Comparative Risk of Psoriatic Arthritis in Type 2 Diabetes: An Emulated Target Trial of SGLT2 Inhibitors vs. GLP-1 Receptor Agonists

Abstract

Background

Patients with psoriatic arthritis (PsA) have a higher prevalence of type 2 diabetes mellitus (T2DM) and cardiovascular events. They also experience increased absenteeism and reduced work productivity, all of which can negatively affect both life expectancy and quality of life.

Objective

We conducted this emulated target trial to compare the risk of incident PsA among patients with T2DM treated with sodium-glucose cotransporter-2 inhibitors (SGLT2i) versus those treated with glucagon-like peptide-1 receptor agonists (GLP-1 RA).

Methods

We identified 188,378 users of SGLT2 inhibitors and 213,218 users of GLP-1 receptor agonists within the TriNetX network between January 1, 2016, and December 31, 2024. Following propensity score matching, 146,810 matched pairs of SGLT2i and GLP-1 RA users were included for analysis. The primary causal estimands were the intention-to-treat (ITT) effects of the respective treatment strategies. Kaplan–Meier analysis was employed to estimate outcome probabilities, and hazard ratios (HRs) with corresponding confidence intervals (CIs) were calculated, along with tests for proportionality.

Results

In this emulated target trial, users of SGLT2i showed a significantly lower risk of developing PsA compared to users of GLP-1 RA. At 5 years of follow-up, the hazard ratio for PsA was 0.793 (95% CI, 0.667–0.944). The proportional hazards assumption was tested and met (p > 0.05). Kaplan–Meier curves further showed a significantly lower cumulative incidence of PsA among SGLT2i users compared to GLP-1 RA users (Log rank test p = 0.008). These associations remained consistent after adjusting for multiple covariates and were further supported by sensitivity analyses using a per-protocol approach.

Conclusion

This multicenter emulated target trial found that SGLT2i use was associated with a lower risk of incident PsA compared with GLP-1 RA use. However, given the observational nature of the study and the potential for residual confounding despite extensive adjustment, these findings should be interpreted with caution. Further prospective and randomized studies are warranted to confirm this association.

Introduction

Psoriatic arthritis (PsA) is a chronic, immune-mediated inflammatory arthropathy that affects both the joints and entheses, including those in the axial skeleton.1 It affects approximately 0.3–1% of the global population,2 with a prevalence in the United States ranging from 6 to 25 cases per 10,000 people.3 It is estimated that 11–40% of individuals with psoriasis also have PsA.4 For most patients, the onset of arthritis occurs about 10 years after the first signs of psoriasis, though in 15% of cases, arthritis appears before the skin symptoms.3 The annual incidence of PsA in individuals with psoriasis has been reported to range from 2% to 3%, according to a prospective study.5 PsA is notably associated with higher rates of absenteeism and decreased work productivity.3,6 Furthermore, patients with PsA have an increased prevalence of obesity, hypertension, hyperlipidemia, type 2 diabetes mellitus (T2DM), and cardiovascular events compared to the general population, which can impact both life expectancy and quality of life (QoL).3,7 Thus, in clinical practice, preventing the onset of PsA may help reduce the overall disease burden and enhance patients’ quality of life.

Sodium–glucose cotransporter-2 inhibitors (SGLT2i), glucagon-like peptide-1 receptor agonists (GLP-1 RA), and dipeptidyl peptidase-4 inhibitors (DPP-4i) represent newer classes of antidiabetic agents.8–11 These therapies effectively lower blood glucose levels while carrying a relatively low risk of hypoglycemia. In addition, some agents contribute to weight reduction, improve insulin sensitivity, and may attenuate systemic inflammation. Notably, GLP-1 RA are generally associated with greater weight loss, whereas SGLT2i tend to result in more modest reductions in body weight.8–11 Observational studies and meta-analyses have suggested that DPP-4i use may be associated with a reduced risk of autoimmune conditions such as rheumatoid arthritis (RA).12 However, findings across studies remain inconsistent, and the extent to which these immunomodulatory effects translate into clinically meaningful outcomes is still under investigation.13 Given their relatively neutral effects on body weight and cardiovascular outcomes compared with SGLT2i and GLP-1 RA, DPP-4i are often used as a comparator in pharmacoepidemiologic studies evaluating the broader systemic effects of antidiabetic therapies. Preclinical evidence suggests that GLP-1 RA have anti-inflammatory properties, protect joint cartilage by preventing chondrocyte apoptosis, and provide protection in both osteoarthritis and RA models.14 Clinical studies also indicated that GLP-1 RA therapy could improve psoriasis in individuals with T2DM.15,16 In a laboratory study, canagliflozin reduced T cell activation markers, such as IL-2, in CD4+ T cells from patients with systemic lupus erythematosus (SLE) and RA. Additionally, in synovial fluid mononuclear cells from RA patients, canagliflozin decreased IL-17 expression and slightly lowered the expression of other cytokines.17 To date, there have been no clinical studies directly comparing the impact of these antidiabetic drugs on the risk of PsA. Therefore, we conducted this emulated target trial18 to evaluate the risk of developing incident PsA in T2DM patients treated with SGLT2i, GLP-1 RA, and DPP-4i.

Patients and Methods

Study Design and Data Source

This study was performed as a retrospective cohort analysis using data from TriNetX, the largest and most comprehensive real-world data and evidence ecosystem in the life sciences and healthcare sectors. TriNetX includes de-identified electronic health records from over 250 million individuals across more than 120 global healthcare organizations (HCOs). The platform employs a standardized framework to ensure high data quality, focusing on three key quality metrics: conformance, completeness, and plausibility.19

The data analysis was conducted in May 2025 using the US Collaborative Network, a subset of the TriNetX platform, which includes 68 healthcare organizations. To align with the study’s goals, the analysis was limited to data collected between January 1, 2016, and December 31, 2024.

