Immune-Related Adverse Events as a Biomarker: What Toxicity Tells Us About Response

Immune-Related Adverse Events as a Biomarker: What Toxicity Tells Us About Response

Immune checkpoint inhibitors work by releasing brakes on T cells, and some of those T cells attack healthy tissue. The resulting immune-related adverse events (irAEs) range from a rash that needs a topical steroid to myocarditis that kills. They are also, in many cohorts, more common in patients whose tumours respond.

That observation has made immune related adverse events a biomarker question as well as a safety one. Does toxicity tell us the drug is working? Can we predict who will get it before the first dose? And what should translational, clinical and safety biomarker teams build into trials now? This article looks at the evidence on each point, including where it is weaker than it first appears.

Key takeaways

• In a meta-analysis of 125 trials, 66% of patients on PD-1 or PD-L1 inhibitors had a treatment-related adverse event of any grade and 14% had grade 3 or higher. Combination PD-1 plus CTLA-4 blockade carries far higher rates.
• Meta-analyses link irAEs to better response and survival (pooled OS hazard ratios around 0.54–0.57), but most underlying studies are retrospective and exposed to immortal time bias. The association weakens in landmark and time-dependent analyses and largely disappears for later landmarks in one meta-analysis.
• The cleanest evidence comes from a randomised trial: in KEYNOTE-054, irAEs were associated with longer recurrence-free survival on pembrolizumab but not on placebo.
• Candidate predictors of irAE risk include baseline autoantibodies, cytokine signatures, T-cell receptor diversity and early CD8 clonal expansion, HLA alleles, gut microbiome and germline variants. None is validated for routine clinical use.
• Cytokine release syndrome (CRS) is a different problem, driven by T-cell engagers and cell therapies, with its own grading system and its own biomarker literature.

What are immune-related adverse events?

irAEs are inflammatory toxicities caused by the immune activation that checkpoint blockade produces. They can affect almost any organ, though skin, gut, liver, endocrine glands and lungs are the most common sites. Many are low grade and manageable. Some, such as hypophysitis or hypothyroidism, leave permanent hormone deficiency. A small number are fatal.

Checkpoint inhibitor toxicity by regimen

Incidence depends heavily on the regimen. The figures below come from published meta-analyses and the CheckMate 067 melanoma trial.

The 2019 JAMA Oncology meta-analysis also found that PD-1 inhibitors had a borderline higher rate of grade 3 or higher events than PD-L1 inhibitors. The 2018 analysis of fatal toxic effects found that deaths tended to occur early, with a median of 14.5 days after starting combination therapy, and that colitis was the most common fatal event with anti-CTLA-4 while pneumonitis and hepatitis were the leading causes with PD-1 and PD-L1 blockade.

Do irAEs predict better outcomes?

The case for an association

The association has been reported many times. A 2020 meta-analysis in BMC Medicine pooled 30 cohort studies (4,971 patients) and found that patients with irAEs had better overall survival (HR 0.54) and progression-free survival (HR 0.52). The association held for PD-1 inhibitors but not for CTLA-4 inhibitors, and it held for low-grade irAEs but not for grade 3–4 events, where the OS hazard ratio was 0.99. A second meta-analysis of 34 studies (5,840 patients) reported an odds ratio of 4.72 for objective response and an OS hazard ratio of 0.57.

Single-centre work drove much of the early interest. In a Japanese multicentre study of 134 patients with NSCLC on nivolumab, the response rate in a 6-week landmark analysis was 52.3% in patients with irAEs and 27.9% in those without.

Immortal time bias and other pitfalls

The problem is structural. A patient has to be alive and on treatment to develop an irAE. Patients who progress quickly stop treatment early, receive fewer doses and have less time to develop toxicity. Comparing “ever had an irAE” against “never had one” therefore builds survival time into the irAE group. This is immortal time bias, and most early studies were exposed to it.

Three corrections are commonly used:

• Landmark analysis. Classify patients by irAE status at a fixed time point and analyse survival only from that point. In the 34-study meta-analysis, the OS benefit was present with landmarks of 12 weeks or less (HR 0.60) but not with later landmarks (HR around 1.0). The authors described irAEs as a possibly time-dependent prognostic factor.
• Time-varying covariates. Model irAE as a status that switches on at onset, so pre-irAE time counts in the no-irAE group.
• A control arm. The strongest test is a randomised trial in which the placebo arm can show whether “toxicity” tracks with outcome without an active drug.

