Autoimmune Disease Risk With GLP-1RA, DPP-4i, and SGLT2i Treatment in Patients With Diabetes.

INTRODUCTION

Dipeptidyl peptidase‐4 inhibitors (DPP‐4i), glucagon‐like peptide‐1 receptor agonists (GLP‐1RA), and sodium‐glucose cotransporter‐2 inhibitors (SGLT2i) have become mainstays of therapy for type 2 diabetes mellitus, together reaching over 1 in 10 adults in the United States.
1
,
2
Although extensive cardiorenal outcome data exist for GLP‐1RA, SGLT2i, and DPP‐4i, the immunologic safety of these agents remains relatively uncertain; current evidence is often limited to small or uncontrolled observational studies and disparate case series that offer conflicting signals of both protection and risk for autoimmune disease.
2
,
3
,
4
,
5
,
6
,
7
,
8
Crucially, data systematically comparing autoimmune disease risk across these different targeted antidiabetic agent classes within the same population remains relatively limited.
4
,
6
,
7
,
8
To address this gap, we used data from a large‐scale federated health network to conduct emulated target trials evaluating incident autoimmune disease among patients with diabetes treated with SGLT2i, GLP‐1RA, and DPP‐4i. These three classes were selected because they represent commonly used antidiabetic agents and exert distinct biologic actions with plausible immunomodulatory consequences, thereby warranting direct comparison of autoimmune outcomes.
2
,
3
,
4
,
5
,
6
,
7
,
8

PATIENTS AND METHODS

Data acquisition

The data for this study were obtained from the TriNetX platform, a Health Insurance Portability and Accountability Act and General Data Protection Regulation‐compliant federated network, accessing deidentified electronic health records from more than 150 million individuals across member health care organizations in over 20 countries. Data were accessed through the Global Collaborative Network (with natural language processing) subset, encompassing 152 health care organizations. This study followed the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) reporting guidelines.

Patient selection and study design

We conducted emulated trials to compare the incidence of autoimmune disease among patients with type 2 diabetes (T2D) after treatment with either DPP‐4i, GLP‐1RA, or SGLT2i.
9
The index date was the first prescription of each medication analyzed.

We included adult patients with a new diagnosis of T2D between 2016 and 2023 with first prescription for each respective agent within a year of diagnosis. Previously used International Classification of Diseases Tenth Edition (ICD‐10)‐based coding approaches, further confirmed by medication prescription, were used for cohort creation.
10
,
11
Exclusion criteria included patients ≤18 years of age, patients who received any study agent before the index date other than the index agent, and individuals diagnosed with any autoimmune disease or exposed to any immunosuppressant before or on the day of the index event (Supplementary Material  1 ). Combination therapy with any other study medication was excluded to prevent confounding from treatment switching and mixed exposure effects.

Because of its deidentified nature, this study was deemed exempt by the Mass General Brigham Institutional Review Board.

Outcomes

Primary outcomes included three‐year incidences of systemic lupus erythematosus, cutaneous lupus erythematosus, rheumatoid arthritis, psoriasis, psoriatic arthritis, multiple sclerosis, inflammatory bowel disease, antineutrophilic cytoplasmic antibody‐associated vasculitis, giant cell arteritis, dermatomyositis, systemic sclerosis, celiac disease, autoimmune thyroiditis, myasthenia gravis, pemphigus, bullous pemphigoid, polymyositis, and polymyalgia rheumatica.

The follow‐up period was selected based on the availability of follow‐up data to ensure reliability and minimize potential bias in outcome analysis. A landmark was set at six months after the index event, requiring patients to survive this landmark period before entering the analysis, with follow‐up beginning from this landmark time to mitigate immortal time bias and reverse causation bias.
12
,
13
Where possible, primary outcomes were identified using ICD‐10‐based coding approaches that have been validated and used in prior research.
14
,
15
,
16
,
17
,
18
,
19
,
20
,
21
,
22
,
23
,
24
,
25
,
26
,
27
Trauma (laceration of the hand), lipoma, and inguinal hernia diagnoses served as negative control outcomes that would not be expected to differ by exposure medication status.
28
Change in body mass index (BMI) and hemoglobin A1c (HbA1c) over the study period was evaluated as a positive control analysis to confirm medication exposure and alignment with expected treatment effects.
29

