Switching to Versus Addition of Incretin-Based Drugs Among Patients With Type 2 Diabetes Taking Sodium-Glucose Cotransporter-2 Inhibitors.

Background Evidence is limited in comparing treatment modification by substitution or add-on of glucose-lowering medications in patients with type 2 diabetes. This observational study aims to compare switching versus add-on of incretin-based drugs among patients with type 2 diabetes on background sodium-glucose cotransporter-2 inhibitors (SGLT2i). Methods and Results This population-based, retrospective cohort study was conducted using the IQVIA Medical Research Data, including adults with type 2 diabetes on background SGLT2i from 2005 to 2020. New users of incretin-based drugs were allocated into the "Switch" group if they had discontinued SGLT2i treatment, or the "Add-on" group if their background SGLT2i was continued. Baseline characteristics of patients were balanced between groups. Study outcomes were all-cause mortality, cardiovascular diseases, kidney diseases, hypoglycemia, and ketoacidosis. Patients were observed from the index date of initiating incretin-based drugs until the earliest of an outcome event, death, or data cut-off date. Changes in anthropometric and metabolic parameters were also compared between groups from baseline to 12-month follow-up. A total of 2888 patients were included, classified into "Switch" (n=1461) or "Add-on" group (n=1427). Median follow-up was 18 months with 5183 person-years. Overall, no significant differences in the risks of study outcomes were observed between groups; however, patients in the "Add-on" group achieved significantly greater reductions in glycated hemoglobin, weight, percentage weight loss, and systolic blood pressure than their "Switch" counterparts. Conclusions Initiating incretin-based drugs as add-on among patients with type 2 diabetes on background SGLT2i was associated with risks of clinical end points comparable to switching treatments, in addition to better glycemic and weight control observed with the combination approach.

Charlson Comorbidity Index

dipeptidyl peptidase‐4 inhibitors

end‐stage kidney disease

glucagon‐like peptide‐1 receptor agonists

IQVIA Medical Research Data

inverse probability of treatment weights

systolic blood pressure

sodium‐glucose cotransporter‐2 inhibitors

type 2 diabetes

Clinical Perspective

What Is New?

In this retrospective cohort study of patients with type 2 diabetes who were on background sodium‐glucose cotransporter‐2 inhibitors (SGLT2i), new users of incretin‐based drugs were allocated into the “Switch” group if they had discontinued SGLT2i treatment, or the “Add‐on” group if their background sodium‐glucose cotransporter‐2 inhibitors was continued.

Over a median follow‐up of 18 months, no significant differences in the risks of all‐cause mortality, cardiovascular diseases, kidney diseases, hypoglycemia, and ketoacidosis were observed between groups.

Patients in the “Add‐on” group achieved significantly greater reductions in glycated hemoglobin, weight, percentage weight loss, and systolic blood pressure than their “Switch” counterparts.

What Are the Clinical Implications?

While no significant differences in the risks of various clinical end points were identified between switching and add‐on approaches in the current study, they should be interpreted with caution given the relatively short follow‐up period and hence the small number of events that occurred.

Meanwhile, several metabolic benefits of the combination (“Add‐on”) approach were significantly greater than that of switching, including better glycemic control, reduction in weight and blood pressure over 12‐month follow‐up.

Further studies with longer observation periods and randomized controlled trials are needed to clarify the risks and benefits of the 2 treatment modalities.

Considering the progressive nature of type 2 diabetes (T2D), patients often require multiple antidiabetic agents over their course of disease for optimal glycemic control, where the stepwise approach of initiating new glucose‐lowering medications following the failure of existing therapy in meeting individualized glycated hemoglobin (HbA1c) targets remains the preferred regimen by various international guidelines. , , , When treatment intensification is needed sequential to first‐line metformin monotherapy, introduction of antidiabetic drugs with complementary mechanisms of action is recommended to help address the ominous octet of T2D pathophysiology. , , , Among the different drug classes, sodium‐glucose cotransporter‐2 inhibitors (SGLT2i) offer substantial metabolic benefits beyond glycemic control, reducing the risks of cardiovascular diseases (CVD), progression of diabetic nephropathy, and mortality, in addition to promoting weight loss, lowering blood pressure (BP), and incurring a low risk of hypoglycemia. , , , , With increasing availability and its repositioning as a second‐line glucose‐lowering medication, , , , , it can be anticipated that an increasing number of patients will be put on a combination regimen of metformin and SGLT2i, and it would be intriguing to explore the preferred option for subsequent treatment intensification.

Incretin‐based therapy consisting of dipeptidyl peptidase‐4 inhibitors (DPP4i) and glucagon‐like peptide‐1 receptor agonists (GLP1RA) are alternative antidiabetic agents with demonstrated efficacy and general tolerability. , While specific GLP1RA have exerted beneficial effects in terms of cardiovascular outcomes, especially lowering the risks of major adverse cardiovascular events and mortality, alongside considerable weight loss and BP reduction, , , , , DPP4i are less potent in the stimulation of incretin effect. Hence they are mostly associated with cardiovascular neutrality and clinical benefits of a smaller magnitude than GLP1RA. , , , Because both drug classes act by promoting insulin secretion while suppressing that of glucagon in a glucose‐dependent manner, they may compensate for the increased glucagon level and endogenous glucose production induced by SGLT2i to facilitate better glycemic control, and offer distinct mechanisms of action in targeting the metabolic defects of T2D that are complementary to those of metformin and SGLT2i, respectively, all without posing an additional risk of hypoglycemia. , , , , Accordingly, incretin‐based drugs appear to be an attractive option over sulfonylureas or thiazolidinediones as treatment intensification, with respect to cardiorenal outcomes, clinical parameters, and risk of hypoglycemia. ,

