IO Plus TKI Combination Toxicity: The Published Profile

Short answer

What is the known safety profile of IO plus TKI combinations such as pembrolizumab plus lenvatinib? High-grade toxicity is the expected case, not a signal. Across the registrational trials of pembrolizumab plus lenvatinib, grade 3 or higher events occurred in roughly 85 to 89 percent of treated patients, with treatment-related deaths near 1 percent. Hypertension and diarrhoea dominate, dose modification is routine, and the population treated moves the numbers more than the regimen does.

What the registrational trials reported

In the final pre-specified overall survival analysis of CLEAR in advanced renal cell carcinoma (Motzer et al., Journal of Clinical Oncology 2024), grade 3 or higher treatment-emergent adverse events occurred in 84.9% of patients receiving lenvatinib plus pembrolizumab. Diarrhoea was the most common treatment-emergent event overall and hypertension the most common grade 3 or higher event. Fatal treatment-emergent events occurred in 16 patients (4.5%), of which 4 (1.1%) were considered treatment-related. Median overall survival was 53.7 months and median progression-free survival 23.9 months.

In KEYNOTE-775 in advanced endometrial cancer after platinum-based chemotherapy (Makker et al., New England Journal of Medicine 2022), adverse events of grade 3 or higher occurred in 88.9% of patients receiving lenvatinib plus pembrolizumab, against 72.7% of those receiving physician's-choice chemotherapy, while median overall survival was 18.3 versus 11.4 months in the overall population.

In LEAP-002 in first-line hepatocellular carcinoma (Llovet et al., Lancet Oncology 2023), where lenvatinib was dosed by bodyweight at 8 or 12 mg, the most common treatment-related grade 3-4 events were hypertension (17% in both arms), increased aspartate aminotransferase (7% versus 4%) and diarrhoea (6% versus 4%), with treatment-related deaths in 4 patients (1%) on the combination and 3 (1%) on lenvatinib plus placebo.

Two mechanisms, with a short list of shared organs

The toxicity arithmetic follows from the biology. Immune checkpoint blockade produces immune-mediated events across endocrine, hepatic, pulmonary, gastrointestinal and dermatological systems. Multi-kinase VEGFR inhibition produces vascular and constitutional events: hypertension, proteinuria, haemorrhage, hand-foot syndrome, fatigue, diarrhoea. These act on different targets, so a patient can experience both, and most of the combined burden is the two profiles overlaying rather than amplifying.

The exception is an organ both agents affect. Thyroid dysfunction is the clearest example, arising both as an immune-related endocrinopathy and as a recognised effect of multi-kinase inhibition. Sawada and Narukawa (Cancer Control 2024), reviewing 83 clinical studies and 7,951 patients, found that combining a multi-kinase inhibitor with a checkpoint inhibitor raised the risk of diarrhoea (relative risk 1.24), hypothyroidism (1.44) and rash (1.71) relative to multi-kinase inhibitor monotherapy.

Those relative risks are the useful shape of the class effect: a modest increase concentrated in the categories where the two mechanisms meet, rather than a uniform doubling.

Population, not regimen, explains much of the spread

The same drug pair produced an 84.9% grade 3 or higher rate in a first-line renal cell carcinoma population with good performance status, and 88.9% in a platinum-pretreated endometrial cancer population. The regimen was not more toxic in the second trial; the patients had less reserve.

This is why a higher high-grade rate in a later or broader study is usually not a new safety signal. Before escalating one, compare the populations on age, performance status, prior cytotoxic exposure and organ substrate - and compare the actual starting doses, which differed between these programmes.

Dose is a trajectory, and real-world practice knows it

Combination regimens containing a multi-kinase inhibitor titrate downward, so predicting constitutional toxicity from the protocol starting dose overstates it. Community practice goes further than protocols do. Barbi and colleagues (Frontiers in Oncology 2025), in a retrospective cohort of 92 patients with advanced endometrial cancer, found that 62% started lenvatinib at 10 mg or less and only 14.1% received the labelled 20 mg starting dose; grade 2 or higher adverse events occurred in 74%, half had treatment interruptions, and 36% discontinued.

That study also reports the cost of the adaptation: in an age-adjusted analysis the reduced-dose group had a significantly higher hazard of progression or death (hazard ratio 2.92, 95% CI 1.32-6.44). It is a single retrospective cohort and should be read as a caution rather than a dosing rule, but it illustrates that de-escalation is a trade, not a free safety improvement.

For anyone comparing an observed toxicity rate against a published one, the operative comparison is exposure over the window in which the event occurred, not the dose written in the protocol.

What the negative trial adds to the safety picture

LEAP-002 did not meet its pre-specified significance thresholds for overall survival (median 21.2 versus 19.0 months, hazard ratio 0.84, 95% CI 0.71-1.00) or progression-free survival, and its authors concluded that the findings do not support a change in clinical practice.

Read as a safety document rather than an efficacy one, it is the most informative of the three. The toxicity was broadly as expected for the class, and the benefit was not there. Tolerability is only meaningful against demonstrated benefit in that specific population, which is the argument for evaluating a combination indication by indication rather than as a drug-class property.

How Atlas Bio does this

Atlas Bio's Combo Safety and Efficacy platform models combination toxicity as additive monotherapy risk plus a population fragility coefficient, with supra-additive treatment only for organs both agents damage, and validates one parameter set against four IO plus TKI trials - including the negative trial an earlier version predicted wrongly, which is published as a documented miss.

Common questions

Is an 85 percent grade 3 rate a safety signal?

Not on its own for this class. Both registrational trials of this combination reported grade 3 or higher events in the mid-to-high eighties, with treatment-related deaths near 1 percent. A signal is a deviation from that published expectation in a comparable population, not the absolute number.

Which toxicities are genuinely worse in combination than with the TKI alone?

In a review of 83 studies and 7,951 patients, the categories with a raised relative risk versus multi-kinase inhibitor monotherapy were diarrhoea, hypothyroidism and rash. These are the categories where the immune and vascular mechanisms converge on the same organ or tissue.

Why do real-world toxicity rates differ from trial rates?

Community populations are older and more comorbid than trial populations, starting doses are frequently lower, and discontinuation is often chosen where a trial would dose-modify. In one real-world endometrial cohort only 14.1 percent of patients began at the labelled starting dose.

Sources

Talk to Atlas Bio.

Sponsors, researchers and clinicians: tell us the decision you are trying to make and we will show what the platform can and cannot answer.