Gout-Related Medication Use After Initiating Sodium–Glucose Cotransporter 2 Inhibitors in Patients With Gout and Type 2 Diabetes: Population-Based Target Trial Emulation Studies

Abstract

OBJECTIVE

Sodium–glucose cotransporter 2 inhibitors (SGLT2is) lower serum urate and are associated with a lower risk of recurrent gout flares. We used target trial emulation to compare rates of allopurinol initiation and use of anti-inflammatories (high-dose glucocorticoids, nonsteroidal anti-inflammatory drugs [NSAIDs], colchicine) and diuretics (prototypic serum urate-raising medication) among patients with gout using SGLT2is versus dipeptidyl peptidase 4 inhibitors (DPP-4is) (primary comparator), with glucagon-like peptide 1 receptor agonists (GLP-1RAs) as an alternative comparator.

RESEARCH DESIGN AND METHODS

From a general population database, we identified patients with gout and comorbid type 2 diabetes and used Cox proportional hazards and Poisson regressions with inverse probability of treatment weighting to emulate randomization to SGLT2i or DPP-4i/GLP-1RA. We also replicated the analysis in an electronic health record data set with further adjustment for serum urate and BMI.

RESULTS

Among 26,739 adults with gout and type 2 diabetes (mean age 66 years), 67% had polypharmacy. Allopurinol initiation was lower among SGLT2i initiators than DPP-4i, with a hazard ratio of 0.62 (95% CI 0.52–0.73). Associations were stronger among those using diuretics at baseline (P for interaction = 0.03) and persisted when comparing SGLT2i with GLP-1RA and accounting for serum urate and BMI in the secondary data set. SGLT2i was also associated with lower rates of high-dose glucocorticoid, NSAID, colchicine, and diuretic dispensing, with rate ratios of 0.78 (95% CI 0.74–0.83), 0.85 (95% CI 0.80–0.92), 0.87 (95% CI 0.83–0.92), and 0.87 (95% CI 0.85–0.89), respectively.

CONCLUSIONS

For patients with gout and type 2 diabetes, SGLT2is may reduce gout-related medication use, which could, in turn, reduce exposure to the harmful cardiovascular-kidney-metabolic effects of NSAIDs and glucocorticoids in this high-risk population.

Graphical Abstract

Introduction

Gout is the most common inflammatory arthritis, affecting 5.1% of the US population, including >10% of those aged ≥65 years (1). Characterized by excruciatingly painful episodes of arthritis and joint damage, gout is also associated with the metabolic syndrome (2) and elevated burden of cardiovascular-kidney-metabolic (CKM) comorbidities (3); indeed, one in four U.S. adults with gout also has type 2 diabetes (3). In turn, patients with gout are particularly prone to polypharmacy (use of five or more medications), which has multiple clinical consequences (4), including premature mortality (5). Although urate-lowering therapies (ULTs) are effective for preventing flares, landmark trials have found no CKM benefits with conventional ULT (e.g., allopurinol) (6,7), necessitating different medication classes for gout and comorbidity care. However, this makes treatment decisions more complex, with the resultant pill burden, drug-drug interactions, and adverse effects being key concerns for patients living with gout (8,9). As such, pleiotropic interventions that can simultaneously reduce flare rates and address CKM comorbidities, while limiting exposure to the adverse effects of certain gout medications (e.g., nonsteroidal anti-inflammatory drugs [NSAIDs], glucocorticoids), are highly desirable for patients with gout and multiple comorbidities.

Since their initial approval as a glucose-lowering therapy in type 2 diabetes, sodium–glucose cotransporter 2 inhibitors (SGLT2is) have demonstrated multiple CKM benefits (10), including in heart failure where they have reduced the need for diuretic medications (11,12). Randomized trials have also found that SGLT2is reduce serum urate levels (13), a key causal precursor for gout, which has translated into a reduced risk of gout (14,15) associated with SGLT2i initiation.

ULTs are a mainstay treatment for gout management; however, for certain patients (e.g., with CKM comorbidities but without frequent flares or tophi [16,17]), the benefits of SGLT2is for recurrent gout prevention could reduce the need for ULT, as well as use of gout flare medications. This could benefit patients with gout in several ways, including reducing exposure to the adverse CKM, gastrointestinal, and other systemic adverse effects of NSAIDs and glucocorticoids (18,19), as well as reducing polypharmacy and its downstream effects on health care costs (20), medication adherence (4), and clinical outcomes (4,5). However, the impact of SGLT2i initiation on use of ULTs, gout flare medications, and diuretics (major risk factor for recurrent flares) among patients with gout has not been evaluated.

We hypothesized that the pleotropic effects of SGLT2is can reduce reliance on ULT and use of diuretics while also reducing acute gout flares and associated need for acute gout flare anti-inflammatory drugs (e.g., NSAIDs, high-dose glucocorticoids). We used target trial emulation to compare rates of initiation of allopurinol, as well as gout flare and diuretic medication use, and gout flare rates among patients with gout and type 2 diabetes initiating an SGLT2i versus a dipeptidyl peptidase 4 inhibitor (DPP-4i) (primary comparator), with a glucagon-like peptide 1 receptor agonist (GLP-1RA) as an alternative comparator.

Research Design and Methods

We followed the framework of Hernán and Robins (21) to emulate two-arm clinical trials (the target trials) that would randomly assign patients with gout and type 2 diabetes to initiate an SGLT2i versus a DPP-4i (or GLP-1RA) in an unblinded fashion. We first specified the target trial emulation framework that would address our research questions (Supplementary Tables 1 and 2), including eligibility criteria, treatment strategies and assignment, start and end of follow-up, outcomes, causal contrasts, and analysis plan. We then specified how we would use the population data to emulate components of the protocol and conduct the respective analyses (21).

