Comparative effectiveness of sodium-glucose co-transporter 2 inhibitors and dipeptidyl peptidase-4 inhibitors on cardiorenal function and treatment adherence: a prevalent new-user design study in tertiary hospitals

Abstract

Objectives

Data on the long-term effects comparing sodium-glucose co-transporter 2 inhibitors (SGLT2i) and dipeptidyl peptidase-4 inhibitors (DPP4i) are scarce, especially from middle-income countries. To examine the effects of SGLT2i and DPP4i on the cardiorenal function and treatment adherence for people with type 2 diabetes (T2D) using prevalent new-user design in real-world setting.

Methods

We conducted a retrospective cohort study in two tertiary hospitals in Malaysia and matched T2D patients initiated on SGLT2i or DPP4i from 2010 to 2021 using time-conditional propensity score. Outcomes of interest included cardiovascular and renal outcomes, as well as clinical lab outcomes, adherence and non-persistence. The hazard ratios for cardiorenal outcomes were inferred using Cox proportional hazards model.

Key findings

The cohort included 1528 patients, with 406 SGLT2i users matched with 406 DPP4i users. Over a median follow-up of 1.52 years, no differences in cardiorenal outcomes were observed. Patients initiated with SGLT2i had lower HbA1c at 12 months (−0.79%, P < 0.001) compared with DPP4i (−0.49%, P < 0.05; difference: −0.30%, P < 0.05). No differences in the renal, lipid, weight and blood pressure parameters were observed between both groups. Higher medication persistence was noted among SGLT2i users compared with DPP4i users (92% vs 87%, P = 0.03).

Conclusions

Both medications were comparable in exerting distinct effects on cardiorenal risk factors, with better HbA1c control and medication persistence among SGLT2i users. The long-term cardiorenal outcomes remain undetermined.

dipeptidyl peptidase-4 inhibitors (DPP4i), sodium-glucose co-transporter 2 inhibitors (SGLT2i), cardiorenal outcomes, glycaemic outcome, lipid outcome, type 2 diabetes

Introduction

Type 2 diabetes is a chronic progressive disease characterised by an increase in blood glucose levels. It is estimated that 537 million people living with diabetes in 2021, with the prevalence increasing more rapidly in low- and middle-income countries.^[1] As people with type 2 diabetes are associated with a higher risk of developing severe macrovascular and microvascular complications, optimal glycaemic management is essential to avoid downstream health and economic consequences. While dietary and lifestyle changes are the cornerstone for treatment, the majority of people with type 2 diabetes often require pharmacological treatment to achieve the desired glycaemic targets.

Recently, several glucose-lowering agents including dipeptidyl peptidase-4 inhibitors (DPP4i), glucagon-like peptide-1 receptor agonist (GLP1 RA) and sodium-glucose co-transporter 2 inhibitors (SGLT2i) have been touted to be the game changer in diabetes management due to their promising benefits. DPP4i have been shown to confer cardiovascular safety while GLP1RA and SGLT2i offer cardiovascular and renal protection among people with type 2 diabetes.^[2] In particular, studies have shown that the magnitude of cardiorenal benefits was greatest for patients with a pre-existing cardiovascular history. Given these promising results, clinical practice guidelines have been updated to reflect these changes, which led to the increased use of DPP4i and SGLT2i.^[3–7] Observational studies from different countries have consistently shown the beneficial effects of SGLT2i and DPP4i in reducing the risk of cardiorenal events among patients with different risk profiles.^[8–16] Additionally, SGLT2i users have also been shown to have a comparable glycaemic and pleiotropic effect compared with DPP4i users.^[17]

Studies have shown that treatment adherence for both medications was suboptimal, despite the benefits.^{[18, 19]} Studies from high-income countries found that the adherence rate was 72% for SGLT2i users and 56.9% for DPP4i users, respectively.^{[18, 19]} Many of these studies were conducted in developed countries, with limited evidence from low- and middle-income countries due to the restricted accessibility to medications. This is true for the Malaysian public healthcare system as the high cost of the medications has limited their usage in the tertiary hospitals. Given the importance of ensuring judicious use of resources, it is important to understand how the adherence to these medications is in a local context and how these relate to the outcomes reported in clinical trials in a real-world context. In this study, we examined the effects of SGLT2i and DPP4i on clinical outcomes and treatment adherence in Malaysia, a middle-income country with a high prevalence of type 2 diabetes in a real-world setting.