That last design exists. In a secondary analysis of the EORTC 1325/KEYNOTE-054 trial of adjuvant pembrolizumab in stage III melanoma, Eggermont and colleagues (JAMA Oncology, 2020) modelled irAE as a time-varying covariate. irAEs occurred in 37.4% of the pembrolizumab arm and 9.0% of the placebo arm. On pembrolizumab, irAE onset was associated with longer recurrence-free survival (HR 0.61); in the placebo arm there was no such association (HR 1.37). The CheckMate 238 investigators did not find a similar link with adjuvant nivolumab, which they partly attributed to differences in how irAEs were defined (Weber et al., JITC 2022).

Other pitfalls are worth naming. Steroids and treatment interruptions after severe irAEs confound the relationship. irAE definitions vary between studies. Publication bias likely favours positive associations. And the signal is concentrated in PD-1 and PD-L1 monotherapy, low-grade events and early time windows.

What this means in practice

irAEs are a plausible, partly validated marker of on-target immune activation, but they are not a clinical decision tool. They arrive after treatment has started, so they cannot select patients, and severe irAEs carry real harm with no proven survival upside. For trial teams, the practical point is that any analysis linking toxicity to efficacy needs time-dependent methods specified in the statistical analysis plan, not added later.

irAE biomarkers: can we predict who will get toxicity?

A baseline or early on-treatment marker of irAE risk would be useful for monitoring intensity, dose decisions and consent. The candidates below are supported by primary studies, but almost all come from small, single-institution or retrospective cohorts.

Reading the table honestly

Three patterns stand out. First, organ-specific markers do better than general ones: anti-thyroid antibodies predict thyroid irAEs, but no single marker predicts “any irAE”. Second, several findings come from ipilimumab cohorts and may not transfer to PD-1 monotherapy or newer combinations. Third, a recurring theme in the T-cell work (Lozano et al., Subudhi et al.) and the germline IL7 finding is that the immune system’s baseline readiness may drive both toxicity and benefit. That would explain the outcome association and also suggests the two will be hard to separate with one marker.

Moving any of these into practice needs prospective, multi-site validation with pre-specified cut-offs, the same path set out in our guide on how to validate an IO biomarker in a clinical trial. Cytokine panels in particular depend on platform choice, pre-analytics and lot consistency; our multiplex immunoassay platform comparison covers the trade-offs.

Management guidelines at a glance

Three society guidelines are the usual reference points for checkpoint inhibitor toxicity:

• ASCO: Management of immune-related adverse events in patients treated with immune checkpoint inhibitor therapy, updated by Schneider and colleagues in the Journal of Clinical Oncology in 2021 (the original version was published in 2018).
• ESMO: Management of toxicities from immunotherapy: ESMO Clinical Practice Guideline, Haanen and colleagues, Annals of Oncology 2022, replacing the 2017 version.
• SITC: Clinical practice guideline on immune checkpoint inhibitor-related adverse events, Brahmer and colleagues, Journal for ImmunoTherapy of Cancer 2021.

All three grade toxicity by CTCAE and manage it by grade: continue with monitoring at grade 1, hold therapy and start corticosteroids at higher grades, and add other immunosuppressants such as infliximab when steroids fail. The ASCO update advises permanent discontinuation after most grade 4 toxicities, with the exception of endocrinopathies controlled by hormone replacement. The recommendations rest largely on expert consensus, which is itself a reason the field needs better biomarkers.

Cytokine release syndrome: a different toxicity

Cytokine release syndrome is not a classic irAE. It is an acute systemic inflammatory response driven by mass T-cell activation, typical of CAR-T cells and T-cell engagers rather than checkpoint inhibitors. It is graded with the ASTCT consensus criteria (Lee et al., 2019), alongside immune effector cell-associated neurotoxicity syndrome (ICANS), and IL-6 receptor blockade with tocilizumab is a standard treatment.

With T-cell engagers, CRS is frequent but mostly low grade when step-up dosing is used. In the MajesTEC-1 trial of teclistamab in myeloma, CRS occurred in 72% of 165 patients, with one grade 3 case and no grade 4. In the DeLLphi-301 trial of tarlatamab in small cell lung cancer, CRS occurred in about 51% of patients at 10 mg and 61% at 100 mg, with grade 3 events in about 1% and 6%. Our articles on T-cell engagers in solid tumours and bispecific antibodies in solid tumours cover the wider safety and biomarker picture.

