Risk Factors and Characteristics of Checkpoint Inhibitor–Associated Autoimmune Diabetes Mellitus (CIADM): A Systematic Review and Delineation From Type 1 Diabetes

Immune checkpoint inhibitors (ICIs) have resulted in a paradigm shift in the treatment of many cancers, and as a result the indications for their use continue to expand. These agents inhibit specific regulatory immune pathways such as programmed cell death-1 (PD1) and cytotoxic T-lymphocyte activating factor 4 (CTLA4) to increase antitumor immune activity with impressive efficacy across a number of malignancies, most commonly melanoma and non–small cell lung cancer (1–3). A consequence of ICI use is the risk of triggering immune-related adverse events (irAE), a broad and heterogenous group of autoimmune diseases affecting virtually any system within the body. Endocrinopathies are different from other irAE in that resultant hormonal deficits are usually irreversible, except for some thyroid changes (4–6).

Similar to type 1 diabetes (T1D), checkpoint inhibitor–associated autoimmune diabetes mellitus (CIADM) (also termed ICI-DM, CPI-DM) is thought to occur secondary to autoimmune destruction of pancreatic β-cells. Among irAE, CIADM is relatively uncommon, with an incidence of 0.2–1.4% (7–12). Despite this, it is of significant clinical concern due to the high incidence of diabetic ketoacidosis (DKA) at presentation, as well as the lifelong persistent insulin deficiency, and the associated risks of diabetes complications and decreased life expectancy (7–13). Furthermore, as indications for use of ICIs expand into adjuvant and neoadjuvant settings and patient survival improves (2,3), the overall prevalence of CIADM will correspondingly increase.

There are limited data available on CIADM. The relatively uncommon nature of the diagnosis compounded by the heterogenous diagnostic criteria applied thus far has made it difficult to draw conclusions from the case reports, small series, and limited number of systematic reviews published on CIADM to date.

The primary aim for this review is to define clear, easily applied criteria for the diagnosis of CIADM, as well as analyze the presentation characteristics, kinetics, associations, and risk factors for earlier and severe presentations of CIADM.

A systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Supplementary Fig. 1) to address the aims stated above. The review was registered with International prospective register of systematic reviews (PROSPERO) prior to commencement (CRD42021276016). MEDLINE and PubMed databases were reviewed by L.W. for full text English articles published from 1 January 2014 to April 2022. The applied search terms were [“immune checkpoint inhibitor” OR “checkpoint inhibitor” OR “ipilimumab” OR “nivolumab” OR “pembrolizumab” OR “atezolizumab” OR “durvalumab” OR “avelumab”] AND [“diabetes mellitus” OR “type 1 diabetes” OR diabetes*] limited to English language and humans.

Studies were then reviewed by L.W., and all cases of reported CIADM in adults with evidence of ICI exposure and diabetes were included in initial analysis. Where individual patient data were not provided, corresponding authors were contacted via e-mail to request this.

We defined diagnostic criteria for CIADM as evidence of new-onset hyperglycemia in the diabetes range (blood glucose level >11 mmol/L or HbA_1c ≥6.5%) (criteria 1) and insulin deficiency (C-peptide <0.4 nmol/L and/or DKA) (criteria 2) without known treatment with a sodium–glucose cotransporter 2 (SGLT2) inhibitor, in patients who have received checkpoint ICIs.

To avoid “double reporting,” we did not include one review of VigiBase, a World Health Organization database of individual case safety reports, in this review, as no individual data were provided and it would be impossible to determine whether the treating physician had reported the case in another publication (14).

Supplementary Table 2 includes all cases described in the literature as “CIADM” and any reasons for inclusion or exclusion of each patient in our final cohort.

