Defining the correlation between immune-checkpoint inhibitors-related adverse events and clinical outcomes: a narrative review

Immune checkpoint inhibitors (ICIs) have increased modern anticancer armamentarium portfolios, with 15%–60% of cancer patients deriving clinical benefit while others progress, including some occurrences of accelerated progressions. ICIs have also introduced a new pattern of immune-related adverse events (irAEs). Recently, a mechanistic link was proposed in which patients who develop ICIs-related irAEs derive a survival benefit compared to those who do not, suggesting an overlap between toxicities and the treatment efficacy. Identifying predictive biomarkers to optimally identify patients who will benefit from ICIs is a contemporary research area in Oncology. However, the data remains sparse, with only several smaller studies showing a plausible direct proportionality of a therapeutic effect across tumours. In contrast, the overall survival and progression-free survival rate depend on the tumour type, degree of toxicities, duration of exposure, affected system/organs and inherent patient characteristics. Furthermore, the occurrence of irAEs appears to be more associated with a clinical benefit from programmed death 1 and programmed death-ligand 1 inhibitors than anti-cytotoxic T-lymphocyte-associated antigen 4. Several questions remain unanswered, including the association between survival benefit and specific type of organ system toxicities, toxicity grade, if the benefit is entirely due to immortal-time biases (ITBs), presence of patients confounding comorbidities like autoimmune diseases, and finally, immune heterogeneities. Considering ITB represents a key element in interpreting these studies since patients with precipitated death or with an earlier disease progresses rarely develop irAEs; in fact, such patients have not stayed in the study long enough to experience such irAEs. Conversely, patients that stayed in the study for a longer period have a higher risk of developing irAEs. Landmark analysis is key in these studies if a real association is to be found. Overall response and disease control rates are mainly higher in those who develop irAEs due to immune activation. So, this review aims to summarise the evidence from key studies that addressed this important clinical question.


Introduction
Our immune system remains our first line of defence in any disease, including cancer. Harnessing it by enhanced immune activation while maintaining an equilibrium of self-tolerance is what modern science continues to achieve.
Cancer cells harbour mechanisms to evade immune surveillance by hijacking the physiological immune negative regulators, posturing as a 'self' cell, and eventually evading immune distraction. Immune checkpoint inhibitors (ICIs) are a class of anticancer drugs known as monoclonal antibodies that target receptors like anti-programmed death 1 (anti-PD-1), anti-programmed death-ligand 1 (anti-PDL-1) and the anti-cytotoxic T-lymphocyte-associated antigen 4 (anti-CTLA-4). By inhibiting these innate negative T-cell regulators, the immune system is activated, and its ability to recognise and kill tumour cells is enhanced [1][2][3] Immunotherapy is currently considered the 'fifth pillar' of cancer treatment, joining the ranks of chemotherapy, radiotherapy, surgery and targeted therapy. Over the years, several ICI agents have been approved, improving the overall survival (OS) of many cancer patients ( Table 1). Amongst the approved ICI agents, anti-PD1 agents have the largest indication across several tumours, including metastatic non-small cell lung cancer (NSCLC) [4], advanced head and neck cancer (HNC) [5], advanced melanoma [6], genitourinary tumours and others [7,8].
Randomised clinical trials in immuno-oncology include the Keynote 045 trial [9], which showed prolonged OS and preservation of good quality of life (QoL) scores in patients with urothelial cancers (UCs).
The Checkmate 141 trial [5] looked at Nivolumab as a first-line therapy in patients with relapsed or metastatic HNC, and it showed an improvement in progression-free survival (PFS) and QoL.
The Checkmate 214 phase 3 trial [10] looked at renal carcinoma patients receiving either double ICI agents or sunitinib and showed a similar positive outcome in both OS and improved QoL scores in the ICI group. While there is a well-documented clinical benefit from ICIs, the activated immune system causes several off-target and bystander effects to normal tissues, leading to the so-called immune-related adverse events (irAEs) [11]. The similarities between antigens presented on normal tissues and tumour cells are the leading hypotheses for developing these adverse events [12]. Similarly, the aetiology of irAEs is akin to those that promote the anti-tumour effect and mainly involves the expansion of the T cell repertoire. Also, ICIs can affect clonal B cell responses, thus inducing autoantibody production [13].