Methods

Study design and setting

This retrospective study was carried out in two tertiary public hospitals located in the state of Putrajaya and Negeri Sembilan. Both hospitals run weekly diabetes clinics. Data including sociodemographic information, prescribing orders and laboratory results were extracted from the hospital medical records. This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline.

Ethics approval

The study was approved by the National Medical Research Register (NMRR-20-662-52570), Medical Research and Ethics Committee (MREC), Ministry of Health Malaysia and Monash Human Research Ethics Committee (2020-24900-45575) in April 2020.

Study cohort

We included patients who were seeking care in both hospitals from January 2010 to April 2021. The base cohort was inferred by identifying patients who had a first prescription for SGLT2i or a prescription for DPP4i throughout the inclusion period. Records of patients were excluded if a patient was: younger than 18 years; used either drugs for less than 6 months or had missing baseline characteristics such as comorbidities and medical history. As few patients only had a prior history of SGLT2i use before DPP4i, we excluded these patients from our analyses.

We conducted a prevalent new-user design to compare SGLT2i and DPP4i.^[20] Briefly, SGLT2i new users in the base cohort consisted of incident new users (no previous use of DPP4i) and prevalent new users (with previous use of DPP4i before SGLT2i use). The SGLT2i prevalent group were patients in the base cohort who received an initial DPP4i prescription and then switched/add on with SGLT2i during subsequent treatment.

Exposure set

An exposure set was created for each new SGLT2i prescription. We grouped the days since the first DPP4i dispensing into 30 days bracket for every SGLT2i prescription (for those who had previously used DPP4i) and DPP4i prescription (Supplementary Figure S1). Every exposure set included a DPP4i prescription occurring within the same time bracket as the new SGLT2i prescription, based on the duration of DPP4i use and year of dispensing. As such, the exposure set encompassed one SGLT2i new prescription and several corresponding DPP4i prescriptions in each exposure set. The corresponding DPP4i prescriptions were those that were: (1) dispensed within the same time bracket as the SGLT2i index prescription, (2) dispensed in the same year with the SGLT2i index prescription. New SGLT2i prescriptions were classified as either prevalent a new user exposure set (DPP4i prescribed before the new SGLT2i prescription) or an incident new user exposure set (no prior DPP4i prescribed to the new SGLT2i prescription). A patient can contribute DPP4i prescriptions to several exposure sets depending on the time of entry. Multiple imputations with chained equations were used to handle missing data, assuming that data were missing at random. Five imputed datasets were generated using classification and regression tree “cart” (Supplementary Table S1).

Time-conditional propensity score

Propensity scores were estimated using a conditional logistic regression model within each exposure set, with medication prescription (SGLT2i or DPP4i) being the dependent variable. The conditional logistic regression model included covariates such as demographics, comorbidities, lab values and concurrent medications at the time when initial SGLT2i and DPP4i prescriptions were being incorporated into different exposure sets. Scores were estimated independently for incident and prevalent new users in each exposure set, based on the time of inclusion. The time-conditional propensity score for every SGLT2i and DPP4i prescription incorporated in the exposure sets was computed with the estimated model coefficients.

Matching

We then matched one DPP4i prescription to a new SGLT2i prescription in the same exposure set in accordance with the nearest propensity score. In sequential order, we matched a new unique user of SGLT2i one-to-one (without replacement) to a patient using DPP4i within their exposure set using the nearest propensity score. All patients were followed up until the event of interest, which were either death, end of follow-up data or end of study period, whichever came first.

Outcomes

Cardiorenal outcomes of interest were heart failure (HF), coronary heart disease, cardiovascular events, stroke, myocardial infarction and end-stage renal disease, as charted in the medical record. Coronary heart disease included acute coronary syndrome, unstable angina, coronary artery disease, ischaemic heart disease and myocardial infarction. Cardiovascular diseases included all of the above. Changes in clinical outcomes include: glycated haemoglobin (HbA1c), estimated glomerular filtration rate (eGFR), high-density lipoprotein (HDL), low-density lipoprotein (LDL), triglycerides (TG), total cholesterol (TC), serum creatinine, weight, body mass index (BMI), systolic blood pressure (SBP), diastolic blood pressure (DBP) were recorded from baseline to 60 months. Adherence and non-persistence were recorded throughout the follow-up period. Adherence to medication was measured by the proportion of days covered (PDC) by the medication. Non-persistence was defined as the presence of a >60-day gap between the previous supply and the next supply. We have created separate cohorts for eleven clinical outcomes. Eligible patients in the cohort must have index lab values and at least one follow-up lab value.