Statistical analysis

To emulate an intention‐to‐treat analysis, patients were analyzed according to their initial treatment assignment. To mitigate the risk of confounding from treatment switching and mixed exposure effects, combination therapy with any other study medication was excluded. To emulate randomization, we used 1:1 propensity score matching with a greedy nearest‐neighbor algorithm to adjust for between‐group differences in characteristics such as age, sex, race and ethnicity, comorbidities such as hypertension, obesity and overweight, HIV and other infectious exposures, smoking status, medication use, and laboratory data including HbA1c and estimated glomerular filtration rate, and medication use (full list of covariates is shown in Table  1 ).

Table 1 Baseline patient characteristics in DPP‐4i versus GLP‐1RA‐treated diabetes cohorts *
Characteristic Before PSM After PSM DPP‐4i cohort (n = 179,550) GLP‐1RA cohort (n = 253,941) SMD DPP‐4i cohort (n = 118,419) GLP‐1RA cohort (n = 118,419) SMD Demographics Age at index, mean (SD) 64.9 (13.2) 56.9 (13.0) 0.609 61.8 (12.8) 61.6 (11.9) 0.021 Sex, n (%) Female 85,809 (47.8) 138,052 (54.4) 0.132 58,311 (49.2) 57,877 (48.9) 0.007 Male 93,738 (52.2) 115,869 (45.6) 0.132 60,107 (50.8) 60,536 (51.1) 0.007 Unknown 10 (0.0) 16 (0.0) 0.001 ≤10 (0.0) ≤10 (0.0) <0.001 Race, n (%) American Indian or Alaska native 468 (0.3) 1,288 (0.5) 0.040 426 (0.4) 409 (0.3) 0.002 Asian 13,763 (7.7) 13,034 (5.1) 0.104 8,310 (7.0) 8,222 (6.9) 0.003 Black or African American 28,246 (15.7) 47,114 (18.6) 0.075 22,024 (18.6) 21,358 (18.0) 0.015 Native Hawaiian or Other Pacific Islander 3,304 (1.8) 5,335 (2.1) 0.019 2,523 (2.1) 2,514 (2.1) 0.001 Other race 2,549 (1.4) 4,363 (1.7) 0.024 2,021 (1.7) 1,933 (1.6) 0.006 Unknown race 40,848 (22.8) 10,171 (4.0) 0.573 7,666 (6.5) 9,211 (7.8) 0.051 White 90,371 (50.3) 172,632 (68.0) 0.365 75,449 (63.7) 74,772 (63.1) 0.012 Comorbidities, n (%) Hypertension 118,667 (66.1) 182,233 (71.8) 0.123 83,144 (70.2) 82,747 (69.9) 0.007 Overweight and obesity 37,855 (21.1) 134,544 (53.0) 0.700 36,315 (30.7) 37,037 (31.3) 0.013 Chronic viral hepatitis C 1,792 (1.0) 2,074 (0.8) 0.019 1,161 (1.0) 1,111 (0.9) 0.004 Chronic viral hepatitis B without delta‐agent 1,066 (0.6) 615 (0.2) 0.055 354 (0.3) 370 (0.3) 0.002 Chronic viral hepatitis B with delta‐agent 46 (0.0) 21 (0.0) 0.013 14 (0.0) 13 (0.0) 0.001 Alcohol abuse 2,967 (1.7) 5,103 (2.0) 0.027 2,355 (2.0) 2,304 (1.9) 0.003 Smoking status (nicotine dependence) 19,106 (10.6) 38,705 (15.2) 0.137 15,803 (13.3) 15,749 (13.3) 0.001 HIV 595 (0.3) 1,357 (0.5) 0.031 533 (0.5) 510 (0.4) 0.003 CKD 36,766 (20.5) 38,117 (15.0) 0.143 21,474 (18.1) 21,387 (18.1) 0.002 Diabetic nephropathy 32,115 (17.9) 42,344 (16.7) 0.032 20,480 (17.3) 20,419 (17.2) 0.001 Diabetic neuropathy 27,844 (15.5) 48,044 (18.9) 0.090 20,829 (17.6) 20,564 (17.4) 0.006 Diabetic retinopathy 13,888 (7.7) 21,958 (8.6) 0.033 9,914 (8.4) 9,725 (8.2) 0.006 Infectious mononucleosis 59 (0.0) 225 (0.1) 0.023 52 (0.0) 64 (0.1) 0.005 BMI 30–39 kg/m 2
