Safety and clinical impact of SGLT2 inhibitor in patients undergoing peritoneal dialysis: a target-trial emulation study Authors: Yi-Ting Chen, Chung-An Wang, Jui-Yi Chen, Vin-Cent Wu Published: 2026 ABSTRACT ABSTRACTObjectiveSodium–glucose cotransporter 2 inhibitors (SGLT2i) improve cardiovascular outcomes in chronic kidney disease, but their effects in peritoneal dialysis (PD) patients who are infection prone, remain unclear.Study designThe study design of this research is target-trial emulation using the global federated electronic health record database. Adults (18–90 years) with type 2 diabetes on PD for ≥3 months between February 2015 and June 2025 were included. Propensity score matching balanced SGLT2i and nonuser. Primary outcomes were all-cause mortality, severe sepsis, sepsis, and pneumonia; secondary outcomes included major adverse cardiovascular events (MACE) and PD-associated peritonitis.ResultsAmong 29 529 eligible patients, 2815 (9.5%) received SGLT2i. After matching, 2749 patients were retained in each group with well-balanced baseline characteristics. Over a median follow-up of 0.79 years, SGLT2i users were associated with lower risks of all-cause mortality [adjusted hazard ratio (aHR) 0.818], severe sepsis (aHR 0.802), sepsis (aHR 0.661), pneumonia (aHR 0.664), and PD-associated peritonitis (aHR 0.340). MACE was not significantly different (aHR 0.798). SGLT2i was not associated with increased diabetic ketoacidosis, hypoglycemia, genital infection, volume depletion, or amputation.ConclusionsIn this large real-world PD cohort with type 2 diabetes, SGLT2i use was associated with lower risks of death and major infections without safety concerns, supporting potential benefit pending randomized trial confirmation.Keywords: end-stage kidney disease, mortality, peritoneal dialysis, SGLT2 inhibitor What was known: Sodium–glucose cotransporter 2 (SGLT2) inhibitors improve cardiorenal outcomes in chronic kidney disease, but evidence in patients with end-stage kidney disease receiving peritoneal dialysis (PD) is limited, especially in those with type 2 diabetes. Sodium–glucose cotransporter 2 (SGLT2) inhibitors improve cardiorenal outcomes in chronic kidney disease, but evidence in patients with end-stage kidney disease receiving peritoneal dialysis (PD) is limited, especially in those with type 2 diabetes. Patients on PD are highly vulnerable to infection-related complications, including pneumonia, sepsis, and PD-associated peritonitis, which are major contributors to hospitalization and mortality. Patients on PD are highly vulnerable to infection-related complications, including pneumonia, sepsis, and PD-associated peritonitis, which are major contributors to hospitalization and mortality. Use of SGLT2 inhibitors in dialysis populations has been cautious because of safety concerns, including ketoacidosis, hypoglycemia, genital infection, volume depletion, and amputation. Use of SGLT2 inhibitors in dialysis populations has been cautious because of safety concerns, including ketoacidosis, hypoglycemia, genital infection, volume depletion, and amputation. This study adds: In a large global real-world cohort of patients with type 2 diabetes on PD, SGLT2 inhibitor use was associated with lower all-cause mortality and major infection-related outcomes after propensity score matching. In a large global real-world cohort of patients with type 2 diabetes on PD, SGLT2 inhibitor use was associated with lower all-cause mortality and major infection-related outcomes after propensity score matching. SGLT2 inhibitor use was associated with a markedly lower risk of PD-associated peritonitis, a clinically important and novel finding for PD care. SGLT2 inhibitor use was associated with a markedly lower risk of PD-associated peritonitis, a clinically important and novel finding for PD care. No excess risk was observed for key safety outcomes, and findings were supported by subgroup, landmark, and positive/negative exposure control analyses, strengthening robustness. No excess risk was observed for key safety outcomes, and findings were supported by subgroup, landmark, and positive/negative exposure control analyses, strengthening robustness. Potential impact: Y.These findings support reconsidering the role of SGLT2 inhibitors in PD patients with type 2 diabetes, beyond glucose control, for survival and infection risk reduction. Y.These findings support reconsidering the role of SGLT2 inhibitors in PD patients with type 2 diabetes, beyond glucose control, for survival and infection risk reduction. The results provide effect estimates and clinically relevant endpoints that can inform the design of future randomized controlled trials in PD populations. The results provide effect estimates and clinically relevant endpoints that