Some irAEs can be subtle and tolerable, while others are rare but life-threatening. irAEs can result in an array of systemic, multi-organ inflammations like liver, thyroid, pituitary, colon and lungs [14].
Studies show that ICIs have lower nausea, vomiting, anorexia, fatigue, diarrhoea, neuropathy, haematologic toxicities and treatment discontinuation rate than conventional chemotherapy. Conversely, the signature irAEs like skin rash, pruritus, colitis, pneumonitis, transaminitis and endocrine toxicities are more common with ICIs than chemotherapy. Lastly, ICIs are overall better tolerated than chemotherapy [15].
Although most irAEs are identified during pre-clinical development, the spectrum, timing and outcomes were only recently investigated through real-world, large-scale pharmacovigilance analyses like the spontaneous reporting systems that looked at the real-time monitoring of these drugs' safety profiles [16]. The term 'immunovigilance' and disproportionality analyses (DAs) have also been proposed. DAs are now the most common epidemiological approach for postmarketing safety assessment and documentation in many ICI agents [17].
There is growing evidence suggesting a potential association between irAEs and clinical benefit in some cancer types [18], and in fact, several reports have shown that irAEs are not only associated with ICI efficacy but also precede response [19][20][21]. However, a possible bias exists in which responders benefit from long-term ICI therapy and are therefore exposed to a higher risk of developing irAEs throughout time [22].
The main bias in the association between irAEs and clinical outcomes is the immortal time bias (ITB); defined as bias arising from methodologically wrong analyses of time-dependent events in survival analyses [23].
Considering ITB represents a key element in interpreting these studies since patients with precipitated death or with an earlier disease progresses rarely develop irAEs; in fact, such patients have not stayed in the study long enough to experience such irAEs. Conversely, patients that stayed in the study for a longer period have a higher risk of developing irAEs.
Furthermore, despite several studies providing interesting data, this association should be interpreted with caution because the majority of these studies did not report or adjusted for the effect of ITB in their survival analyses [24].
Landmark analysis is the recommended solution to such biases since it splits the follow-up period at a given time point. Patients groups are then defined by time-dependent development of irAEs, such as before and after the landmark, with clinical outcomes only considered if occurring after the landmark [23].  Neurologic toxicities are fatal irAEs and usually involve the brain, peripheral nervous system, leptomeninges or even autoimmune destruction of the autonomic nervous system. The antiviral response resulting from the use of ICI agents can result in encephalitis.

Review
The endogenous expression of CTLA-4 on the pituitary can also lead to hypophysitis. Musculoskeletal toxicities usually include myalgias, arthralgias with elevated muscle enzymes signifying significant muscle destruction [26,27].
Cardiotoxicity, which can present as fulminant myocarditis, is one of the rare but catastrophic irAEs. Myocytes express shared antigens with the tumours, which can lead to autoimmune distraction. It usually mimics heart failure symptoms and can also present with arrhythmias and conduction abnormalities. These patients are usually managed with joint help from cardiologists, even as the cardio-oncology sub-speciality continues to become popular [29,30]. A pictorial view of the organ system-based irAEs is shown in Figure 1.

Incidence
Data emerging from randomized clinical trials (RCTs) shows that the frequency and the incidence of irAEs usually depend on the ICI agent used, duration of exposure, dose and patient's risk factors profile. On the other hand, the kinetics and the timing of irAEs are dependent on the affected organ system [21]. A higher incidence of G1 to G5 toxicities is usually common in anti-CTLA-4, moderate with PD-1 class and lower in PD-L1 agents. Intriguingly, different tumour types had different tissue type inflammation resulting from ICI use. For example, colitis was mostly found in melanoma patients [31], whereas pneumonitis was more common in lung cancer patients [32].