16,335 (9.1) 61,853 (24.4) 0.418 15,873 (13.4) 16,362 (13.8) 0.012 BMI 40.0–44.9 kg/m 2
5,251 (2.9) 33,057 (13.0) 0.379 5,213 (4.4) 5,811 (4.9) 0.024 BMI 45.0–49.9 kg/m 2
2,713 (1.5) 20,371 (8.0) 0.309 2,705 (2.3) 3,129 (2.6) 0.023 Medications Immune checkpoint inhibitors 129 (0.1) 136 (0.1) 0.007 76 (0.1) 64 (0.1) 0.004 Estrogen‐containing oral contraceptives 6,596 (3.7) 13,662 (5.4) 0.082 4,438 (3.7) 4,517 (3.8) 0.003 Methimazole 643 (0.4) 993 (0.4) 0.005 444 (0.4) 438 (0.4) 0.001 Propylthiouracil 86 (0.0) 58 (0.0) 0.013 25 (0.0) 33 (0.0) 0.004 Sulfonamides 31,991 (17.8) 41,997 (16.5) 0.034 20,102 (17.0) 19,847 (16.8) 0.006 Spironolactone 7,522 (4.2) 14,349 (5.7) 0.068 5,330 (4.5) 5,275 (4.5) 0.002 Hydralazine 22,041 (12.3) 27,856 (11.0) 0.041 14,085 (11.9) 13,904 (11.7) 0.005 Procainamide 38 (0.0) 73 (0.0) 0.005 29 (0.0) 21 (0.0) 0.005 Isoniazid 584 (0.3) 517 (0.2) 0.024 344 (0.3) 324 (0.3) 0.003 Minocycline 519 (0.3) 979 (0.4) 0.017 308 (0.3) 344 (0.3) 0.006 Metformin 78,710 (43.8) 134,821 (53.1) 0.186 55,717 (47.1) 56,328 (47.6) 0.010 Sulfonylureas 45,015 (25.1) 52,647 (20.7) 0.103 28,159 (23.8) 28,098 (23.7) 0.001 Alpha‐glucosidase inhibitors 1,728 (1.0) 664 (0.3) 0.090 389 (0.3) 433 (0.4) 0.006 Insulin 62,544 (34.8) 103,918 (40.9) 0.126 44,690 (37.7) 44,920 (37.9) 0.004 Laboratory Hemoglobin A1c, mean (SD) 8.1 (2.1) 8.2 (2.1) 0.035 8.2 (2.1) 8.2 (2.1) 0.001 6.5%–7%, n (%) 70,163 (39.1) 125,979 (49.6) 0.213 50,245 (42.4) 49,394 (41.7) 0.015 7%–8%, n (%) 69,043 (38.5) 105,772 (41.7) 0.065 47,478 (40.1) 47,528 (40.1) 0.001 8%–9%, n (%) 45,303 (25.2) 76,657 (30.2) 0.111 32,836 (27.7) 33,025 (27.9) 0.004 ≥9%, n (%) 47,113 (26.2) 92,709 (36.5) 0.223 36,415 (30.8) 36,966 (31.2) 0.010 Glomerular filtration rate, mean (SD) 71.7 (32.8) 80.9 (28.3) 0.299 76.2 (31.8) 76.0 (28.3) 0.006 60–90 mL/min/1.73 m 2 , n (%) 88,204 (49.1) 152,195 (59.9) 0.218 64,742 (54.7) 64,246 (54.3) 0.008 45–60 mL/min/1.73 m 2 , n (%) 53,798 (30.0) 70,471 (27.8) 0.049 35,226 (29.7) 34,828 (29.4) 0.007 30–45 mL/min/1.73 m 2 , n (%) 34,939 (19.5) 35,786 (14.1) 0.144 20,300 (17.1) 20,218 (17.1) 0.002 15–30 mL/min/1.73 m 2 , n (%) 18,229 (10.2) 14,590 (5.7) 0.163 9,256 (7.8) 9,214 (7.8) 0.001 <15 mL/min/1.73 m 2 , n (%) 23,795 (13.3) 18,915 (7.4) 0.191 11,512 (9.7) 11,574 (9.8) 0.002 * One representative trial (DPP‐4i vs GLP‐1RA) shown for illustrative purposes; all trial cohorts matched on identical covariates. BMI, body mass index; CKD, chronic kidney disease; DPP‐4i, dipeptidyl peptidase‐4 inhibitor; GLP‐1RA, glucagon‐like peptide‐1 receptor agonist; PSM, propensity score matching; SMD, standardized mean difference.