Data were collected including patient baseline demographics, oncological characteristics, CIADM presentations, and follow-up data. Patient data included age, sex, ethnicity, BMI, preexisting type 2 diabetes and medications associated, history of autoimmune disease, family history of autoimmune diabetes, and HLA haplotyping. Oncological characteristics included cancer type, ICI type, ICI response, associated irAE, and previous systemic therapy. Presentation characteristics included time to onset from initial ICI exposure, DKA at presentation, pH, blood glucose, β-hydroxybutyrate, HbA_1c, C-peptide, lipase, presence of acute kidney injury, T1D autoantibody status, steroid use, and immunosuppression. Follow-up data included further C-peptide measurements, insulin use, and exocrine insufficiency. Results that were not definitively reported (e.g., “C-peptide was low”) were reported as missing data and excluded from analysis.

An adapted Newcastle-Ottawa scale was used to assess the quality of included studies. The modified criteria for quality assessment were as follows:

1. Is the case definition adequate—meeting diagnostic criteria for hyperglycemia and insulin deficiency? (Score 0/1)

2. Do the reported cases represent all eligible cases over a defined time period, in a defined catchment area, or an appropriate sample of those cases? (Score 0/1)

3. Have other diagnostic possibilities been excluded, including demonstration that the outcome of interest was not present at the start of the study? (Score 0/1)

4. Is the duration of follow-up adequate to confirm/exclude the diagnosis (minimum 1 month if not meeting criteria at presentation)? (Score 0/1)

Statistical analyses were conducted with R software (version 4.1) and GraphPad Prism (version 9.4.0). Continuous variables were compared with Mann-Whitney U test and/or ANOVA and categorical variables compared with χ² test. Multivariate logistic regression with ANOVA was performed to determine risk factors for occurrence of DKA.

The data set generated from this review is available from the corresponding author on reasonable request.

There were 1,206 articles meeting the search criteria (Supplementary Fig. 1). Of these, a total of 278 patients were identified in the 146 included studies that had reported “CIADM.” These patients were reported to have “CIADM” by the authors on the basis of new-onset diabetes after ICI use, with only some authors applying additional criteria such as autoantibody status or C-peptide levels.

From these, 192 patients met both of our proposed diagnostic criteria for CIADM. Forty-five patients were excluded due to lack of individual data and 31 for insufficient results to determine whether they met criteria. Ten patients did not meet our criteria for CIADM but had been reported to have CIADM in the literature: five patients did not meet criteria for C-peptide (i.e., serum C-peptide ≥0.4 nmol/L at presentation or follow-up), two patients had concurrent use of SGLT2 inhibitors at time of DKA confounding C-peptide measurements and no demonstration of low C-peptide after cessation of SGLT2 inhibitor, two patients had type 2 diabetes that was insulin requiring prior to diagnosis of CIADM, and one patient was excluded due to prior partial pancreatectomy. Details of the quality assessment and exclusion criteria are available in Supplementary Table 2.

Table 1 summarizes the key baseline characteristics for the CIADM patients (n = 192). Mean ± SD age was 63.4 ± 12.8 years, and 62.3% of patients were male, likely reflecting the profile of patients treated with ICIs (e.g., melanoma is most common in older males). Type 2 diabetes was a preexisting condition in 5.2%, which is within the usual range for that age-group. Mean HbA_1c of the subgroup with preexisting type 2 diabetes was 6.6% (range 5.9–8.0), reported in 6 of 17 patients. Mean BMI was 24.7 kg/m², with 32 of 48 (67%) patients having a normal-range BMI at presentation. Melanoma was the most common cancer type (47.8%). All but one patient had exposure to either anti-PD1 or anti-PDL1 therapy as monotherapy or in combination with anti-CTLA4. ICI was first-line therapy for malignancy in 49.7% of patients, while 35.2% had been treated with chemotherapy previously and 8.5% treated with a targeted therapy. Concurrent other irAE at/before time of presentation were present in 43.8%, with thyroid involvement being the most common followed by pituitary. HLA haplotyping was reported in 91 patients (47%), with 54 (59%) having a typical HLA susceptibility haplotype for T1D, while 28 (31%) were neutral and 7 (8%) developed CIADM despite having a haplotype that typically protects against T1D.

