Association of sodium-glucose co-transport protein 2 inhibitor use with clinical outcomes in patients receiving immune checkpoint inhibitors: a pan-tumor propensity-matched analysis

Abstract

Background

Recently, sodium–glucose cotransporter-2 inhibitors (SGLT2i) have emerged as a treatment option for type 2 diabetes mellitus (T2DM) and have also shown promising anti-tumor activity in preclinical models. Limited clinical data exist regarding the use of SGLT2i in patients with metastatic solid tumor malignancies treated with immune checkpoint inhibitors (ICIs).

Patients and Methods

A retrospective, multi-center, matched cohort analysis of patients with a diagnosis of either T2DM and/or congestive heart failure and an advanced solid tumor malignancy treated with ICIs was performed using the Epic Cosmos database. Ten different solid tumor types, 11 ICIs, and 3 SGLT2i were analyzed. Patients in the exposure group also received an SGLT2i and had at least 1 overlapping cycle with ICI.

Results

Of the 1146 patients included, 573 patients received SGLT2i + ICIs and 573 received ICIs alone. Compared to patients treated with ICIs, individuals who received SGLT2i + ICIs had improved median overall survival (mOS) (27.4 vs 17.4 months, hazard ratio 0.68, log-rank test, P < .001). When stratified by histology, significant improvements in mOS were seen for non-small-cell lung cancer and renal cell carcinoma patients. Survival benefit was attenuated when accounting for immortal time bias in late initiators of SGLT2i, though persisted among patients prescribed SGLT2i at or prior to initiation of ICI. Patients in the exposure group experienced longer time on immunotherapy compared to the control group.

Conclusion

Patients prescribed SGLT2i at or prior to ICI initiation had a modest improvement in survival compared to patients treated with ICIs alone. SGLT2i may be a beneficial adjunctive therapy in patients treated with ICI, though further studies are needed.

Implications for Practice.

In this multicenter, propensity-matched, retrospective study, concomitant use of sodium-glucose cotransporter-2 inhibitors (SGLT2i) with immune checkpoint inhibitors (ICIs) was associated with improved overall survival in patients with metastatic solid tumors, particularly non-small cell lung cancer and renal cell carcinoma. Survival benefit was attenuated when accounting for immortal time bias in late initiators of SGLT2i, though persisted among patients prescribed SGLT2i at or prior to ICI initiation. These findings suggest that SGLT2i may provide further benefit in patients with cancer and type 2 diabetes while maintaining established metabolic and cardiovascular advantages. Prospective clinical trials are needed to validate these results.

Introduction

Following their approval for the treatment of metastatic melanoma in 2011, immune checkpoint inhibitors (ICIs) have rapidly altered the therapeutic landscape for many solid tumor malignancies.^1–3 By blocking inhibitory immunologic molecules such as programmed cell death 1 (PD-1), programmed cell death ligand 1 (PD-L1), cytotoxic T-lymphocyte–associated protein 4 (CTLA-4), and lymphocyte activation gene 3 (LAG-3), ICIs promote an antitumor immune response through re-activation of exhausted T-cells and other effects on the tumor immune microenvironment (TME).^4–7 The United States Food and Drug Administration (FDA) has approved use of multiple ICIs alone or in combination in over 20 different malignancies.^8,9 These agents are commonly used in the first-line metastatic setting, particularly in melanoma, renal cell carcinoma (RCC), and non-small-cell lung cancer (NSCLC).^1,10–18 However, through overactivation of the immune system, ICIs can produce autoimmune-like side effects, termed immune-related adverse events (irAEs), impacting virtually any organ system within the body.^19–21

While use of ICIs improves outcomes in many malignancies, there continues to be significant heterogeneity in response rates. These differences are complex and likely multifactorial, due in part to variations in the TME, tumor mutational burden, gut microbiome, and patients’ underlying comorbidities, among other factors.^22,23 Type 2 diabetes mellitus (T2DM) is well-established as a relevant comorbidity in patients with cancer and has been associated with increased risk for development of several malignancies, including pancreatic, endometrial, colon, and breast cancer.^24–27 Due to increasing prevalence of T2DM world-wide, it is estimated that 15% of patients have comorbid T2DM at the time of cancer diagnosis.²⁸ Moreover, Cortellini et al. also showed that concomitant use of glucose-lowering medications, particularly metformin, is associated with worse outcomes in cancer patients with comorbid T2DM treated with ICI.²⁹