Ethics Statement

The TriNetX platform complies with the Health Insurance Portability and Accountability Act (HIPAA) and the General Data Protection Regulation (GDPR). The Western Institutional Review Board (WIRB) has granted a waiver to TriNetX, as it only aggregates de-identified data in the form of counts and statistical summaries. Additionally, the use of TriNetX for this study was approved by the Institutional Review Board of Chung Shan Medical University Hospital (CSMUH No: CS1-25171).

The Target Trial

We utilized the framework outlined by Hernán and Robins (2016)18 to emulate clinical trials comparing SGLT2i treatment with GLP-1 RA treatment in patients with T2DM using an unblinded design. First, we established the target trial emulation framework to align with our research goals, incorporating key elements such as eligibility criteria, treatment strategies, allocation methods, follow-up procedures, outcomes of interest, causal comparisons, and the analysis plan. Then, we defined the approach for using observational data to replicate these protocol components and perform the necessary analyses (Supplementary Table 1).

Study Eligibility Criteria

Individuals eligible for this study were those diagnosed with T2DM at least twice during the study period, as indicated by the International Statistical Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes E08–E13, excluding E10.20 We excluded individuals under the age of 20 and those with a prior diagnosis of type 1 diabetes mellitus (Supplementary Table 2). To ensure the inclusion of newly diagnosed T2DM cases, we also excluded individuals with a T2DM diagnosis before December 31, 2015. Patients who had undergone dialysis or kidney transplantation were excluded. Furthermore, individuals were excluded if they had been diagnosed with psoriasis, psoriatic arthritis or had died on or before the index date. We excluded prevalent psoriasis to reduce misclassification, as PsA is often miscoded in the absence of psoriasis.

Assigned Treatment Strategies and Groups

We categorized the subjects into two groups based on their treatment regimen. The SGLT2i cohort included individuals who were prescribed SGLT2i, identified by the Anatomical Therapeutic Chemical (ATC) code A10BK, after their first T2DM diagnosis. The index date for this cohort was defined as the date of the first SGLT2i prescription. Patients who received fewer than two prescriptions of SGLT2i or who were prescribed it before their initial T2DM diagnosis were excluded. The GLP-1 RA cohort comprised individuals prescribed GLP-1 RA, identified by the ATC code A10BJ, after their initial T2DM diagnosis. The index date for the GLP-1 RA cohort was set as the date of the first GLP-1 RA prescription. Patients were excluded if they received fewer than two prescriptions of GLP-1 RAs or if the prescription was given prior to their first T2DM diagnosis. No explicit grace period between prescriptions was applied. The requirement for at least two prescriptions was used to identify active users at cohort entry; however, this approach may introduce immortal time bias. To mitigate this potential bias, cohort entry was aligned with treatment initiation, and follow-up commenced at the index date.

Study Outcomes and Follow-Up

The primary outcome of interest was the incidence of PsA, identified by ICD-10-CM code L40.5.21 Secondary outcomes included psoriasis vulgaris, defined by code L40.0, and overall psoriasis, which encompasses all conditions classified under code L40. This algorithm for identifying psoriasis and PsA has demonstrated positive predictive values of approximately 81–100% for psoriasis and 63–92% for PsA in a validation study.21 To evaluate the risk of these outcomes, each participant was followed from the start of their assigned treatment for a period of up to five years.

Emulation of Target Trials

To reduce the impact of confounding factors and simulate randomization, we utilized TriNetX’s built-in propensity score generation feature and performed 1:1 matching using greedy nearest neighbor matching. A caliper of 0.1 pooled standard deviations between the two groups was applied during the matching process. To address missing data, continuous covariates were transformed into categorical variables with a distinct “Missing” category during propensity score matching, effectively balancing data-recording patterns and maintaining sample power. Furthermore, categorical variables derived from ICD-10 codes were identified based on their presence in electronic health records; as per standard convention in real-world evidence studies, the absence of a code was interpreted as the absence of the condition. We evaluated the comparability of the two groups both before and after matching using standardized mean differences (SMD), with an SMD below 0.1 indicating good balance between the cohorts. To address potential confounding, variables from the year prior to the index date were included in the analysis. These variables included: (1) Demographic and socioeconomic factors: age at the index date, sex, race, and socioeconomic status. (2) Lifestyle factors: smoking status (including tobacco use, nicotine dependence, personal history of nicotine dependence, and alcohol-related disorders). (3) Diabetes severity indicators: complications related to T2DM such as hyperosmolarity, ketoacidosis, kidney complications, ophthalmic complications, neurological complications, circulatory complications, other specified complications, unspecified complications, and T2DM without complications. (4) Medical utilization: use of office or outpatient services, emergency services, inpatient services, and preventive care services. Additionally, treatments like phototherapy (ultraviolet light,) and photochemotherapy were included. (5) Comorbidities: defined by ICD-10 codes and categorized as present or absent. Comorbidities included hypertensive diseases, ischemic heart disease, heart failure, cerebrovascular diseases, diseases of arteries, arterioles, and capillaries, disorders of lipoprotein metabolism and other dyslipidemias, obesity, chronic lower respiratory diseases, vitamin D deficiency, neoplasms, chronic kidney disease, depressive episodes, major depressive disorder, recurrent, noninfective enteritis and colitis, liver diseases, infections of the skin and subcutaneous tissue, viral hepatitis, rheumatoid arthritis with rheumatoid factor, other rheumatoid arthritis, gout, osteoarthritis, systemic lupus erythematosus, and ankylosing spondylitis. (6) Medication use: medications such as metformin, sulfonylureas, thiazolidinediones, DPP-4i, insulin and analogues, angiotensin-converting enzyme (ACE) inhibitors, angiotensin II receptor blockers (ARBs, plain), ARB combinations, alpha-adrenoreceptor antagonists, beta-blockers, calcium channel blockers, diuretics, HMG-CoA reductase inhibitors, non-steroidal anti-inflammatory drugs (NSAIDs), aspirin, corticosteroids for systemic use, and proton pump inhibitors. (7) Laboratory results: key baseline characteristics, including Hemoglobin A1C (HbA1c), glomerular filtration rate (eGFR) estimated by the creatinine-based formula (Modification of Diet in Renal Disease, MDRD), and body mass index (BMI). Participants with incomplete data were retained in the analysis, and no imputation methods were applied. Instead, the propensity score matching accounted for patterns of missingness to ensure balance between the comparison groups.