ICI monotherapy versus combined ICI is associated with fewer irAEs. A meta-analysis showed that the ICI combo significantly increased the development of any-grade irAEs, including colitis, pneumonitis, hypothyroidism, hypophysitis, amongst others [35]. Similarly, an analysis of 18 studies with 2,767 patients across cancer types showed that combination ICI therapy was associated with a higher risk of all-grade irAEs (Response rate (RR): 1.07; 95% confidence interval (CI): 1.03-1.11) and markedly greater risk of G3 or higher adverse events (RR: 2.21; 95% CI: 1.57-3.10) compared to monotherapy ICI [36].
Retreatment with ICI after a clinically significant previous irAE is common. A multicentre retrospective study of 499 patients identified ICI patients with renal cell carcinoma (RCC) in which 80 patients developed severe irAEs warranting treatment cessation, 36 of whom were able to be restarted on an ICI agent and 44 permanently discontinued. Time to initial irAE was similar between the two groups (2.8 versus 2.7 months; p = 0.59) [37].
Similarly, in a retrospective study of 80 patients with metastatic melanoma who received a combined CTLA-4 and a PD-1 inhibitor, patients who developed irAEs warranting treatment discontinuation were later rechallenged with a single-agent PD-1 inhibitor. The rates of toxicities amongst patients who resumed single-agent anti-PD-1 were notably lower (for example, G3/4 colitis declined to 3%, and G3/4 hepatitis declined to 7%) [38].
Interestingly, it is also common to diagnose irAEs even after the discontinuation of ICI agents. In a retrospective review of 64 patients with metastatic melanoma treated with Nivolumab plus Ipilimumab, 31 patients experienced early toxicity warranting discontinuation of treatment, and 4 of these (13%) further developed a significant irAEs 4 months after treatment cessation. Therefore, both clinicians and patients should be aware of this trend [40].
Therefore, it is crucial to anticipate the occurrence of irAEs and educate patients on the risks. Weber et al [41] conducted a pooled safety data analysis from four studies consisting of 576 patients with irAEs of any grade. They found most of these events occurred within the first 4 months of therapy, and skin was the commonly involved site, followed by the gastrointestinal and the endocrine system. The authors also noted some toxicities occurred as late as day 60.
Lastly, tumour and class-specific irAEs are an unresolved and controversial topic. While anti-CTLA-4 agents are believed to be more toxic than anti-PD-1/PD-L1 drugs, a definitive comparison is lacking due to their different therapeutic uses. However, others have speculated that anti-PD-L1 drugs could be less toxic than anti-PD-1 agents due to the preservation of PD-L2 signalling [43].

Tumour-specific irAEs and clinical outcomes
A mechanistic link was proposed by the investigators of a prospective cohort study in which 73 NSCLC patients treated ICI showed that there was shared cancer and tissue antigens and T cell infiltration in both the lung tumour and the skin, suggesting an overlap between the toxicities and the efficacy [44].
In a similar study of 559 NSCLC patients, the incidence of irAEs was 41.3% (231 patients), and 50 patients (8.9%) developed grade 3 or 4 events. Using a landmark analysis at 6 weeks, the authors confirmed that irAEs of any grade were an independent predictor of higher overall response rate (ORR), longer OS and more prolonged PFS [45].
A study of 99 Phase III oncology RCTs found both toxicity and efficacy were higher amongst the treatment arm than the control (all grade toxicities were 3.5 times higher in the treatment than the control arm, p < 0.001; mean OS of 18.6 versus 16.9 months; p < 0.001; mean PFS of 9.1 versus 7.1 months; p < 0.001, respectively). Notably, the authors concluded that, across the RCTs, irAEs were strongly associated with higher PFS in ICI patients versus controls (p < 0.001), but not OS (p = 0.44) [46].
A systematic review looking at the emerging evidence on the association between IrAEs and clinical outcomes highlighted the existence of a plausible direct proportionality of a therapeutic effect across ICI agents and solid tumours [47]. In a similar melanoma study in the Chinese population of 93 patients who received ICI therapy, frequent irAEs were pruritus, vitiligo, rash and fatigue. The median onset of irAEs was 6.1 weeks. Both disease control rates (DCRs) and ORR were higher in the irAEs group than those without (p = 0.004 and p = 0.003, respectively) [53].
