling pathway activity. This biological validation strongly supports the interpretation of CT-derived thymic health as a functional measure of adaptive immune competence rather than a nonspecific imaging artifact.

Across the pan-cancer Harvard-PAN cohort including melanoma, breast cancer, and renal cell carcinoma, thymic health retained its significant association with survival outcomes, demonstrating tumor-agnostic relevance. The deep-learning model successfully extracted meaningful thymic signal from routine imaging in diverse cancer contexts without requiring dedicated thymus-focused scans.

These findings collectively argue that thymic health represents a previously unrecognized host-immune determinant of immunotherapy efficacy. The ability to extract this information non-invasively from already-acquired CT scans is a major practical advantage. The authors propose thymic health as a potential tool for patient stratification, treatment timing optimization, and as a rationale for developing thymic rejuvenation strategies—such as IL-7, growth hormone, or other thymopoietic interventions—to enhance ICI response in patients with poor thymic function.