Baseline patient characteristics in DPP‐4i versus GLP‐1RA‐treated diabetes cohorts *

One representative trial (DPP‐4i vs GLP‐1RA) shown for illustrative purposes; all trial cohorts matched on identical covariates. BMI, body mass index; CKD, chronic kidney disease; DPP‐4i, dipeptidyl peptidase‐4 inhibitor; GLP‐1RA, glucagon‐like peptide‐1 receptor agonist; PSM, propensity score matching; SMD, standardized mean difference.

Follow‐up for all cohorts extended until the onset of autoimmune disease (first receipt of relevant ICD‐10 code), withdrawal from the TriNetX database, death, or the study's conclusion on February 25, 2025. Censoring occurred at the earliest of the following events: development of an outcome of interest, death, loss to follow‐up, or the end of the study period.

Hazard ratios (HRs) for all outcomes were estimated using Cox regression models on weighted samples. Statistical analyses were conducted using Python version 3.7 (Python Software Foundation). Given the multiple comparisons evaluated in our study, P values were adjusted to control the false discovery rate at 5% using the Benjamini–Hochberg procedure.

Data sharing statement

Data used in this study were obtained from the TriNetX Research Network, a federated health research network. Because of proprietary restrictions and institutional data sharing agreements, individual‐level data cannot be shared publicly.

RESULTS

Study sample and baseline characteristics

We performed three emulated target trials that included 118,419 matched pairs of DPP‐4i and GLP‐1RA initiators, 102,810 matched pairs of DPP‐4i and SGLT2i initiators, and 105,869 matched pairs of GLP‐1RA and SGLT2i initiators. Before propensity‐score matching, DPP‐4i initiators were the oldest of the three groups (mean 64.9 [SD 13.2] years), followed by SGLT2i users (mean 64.2 [SD 12.7] years) and GLP‐1RA users (mean 56.9 [SD 13.0] years); they also had the lowest proportion of females (47.8% vs 54.4% and 39.4%, respectively) and had the lowest proportion of White race (50.3% vs 68.0% and 63.9%, respectively). The median follow‐up time was 1,095 days for the DPP‐4i cohort, 861 days for the GLP‐1RA cohort, and 743 days for the SGLT2i cohort. After matching, there were minimal differences in all baseline characteristics between groups (standardized mean difference <0.10) (Table  1 ).

DPP ‐4i versus GLP ‐ 1RA trial results

In the DPP‐4i versus GLP‐1RA trial, DPP‐4i treatment was associated with lower risks of psoriasis (HR 0.79, 95% confidence interval [CI] 0.70–0.85), PsA (HR 0.65, 95% CI 0.53–0.79), and autoimmune thyroiditis (HR 0.68, 95% CI 0.59–0.76), but higher risks of dermatomyositis (HR 2.18, 95% CI 1.24–3.53) and bullous pemphigoid (HR 1.78, 95% CI 1.24–2.46) (Table  2 , Figure  1 ). No risk differences were observed for the prespecified negative controls (hand laceration, inguinal hernia, and lipoma). For the positive control analysis, both cohorts demonstrated statistically significant declines in mean BMI and HbA1c from baseline to three years ( P < 0.001 for each marker); the reduction in BMI was significantly greater with GLP‐1RA than with DPP‐4i ( P < 0.001) (Supplementary material  1 ).