Predictive CRS biomarkers have a longer track record than irAE biomarkers. In 51 patients treated with CD19 CAR-T, Teachey et al. (Cancer Discovery, 2016) found that peak levels of 24 cytokines, including IFN-gamma, IL-6 and sgp130, were associated with severe CRS, and a three-cytokine model predicted it and was validated in an independent cohort. CRP and ferritin tracked CRS but did not predict severe events.

Implications for trial design and safety biomarker vendors

• Pre-specify toxicity-efficacy analyses. Use landmark or time-varying methods, and randomised controls where possible. Naive “ever versus never” comparisons will not survive review.
• Bank baseline samples properly. Serum for autoantibodies and cytokines, PBMCs for TCR sequencing and flow, germline DNA, and stool where feasible. Most of the candidate markers above need a pre-dose sample.
• Sample early. Fatal toxicities cluster in the first weeks, and clonal expansion and cytokine shifts appear before clinical onset. Schedules that only sample at response assessment miss the window.
• Standardise irAE definitions. Inconsistent definitions are one reason trials disagree. Organ-specific endpoints are more tractable than a composite “any irAE”.
• Treat CRS monitoring as a separate assay plan. T-cell engager and cell therapy programmes need rapid-turnaround cytokine testing tied to dosing steps.

For assay, sequencing and CRO companies, the opening is clear. The field has many candidate safety biomarkers and almost no validated, transferable assays. Vendors that can run multiplex cytokines, autoantibody panels and TCR sequencing to a clinical standard, across sites, are what sponsors need to move these markers from association to use.

Discuss safety biomarkers at ImmunoMark Boston

Immune toxicity sits squarely in Stream 1: Next-Gen Immuno-Oncology at ImmunoMark Summit Boston, the East Coast Edition in June 2027. More than 300 senior pharma, biotech and diagnostics leaders meet across two parallel streams, with Stream 2 covering biomarkers and companion diagnostics. The format combines keynotes, panels, case studies, pre-arranged 1:1 meetings and a Day 1 drinks reception, so clinical, translational and safety teams can compare notes with the vendors who run their assays.

Register for ImmunoMark Boston 2027

If your company wants to meet this audience directly, whether you offer immune monitoring assays, multiplex cytokine platforms, TCR sequencing or safety biomarker services, see the Boston 2027 sponsorship opportunities, or read why companies sponsor our cancer immunology conference series.

FAQs

What are immune-related adverse events?

Immune-related adverse events (irAEs) are inflammatory toxicities caused by the immune activation that checkpoint inhibitors produce. They can affect almost any organ, most often the skin, gut, liver, endocrine glands and lungs. In a meta-analysis of 125 trials, 66% of patients on PD-1 or PD-L1 inhibitors had a treatment-related adverse event and 14% had one of grade 3 or higher, and rates are much higher with PD-1 plus CTLA-4 combinations.

Do irAEs mean immunotherapy is working?

Often, but not reliably. Meta-analyses link irAEs to better response and survival, mainly for low-grade events on PD-1 or PD-L1 inhibitors. Many of those studies are affected by immortal time bias, and the association weakens in landmark analyses. A randomised analysis of KEYNOTE-054 found irAEs associated with longer recurrence-free survival on pembrolizumab but not on placebo, which supports a real link, but irAEs are not a tool for treatment decisions.

Are there validated biomarkers that predict irAEs?

No. Candidates include baseline anti-thyroid and other autoantibodies, cytokine signatures, T-cell receptor diversity, early CD8 clonal expansion, HLA alleles, gut microbiome composition and germline variants such as one near IL7. The strongest evidence is for organ-specific markers such as anti-thyroid antibodies, and none is validated for routine clinical use.

How is cytokine release syndrome different from irAEs?

Cytokine release syndrome is an acute systemic inflammatory reaction to mass T-cell activation, seen mainly with CAR-T cells and T-cell engagers. It usually occurs within days of dosing, is graded with ASTCT consensus criteria and is commonly treated with tocilizumab. irAEs from checkpoint inhibitors are typically organ-specific and can arise weeks or months into treatment.