Statistical analysis

We used Cox proportional hazards models to compute hazard ratios and 95% confidence intervals for outcomes with SGLT2i and DPP4i. For clinical outcomes, analysis of variance (ANOVA) was conducted to compare the differences in means of clinical outcomes at 6-month intervals compared with baseline, and between the treatment groups. ANOVA was used to compare mean values in treatment adherence and non-persistence. All data were reported and analysed using descriptive statistics. Categorical variables were expressed as a number and percentage, whereas continuous variables were presented as mean, standard deviation, 95% confidence intervals, as applicable. As part of our sensitivity analyses, we repeated the analysis without propensity score matching and examined the outcome independently for incident new users of SGLT2i. We also limited the period for observing adherence and non-persistence to one year. Analyses were conducted in R version 4.1.1.^[21]

Results

Patient cohort

A total of 1823 patients received SGLT2i and DPP4i during this period. A total of 1528 patients met the inclusion criteria (1113 DPP4i and 415 SGLT2i) and were included in the study (Figure 1).

Figure 1

Flowchart of patient selection. (Please see attached.) Data are presented as mean (standard deviation) or as specified. Abbreviations: DPP4i, dipeptidyl peptidase-4 inhibitors; SGLT2i, sodium-glucose co-transporter 2 inhibitors; GLP1 RA, glucagon-like peptide-1 receptor agonists; BMI, body mass index; CVD, cardiovascular disease; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate; HbA1c, glycated haemoglobin; HDL, high density lipoprotein; TC, total cholesterol; TG, triglycerides; LDL, low density lipoprotein; SBP, systolic blood pressure; SMD, standardized mean difference.

After matching within exposure sets, 406 SGLT2i users were matched to 406 DPP4i users, with 199 SGLT2i prevalent new users and 207 incident new users. The matched cohort comprised 50.6% males, with a mean age of 56.1 (12.8) years and a mean of 13.6 (8.8) years of diabetes diagnosis. Almost one-third of patients have nephropathy, retinopathy, neuropathy or established cardiovascular disease at baseline. The majority (91%) of the patients were on more than one glucose-lowering therapy, the most common being metformin, followed by insulin and sulfonylureas. Baseline characteristics were balanced in the two groups, except for ethnicity (Table 1). Baseline characteristics in each group for an unmatched cohort before imputation and after imputation is shown in Supplementary Tables S2 and S3.

Table 1

Characteristics of SGLT2i and propensity score-matched DPP4i users within exposure sets