Aside from the selection of antidiabetic agents based on patient preferences, cardiorenal status, and drug safety profile, the choice of drug initiation approach may also influence therapeutic efficacy via factors such as medication burden and patient adherence, correction of T2D pathophysiology, time to achieving individualized targets, clinical inertia, and overall cost‐effectiveness that takes diabetic complications into account. , , A retrospective cohort study utilizing electronic medical records from the UK Clinical Practice Research Datalink (CPRD) found that among patients with T2D with inadequate glycemic control, adding a new glucose‐lowering medication was associated with clinically significant reduction in HbA1c, which was not evident among those switching to another therapy or continuing with the original treatment. Recently, several clinical trials and meta‐analyses have demonstrated that the combination of SGLT2i with incretin‐based drugs may produce subadditive or additive effects in glycemic control and improvements in metabolic parameters than either drug class with placebo , , , , , , ; yet, there is very limited evidence on the comparison of cardiorenal end points and mortality for combination therapy versus each treatment alone. ,

With reference to clinical guidelines recommending the substitution and/or addition of new antidiabetic agents upon limited response to existing glucose‐lowering therapy, as well as the research gap in evaluating any additional cardiorenal benefits of combining SGLT2i with incretin‐based drugs over individual treatments and across different patient subgroups, , , , , , , this observational study aims to compare the all‐cause mortality, cardiorenal outcomes, adverse effects, and changes in clinical parameters associated with incretin‐based drugs as switching versus add‐on therapy among patients with T2D on background SGLT2i in a real‐life setting. Because glucose‐lowering medications with duplicating mechanisms of action are generally not recommended in combination regimens, this study will consider the initiation of DPP4i or GLP1RA as substitution versus add‐on to SGLT2i separately, and compare their safety and efficacy under respective treatment condition.

Methods

Data Source and Study Design

This population‐based, retrospective cohort study was conducted using the IMRD, a database comprising anonymized electronic primary health care records for 15 million patients from >750 general practices across the United Kingdom. IMRD incorporates data supplied by The Health Improvement Network, a propriety database of Cegedim SA. It contains coded patient‐level longitudinal information on demographics, symptoms, clinical diagnoses recorded using Read Codes, medication prescriptions, consultations, and anthropometric, clinical, and laboratory measures. The data set is representative of the UK population by age, sex, medical conditions, and death rates adjusted for demographics, and has similar distribution of major chronic diseases, including diabetes, CVD, and mental illnesses, compared with the UK national statistics. , Validity of the diagnoses of ischemic cerebrovascular events and chronic kidney disease (CKD) with Read Codes in The Health Improvement Network database has been confirmed, , in addition to the accuracy of diabetes, hypertension, and CVD. Studies have utilized this database to explore the associations between glucose‐lowering medications and mortality, macrovascular, and microvascular diseases in patients with T2D. , , We implemented a new user design based on IMRD data. New users of incretin‐based drugs were first‐time‐ever users of GLP1RA or DPP4i drugs.

Study Population

General practices were included in the study from the latest of the following dates: 12 months after reporting acceptable mortality rates (a measure of data‐recording quality), 12 months after beginning the use of electronic medical records, and study start date (January 1, 2005). This was to maximize data and recording quality.

People aged ≥18 years who had registered with an eligible general practice for a minimum of 12 months, with a record of T2D (using Read codes in Table S1 or Chapter 6.1 of the British National Formulary), and received 2 or more consecutive prescriptions for SGLT2i drug, were eligible for inclusion. Prescriptions of SGLT2i, GLP1RA, and DPP4i were identified using drug codes (Table S1). Eligible patients were categorized into the “Switch” group if they had initiated prescriptions for index incretin‐based drugs, either GLP1RA or DPP4i drug, but discontinued that of SGLT2i, defined by either the absence of ongoing refills or a gap of 60 days; or “Add‐on” group if they had received prescriptions for incretin‐based drugs while not discontinuing that of background SGLT2i. Patients in the “Add‐on” group with overlapping duration of 2 drug classes of <60 days were excluded. The date of initiating incretin‐based drugs was considered the index date (baseline).

Follow‐Up Period

Participants were followed up from the index date until the earliest of the following occurrences: outcome diagnosis, death, participant left the practice, practice ceased to contribute to the database, or the end of study (June 30, 2020).

Baseline Covariates

Baseline covariates of patients included age, sex, smoking status, drinking status, duration of T2D, duration of SGLT2i prescription, anthropometric and clinical measurements, laboratory readings, drug prescription within 1 year, and comorbidity status at baseline. Baseline body mass index, fasting glucose, HbA1c, average systolic blood pressure (SBP) and diastolic blood pressure within 1 year before baseline, total cholesterol to high‐density lipoprotein‐cholesterol ratio, low‐density lipoprotein‐cholesterol, and triglycerides were taken from the closest reading before the index date. The estimated glomerular filtration rate (eGFR) was estimated by serum creatinine, age, and sex based on the Modification of Diet in Renal Disease Study formula. Use of insulin, oral antidiabetic drugs (metformin, sulfonylureas, and thiazolidinediones), antihypertensive drugs (in particularly angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers), lipid‐lowering agents, antiplatelets, and anticoagulants at baseline were identified using the prescription records within 1‐year window before the index date. Past medical records of bariatric surgery were also extracted. Presence of any CVD, heart failure (HF), atrial fibrillation, hypertension, CKD, end‐stage kidney disease (ESKD), diabetic retinopathy, peripheral neuropathy, mental or psychiatric disorder, and cancer were documented at baseline, as well as the comorbidity status determined by Charlson Comorbidity Index. The occurrence of hypoglycemia and ketoacidosis within 1 year before the index date was also recorded.