Analysis Plan and Statistical Methods

The primary causal estimands in this study were the intention-to-treat (ITT) effects of the assigned treatment strategies. Kaplan-Meier analysis was used to estimate the probability of the outcomes. Hazard ratios (HRs) and their corresponding confidence intervals (CIs) were estimated using the Survival package in R (version 3.2–3) within the TriNetX platform. Robust standard errors were applied to account for within-pair dependence in the 1:1 propensity score–matched cohorts. The proportional hazards assumption was assessed using Schoenfeld residuals, with no evidence of violation detected (p > 0.05). The Log Rank test was employed to assess differences in survival curves between the cohorts, and these analyses were conducted within the TriNetX platform.

We performed six subgroup analyses to explore variations among cohorts based on age (20–64 years, ≥65 years), sex (male, female), race (White, Black or African American, Asian), HbA1c levels (< 7%, ≥ 9%), obesity status (obese vs. non-obese), and eGFR levels (< 60, 60–90, >90 mL/min/1.73 m²). Additionally, several sensitivity analyses were conducted to test the robustness of the results. First, we modified the study design from an intention-to-treat to a per-protocol approach, excluding patients who switched treatments (ie, replaced the index drug). Second, we applied the same design but compared the cohort to DPP-4i users. Lastly, we compared GLP-1 RA users to DPP-4i users using the same study design. Third, we conducted a negative control outcome analysis using appendicitis risk to assess the presence of potential unmeasured confounding.

Results

Characteristics of Study Subjects

This study included a total of 188,378 SGLT2i users and 213,218 GLP-1 RA users who met the inclusion and exclusion criteria. After implementing propensity score matching, a total of 146,810 patients were classified as users of SGLT2i, with an equal number in the GLP-1 RA user group. The selection process is illustrated in Figure 1.

Figure 1.

The selection process between SGLT2i and GLP-1 RA users.

The study subjects’ baseline characteristics were displayed in Table 1, both before and after the matching process. There are significant disparities between SGLT2i and GLP-1RA users in the distribution of age, sex, race, personal history of nicotine dependence, type 2 diabetes mellitus with kidney complications, medical utilization, comorbidities, comedications, and laboratory results. After propensity score matching (PSM), the baseline characteristics differences between SGLT2i users and the GLP-1RA users were minimized and fell within an acceptable range (SMD < 0.1). Specifically, this approach achieved comparable missingness proportions across most variables (SMD < 0.1). However, residual imbalance remained for BMI missingness (SMD = 0.167), although this represented substantial improvement compared with the unmatched cohort.

Table 1.

Baseline Characteristics of Study Subjects (Before and After Matching)

