BERLIN — The health of a cancer patient’s thymus could be linked to survival outcomes with immunotherapy, regardless of the tumor type and independent of established tumor-based biomarkers, suggests an AI-based study.

While it is known that the thymus decays with age, “this rate of decay is very different from person to person,” said study author Simon Bernatz, MD, in his explanation of why the study was done, during his October 18 presentation of the results at the ESMO Congress 2025. Consequently, its relevance in adults is unclear, he continued. 

The research revealed that high thymic health was associated with a 37% lower risk for cancer progression and 44% lower risk for death among patients with lung cancer treated with immunotherapy, while further analysis indicated a similar response across tumor types. Thymic health was measured with standard CT scans.

“This could be relevant because the patient's general adaptive immune system is currently not considered in clinical trials and in clinical practice,” said Bernatz, MD, a radiology resident and research fellow at AI in Medicine (AIM), Mass General Brigham, Boston.

Nevertheless, thymic health could “point to a novel paradigm in holistic medicine that we could potentially align to the patient,” he said.

“[W]e think thymic health is one of the missing pillars from current cancer biomarker panels and can start bringing the patient’s immune system into clinical decision-making alongside established tumor-centered biomarkers,” Bernatz added in a press release.

During the meeting, Bernatz said: As T cells are matured in the thymus, and T cells drive the response to immunotherapy, the question then becomes: “Does a healthy thymus impact immunotherapy outcomes?”

Bernatz said he believes one of the reasons why this question has yet to be answered fully is that a practical way of assessing the thymus in real-world settings was lacking. In attempting to address this, Bernatz and colleagues developed a novel AI-driven means of quantifying the thymus from CT scans.

While an obviously healthy thymus can be easily distinguished from one that has completely decayed into fatty tissue, “the changes in between can be quite subtle and can be very hard for a radiologist to see and grade accurately,” he added.

Methods and Results

To develop the AI model, the researchers used an independent dataset of more than 5000 individuals, with the algorithm locating the thymus, then extracting a range of different imaging features and combining them into one output, known as the thymic health score (range 0-100).

Patients with a score in the lowest quartile were classified as having low thymic health, those in the top quartile as having high thymic health, and everyone in between as average thymic health.

To independently evaluate the thymic health algorithm in clinical cohorts, they initially studied a non-small cell lung cancer dataset from Harvard of 1218 patients who had been treated with immunotherapy and followed-up for progression-free and overall survival.

This showed that high thymic health was, in comparison with low thymic health, associated with a significant improvement in progression-free survival (PFS), at a hazard ratio of 0.63 (P < .001) after adjusting for sex, age, ECOG performance status, histologic subtype, PD-L1 status, and the treatment line and type. Patients with average thymic health also did better than their low health counterparts, at a hazard ratio of 0.69 (P < .001).

Bernatz also reported similar findings for overall survival, with high thymic health associated with an adjusted hazard ratio for death of 0.56 vs low thymic health, while average thymic health was associated with a hazard ratio of 0.61 (P < .001 for both). 

Next, the team compared the performance of thymic health with that for PD-L1 status and tumor mutational burden (TMB), finding similar effect sizes for both PFS and overall survival. Further analysis demonstrated that thymic health was prognostic across PD-L1 and TMB subgroups, indicating that it is an independent biomarker.

To examine the underlying biology of the impact of thymic health, Bernatz said the investigators turned to the TRACERx study, using blood-based assessments to show that higher thymic health was linked to higher thymic output, higher diversity of the T-cell receptor, and a higher T-cell fraction in the blood.

Finally, the researchers leveraged data on 2258 Harvard patients with a range of tumors, including melanoma, breast cancer, kidney cancer, bladder cancer, esophageal cancer, and other cancers.

They found “very similar and consistent results across all these different tumor types.” This suggested a pan-cancer impact of thymic health on immunotherapy outcomes, Bernatz said.

Research Moves Beyond Traditional Tumor-Intrinsic Predictors to a Host-Immune Biomarker

The study has several strengths, according to Samra Turajlić, MD, PhD, director of Cancer Research UK Manchester Institute, and a consultant medical oncologist at the Christie NHS Foundation Trust, Manchester, UK.

In a discussion of the analysis, she said it moves beyond traditional tumor-intrinsic predictors to a host-immune biomarker, and “the other compelling thing about this is [it relies on] a routinely acquired piece of data in clinic” that relates to PFS and overall survival outcomes.

It also conceptually reframes the notion of immune aging as a modifiable, or at least quantifiable, aspect of immune-oncology, noted Turajlić, who was not involved in the study.

Questions and Limitations Related to New Research Remain

There remain a number of open questions, Turajlić underlined. The first being that of causality vs correlation in terms of whether thymic atrophy causes reduced immunotherapy responses or simply correlates with aging, she said.

The association between thymic health and outcomes may also be confounded by prior therapies, steroids, radiation therapy, infections, BMI, and comorbid conditions, she said. Moreover, age was included in the analysis, but perhaps immunobiological age could be captured and may have an impact.

Turajlić said she wondered whether changes in thymic health track with disease relapse and emergent immunotherapy resistance, or immune-related adverse events. Finally, there is a question mark over the basis of tumor histology-specific responses to immunotherapy, she said.

There are a number of next steps needed to be taken to validate thymic health as a biomarker, Turajlić suggested, including benchmarking against measures of immune fitness, such as lymphocyte count and conducting a prospective study controlling for tumor histology and the type and combination of immunotherapy uses.

Turajlić also wondered if imaging studies could consider immunity more broadly, perhaps by looking at changes in lymph nodes, the spleen, and bone marrow.

“The main limitation of the finding is that it has not been validated prospectively,” said Alessandra Curioni-Fontecedro, MD, professor of oncology at the University of Fribourg, Switzerland, in the press release. Curioni-Fontecedro was not associated with the research.

She did note, however, that the inclusion of a validation cohort added to the study’s quality, and said that while thymic health is not routinely assessed, chest CT scans are commonly performed in patients with cancer.

“Randomized clinical trials will be needed to establish this in clinical practice,” Bernatz stated in the press release.

The authors received financial support from the National Institutes of Health and the European Union - European Research Council. The study was partially funded by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation).

Bernatz declared relationships with Ambient, Inc, German Research Foundation, DFG. Other authors also declared numerous relationships. Curioni-Fontecedro disclosed relationships with Amgen, AstraZeneca, Boehringer Ingelheim, Bristol-Myers Squibb, Daichii Sankyo, Janssen, Medscape, Merck Sharp & Dohme, Roche/Genentech, Takeda, and Foundation Medicine.

Turajlić declared no relevant relationships.