	Overall	DPP4i	SGLT2i	P-value	SMD
n	812	406	406
Year of matched prescription (%)				0.185	0.224
2014	2 (0.2)	2 (0.5)	0 (0.0)
2015	9 (1.1)	4 (1.0)	5 (1.2)
2016	3 (0.4)	2 (0.5)	1 (0.2)
2017	12 (1.5)	10 (2.5)	2 (0.5)
2018	96 (11.8)	41 (10.1)	55 (13.5)
2019	366 (45.1)	181 (44.6)	185 (45.6)
2020	316 (38.9)	162 (39.9)	154 (37.9)
2021	8 (1.0)	4 (1.0)	4 (1.0)
Demographics
Male, n (%)	411 (50.6)	203 (50.0)	208 (51.2)	0.779	0.025
Age	56.07 (12.77)	56.59 (13.75)	55.54 (11.70)	0.24	0.083
Family history of diabetes, n (%)	124 (15.3)	59 (14.5)	65 (16.0)	0.626	0.041
Diabetes duration, years	13.58 (8.76)	13.76 (9.16)	13.39 (8.34)	0.546	0.042
Smoking history, n (%)	52 (6.4)	25 (6.2)	27 (6.7)	0.886	0.02
Ethnicity, s (%)				0.004	0.257
Malay	575 (70.8)	279 (68.7)	296 (72.9)
Chinese	84 (10.3)	57 (14.0)	27 (6.7)
Indian	131 (16.1)	62 (15.3)	69 (17.0)
Others	22 (2.7)	8 (2.0)	14 (3.4)
Biochemical parameters
Serum creatinine (μmol/L)	87.31 (40.95)	87.15 (42.29)	87.46 (39.61)	0.915	0.007
BMI (kg/m2)	31.69 (7.32)	31.44 (7.55)	31.94 (7.07)	0.331	0.068
Weight (kg)	83.14 (21.41)	82.09 (21.65)	84.18 (21.14)	0.164	0.098
SBP (mmHg)	139.90 (16.61)	139.84 (16.62)	139.95 (16.62)	0.922	0.007
DBP (mmHg)	76.92 (12.70)	76.87 (14.09)	76.97 (11.16)	0.904	0.009
LDL (mmol/L)	2.66 (1.03)	2.66 (1.02)	2.67 (1.04)	0.849	0.013
HDL (mmol/L)	1.24 (0.32)	1.24 (0.30)	1.23 (0.34)	0.659	0.031
TG (mmol/L)	1.78 (1.03)	1.80 (1.06)	1.77 (0.99)	0.732	0.024
TC (mmol/L)	4.68 (1.21)	4.69 (1.18)	4.67 (1.25)	0.852	0.013
HbA1c (%)	8.59 (1.85)	8.57 (1.88)	8.61 (1.83)	0.756	0.022
HbA1c (mmol/mol)	70.4 (20.2)	70.2 (20.5)	70.6 (20.0)	0.756	0.022
eGFR (ml/min/1.73 m2)	83.67 (30.06)	83.76 (29.00)	83.58 (31.13)	0.932	0.006
Comorbidities, n (%)
Hypertension	601 (74.0)	307 (75.6)	294 (72.4)	0.337	0.073
Dyslipidaemia	574 (70.7)	290 (71.4)	284 (70.0)	0.7	0.032
Myocardial infarction	42 (5.2)	19 (4.7)	23 (5.7)	0.635	0.044
Stroke	47 (5.8)	27 (6.7)	20 (4.9)	0.367	0.074
Chronic kidney disease	160 (19.7)	81 (20.0)	79 (19.5)	0.93	0.012
Nephropathy	260 (32.0)	129 (31.8)	131 (32.3)	0.94	0.011
Retinopathy	264 (32.5)	120 (29.6)	144 (35.5)	0.085	0.126
Neuropathy	292 (36.0)	147 (36.2)	145 (35.7)	0.942	0.01
Established CVD	256 (31.5)	121 (29.8)	135 (33.3)	0.326	0.074
Medications, n (%)
Blood pressure medication	622 (76.6)	313 (77.1)	309 (76.1)	0.804	0.023
Lipid-lowering medication	637 (78.4)	321 (79.1)	316 (77.8)	0.733	0.03
Metformin	629 (77.5)	316 (77.8)	313 (77.1)	0.867	0.018
Sulfonylureas	172 (21.2)	96 (23.6)	76 (18.7)	0.103	0.121
Insulin	487 (60.0)	239 (58.9)	248 (61.1)	0.567	0.045
GLP1RA	4 (0.5)	1 (0.2)	3 (0.7)	0.616	0.07
>1 glucose-lowering therapy	739 (91.0)	369 (90.9)	370 (91.1)	1	0.009

Data are presented as mean (standard deviation) or as specified.

Abbreviations: DPP4i, dipeptidyl peptidase-4 inhibitors; SGLT2i, sodium-glucose co-transporter 2 inhibitors; GLP1 RA, glucagon-like peptide-1 receptor agonists; BMI, body mass index; CVD, cardiovascular disease; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate; HbA1c, glycated haemoglobin; HDL, high-density lipoprotein; TC, total cholesterol; TG, triglycerides; LDL, low-density lipoprotein; SBP, systolic blood pressure; SMD, standardized mean difference

Cardiorenal outcomes

Table 2 depicts the incidence rates (IRs) and hazard ratios for cardiovascular events and end-stage renal disease throughout the follow-up periods for patients using SGLT2i and DPP4i. Across a median follow-up of 1.52 years, SGLT2i users had a lower IR for heart failure (IR: 2.84 vs 6.44, per 1000 person-years) and cardiovascular disease (IR: 14.4 vs 16.3, per 1000 person-years). DPP4i users had a lower IR for coronary heart disease (IR: 4.83 vs 5.72, per 1000 person-years) and stroke (IR: 4.83 vs 5.72, per 1000 person-years). No significant difference in cardiorenal outcomes between the two groups was observed.