Variables	Before PSMa			After PSMa
	SGLT2i Users (n=188378)	GLP-1 RA Users (n=213218)	SMD	SGLT2i Users (n=146810)	GLP-1 RA Users (n=146810)	SMD
Current age, mean ± SD	64.1 ± 12.1	59.4 ± 12.5	0.380	62.2 ± 11.9	62.1 ± 11.8	0.004
Age at index, mean ± SD	60.9 ± 12.4	56.2 ± 12.4	0.379	58.9 ± 12.1	58.8 ± 11.7	0.006
Sex, n (%)
Male	104910 (55.7)	90,039 (42.2)	0.272	74,567 (50.8)	74,401 (50.7)	0.002
Female	76139 (40.4)	113,078 (53.0)	0.255	66,063 (45.0)	66,241 (45.1)	0.002
Unknown Gender	7329 (3.9)	10,101 (4.7)	0.042	6180 (4.2)	6168 (4.2)	0.000
Race, n (%)
White	108906 (57.8)	125,930 (59.1)	0.025	86,544 (59.0)	86,890 (59.2)	0.005
Black or African American	36133 (19.2)	44,876 (21.0)	0.047	28,400 (19.3)	28,212 (19.2)	0.003
Unknown Race	18801 (10.0)	21,327 (10.0)	0.001	14,854 (10.1)	14,947 (10.2)	0.002
Asian	12960 (6.9)	9038 (4.2)	0.115	8131 (5.5)	7851 (5.3)	0.008
Other Race	9458 (5.0)	9713 (4.6)	0.022	7283 (5.0)	7291 (5.0)	0.000
Native Hawaiian or Other Pacific Islander	1524 (0.8)	1558 (0.7)	0.009	1103 (0.8)	1124 (0.8)	0.002
American Indian or Alaska Native	596 (0.3)	776 (0.4)	0.008	495 (0.3)	495 (0.3)	0.000
Social economic status, n (%)
Problems related to housing and economic circumstances	2126 (1.1)	1272 (0.6)	0.058	956 (0.7)	948 (0.6)	0.001
Problems related to employment and unemployment	444 (0.2)	421 (0.2)	0.008	264 (0.2)	279 (0.2)	0.002
Problems related to education and literacy	246 (0.1)	138 (0.1)	0.021	92 (0.1)	107 (0.1)	0.004
Lifestyles, n (%)
Personal history of nicotine dependence	21454 (11.4)	16,222 (7.6)	0.129	12,148 (8.3)	11,907 (8.1)	0.006
Nicotine dependence	17285 (9.2)	14,832 (7.0)	0.082	10,728 (7.3)	10,749 (7.3)	0.001
Tobacco use	6394 (3.4)	5625 (2.6)	0.044	4045 (2.8)	3957 (2.7)	0.004
Alcohol related disorders	3791 (2.0)	2391 (1.1)	0.072	1939 (1.3)	1899 (1.3)	0.002
DM severity, n (%)
Type 2 diabetes mellitus without complications	109435 (58.1)	123,497 (57.9)	0.004	82,793 (56.4)	82,509 (56.2)	0.004
Type 2 diabetes mellitus with other specified complications	64506 (34.2)	68,467 (32.1)	0.045	47,991 (32.7)	48,210 (32.8)	0.003
Type 2 diabetes mellitus with kidney complications	25947 (13.8)	16,964 (8.0)	0.188	14,453 (9.8)	14,284 (9.7)	0.004
Type 2 diabetes mellitus with neurological complications	19540 (10.4)	17,958 (8.4)	0.067	13,639 (9.3)	13,577 (9.2)	0.001
Type 2 diabetes mellitus with circulatory complications	12118 (6.4)	8941 (4.2)	0.100	7051 (4.8)	6945 (4.7)	0.003
Type 2 diabetes mellitus with unspecified complications	7648 (4.1)	8946 (4.2)	0.007	6164 (4.2)	6151 (4.2)	0.000
Type 2 diabetes mellitus with ophthalmic complications	6798 (3.6)	6153 (2.9)	0.041	4945 (3.4)	4817 (3.3)	0.005
Type 2 diabetes mellitus with hyperosmolarity	1779 (0.9)	1920 (0.9)	0.005	1318 (0.9)	1324 (0.9)	0.000
Type 2 diabetes mellitus with ketoacidosis	1256 (0.7)	1475 (0.7)	0.003	912 (0.6)	915 (0.6)	0.000
Medical utilization, n (%)
Office or Other Outpatient Services	87764 (46.6)	111,305 (52.2)	0.112	69,923 (47.6)	69,221 (47.2)	0.010
Emergency Department Services	41897 (22.2)	33,431 (15.7)	0.168	24,582 (16.7)	24,321 (16.6)	0.005
Preventive Medicine Services	15420 (8.2)	27,486 (12.9)	0.154	13,900 (9.5)	13,482 (9.2)	0.010
Hospital Inpatient and Observation Care Services	28215 (15.0)	13,822 (6.5)	0.277	12,082 (8.2)	11,932 (8.1)	0.004
Phototherapy (ultraviolet light)	10 (0.0)	12 (0.0)	0.000	10 (0.0)	10 (0.0)	0.000
Photochemotherapy	10 (0.0)	10 (0.0)	0.001	10 (0.0)	10 (0.0)	0.000
Comorbidities, n (%)
Hypertensive diseases	120893 (64.2)	123,073 (57.7)	0.133	86,979 (59.2)	86,576 (59.0)	0.006
Disorders of lipoprotein metabolism and other lipidemias	110984 (58.9)	114,418 (53.7)	0.106	81,081 (55.2)	80,765 (55.0)	0.004
Overweight and obesity	51112 (27.1)	73,726 (34.6)	0.162	40,419 (27.5)	40,109 (27.3)	0.005
Ischemic heart diseases	46100 (24.5)	24,260 (11.4)	0.346	22,297 (15.2)	22,007 (15.0)	0.006