can inform the design of future randomized controlled trials in PD populations. If confirmed in trials, SGLT2 inhibitors may be incorporated into treatment strategies for high-risk PD patients to reduce infection-related morbidity, mortality, and healthcare utilization. If confirmed in trials, SGLT2 inhibitors may be incorporated into treatment strategies for high-risk PD patients to reduce infection-related morbidity, mortality, and healthcare utilization. INTRODUCTION Sodium–glucose cotransporter 2 inhibitors (SGLT2i), originally developed to treat type 2 diabetes mellitus (T2DM), have demonstrated broad organ-protective effects [1–5]. However, the safety and organ-protection effects of SGLT2i in end-stage kidney disease (ESKD) patients undergoing dialysis remain unknown. Recent updates from the UK Kidney Association and Kidney Disease: Improving Global Outcomes now recommend initiating SGLT2i therapy at an estimated glomerular filtration rate threshold as low as 20 ml/min/1.73 m2 and continuing treatment until dialysis required [6, 7]. In parallel, regulatory guidance has evolved: as of May 2023, the U.S. Food and Drug Administration removed its prior contraindication on dapagliflozin use in dialysis patients [8], and the European Medicines Agency likewise does not list dialysis as a contraindication [9]. Although definitive large-scale randomized controlled trial (RCT) data are lacking, mechanistic studies suggest SGLT2i may confer cardiovascular and renal protection in dialysis-dependent ESKD through nonglycemic pathways-including anti-inflammatory effects, anemia amelioration, and oxidative stress reduction [10, 11]. Pharmacokinetic evaluation in ESKD patients undergoing dialysis revealed that dapagliflozin reaches a peak plasma concentration of 117 ng/ml, modestly higher than in diabetic individuals with preserved renal function (97.6 ng/ml), indicating limited dialyzability and overall safety in this population [12]. A 2025 narrative review on SGLT2 inhibitor use in dialysis noted that evidence in peritoneal dialysis (PD) is limited with inconsistent effects on ultrafiltration, blood pressure, and peritoneal transport, and no hard outcomes, concluding that SGLT2i use in PD, should remain cautions pending robust efficacy and safety data [13]. Notably, emerging data suggest SGLT2i could reduce the risk of sepsis and pneumonia, especially among patients with chronic kidney disease (CKD), who are inherently at higher risk for such infections due to impaired immunity, metabolic dysfunction, and frequent hospital exposures [14–16]. Patients undergoing PD are particularly susceptible to pneumonia, a vulnerability attributable to the combined effects of malnutrition, chronic inflammation, the presence of an indwelling catheter-related peritonitis, and an overall heightened infection burden [17]. Across CKD cohorts, treatment with SGLT2i has been consistently associated with lower incidences of pneumonia and sepsis [14, 18, 19]. These benefits are thought to arise from pleiotropic improvements in renal and cardiovascular physiology, most notably anti-inflammatory, hemodynamic, and metabolic effects, rather than from any intrinsic antimicrobial activity [3, 20–22]. To determine whether these associations extend to PD populations treated with SGLT2i, with particular focus on dialysis-related complications including sepsis and long-term outcomes, we emulated a target trial using the TriNetX global electronic health record network. MATERIALS AND METHODS Data source Data for this study were accessed through the TriNetX analytics platform, a global collaborative health research network widely utilized in numerous prominent epidemiological studies. The data used in this investigation were extensive, covering various aspects of patient information including demographic details, diagnoses [according to International Classification of Disease, Tenth Revision, Clinical Modification (ICD-10-CM) codes], procedures [classified according to the International Classification of Disease, Tenth Revision, Procedure Coding System (ICD-10-PCS) or Current Procedural Terminology], and medications (coded as per the Veterans Affairs National Formulary). It also encompassed laboratory tests (organized using Logical Observation Identifiers Names and Codes) and healthcare utilization records from a network of 146 healthcare organizations including hospitals, primary care facilities, and specialist care providers. The dataset comprised records from both insured and uninsured individuals, representing a total population of more than 250 million patients. Data collection spanned from 1 February 2015 to 1 January 2025, with follow-up through 11 June 2025, ensuring a minimum observation period of 6 months. To enhance data reliability, rigorous preprocessing procedures were undertaken to address missing or implausible values and to harmonize heterogeneous data structures across contributing institutions. Detailed documentation of the TriNetX platform is publicly available [23] and has been validated in multiple peer-reviewed studies [24]. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline for cohort studies (Supplementary Table 8). Study design Using a target-trial emulation design, we compared adults with type 2 diabetes who were undergoing PD and initiated an SGLT2i with a contemporaneous cohort who began other oral hypoglycemic agents. Patients were identified in the TriNetX network between 1 February 2015 and 11 June 2025: eligibility required an age of 18–90 years, at least 3 months of PD, and complete baseline and follow-up data. The index date was defined as 30 days after the first SGLT2i prescription (exposure group) or, for controls, 30 days after initiation of a comparator agent [sulfonylurea, metformin, or dipeptidyl peptidase-4 (DPP-4) inhibitor]. This 30-day latency minimizes bias arising from early, transient drug effects and incomplete first-fill documentation for the dispensing [25]. Diagnostic, procedural, and encounter codes used to define exposures, covariates, and outcomes are list in Supplementary Tables 1–4. Ethics statement Because TriNetX supplies only de-identified, aggregated data, the Western Institutional Review Board waived the requirement for informed consent [26]. The protocol was also approved by the Chi-Mei Hospital IRB (No. 11210-E01). All study procedures adhered to the Declaration of Helsinki, and reporting followed STROBE guidelines for observational cohort studies [27]. Prespecified outcomes The primary outcomes included all-cause mortality, pneumonia, severe sepsis, and sepsis. The secondary outcomes were major adverse cardiovascular events (MACE) and PD-associated peritonitis. Covariates To address disparities between groups, we collected various factors from the studied population. Key demographic data included age, gender, and ethnicity, along with existing comorbidities and medication usage. To minimize potential discrepancies in baseline characteristics between the two study groups, we thoroughly integrated and algorithmically selected high-dimensional covariates assessed within 1 year prior to the index time. Comorbidities were identified based on ICD-10-CM codes (Supplementary Tables 2 and 3). Additionally, we gathered potentially influential factors from physical examinations, such as systolic blood pressure and body mass index (BMI). A comprehensive set of laboratory tests was also conducted, including albumin, C-reactive protein, hemoglobin, glycohemoglobin [i.e. hemoglobin A1c (HbA1C)], potassium, calcium, and phosphate levels. Subgroup analyses Subgroup analyses were stratified by demographic and clinical characteristics, including sex, age, race, BMI, HbA1C, systolic blood pressure, hypertension, carotid artery disease, atrial fibrillation, heart failure, chronic obstructive pulmonary disease, dyslipidemia, gout, and prior history of SGLT2i uses before PD. To further assess the robustness of our findings, we conducted sensitivity analyses using the Cox proportional hazards model with alternative covariate adjustments and varied cohort exclusion criteria. Positive and negative controls outcome To validate our analysis, we performed both positive and negative outcome and exposure controls. We selected ketoacidosis recognized as adverse effects of SGLT2i [28], as positive outcome controls. On the other hand, hypoglycemia, volume depletion, osteoporotic fracture, and amputation were chosen as negative outcome controls. For exposure controls, renin–angiotensin–aldosterone inhibitors were selected as the positive control of exposure. Conversely, a nonsteroid anti-inflammatory drug, not expected to influence outcomes, was used as a negative control of exposure. Landmark analysis To mitigate immortal time bias, we performed a landmark analysis by redefining the cohort selection window reflecting the legacy effect of SGLT2i, shifting it from the original 3 months after initiating PD to fixed time points at 45, 120, and 150 days [29]. This strategy ensured that the effect of SGLT2i on outcomes was consistently assessed across varying time intervals. Statistical analysis Baseline characteristics of SGLT2i users and nonusers were summarized as means with standard deviations (SDs) for continuous variables and frequencies with percentages for categorical variables. Between-group differences were assessed using independent two-sample t tests for continuous variables and χ² tests for categorical variables. To reduce confounding, one-to-one propensity score matching (PSM) was performed using a greedy nearest-neighbor