Outcome Measures

Study outcomes were all‐cause mortality, CVD (composite of coronary heart disease, acute myocardial infarction, other ischemic heart disease, HF, stroke, transient ischemic attack, and peripheral vascular disease), HF (an outcome of interest with SGLT2i use), CKD, ESKD, hypoglycemia, and ketoacidosis by treatment groups. Outcome events and comorbidities were identified by Read Codes (Table S1). The diagnosis of CKD was identified by relevant Read Codes, 2 consecutive measurements of eGFR <60 mL/min per 1.73 m2, or 2 consecutive measurements of urine albumin‐creatinine ratio ≥30 mg/g ; and ESKD by recorded eGFR of <15 mL/min per 1.73 m2. Secondary outcomes were changes in anthropometric (SBP, diastolic blood pressure, body mass index, percentage total weight loss) and metabolic (HbA1c, low‐density lipoprotein‐cholesterol, total cholesterol/high‐density lipoprotein‐cholesterol, triglycerides, eGFR) parameters from baseline to 12‐month follow‐up (the assessment closest to 12‐month follow‐up over the period of 6–18 months).

Statistical Analysis

To account for incomplete baseline data, multiple imputation by chained equations was performed. Each missing baseline datum was imputed 5 times by random chained equation using other known baseline covariates. Five complete imputed data sets were analyzed individually to generate model estimates, which were then pooled into to a single estimate using Rubin’s rules.

For confounding adjustment, inverse probability of treatment weights (IPTW) using the propensity score was applied to balance covariates across 2 treatment groups. Logistic regression models were fitted by using the indicator variables of treatment group as the dependent variable and baseline covariates as independent variables. The predicted probability of receiving treatment based on the patient’s baseline covariates in the model is called propensity score. Patients with similar propensity scores were classified as having similar characteristics. We applied IPTW based on the propensity scores. Propensity score weights <1st percentile or ˃99th percentile in each group were trimmed. In the context of IPTW, multiple imputation followed by pooling treatment effect estimates across imputed data sets is the preferred approach. Balance of baseline covariates between groups were assessed using the standardized mean difference, with a value of ˂0.1 indicating balance.

Number of outcome events, person‐years, and incidence rate with 95% PoissonCI for each treatment group were calculated. Cox proportional hazards regression model was used to examine the association between treatment groups and incidence of events, and estimate hazard ratios (HR) of treatment effects and their 95% CI. Proportional hazard assumption was tested by Schoenfeld residuals with P values adjusted by Bonferroni method.

Secondary outcomes were compared between baseline and 12‐month follow‐up by paired t test within the same treatment group. Effects of switching from SGLT2i (dapagliflozin or empagliflozin) to either GLP1RA (exenatide or liraglutide) or DPP4i (sitagliptin, linagliptin, or alogliptin) were assessed, whereas the effects of initiating GLP1RA or DPP4i in addition to SGLT2i were investigated within the Add‐on group.

Subgroup analyses were conducted based on incretin‐based drug class (GLP1RA or DPP4i); stratification of baseline HbA1c (≤9% versus >9%); any prescription records of insulin, metformin, or sulfonylureas within 1 year before baseline; and types of SGTL2i (dapagliflozin or empagliflozin), GLP1RA (exenatide or liraglutide), and DPP4i (sitagliptin, linagliptin, or alogliptin) used (which were administered by >20% of patients). In sensitivity analyses, different scenarios were tested to assess the robustness of treatment effects, including (1) “as‐treated” analysis to censor the follow‐up period at the discontinuation of incretin‐based drugs, subsequent switch from GLP1RA to DPP4i, or switch from DPP4i to GLP1RA; (2) competing risk analysis accounting for competing risk of death; (3) multiple imputation of missing baseline covariates without IPTW; and (4) complete‐case with IPTW.

All statistical analyses were performed using Stata version 16.0 (StataCorp LP, College Station, Texas). All significance tests were 2‐tailed and P values of ˂0.05 were taken to indicate statistical significance.

Ethical Approval

Use of the IMRD database has been approved by the NHS Health Research Authority (NHS Research Ethics Committee reference: 18/LO/0441); in accordance with this approval, the study protocol was reviewed and approved by an independent Scientific Review Committee (reference number: 20SRC070). This study used de‐identified data provided by patients as part of their routine primary care, and no informed consent was required for this study.