Table 2 Occurrence and risk of autoimmune disease in patients with diabetes treated with DPP‐4i versus GLP‐1RA, DPP‐4i versus SGLT2i, and GLP‐1RA versus SGLT2i *
DPP‐4i vs GLP‐1RA trial DPP‐4i vs SGLT2i trial GLP‐1RA vs SGLT2i trial Outcome DPP‐4i cohort incident cases (n = 118,419) GLP‐1RA cohort incident cases (n = 118,419) HR (95% CI) Adjusted P value DPP‐4i cohort incident cases (n = 102,810) SGLT2i cohort incident cases (n = 102,810) HR (95% CI) Adjusted P value GLP‐1RA cohort incident cases (n = 105,869) SGLT2i cohort incident cases (n = 105,869) HR (95% CI) Adjusted P value SLE 179 145 1.04 (0.89–1.35) 0.782 139 85 1.24 (0.91–1.62) 0.293 120 100 1.01 (0.78–1.32) 0.968 CLE 57 33 1.53 (0.94–2.34) 0.203 44 19 1.74 (1.01–2.99) 0.148 28 25 0.99 (0.55–1.68) 0.972 Rheumatoid arthritis 966 877 0.92 (0.83–1.07) 0.595 773 512 1.10 (1.00–1.25) 0.160 683 565 1.06 (0.83–1.24) 0.136 Psoriasis 756 851 0.79 (0.70–0.85) 0.006 519 621 0.81 (0.70–0.91) 0.035 777 632 1.02 (0.91–1.15) 0.084 Psoriatic arthritis 148 202 0.65 (0.53–0.79) 0.006 119 92 0.96 (0.73–1.25) 0.869 163 107 1.21 (0.93–1.52) 0.158 Multiple sclerosis 94 87 0.96 (0.72–1.21) 0.784 77 54 1.07 (0.73–1.51) 0.847 89 65 1.15 (0.83–1.60) 0.509 IBD 511 472 0.98 (0.87–1.10) 0.595 439 365 0.88 (0.74–1.04) 0.196 436 404 0.91 (0.79–1.08) 0.517 Crohn disease 178 131 1.22 (0.94–1.52) 0.280 148 97 1.14 (0.87–1.52) 0.508 89 110 0.70 (0.54–0.93) 0.072 ANCA vasculitis 38 26 1.29 (0.72–2.10) 0.595 32 18 1.38 (0.73–2.45) 0.491 23 19 1.05 (0.54–1.89) 0.935 Giant cell arteritis 136 103 1.17 (0.81–1.51) 0.595 114 44 1.83 (1.31–2.65) 0.010 90 55 1.38 (1.00–1.92) 0.180 Dermatomyositis 80 31 2.18 (1.24–3.53) 0.033 78 20 2.03 (1.35 – 3.02) 0.026 27 28 0.85 (0.51–1.43) 0.744 Systemic sclerosis 39 33 1.02 (0.63–1.60) 0.964 35 25 1.02 (0.62–1.69) 0.996 27 31 0.72 (0.46–1.26) 0.504 Celiac disease 83 69 1.05 (0.73–1.41) 0.874 68 44 1.11 (0.76–1.68) 0.772 57 51 0.91 (0.62–1.35) 0.910 Autoimmune thyroiditis 350 450 0.68 (0.59–0.76) 0.006 261 248 0.76 (0.63–0.90) 0.025 354 269 1.15 (0.97–1.34) 0.405 Myasthenia gravis 63 51 1.11 (0.77–1.59) 0.784 52 39 0.98 (0.65–1.48) 0.906 56 43 1.10 (0.73–1.64) 0.757 Pemphigus 20 10 1.94 (0.88–4.24) 0.269 22 18 0.92 (0.47–1.70) 0.891 13 14 0.77 (0.39–1.65) 0.759 Bullous pemphigoid 110 46 1.78 (1.24–2.46) 0.041 94 34 2.07 (1.34–3.05) 0.005 44 29 1.35 (0.84–2.11) 0.481 Polymyositis 18 12 1.36 (0.62–2.75) 0.637 18 12 1.14 (0.57–2.45) 0.852 10 13 0.63 (0.28–1.44) 0.542 Polymyalgia rheumatica 196 163 1.01 (0.81–1.26) 0.834 165 114 1.28 (0.94–1.56) 0.157 145 135 1.09 (0.90 – 1.20) 0.962 Negative controls Hand laceration 384 325 1.03 (0.90–1.16) 0.963 365 302 0.88 (0.74–1.03) 0.206 39 51 1.06 (0.67–1.58) 1.00 Inguinal hernia 328 283 0.97 (0.83–1.14) 0.967 335 266 0.93 (0.78–1.09) 0.389 223 201 0.92 (0.82–1.13) 1.00 Lipoma 83 93 0.78 (0.58–1.05) 0.198 76 51 1.01 (0.72–1.48) 0.933 18 13 1.73 (0.89–3.26) 0.679 * ANCA, antineutrophilic cytoplasmic antibody; CI, confidence interval; CLE, cutaneous lupus erythematosus; DPP‐4i, dipeptidyl peptidase‐4 inhibitor; GLP‐1RA, glucagon‐like peptide‐1 receptor agonist; HR, hazard ratio; IBD, inflammatory bowel disease; SGLT2i, sodium‐glucose cotransporter‐2 inhibitor; SLE, systemic lupus erythematosus.

