A new analysis of the phase 3 ORIENT-11 trial suggests that the future of lung-cancer immunotherapy may depend not only on identifying patients most likely to respond, but also on predicting who could suffer dangerous immune complications. Researchers from Sun Yat-Sen University Cancer Center report that a gene-expression signature associated with T helper 17, or Th17, cell differentiation can help estimate the risk of severe immune-related adverse events in people with advanced non-small cell lung cancer receiving the anti-PD-1 antibody sintilimab together with chemotherapy. When this toxicity signal was combined with established measures of treatment benefit—PD-L1 expression and tumor-infiltrating lymphocytes—the investigators created a four-quadrant system that separated patients according to both expected efficacy and potential harm. The results, published in Cancer Immunology, Immunotherapy, point toward a more individualized approach to immune checkpoint therapy, a treatment strategy that has transformed oncology but remains difficult to tailor before therapy begins.
Immune checkpoint inhibitors work by releasing molecular brakes that restrain T cells. In cancer, proteins such as PD-1 on immune cells and PD-L1 on tumor or surrounding cells can suppress an antitumor response, allowing malignant cells to evade immune surveillance. Blocking PD-1 with an antibody such as sintilimab can restore T-cell activity and produce durable tumor control. The same immune activation, however, can sometimes turn against healthy organs. These complications, known as immune-related adverse events, may affect the skin, colon, liver, lungs, endocrine glands, heart, nervous system, or other tissues. Although many are manageable, severe events can require hospitalization, high-dose corticosteroids or other immunosuppressive treatments, interruption of cancer therapy, and, in rare cases, can be fatal. Clinicians therefore face a central dilemma: the immune profile that supports tumor rejection may also increase the risk of uncontrolled inflammation.
The ORIENT-11 analysis focused on 266 patients with advanced NSCLC who received sintilimab plus chemotherapy. Within this group, 19 patients, or 7.14 percent, developed severe immune-related adverse events. Because these events were relatively uncommon, the researchers used baseline tumor RNA-sequencing data to search for biological pathways that distinguished affected patients from those who did not experience severe toxicity. RNA sequencing measures the abundance of thousands of RNA molecules in a tissue sample, offering a molecular snapshot of the genes and cellular programs active inside the tumor microenvironment. The team applied gene set enrichment analysis and other pathway-based methods rather than examining isolated genes alone. This approach is designed to identify coordinated biological processes, which can be more robust than relying on a single molecular marker whose level may vary between laboratories or tumor samples.
The strongest signal was the Th17 cell differentiation pathway. Th17 cells are a subset of CD4-positive T cells that produce inflammatory cytokines, including interleukin-17, and help coordinate immune responses at mucosal barriers. They are important in protection against certain pathogens, but excessive or misdirected Th17 activity has also been linked to autoimmune and inflammatory diseases. The enrichment observed in the tumors of patients who later developed severe irAEs does not prove that Th17 cells directly cause treatment toxicity. It does, however, suggest that a pre-existing inflammatory immune state may identify patients whose immune systems are more likely to become pathologically activated after checkpoint blockade. The finding also offers a plausible biological bridge between local immune activity in the tumor and systemic toxicities that emerge in organs far from the original cancer.
To convert this biological observation into a clinically usable tool, the investigators developed a predictive model from genes within the enriched pathway. They used repeated least absolute shrinkage and selection operator, or LASSO, regression 50 times. LASSO is a statistical technique that reduces the influence of redundant variables and selects a smaller group of features, an important safeguard when molecular datasets contain many genes but relatively few clinical events. The repeated analyses were intended to identify a stable gene combination rather than a signature dependent on one random division of the data. The resulting model achieved an area under the receiver operating characteristic curve of 0.904 in the training cohort and 0.769 in the validation cohort. An AUC of 0.5 represents chance discrimination, whereas a value of 1.0 indicates perfect separation, placing the model’s performance between strong and moderate depending on the dataset used.
The researchers then asked whether toxicity prediction could be integrated with markers of antitumor benefit. PD-L1 expression is already used in many NSCLC treatment decisions because it can reflect the likelihood of response to PD-1 or PD-L1 blockade, although its predictive accuracy is imperfect. Tumor-infiltrating lymphocytes provide a complementary view of the immune contexture: rather than measuring a tumor’s ability to display an immune target, TIL assessment considers whether immune cells are already present within or around the cancer. By combining PD-L1 and TIL-defined efficacy categories with the Th17-based severe-irAE score, the team assigned patients to four risk–benefit groups. This framework was designed to distinguish patients with a favorable likelihood of response and low toxicity risk from those who might have a less attractive therapeutic balance.