Table 2

Cardiorenal outcomes for matched cohort

	Total patients	Number of outcomes	Person-years in total	Incidence ratea	Hazard ratio (95% CI)
Heart failure	812	6
SGLT2i	406	2	704.47	2.84	0.45 (0.08, 2.48)
DPP4i	406	4	621.49	6.44	Reference
Coronary heart disease	812	7
SGLT2i	406	4	698.76	5.72	1.25 (0.28, 5.59)
DPP4i	406	3	621.26	4.83	Reference
Cardiovascular disease	812	20
SGLT2i	406	10	693.70	14.42	0.92 (0.38, 2.22)
DPP4i	406	10	614.83	16.26	Reference
Stroke	812	7
SGLT2i	406	4	699.52	5.72	1.23 (0.28, 5.49)
DPP4i	406	3	620.87	4.83	Reference
Myocardial infarction	812	2
SGLT2i	406	2	702.84	2.85	N/A
DPP4i	406	0	624.39	0
End-stage renal disease	812	0
SGLT2i	406	0	704.53	0	N/A
DPP4i	406	0	624.39	0

^aIncidence rate per 1000-person year.

Abbreviations: DPP4i, dipeptidyl peptidase-4 inhibitors; SGLT2i, sodium-glucose co-transporter 2 inhibitors; CI, confidence intervals; N/A: sample size or event rate too low to calculate a hazard ratio.

Clinical outcomes

Glycaemic outcomes

Both SGLT2i and DPP4i lowered HbA1c throughout the follow-up period, with a significant reduction in the first 24 months (Figure 2, Table 3, Supplementary Figure S2). The reduction in HbA1c at 6 months was −0.67% or −7.4 mmol/mol (P < 0.001) for SGLT2i users and −0.65% or −7.1 mmol/mol (P < 0.001) for DPP4i users. SGLT2i lowered HbA1c (−0.79% or −8.7 mmol/mol; P < 0.001) to a greater extent than DPP4i (−0.49% or −5.4 mmol/mol; P < 0.05; difference: −0.30% or −3.3 mmol/mol, P < 0.05) at 12 months.