Results

Among 31 171 adults with T2D receiving 2 or more consecutive prescription records of SGLT2i, a total of 2888 patients had initiated incretin‐based drugs and received 2 or more consecutive prescription records of GLP1RA or DPP4i on or after January 1, 2005, of whom 1461 were switched from SGLT2i to incretin‐based drugs (Switch group: GLP1RA n=412; DPP4i n=1049), while 1427 were prescribed with a combination of SGLT2i and incretin‐based drugs (Add‐on group: GLP1RA n=409; DPP4i n=1018) (Figure 1). Background SGLT2i therapy had been initiated for a mean of 1.4 (SD 1.1) years at baseline (Table 1). The 3 types of SGLT2i used were dapagliflozin (60.2%), empagliflozin (27.7%), and canagliflozin (12.1%). Over half (52.6%) of the patients used exenatide for GLP1RA initiation, followed by liraglutide (32.3%), dulaglutide (10.7%), and lixisenatide (4.4%). For patients initiating DPP4i, 39.2% used sitagliptin, 25.0% used linagliptin, 24.6% used alogliptin, 10.8% used saxagliptin, and 0.3% used vildagliptin. Baseline characteristics of patients in the 2 treatment groups after multiple imputation and weighting are listed in Table 1. Overall, the mean age of this cohort was 57.9 (SD 11.2) years, with baseline HbA1c of 9.0% (1.5%), duration of T2D for 8.7 (6.4) years, and Charlson Comorbidity Index of 4.1 (1.9). Demographic and clinical characteristics of patients were balanced between groups. Data completion rates of baseline covariates are detailed in Table S2.

Figure 1

Flowchart of identifying eligible patients with type 2 diabetes who had initiated incretin‐based drugs as substitution (“Switch”) or add‐on (“Add‐on”) to background SGLT2i therapy.

DPP4i indicates dipeptidyl peptidase‐4 inhibitors; GLP1RA, glucagon‐like peptide‐1 receptor agonists; and SGLT2i, sodium‐glucose cotransporter‐2 inhibitors.

Table 1

Baseline Characteristics of Patients With Type 2 Diabetes Who Initiated Incretin‐Based Drugs as Substitution (“Switch”) or Add‐on to SGLT2i Before and After Propensity Score Weighting

	Before weighting				After weighting
Baseline characteristics	Total (N=2888)	Switch (N=1461)	Add‐on (N=1427)	SMD	SMD
Socio‐demographics
Sex (%)				0.15	0.01
Female	46.3%	50.0%	42.5%
Male	53.7%	50.0%	57.5%
Age (mean±SD), y	57.9 (11.2)	58.8 (11.6)	57.0 (10.8)	0.16	0.03
Clinical characteristics (mean±SD)
SBP, mm Hg	131.6 (13.9)	132.1 (13.8)	131.1 (14.1)	0.07	0.00
DBP, mm Hg	77.9 (9.0)	77.8 (8.8)	78.0 (9.3)	0.02	0.00
BMI, kg/m2	34.7 (7.0)	34.8 (7.0)	34.5 (7.0)	0.03	0.01
<25	4.9%	5.3%	4.5%	0.07	0.08
25 to <30	22.4%	21.2%	23.7%
30 to <35	28.8%	28.5%	29.0%
≥35	43.9%	45.0%	42.8%
Weight, kg	99.1 (21.9)	98.7 (22.0)	99.5 (21.7)	0.03	0.01
TC, mmol/L	4.5 (1.2)	4.5 (1.1)	4.5 (1.2)	0.01	0.02
LDL‐C, mmol/L	2.7 (1.2)	2.7 (1.2)	2.8 (1.1)	0.04	0.02
TC/HDL‐C ratio	4.2 (1.5)	4.2 (1.5)	4.2 (1.5)	0.01	0.00
Triglyceride, mmol/L	2.7 (2.0)	2.6 (1.9)	2.7 (2.1)	0.04	0.03
Fasting glucose, mmol/L	11.1 (4.8)	11.1 (4.9)	11.1 (4.8)	0.00	0.01
HbA1c, %	9.0 (1.5)	9.0 (1.6)	9.0 (1.4)	0.02	0.00
≤7	3.3%	3.8%	2.7%	0.07	0.05
>7 to 9	54.4%	53.5%	55.4%
>9	42.3%	42.7%	41.9%
Creatinine (serum), µmol/L	74.7 (20.4)	75.5 (23.8)	73.8 (16.3)	0.08	0.06
eGFR, mL/min per 1.73 m2	114.1 (29.6)	112.3 (30.4)	116.0 (28.7)	0.12	0.01
Urine ACR, mg/g	58.2 (257.5)	64.4 (303.9)	51.5 (195.7)	0.05	0.00
Lifestyle factors (%)
Smoking status				0.03	0.06
Nonsmoker	47.8%	47.6%	47.9%
Current smoker	16.6%	16.2%	17.1%
Ex‐smoker	35.6%	36.1%	35.0%
Drinking status				0.04	0.02
Nondrinker	26.2%	26.9%	25.5%
Current drinker	67.6%	66.7%	68.4%
Ex‐drinker	6.2%	6.3%	6.1%
Comorbidity status (%)
Cardiovascular diseases	19.0%	20.5%	17.4%	0.08	0.02
Heart failure	2.5%	2.9%	2.1%	0.05	0.02
Atrial fibrillation	4.7%	5.9%	3.6%	0.11	0.01
Hypertension	59.0%	60.3%	57.7%	0.05	0.01
Chronic kidney disease	19.6%	21.8%	17.4%	0.11	0.02
End‐stage kidney disease	0.1%	0.1%	0.1%	0.02	0.01
Diabetic retinopathy	20.7%	19.7%	21.7%	0.05	0.00
Peripheral neuropathy	10.2%	11.6%	8.8%	0.09	0.01
Mental or psychiatric disorder	19.2%	19.6%	18.9%	0.02	0.02
Cancer	5.5%	6.0%	4.9%	0.05	0.00
Hypoglycemia within 1 y	1.0%	1.2%	0.8%	0.05	0.00
Ketoacidosis within 1 y	0.1%	0.1%	0.1%	0.02	0.01
Charlson comorbidity index*	4.1 (1.9)	4.3 (2.0)	3.9 (1.8)	0.20	0.03
Charlson comorbidity index*, (%)				0.18	0.10
1–2	19.3%	18.5%	20.0%
3	24.4%	20.9%	27.9%
4 or above	56.4%	60.5%	52.1%
Duration of type 2 diabetes, y	8.7 (6.4)	8.8 (6.6)	8.6 (6.1)	0.03	0.00
Treatment use within 1 y (%)
Insulin	57.3%	61.3%	53.1%	0.17	0.02
Basal insulin	11.3%	13.3%	9.1%	0.13	0.10
Oral antidiabetic drugs
Metformin	91.9%	92.1%	91.6%	0.02	0.00
SU	45.9%	50.8%	40.9%	0.20	0.01
TZD	8.3%	9.7%	6.9%	0.10	0.01
Antihypertensive drugs	75.8%	76.5%	75.1%	0.03	0.00
ACEI/ARB	64.7%	65.0%	64.4%	0.01	0.00
Lipid‐lowering drugs	84.0%	82.8%	85.4%	0.07	0.01
Antiplatelet drugs	28.9%	29.6%	28.2%	0.03	0.00
Anticoagulant	7.9%	9.8%	5.9%	0.15	0.03
Bariatric surgery	0.5%	0.4%	0.5%	0.01	0.02
Duration of SGLT2i, y	1.4 (1.1)	1.3 (1.1)	1.5 (1.2)	0.14	0.02
Drug type (%)
SGLT2i				0.12	0.03
Canagliflozin	12.1%	14.0%	10.2%
Dapagliflozin (Propanediol)	60.2%	58.8%	61.6%
Empagliflozin	27.7%	27.2%	28.2%
GLP1RA				0.28	0.04
Exenatide	52.6%	48.8%	56.5%
Dulaglutide	10.7%	14.8%	6.6%
Liraglutide	32.3%	32.5%	32.0%
Lixisenatide	4.4%	3.9%	4.9%
DPP4i				0.10	0.03
Sitagliptin	39.2%	39.5%	39.0%
Vildagliptin	0.3%	0.6%	0.1%
Saxagliptin	10.8%	11.0%	10.5%
Linagliptin	25.0%	25.4%	24.7%
Alogliptin	24.6%	23.6%	25.7%