Occurrence and risk of autoimmune disease in patients with diabetes treated with DPP‐4i versus GLP‐1RA, DPP‐4i versus SGLT2i, and GLP‐1RA versus SGLT2i *

ANCA, antineutrophilic cytoplasmic antibody; CI, confidence interval; CLE, cutaneous lupus erythematosus; DPP‐4i, dipeptidyl peptidase‐4 inhibitor; GLP‐1RA, glucagon‐like peptide‐1 receptor agonist; HR, hazard ratio; IBD, inflammatory bowel disease; SGLT2i, sodium‐glucose cotransporter‐2 inhibitor; SLE, systemic lupus erythematosus.

(Left) Risk of autoimmune disease in patients with diabetes treated with DPP‐4i versus GLP‐1RA. Forest plot showing hazard ratios and 95% CIs for incident autoimmune diseases and negative control outcomes among patients with diabetes treated with DPP‐4i compared with GLP‐1RA. Values <1 favor DPP‐4i and values >1 favor GLP‐1RA. (Right) Risk of autoimmune disease in patients with diabetes treated with DPP‐4i versus SGLT2i. Forest plot showing hazard ratios and 95% CIs for incident autoimmune diseases and negative control outcomes among patients with diabetes treated with DPP‐4i compared with SGLT2i. Values <1 favor DPP‐4i and values >1 favor SGLT2i. (Bottom) Risk of autoimmune disease in patients with diabetes treated with GLP‐1RA versus SGLT2i. Forest plot showing hazard ratios and 95% CIs for incident autoimmune diseases and negative control outcomes among patients with diabetes treated with GLP‐1RA compared with SGLT2i. Values <1 favor GLP‐1RA and values >1 favor SGLT2i. ANCA, antineutrophilic cytoplasmic antibody; CI, confidence interval; DPP‐4i, dipeptidyl peptidase‐4 inhibitor; GLP‐1RA, glucagon‐like peptide‐1 receptor agonist; SGLT2i, sodium‐glucose cotransporter 2 inhibitor.

DPP ‐4i versus SGLT2i trial results

In the DPP‐4i versus SGLT2i trial, DPP‐4 inhibitor use was associated with a decreased risk of psoriasis (HR 0.81, 95% CI 0.70–0.91) and autoimmune thyroiditis (HR 0.76, 95% CI 0.63–0.90), but increased risks of dermatomyositis (HR 2.03, 95% CI 1.35–3.02), bullous pemphigoid (HR 2.07, 95% CI 1.34–3.05), and giant cell arteritis (HR 1.83, 95% CI 1.31–2.65) (Table  2 , Figure 2). No risk differences were observed for any of the negative control outcomes.

GLP ‐ 1RA versus SGLT2i trial results

The GLP‐1RA versus SGLT2i trial showed no statistically significant differences in autoimmune disease risk between treatment groups (Table  2 , Figure 3). Crohn disease (HR 0.70, 95% CI 0.54–0.93) and giant cell arteritis (HR 1.38, 95% CI 1.00–1.92) demonstrated point estimates suggestive of associations, but did not reach statistical significance after adjustment for multiple comparisons ( P > 0.05).

