A Real‐World Cohort Study on the Association of the Use of Sodium‐Glucose Cotransporter‐2 Inhibitor With the Incidence of Chronic Kidney Disease (CKD) in Patients With Type 2 Diabetes Mellitus Without CKD Using the SingHealth Diabetes Registry

ABSTRACT

Background

Sodium‐glucose cotransporter‐2 inhibitors (SGLT2i) are well‐established for slowing chronic kidney disease (CKD) progression. However, their role in reducing the incidence of CKD in individuals with diabetes mellitus remains unclear.

Objectives

We evaluate the association of SGLT2i use with CKD incidence in a pre‐CKD multi‐ethnic Asian population with type 2 diabetes mellitus (T2D) with no CKD.

Materials and Methods

This retrospective cohort study analysed patients with T2D from 2016 to 2023 using the SingHealth Diabetes Registry. Patients with baseline estimated glomerular filtration rate (eGFR) ≥ 60 mL/min/1.73 m² and urinary albumin‐to‐creatinine ratio (uACR) < 30 mg/g were included. Cases and controls were matched based on age, gender, housing type, body mass index, HbA1c, baseline eGFR, renin‐angiotensin‐system inhibitors, metformin and glucagon‐like peptide‐1 receptor agonist using propensity score matching (PSM). Primary outcome was incident CKD and secondary outcomes included eGFR and uACR change over time. We used conditional logistic regression to estimate the odds ratio of incident CKD and mixed‐effects linear regression to assess eGFR change over time.

Results

Of 21 780 patients, 4450 patients (2225/group) were analysed after PSM. Median baseline eGFR were 103.58 mL/min/1.73 m² and 103.53 mL/min/1.73 m² in SGLT2i users and non‐users, respectively. SGLT2i users had lower CKD incidence (OR 0.36, 95% CI 0.31–0.42; p < 0.001) compared to non‐users. Longitudinal analysis demonstrated a slower rate of GFR decline and lower uACR increase over time in SGLT2i users.

Conclusion

SGLT2i use in individuals with T2D and preserved renal function was associated with lower CKD incidence, suggesting its potential preventive role in non‐CKD population.

1. Introduction

Chronic kidney disease (CKD) is a major risk factor and contributor to cardiovascular mortality, progression to renal failure requiring renal replacement therapy, and acute kidney injury [1, 2]. CKD has been forecasted to be the fifth leading cause of death worldwide by 2050, and is a significant burden on global resources and healthcare costs [3, 4]. Reported median CKD prevalence in the region (South‐east Asia and Oceania) at 10.4% is higher than the global prevalence of 9.5%, and the prevalence of treated renal failure in Singapore is double that of the global median [5]. Diabetic kidney disease is the leading cause of end stage kidney disease (ESKD) worldwide and in Singapore [4, 5]. In Singapore, the prevalence of CKD has risen significantly from 8.8% in 2019–2020 to 13.8% in 2021–2022 [6]. In the period of 2023–2024, the age‐standardised prevalence of diabetes mellitus in Singapore was 8.8%, with approximately every 1 in 5 persons between 60 to 74 years old being affected by type 2 diabetes mellitus (T2D) [7]. Singapore had one of the highest incidences of treated kidney failure on kidney replacement therapy attributed to diabetes in the world, and the proportion of patients on dialysis with diabetic nephropathy in Singapore has risen from 48.5% in 2012 to 56.0% in 2022 [6, 8].

The role of sodium‐glucose cotransporter‐2 inhibitors (SGLT2i) in CKD management has been well established and incorporated into existing clinical guidelines as one of the main pillars of pharmacotherapy. The effect of SGLT2i on kidney composite outcomes (comprising renal function, ESKD, death from renal causes) in the population with T2D or diabetic kidney disease has been well described in systematic reviews and meta‐analyses comprising several large randomised controlled trials (RCTs) [9, 10, 11]. Large trials including CREDENCE, EMPA‐KIDNEY and DAPA‐CKD have demonstrated cardiorenal protection of SGLT2i in CKD, with or without T2D [12, 13, 14].