The differences between the resulting groups were substantial. Reported objective response rates ranged from 92.9 percent in the most favorable category to 51.1 percent in another group. Severe irAE incidence ranged from zero to 46.7 percent, indicating that some molecularly defined subsets appeared to carry a markedly higher risk of serious immune complications. The groups also differed in progression-free survival, the interval before cancer progression or death, although the abstract does not provide the exact survival estimates. These findings are important because efficacy and toxicity were not treated as opposite ends of a single scale. Instead, the model suggested that the biological factors associated with tumor response and those associated with severe immune injury may be at least partly independent. A patient could therefore have a strong predicted response but also a high toxicity risk, or a lower predicted benefit without an obviously elevated risk of severe irAEs.
The concept could eventually reshape how oncologists discuss immunotherapy with patients. A person in a high-benefit, low-risk group might be an especially strong candidate for treatment, while someone in a high-risk, lower-benefit group could require a more cautious evaluation of alternatives, intensified monitoring, or a different therapeutic strategy. The model might also help researchers design clinical trials that prospectively test toxicity-prevention measures, including closer surveillance for early organ inflammation. However, the findings are not yet a validated diagnostic test. This was a post hoc analysis of a single randomized trial, and the model was developed from patients treated with one specific immunotherapy-plus-chemotherapy regimen. Severe irAEs were observed in only 19 patients, a small number for training a multigene predictor, and performance declined from the training cohort to the validation cohort. External validation in independent populations is essential before the score can guide routine care.
Several additional challenges must also be addressed before a Th17-based model could move from research into hospitals. Tumor RNA sequencing requires adequate tissue, standardized laboratory procedures, computational analysis, and a clinically defined threshold for a positive or high-risk result. Tumors are heterogeneous, meaning that a small biopsy may not represent the entire cancer or its changing immune environment. Th17-related activity could also vary with previous treatments, infections, medications, microbiome composition, and the organ-specific mechanisms of individual irAEs. Moreover, the analysis combined severe immune toxicities as a broad outcome, even though pneumonitis, colitis, hepatitis, myocarditis, and endocrine events may arise through different biological pathways. Future studies will need to determine whether the signature predicts severe irAEs generally or is particularly informative for specific organs and syndromes.
Despite these limitations, the ORIENT-11 study illustrates a wider shift in cancer medicine: predictive biomarkers are beginning to incorporate treatment risk as well as treatment benefit. For years, immunotherapy selection has focused primarily on whether a tumor appears vulnerable to immune attack. The new analysis argues that the patient’s capacity for harmful immune activation deserves equal attention. Its Th17-associated signal is not a final answer, but it provides a mechanistic hypothesis and a measurable framework for testing it. If confirmed in larger, prospective and ethnically diverse cohorts, the approach could help replace one-size-fits-all checkpoint blockade with a more precise risk–benefit strategy—one that seeks not merely to activate the immune system, but to activate it where it is most likely to help and least likely to cause lasting harm.
Subject of Research: Severe immune-related adverse-event prediction and efficacy–toxicity stratification in advanced non-small cell lung cancer immunotherapy
Article Title: Integration of a severe immune-related adverse events predictive model with efficacy biomarkers enables precise stratification in NSCLC immunotherapy: insights from the phase 3 ORIENT-11 study
Article References: Peng Y, Wen L, Shen J, et al. Cancer Immunology, Immunotherapy. 2026. Springer Nature.
Image Credits: AI Generated
DOI: 10.1007/s00262-026-04511-y
Keywords: Severe immune-related adverse events; non-small cell lung cancer; Th17 differentiation pathway; sintilimab; immune checkpoint inhibitors; PD-L1; tumor-infiltrating lymphocytes; predictive model; risk–benefit stratification
Tags: anti-PD-1 sintilimabimmune checkpoint inhibitor stratificationimmune response biomarkersimmune toxicity predictionimmune-related adverse event risk assessmentlung cancer immunotherapyNSCLC immune-related adverse eventsPD-L1 expression biomarkerspersonalized cancer immunotherapyphase 3 ORIENT-11 trialT helper 17 cell gene-expression signaturetumor-infiltrating lymphocytes