Table 1 summarizes the characteristics at presentation for patients with CIADM. Median time to onset of CIADM from initial ICI exposure was 12 weeks (interquartile range [IQR] 6–24) (Fig. 1A).

DKA was present at diagnosis in 69.7%, and the remainder presented with hyperglycemia without acidosis. Mean HbA_1c at presentation was 8.1%, and HbA_1c was overtly elevated (>6.5%) in 84.3%. T1D autoantibodies were present in 40.4% of the 178 patients tested, with anti-GAD in 66 of 72 tested.

Acute kidney injury was present in 27 of 47 (57.4%) patients who had assessment of renal function reported. Steroid use was ongoing and deemed a potential contributor to CIADM presentation in 9 (4.6%) patients. Lipase was tested in 36 of 192 patients, with 69.4% having levels above the reference range and 55% having lipase levels more than two times the upper limit of normal (CTCAE v4.0 grade 3). Two patients had fecal elastase testing performed, both with low values consistent with exocrine insufficiency.

C-peptide was <0.4 nmol/L in 141 of 154 (91.6%) at presentation (n = 154). Of the 13 patients with initial C-peptide ≥0.4 nmol/L, 12 had follow-up C-peptide <0.4 nmol/L and 1 was diagnosed with CIADM based on DKA alone without C-peptide reported. Follow-up C-peptide was reported in 30 of 192 patients, with a mean ± SD value of 0.02 ± 0.04 nmol/L [range 0–0.19]). The time of repeat C-peptide varied from 1 week to 10 months, but all reported values were low (<0.4 nmol/L).

In taking the 141 patients identified with our CIADM criteria as “true positives,” use of lower thresholds of C-peptide would reduce the sensitivity of detection of CIADM to 97.2% for a cutoff of 0.3 nmol/L (4 patients excluded) and 93.6% for a cutoff of 0.2 nmol/L (9 patients excluded). This is largely due to the lack of follow-up C-peptide testing.

All patients included in this review were managed with insulin therapy alone, which could not be stopped during follow-up. Of the 83 of 192 patients with details regarding insulin regimen reported, 75 had multiple daily injections, 5 were on premixed insulin, and 4 were on insulin pump therapy. Insulin requirements were reported for 11 patients, with a mean ± SD requirement of 0.51 ± 0.25 units/kg/day.

Eight patients were given a course of steroids as a trial of therapy for new-onset diabetes and/or concurrent irAE, and 8 patients were given glucagon-like peptide 1 agonists. None of these patients were able to be weaned off insulin, and ultimately they were taken off these medications.

A total of 105 patients had response to ICI therapy reported. The overall response rate was 64% (n = 31, 30% complete, and n = 36, 34% partial), 18 had stable disease, and 20 patients had progressive disease.

Median follow-up duration was 6 months (IQR 3–9), reported in 76 of 192 patients.

Univariate analysis was performed initially to determine associations for each variable with time to onset of CIADM (Supplementary Table 1). Patients with positive T1D autoantibodies (N = 72) presented a mean 3.5 weeks earlier (P = 0.021) than those who were seronegative (N = 106), and associations of other variables (age, sex, cancer type, ICI type, irAE, HLA haplotype, elevation in lipase) were not significant on univariate analysis. Figure 1B depicts the time to onset of autoantibody-positive patients with CIADM in comparison with those without T1D autoantibodies. In one-way ANOVA comparing time to onset of CIADM by type of ICI (anti-PD1, anti–PD-L1, or combination therapy anti-PD1/anti–PD-L1 with anti-CTLA4) association was not significant (P = 0.780, F = 0.25).