ACEI indicates angiotensin‐converting enzyme inhibitor; ARB, angiotensin receptor blockers; BMI, body mass index; DBP, diastolic blood pressure; DPP4i, dipeptidyl peptidase‐4 inhibitor; eGFR, estimated glomerular filtration rate; GLP‐1RA, glucagon‐like peptide‐1 receptor agonists; HbA1c, hemoglobin A1c; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; SBP, systolic blood pressure; SGLT2i, sodium glucose cotransporter‐2 inhibitors; SMD, standardized mean difference; SU, sulfonylureas; TZD, thiazolidinedione; and Urine ACR, urine albumin‐to‐creatinine ratio.

The calculation of Charlson Comorbidity Index does not include acquired immune deficiency syndrome (AIDS).

The median follow‐up period of patients in Switch and Add‐on groups were 19.2 (interquartile range, 9.1–34.6) and 17.0 (8.0–28.5) months, respectively (Table 2). After weighting, incidence rate of all‐cause mortality during follow‐up was 11.82 and 12.57 per 1000 person‐years among Switch and Add‐on users, respectively. Overall, there were no significant difference in risks of all‐cause mortality (HR, 0.908 [95% CI, 0.541–1.523]; P=0.713), CVD (HR, 0.746 [95% CI, [0.464–1.198]; P=0.225), HF (HR, 1.238 [95% CI, 0.501–3.058]; P=0.644), CKD (HR, 1.128 [95% CI, 0.761–1.670]; P=0.549); ESKD (HR, 1.942 [95% CI, 0.205–18.433]; P=0.563), hypoglycemia (HR, 1.180 [95% CI, 0.595–2.342]; P=0.636), and ketoacidosis (HR, 0.854 [95% CI, 0.113–6.480]; P=0.879) between treatment groups (Table 3). Similar risks of outcome events were observed between the 2 groups across subgroup and sensitivity analyses (Tables S3 and S4, respectively). Test for proportional hazard assumption by Schoenfeld residuals showed there is no evidence that the proportional hazard assumption has been violated.

Table 2

Number and Incidence Rate of All‐Cause Mortality, Cardiovascular Diseases, Heart Failure, Chronic Kidney Disease, End‐Stage Kidney Disease, Hypoglycemia, and Ketoacidosis Events