The shift in focus to primary prevention of cardiovascular disease had successfully reduced mortality and healthcare burden of disease. With the expected rise of healthcare burden of CKD globally, there is an impetus to shift management of CKD to primary prevention in an at‐risk population.

Growing evidence suggests that the protective role of SGLT2i extends beyond CKD and may lower the incidence of CKD in high‐risk populations. Post hoc analyses of existing RCTs, including DECLARE‐TIMI and EMPA‐REG OUTCOME, reported that SGLT2i use in low KDIGO risk populations is associated with improved renal outcomes, with reduced CKD incidence or a reduction in the rate of estimated glomerular filtration rate (GFR) decline [15, 16]. A retrospective cohort, nested case–control study in Spain reported that the use of SGLT2i and glucagon‐like peptide‐1 receptor agonist (GLP1‐RA) for more than 3 years was associated with decreased risk of CKD incidence [17].

In this study, we aim to evaluate the real‐world association of SGLT2i use on the incidence of CKD in a multi‐ethnic Southeast Asian population with T2D.

2. Materials and Methods

2.1. Study Design and Setting

This retrospective cohort study analysed patients with T2D from 2016 to 2023 using the SingHealth Diabetes Registry (SDR) [18]. The SDR was established in 2015 and included patient data retrospectively from 2013 and data collection remains ongoing. It is updated with electronic medical records across the primary healthcare and hospital‐based continuum within SingHealth, which is the largest healthcare cluster in Singapore, comprising acute hospitals, national specialty centres, community hospitals and primary care clinics. The SDR comprises data on disease conditions such as diabetes and ischemic heart disease (IHD), identified through diagnosis codes on the electronic medical records (Tables S1.1 and 1.2).

The study was approved by the SingHealth Centralised Institutional Review Board (IRB registration number 2024‐2166). All SDR data are anonymized before release to researchers. The use of anonymised data included a data use agreement that prohibits reidentification and unauthorised data redistribution.

2.2. Participants

Patients with T2D from 2016 to 2023 identified using the SDR were included in this study. Inclusion criteria were baseline eGFR ≥ 60 mL/min/1.73 m² and urinary albumin‐to‐creatinine ratio (uACR) < 30 mg/g. The exclusion criteria included age < 21 years, history of renal transplant or maintenance dialysis (Figure 1).

FIGURE 1.

Patient flowchart.

2.3. Variables and Data Sources

The primary outcome was incident CKD, defined as sustained eGFR < 60 mL/min/1.73 m² or uACR > 30 mg/g with consecutive readings of > 90 days apart. Secondary outcomes included change in eGFR over time, assessed at yearly time‐points following the index year of SGLT2i initiation.

eGFR was calculated using the 2009 Chronic Kidney Disease Epidemiology Collaboration (CKD‐EPI) equation with serum creatinine measured in μmol/L and standardised to isotope dilution mass spectrometry‐traceable method [19]. uACR was quantified in mg/g. Ethnicity was classified into the main ethnic groups in Singapore including Chinese, Indian, Malay and others. Residential housing type served as a surrogate marker for socioeconomic status. Body mass Index (BMI) was calculated by weight in kilograms divided by the square of height in meters. Diagnosis of hypertension was defined by any use of anti‐hypertensive medications.

The index date of exposure was defined as the date of first prescription obtained in the study period. This was determined by the date of first SGLT2i prescription in SGLT2i users and the first date of non‐SGLT2i prescription among non‐SGLT2i users. Exposure was defined as ever use of SGLT2i and this status was assigned at the start of the follow‐up period at the point where participants were eligible to limit the effect of immortal time bias.

2.4. Bias

Potential confounding was addressed using propensity score matching (PSM). PSM with 1:1 matching was conducted on SGLT2i users and non‐users based on age, sex, housing type, body mass index, glycated haemoglobin (HbA1c), baseline eGFR, IHD and use of metformin, renin‐angiotensin‐system inhibitor (RASi), mineralocorticoid receptor antagonist (MRA) and GLP‐1RA based on a calliper of 0.1 standard deviation without replacement using nearest neighbour method. The propensity score distribution is shown in Figure 2.

FIGURE 2.

Distribution of propensity scores.