Study Population and Design

For our emulated target trials, we performed inverse probability–weighted, new-user cohort studies among patients with preexisting gout and type 2 diabetes (14) (Supplementary Table 3) initiating an SGLT2i or a DPP-4i. The primary source population was the entire population of the province of British Columbia in Canada. We used Population Data BC, population-based linked administrative databases that provide deidentified, individual-level data for nearly all of British Columbia’s >5 million residents on all provincially funded outpatient medical visits and hospital discharges, emergency department (ED) encounters, and dispensed prescriptions, linked with vital statistics. Access to data provided by the Data Stewards is subject to approval but can be requested for research projects through the Data Stewards or their designated service providers. The following data sets were used in this study: BC Medical Services Plan, Consolidation Database, Hospital Separations, PharmaNet, National Ambulatory Care Reporting System, and Vital Statistics Deaths. Further information regarding these data sets can be found by visiting the PopData project webpage at: https://my.popdata.bc.ca/project_listings/21-048/. All inferences, opinions, and conclusions drawn in this publication are those of the author(s), and do not reflect the opinions or policies of the Data Steward(s).

The study population included patients aged ≥18 years with gout and type 2 diabetes who had at least 1 year of continuous enrollment in the database and initiation of any of the study medications (i.e., first dispensing after a 6-month washout period [22]) between January 2014 and June 2022. For target trials assessing the outcome of allopurinol initiation, we restricted to individuals with no recorded dispensing of a ULT (i.e., allopurinol, febuxostat, probenecid) during the prior 12 months. Diagnoses of gout and type 2 diabetes were based on the presence of one or more codes from the ICD-9 (for outpatient encounters) or ICD-10 (for inpatient and ED encounters) (14) (Supplementary Table 3). We then sought to replicate our findings with confirmation of balanced serum urate levels at baseline in an electronic health record (EHR) database from another general population (i.e., Clalit Health Services [CHS] health maintenance organization in Israel) (23,24) for which laboratory data were available.

Outcome Assessment

The primary outcome was the first dispensing of allopurinol. Secondary outcomes included counts of prescriptions for medications used to treat gout flares, including indomethacin (prototypic NSAID for gout flares), high-dose glucocorticoids (≥30 mg prednisone-equivalent daily dose), and colchicine. We also examined the use of diuretic medications, which not only are used to treat gout’s CKM comorbidities but also are a major risk factor for recurrent gout (25).

To simultaneously examine gout flare control together with the drug use trends above, we also assessed counts of gout-related primary ED visits, hospitalizations, and outpatient encounters during follow-up, as well as recurrent gout flares. Gout flares were defined by 1) an ED visit or hospitalization with a primary diagnosis of gout or 2) a gout-coded outpatient encounter together with at least one of the following treatments dispensed within 7 days: intra-articular or oral corticosteroids, colchicine, or NSAIDs (14,15). Gout-related primary ED visits and hospitalizations (27) and the combination of gout visit and drug dispensing or procedure (28) have been found to accurately ascertain gout flares, with a positive predictive value of 95% for that combination in identifying patients with at least one flare during a 15-month study period (28), closely resembling our ascertainment of flare counts over a similar follow-up period.

Covariate Assessment

Covariates, listed in Table 1, included sociodemographic factors, calendar year of first dispensing, time since first diagnosis of gout and diabetes, comorbidities, diabetes complications, gout flare rate, risk factors for flares and ULT initiation (e.g., cardiovascular disease, chronic kidney disease [CKD], obesity, diuretics) (25,26), gout flare medications (colchicine, corticosteroids, NSAIDs), other diabetes and relevant (e.g., cardiovascular, antihypertensive) medications, and health care utilization. Serum urate, other laboratory values, BMI, and smoking status were included as covariates in the secondary replication cohort. We also determined baseline prevalence of polypharmacy, defined as active prescriptions for five or more distinct medications on target trial index date.

Table 1.

Baseline characteristics among patients with gout and type 2 diabetes and no recent ULT use initiating SGLT2is vs. DPP-4is before and after IPTW

