Immunotherapy Safety Data May Not Apply to Borderline Resectable Lung Cancer

For a subset of patients with stage III non-small cell lung cancer, the decision to operate is genuinely contested. An international multidisciplinary survey of stage III TNM subsets found that surgeons and oncologists could reach no agreement on resectability for 11 of 37 anatomical configurations, and a multisocietal consensus exercise run alongside it left the same configurations unresolved. A new narrative review in Frontiers in Pharmacology takes those non-consensus subsets as its working definition of borderline resectable disease and asks a question the pivotal immunotherapy trials were never designed to answer: whether the safety profiles of current and emerging neoadjuvant therapies hold up when the patient was already marginal for surgery before the first infusion.

The evidence base for preoperative immune checkpoint blockade is now substantial. Six phase 3 trials support neoadjuvant or perioperative PD-1 or PD-L1 blockade combined with platinum doublet chemotherapy in resectable disease, reporting pathological complete response rates ranging from 17.2 percent to 40.7 percent and event-free survival hazard ratios between 0.40 and 0.68. CheckMate 816, the neoadjuvant-only trial of nivolumab with chemotherapy in 358 patients, achieved pathological complete response of 24.0 percent against 2.2 percent with chemotherapy alone, a hazard ratio of 0.63, and five-year overall survival of 65.4 percent against 55.0 percent at a median follow-up of 68.4 months. KEYNOTE-671, AEGEAN, CheckMate 77T, NEOTORCH and RATIONALE-315 extended the approach with perioperative designs, each reporting higher pathological response and longer event-free survival than chemotherapy alone.

The safety picture across those trials is more complicated than the headline efficacy figures suggest. Grade 3 or higher treatment-related adverse events ranged from 32.5 percent in the CheckMate 77T experimental arm to 73.0 percent in RATIONALE-315, and aggregate immune-mediated adverse event rates, where reported, spanned 25.3 percent to 42.1 percent. Yet the figure that matters most to a surgeon, treatment-related cancellation of surgery, stayed below 2 percent in every trial reporting it, and most patients who never reached the operating room did so for reasons other than toxicity, including disease progression, patient choice and intraoperative findings of unresectability. In AEGEAN, all-cause delay to surgery was actually more frequent in the control arm, at 22.2 percent against 17.3 percent.

The review’s central argument is that these reassuring denominators consist entirely of patients whom investigators had already judged fit for and suitable for resection at enrolment. None of the six pivotal trials was designed around the population in which the operative decision is contested, and none reported outcomes for a prospectively defined borderline subgroup. The authors draw a sharp distinction between technical resectability, an anatomical judgement about the tumour, and functional operability, which covers pulmonary reserve, cardiopulmonary fitness and frailty. Immune-related toxicity acts on the second axis rather than the first: pneumonitis, myocarditis, colitis with deconditioning and prolonged corticosteroid exposure all erode physiological reserve without changing the T or N descriptor.

Pneumonitis is the event most likely to remove surgical candidacy outright, and it is uncommon but not negligible. CheckMate 816 reported any-grade pneumonitis in 1.1 percent of the experimental arm with no grade 3 or higher events. AEGEAN reported grade 3 or higher pneumonitis in approximately 3.0 percent. Two treatment-related deaths from pneumonitis occurred in the experimental arm of CheckMate 77T, and one pneumonitis death, 0.4 percent of the tislelizumab arm, occurred in RATIONALE-315. A patient with grade 2 pneumonitis on corticosteroids is not a candidate for lobectomy until the infiltrate and the steroid have resolved, and in a borderline case that interval may be the interval in which the window for resection closes.

Perioperative mortality data are available in detail from only one of the six trials. KEYNOTE-671 recorded 30-day mortality of 1.8 percent against 0.6 percent and 90-day mortality of 4.0 percent against 1.6 percent, with any surgical-phase adverse event in 71.1 percent against 71.3 percent. That 4.0 percent figure sits inside the ASCO threshold for offering surgery in stage III disease, which requires R0 resection to be judged achievable and perioperative 90-day mortality not to exceed approximately 5 percent, but it leaves little headroom in a patient whose baseline reserve is already marginal. The review also notes that the guideline predates routine perioperative immunotherapy and was written without the corticosteroid exposure, delayed operations and interstitial lung disease events that now form part of the induction pathway.