2.5. Statistical Methods

Conditional logistic regression was used to estimate the odds ratio (OR) of incident CKD. Sensitivity analysis using Cox regression was conducted to examine the association between SGLT‐2i and CKD. Patients were followed up from the date of first prescription to the date of incident CKD or up to 1 January 2024. In patients with CKD diagnosed prior to SGLT2i prescription, follow‐up was between the date of first documented renal function (creatinine) to the date of CKD diagnosis. Longitudinal analysis was conducted using mixed effects linear regression to evaluate changes in eGFR and uACR over time from index year, between SGLT2i users and non‐users.

Supplementary analysis was conducted using a nested case–control approach. Patients with incident CKD were matched with the patients without CKD using attained age, gender, housing type, BMI, HbA1c, eGFR, IHD, use of RASi, metformin, GLP‐1RA, MRA and analysis time (since date of diagnosis of DM) using sttocc in the estimation of the effect of SGLT2i on incident CKD.

3. Results

3.1. Baseline Characteristics

A total of 21 780 participants in the registry during the study period met the inclusion criteria. After propensity score matching, 4450 patients (2225 in each group) were analysed. Median baseline eGFR and uACR were 103.6 mL/min/1.73 m² (IQR 94.2–111.4) and 10.6 mg/g (IQR 7.1–17.3), respectively, in SGLT2i users; and 103.5 mL/min/1.73 m² (IQR 94.0–112.9) and 11.5 mg/g (IQR 7.1–17.7), respectively, in controls or non‐users. SGLT2i users had a lower systolic blood pressure, higher incidence of prior IHD, and a higher proportion who were also prescribed sulfonylurea, Dipeptidyl peptidase‐4 inhibitors (DPP‐4i) and diuretics (Table 1).

TABLE 1.

Baseline characteristics of participants.

Characteristics, N (%), unless otherwise stated	Before PSM			After PSM
	Non‐SGLT2i users (n = 17 828)	SGLT2i users (n = 3952)	p	Non‐SGLT2i users (n = 2225)	SGLT2i users (n = 2225)	p
Age			< 0.001			0.21
< 60 years old	9176 (51.5)	2978 (75.4)		1660 (74.6)	1696 (76.2)
≥ 60 years old	8652 (48.5)	974 (24.7)		565 (25.4)	529 (23.8)
Age; mean (SD)	59.4 (11.7)	52.8 (11.2)	< 0.001	50.94 (12.0)	51.28 (10.9)	0.57
Male	9832 (55.2)	2426 (61.4)	< 0.001	1327 (59.6)	1369 (61.5)	0.20
Ethnicity			< 0.001			0.02
Chinese	12 403 (69.6)	2339 (59.2)		1328 (59.7)	1310 (58.9)
Malays	2732 (15.3)	736 (18.6)		471 (21.2)	417 (18.7)
Indians	1803 (10.1)	598 (15.1)		283 (12.7)	343 (15.4)
Others	890 (5.0)	279 (7.1)		143 (6.4)	155 (7.0)
Housing type			0.003			0.96
Studio‐/1‐/2‐/3‐room	4992 (29.3)	992 (26.5)		586 (26.3)	573 (25.8)
4‐room	5570 (32.7)	1236 (33.0)		752 (33.8)	760 (34.2)
5‐room, executive apartment	4448 (26.1)	1058 (28.3)		616 (27.7)	625 (28.1)
Condominium, private	2033 (11.9)	455 (12.2)		271 (12.2)	267 (12.0)
Missing	785	211		0	0
BMI; median (IQR)	26.6 (23.9, 30.1)	28.4 (25.1, 32.4)	< 0.001	28.2 (24.8, 32.6)	28.3 (25.0, 32.3)
Missing	6640	1591		0	0	0.82
SBP; median (IQR)	133 (123, 144)	130 (120, 141)	< 0.001	132 (122, 141)	131 (120, 140)	0.04
Missing	1706	557		394	426
Hypertension (prevalent)	13 459 (75.5)	2759 (69.8)	< 0.001	1491 (67.0)	1529 (68.7)	0.22
History of Prior IHD	2000 (11.2)	754 (19.1)	< 0.001	337 (15.2)	387 (17.4)	0.04
History of Ischemic stroke	945 (5.3)	146 (3.7)	< 0.001	65 (2.9)	62 (2.8)	0.79
History of Haemorrhagic stroke	161 (0.9)	11 (0.3)	< 0.001	7 (0.3)	5 (0.2)	0.56
History of heart failure	64 (0.4)	87 (2.2)	< 0.001	8 (0.4)	40 (1.8)	< 0.001
RASi	4785 (26.8)	1107 (28.0)	0.13	585 (26.3)	633 (28.4)	0.11
MRA	139 (0.8)	66 (1.7)	< 0.001	35 (1.6)	40 (1.8)	0.56
Diuretic	282 (1.6)	124 (3.1)	< 0.001	35 (1.6)	68 (3.1)	0.001
GLP1‐RA	8 (0.04)	8 (0.2)	0.001	2 (0.09)	2 (0.09)	> 0.99
Metformin	9191 (51.6)	2801 (70.9)	< 0.001	1591 (71.5)	1571 (70.6)	0.51
Sulfonylurea	2102 (11.8)	962 (24.3)	< 0.001	398 (17.9)	517 (23.2)	< 0.001
Baseline DM medications excluding SGLT2i and insulin	9523 (53.42)	2945 (74.52)	< 0.001	1622 (72.90)	1653 (74.29)	0.29
Premixed insulin	68 (0.4)	45 (1.1)	< 0.001	11 (0.5)	21 (0.9)	0.08
DPP‐4i	529 (3.0)	317 (8.0)	< 0.001	110 (4.9)	184 (8.3)	< 0.001
Any insulin use	2248 (12.6)	825 (20.9)	< 0.001	368 (16.5)	403 (18.1)	0.17
eGFR; median (IQR)	97.2 (86.1, 105.5)	103.4 (93.8, 111.3)	< 0.001	103.53 (94.0, 112.9)	103.6 (94.2, 111.4)	0.25
uACR; median (IQR)	11.5 (7.7, 17.7)	11.5 (7.7, 17.3)	0.23	11.49 (7.1, 17.7)	10.61 (7.1, 17.3)	0.23
HbA1c (%); median (IQR)	6.9 (6.3, 8.1)	8.1 (7.0, 10.1)	< 0.001	7.7 (6.7, 10.3)	7.9 (6.9, 9.9)	0.046