	Before weighting			After weighting
Variable	SGLT2i (n = 10,939)	DPP-4i (n = 8,045)	Standardized differencea	SGLT2i (n = 10,726)	DPP-4i (n = 8,343)	Standardized differencea
Age, mean (SD), years	64.2 (11.2)	67.8 (12.0)	0.31	65.37 (11.07)	65.09 (12.44)	0.02
Male sex (%)	7,793 (71.2)	5,350 (66.5)	0.10	7,444.2 (69.4)	5,773.5 (69.2)	<0.01
Neighborhood income quintileb			0.09			0.02
1 (Lowest)	2,287 (20.9)	1,854 (23.0)		2,326.6 (21.7)	1,828.0 (21.9)
2	2,376 (21.7)	1,879 (23.4)		2,380.3 (22.2)	1,880.1 (22.5)
3	2,331 (21.3)	1,610 (20.0)		2,225.1 (20.7)	1,704.3 (20.4)
4	2,176 (19.9)	1,431 (17.8)		2,061.3 (19.2)	1,611.8 (19.3)
5 (Highest)	1,667 (15.2)	1,166 (14.5)		1,605.4 (15.0)	1,217.4 (14.6)
Health regionc			0.08			0.04
Interior	1,563 (14.3)	1,126 (14.0)		1,530.4 (14.3)	1,176.0 (14.1)
Fraser	4,903 (44.8)	3,571 (44.4)		4,812.6 (44.9)	3,692.3 (44.3)
Vancouver Coastal	2,387 (21.8)	1,975 (24.5)		2,427.8 (22.6)	1,871.1 (22.4)
Vancouver Island	1,364 (12.5)	928 (11.5)		1,280.8 (11.9)	1,006.9 (12.1)
Northern	709 (6.5)	437 (5.4)		661.8 (6.2)	579.2 (6.9)
Type 2 diabetes duration, mean (SD), years	13.70 (7.40)	14.31 (7.11)	0.08	13.90 (7.25)	13.90 (7.23)	<0.01
Gout duration, mean (SD), years	11.36 (7.88)	11.21 (7.70)	0.02	11.26 (7.84)	11.26 (7.69)	<0.01
Gout flares, mean (SD)	0.08 (0.34)	0.10 (0.39)	0.05	0.09 (0.36)	0.09 (0.35)	0.01
Comorbidity
Obesity	1,357 (12.4)	664 (8.3)	0.14	1,161.7 (10.8)	961.3 (11.5)	0.02
Hypertension	7,536 (68.9)	5,230 (65.0)	0.08	7,149.9 (66.7)	5,535.8 (66.4)	<0.01
Myocardial infarction	1,656 (15.1)	1,142 (14.2)	0.03	1,529.2 (14.3)	1,136.1 (13.6)	0.02
Stroke	1,526 (14.0)	1,293 (16.1)	0.06	1,521.7 (14.2)	1,163.6 (13.9)	<0.01
Heart failure	1,943 (17.8)	1,496 (18.6)	0.02	1,842.1 (17.2)	1,360.7 (16.3)	0.02
Ischemic heart disease	5,162 (47.2)	3,593 (44.7)	0.05	4,884.6 (45.5)	3,767.4 (45.2)	<0.01
Atrial fibrillation	694 (6.3)	547 (6.8)	0.02	682.1 (6.4)	503.8 (6.0)	0.01
Varicose veins	677 (6.2)	488 (6.1)	<0.01	647.4 (6.0)	513.3 (6.2)	<0.01
Venous thromboembolism	53 (0.5)	57 (0.7)	0.03	58.6 (0.5)	48.9 (0.6)	<0.01
Peripheral vascular disease	1,527 (14.0)	1,186 (14.7)	0.02	1,473.4 (13.7)	1,125.0 (13.5)	<0.01
CKD	1,817 (16.6)	1,806 (22.4)	0.15	1,944.9 (18.1)	1,503.8 (18.0)	<0.01
Diabetes complications
Nephropathy	942 (8.6)	1,305 (16.2)	0.23	1,188.7 (11.1)	949.1 (11.4)	<0.01
Retinopathy	6,357 (58.1)	4,911 (61.0)	0.06	6,359.7 (59.3)	4,911.0 (58.9)	<0.01
Neuropathy	2,222 (20.3)	1,505 (18.7)	0.04	2,085.6 (19.4)	1,642.4 (19.7)	<0.01
Medicationsd (%)
Diabetes
Metformin	8,813 (80.6)	6,323 (78.6)	0.05	8,626.2 (80.4)	6,752.0 (80.9)	0.01
Sulfonylureas	4,786 (43.8)	4,119 (51.2)	0.15	5,001.5 (46.6)	3,929.6 (47.1)	0.01
GLP-1RAs	1,329 (12.1)	224 (2.8)	0.36	914.7 (8.5)	848.8 (10.2)	0.06
Insulin	2,800 (25.6)	1,413 (17.6)	0.20	2,404.4 (22.4)	1,993.5 (23.9)	0.04
Gout
Colchicine	987 (9.0)	730 (9.1)	<0.01	957.0 (8.9)	718.5 (8.6)	0.01
Corticosteroid	721 (6.6)	621 (7.7)	0.04	747.7 (7.0)	560.3 (6.7)	0.01
NSAIDs	1,065 (9.7)	725 (9.0)	0.03	1,033.9 (9.6)	819.0 (9.8)	<0.01
Aspirin	237 (2.2)	317 (3.9)	0.10	296.3 (2.8)	236.1 (2.8)	<0.01
Proton pump inhibitors	3,230 (29.5)	2,344 (29.1)	0.01	3,116.0 (29.1)	2,419.1 (29.0)	<0.01
Statins	668 (6.1)	567 (7.0)	0.04	693.6 (6.5)	552.2 (6.6)	<0.01
Nitrates	746 (6.8)	596 (7.4)	0.02	727.7 (6.8)	531.2 (6.4)	0.02
Fibrates	110 (1.0)	69 (0.9)	0.02	99.3 (0.9)	98.4 (1.2)	0.03
Other cardiovasculare	3,675 (33.6)	2,690 (33.4)	<0.01	3,492.4 (32.6)	2,679.8 (32.1)	<0.01
Thiazide diuretics	2,577 (23.6)	1,868 (23.2)	<0.01	2,515.7 (23.5)	1,946.5 (23.3)	<0.01
Opioids	2,880 (26.3)	2,267 (28.2)	0.04	2,902.5 (27.1)	2,266.1 (27.2)	<0.01
Loop diuretics	1,258 (11.5)	1,105 (13.7)	0.07	1,270.5 (11.8)	938.0 (11.2)	0.02
Potassium-sparing diuretics	814 (7.4)	419 (5.2)	0.09	673.1 (6.3)	464.2 (5.6)	0.03
Oral immunosuppressants	324 (3.0)	312 (3.9)	0.05	366.9 (3.4)	275.3 (3.3)	<0.01
Biologics	91 (0.8)	55 (0.7)	0.02	84.6 (0.8)	74.8 (0.9)	0.01
Health care utilization, mean (SD)d
Hospitalizations	0.48 (1.00)	0.61 (1.20)	0.12	0.51 (1.04)	0.52 (1.06)	<0.01
ED visits	0.60 (1.51)	0.75 (1.77)	0.09	0.64 (1.60)	0.66 (1.60)	0.01
Hospitalizations and ED visits for diabetes	0.34 (0.81)	0.47 (1.04)	0.15	0.38 (0.87)	0.39 (0.91)	0.02
Total outpatient encounters	23.86 (17.15)	27.41 (24.68)	0.17	24.63 (18.12)	24.96 (21.88)	0.02
Rheumatology encounters	0.10 (0.63)	0.11 (0.80)	<0.01	0.10 (0.61)	0.11 (0.78)	<0.01
Endocrinology encounters	0.62 (1.90)	0.49 (1.58)	0.07	0.56 (1.72)	0.61 (2.45)	0.03
Internal medicine encounters	1.11 (2.69)	1.38 (3.98)	0.08	1.15 (2.86)	1.16 (3.37)	<0.01

Data are n (%) unless otherwise indicated.

^aStandardized difference <0.1 indicates negligible differences.

^bData missing for 1.1%

^cData missing for 0.1%.

^dFrequency during the past 1 year.

^eIncludes α-adrenergic blocking agents, calcium-channel blocking agents, β-blockers, angiotensin II receptor antagonists, ACE inhibitors, antiarrhythmic agents, anticoagulants, and cardiac glycosides.