Among emerging targets, the landscape has narrowed. Monospecific TIGIT blockade failed in three randomised settings: SKYSCRAPER-01 in PD-L1-high metastatic disease reported an overall survival hazard ratio of 0.87 with a p value of 0.22, SKYSCRAPER-02 missed in extensive-stage small cell lung cancer, and SKYSCRAPER-03 missed its primary endpoint in unresectable stage III disease, the setting anatomically closest to borderline resectable. The most advanced TIM-3 programme, cobolimab in COSTAR Lung, showed no overall survival improvement over docetaxel and was discontinued on company-reported evidence with no peer-reviewed phase 3 publication. These failures recontextualise the earlier phase 2 signals that had made both targets attractive.

The agents that still carry a neoadjuvant signal rest on thin evidence. NKG2A and CD73 combinations with durvalumab produced pathological complete response rates of 25.7 percent and 20.3 percent respectively in NeoCOAST-2 platform arms of 70 and 74 patients, with surgery performed in over 93 percent of each arm, but each rests on a single-arm cohort without a randomised comparator and without a borderline subgroup. Datopotamab deruxtecan, a TROP2-directed antibody-drug conjugate, achieved pathological complete response of 35.2 percent in a 54-patient arm with surgery in 94.4 percent, but its toxicity profile is mucosal rather than immune: oral mucositis or stomatitis in 40.7 percent, ocular-surface events in 20.4 percent, and one grade 2 drug-related interstitial lung disease event. Ivonescimab, a PD-1 and VEGF bispecific antibody, has a positive peer-reviewed phase 3 result in advanced disease, with progression-free survival of 11.1 against 5.8 months and a hazard ratio of 0.51, but no neoadjuvant data at all.

The review is equally direct about what the trials have not measured. No pivotal trial reported a functional outcome: no postoperative pulmonary function, no exercise capacity, no patient-reported physical function stratified by whether an immune-mediated event occurred. Pathological downstaging in N2 subgroup analyses is frequently read as evidence that induction chemoimmunotherapy converts borderline cases to resectable ones, but every one of those proportions is calculated among patients who already reached surgery in trials that had selected patients as resectable. Downstaging is a treatment-effect measure in operated patients; conversion is a change in a multidisciplinary decision, and no pivotal trial measured the second.

Answering the operative question, the authors argue, requires prospective cohorts in which borderline status is defined by a multidisciplinary team before treatment begins. Two such trials exist. MDT-BRIDGE will enrol approximately 140 patients with stage IIB to IIIB N2 disease, give two cycles of neoadjuvant durvalumab with chemotherapy, reassess at a multidisciplinary meeting, and use resection rate as its primary endpoint, with primary completion anticipated in 2026. NEOINSPIRE will randomise ivonescimab with chemotherapy against penpulimab with chemotherapy preoperatively. Neither has reported any outcome. Until they do, the authors conclude, the decision to give a borderline resectable patient preoperative checkpoint inhibition rests on toxicity data generated in patients who were never borderline, and the useful endpoints for the next generation of trials are resection rate after multidisciplinary reassessment, time from diagnosis to definitive treatment, and postoperative functional recovery.

Subject of Research: Safety of neoadjuvant immunotherapy strategies in borderline resectable stage III non-small cell lung cancer

Article Title: Safety profiles of current and emerging neoadjuvant immune-modulatory and combination strategies in borderline resectable stage III non-small cell lung cancer: a narrative review

Article References: Berjaoui, N., Siu, Y. T. R., Bille, A., & Patel, A. J. (2026). Safety profiles of current and emerging neoadjuvant immune-modulatory and combination strategies in borderline resectable stage III non-small cell lung cancer: a narrative review. Frontiers in Pharmacology, 17, Article 1929734. https://doi.org/10.3389/fphar.2026.1929734

Image Credits: AI Generated

DOI: 10.3389/fphar.2026.1929734

Keywords: non-small cell lung cancer, borderline resectable, stage III, neoadjuvant immunotherapy, checkpoint inhibitors, pneumonitis, surgical feasibility, TIGIT, antibody-drug conjugates, bispecific antibodies, MDT-BRIDGE, NEOINSPIRE