algorithm with a caliper width of 0.1 pooled SD. Covariate balance was evaluated using standardized differences, with values <0.1 indicating adequate balance. Time-to-event outcomes were analyzed using Kaplan–Meier estimates and compared via log-rank tests. Adjusted hazard ratios (aHRs) with 95% confidence intervals (CIs) were calculated using Cox proportional hazards models. The proportional hazards assumption was verified with Schoenfeld residuals. To quantify the potential impact of unmeasured confounding, E-values were calculated for primary outcomes [30]. All statistical tests were two-sided, with significance defined as P < .05. Prespecified analyses included specificity assessments, positive outcome controls, and negative outcome controls, Analyses were conducted using SAS software, version 9.4 (SAS Institute Inc.). For all analyses, statistical significance was set at a two-tailed P-value <.05, with results reported as point estimate and 95% CIs. RESULTS Study population In this cohort of 29 529 patients with T2DM and ESKD undergoing PD, the mean age was 61.3 years; 57.0% (16 828 of 29 529) were male. Among them, 2749 patients (9.3%) received SGLT2i and were classified as the exposure group (Fig. 1), while the remaining 26 714 patients comprised the nonuser comparator group. The mean (SD) age of the SGLT2i group was 62.1 (12.2) years, which was younger than that in the control group [62.4 (12.5) years]. The SGLT2i group consisted of 1810 (65.8%) men and 939 (34.2%) women, with most patients being White [1570 (57.1%)]. Flow diagram of all participants. The diabetic patients with ESKD undergoing PD between February 2015 and June 2025 were enrolled then classified into SGLT2i users and nonusers after PSM. After PSM, there were 2749 SGLT2i users and an equal number of nonusers (controls) for analysis (Table 1). After PSM, these two groups had small and well-matched differences in age, gender, race and ethnicity, comorbidities, medication use, and most other laboratory results. Table 1:Baseline characteristics were compared among all participants before and after PSM.Before matchingAfter matchingSGLT2i usersNonusersSGLT2i usersNonusers(n = 2815)(n = 26 714)Std. diff.(n = 2749)(n = 2749)Std. diff.Demographics Age at index date62.1 ± 12.260.5 ± 14.20.12262.1 ± 12.262.4 ± 12.50.026 Male, n (%)1869 (66.4%)14 959 (56.0%)0.2151810 (65.8%)1830 (66.6%)0.015 White, n (%)1607 (57.1%)14 747 (55.2%)0.0381570 (57.1%)1586 (57.7%)0.012 Not Hispanic or Latino, n (%)2542 (90.3%)22 148 (82.9%)0.2172476 (90.1%)2461 (89.5%)0.018Comorbidities, n (%) Hypertension1848 (65.6%)17 916 (67.1%)0.0301813 (66.0%)1788 (65.0%)0.019 Hyperlipidemia2153 (76.5%)17 571 (65.8%)0.2382097 (76.3%)2140 (77.8%)0.037 Heart failure1944 (69.1%)11 997 (44.9%)0.5031880 (68.4%)1901 (69.2%)0.016 Disease of liver500 (17.8%)4282 (16.0%)0.046488 (17.8%)516 (18.8%)0.026 Chronic lower respiratory disease749 (26.6%)6449 (24.1%)0.057738 (26.8%)712 (25.9%)0.021 Overweight and obesity1570 (55.8%)12 327 (46.1%)0.1941533 (55.8%)1532 (55.7%)0.001 Diabetic retinopathy124 (4.4%)2948 (11.0%)0.250123 (4.5%)118 (4.3%)0.009Comedications at SGLT2i initiation, n (%) ACEi/ARB1721 (61.1%)12 329 (46.2%)0.3041675 (60.9%)1660 (60.4%)0.011 Beta-blockers2414 (85.8%)19 813 (74.2%)0.2932349 (85.4%)2328 (84.7%)0.021 CCB1432 (50.9%)12 764 (47.8%)0.0621400 (50.9%)1395 (50.7%)0.004 Diuretics2263 (80.4%)16 075 (60.2%)0.4542197 (79.9%)2170 (78.9%)0.024 HMG-CoA reductase inhibitors2275 (80.8%)17 573 (65.8%)0.3452211 (80.4%)2213 (80.5%)0.002 Insulin2520 (89.5%)23 970 (89.7%)0.0072463 (89.6%)2455 (89.3%)0.009 Biguanides827 (29.4%)5395 (20.2%)0.214808 (29.4%)811 (29.5%)0.002 Thiazolidinediones53 (1.9%)509 (1.9%)0.00253 (1.9%)49 (1.8%)0.011 Sulfonylureas315 (11.2%)3252 (12.2%)0.031312 (11.3%)312 (11.3%)<0.001 GLP-1 agonists331 (11.8%)1765 (6.6%)0.179316 (11.5%)328 (11.9%)0.014 DPP-4 inhibitors206 (7.3%)2106 (7.9%)0.021203 (7.4%)216 (7.9%)0.018 Alpha glucosidase inhibitors10 (0.4%)61 (0.2%)0.02410 (0.4%)10 (0.4%)<0.001Biochemistry data at SGLT2i initiation Hemoglobin, g/dl10.8 ± 2.510.6 ± 2.50.10110.9 ± 2.511.0 ± 2.50.076 Sodium, mmol/l136.1± 5.0135.9 ± 5.20.042136.2 ± 5.0136.3 ± 5.10.026 Calcium, mg/dl8.7 ± 0.78.5 ± 0.90.1798.7 ± 0.78.6 ± 0.80.073 HbA1c, %7.8 ± 2.17.9 ± 2.20.0317.8 ± 2.17.9 ± 2.10.029 Urea nitrogen, mg/dl26.7 ± 16.934.9 ± 25.40.38126.8 ± 17.028.0 ± 20.00.063 Creatinine, mg/dl2.5 ± 11.73.5 ± 7.40.1092.5 ± 11.82.0 ± 6.00.054 LDL, mg/dl84.8 ± 42.084.6 ± 43.50.00384.9 ± 42.085.2 ± 43.10.007 BMI, kg/m231.6 ± 7.831.5 ± 7.80.01331.6 ± 7.831.7 ± 7.60.019* Plus–minus values are means ±SD.ACEi, angiotensin-converting enzyme inhibitor; ARB, angiotensin II receptor blocker; CCB, calcium channel blocker; CRP, C-reactive protein; SBP, systolic blood pressure; DBP, diastolic blood pressure; GLP-1 