Table 3

Clinical outcomes for matched cohort

	DPP4i	N	Change from baseline (95% CI)	P-value	SGLT2i	N	Change from baseline (95% CI)	P-value	Difference between SGLT2i DPP4i	Change from baseline (95% CI)	P-value
	Month				Month				Month
HbA1c (%)	6	289	−0.65 (−1.08, −0.21)	<0.01	6	322	−0.67 (−1.06, −0.29)	<0.01	6	−0.03 (−0.27, 0.22)	0.82
	12	225	−0.49 (−0.97, −0.02)	0.03	12	238	−0.8 (−1.22, −0.37)	<0.01	12	−0.3 (−0.6, 0)	<0.05
	18	119	−0.33 (−0.91, 0.26)	0.79	18	122	−0.75 (−1.28, −0.21)	<0.01	18	−0.42 (−0.81, −0.03)	0.03
	24	53	−0.96 (−1.77, −0.14)	<0.01	24	69	−0.69 (−1.36, −0.02)	0.04	24	0.27 (−0.2, 0.73)	0.26
	30	29	−0.46 (−1.53, 0.61)	0.95	30	30	−0.58 (−1.55, 0.39)	0.7	30	−0.12 (−1, 0.76)	0.79
	36	12	−0.75 (−2.37, 0.88)	0.93	36	18	−0.63 (−1.86, 0.61)	0.87	36	0.12 (−0.86, 1.1)	0.8
	42	7	−1.2 (−3.31, 0.91)	0.76	42	6	−1.3 (−3.4, 0.8)	0.65	42	−0.1 (−1.08, 0.88)	0.83
	48	4	−1.54 (−4.32, 1.25)	0.79	48	6	−1.02 (−3.12, 1.08)	0.9	48	0.52 (−0.41, 1.45)	0.23
	54	3	−1.84 (−5.04, 1.37)	0.75	54	6	−1.87 (−3.97, 0.23)	0.13	54	−0.03 (−2.65, 2.59)	0.98
	60	3	−1.65 (−4.86, 1.55)	0.85	60	5	−1.48 (−3.78, 0.81)	0.59	60	0.17 (−0.59, 0.93)	0.6
HbA1c (mmol/mol)	6	289	−7.1 (−11.8, −2.3)	<0.01	6	322	−7.4 (−11.6, −3.2)	<0.01	6	−0.4 (−3.0, 2.4)	0.82
	12	225	−5.4 (−10.6, −0.3)	0.03	12	238	−8.8 (−13.4, −4.1)	<0.01	12	−3.3 (−6.6, 0)	<0.05
	18	119	−3.6 (−10.0, 2.9)	0.79	18	122	−8.2 (−14.0, −2.3)	<0.01	18	−4.6 (−8.9, −0.4)	0.03
	24	53	−10.5 (−19.4, −1.6)	<0.01	24	69	−7.6 (−14.9, −0.3)	0.04	24	3.0 (−2.2, 8.0)	0.26
	30	29	−5.1 (−16.8, 6.7)	0.95	30	30	−6.4 (−17.0, 4.3)	0.7	30	−1.3 (−11.0, 8.3)	0.79
	36	12	−8.2 (−25.9, 9.7)	0.93	36	18	−6.9 (−20.4, 6.7)	0.87	36	1.3 (−9.4, 12.1)	0.8
	42	7	−13.1 (−36.2, 10.0)	0.76	42	6	−14.2 (−37.2, 8.8)	0.65	42	−1.1 (−11.8, 9.7)	0.83
	48	4	−16.9 (−47.2, 13.7)	0.79	48	6	−11.2 (−34.1, 11.8)	0.9	48	5.7 (−4.5, 15.9)	0.23
	54	3	−20.1 (−55.1, 15.0)	0.75	54	6	−20.5 (−43.4, 2.5)	0.13	54	−0.4 (−29.0, 28.3)	0.98
	60	3	−18.1 (−53.1, 17.0)	0.85	60	5	−16.2 (−41.3, 8.9)	0.59	60	1.9 (−6.5, 10.2)	0.6
eGFR (ml/min/1.73m2)	6	274	−3.05 (−10.52, 4.42)	0.97	6	312	−2.09 (−9.64, 5.47)	1	6	0.96 (−3.73, 5.65)	0.69
	12	210	−2.32 (−10.39, 5.76)	1	12	235	−1.19 (−9.38, 7)	1	12	1.13 (−4.44, 6.69)	0.69
	18	120	−2.61 (−12.39, 7.18)	1	18	119	−0.92 (−11.24, 9.39)	1	18	1.68 (−6.61, 9.98)	0.69
	24	49	−7.41 (−21.51, 6.69)	0.84	24	63	−3.22 (−16.53, 10.09)	1	24	4.19 (−6.69, 15.07)	0.45
	30	27	−5.91 (−24.37, 12.54)	0.99	30	28	−5.48 (−24.59, 13.63)	1	30	0.43 (−14.08, 14.95)	0.95
	36	9	−6.89 (−38.08, 24.29)	1	36	16	−1.31 (−26.19, 23.57)	1	36	5.58 (−14.61, 25.78)	0.57
	42	6	−15.1 (−53.12, 22.93)	0.97	42	6	−1.64 (−41.72, 38.43)	1	42	13.45 (−21.05, 47.96)	0.41
	48	4	3.46 (−42.98, 49.9)	1	48	6	−2.56 (−42.63, 37.51)	1	48	−6.02 (−35.49, 23.45)	0.65
	54	2	5.13 (−60.36, 70.62)	1	54	6	−11.85 (−51.92, 28.22)	1	54	−16.98 (−45.99, 12.03)	0.2
	60	2	−1.5 (−66.99, 63.99)	1	60	5	−7.55 (−51.39, 36.28)	1	60	−6.05 (−44.32, 32.22)	0.7
Adherence (PDC)	Total	406	0.97 (0.07)		Total	406	0.97 (0.07)		Total	0 (−0.01, 0.01)	0.47
Non-persistence	Total	406	0.13 (0.33)		Total	406	0.08 (0.27)		Total	−0.05 (−0.09, −0.01)	0.03

Change in mean value from baseline over follow-up period for patients treated with SGLT2i or DPP4i. Data are presented as mean (standard deviation) or as specified.