Using multivariate logistic regression analysis with ANOVA we found that T1D autoantibody positivity was associated with significantly increased risk of presenting with DKA (odds ratio 3.4, 95% CI 1.6–7.5, P < 0.001) after adjustment for sex, age, cancer type, and ICI type. Figure 1C depicts the increased incidence, and earlier onset of DKA, in autoantibody-positive patients.

Patients who had positive autoantibodies were more likely to have elevated lipase (P = 0.022), and there was greater prevalence of preexisting type 2 diabetes among these patients (P = 0.013). No significant differences between seropositive and seronegative groups were found in serum C-peptide, sex, age, HLA haplotype, presence of other irAE, cancer subtype, or ICI type. As depicted in Fig. 1D, 82% of T1D autoantibody-positive CIADM patients had a single autoantibody positive only, which contrasts with typical findings in patients with T1D.

In subgroup analyses we compared patients with elevated lipase values above the reference range (n = 36) with those who had normal lipase at CIADM presentation. Patients with elevated lipase had significantly lower pH values (mean difference 0.2, 95% CI 0.06–0.32, P = 0.012) and higher glucose levels at presentation (mean difference 16.7 mmol/L, 95% CI 5–27, P = 0.005). Age, sex, cancer subtype, ICI type, presence of irAE, BMI, presence of type 2 diabetes, and HbA_1c, C-peptide, and ketone levels were not significantly different between the groups.

Supplementary Table 2 presents the modified Newcastle Ottawa Scale scores for the studies identified during initial literature review. Overall the quality of the data was poor, with a mean score of 2.6 of 4.0. Case definition criteria were met in 92% of cases and for exclusion of other causes in 89%; however, representation criteria was met in only 12%, and follow-up was adequate in 75%.

We have systematically reviewed reported cases of CIADM and filtered these cases on an individual level based on a combination of hyperglycemia and insulin deficiency. In doing so we present for the first time, and on the largest possible scale, a phenotype of diabetes that is clinically, biochemically, and immunologically distinct (Table 2).

We applied our diagnostic criteria (Fig. 2) to determine with confidence the characteristics of patients with CIADM. The preexisting diagnostic criteria for CIADM are heterogenous, varying from all patients with new-onset diabetes after ICI use to those who are autoantibody positive and those who are also insulin deficient. The inclusion of all patients with new-onset diabetes after ICI use creates a heterogenous population of patients with type 2 diabetes and steroid-induced hyperglycemia, which, as shown in one study, is the most common etiology of new-onset hyperglycemia after ICI use (15). Furthermore, autoantibodies cannot be relied on in the diagnosis of CIADM, as only 40.4% of our cohort was antibody positive, and in the literature seropositivity ranges from 0 to 71% (7–12). Therefore, we sought to use evidence of insulin deficiency (DKA and/or low C-peptide <0.4 nmol/L) to define our cases and exclude differentials.

C-peptide has been used extensively in T1D to aid in the diagnosis of autoantibody-negative T1D and to risk stratify patients with latent autoimmune diabetes of adulthood and, thus, makes a widely available and useful diagnostic tool (16). As our review highlights, serum C-peptide is generally low in patients presenting with CIADM (91.6%), but also all patients who had follow-up testing after initially “normal” C-peptide levels progressed to overtly low C-peptide levels. Early insulin therapy should be considered in all patients with a presumptive diagnosis of CIADM even if initial C-peptide is not low until subsequent C-peptide testing can be performed on follow-up to confirm the diagnosis and guide ongoing use of insulin (Fig. 2).

T1D consensus guidelines from American Diabetes Association and European Association for the Study of Diabetes recommend a threshold of <0.2 nmol/L to confirm a diagnosis of T1D, while C-peptide 0.2–0.6 nmol/L is considered a gray area (17). One key point of difference in CIADM is that the expected prevalence of T1D antibodies is significantly lower than in T1D, and thus we feel that C-peptide cutoffs need to be set at a higher, more sensitive threshold. In the right clinical context, a C-peptide <0.4 nmol/L collected at the time of hyperglycemia appears to be specific for CIADM; we identified no cases of recovery from diabetes with C-peptide below this level. We did identify four patients who presented with symptomatic diabetes but did not meet our suggested C-peptide/DKA criteria for CIADM (18–21). These four people were able to cease insulin during follow-up. However, in the absence of clinical features pointing at type 2 diabetes or steroid use, the added value of C-peptide testing has not been shown.