agonists, glucagon-like peptide-1 agonists; Std. diff., standardized difference. Baseline characteristics were compared among all participants before and after PSM. * Plus–minus values are means ±SD. ACEi, angiotensin-converting enzyme inhibitor; ARB, angiotensin II receptor blocker; CCB, calcium channel blocker; CRP, C-reactive protein; SBP, systolic blood pressure; DBP, diastolic blood pressure; GLP-1 agonists, glucagon-like peptide-1 agonists; Std. diff., standardized difference. Primary and secondary outcomes After a median follow-up of 0.79 years [0.68–2.10 years (25%,75%)], all-cause mortality rate was 11.7% (321 of 2749), which was lower in the SGLT2i user group than in the nonuser group (15.5%, 427 of 2749). The use of SGLT2i was significantly associated with a lower mortality rate [aHR, 0.818 (95% CI, 0.707–0.945); P = .006] (Table 2 and Fig. 2A). Compared with nonuser group, SGLT2i users had a lower risk of pneumonia, severe sepsis, and sepsis [aHR, 0.664 (95% CI, 0.559–0.790); P < .001; aHR, 0.802 (95% CI, 0.654–0.982); P = .033; and aHR, 0.661 (95% CI, 0.564–0.775); P < .001, respectively] (Table 2 and Fig. 2B and C). Of note, a lower risk of peritonitis was observed in the SGLT2i user group compared with the nonuser group [aHR, 0.340 (95% CI, 0.202–0.573); P < .001] (Table 2). Table 2:Primary and secondary outcomes among SGLT2i users compared to the control group after PSM.Patients with outcomeaHR P-value E-valueOutcomeSGLT2i usersNonusers(95% CI)For aHRFor upper bound of 95% CIPrimary outcomes All-cause mortality3214720.818.0061.891.41(0.707, 0.945) Pneumonia2103290.664<.001N/AN/A(0.559, 0.790) Severe sepsis1622200.802.033N/AN/A(0.654, 0.982) Sepsis2503920.661<.0012.351.82(0.564, 0.775)Secondary outcomes MACE931400.798.0921.961(0.614, 1.083) Peritonitis19550.340<.0016.433.73(0.202, 0.573) Primary and secondary outcomes among SGLT2i users compared to the control group after PSM. Kaplan–Meier estimates overall and event-free survival according to treatment group among SGLT2i users and nonusers after PSM. (A) All-cause mortality (long-rank P = .006). (B) Severe sepsis (long-rank P = .033). (C) Sepsis (long-rank P < .001). No significant difference was observed in the incidence of MACE between the two groups (aHR, 0.6798 (95% CI, 0.614–1.083); P = .092] (Table 2). E-values were calculated to assess sensitivity to unmeasured confounding. For all-cause mortality (aHR 0.818), the E-value was 1.89 (1.41 for the upper 95% CI limit), indicating that an unmeasured confounder wound be limited. The association was more robust for sepsis (E-value 2.35; 1.82 for CI) (Table 2). Figure 2 presents the Kaplan–Meier survival curves for our primary outcomes. Subgroup analysis Subgroup analysis presented in the forest plot (Fig. 3) demonstrated that the association between SGLT2i use and reduced risk of adverse outcomes was consistent across all evaluated clinical strata. The protective effects observed in the primary analysis were preserved irrespective of age, sex, BMI, blood pressure, HbA1C, or the presence of major cardiometabolic comorbidities, including hypertension (HTN), coronary artery disease, atrial fibrillation, heart failure, and chronic obstructive pulmonary disease. Notably, patients with a history of coronary artery disease demonstrated a greater reduction in pneumonia risk (P for interaction = .044), and those not receiving renin–angiotensin–aldosterone system (RAAS) blockers exhibited a more pronounced decreased in sepsis risk (P for interaction = .034). No other significant effect modification was detected for any subgroup, and all point estimates favored SGLT2i use with a direction and magnitude comparable to the overall aHRs. Importantly, sensitivity analyses restricted to “new users” compared with analyses including “prior users” yielded consistent results, confirming that the observed benefits of SGLT2 inhibitor therapy were not driven by prior exposure or survivor bias. These findings indicate that the beneficial impact of SGLT2i on mortality, infection-related outcomes, and major cardiovascular events is broadly applicable and not confined to specific patient phenotypes. Collectively, the consistency across subgroups supports the robustness and generalizability of the treatment effect among PD patients with diabetes. Forest plots comparing primary outcomes of SGLT2i users compared to the controls after PSM. Safety and landmark analysis We conducted variable safety analysis to evaluate the safety profile of SGLT2i use and found no elevated risk of ketoacidosis, hypoglycemia, volume depletion, and amputation. The risk of diabetic ketoacidosis [aHR, 0.713 (95% CI, 0.418–1.217); P = .212] and genital infection [aHR, 1.046 (95% CI, 0.649–1.685); P = .854] were comparable in SGLT2i users and