Events	Before weighting							After weighting
	Cumulative incidence		Crude incidence rate (Cases / 1000 person‐y)			Median follow‐up periods (Months)	Mean follow‐up periods (Months)	Incidence rate (Cases/1000 person‐y)
	Cases with event	Rate	Estimate	95% CI*	Person‐y			Estimate	95% CI*
Total (N=2888)
All‐cause mortality	64	2.22%	12.35	(9.51, 15.77)	5183	18	22	12.20	(10.15, 14.48)
Cardiovascular diseases	75	3.21%	18.43	(14.49, 23.10)	4070	17	21	19.53	(16.58, 22.74)
Heart failure	21	0.75%	4.17	(2.58, 6.37)	5041	18	21	4.11	(2.92, 5.51)
Chronic kidney disease	112	4.83%	28.13	(23.16, 33.85)	3981	17	21	27.39	(23.89, 31.23)
End‐stage kidney disease	4	0.14%	0.77	(0.21, 1.98)	5170	18	22	0.76	(0.30, 1.44)
Hypoglycemia	38	1.33%	7.47	(5.28, 10.25)	5089	18	21	7.81	(6.17, 9.68)
Ketoacidosis	4	0.14%	0.77	(0.21, 1.98)	5173	18	22	0.75	(0.30, 1.44)
Switch (N=1461)
All‐cause mortality	36	2.46%	12.90	(9.04, 17.87)	2790	19	23	11.82	(9.02, 15.08)
Cardiovascular diseases	37	3.19%	17.06	(12.02, 23.52)	2168	19	22	17.04	(13.28, 21.45)
Heart failure	13	0.92%	4.82	(2.57, 8.24)	2699	19	23	4.55	(2.85, 6.74)
Chronic kidney disease	64	5.60%	31.04	(23.90, 39.63)	2062	17	22	28.95	(23.93, 34.70)
End‐stage kidney disease	3	0.21%	1.08	(0.22, 3.15)	2779	19	23	0.98	(0.31, 2.22)
Hypoglycemia	23	1.59%	8.43	(5.35, 12.65)	2727	19	23	8.41	(6.03, 11.22)
Ketoacidosis	2	0.14%	0.72	(0.09, 2.59)	2786	19	23	0.73	(0.16, 1.81)
Add‐on (N=1427)
All‐cause mortality	28	1.96%	11.70	(7.77, 16.91)	2393	17	20	12.57	(9.64, 15.92)
Cardiovascular diseases	38	3.23%	19.98	(14.14, 27.42)	1902	16	19	22.10	(17.70, 27.09)
Heart failure	8	0.57%	3.42	(1.47, 6.73)	2342	17	20	3.67	(2.17, 5.71)
Chronic kidney disease	48	4.07%	25.01	(18.44, 33.16)	1919	16	20	25.85	(21.14, 31.26)
End‐stage kidney disease	1	0.07%	0.42	(0.01, 2.33)	2391	17	20	0.53	(0.08, 1.54)
Hypoglycemia	15	1.06%	6.35	(3.55, 10.47)	2362	17	20	7.20	(5.01, 9.85)
Ketoacidosis	2	0.14%	0.84	(0.10, 3.03)	2387	17	20	0.77	(0.16, 1.83)

DPP4i indicates dipeptidyl peptidase‐4 inhibitor; GLP‐1RA, glucagon‐like peptide‐1 receptor agonists; and SGLT2i, sodium glucose cotransporter‐2 inhibitors.

95% CI of incidence rates were constructed by Poisson distribution.

Table 3

HR of All‐cause Mortality, Cardiovascular Diseases, Heart Failure, Chronic Kidney Disease, End‐Stage Kidney Disease, Hypoglycemia, and Ketoacidosis Events

Events	Switch vs Add‐on
	HR	95% CI	P value
All‐cause mortality	0.908	(0.541–1.523)	0.713
Cardiovascular disease	0.746	(0.464–1.198)	0.225
Heart failure	1.238	(0.501–3.058)	0.644
Chronic kidney disease	1.128	(0.761–1.670)	0.549
End‐stage kidney disease	1.942	(0.205–18.433)	0.563
Hypoglycemia	1.180	(0.595–2.342)	0.636
Ketoacidosis	0.854	(0.113–6.480)	0.879

HR indicates hazard ratio.

Changes in anthropometric and laboratory parameters from baseline to 12‐month follow‐up were also compared within each treatment group (Figure 2) and by differences between the 2 groups (Figure S1). A significantly greater reduction in mean HbA1c (−0.7% versus −0.5%, P<0.001) was observed in the Add‐on group compared with the Switch group, which were also evident among DPP4i users. When stratified by glycemic control at baseline, considerably larger decreases in HbA1c were noted at 12‐month follow‐up among patients with baseline level of ˃9% than those with ≤9%. In addition, patients in the Add‐on group managed to achieve greater mean reduction in weight (−2.4 versus −0.7 kg, P<0.001) and percentage total weight loss (2.2% versus 0.5%, P<0.001) than those in the Switch group, regardless of the incretin‐based drug class. A significantly larger decrease in body mass index (−0.8 versus −0.2 kg/m2, P<0.001) was evident among Add‐on versus Switch users, particularly with DPP4i. While within‐group changes in SBP were statistically insignificant, a trend towards BP lowering among patients in the Add‐on group resulted in a significant difference from those in the Switch group (−1.1 versus 0.5 mm Hg, P=0.047). Notably, a larger decrease in total cholesterol/high‐density lipoprotein‐cholesterol ratio was only significant among DPP4i users of Add‐on versus Switch treatment groups. Overall, there were no significant differences in 12‐month changes of DBP, low‐density lipoprotein‐cholesterol, triglycerides, and eGFR between the Switch and Add‐on groups.

Figure 2

Mean and 95% CI of 12‐month changes in anthropometric and laboratory parameters of patients with type 2 diabetes who had initiated incretin‐based drugs as substitution (“Switch”) or add‐on (“Add‐on”) to background SGLT2i therapy.

%WL indicates percentage weight loss; BMI, body mass index; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate; HbA1c, glycated hemoglobin; HDL‐C, high‐density lipoprotein‐cholesterol; LDL‐C, low‐density lipoprotein‐cholesterol; SBP, systolic blood pressure; SGLT2i, sodium‐glucose cotransporter 2 inhibitors; and TC, total cholesterol. *Significant difference (P<0.05) in mean of change from baseline to 12‐month follow‐up.