Furthermore, apart from the established impact of cutaneous irAEs on OS of melanoma patients, gastrointestinal irAEs, especially diarrhoea, which usually limits ICI treatment, was paradoxically associated with improved OS and PFS in metastatic melanoma patients (HR 0.56; 95% CI: 0.41-0.76; p < 0.01) [31].
Thyroid irAEs induced by Nivolumab in a cohort of 200 Japanese patients found that patients with overt thyroid uptake as depicted by fluorodeoxyglucose-positron emission tomography (FDG-PET) had a high incidence of irAEs but with an OS benefit ( [56]. Although cutaneous irAEs like vitiligo confers a clinical efficacy in melanoma, the benefit in other cancers remains unknown. A retrospective 6-week landmark analysis of 155 NSCLC patients who received either Pembrolizumab or Nivolumab monotherapy showed that the median time to rash onset was 6.4 weeks. The ORR was higher in patients with cutaneous reactions (57% versus 19%; p < 0.001) and a median PFS of 12.9 versus 3.5 months. Multivariate analysis identified rheumatoid factor positivity as an independent predictor of cutaneous irAEs [57].
A similar retrospective analysis of 40 NSCLC patients on Nivolumab found a better RR in those with skin toxicity (42%) than those without (7%) [12].
Rheumatological irAEs are not commonly reported as signature irAEs in patients receiving immunotherapy. Kostine et al [58] looked at the impact of rheumatologic side effects in all ICI agents and irAEs sites on ORR. The ORR was higher in patients with rheumatologic side effects (85.7% versus 35.3%; p < 0.0001) compared to their counterparts [58].
The association of irAEs and clinical outcomes in NSCLC patients with existing brain metastasis who received ICIs is also not well studied. A small retrospective review of 63 patients found that any grade irAEs were associated with longer OS (21 versus 10 months; p = 0.004) and delayed treatment failure (14 versus 5 months; p = 0.001), especially in patients with high PD-L1 expression [59].

ICI-related pneumonitis (ICIP) is a common and potentially life-threatening irAE in NSCLC patients.
A real-world cohort study looked at the risks of developing ICIP in NSCLC patients who received prior radiotherapy and are currently on ICI therapy. Expectedly, the study found prior radiotherapy was not only associated with a higher risk of ICIP but was an independent prognostic factor for low survival in NSCLC patients [60].
Immunotherapy data in GC is not as robust, and the survival benefit from ICI has been modest. Nonetheless, a multicentre study of 76 GC patients who received Food and Drug Administration (FDA) approved ICI agents showed irAEs were associated with a better outcome in PFS and OS (PFS: 3.9 months versus not reached; HR: 0.13; 95% CI: 0.05-0.3; p < 0.001; OS: 7.4 months versus not reached; HR: 0.11; 95% CI: 0.03-0.36; p < 0.001 [61]. On the other hand, UCs have received more ICI agents FDA approvals than many other cancers. The occurrence and the spectrum of IrAEs corresponded with higher efficacy, higher ORR (52% versus 16%; p < 0.01), better PFS (11.0 versus 3.6 months; p < 0.01) and OS (13.1 versus median not reached) as compared to those without IrAEs [62]. In another study of UC patients, those who developed irAEs had an ORR of 33.0% versus 8%, and at 15 months median follow-up, all the patients with irAEs were still alive, compared to only 38% of their counterparts [63].
Immunotherapy is also approved for metastatic HNC [66]. A similar benefit on RR, PFS and OS has been reported in patients receiving ICI agents who developed irAEs [67]. A single-centre study of 89 patients with relapsed or metastatic squamous cell HNC treated with Nivolumab showed that the median OS was significantly longer in patients with irAEs than those without (17.9 versus 6.3 months; log-rank p = 0.004). The authors also reported a median PFS (10.2 versus 2.8 months) in those with versus without irAEs [68].
Although not commonly reported, ICI therapy-related pyrexia was associated with a longer PFS in one study (6.4 versus 2.1 months; p = 0.26) in patients with advanced NSCLC. Similarly, a better ORR trend was also seen in those who developed pyrexia (65% versus 35%; p = 0.29) [69].