nonusers. Similarly, the risks of volume depletion [aHR, 0.792 (95% CI, 0.624–1.006); P = .055] and amputation [aHR, 0.661 (95% CI, 0.416–1.051); P = .078] showed nonsignificant trends without evidence of excess harm (Supplementary Table 5). In the landmark analyses comparing SGLT2i users with nonusers, patients treated with SGLT2i exhibited consistently lower risk of adverse outcomes across all time horizons. These landmark analyses, which evaluated the impact of SGLT2i use with each time window on mortality, pneumonia, severe sepsis, and sepsis, consistently demonstrated significant improvements in clinical outcomes (Supplementary Table 6). Positive and negative exposure controls Use of RAAS blockade was associated with significant reductions in all-cause mortality [aHR, 0.792 (95% CI, 0.756–0.829); P < .001], pneumonia [aHR, 0.881 (95% CI, 0.837–0.927); P < .001], severe sepsis [aHR, 0.930 (95% CI, 0.875–0.989); P = .020], and sepsis [aHR, 0.858 (95% CI, 0.820–0.897); P < .001]. In contrast, NSAID exposure was not associated with any outcome differences. Together, these findings demonstrate concordance with established therapeutic effects of RAAS blockade while showing null effects for NSAID, thereby suggesting limited systemic bias in our analytic model (Supplementary Table 7). DISCUSSION In this large, multinational, real-world cohort of patients with type 2 diabetes receiving PD, SGLT2i therapy was associated with significantly lower risks of all-cause mortality, sepsis, pneumonia, and PD-associated peritonitis, without evidence of increased adverse events. These findings extend the established renal and cardiovascular benefits of SGLT2i to a population historically excluded from randomized clinical trials. The consistent risk reduction across multiple infection-related outcomes, particularly in patients with CAD and in those not receiving RAAS blockade, suggests potential additive or complementary benefits of SGLT2i in infection-prone, high-risk population. Although causality cannot be inferred from this observational analysis, these data support the growing clinical rationale for evaluating SGLT2i as part of comprehensive management in patients with diabetes and ESKD on PD. The use of SGLT2i decreases the risk of mortality Our results demonstrated that the use of SGLT2i was associated with a significant effect on survival, even after adjusting important confounding factors. This survival benefit aligns with evidence from previous studies in nondialysis populations [1, 31, 32], suggesting that the favorable effects of SGLT2 inhibition may extend beyond the predialysis stages of CKD. Real-world data on patients with type 2 diabetes initiating dialysis showed significantly lower all-cause mortality in SGLT2i users [11], and large cohort analyses further indicate reduced mortality and cardiovascular event rates in patients starting dialysis who use SGLT2i versus nonusers [13]. Beyond glycemic control [33], SGLT2i exert favorable hemodynamic and metabolic effects, including reductions in blood pressure, inflammation [34], body weight, and lipid abnormalities—factors that may collectively contribute to improved outcomes in this high-risk population. Recent investigations suggest SGLT2i may exert a protective effect on RKF (residual kidney function) even in ESKD populations, which heavily influences survival and infection rates. The inability to adjust for baseline urine output remains a major hurdle in determining if the mortality benefit is mediated through direct renal protection or systemic anti-inflammatory pathway [21, 35]. The use of SGLT2i decreases the risk of sepsis We found that SGLT2 inhibitor therapy was associated with lower risks of pneumonia, severe sepsis, sepsis, and PD-associated peritonitis compared with nonusers, whereas no significant reduction was observed in MACE. In subgroup analyses, the reduction in pneumonia risk was particularly evident among patients with a history of CAD and better decreased sepsis in patients win non-RAAS blockaders users. SGLT2i may mitigate sepsis-induced organ dysfunction, including sepsis-related cardiomyopathy, through their anti-inflammatory and hemodynamic effects [36]. By promoting natriuresis and osmotic diuresis, SGLT2i help maintain fluid balance, reduce volume overload, and attenuate organ congestion—factors that are particularly important in mitigate heart failure exacerbations during sepsis among PD patients with residual urine output [36]. Although the underlying mechanisms remain to be clarified, emerging evidence suggests that SGLT2i may exert anti-inflammatory effects though several pathways, including attenuation of macrophage and inflammasome cytokine production, mitigation on lipopolysaccharide-induced endothelial hyperpermeability, and preservation