Chronic lower respiratory diseases	27750 (14.7)	28,388 (13.3)	0.041	18,124 (12.3)	17,958 (12.2)	0.003
Vitamin D deficiency	21292 (11.3)	29,981 (14.1)	0.083	17,564 (12.0)	17,353 (11.8)	0.004
Neoplasms	23701 (12.6)	26,350 (12.4)	0.007	17,477 (11.9)	17,204 (11.7)	0.006
Osteoarthritis	21788 (11.6)	25,796 (12.1)	0.016	16,296 (11.1)	16,156 (11.0)	0.003
Chronic kidney disease	27604 (14.7)	16,426 (7.7)	0.222	14,469 (9.9)	14,265 (9.7)	0.005
Heart failure	36100 (19.2)	12,130 (5.7)	0.417	12,158 (8.3)	11,974 (8.2)	0.005
Depressive episode	15620 (8.3)	20,281 (9.5)	0.043	11,647 (7.9)	11,542 (7.9)	0.003
Diseases of liver	13656 (7.2)	15,936 (7.5)	0.009	9887 (6.7)	9822 (6.7)	0.002
Diseases of arteries, arterioles and capillaries	17939 (9.5)	11,276 (5.3)	0.162	9484 (6.5)	9329 (6.4)	0.004
Infections of the skin and subcutaneous tissue	10023 (5.3)	10,214 (4.8)	0.024	6880 (4.7)	6874 (4.7)	0.000
Cerebrovascular diseases	12479 (6.6)	7566 (3.5)	0.140	6595 (4.5)	6460 (4.4)	0.004
Major depressive disorder, recurrent	5469 (2.9)	8841 (4.1)	0.067	4588 (3.1)	4577 (3.1)	0.000
Gout	6769 (3.6)	5100 (2.4)	0.071	3980 (2.7)	3950 (2.7)	0.001
Noninfective enteritis and colitis	4502 (2.4)	4539 (2.1)	0.018	3134 (2.1)	3104 (2.1)	0.001
Other rheumatoid arthritis	1952 (1.0)	2216 (1.0)	0.000	1371 (0.9)	1393 (0.9)	0.002
Viral hepatitis	1706 (0.9)	1287 (0.6)	0.035	1001 (0.7)	991 (0.7)	0.001
Rheumatoid arthritis with rheumatoid factor	645 (0.3)	805 (0.4)	0.006	489 (0.3)	490 (0.3)	0.000
Systemic lupus erythematosus	562 (0.3)	702 (0.3)	0.006	409 (0.3)	433 (0.3)	0.003
Ankylosing spondylitis	159 (0.1)	210 (0.1)	0.005	123 (0.1)	115 (0.1)	0.002
Concomitant medications, n (%)
HMG CoA reductase inhibitors	89757 (47.6)	86,620 (40.6)	0.142	63,727 (43.4)	61,987 (42.2)	0.024
Metformin	73927 (39.2)	94,563 (44.4)	0.104	60,924 (41.5)	61,509 (41.9)	0.008
Diuretics	60938 (32.3)	54,033 (25.3)	0.155	38,406 (26.2)	37,924 (25.8)	0.007
Beta blocking agents	60789 (32.3)	45,834 (21.5)	0.245	36,849 (25.1)	34,737 (23.7)	0.034
Insulins and analogues	55610 (29.5)	47,187 (22.1)	0.169	35,515 (24.2)	36,107 (24.6)	0.009
Corticosteroids for systemic use	48012 (25.5)	55,291 (25.9)	0.010	33,996 (23.2)	35,583 (24.2)	0.025
ACE inhibitors, plain	41943 (22.3)	43,541 (20.4)	0.045	31,516 (21.5)	30,647 (20.9)	0.014
Angiotensin II receptor blockers (ARBs), plain	43189 (22.9)	40,159 (18.8)	0.101	29,153 (19.9)	28,488 (19.4)	0.011
Calcium channel blockers	41759 (22.2)	36,538 (17.1)	0.127	28,051 (19.1)	26,557 (18.1)	0.026
NSAIDs	34845 (18.5)	46,084 (21.6)	0.078	27,344 (18.6)	28,574 (19.5)	0.021
Proton pump inhibitors	38896 (20.6)	37,802 (17.7)	0.074	26,081 (17.8)	25,795 (17.6)	0.005
Aspirin	40313 (21.4)	24,928 (11.7)	0.264	22,154 (15.1)	20,383 (13.9)	0.034
Sulfonylureas	27061 (14.4)	29,683 (13.9)	0.013	21,452 (14.6)	21,838 (14.9)	0.007
Dipeptidyl peptidase 4 inhibitors	17036 (9.0)	19,346 (9.1)	0.001	13,578 (9.2)	14,362 (9.8)	0.018
Thiazolidinediones	4728 (2.5)	5364 (2.5)	0.000	3822 (2.6)	4043 (2.8)	0.009
Angiotensin II receptor blockers (ARBs), combinations	6878 (3.7)	1909 (0.9)	0.186	2562 (1.7)	1841 (1.3)	0.040
Alpha-adrenoreceptor antagonists	1721 (0.9)	1571 (0.7)	0.020	1114 (0.8)	1135 (0.8)	0.002
Laboratory, n (%)
Hemoglobin A1C/hemoglobin, total in blood (%)	114,948 (61.0)	138,725 (65.1)	0.034	89,187 (60.7)	88,681 (60.4)	0.010
< 7	46,374 (24.6)	62,568 (29.3)	0.107	34,989 (23.8)	33,851 (23.1)	0.018
≧9	39,671 (21.1)	47,579 (22.3)	0.030	32,649 (22.2)	33,026 (22.5)	0.006
Estimated Glomerular filtration rate (eGFRb, mL/min/1.73m2)	128,907 (68.4)	141,161 (66.2)	0.217	95,976 (65.4)	93,435 (63.6)	0.027
< 60	48,306 (25.6)	35,758 (16.8)	0.218	28,767 (19.6)	28,413 (19.4)	0.006
60~90	77,326 (41.0)	82,275 (38.6)	0.050	56,339 (38.4)	55,920 (38.1)	0.006
Body mass index (BMI, kg/m2)	115,040 (61.1)	132,248 (62.0)	0.455	88,364 (60.2)	82,992 (56.5)	0.167
≧30	78,074 (41.4)	110,696 (51.9)	0.211	64,515 (43.9)	64,186 (43.7)	0.005