Discussion

In this cohort of patients with T2D with inadequate glycemic control despite being on a background glucose‐lowering therapy of SGLT2i and other antidiabetic agents, no significant differences in the risks of all‐cause mortality, cardiorenal outcomes, and other clinical end points were identified between the initiation of incretin‐based drugs as substitution or addition to the existing drug regimen. Nevertheless, treatment modification with the stepwise combination approach (add‐on) resulted in significant improvements of several metabolic parameters over 12‐month follow‐up compared with replacing SGLT2i with another new drug class (switch).

To our knowledge, the study design of this “new user” retrospective cohort analysis is unique in terms of comparing multiple clinical end points and metabolic changes with respect to the adjustment of treatment modalities and the selection of newer antidiabetic agents (namely, SGLT2i and incretin‐based drugs). The current literature is limited and inconclusive on any additional benefits of combining SGLT2i with incretin‐based drugs in reducing the macrovascular and microvascular complications of diabetes. While a post hoc analysis of DECLARE‐TIMI 58 concluded that the addition of dapagliflozin to baseline use of GLP1RA could lower the risks of hospitalization for heart failure and a composite of cardiovascular mortality and hospitalization for heart failure versus placebo, another post hoc analysis of EXSCEL could only observe significant risk reduction in all‐cause and cardiovascular death with the combination of exenatide plus SGLT2i versus either placebo or exenatide alone, alongside a trend towards reducing the risk of major adverse cardiovascular events. , Regarding specific renal outcomes (composite of eGFR reduction, ESKD, or renal death; and new‐onset albuminuria), the former study also demonstrated a trend towards benefit for the addition of dapagliflozin versus placebo to baseline DPP4i or GLP1RA therapy. Similarly, using sulfonylureas as an active comparator, an observational cohort study of propensity score‐matched patients with T2D found that adding SGLT2i to background GLP1RA therapy could lower the risks of composite cardiovascular outcomes and hospitalization for heart failure.

Contrary to the few existing studies that explored the cardiorenal outcomes and mortality of SGLT2i and incretin‐based drug combination relative to placebo, either treatment alone, or an active comparator, this study focused on evaluating these effects on new users of GLP1RA or DPP4i who had received SGLT2i therapy for a mean of 1.4 years, and attempted to answer the intriguing question of whether switching to another new drug class or adding it to the existing drug regimen would influence patient outcomes in real‐world clinical practice. This research question is of clinical relevance because patient adherence could be affected by factors including pill burden, treatment complexity, and medication cost; whereas a combination of antidiabetic agents with distinct mechanisms of action could potentially offer additional benefits to glycemic and metabolic control by targeting different pathophysiological defects of T2D, , , which remains to be proven and justified. While no significant differences in the risks of developing various clinical end points between switching and add‐on could be identified in the current study, they should be interpreted with caution given the relatively short follow‐up period and hence the small number of events that occurred.

In theory, the combination of SGLT2i with incretin‐based drugs could exert complementary actions on cardiorenal protection and ameliorating adverse effects, with SGLT2i mainly lowering the risks of HF and diabetic nephropathy via hemodynamic benefits, GLP1RA acting to reduce major adverse cardiovascular events with anti‐atherogenic and anti‐inflammatory properties, and DPP4i attenuating the elevated risk of genital infections associated with SGLT2i use through modulating the immune system. , , , , Furthermore, SGLT2i may compensate for the possible negative actions of GLP1RA and potential risk of specific DPP4i in HF progression, while incretin‐based drugs may alleviate the development of ketoacidosis associated with SGLT2i use by counteracting its increased glucagon secretion and subsequent ketogenesis. , , , Nevertheless, it has also been proposed that the production of ketone bodies induced by SGLT2i may partly be responsible for its decrease in cardiac and renal workload, and hence the observed clinical benefits; therefore, any complementary effects of SGLT2i and incretin‐based drug combination may depend on the degree of glucagon suppression, duration of pharmacological treatment, and any changes in drug efficacy over time.

Regarding the choice of treatment modality, our results were consistent with that of the retrospective cohort study utilizing the UK CPRD, demonstrating that the add‐on approach could achieve HbA1c reduction substantially larger than that of switching therapy, when patients were showing limited response to the original drug regimen ; however, changes in other anthropometric and metabolic parameters have not been compared between the 2 treatment approaches. This study suggested that, in addition to better glycemic control, the stepwise combination (add‐on) therapy could produce reduction in weight and SBP significantly larger than that of substituting SGLT2i with incretin‐based drugs over 12‐month follow‐up, which were generally in line with several clinical trials observing greater improvements with the addition of GLP1RA or DPP4i to SGLT2i versus placebo add‐on or either drug class alone. , , , , , While these studies would be classified as the comparison between “adding a new drug class” and “continuing the original therapy,” our study provided further evidence to support the use of “combination therapy” (add‐on) over “replacing SGLT2i with incretin‐based drugs” (switching) in terms of metabolic changes.

With reference to the pharmacological profile of these 3 drug classes, it can be postulated that GLP1RA would exert compensatory effects on the increased glucagon level and endogenous glucose production of SGLT2i to further reduce the HbA1c level, promote additive weight loss via the suppression of appetite to counteract the reported increase in food intake associated with SGLT2i use, and produce a synergistic effect on BP lowering with vasodilation and mild natriuresis that are distinct from SGLT2i‐induced natriuresis and reduction of intravascular volume. , , , Notably, reduction in HbA1c has also been consistently shown to be sub‐additive with the combination of SGLT2i and incretin‐based drugs versus either treatment alone, which could be attributed to the interference of drugs combined and the failure of GLP1RA or DPP4i in adequately blocking the elevated endogenous glucose production of SGLT2i, especially at higher HbA1c levels. , , , , , Yet, our results reinforced the proposition that add‐on or combination therapy would facilitate better glycemic control, even when compared with switching from a drug class with “limited response” to another with different mechanisms of action.