Recently, sodium–glucose cotransporter 2 inhibitors (SGLT2i) have emerged as a treatment option for the management of T2DM.^30–32 SGLT2i inhibit sodium-dependent glucose reabsorption in renal proximal tubules and were initially approved as antihyperglycemic medications.³³ Since their initial approval, they have been studied for their far-reaching effects beyond glycemic control. Landmark studies such as the EMPEROR and DAPA-HF trials revealed cardioprotective effects in congestive heart failure (CHF) patients, irrespective of diabetes status, prompting US FDA approval for use in CHF with reduced or preserved ejection fraction.^34–37 These studies also demonstrated decreased rates of renal function deterioration in patients with chronic kidney disease (CKD).^38–41 SGLT2i are generally well-tolerated, with potential side effects including euglycemic diabetic ketoacidosis (DKA) and genitourinary infections.⁴²

Following the observed improvements in clinical outcomes with use of SGLT2i for the management of T2DM, CHF, and CKD, SGLT2i are under investigation for potential anti-neoplastic properties.^43,44 Pre-clinical research involving in-vitro and in-vivo models suggests that SGLT2i prevent and slow tumor growth across multiple cancer types.^45–50 Retrospective clinical studies show that patients with NSCLC or hepatocellular carcinoma (HCC) and preexisting diabetes who received SGLT2i in conjunction with their cancer treatments experienced improved overall survival (OS).^51,52

The expanding overlap between metabolic and oncologic pathways underscores the need to understand how therapies such as SGLT2i may impact cancer-directed therapies, including ICIs. However, clinical data examining the use of SGLT2i in patients with advanced malignancies receiving ICIs remain limited. The primary objective of this study was to evaluate the association between SGLT2i use and OS in patients treated with ICIs for advanced solid tumors. The secondary objectives were to assess the effect of SGLT2i use on time on immunotherapy (TOI) and to evaluate the safety of concomitant SGLT2i and ICI use.

Methods

Study populations

A cohort of patients from the Epic Cosmos database and a separate cohort of patients from our single institution were analyzed in this retrospective study. Epic Cosmos is a collaborative dataset created by Epic Systems comprised of 296 million patients treated at over 1700 hospitals and 39 000 clinics across all 50 states, Washington, DC, Saudi Arabia, and Lebanon that use Epic’s electronic health record software. The eligible diagnosis window spanned January 1, 2013 to January 1, 2025. Patients from our single institution were excluded from the survival analysis and only included in the toxicity analysis. All patients in our study were diagnosed with one of the following malignancies: RCC, small-cell lung cancer (SCLC), non-small-cell lung cancer (NSCLC), hepatocellular carcinoma (HCC), esophageal cancer (adenocarcinoma and squamous cell carcinoma histologies) urothelial cell carcinoma, cutaneous melanoma, gastric cancer (adenocarcinoma histology only), Merkel cell carcinoma (MCC), and endometrial cancer. SNOMED CT codes were used to identify patients with the corresponding cancer type within the Epic Cosmos database. Patients with multiple malignancies were excluded. The SNOMED CT codes used for identifying specific cancer types and comorbidities can be found in Tables S1 and S2, respectively. Information pertaining to treatment-related adverse events (TRAEs) was obtained by review of medical records. IrAEs were identified graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events, version 5.0.

All patients included in this study had American Joint Committee on Cancer (AJCC) stage IV disease and at least one of the following co-morbidities prior to their stage IV cancer diagnosis: T2DM or CHF. Patients without AJCC staging information or those with AJCC Stage I, II, or III disease were excluded. All patients received at least 1 cycle of treatment with at least one of the following ICI: pembrolizumab, atezolizumab, nivolumab, ipilimumab, ipilimumab + nivolumab, durvalumab, durvalumab + tremelimumab, avelumab, nivolumab-relatlimab, and tremelimumab. ICI types were defined by the first ICI administered after metastatic cancer diagnosis. Patients who received chemotherapy in addition to ICI were included.

Patients in the exposure group also received an SGLT2i and had at least 1 overlapping cycle with ICI; the SGLT2i included were canagliflozin, dapagliflozin, and empagliflozin. Overlapping cycles were defined as periods when a patient was treated with an ICI and was prescribed an SGLT2i simultaneously. Patients in the control group did not receive an SGLT2i at any point. Information pertaining to medication use was identified from medication records in Epic Cosmos using generic drug names. For each patient, the start and end dates of all relevant orders were extracted. For ICIs, the start date of each order was treated as the administration date. For SGLT2i, both start and end dates were used to define the duration of therapy. After applying all inclusion and exclusion criteria, the exposure group included 573 patients, and the control group included 5563 patients (see Figure S1 [see online supplementary material for a color version of this figure] for details regarding sample selection).