Notes: Bold font represents a standardized difference was more than 0.1. If the patient is less or equal to 10, results show the count as 10. ^aPropensity score matching was performed on age, sex, race, socioeconomic status, DM severity, lifestyles, medical utilization, comorbidities, and HbA1c, eGFR, BMI. ^bPredicted by Creatinine-based formula (modification of diet in renal disease, MDRD).

Abbreviations: PSM, Propensity score matching; SGLT2i, sodium–glucose cotransporter 2 inhibitors; GLP-1 RA, Glucagon-like peptide-1 receptor agonist; SMD, Standardized mean difference; SD, Standard deviation, DM, Diabetes mellitus; HMG CoA, β-Hydroxy β-methylglutaryl-CoA; NSAIDs, Anti-inflammatory and anti-rheumatic products, non-steroids.

Outcomes

Table 2 summarizes the number of patients with outcomes in each cohort and presents the HRs for the risk of PsA across different follow-up durations. The findings consistently demonstrate that, irrespective of whether the follow-up period was 1, 3, or 5 years, SGLT2i users exhibited a lower risk of developing PsA compared to GLP-1 RA users. Over a comparable median follow-up of approximately 813 days for both cohorts, the SGLT2i group showed a lower incidence rate of PsA (0.635 vs 0.801 per 1000 person-years), with an absolute risk difference of −0.042% (95% CI: −0.073% to −0.012%). The corresponding HRs were 0.756 (95% CI: 0.590–0.967), 0.812 (95% CI: 0.674–0.979), and 0.793 (95% CI: 0.667–0.944), respectively. The estimated numbers needed to treat (NNTs) were 3968, 3059, and 2368, respectively, and the corresponding E-values were 1.98, 1.77, and 1.83. These associations remained robust after adjusting for multiple covariates (Supplementary Table 3). Kaplan–Meier curves depicting the cumulative incidence of PsA between the SGLT2i and GLP-1 RA cohorts are shown in Figure 2. The Log rank test indicated a statistically significant difference between the two groups (p = 0.008).

Table 2.

Risk of Psoriatic Arthritis

Follow Up	Patients with Outcome		NNT	Hazard Ratio (95% CI)a	E-value for Point (CI)
	SGLT2i Users (n=146,810)	GLP-1 RA Users (n=146,810)
1 day to 1 year	110	147	3968	0.756 (0.590–0.967)	1.98 (1.22)
1 day to 3 years	200	248	3059	0.812 (0.674–0.979)	1.77 (1.17)
1 day to 5 years	228	290	2368	0.793 (0.667–0.944)	1.83 (1.31)

Notes: The bold numerical values in this table indicate statistical significance. ^aPropensity score matching was performed on age at index, sex, race, socioeconomic status, DM severity, lifestyles, medical utilization, comorbidities, and HbA1C, eGFR, BMI.

Abbreviations: SGLT2i, sodium–glucose cotransporter 2 inhibitors; GLP-1 RA, Glucagon-like peptide-1 receptor agonist; NNT, Number needed to treat; CI, Confidence interval.

Figure 2.

Kaplan-Meier curves of psoriatic arthritis incidence between SGLT2i and GLP-1 RA users.

In contrast, no significant difference was observed in the risk of developing psoriasis vulgaris between SGLT2i and GLP-1 RA users across 1-, 3-, and 5-year follow-up periods. The corresponding HRs were 1.150 (95% CI: 0.903–1.465), 1.097 (95% CI: 0.920–1.307), and 1.033 (95% CI: 0.887–1.203), respectively (Supplementary Table 4).

Similarly, a separate analysis revealed no significant difference in the risk of overall psoriasis between the two groups over the same follow-up intervals, with HRs of 0.963 (95% CI: 0.859–1.080), 0.975 (95% CI: 0.896–1.061), and 0.936 (95% CI: 0.866–1.012), respectively (Supplementary Table 5).

Subgroup Analyses

We conducted additional analyses to evaluate potential variations in PsA incidence across different subgroups (Supplementary Tables 6–11), with a summary of these findings presented in Figure 3. Our results indicate that among elderly individuals (aged 65 years and older), SGLT2i users had a significantly lower risk of PsA compared to GLP-1 RA users (HR: 0.730; 95% CI: 0.558–0.957). Similarly, among male participants, SGLT2i users exhibited a significantly lower risk of PsA (HR: 0.706; 95% CI: 0.538–0.926). In the White population, SGLT2i users showed significantly lower risks of PsA (HR: 0.732; 95% CI: 0.593–0.905), psoriasis vulgaris (HR: 0.742; 95% CI: 0.603–0.913), and overall psoriasis (HR: 0.892; 95% CI: 0.806–0.987), compared to GLP-1 RA users. Stratification by glycemic control showed no significant difference in PsA risk between the two groups, regardless of HbA1c levels in the year prior to the index date. Specifically, HRs were 0.788 for HbA1c < 7% and 1.046 for HbA1c ≥ 9%. Among individuals with comorbid obesity, there was no significant difference in PsA incidence between the two treatment groups. However, among non-obese individuals, SGLT2i users had a significantly lower risk of PsA compared to GLP-1 RA users (HR: 0.694; 95% CI: 0.496–0.970). We also examined whether renal function, as measured by eGFR, modified the association between medication use and PsA risk. Although SGLT2i users generally showed a slightly reduced risk of PsA across different levels of renal function, the differences did not reach statistical significance. The interaction p-values from the subgroup analyses were not statistically significant (Figure 3). However, these analyses should be considered exploratory.

Figure 3.

Forest plot of psoriatic arthritis _ Subgroup analyses between SGLT2i and GLP-1 RA users.

Sensitivity Analyses

Our findings remained generally consistent when the study design was modified from an intention-to-treat to a per-protocol approach. Under this exploratory analysis, SGLT2i users continued to exhibit a lower risk of PsA compared to GLP-1 RA users (HR: 0.608; 95% CI: 0.447–0.827; Supplementary Table 12). However, because treatment switching and discontinuation may introduce informative censoring and adherence-related bias, these per-protocol findings should be interpreted cautiously.

However, when the comparator cohort was changed to DPP-4i users, no significant difference was observed in the risk of PsA between SGLT2i and DPP-4i users (HR: 0.928; 95% CI: 0.721–1.194; Log rank test, p = 0.561; Supplementary Figures 1,2 and Supplementary Table 13).

Similarly, there was no significant difference in PsA risk between GLP-1 RA and DPP-4i users (HR: 0.890; 95% CI: 0.721–1.098; Log rank test, p = 0.276; Supplementary Figures 3, 4 and Supplementary Table 14).