of intestinal barrier integrity [37–39]. These combined anti-inflammatory, cardioprotective, metabolic, and volume-regulating properties may contribute to the observed reductions in sepsis-related morbidity and mortality [34]. Beyond the systemic immunomodulatory effects, the reduction in peritonitis risk may be attributable to the local protective effects of SGLT2 inhibitors on the peritoneal membrane integrity. Chronic exposure to bioincompatible, high-glucose dialysate is known to trigger epithelial–mesenchymal transition in peritoneal mesothelial cells (PMCs), a key pathological process leading to peritoneal fibrosis and barrier dysfunction [40]. Furthermore, SGLT2 inhibitors have been shown to suppress the NLRP3 inflammasome and downstream profibrotic cytokines (e.g. TGF-beta1) in renal tubular cells [41], a similar mechanism in the peritoneal cavity would protect against dialysate-induced injury, thereby reducing the susceptibility to transmural bacterial invasion. Although the clinical definitions of sepsis evolved following the adoption of the “Sepsis-3” criteria [34], our findings remained consistent through 2023, suggesting that changes in diagnostic criteria did not materially affect the results. Notably, previous studies have shown that the epidemiological trends and outcomes for sepsis and severe sepsis remain comparable when applying either “Sepsis-2” or “Sepsis-3” definitions [42], despite modest differences in case identification [43]. Our analysis demonstrated similar results regardless of whether sepsis or severe sepsis criteria were applied, reinforcing the robustness of the observed association between SGLT2 inhibitor use and reduced infection-related morbidity and mortality. SGLT2i on PD patients SGLT2 inhibitor use in PD is not yet formally endorsed by current guidelines, and clinical decisions remain individualized, balancing potential cardiovascular and RKF benefits against uncertain efficacy and regulatory limitations. The 2025 review notes that while mechanistic and cardiovascular data support interest in SGLT2i use during the transition to dialysis, clinical benefits in established PD remain speculative [13]. Ongoing trials, including studies evaluating empagliflozin and its effects on ultrafiltration in dialysis patients, may provide PD-specific evidence to inform future recommendations [44]. Importantly, our study—demonstrating significant reductions in mortality, sepsis, pneumonia, and peritonitis among PD patients receiving SGLT2i—provides robust real-world data that may inform and potentially reshape future clinical practice guidelines. Furthermore, while our data demonstrated a favorable safety profile, clinical application requires continued vigilance regarding volume homeostasis and metabolic monitoring, given the unique physiological vulnerability of patients undergoing PD. Study limitations There are several limitations in this study. First, although we employed a target-trial emulation framework with one-to-one PSM to approximate the conditions of an RCT, the retrospective observational design inherently limits causal inference. To address this limitation, we performed an E-value analysis, which indicated that only a confounder of considerable strength could fully explain the observed associations between SGLT2 inhibitor use and the primary outcomes. Second, we were unable to account for the exact duration of PD; however, we adjusted for dialysis-related complications and cardiovascular risk factors as proxies for disease severity, which likely reduced potential bias. Third, diseases were identified through diagnostic coding, which may have led to underreporting of milder conditions or events occurring outside participating healthcare systems, thereby introducing potential ascertainment bias. Nevertheless, our positive and negative control analyses indicated that such bias was unlikely to have materially influenced the results. Fourth, medication adherence and dosing information were unavailable in the TriNetX dataset. Thus, exposure misclassification cannot be fully excluded, as prescription records may not accurately reflect actual drug or persistence. Fifth, information on PD modality (e.g. continuous ambulatory vs automated PD), dialysate composition, and RKF was lacking, which could influence infection risk and treatment response. Sixth, SGLT2i may modulate the peritoneal equilibration test (PET) profiles by reducing local inflammation in the PMCs [45]. The absence of PET data in the current cohort limits the understanding of whether the observed reduction in peritonitis is linked to improved membrane integrity. Seventh, because granular adherence data and complete longitudinal information on SGLT2i treatment persistence, discontinuation, or switching are not uniformly