Patients with CIADM have distinct clinical differences in comparison with those presenting with classic T1D, as summarized in Table 2. Our results confirm previous associations in case series and reports that CIADM is a condition nearly exclusively associated with anti-PD1 and anti–PD-L1 therapy, and 50% of patients will present by 12 weeks of ICI commencement. Disease onset is not only more rapid but it also more fulminant, with a high incidence of DKA and rapid loss of endogenous insulin production. Meanwhile, in classic T1D no unifying precipitant has been identified, and the onset of disease is commonly much more gradual with lower incidence of DKA and with autoantibodies preceding clinical disease by years (22,23).

The prevalence of typical T1D susceptibility haplotypes is lower in the CIADM cohort than in classic T1D patients (24). In particular, a significant proportion of patients developed CIADM despite protective haplotypes. This is likely to reflect that CIADM patients are derived from a population that lacks a strong propensity for T1D, as demonstrated by the protection from T1D until a mean of 63.4 years of age.

T1D autoantibody prevalence is substantially lower in the CIADM cohort, perhaps reflective of differing autoimmune responses to ICI therapy. Studies in humans and animals indicate that ICIs predominantly lead to T cell activation responses (25–28), and it is possible that traditional T1D-associated B cell–mediated autoantibody production is bypassed in CIADM. Similar to T1D, anti-GAD was the most commonly positive autoantibody, but interestingly the prevalence of other autoantibodies was significantly lower than for T1D as well (Fig. 1D).

A key finding of this work is that patients who were seropositive for classic T1D autoantibodies exhibited more rapid and severe onset of diabetes. This is a confirmatory finding consistent with those of two previous systematic reviews in this area (29,30). This difference may be a result of differential immune pathway activation between these two groups or the presence of a “primed” immune system leading to a more fulminant response to checkpoint inhibition. Of the six patients with CIADM with autoantibody results reported from pretreatment time points, three had positive autoantibodies prior to ICI exposure, two seroconverted to autoantibody positivity, and one remained seronegative throughout (8,31–33). Unfortunately, these numbers are too limited for drawing strong conclusions and likely skewed toward retrospective testing of autoantibody-positive individuals.

Understanding the immunology of CIADM is challenging, due to not only the rarity of the disease but also the limited accessibility of the human pancreas and challenges in identifying an islet-specific immune cell population in peripheral blood. In studies of human peripheral blood mononuclear cells in four HLA-A2⁺ CIADM patients, investigators found that only two had increased levels of classic T1D antigen–specific CD8⁺ T cells, of which the majority were effector or memory T cells (34). In the two patients with CIADM for whom postmortem pancreatic tissue was available for histological analysis, both showed inflammatory infiltrate within the pancreas, in both endocrine and exocrine compartments (28,35). Additional staining in a recent case reported by Perdigoto et al. (28) also showed β-cell expression of PDL1 and IDO1 ligands in manner similar to that in pancreatitis control subjects, suggestive of inhibitory immune ligand expression in response to inflammation. Interferon-γ and TNF-α expression was also high in the peri-islet inflammatory infiltrate.

Akin to classic T1D, of the myriad of treatments with demonstrated efficacy in CIADM prevention in NOD mice (anti-CD3, repeated antigen exposure, anti–IFN-γ, anti–TNF-α, JAK1/2 inhibitor), none have yet been used successfully in a patient with definite CIADM. Although two patients labeled as having CIADM became insulin independent after infliximab treatment, neither demonstrated a low C-peptide or manifest DKA to suggest overt insulin deficiency, which makes it difficult to determine the cause for hyperglycemia (19,20).