Cohort Follow-up

Follow-up started the day of first dispensing of an SGLT2i or active comparator (target trial index date) and continued until the end of the study period (June 2022), deregistration, prescription of allopurinol (for primary outcome analysis), discontinuation of the index medication (>60 days following expiration date of the last dispensed supply [14,15]), switching to or adding the comparator medication, or death.

Statistical Analysis

To emulate randomization, we used stabilized inverse probability of treatment weighting (IPTW) for the marginal treatment effect (i.e., average treatment effect in the whole population of eligible patients) (29,30). Weights were calculated based on the covariates described above in a multivariable logistic regression model to predict each patient’s probability (i.e., propensity score [PS]) of filling a prescription for SGLT2i or DPP-4i before the start of follow-up. IPTWs were based on PSs of the cohort resulting in a weighted cohort (pseudo-population) where the distribution of observed covariates is balanced between treatment groups, thus emulating the conditions of a randomized trial (31). Stabilized weights are recommended to account for the otherwise unstable weights assigned to individuals with a low probability of receiving the treatment of interest and the resultant large variance for the IPTW estimator (29,32). We assessed the balance of distribution of covariates before and after weighting using standardized differences between treatment groups (<0.1 denoting a negligible difference) (31). In a sensitivity analysis, we used overlap weighting of the PS for which the estimand is the average treatment effect in the population most likely to receive either treatment (i.e., PS close to 0.5) (30,33,34). This approach retains all eligible patients and results in exact balance of the mean of every measured covariate (35). We also matched SGLT2i users one to one to DPP-4i users on the PS, with replacement, for the conditional treatment effect (i.e., average treatment effect among the treated population) (30).

We calculated incidence rates for allopurinol dispensing and estimated hazard ratio (HRs) and rate differences (RDs) and 95% CIs between treatment groups, using robust sandwich-type variance estimation overall and by subgroups defined by sex, age, baseline use of diuretic medications (loop, thiazide and potassium sparing), and gout intensity (one or more gout-coded encounters or colchicine dispensing over the past year) (14). Within each subgroup, PSs were recalculated with patients reweighted. Cox proportional hazard models were used to obtain HRs for first-event analysis, accounting for the competing risk of death using the Fine-Gray approach (36). In our analyses of gout flare/diuretic medication dispensing and flare counts, we performed Poisson regression to obtain rate ratios (RRs) and calculate RDs. The E-value was calculated to evaluate the robustness of our primary outcome to unmeasured confounders.

Additional sensitivity analyses included truncating follow-up after 2 years and carrying forward the index medication exposure for up to 1 and 2 years (akin to an intention-to-treat analysis). To evaluate reproducibility with established relations and unmeasured systematic bias, we assessed the risk of genital infection, which is associated with SGLT2i exposure (37), as a positive control outcome and risk of an osteoarthritis encounter and appendicitis as negative control outcomes, as done previously (14,15).

Replication Analysis With Laboratory Data

For the primary outcome analysis of allopurinol initiation, we sought to confirm serum urate balance at baseline and replicate and meta-analyze the findings in the CHS EHR data set (38). Methods for establishing the cohort of patients with gout and type 2 diabetes and the statistical analysis were the same as the primary analysis. We constructed three PSs and compared the results associated with each: one with the same set of variables included in the main analysis from British Columbia (where laboratory values were not available), one adding serum urate levels to the PS, and a third adding levels of serum urate, creatinine, triglycerides, total cholesterol, HbA_1c, albumin-creatinine ratio, BMI, and smoking status.

Statistical analyses were performed using SAS (version 9.4) and R (version 4.3.1). Ethical approval was obtained from The University of British Columbia’s Behavioral Research Ethics Board, Vancouver, British Columbia, Canada (H15-00887) and Soroka University Medical Center Institutional Review Board, Beer-Sheva, Israel (0174-24).

Data and Resource Availability

Data for this study are not publicly available because of a data use agreement. Requests to access the study should be made to the corresponding author.

Results

There were 18,984 patients with gout and type 2 diabetes and no baseline ULT use eligible for the emulated trial (Supplementary Fig. 1). Even before inverse probability weighting, gout-related covariates, including gout duration, flare rates, use of flare medications and diuretics, and number of rheumatology encounters, were similar between treatment groups, although obesity was more prevalent among those initiating an SGLT2i, and CKD was more prevalent among those initiating a DPP-4i. All baseline covariates, including obesity and CKD, were well balanced after IPTW (standardized differences <0.1), with 10,726 weighted initiators of SGLT2i and 8,343 weighted initiators of DPP-4i included in the target trial (Table 1). Sixty-nine percent were males with a mean age at index date of 65 years (median 65 years, Q1 57 years among SGLT2i users; median 69 years, Q1 61 years among DPP-4i users). Prevalence of polypharmacy in this cohort (without ULT use in the 12-month baseline period) was 62%.

Mean follow-up was 1.4 years among SGLT2i initiators and 1.3 years among DPP-4i initiators, with empagliflozin and linagliptin accounting for the majority of prescriptions (68% and 64%, respectively) (Supplementary Table 4). The most common reason for censoring was discontinuation of index treatments, followed by end-of-study period (Supplementary Table 5). Follow-up ceased due to death for 4.1% of DPP-4i initiators and 1.9% of SGLT2i initiators.

Allopurinol Initiation

There were 311 SGLT2i users who initiated allopurinol (21.6 per 1,000 person-years) compared with 365 DPP-4i users (36.7 per 1,000 person-years) (Supplementary Fig. 2). This corresponded to a weighted HR of 0.62 (95% CI 0.52–0.73) associated with SGLT2i use and RD of −15 (95% CI −20 to −11) per 1,000 person-years (Table 2). Findings were consistent when truncating follow-up after 2 years, when carrying forward the index medication exposure up to 1 and 2 years and one-to-one PS-matching with replacement (Table 2), and regardless of sex and age-group and gout intensity at baseline, though the HR was numerically stronger among patients aged ≥65 years (Supplementary Table 6). Findings were also consistent with the primary analysis when comparing inverse probability–weighted initiators of SGLT2i with initiators of GLP-1RA (n = 13,038 for SGLT2i and = 4,490 for GLP-1RA) (Supplementary Table 7 and Supplementary Figs. 3 and 4), with a weighted HR of 0.75 (95% CI 0.56–0.98) and RD of −12 per 1,000 person-years (Table 2), and closely agreed when applying overlap weighting (Supplementary Table 8 and Table 2).