Abbreviations: DPP4i, dipeptidyl peptidase-4 inhibitors; SGLT2i, sodium-glucose co-transporter 2 inhibitors; CI, Confidence intervals; BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; LDL, low-density lipoprotein; HDL, high-density lipoprotein; TG, triglycerides; TC, total cholesterol; HbA1c, glycated haemoglobin; eGFR, estimated glomerular filtration rate

Figure 2

Change in clinical outcomes for the matched cohort over follow-up period. Abbreviations: DPP4i, dipeptidyl peptidase-4 inhibitors; SGLT2i, sodium-glucose co-transporter 2 inhibitors; Trt: Treatment; HbA1c, glycated haemoglobin; eGFR, estimated glomerular filtration rate.

Renal outcomes

There were no significant changes in eGFR across the study period for the two medications (Figure 2, Table 3, Supplementary Figure S2). There was a smaller decline in the eGFR with SGLT2i use, with a decline of 2.09 ml/min/1.73 m² in the first 6 months, which was further reduced to 1.64 ml/min/1.73 m² at 42 months. In comparison, the decline was 3.05 ml/min/1.73 m² for the first 6 months and a decline of 15.1 ml/min/1.73 m² at 42 months with DPP4i.

Other outcomes

The use of both drugs resulted in minimal change in lipid parameters throughout the study period (Supplementary Figure S2 and Table S5). Nevertheless, there was a reduction in patient weight of 5.1 kg with SGLT2i use compared with an increase of 2.6 kg among DPP4i users at 30 months. Similarly, no difference in blood pressure was observed between both groups throughout the follow-up period.

Treatment adherence

Treatment adherence was marginally higher for SGLT2i users (97.4% vs 97.1%, P = 0.47) compared with DPP4i users (Table 3). The proportion of patients with non-persistence to medication was significantly smaller for SGLT2i than DPP4i (8% vs 13%, P = 0.03) (Figure 2, Table 3).

Sensitivity analysis

Similar to the main analysis, there is no significant difference between SGLT2i users and DPP4i users for the cardiorenal outcomes in the sensitivity analysis (Supplementary Tables S6 and S8). Results of adherence and non-persistence in the first year were also similar to the main analysis (Supplementary Tables S6, S8, S10). Sensitivity analysis using the complete dataset showed a greater reduction in HbA1c for SGLT2i users compared with DPP4i users after the first 24 months of use, while LDL and TC levels were significantly lowered among DPP4i users (Supplementary Table S7). Sensitivity analysis using incident cohort showed that the reduction in HbA1c was significantly larger for both medications in the first 12 months (DPP4i: −0.77% or −8.4 mmol/mol, SGLT2i: −0.92% or −10.1 mmol/mol) (Supplementary Table S9).

Discussion

Statement of key findings

Type 2 diabetes is a progressive disease with various underlying pathophysiological defects. Very often, monotherapy alone cannot maintain adequate glycaemic control among people with type 2 diabetes, leading to treatment failure. Current guidelines recommend a combination of glucose-lowering drugs which have complementary mechanisms of action which can address the various pathophysiological pathways, with no increased risk of hypoglycaemia and cardiovascular events. In this study, we noted both SGLT2i and DPP4i were effective in managing glucose levels but brought some distinct differences in other clinical parameters in our cohort. Although both medications were not significantly different in their clinical effects, SGLT2i use brought a relatively larger point reduction in HbA1c, weight loss and blood pressure in the cohort within the first 6 months, which were important risk factors for cardiovascular event progression.^[22] DPP4i on the other hand brought a larger point reduction in LDL and TC levels.