The negative control outcome analysis for appendicitis demonstrated no significant differences between SGLT2i and GLP-1 RA users at 1-, 3-, or 5-year follow-up (Supplementary Table 15). Because this analysis was conducted using a later TriNetX data extraction, the findings should be considered exploratory.

Discussion

This multicenter, emulated target trial showed that, among patients with T2DM, the use of SGLT2i was associated with a significantly lower risk of developing PsA compared to GLP-1 RA. The exact biological pathways responsible for this association are not yet understood, highlighting the need for additional prospective research to confirm and further explore these observations. However, no significant difference in PsA risk was observed between users of SGLT2i and DPP-4i, nor between GLP-1 RA and DPP-4i users.

Several prior studies have reported that the use of DPP-4i is associated with a significantly lower risk of developing RA compared to non-users.12,22 However, a meta-analysis examining the effects of DPP-4i found no significant association with RA incidence when compared to non-users.13 Additionally, a cohort study from the United States indicated that the use of GLP-1 RA was significantly associated with a higher risk of arthritis compared to both non-users and users of DPP-4i.23 To the best of our knowledge, our study is among the first to show that, in patients with T2DM, treatment with SGLT2i is associated with a significantly lower risk of developing PsA compared to GLP-1 RA. This inverse association remained consistent across 1-, 3-, and 5-year follow-up periods, and persisted across various adjustment models. Sensitivity analyses using a per-protocol approach yielded a lower hazard ratio (HR 0.608) compared with the intention-to-treat estimate (HR 0.793). This stronger association may reflect differences related to sustained SGLT2i exposure or patient adherence patterns; however, the possibility of informative censoring, adherence bias, and selection bias cannot be excluded. Interestingly, subgroup analyses suggested that the association between SGLT2i use and a lower risk of PsA, compared with GLP-1 RA use, was observed in patients without obesity but not in those with obesity. Given the well-established link between obesity and PsA,3,7 this finding may reflect differences in underlying risk profiles or treatment effects across subgroups. Although GLP-1 RA are generally associated with greater weight reduction than SGLT2i, which could potentially influence PsA risk, this explanation remains speculative. Other unmeasured factors, such as differences in disease severity, metabolic status, or residual confounding, may also contribute to the observed heterogeneity. Therefore, these subgroup findings should be interpreted with caution and warrant further investigation.

Although our results suggest that SGLT2i use is associated with a lower PsA risk compared to GLP-1 RA, no significant difference was observed when comparing SGLT2i to DPP-4i. Both GLP-1 RA and DPP-4i exert glucose-lowering effects via incretin-based mechanisms, but their roles in immunomodulation and arthritis risk may differ.8,10 For instance, the aforementioned US cohort study reported a higher arthritis risk with GLP-1 RA than with DPP-4i, which may partially explain our findings.23 Moreover, our study did not find a significant difference in PsA risk between GLP-1 RA and DPP-4i users, which contrasts with the findings reported by Yan Xie et al23 This discrepancy may be attributable to differences in outcome definitions: our study focused specifically on PsA, whereas theirs examined arthritis more broadly.

Potential mechanisms by which SGLT2i may reduce the risk of PsA include the following: (1) SGLT2i promote fat utilization and induce the browning of adipose tissue, thereby attenuating obesity-related inflammation by shifting macrophage polarization toward the anti-inflammatory M2 phenotype.3,24 They also increase circulating levels of β-hydroxybutyrate, which inhibits activation of the NLRP3 inflammasome and consequently reduces the release of interleukins IL-1β, IL-2, and IL-17.11,17,25 In addition, SGLT2i activate key regulatory pathways involving AMP-activated protein kinase (AMPK), sirtuin 1 (SIRT1), SIRT3, and peroxisome proliferator-activated receptor-γ coactivator 1α (PGC-1α), leading to enhanced autophagic activity and suppression of inflammatory responses.24 (2) Tendinopathic tissues exhibit upregulation of hypoxia-inducible factor-1 (HIF-1), which promotes inflammation, apoptosis, and angiogenesis. Hypoxia-driven neovascularization mediated by vascular endothelial growth factor (VEGF) has also been implicated in tendinopathy pathogenesis.1,26 Preclinical studies suggest that SGLT2i can downregulate both HIF-1α and VEGF expression, thereby potentially lowering the risk of tendinopathy24,26 (3) SGLT2i may enhance insulin sensitivity in skeletal muscle by lowering hyperinsulinemia, reducing visceral adiposity, and increasing glucose uptake into muscle tissue.11 They have also been shown to reduce inflammatory cytokine levels, limit macrophage infiltration, and downregulate the mRNA expression of atrogin-1, a key marker of muscle atrophy.24 These mechanisms may help prevent the development of enthesitis.4,7 Nonetheless, these anti-inflammatory actions of SGLT2i have not yet been validated in arthritis models. Further mechanistic studies are warranted to clarify their potential role in PsA prevention and treatment.

Regarding the association between antidiabetic medications and psoriasis, a meta-analysis of 10 randomized controlled trials demonstrated that pioglitazone use significantly improved psoriasis severity compared to placebo in patients with psoriasis.27 Several case series have reported that GLP-1 RA may lead to improvements in psoriasis severity compared to non-GLP-1 RA in patients with T2DM and psoriasis.15,16,28 However, a randomized controlled trial involving 9 patients receiving sitagliptin and 11 receiving gliclazide found no statistically significant difference in psoriasis severity between the two groups.29 A nationwide cohort study found that SGLT2i use was not associated with a significant difference in the overall risk of psoriasis compared to non-SGLT2i use in patients with T2DM.30 However, among T2DM patients with pre-existing renal disease, SGLT2i use was associated with a 2.7-fold increased risk of psoriasis. In our study, SGLT2i use was not associated with a significant difference in the risks of overall psoriasis or psoriasis vulgaris4—the most common subtype—when compared to the use of GLP-1 RAs or DPP-4i. Furthermore, no significant differences in the risks of psoriasis or psoriasis vulgaris were observed between GLP-1 RA versus DPP-4i use. The absence of an effect on psoriasis vulgaris suggests that SGLT2i may preferentially modulate joint-specific inflammation (enthesitis, synovitis) rather than skin inflammation, possibly through effects on entheseal resident cells or mechanical stress pathways. This remains speculative and requires mechanistic study.