Table 2.

Probability of initiation of allopurinol among patients with gout and type 2 diabetes following initiation of SGLT2is vs. DPP-4is before and after IPTW

	Before weighting		After weighting
	SGLT2i	DPP-4i	SGLT2i	DPP-4i
Allopurinol initiation (primary outcome)
Patients, n	10,939	8,045	10,726	8,343
Events, n	310	426	311	365
IR, per 1,000 person-years	22.5	36.9	21.58	36.7
HR (95% CI)	0.59 (0.51 to 0.69)	1.0 (Ref)	0.62 (0.52 to 0.73)	1.0 (Ref)
RD (95% CI)	−14.4 (−18.7 to −10.1)	Ref	−15.1 (−19.6 to −10.7)	Ref
Sensitivity analysis of truncating follow-up after 2 yearsa
Patients, n	10,939	8,045	10,726	8,343
Events, n	292	368	291	329
IR, per 1,000 person-years	27.97	44.54	27.27	43.09
RR (95% CI)	0.64 (0.54 to 0.76)	1.0 (Ref)	0.65 (0.55 to 0.78)	1.0 (Ref)
RD (95% CI)	−16.6 (−22.1 to −11.0)	Ref	−15.8 (−21.4 to −10.2)	Ref
Sensitivity analysis of carrying forward the index medication exposure up to 1 yearb
Patients, n	10,939	8,045	10,726	8,343
Events, n	304	357	306	329
IR, per 1,000 person-years	33.11	48.73	32.67	46.13
HR (95% CI)	0.68 (0.58 to 0.80)	1.0 (Ref)	0.70 (0.59 to 0.85)	1.0 (Ref)
RD (95% CI)	−15.6 (−21.9 to −9.3)	Ref	−13.4 (−19.6 to −7.3)	Ref
Sensitivity analysis of carrying forward the index medication exposure up to 2 yearsc
Patients, n	10,939	8,045	10,726	8,343
Events, n	459	544	473	484
IR, per 1,000 person-years	28.86	40.30	28.23	38.39
HR (95% CI)	0.74 (0.64 to 0.85)	1.0 (Ref)	0.74 (0.64 to 0.87)	1.0 (Ref)
RD (95% CI)	−11.4 (−15.1 to −7.1)	Ref	−10.1 (−14.4 to −5.9)	Ref
Sensitivity analysis applying overlap weighting instead of IPTW
Patients, n	10,939	8,045	10,939	8,045
Events, n	310	426	310	426
IR, per 1,000 person-years	22.5	36.9	21.55	37.89
HR (95% CI)	0.59 (0.51 to 0.69)	1.0 (Ref)	0.59 (0.50 to 0.70)	1.0 (Ref)
RD (95% CI)	−14.4 (−18.7 to −10.1)	Ref	−16.4 (−23.9 to −9.5)	Ref
Sensitivity analysis of one-to-one PS matching with replacement instead of IPTWd
Patients, n	10,939	8,045	10,939	4,080
Events, n	310	426	310	200
IR, per 1,000 person-years	22.5	36.9	22.5	38.7
HR (95% CI)	0.59 (0.51 to 0.69)	1.0 (Ref)	0.58 (0.49 to 0.70)	1.0 (Ref)
RD (95% CI)	−14.4 (−18.7 to −10.1)	Ref	−16.2 (−22.2 to −10.3)	Ref
Sensitivity analysis using GLP-1RA as the comparatore
Patients, n	13,016	4,363	13,038	4,490
Event, n	383	124	364	138
IR, per 1,000 person-years	23.18	36.08	22.67	34.33
HR (95% CI)	0.76 (0.62 to 0.94)	1.0 (Ref)	0.75 (0.56 to 0.98)	1.0 (Ref)
RD (95% CI)	−12.9 (−19.7 to −6.1)	Ref	−11.8 (−17.9 to −5.6)	Ref

IR, incidence rate; Ref, reference.

^aPatients were censored at the earliest of deregistration from the provincial medical plan, discontinuation of the index medication, switching to or addition of the comparator medication, death, or accrual of 730 days of follow-up.

^bNo censoring if index medication was discontinued or switched within 365 days of initiation.

^cNo censoring if index medication was discontinued or switched within 730 days of initiation.

^dNo caliper was imposed. We computed the distribution of the number of times each control patient was used in matched sets. Under matching with replacement, the median number of times a control patient (DPP-4) was used was 2 (25th–75th percentiles 1–3; 10th–90th percentiles 1–4). Because replacement creates nonindependent observations, we accounted for repeated selection of the same control patient via a cluster-robust variance estimator, with patients treated as clusters.

^eData in the third and fifth columns are for GLP-1RA use.

One subgroup that showed evidence of effect modification was those with baseline diuretic use, with an HR of 0.49 (95% CI 0.38–0.63) and RD of −27 (95% CI −36 to −19) per 1,000 person-years among those using diuretics at baseline (n = 3,734 SGLT2i [34% of the cohort]) compared with an HR of 0.76 (95% CI 0.61–0.95) and RD of −11 (95% CI −16 to –6.2) per 1,000 person-years among those not using diuretics at baseline (P for interaction = 0.03) (Supplementary Table 6). This effect modification was also present when comparing SGLT2i with GLP-1RA, with HRs of 0.58 (95% CI 0.38–0.88) and 0.89 (95% CI 0.66–1.20) among those with and without baseline diuretic use (P for interaction = 0.04). There was no effect modification observed for sex, age, or gout intensity.