available across contributing health systems in the federated electronic health record (HER) network, our primary analysis followed an intention-to-treat-like approach: exposure was assigned at baseline (index date) and patients were followed forward regardless of subsequent discontinuation or switching. Eighth, the median follow-up in our cohort was relatively short, which may limit statistical power for outcomes with longer latency such as MACE and may partly explain the nonsignificant association observed for this endpoint. Finally, the generalizability of these findings may be constrained by regional variations in dialysis practices, infection control measures, and prescribing patterns across healthcare systems contributing to TriNetX. CONCLUSION In this large, real-world cohort of patients with type 2 diabetes and ESKD undergoing PD, the use of SGLT2i was associated with significantly lower risks of all-cause mortality, sepsis, pneumonia, and PD-associated peritonitis, without an increase in adverse events. These findings suggest that SGLT2i may provide meaningful clinical benefits beyond glycemic control in this high-risk population. These associations were consistent across clinical subgroups and unaffected by prior drug exposure, reinforcing the robustness of the findings. Collectively, these data suggest that SGLT2i may offer meaningful survival and infection-related benefits in this high-risk population and provide a foundation for future randomized trials and potential updates to clinical practice guidelines. Supplementary Material ACKNOWLEDGEMENTS Data analysis utilizing the TriNetX platform received approval from the Institutional Review Board of Chi-Mei Hospital (No. 11202-002), as well as approval from the institutional review boards of all participating hospitals. Compliance with both the Health Insurance Portability and Accountability Act and General Data Protection Regulation is maintained by the TriNetX platform [46, 47]. Since the platform consolidates only de-identified data summaries and counts, TriNetX has been granted an informed consent waiver by the Western Institutional Review Board [46]. This study was conducted following the principles of the Declaration of Helsinki. This study was supported by Mrs Hsiu-Chin Lee Kidney Research Foundation. Contributor Information Yi-Ting Chen, Division of Nephrology, Department of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan; Division of Blood Purification, Department of Integrated Diagnostics & Therapeutics, National Taiwan University Hospital, Taipei, Taiwan. Chung-An Wang, Department of Medical Education, Far Eastern Memorial Hospital, New Taipei City, Taiwan. Jui-Yi Chen, Division of Nephrology, Department of Internal Medicine, Chi-Mei Medical Center, Tainan, Taiwan; Department of Health and Nutrition, Chia Nan University of Pharmacy and Science, Tainan, Taiwan. Vin-Cent Wu, Division of Nephrology, Department of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan; NSARF (National Taiwan University Hospital Study Group of ARF), TAIPAI, (Taiwan Primary Aldosteronism Investigators), and Division of Nephrology, Primary Aldosteronism Center (PAC) of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan. AUTHORS' CONTRIBUTIONS Y.-T.C., J.-Y.C., and V.-C.W. designed the studies. Y.-T.C., C.-A.W., and V.-C.W. assisted in the experiments. C.-A.W. and V.-C.W. assisted in the statistical analysis. Y.-T.C. and V.-C.W. drafted the manuscript. CONFLICT OF INTEREST STATEMENT None declared. FUNDING None declared. DATA AVAILABILITY STATEMENT The datasets used and/or analysed during this study consist of aggregate data sourced from the TriNetX platform. Due to TriNetX’s data sharing polices, we do not have access to individual-level data. Requests for access to the datasets should specify the intended use and will be reviewed and responded to within 2 weeks. Once approved, access to the datasets will be granted within 4 weeks. Access may be protection policies. For inquiries regarding access, please contact V.-C.W., the corresponding author, at q91421028@ntu.edu.tw. Source data are provided with this paper. REFERENCES Associated Data Supplementary Materials Data Availability Statement The datasets used and/or analysed during this study consist of aggregate data sourced from the TriNetX platform. Due to TriNetX’s data sharing polices, we do not have access to individual-level data. Requests for access to the datasets should specify the intended use and will be reviewed and responded to within 2 weeks. Once approved, access to the datasets will be granted within 4 weeks. Access may be protection policies. For inquiries regarding access, please contact V.-C.W., the corresponding author, at q91421028@ntu.edu.tw. Source data are provided with this paper.