This study has several notable strengths. First, to the best of our knowledge, it is the first to investigate the relationship between the use of SGLT2i, GLP-1 RA, or DPP-4i and the development of PsA, psoriasis, and psoriasis vulgaris. Second, we utilized propensity score matching to create comparable groups, which was essential for analyzing baseline factors—such as smoking status, body mass index, socioeconomic status, and hemoglobin A1C levels—that are known to influence PsA risk.3,7 Third, we conducted an emulated target trial and used a new-user, active comparator design to mitigate immortal time bias and reduce the impact of confounding by indication and other unmeasured factors. Fourth, the comprehensive patient database provided by TriNetX allowed for the identification of a sufficient cohort of patients with T2DM using SGLT2i, GLP-1 RA, and DPP-4i. The large dataset also facilitated subgroup analyses based on variations in obesity, diabetes severity, and kidney disease. Finally, for T2DM patients with additional risk factors for PsA (eg, family history, personal history of psoriasis, enthesitis symptoms), SGLT2i may be preferred over GLP-1 RA, pending confirmation in randomized trials.

However, this study has several limitations. First, the identification of PsA, psoriasis, and psoriasis vulgaris was based on ICD codes, which may introduce misclassification. To reduce this risk, we excluded patients with prior diagnoses of PsA or psoriasis before the index date, thereby minimizing the inclusion of prevalent cases. In addition, previously validated algorithms for identifying T2DM, psoriasis, and PsA using ICD codes have demonstrated acceptable accuracy.20,21 However, these definitions did not require confirmation by a rheumatologist, the use of disease-modifying antirheumatic drugs (DMARDs), or imaging studies, which may still result in misclassification. Furthermore, although psoriasis was adjusted for as a binary variable, information on disease severity was unavailable, potentially leading to residual confounding. Accordingly, the findings should be interpreted with caution. Second, several lifestyle factors—such as exercise and alcohol consumption, sleep conditions, genetic predisposition, and environmental exposures may influence the risk of PsA. While this introduces the possibility of unmeasured confounding, we attempted to address this using various sensitivity tests to minimize its impact. Third, medication adherence could not be directly evaluated, which is a common limitation of studies based on administrative databases. In addition, the TriNetX platform does not allow assessment of cumulative dose or treatment duration. To address these limitations, we performed a per-protocol analysis and conducted analyses across multiple follow-up periods (1, 3, and 5 years) to provide a more comprehensive interpretation of the findings. Fourth, death represents a competing event for PsA, and conventional Kaplan–Meier methods may overestimate incidence if mortality differs between groups. However, due to analytical limitations of the TriNetX platform, competing risk analyses could not be performed. Finally, despite using an emulated target trial and a new-user, active comparator design, there remains an inherent risk of unmeasured confounding factors, even after adjusting for key variables. Therefore, our findings should be considered as associations rather than direct evidence of causation. Rigorous clinical trials are needed to confirm and validate these results.

Conclusions

This multicenter emulated target trial showed that SGLT2i use was associated with a lower risk of incident PsA compared with GLP-1 RA use. However, given the observational nature of the study, residual confounding and potential biases cannot be excluded, and these findings should not be interpreted as evidence of a preventive effect. Further clinical and mechanistic studies are warranted to clarify the relationship between SGLT2i use and PsA risk.

Funding Statement

This study received funding support from Chung Shan Medical University Hospital (CSH-2025- D −005). The corresponding author had full access to all data in the study and final responsibility for the decision to publish. The funders had no role in study design, data collection, data analysis, decision to publish, or manuscript preparation.

Abbreviations

PsA, psoriatic arthritis; T2DM, type 2 diabetes mellitus; SGLT2i, sodium-glucose cotransporter-2 inhibitors; GLP-1 RA, glucagon-like peptide-1 receptor agonists; DPP-4i, dipeptidyl peptidase-4 inhibitors; HbA1c, hemoglobin A1C; eGFR, estimated glomerular filtration rate.

Data Sharing Statement

Requests should be directed to the corresponding author and will require approval from TriNetX and the relevant institutional review boards. Further information about the TriNetX database can be accessed at https://trinetx.com.

Ethics Approval and Consent to Participate

Not applicable. This study used de-identified electronic health records data from the TriNetX Global Collaborative Network; therefore, no individual consent or institutional review board approval was required.

Author Contributions

FSY: study concept and design, drafting of the manuscript, critical revision of the manuscript for important intellectual content. SIW: data acquisition, analysis, and interpretation, statistical analysis, drafting of the manuscript. CMH: data analysis and interpretation, statistical analysis, critical revision of the manuscript for important intellectual content. CCH: study concept and design, data interpretation, critical revision of the manuscript for important intellectual content. KYC: study concept and design, data interpretation, critical revision of the manuscript for important intellectual content. JCCW: critical revision of the manuscript for important intellectual content, technical and material support, and study supervision. All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors confirm that they have not received any personal fees or honoraria from manufacturers of SGLT2i or GLP-1 RA within the past 36 months. The authors report no conflicts of interest in this work.