Gout Flare Medication and Diuretic Use

The target trials for the medication use and flare end points included 14,876 weighted patients with gout and type 2 diabetes initiating an SGLT2i and 11,979 weighted patients initiating a DPP-4i, (standardized difference <0.1) (Supplementary Table 9). Mean age was 66 years, and polypharmacy was present in 65% overall, including 74% of those aged ≥65 years and 76% of those with concomitant CKD. Compared with DPP-4i and GLP-1RA use, SGLT2i use was associated with lower rates of dispensing of high-dose glucocorticoids, indomethacin, and colchicine over the study period, after adjusting for baseline use (Table 3). For example, the weighted RR for high-dose glucocorticoid prescriptions was 0.78 (95% CI 0.74–0.83) versus DPP-4i, with an RD of −30 (95% CI −38 to −23) per 1,000 person-years, while the corresponding RR for indomethacin prescriptions was 0.85 (95% CI 0.80–0.92), with an RD of −9 (95% CI −15 to −3) per 1,000 person-years. However, unlike for the allopurinol initiation end point, there was no evidence of effect modification by diuretic use.

Table 3.

Counts of prescriptions for gout flare medications dispensed to patients with gout and type 2 diabetes following initiation of SGLT2i vs. DPP-4i or GLP-1RA Before and After IPTW

	Before weighting		After weighting
	SGLT2i	DPP-4i	SGLT2i	DPP-4i
High-dose glucocorticoidsa
Patients, n	15,237	11,502	14,876	11,979
Events, n	2,083	2,601	2,226	2,118
IR, per 1,000 person-years	105.74	145.14	108.36	138.70
RR (95% CI)	0.73 (0.69 to 0.77)	1.0 (Ref)	0.78 (0.74 to 0.83)	1.0 (Ref)
RD (95% CI)	−39.4 (−46.6 to −32.2)	Ref	−30.3 (−37.8 to −22.9)	Ref
GLP-1RA as comparatorb
Patients, n	18,120	6,240	18,140	6,603
Events, n	2,413	779	2,367	829
IR, per 1,000 person-years	102.10	147.97	103.11	135.81
RR (95% CI)	0.69 (0.64 to 0.75)	1.0 (Ref)	0.75 (0.70 to 0.82)	1.0 (Ref)
RD (95% CI)	−45.9 (−57.0 to −34.7)	Ref	−32.8 (−42.9 to −22.6)	Ref
Indomethacin
Patients, n	15,237	11,502	14,876	11,979
Events, n	1,660	1,437	1,739	1,427
IR, per 1,000 person-years	84.27	80.19	84.64	93.48
RR (95% CI)	1.05 (0.98 to 1.13)	1.0 (Ref)	0.85 (0.80 to 0.92)	1.0 (Ref)
RD (95% CI)	4.1 (−1.7 to 9.9)	Ref	−8.8 (−15.1 to −2.5)	Ref
GLP-1RA as comparatorb
Patients, n	18,120	6,240	18,140	6,603
Events, n	2,080	479	1,960	545
IR, per 1,000 person-years	88.01	90.99	85.37	89.37
RR (95% CI)	0.97 (0.88 to 1.07)	1.0 (Ref)	0.91 (0.83 to 1.00)	1.0 (Ref)
RD (95% CI)	−3.0 (−11.9 to 6.0)	Ref	−4.0 (−12.3 to 4.4)	Ref
Colchicine
Patients, n	15,237	11,502	14,876	11,979
Events, n	3,268	3,314	3,230	2,732
IR, per 1,000 person-years	165.89	184.93	157.22	178.97
RR (95% CI)	0.90 (0.85 to 0.94)	1.0 (Ref)	0.87 (0.83 to 0.92)	1.0 (Ref)
RD (95% CI)	−19.0 (−27.5 to −10.5)	Ref	−21.7 (−30.3 to −13.1)	Ref
GLP-1RA as comparatorb
Patients, n	18,120	6,240	18,140	6,603
Events, n	3,936	1,116	3,825	1,314
IR, per 1,000 person-years	166.54	211.98	166.61	215.32
RR (95% CI)	0.79 (0.74 to 0.84)	1.0 (Ref)	0.77 (0.73 to 0.82)	1.0 (Ref)
RD (95% CI)	−45.4 (−59.4 to −32.3)	Ref	−48.7 (−61.5 to −40.0)	Ref

IR, incidence rate; Ref, reference.

^aGreater than or equal to 30 mg prednisone-equivalent daily dose.

^bData in the third and fifth columns are for GLP-1RA use.

SGLT2i use was also associated with lower rates of dispensing of diuretic medications, with weighted RRs of 0.87 (95% CI 0.85–0.89) versus DPP-4i and 0.82 (95% CI 0.80–0.85) versus GLP-1RA (Table 4). The RRs were even stronger for dispensing of loop diuretics alone (e.g., 0.58 [95% CI 0.56–0.60] vs. DPP-4i).

Table 4.

Counts of prescriptions for diuretic medications dispensed to patients with gout and type 2 diabetes following initiation of SGLT2i vs. DPP-4i or GLP1-RA, after IPTW

	SGLT2i	DPP-4i
All diuretics (loop, thiazide, or potassium-sparing)
Patients, n	14,876	11,979
Events, n	12,099	10,822
IR, per 1,000 person-years	588.96	708.83
RR (95% CI)	0.87 (0.85 to 0.89)	1.0 (Ref)
RD (95% CI)	−119.9 (−142.2 to −97.5)	Ref
Loop diuretics
Patients, n	14,876	11,979
Events, n	4,394	4,955
IR, per 1,000 person-years	213.89	324.53
RR (95% CI)	0.58 (0.56 to 0.60)	1.0 (Ref)
RD (95% CI)	−110.6 (−121.7 to −99.6)	Ref
Thiazide diuretics
Patients, n	14,876	11,979
Events, n	8,073	6,579
IR, per 1,000 person-years	392.97	430.91
RR (95% CI)	0.92 (0.89 to 0.95)	1.0 (Ref)
RD (95% CI)	−37.9 (−51.4 to −24.4)	Ref
GLP-1RA as comparatora
All diuretics (loop, thiazide, or potassium-sparing)
Patients, n	18,140	6,603
Events, n	13,807	4,574
IR, per 1,000 person-years	601.40	749.65
RR (95% CI)	0.82 (0.80 to 0.85)	1.0 (Ref)
RD (95% CI)	−148.2 (−172.2 to −124.3)	Ref
Loop diuretics
Patients, n	18,140	6,603
Events, n	4,861	1,707
IR, per 1,000 person-years	211.74	279.82
RR (95% CI)	0.81 (0.77 to 0.86)	1.0 (Ref)
RD (95% CI)	−68.0 (−82.6 to −53.5)	Ref
Thiazide diuretics
Patients, n	18,140	6,603
Events, n	9,682	3,130
IR, per 1,000 person-years	421.73	512.98
RR (95% CI)	0.83 (0.80 to 0.86)	1.0 (Ref)
RD (95% CI)	−91.2 (−111.1 to −71.4)	Ref

IR, incidence rate; Ref, reference.