Note: Intention to highlight p < 0.05 for statistical significance.

Abbreviations: BMI: body mass index; DFPP4i: dipeptidyl peptidase‐4 inhibitor; eGFR: estimated glomerular filtration rate; GLP1‐RA: glucagon‐like peptide‐1 receptor agonist; HbA1c: glycated haemoglobin; IHD: ischemic heart disease; IQR: interquartile range; MRA: mineralocorticoid receptor antagonist; PSM: propensity score matching; RASi: renin‐angiotensin system inhibitor; SBP: systolic blood pressure; SD: standard deviation; SGLT2i: sodium‐glucose co‐transporter‐2 inhibitor; uACR: urinary albumin‐creatinine ratio.

The odds of incident CKD in SGLT2i users are lower compared with non‐users (OR: 0.36, 95% CI: 0.31–0.42; p < 0.001). There were no significant interactions observed between SGLT2i and baseline uACR (p for interaction = 0.67), use of RASi (p for interaction = 0.31) or metformin use (p for interaction = 0.35) (Table 2). In Cox proportional hazards models, SGLT2i use was associated with a significantly lower hazard of incident CKD compared with non‐users (HR 0.59, 95% CI 0.51–0.67; p < 0.001), consistent with the findings above.

TABLE 2.

Odds ratio of SGLT2i on incident CKD (post PSM).

	OR (95% CI)	p
SGLT2i	0.36 (0.31, 0.42)	< 0.001

Note: No significant interaction between SGLT2i and RAS blocker (p for interaction = 0.31), between SGLT2i and metformin (p for interaction = 0.35), as well as between SGLT2i and baseline uACR (p for interaction = 0.67).

