For the past decades, immunotherapy by targeting programmed cell death-1 (PD-1), programmed cell death-ligand 1 (PD-L1), or cytotoxic T lymphocyte associated antigen 4 (CTLA-4) has revolutionized the treatment of cancer. Among these regimens, nivolumab monotherapy (NIVO) or nivolumab plus ipilimumab (NIVO+IPI) have been approved by the US Food and Drug Administration for indications including advanced lung cancer, melanoma, renal cell carcinoma (RCC), head and neck squamous cell carcinoma (HNSCC), hepatocellular carcinoma (HCC), urothelial carcinoma, colorectal carcinoma (CRC) and classical Hodgkin lymphoma. Despite the impressive anti-tumour activity by removing the barrier of immune checkpoint, anti-PD-1/PD-L1 and anti-CTLA-4 reactivate the T cell-mediated anti-tumour immunity, and meanwhile, inevitably break the innate immuno-homeostasis via facilitating the loss of immune tolerance to autoantigens [1], which is associated with the generation of adverse events, known as immune-related adverse events (irAEs).

IrAEs are varied in terms of tissues affected, the severity, and the time of onset relative to the initiation of treatment [2–8]. Numerous clinical trials have outlined a crude profile of irAEs, including skin, gastrointestinal, pulmonary, hepatic, endocrine and renal toxicities [1, 9]. The most common irAEs include pruritus, rash, nausea, diarrhea and thyroid disorders [9]. The vast majority of these irAEs develop within the first few weeks to months after treatment initiation, while others like liver toxicity or hypophysitis appear later [1, 10]. Most irAEs are mild to moderate, except some are potentially fatal, e.g., colitis, pneumonitis, hepatitis, myocarditis, and neurotoxic effects [11]. Hence, there is an urgent need to be acquainted with the toxicological profile. A recent meta-analysis including 125 clinical trials, provides a comprehensive profile for the irAEs of single-agent immunotherapy [12]. However, this meta-analysis merely provided the pooled estimate of irAEs from all anti-PD-1/PD-L1 monotherapy, without announcing the specific data for each one and comparing the difference among these agents. In addition, severer irAEs owing to the combination with anti-CTLA-4 worth more awareness [12], and thus the incidence rates of irAEs for NIVO and NIVO+IPI are of great vitality and remain to be studied.

Not only do irAEs belong to side effects that require intensive care, they also serve as windows into the anti-tumour response of ICIs. The association between irAEs and survival of patients treated with NIVO was first reported in melanoma. In 148 patients with melanoma treated with NIVO, irAEs were associated with better OS using a 12-week landmark [13]. Recent studies have also demonstrated that in non-small cell lung cancer (NSCLC) treated with NIVO, the patients with irAE occurring within 6 weeks post treatment, reached longer progressive-free survival (PFS) and overall survival (OS) than those without irAE [14]. However, irAEs of different organs vary in severity and time of onset [2–8]. The irAEs of grade 3 or 4 according to CATAE v4.0 demand extra usage of prednisone and reduction of ICI dose (or even cessation) [15, 16], relative to poorer prognosis of ICIs [12, 15–18]. Several irAEs are late-onset [3–8], appearing after the confirmation of objective response, which might result in lower relativity. Whether the irAEs of different organs contribute equally in the association with ICI benefit needs to be explored. Besides, whether the association between irAEs and survival benefit can be applied to other tumour types needs to be further studied.

In order to address these issues above, we did this meta-analysis to depict the landscape of irAE incidence rates and to investigate their correlations with the response in patients with advanced solid tumours treated with NIVO or NIVO+IPI.

We completed the largest and most comprehensive meta-analysis to our knowledge concerning the incidence rates of irAEs induced by NIVO and NIVO+IPI, and firstly investigated their correlations with ORR in a pan-cancer setting. From the data of 48 trials including 7936 patients treated with NIVO or NIVO+IPI, we unmasked the positive associations between ORR and the incidence rates of skin and gastrointestinal irAEs. This meta-analysis provides an insight that skin and gastrointestinal irAEs might be potential indicators for the response to NIVO or NIVO+IPI.

This meta-analysis firstly summarizes the irAE profiles of NIVO and NIVO+IPI published to date. The irAE profiles of these two regimens are analogous, of which the most common all-grade irAEs occurred in skin, gastrointestinal and endocrine systems, and the most common grade 3 or higher irAEs took place in hepatic, gastrointestinal and skin systems. Despite this similarity in the relative order, the incidence rates of categorical irAEs were markedly higher when combined with ipilimumab. The deleterious effects of severe irAEs might outweigh the benefit from the addition of ipilimumab, which requires further evaluation on the cost-effective issue. As for clinical practice, close monitoring of the high-incidence irAEs might enable the early diagnosis and treatment, which thereby potentially reduce the usage of steroids and the irAE-related death.