Concerning the initiation of DPP4i to existing SGLT2i therapy, our study revealed that the add‐on approach could result in significantly larger reduction in HbA1c, weight, and total cholesterol/high‐density lipoprotein‐cholesterol ratio than that of substitution or switching. While some studies argued that beyond glycemic control, the addition of DPP4i to SGLT2i might not confer any additional benefits on weight loss, lowering BP, or improving the lipid profile compared with SGLT2i alone, , , , our study suggested that the combination therapy would be preferred to discontinuing SGLT2i and replacing it with DPP4i. Consistent with the fact that DPP4i is weight neutral and generally less potent than GLP1RA (including the suppression of endogenous glucose production), initiation of the latter could produce more clinically relevant reduction in HbA1c, weight, and BP. , , , , Nonetheless, DPP4i may still offer renal benefits in terms of decreasing albuminuria, and can be an alternative to patients preferring an oral route of administration.

Utilizing the IMRD representative of the United Kingdom population, this study attempted to evaluate the clinical and metabolic outcomes of patients with T2D initiating incretin‐based drugs as substitution for (switching) or in combination with (add‐on) background SGLT2i therapy in the real‐world setting. Various baseline characteristics of patients had been taken into account, which were further adjusted with multiple imputations and propensity score weighting to balance the confounding factors between groups. Despite such unique study design in addressing the clinical question of whether switching or add‐on would be the preferred treatment approach, and the focus on newer antidiabetic agents with demonstrated cardiorenal safety or benefits, several limitations of this study should be acknowledged. First, given that SGLT2i is a relatively new drug class approved for T2D management, the follow‐up period of new users of incretin‐based drugs who had been on previous SGLT2i therapy would be fairly short, and hence the small number of events occurred over a median of 18 months. This could limit the interpretation of our results, because differences in cardiovascular or renal events might not be evident within this short observation period. Accordingly, our study might be underpowered to draw definite conclusions about cardiorenal outcomes, in addition to our limited sample size. Second, this patient cohort had relatively poor glycemic (mean HbA1c 9.0%) and metabolic control at baseline; thus the current findings might not be generalizable to other patient populations with different clinical characteristics. Furthermore, this patient cohort had a mean duration of diabetes of 8.7 years and were prescribed various glucose‐lowering medications within 1 year at baseline; hence the results would not be applicable to patients with T2D at an earlier stage of the disease. Third, over half of the GLP1RA users in this cohort were prescribed exenatide, which is not associated with cardio‐ or renoprotective effects, while none were given semaglutide, which is associated with reduction in major adverse cardiovascular events, stroke, composite renal outcome, and mortality. Such drug type distribution might have influenced our results. Fourth, biological mechanisms of the greater metabolic benefits observed with the add‐on approach versus switching therapy remain to be elucidated. Some unmeasured confounding factors might have also played a role in the significant differences, such as more intensive therapy and lifestyle management of the metabolic risk factors in patients managed by physicians pursuing the add‐on approach. Lastly, cost‐effectiveness of different treatment modalities and quality of life indices of patients were not evaluated in the current study, which would also be relevant in the decision‐making process.

Conclusions

In this patient cohort with T2D with inadequate glycemic control on background SGLT2i therapy, no significant differences in the risks of developing various clinical end points could be identified in the initiation of incretin‐based drugs as substitution (switching) or add‐on to the existing drug regimen. Meanwhile, several metabolic benefits of the combination approach were significantly greater than that of switching, including the reduction of HbA1c, weight, and SBP over 12‐month follow‐up. Further studies with longer observation periods and randomized controlled trials are needed to clarify the risks and benefits of the 2 treatment modalities.

Sources of Funding

None.

Disclosures

No financial relationships exist with any organizations that might have an interest in the submitted work in the previous 2 years. No other relationships or activities that could appear to have influenced the submitted work exist. CKHW reports receipt of research funding from the EuroQoL Group Research Foundation, the Hong Kong Research Grants Council, and the Hong Kong Health and Medical Research Fund. KKCM reports receipt of CW Maplethorpe Fellowship and received personal fees from IQVIA Holdings, Inc., unrelated to this work. CSLC has received grants from the Food and Health Bureau of the Hong Kong Government, Hong Kong Research Grant Council, Hong Kong Innovation and Technology Commission, Pfizer, IQVIA, and Amgen; personal fee from Primevigilance Ltd.; outside the submitted work. ICKW reports receipt of research funding from Wellcome Trust, United Kingdom; National Natural Science Fund of China, China; The Hong Kong Research Grants Council, The Research Fund Secretariat of the Food and Health Bureau, Narcotics Division of the Security Bureau of HKSAR, Hong Kong; Bristol‐Myers Squibb, Pfizer, Bayer, and Janssen, a Division of Johnson & Johnson Takeda. ICKW also reports research funding outside the submitted work from Amgen, GSK, Novartis, and the Hong Kong Health and Medical Research Fund, National Institute for Health Research in England, European Commission, National Health and Medical Research Council in Australia, and also received speaker fees from Janssen and Medice in the previous 3 years. The remaining authors have no disclosures to report.

Data S1

Tables S1–S4

Figure S1

Click here for additional data file.

Click here for additional data file.