^aData in the third column are for GLP-1RA use.

Recurrent Gout Flares and Gout-Related Primary Encounters

Alongside the lower rates of gout-related medication use, SGLT2i initiation was associated with lower rates of recurrent gout flares, with a weighted RR of 0.65 (95% CI 0.60–0.72) and RD of −20 (95% CI −24 to −15) per 1,000 person-years. Findings were consistent when tabulating all gout-related primary health care encounters, regardless of flare medication dispensing, and when comparing SGLT2i with GLP-1RA (n = 18,140 and 6,603 weighted initiators of SGLT2i and GLP-1RA, respectively) (Table 3 and Supplementary Table 10).

Potential Effect of Unmeasured Confounding

The E-value corresponding to the upper bound for allopurinol initiation was 2.08 (E-value for the point estimate was 2.61). Thus, our observed HR could be explained away by an unmeasured confounder that occurs 2.08 times more often in the DPP-4i group than the SGLT2i group, increases the rate of allopurinol initiation by 2.08 times, and is not correlated with the measured confounders we included in the weighting. In the positive control outcome analyses, SGLT2i use was associated with a higher risk of genital infection but no increased risk of an osteoarthritis encounter or appendicitis, all as expected (Supplementary Table 11).

Replication Analysis With Laboratory Data

Even before inverse probability weighting, among 6,845 CHS members with gout and type 2 diabetes, relevant laboratory variables, including HbA_1c and serum urate, were well balanced, similar to gout flare rates and rheumatologist visits as in our primary data set (Supplementary Table 12). The weighted cohort consisted of 12,823 CHS members with gout and type 2 diabetes who were part of the emulated target trial comparing SGLT2i and DPP-4i (Supplementary Fig. 5). The mean age of the weighted population was 67 years, and 69% were males. Among those with no baseline ULT use, weighted HRs for allopurinol initiation were 0.65 (95% CI 0.53–0.80) before addition of any laboratory values in the PS, 0.66 (95% CI 0.54–0.81) when serum urate levels were added, and 0.66 (95% CI 0.55–0.84) when all laboratory values plus BMI and smoking were included. Corresponding RRs for the outcome of recurrent gout flare were 0.71 (95% CI 0.66–0.76), 0.71 (95% CI 0.67–0.76), and 0.77 (95% CI 0.72–0.82).

Findings remained similar when using overlap weighting to emulate randomization (Supplementary Table 13). When stratifying by baseline diuretic use, the HRs for ULT initiation were 0.59 (95% CI 0.43–0.80) and 0.75 (95% CI 0.57–1.00) among those using and not using diuretics at baseline, respectively. When the primary and CHS findings were meta-analyzed, summary HRs for ULT initiation were 0.63 (95% CI 0.55–0.72) overall and 0.53 (95% CI 0.43–0.64) and 0.76 (95% CI 0.64–0.90) among those using and not using diuretics at baseline.

Conclusions

In these emulated trials leveraging population-based cohorts of mainly older adults with gout and type 2 diabetes, SGLT2i use was associated with a 38% lower probability of subsequent ULT initiation compared with DPP-4i use, as well as a lower need for acute gout flare anti-inflammatory drugs (e.g., glucocorticoids, indomethacin) and lesser use of diuretics. Gout-related risk factors (e.g., prior flare rates, serum urate levels) were well balanced even before weighting, most likely because the compared medications were not clinically considered for gout care during the study period and the protective associations were similarly evident before adjusting for covariates. Furthermore, they persisted among demographic subgroups, regardless of baseline gout intensity, and were stronger among patients with diuretic use at baseline than those without. As expected, SGLT2i use was associated with a higher risk of genital infection but no increased risk of osteoarthritis or appendicitis. While SGLT2is are unlikely to displace classical urate-lowering agents in the gout armamentarium, these real-world findings suggest that in addition to their glucose-lowering and cardiovascular and kidney benefits, SGLT2is may provide a mechanism for reducing pill burden and exposure to the often-harmful effects of NSAIDs and glucocorticoids in this high-risk population.

We previously reported on the urate-lowering effect of SGLT2is translating to lower rates of recurrent gout flares (14,15). SGLT2i’s anti-inflammatory effects, including inhibition of IL-1β (39–41), a key mediator of the inflammatory cascade associated with gout flares (42), as well as the reduction in oxidative stress through enhancement of the SIRT-1 signaling pathway (43), likely also contribute. Importantly, this protective association persisted regardless of baseline ULT, and appeared during the first year of SGLT2i use, and we have now replicated this association when directly accounting for baseline serum urate, other laboratory findings, and BMI.

The current work provides real-world evidence that SGLT2is not only lower flare rates but also can be gout-related medication sparing, with multiple potential downstream benefits. For one, although allopurinol is generally well tolerated, its initiation can lead to a prescribing cascade and increased risk of adverse effects if medications for paradoxical flares are used (16,17). Indeed, in a recent observational study of patients with gout initiating allopurinol, NSAID prophylaxis was associated with a ≥60% increased risk of acute kidney injury, angina, peptic ulcer disease, and myocardial infarction (44). Nearly every prescription for high-dose glucocorticoids worsens glycemia in type 2 diabetes, and long-term use can cause weight gain, undermining effective diabetes management. While colchicine has a safer adverse effect profile, both colchicine and NSAIDs should be used cautiously in patients with CKD, a frequent comorbidity of gout (45). This prescribing cascade can extend if adverse effects require treatment or warrant prescription of additional prophylactic therapies (e.g., proton pump inhibitors for protection against NSAID-induced peptic ulcers) (45).