DISCUSSION

In this series of large, emulated trials, we found that DPP‐4i treatment compared to GLP‐1RA and SGLT2i was associated with an increased risk of bullous pemphigoid and dermatomyositis, and a decreased risk of psoriasis and autoimmune thyroiditis across multiple comparisons. These results represent distinct safety signals across antidiabetic medication classes and may suggest novel metabolic pathways involved in the pathogenesis of specific autoimmune diseases.

The observed increase in bullous pemphigoid and reduction in psoriasis risk are consistent with emerging pharmacovigilance reports.
30
,
31
,
32
Although this latter, partly counterintuitive, association may in part reflect residual confounding by obesity‐related factors or diabetes severity—factors imperfectly captured in claims data—the potential for bias is tempered by our positive control analyses, which reproduced the expected GLP‐1RA‐associated weight loss and glycemic improvements in our cohorts. Missing data for continuous variables, such as BMI, may contribute to residual confounding related to baseline obesity and could influence the observed associations. The study's findings do not establish mechanistic certainty and should be regarded as preliminary signals that may help frame the scope and direction of more targeted mechanistic inquiry to identify potential unifying biologic principles underlying the observed associations.

The observed differential autoimmune thyroiditis risk is consistent with mechanistic research demonstrating that DPP‐4 inhibition reduces lymphocyte‐mediated inflammatory factor production in thyroid tissue.
33
The observed increased risk of dermatomyositis with DPP‐4i use represents a potentially novel safety signal, because this association has not been previously reported in large‐scale epidemiologic studies and warrants further investigation given the consistency across multiple comparator groups.
6
,
8
,
34
,
35
Similarly, the elevated risk of giant cell arteritis observed with DPP‐4i treatment, and the directionally similar, though nonsignificant, trend with GLP‐1RA treatment suggest the possibility of class‐specific immunologic effects on these outcomes; however, confirmation in independent datasets and prospective studies is needed to further clarify associations and any causal relationship.

The absence of significant autoimmune disease risk differences between GLP‐1RA and SGLT2i suggests that any immunomodulatory effects of these antidiabetic agents may be more subtle than those observed with DPP‐4i, potentially reflecting distinct mechanisms of action on immune system regulation.
36
,
37
,
38
In addition, a lack of statistically significant findings for other autoimmune diseases across comparisons may reflect insufficient follow‐up time or limited statistical power, given the relative rarity of diseases and disease subtypes.
39

Study limitations include observational study‐related selection biases, outcome and exposure misclassification risk from reliance on structured clinical data, confounding by indication, lack of socioeconomic and insurance status data, and incomplete capture of risk factors such as smoking due to inherent limitations of administrative data.
40
Additionally, unmeasured confounders, including diabetes severity and duration, may introduce bias, though controlling for laboratory data, diabetic complications, and medication use as proxies may partially address these limitations. Despite extensive matching and adjustment, residual confounding by indication and sociodemographic factors may still influence the associations observed, including differential access to care that may correlate with prescribing patterns across drug classes. Given the lack of true untreated comparators, the findings reflect relative differences between antidiabetic classes rather than absolute effects on autoimmune disease risk; future work incorporating mechanistic endpoints and external untreated comparators is essential to further elucidate the degree to which relative differences may translate to clinically relevant impacts.

Study strengths include its robust covariate matching, landmark analysis to mitigate immortal time bias and reverse causation, positive control analysis, and evaluation of multiple negative controls.

Leveraging real‐world data from a multi‐institutional cohort, these emulated target trials provide comparative insights into differential autoimmune safety profiles across these antidiabetic medications used by millions of patients
1
,
2
—findings that may provide the foundation for future mechanistic and prospective research.

AUTHOR CONTRIBUTIONS

All authors contributed to at least one of the following manuscript preparation roles: conceptualization AND/OR methodology, software, investigation, formal analysis, data curation, visualization, and validation AND drafting or reviewing/editing the final draft. As corresponding author, Dr Mahajan confirms that all authors have provided the final approval of the version to be published and takes responsibility for the affirmations regarding article submission (eg, not under consideration by another journal), the integrity of the data presented, and the statements regarding compliance with institutional review board/Declaration of Helsinki requirements.

Supporting information

Data S1 Supporting Information

Disclosure form .