Apart from the common ICI monoclonal antibody-based drugs, monocyte-derived vaccines or dendritic cells (DC) loaded with tumourassociated antigens gp100 are also part of the immuno-oncology armamentarium. In a retrospective analysis of advanced melanoma patients vaccinated with DC vaccines, 84% of the patients developed irAEs, mostly injection site reactions (50%) and flu-like symptoms (67%), and both correlated with immunologic and clinical outcomes (both p < 0.001) [70].
Further compounding the association between the occurrence of irAEs and efficacy conundrum is the common use of steroids to counteract adverse events in patients receiving ICI therapy. In a study by Riudavets et al [71], Prednisone has been found to significantly affect median OS in patients receiving more than 10 mg a day than those with less than 10 mg (6 versus 15.9 months; p < 0.001) [71].
Conversely, a retrospective analysis of 157 patients with different tumour types assessed the effects of irAEs and steroid use on PFS. Forty-five of these patients developed irAEs, in which 21 patients received steroids. Development of irAEs and steroid use was found to improve PFS by Kaplan-Meier estimate. Also, multivariate regression showed that irAEs improved PFS (HR: 0.33; p < 0.001) and persisted even with steroid use (HR: 0.38; p = 0.03). Indeed, steroids' role and their effect on ICI therapy remain an evolving space with many discrepancies [72].
While there are no extensive RCTs data that discredit the association of irAEs with good clinical outcomes, a landmark analysis of a retrospective review of only 97 patients showed no difference in OS at 3 months in patients on single-agent ICI, even though the incidence of irAEs was associated with an OS (24.3 versus 5.3 months; HR: 2.75; 95% CI: 1.54-4.92; p < 0.001) for those with and without irAEs, respectively [32].
A systemic review of 16 studies reported that specific irAEs like mucositis, pneumonitis, myalgias and thyroid disorders are not statistically correlated with OS [73]. Moreover, potential bias and confounding factors were identified in the assumption that patients who develop irAEs remain on treatment longer and hence have a better outcome than their counterparts, and as such, could only be attributed to guaranteetime bias [74,75].
Amongst the coterie of the larger studies looking at this association, the consensus is that there is potentially an association between the development of irAEs and OS, ORR and PFS in patients treated with ICIs, irrespective of disease site, type of ICI and irAEs spectrum. Conversely, higher degrees of toxicity grade (G3 or higher) led to a better ORR only but a worse OS [52]. More similar studies that investigated the association of irAEs and patients' outcomes are shown in Table 2. There is a need for a global initiative to develop an international registry on the reporting of irAEs with well-curated data to gauge the harm and the benefits of ICI agents and further understand if there is a clinically meaningful association between toxicities and survival benefits [76]. Also, an enhanced selection criterion to identify patients with advanced-stage cancers who will or will not benefit from ICI agents is essential, and the need to involve early primary palliative care for the potential non-responders. Patient education and a dedicated multi-disciplinary team are crucial for diagnosing and timely treatment of irAEs [77].

Conclusions and perspectives
Despite all the controversy in this space, the general consensus is that irAEs are not a pre-requisite for ICI benefit but rather a positive heralding sign of clinical benefit. According to all the studies we reviewed, the association between irAEs and clinical outcomes remains disperse. While the objective response rate and DCR are mostly higher in those who develop irAEs, the OS and PFS data are dependent on the tumour type, degree of toxicities, affected spectrum of system/organs and inherent patient characteristics.
The development of irAEs appears to be more associated with both PD-1 and PD-L1 response and benefits as compared to anti-CTLA-4. There is a paucity of large well-powered studies to refute or confirm the exact benefit of the occurrence of irAEs during ICI therapy. Several questions remain unanswered, including the association between irAEs onset and the type of organ system toxicities, toxicity grading, ITBs, confounding comorbidities like autoimmune diseases, immune heterogeneity and patient intrinsic factors. Therefore, a harmonised global registry on the incidence of irAEs and their associated clinical outcomes is required to open a new era of immuno-oncology.