3.2. Longitudinal Analysis

In our longitudinal analysis, there is a significant interaction between SGLT2i use and time, suggesting a time‐dependent effect (p value < 0.001) for change in eGFR (Figure 3) and uACR (Figure 4) over time with SGLT2i use. The change in eGFR per year was −1.72 mL/min/1.73 m² among non‐SGLT2i users compared to −0.94 mL/min/1.73 m² among SGLT2i users. The change in uACR per year was 5.19 mg/g among non‐SGLT2i users and 3.91 mg/g among SGLT2i users. Users of SGLT2i had a slower rate of GFR decline and a lower increase in uACR over time compared to non‐users.

FIGURE 3.

Change in eGFR over time.

FIGURE 4.

Change in uACR over time.

3.3. Supplementary Analysis

In supplementary analysis, using a nested case control approach, a total of 19 489 patients who were non‐users of SGLT2i were matched to 2265 SGLT2i users using attained age, gender, housing type, BMI, HbA1c, eGFR, use of RASi, metformin, GLP‐1RA and analysis time. The results were congruent and showed lower incidence of CKD among SGLT2i users with odds of 0.17 (95% CI 0.14–0.21; p < 0.001) compared to never‐users.

4. Discussion

This study demonstrated that SGLT2i use was associated with a lower risk of incident CKD in a large multi‐ethnic population with T2D and normal renal function. SGLT2i use was also associated with a slower rate of GFR decline and lower uACR increase over time.

The study population is classified as having low existing CKD risk given that they had eGFR > 90 and uACr < 30 mg/g at the time of recruitment. This demographic is of particular interest as they have existing T2D, which places them at risk of developing CKD in later life.

These findings are congruent with those from large RCTs, including those from a post hoc analysis of the DECLARE‐TIMI 58 trial [16]. The study's post hoc analysis demonstrated reduction in kidney outcomes across all KDIGO risk categories, including their low‐risk category, defined as eGFR ≥ 60 mL/min/1.73 m² and UACR < 30 mg/g, similar to our study population. The DECLARE‐TIMI 58 trial reported a 46% risk reduction in the low kidney risk category in primary renal outcome (defined as time to 40% eGFR reduction or progress to eGFR < 60 or ESKD), and reduced eGFR slope decline with dapagliflozin use. The generalisability of these findings from a post hoc analysis remains limited but is useful for generating hypotheses. Our study confirmed that these findings are reproducible in a retrospective cohort study and in an Asian multi‐ethnic population, compared to a predominantly White population (75%–80%) in the DECLARE‐TIMI trial.

The EMPA‐REG OUTCOME trial [15] similarly reported that empagliflozin was associated with a 46% lower relative risk of kidney outcomes compared with placebo across the KDIGO risk categories. Their low KDIGO risk category had a median uACR of 7.1–8.0 mg/g and a mean eGFR of 83–84, of which the eGFR is notably lower compared to our study population. In contrast, this trial examined composite renal outcomes, including incident CKD, worsening nephropathy and progression of uACR, capturing outcomes at a later stage of CKD progression than our study. Our study showed congruent findings that SGLT2i use was associated with improved renal outcomes but also demonstrated that these associations are seen even at an earlier pre‐disease stage and at a higher eGFR.

The results of our study contribute to the increasing pool of data demonstrating the role of SGLT2i in primary prevention of CKD and extends these findings to a low‐risk population in the pre‐CKD stage, but with the propensity to develop diabetic kidney disease. Given that diabetic kidney disease is a leading cause of ESKD, these findings will translate to clinical significance in implementing early intervention in a target population with T2D. Evidence of time‐dependent effect on eGFR decline and uACR control also raises the implication that earlier intervention may translate to greater beneficial effects on renal outcomes in future for the individual.

There are a few limitations in this study. First, this is a retrospective analysis of observational registry data and cannot demonstrate causality. The potential effect of confounding factors and missing data was addressed with the use of propensity score matching. Despite the inherent limitations, the large study size and representation of the population allow greater generalisability in applying the study findings in real‐world clinical practice. The findings of the study were also consistent with those of other large RCTs done in a low‐risk study population.