In the present study, we elucidated the positive correlation between ORR and the incidence rates of skin, gastrointestinal and endocrine irAEs in NIVO and/or NIVO+IPI treatments in solid tumours. Autoimmune reactions contribute greatly to the induction of irAEs. Shared T-cell receptor repertoire was revealed in tumour and other tissues appearing irAE [21, 22], partially due to the tumour-tissue similarity [21]. In a recent study of NSCLC, skin ranked 2nd after lung itself and colon ranked 4th across multiple tissues in the tumour-tissue similarity score [21]. T-cells recognize shared antigens of tumour and skin and therefore target both organs, which is associated with the development of skin irAEs and likely with tumour regression as well [23]. Besides the identical T-cell clones within tumour and normal tissue, the mechanism underlying irAEs might include the pre-existing autoimmunity liberated after the successful blockade of immune checkpoint. NSCLC subjects with positive thyroid antibody at baseline before the initiation of ICIs are significantly at the risk of getting thyroid dysfunction as irAE [24, 25], and these preexisting antibodies are associated with better response to ICIs [24, 25]. Thus far, previous studies were almost implemented in NSCLC cohorts, while in our results, despite the omission of NSCLC trials, the correlation between the incidence rates of skin or gastrointestinal irAEs and ORR remained significant among other solid tumours. This result indicates that the principles for the association between irAEs and ORR in NSCLC patients discussed above, could be conceivably extended to other neoplasms. Future discoveries are needed to further elaborate the mechanism underlying this correlation.

Lead time bias critically contaminates the predictive efficacy of irAEs to better response to ICIs as patients who progress either switch to other therapies, while those who respond to immunotherapies have longer treatment duration and more time to develop autoimmune toxicities, especially in studies without a landmark design [23]. However, if all the occurrence of irAEs were on account of the longer period of immunotherapy, then we would expect the same positive correlation between ORR with the incidence rates of all kinds of irAEs, especially for those irAEs with low incidences. However, we didn’t observe the correlation between ORR and renal, hepatic and pulmonary toxicities. In addition, the median time of onset for skin and gastrointestinal irAEs was within 2 months [3–8], indicating that above half of these irAEs occurred before the first evaluation of response. Altogether, these enable skin and gastrointestinal irAEs to be a potential candidate for monitoring the response of NIVO or NIVO+IPI, even before the first CT scan.

We also observed a negative correlation between the incidence rates of pulmonary irAEs and ORR. According to a previous meta-analysis, immune-related pneumonitis accounted for 28.0% of death in clinical trials [9]. In severe cases of potentially fatal irAEs, systemic administration of steroid and discontinuation of ICI treatment will be performed [15, 16] and in most of the clinical trials, steroid use was allowed to manage irAEs. In a retrospective study of NSCLC patients receiving PD-1 checkpoint blockade, patients receiving > 10 mg/day of the steroid prednisone exhibited poorer outcomes (decreased PFS and OS) than patients taking < 10 mg/day [18]. These results might explain the negative correlation between the incidence rates of pulmonary irAEs and ORR in patients treated with NIVO. However, the incidence of steroid use was seldom reported, making it difficult to further analysis the interference of steroid use. The underlying mechanism needs to be further studied.

Our findings may provide important implications for the clinical practice in immunotherapy. Firstly, compared to NIVO, extensive irAEs were observed when combining with ipilimumab, which requires intensive monitoring to prevent the deterioration of lethal irAEs, e.g., pneumonia, and fulminant myocarditis. Secondly, our study indicates the existence of skin and gastrointestinal irAEs might be associated with favorable response to NIVO or NIVO+IPI. Early prediction of responses to NIVO or NIVO+IPI is of great clinical value. Compared with other irAEs, the skin and gastrointestinal irAEs may be used to predict the response to NIVO or NIVO+IPI in clinic.

Several limitations of this meta-analysis need to be stated. First, the possibility of lead time bias cannot be ruled out based on the data extracted from published articles, especially for those irAEs with low incidence rates. However, we didn’t observe the positive correlation between ORR and hepatic, pulmonary and renal irAEs. The median time of onset for skin and gastrointestinal irAEs is within 2 months in most trials [3–8]. The early-onset of these irAEs suggests that this correlation is not simply related to patients who remain on therapy longer being at greater risk to toxicity. Whether these two categories of irAEs can be applied to predict the response to NIVO and NIVO+IPI needs to be further studied with a land-mark analysis. Second, we only performed the meta-analysis in studies of NIVO or NIVO+IPI, where AEs were often reported according to system organ class. Clinical trials of other agents seldom reported irAEs categorized by system organ class, which make it hard to explore the correlation. Third, small-study effects might influence the correlation analysis when included studies with smaller sample sizes and relatively deviated irAE incidence rates. Forth, the number of clinical trials for NIVO+IPI was relatively limited. The correlation between irAEs and ORR was no longer significant if studies of melanoma (n = 9) or NSCLC (n = 4) were omitted. More clinical trials are needed to study the correlation between irAEs and ORR of NIVO+IPI.

This is the first meta-analysis to our knowledge to summarize the incidence rates of irAEs in patients with advanced solid tumours treated with NIVO or NIVO+IPI. The detriment of severe irAEs from the addition of ipilimumab might outweigh the benefit, which requires further comparative studies on the cost-effectiveness. Following correlation analysis uncovers the association between irAEs and ORR in NIVO and/or NIVO+IPI trials across multiple neoplasms, which highlights the potential of predictive value of irAEs to favorable response of immunotherapy. Such an understanding might help in auxiliary distinguishing pseudo-progression and determining whether resuming immunotherapy after the recovery from manageable irAEs, thereby fostering the clinical application of ICIs in patients with advanced solid tumours.