These effects also have broader implications for improving multimorbidity care for these high-risk patients. Patients with gout are particularly prone to polypharmacy due to their older age on average and elevated burden of metabolic syndrome components (2) and CKM comorbidities, including hypertension, dyslipidemia, CKD, heart failure, and type 2 diabetes (3), that typically require treatment with multiple medications. Indeed, nearly half of all U.S. adults aged ≥65 years have reported polypharmacy (46). Prevalence of polypharmacy was 74% for older adults in this cohort of patients with gout and type 2 diabetes and 76% among the subset with concomitant CKD. Moreover, patients with gout have identified medication side effects, drug-drug interactions, and the burden of taking multiple medications for gout and its comorbidities as critical concerns (8,9). Our findings suggest that the pleiotropic urate-lowering benefits of SGLT2is could be leveraged to help reduce the risk of potential adverse effects of allopurinol prescribing (e.g., paradoxical flares, allergic rash) while improving the use and adherence of other, medically necessary medications in this high-risk population (47).

The serum urate-lowering effects of SGLT2is have been demonstrated in randomized trials (13). SGLT2 inhibition increases urinary glucose concentration, which competes with urate for GLUT9-mediated reabsorption in the renal proximal tubule, thus enhancing urate excretion (48). Other urate transporters may also be affected (48,49). Furthermore, SGLT2is may inhibit insulin-activated urate transport (49) and have anti-inflammatory effects beyond urate lowering, including inhibition of IL-1β (41) and reduction in oxidative stress (43). Diuretic medication use increases reabsorption of urate in the renal proximal tubule, potentially involving stimulation of the URAT1 transporter (50), and is a major risk factor for gout. As such, the effect modification we observed is likely explained by a greater magnitude of reduction in serum urate levels among diuretic users at baseline compared with nonusers. As well, the lower rates of diuretic use during follow-up may be due to SGLT2is’ known heart failure and blood pressure benefits, reducing the need for loop and thiazide diuretics, respectively. In a post hoc analysis of the Empagliflozin Outcome Trial in Patients With Chronic Heart Failure (EMPEROR)-Preserved study (11), empagliflozin increased the likelihood of diuretic dose deescalation and permanent discontinuation over 36 months, while in a prespecified analysis of the Dapagliflozin Evaluation to Improve the Lives of Patients With Preserved Ejection Fraction Heart Failure (DELIVER) study (12), dapagliflozin reduced the likelihood of loop diuretic initiations and dose increases and reduced the mean loop diuretic dose over time.

Given their more modest urate-lowering effects, SGLT2is are unlikely to displace classical ULT, particularly for patients with a substantial urate burden. Still, these patients with concomitant type 2 diabetes may benefit from using an SGLT2i (instead of an alternative glucose-lowering agent) in combination with conventional ULT (48). SGLT2is should be used with caution and monitoring of adverse effects in older adults, who may be at increased risk of diabetic ketoacidosis, although SGLT2is have conferred greater absolute benefits among older or more frail individuals than those who were younger or healthier (51,52).

Strengths and Limitations

We used a population-based database, which makes our findings more generalizable, with data on nearly all dispensed medications regardless of indication for gout or funding source (as opposed to issued prescriptions), thus reducing exposure misclassification. We also conducted multiple sensitivity analyses, and results were consistent with our primary findings, supporting their robustness. The results for the control outcomes are consistent with prior reports (15,37) and lent specificity to our findings, reducing the likelihood of them being explained by spurious associations or unmeasured confounding.

As in any observational study, there is a potential for residual unmeasured confounding, though it is unlikely that any unmeasured factors would affect the choice between initiating SGLT2i and DPP-4i or GLP-1RA with regard to gout because these medications were not approved or clinically considered for gout care in practice during the study period. As well, our E-value indicated that an unmeasured covariate would need to be associated with both allopurinol initiation and use of a DPP-4i by an RR ≥2.61 to nullify our findings. Although this is the minimum strength of association that an unmeasured confounder would need to have with both DPP-4i use and allopurinol initiation to fully explain away the association, it could also be explained away in a scenario in which one association (e.g., RR for the confounder-allopurinol initiation relationship) was <2.61, as long as the other association (e.g., prevalence ratio for the confounder in DPP-4i vs. SGLT2i) was considerably >2.61 to compensate (Supplementary Fig. 6). Indeed, baseline gout flare rates, gout flare medication use, rheumatologist encounters, and laboratory measures from the EHR data set (e.g., serum urate, HbA_1c) were similar between exposure arms, even before inverse probability weighting. As such, when we incorporated these and other measures of gout and diabetes severity and health care access into the PS, results remained similar, supporting the validity of our findings without confounding from these variables. Although we could not directly adjust for laboratory values in the British Columbia data set, we believe that the population characteristics and health care systems in the two data sets (e.g., both publicly funded with universal coverage) are similar enough to make inferences about these adjustments in the British Columbia setting.

In conclusion, these target trial emulation studies of patients with gout and type 2 diabetes showed that use of SGLT2is was consistently associated with a lower probability of allopurinol initiation compared with DPP-4is or GLP-1RAs, as well as lower rates of gout flare and diuretic medication use, gout-related primary health care encounters, and recurrent flares. In addition to the established CKM benefits for patients with gout and type 2 diabetes, the pleiotropic benefits of SGLT2is therefore reduce the need for gout-related medications. This, in turn, reduces medication burden and exposure to potentially harmful effects of NSAIDs and glucocorticoids in these high-risk patients.

This article contains supplementary material online at https://doi.org/10.2337/figshare.30916430.

Funding Statement

This research was supported by grants P50-AR060772 and R01-AR065944 from the National Institutes of Health and by THC-135235 (Preventing Complications From Inflammatory Skin, Joint and Bowel Conditions [PRECISION]), a team grant funded by the Canadian Institutes of Health Research. N.M. is supported by Career Development Award R00-AR080243 from the National Institutes of Health. C.Y. is supported by K23-AR081425. S.K.R. is supported by a fellowship award from the Canadian Institutes of Health Research.