Second, the study participants were not matched on the use of certain glucose‐lowering medications including sulfonylurea and DPP‐4i. We postulate that the effect of these medications on renal outcomes is unlikely to be significant, and any potential effects on uACr or T2D control, as evidenced by other markers such as HbA1c, have been addressed and adjusted for. The standardised mean differences (SMDs) for baseline variables before and after propensity score matching were less than 0.20 (Table S2), which is the proposed threshold for residual bias after matching. The covariate balance is adequate and no further adjustment is required post‐matching. However, we acknowledge that residual confounding from these medications cannot be entirely excluded, while noting that their direct renal effects are generally considered modest relative to those of SGLT2 inhibitors.

Thirdly, the proportion of patients receiving GLP‐1RA in the cohort was very low, which may limit the generalisability to populations with higher GLP‐1RA use. However, this reflected real‐world prescribing patterns at the time of the study.

We cannot entirely exclude the possibility of participants having subclinical or normoalbuminuric DKD at enrolment, despite applying inclusion criteria based on eGFR and albuminuria thresholds at baseline. The inclusion of such patients could have introduced heterogeneity in the study population and may have influenced the observed associations. Future studies incorporating kidney biopsy data, eGFR trend or more sensitive biomarkers may better delineate this population.

We have minimised the effect of immortal time bias by defining exposure as any SGLT2i use based on the index date corresponding to first SGLT2i use and non‐SGLT2i medication use in the non‐user group. There is residual potential for immortal time bias if there was any misclassification of exposure status and follow‐up time in the non‐SGLT2i user group.

Another limitation is the potential influence of competing risk of death before CKD onset. In our pre‐matched dataset, 70 out of 21 780 (0.32%) patients died before incidence of CKD. In the post‐matched dataset, 10 out of 4450 (0.22%) patients died before incidence of CKD. Standard Cox regression may potentially overestimate the cumulative incidence of CKD in the presence of competing risk of death, although this is unlikely to be significant in our study as the overall mortality is low within the 7‐year study period.

While causality cannot be established due to the retrospective design of this study, this limitation should be interpreted in context. Given the widespread use of SGLTi and well‐established renal and cardiac benefits, the feasibility of conducting future prospective RCTs in this area remains limited. As such, this study still provides meaningful and clinically relevant insights that add to the existing body of evidence for SGLTi.

5. Conclusion

In patients with T2D and preserved renal function, the use of SGLT2i can potentially lower rates of incident CKD and slow down eGFR decline with a time‐dependent effect. There is a potential role of SGLT2i in primary prevention in the non‐CKD population, and our findings suggest that earlier initiation of SGLT2i can have greater benefits on renal protection. Early intervention and potential primary prevention of CKD in a low‐risk population can translate into reduced burden of ESKD on healthcare in future. Future prospective studies in a non‐CKD population may shed more light on the long‐term benefit of SGLT2i on eGFR decline and other mortality outcomes.

Funding

This work was supported by the National Kidney Foundation Singapore. The institution was not involved in the study design, collection, analysis and interpretation of data, writing and submission of the manuscript.

Conflicts of Interest

Jia Liang Kwek reports receiving grants provided by the National Kidney Foundation Singapore and Lien Foundation, which have supported this study. He has also received honoraria for speaker or advisory engagement with the following institutions: AstraZeneca, Boehringer Ingelheim, Otsuka Pharmaceutical, Bayer, Vantive, Novo Nordisk, Lien Foundation for Palliative Care, Ngee Ann Polytechnic, Nanyang Polytechnic and Singapore Institute of Technology. Jason Chon Jun Choo is the Medical Director of the National Kidney Foundation Singapore. The other authors declare no conflicts of interest.

Supporting information

Table S1:1. List of ICD/SNOMED‐CT/SHPWKC codes to identify diagnosis of IHD.

Table S1:2 List of SNOMED‐CT codes to identify diagnosis of hypertension.

Table S2: Standardised mean differences (SMD) of covariates pre and post PSM.

Ong Y. L., Liew Z. H., Lim G. H., et al., “A Real‐World Cohort Study on the Association of the Use of Sodium‐Glucose Cotransporter‐2 Inhibitor With the Incidence of Chronic Kidney Disease (CKD) in Patients With Type 2 Diabetes Mellitus Without CKD Using the SingHealth Diabetes Registry,” Diabetes, Obesity and Metabolism 28, no. 10 (2026): 9188–9195, 10.1111/dom.71137.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.