Immune checkpoint inhibitors (ICIs), including the CTLA-4 inhibitor ipilimumab and PD-1 inhibitors nivolumab and pembrolizumab, are now first-line therapy for a variety of locally-advanced and metastatic malignancies. These potent immune-modulating agents are associated with serious and potentially fatal immune-related adverse events (irAEs).1 Therefore, caution is warranted when considering ICI therapy in patients with potentially complicating pre-existing conditions, including patients with autoimmune disease (AID).

Approximately 15% to 30% patients treated with ICI monotherapy and significantly higher percentage (>50%) of patients treated with dual ICI therapy develop ≥3 grade irAEs.2 irAEs can affect nearly every organ system,3 though most commonly involve the gastrointestinal (GI) tract, the endocrine system, lung, skin, and liver.4 It is hypothesized that patients with a genetic predisposition or a pre-existing AID are at higher risk for irAEs.5 In the setting of safety concerns and a paucity of data regarding ICI use in patients with pre-existing AID, these patients were excluded from early clinical trials that have served as the foundation for Food and Drug Administration approval and clinical use. It is estimated that AIDs effect at least 10% of the general population (with recent estimates in cancer patients high as of 13% to 25%).6

A number of retrospective studies have examined the safety and efficacy of using checkpoint inhibitors in patients with AID, with mixed results concerning the relative safety of ICIs in patients with pre-existing AID. Several studies suggested that immune toxicities in these patients were common but comparable to rates in patients without AID, with 23% to 50% of patients developing AID flares and 29% to 42% developing irAEs.7–12 irAEs in these studies were generally manageable with immunosuppression but severe or fatal reactions did occur, though infrequently. One large multicenter study directly compared rates of GI adverse events in patients with pre-existing inflammatory bowel disease (IBD) versus non-IBD controls, finding significantly higher rates of GI adverse events in patients with pre-existing IBD (41% vs 11%), including 21% of patients with IBD experiencing grade 3 or 4 diarrhea and 4% experiencing colonic perforation.13 These data suggest that while toxicity risk for patients with AID might not be dramatically increased, certain subgroups may be at increased risk, and further data on this question is warranted.

To contribute to this expanding literature on the safety of ICIs in patients with pre-existing AID, we conducted a retrospective analysis of a large claims database to assess toxicity-related interventions occurring subsequent to ICI treatment in patients both with and without AID. We compared rates of immunosuppressant use and hospitalization following checkpoint inhibitor treatment in 284 patients with lung cancer or advanced melanoma and pre-existing AID compared with 3230 control patients with a diagnosis of either lung cancer or melanoma without a diagnosis of AID.

To our knowledge, this is the largest study evaluating the toxicity of ICI therapy in cancer patients with AIDs and comparing with patients with no AID. This also the first to use a large medical-pharmacy claims data set from numerous facilities. Here, we demonstrate a significant association of increased corticosteroid treatment and hospitalization rates following ICI therapy in cancer patients with AID compared with patients without AID.

Due to the intervention-based nature of medical claims data, we employed steroid use as proxy measurements of clinical irAEs that plausibly represent clinical responses to ICI toxicity, though this represents a ceiling on the rate of irAEs as steroids could plausibly be used for alternative purposes. Because intravenous corticosteroids are more likely to be used in the context of more serious toxicities and oral corticosteroid treatment more likely used for moderate toxicity,4 our data could be interpreted as approximations of moderate and severe irAEs for rates of oral steroids and intravenous steroids, respectively.

Rates of corticosteroid use following treatment with PD-1 inhibitors in our study are similar to previous clinical studies that used clinical records alone. In a prior study that determined rates of treatment with prednisone for PD-1 therapy toxicity in patients with AID and NSCLC, 14.3% of patients (8/56) were treated with prednisone following treatment,8 compared with 14.1% of patients with AID with all types of lung cancer treated with PD-1 inhibitors in our study (10/71). In a study of the toxicity profile of PD-1 inhibitors in patients with advanced melanoma, 38% of patients (20/52) experienced a reaction requiring immunosuppression, compared with 24% of patients (6/25) requiring oral prednisone in our study.9

Similar to a single study comparing toxicity in a group of AID patients to non-AID controls, our study found increased rates of both oral and intravenous steroid use following treatment with ICIs.13 This previous study specifically evaluated patients with IBD. Our results support increased toxicity risk for patients with AID, but limited sample size for individual diseases prevent subgroup analyzes at the level of specific conditions. It is likely that both the severity of the disease and certain conditions—for example, IBD—place patients at varying degrees of increased risk. Several other retrospective series of AID patients do not compare the toxicity rates with non-AID patients, so it is unclear in these studies if the treatment toxicities are notably worse in patients with AID.7–12 Of note, the rates of steroid use in AID patients in our study are comparable to these studies. However, caution should be applied when comparing rates of steroid use in this study to irAE rates more generally. Missing interventions may have arisen from incomplete data collection, and some irAEs may not be treated with systemic steroids at all. Similar to the study of patients with IBD, our study supports that patients with AID have increased toxicity risk and provided an estimate of this risk, but it is not suggestive of AID being a clear contraindication to treatment.

An unexpected result of our study is that there were lower rates of corticosteroid treatment and hospitalization in patients with AID following treatment with ipilimumab relative to PD-1 inhibitors (tables 2 and 3), despite the fact that ipilimumab has been shown to be more toxic than PD-1 inhibitors in previous studies.9 One potential explanation is that prior knowledge of the toxicity of ipilimumab caused increased hesitation from medical providers when prescribing and continuing ipilimumab, leading to fewer cycles of treatment and ultimately less toxicity. This is consistent with our data, as ipilimumab had the lowest median number of infusions per patient across all checkpoint inhibitors at 1.

Several methodological limitations must be taken into account when interpreting our results. Retrospective analysis of medical claims data does not indicate causation. It is possible that the interventions used in this study as indications of toxicity were not in fact driven by checkpoint inhibitor toxicity, but from another cause, such as cancer morbidity. This is particularly relevant when interpreting the hospitalization results, given that overall rates of hospitalization following checkpoint inhibitor therapy were much higher among lung cancer patients relative to advanced melanoma patients despite the fact that rates of corticosteroid use were similar across cancer type. Our analysis also did not distinguish patients with active AID from inactive AID, and in previous studies of patients with AID receiving ICI treatment the majority of patients did not have active disease.8 Similarly, our results may not be generalizable to patients with active AID.

Our inclusion criteria led to the exclusion of numerous patients within the database, which is typical for large claims-based analysis such as Surveillance, Epidemiology, and End Results (SEER)-Medicare. Because these criteria were solely based on measures of completeness of the data set for individual patients, it is unlikely to skew comparisons across groups, but may lead to the exclusion of patients who were deceased prior to 180 days following their final ICI infusion. The database used had limited survival data, rendering it impossible to discern whether a lack of medical claims during the period following therapy was due to missing data from the database or from patient mortality. Because of this, our exclusion criteria for medical claims analysis included the requirement that the patient had at least a single claim 180 days following the final administration of immunotherapy, implying that the patient survived throughout this period. While this criterion reduces the likelihood of false negatives due to missing data, it also ensures that patients who did not survive during this time period—potentially as a result of fatal treatment-related toxicity—are not included within our analysis.

In summary, our findings suggest an association of increased toxicity risk for AID patients being treated with ICIs compared with patients with no AID. This is consistent with a cooperative effect between AID and ICI, and AID patients should be monitored carefully for immune-related toxicities. Higher levels of evidence with prospective clinical trials are needed to determine the immune-related toxicity profile and safety.