Our study found that SGLT2i use resulted in a smaller non-significant decline in eGFR point estimate compared with DPP4i use, suggesting that there were some renoprotective effects with SGLT2i use. This concurs with previous clinical trials that have shown that SGLT2i could slow down the annual eGFR decline for patients with type 2 diabetes.^[23–25] Nevertheless, ‘eGFR dip’ was not observed in this study as seen in previous clinical trials.^[26–28] We only noted an increase in serum creatinine with both medications use in the first 30 months, followed by a decline in serum creatinine from 36 months onwards, mirroring results from previous studies.^{[23, 29]}

However, no significant differences were observed for the other outcomes including long-term cardiovascular and renal outcomes, possibly due to the small sample size and relatively short follow-up duration. Nevertheless, our study population closely reflected real-world practice,^[11] with some distinct differences from patients in clinical trials.^[30] Patients were relatively younger, had a longer duration of diabetes, and had a significant proportion with established cardiovascular disease and microvascular disease compared with those recruited in previous clinical trials. This could be due to the fact that the study was conducted in tertiary hospital settings where patients were at a more advanced stage of the disease, being referred from primary care settings for less well-controlled diabetes. Indeed, the mean HbA1c for patients in our study (8.6%) was much higher than the national average (7.9%).^[31]

Our cohort had a high treatment adherence rate, with a PDC of more than 95%, as opposed to previous studies.^{[18, 19, 32, 33]} We offer several reasons for this. Firstly, the restricted access to SGLT2i and DPP4i in our setting could be strong incentives for better adherence, due to the high cost of medications. Secondly, most patients were also enrolled in the medication treatment adherence clinic which may have improved their health literacy and thus adherence. Additionally, as these medications have a good tolerability profile, they may serve as a positive reinforcement for patients to adhere to their medication.

Strengths and limitations

In this study, we have used a prevalent-user design. Through this, we are able to identify people that start a new therapy and adjust any confounders. We were also able to preserve the number of patients that could be studied as opposed to conventional new user design which will lead to a larger loss of samples. While patients selected in the prevalent-user design were in a similar stage of disease and treatment, DPP4i were introduced much earlier compared with SGLT2i, which might have introduced confounding effects. To account for this, we used a time-conditional propensity score matching to reduce confounding. Sensitivity analysis using a new user design with the incident cohort was largely in line with the main analysis. Our analysis had some limitations. First, we were unable to account for confounding bias as this was a retrospective study. We did not use calliper in the process of matching, however, baseline characteristics post-matching were well-balanced. Secondly, our sample size was relatively small with a relatively short follow-up duration of 1.52 years, which is likely underpowered to observe differences in cardiovascular and renal endpoints. However, as with most low-resource settings, access to newer medications is highly restricted, which explains the small cohort of patients we had. As such, findings should be interpreted with caution as the number of patients decreased over the span of a 60-month follow-up. We also limited our analysis using DPP4i as a comparator to SGLT2i in this prevalent new user design as there was minimal SGLT2i usage before DPP4i initiation. Thirdly, adherence measures using a secondary database assumed that medication consumption corresponded to medication prescription.^[34] Finally, this study was carried out in tertiary hospitals and therefore could not be generalised to other settings such as primary care clinics. Nevertheless, we believe that the results are applicable to the Malaysia hospital settings, since patients with comorbidities are more likely to benefit from the use of these medications. We relied on medical records for our analyses. As such, it is possible that some diagnoses or outcomes of interest were not captured in the records. Indeed, one missed opportunity was the absence of safety data for the two therapies, impeding our ability to identify rare events that may not be detected in clinical trials or represented in pre-licensure studies.

Conclusions

The use of SGLT2i and DPP4i were comparable in exerting distinct effects on cardiorenal risk factors. However, new users of SGLT-2 inhibitors had larger HbA1c reduction and were more persistent to their medication. These data can potentially provide new insights into the prescribing patterns of clinicians in a low resource middle-income country and confirm the effectiveness in broader, non-trial participants.

Acknowledgement

We would like to thank the Director General of Health Malaysia for his permission to publish this article.

Author Contributions

The study was designed by R.S. and S.W.H.L. Data analysis was planned and conducted by R.S. and S.W.H.L. Project was supervised by S.W.H.L. and C.W.C. The first draft was written by R.S. and S.W.H.L. Critical input for important intellectual content was provided by all authors (R.S., C.W.C., N.K.L., N.L.A., Z.H., S.W.H.L.).

Funding

This research did not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors.

Conflict of Interest

The author(s) declare that there are no conflicts of interest.

Data Availability

The datasets generated during and/or analysed during the current study are available from the corresponding author upon reasonable request.