The cohort of CIADM patients reported exhibited higher prevalence of lipase elevation than the general ICI-treated cohort (55.2% vs. 26.9%, respectively, with grade 3 lipase elevation) (36). Lipase elevations in T1D otherwise have been reported in the Asian-predominant fulminant phenotype (37) but generally are lower at presentation than in normal control subjects by time of presentation of classic T1D patients (38). Studies of pancreatic volumetry in CIADM patients have shown rapid loss of pancreatic volume by time of presentation beyond what can be attributed to β-cell mass alone (20,39). Volume loss has also been reported on MRI scans of newly diagnosed patients with classic T1D, although to a lesser extent (38). More research would be required to determine whether presence of pancreatic inflammation increases risk of subsequent CIADM, which would suggest that pancreatic inflammation may expose pancreatic epitopes that promote pancreatic autoimmunity. Furthermore, studies in T1D have shown that loss of endogenous insulin production directly impacts on acinar exocrine function and that subcutaneous insulin administration delivers much lower concentrations of insulin to acinar cells than local production by β-cells (40).

This review represents the largest number of confirmed CIADM cases reviewed in literature, with the inclusion of a stringent diagnostic criteria to better define the clinical phenotype of CIADM and exclude other differentials. Although previously Wright et al. (14) reported 283 cases of new-onset diabetes after ICI use, as the authors acknowledge there were insufficient data available to distinguish between different forms of diabetes. Two other systematic reviews in the area have included all patients with new-onset diabetes after ICI, but the investigators acknowledged the limitations of including a heterogenous population. Akturk et al. (29) found that 100% of patients with C-peptide measured in their cohort had low values consistent with insulin deficiency, and Lo Preiato et al. (30) found that “C-peptide levels were usually and permanently compromised.”

A further strength of this review is in the exclusive use of individual-level data, which allows for more informative analysis. We believe that these findings will allow clinicians to assess patients exposed to ICI who present with hyperglycemia with a greater degree of confidence. The implication of a missed diagnosis of CIADM is serious, given the high prevalence of DKA at diagnosis, and thus increased awareness of the defining characteristics of this condition is important.

The main limitation of this review is in the fragmented nature of reporting found in collated articles. Certain variables such as HLA haplotyping, cancer response, lipase, fecal elastase, and follow-up data were infrequently reported. We acknowledge that this review is subject to language bias (inclusion of English articles only), thus potentially skewing the representation of other populations, and publication bias. Given the rare nature of CIADM diagnoses there are also no data on a relevant control population. Future prospective studies in this area would be useful in clarifying this condition.

In this review we analyze the individual characteristics of patients presenting with CIADM and find that CIADM patients have many areas of distinction from patients with T1D. In particular, the prevalence of T1D autoantibodies is significantly lower than in classic T1D, and autoantibody-positive CIADM patients present earlier with diabetes and with significantly greater risk of DKA at presentation. This necessitates the development of a separate set of diagnostic criteria and further reporting of characteristics such as exocrine pancreatic function for better understanding of this novel form of diabetes.

Funding. A.M.M. is supported by a National Health and Medical Research Council Investigator Grant, Nicholas and Helen Moore, and Melanoma Institute Australia.

Duality of Interest. A.M.M. has served on advisory boards for Bristol-Myers Squibb, Merck Sharp & Dohme, Novartis, Roche, Pierre-Fabre, and QBiotics. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. L.W., V.T., and J.E.G. were involved in the conception and design of the study. L.W. conducted the review and analysis and wrote the first draft of the manuscript. All authors were involved in the editing, review, and approval of the final version of the manuscript.

Prior Presentation. Parts of this study were presented as an oral presentation at the Australasian Diabetes Congress 2022, Brisbane, Australia, 8–10 August 2022, and Endocrine Society of Australia Annual Scientific Meeting, Christchurch, New Zealand, 11–13 November 2022.