Statistical analysis

A 1:1 propensity score matching was performed based on the following variables: malignancy type, comorbidities (T2DM and CHF), age at malignancy diagnosis, sex, and time from metastatic malignancy diagnosis to ICI initiation. Patients with missing data for any variable used in the propensity score calculation were excluded (Figure S1, see online supplementary material for a color version of this figure). After 1:1 matching, the final matched sample included 1146 patients, with 573 in the exposure group and 573 in the control group. The density function of propensity scores for both groups was closely aligned after matching, indicating successful matching between the 2 groups (Figure S2, see online supplementary material for a color version of this figure). The subsequent analysis of OS and subgroup analyses was conducted using the matched sample (n = 1146).

Kaplan–Meier (KM) analysis was conducted on the matched cohort to report OS and TOI outcomes. OS was defined as time (in months) from date of metastatic malignancy diagnosis to date of death (if applicable). Patients alive at the end of the study period (May 15, 2025) were censored. For censored patients, survival time was calculated as the time from date of metastatic malignancy diagnosis to date of last recorded follow-up while alive. TOI was defined as the period from the first ICI administration to 3 weeks after treatment discontinuation for any reason. Any gaps between the end of one ICI administration’s effective period, which was considered to be 3 weeks post-treatment, and the start of the next cycle of the same ICI, were excluded. For patients who received more than 1 ICI, the total TOI was defined as the sum of the treatment periods for each ICI. Patients who were alive and within the 3-week therapeutic window after last ICI administration at the end of the study period were censored. Cox proportional hazards modeling was used to calculate hazard ratios (HR); all statistical analyses were conducted using R software.

To evaluate for the potential impact of immortal time bias, sensitivity analyses were performed, stratifying the exposure group by timing of SGLT2i initiation relative to ICI initiation. Concurrent initiators were defined as patients who were prescribed SGLT2i at or prior to ICI start, and late initiators were defined as patients who initiated SGLT2i at any point after ICI start. Cox proportional hazards models were run separately for each group using ICI initiation as the time origin. A time-dependent covariate Cox proportional hazards model was additionally performed for the late initiator cohort and the full matched cohort. SGLT2i overlap was treated as a time-varying covariate, transitioning from 0 (no overlap) to 1 (overlap) at the date of first overlapping SGLT2i and ICI use. Cox proportional hazards models were additionally adjusted for metformin and GLP-1 receptor agonist use.

Results

Patient characteristics

This retrospective, matched cohort study included 573 patients in the exposure group (SGLT2i + ICI) and 573 patients in the control group (ICI). Patient characteristics are summarized in Table 1. The mean age of patients was 67.7 years (standard deviation [SD], 8.5 years) in the exposure group and 67.6 years (SD 8.4 years) in the control group. Male gender was more common in both cohorts. The exposure group had higher baseline body mass indices (2-sample t-test, P = .002) and hemoglobin A1c values (HbA1c) (2-sample t-test, P < .001) compared to the control group. Baseline T2DM medications are shown in Table S3; the exposure group had higher rates of metformin, dipeptidyl peptidase-4 (DPP-4), sulfonylurea, glucagon-like-peptide-1 (GLP-1) receptor agonist, and insulin use compared to the control group. No statistically significant differences in cancer type, comorbidities, or ICI type were seen between the 2 cohorts. 378 patients (66%) received empagliflozin, which was the most frequently prescribed SGLT2i in the exposure cohort.

Table 1.

Patient demographics.

Demographic	Exposure (N = 573)	Control (N = 573)	P-value
Sex—no. (%)			.584a
Male	427 (74.5)	436 (76.1)
Female	146 (25.5)	137 (23.9)
Age at diagnosis—years, mean (SD)	67.6 (8.4)	67.7 (8.9)	.845b
Ethnicity—no. (%)			.055a
Hispanic or Latino	27 (4.7)	47 (8.2)
Not Hispanic or Latino	530 (92.5)	510 (89.0)
Unspecified	16 (2.8)	16 (2.8)
Clinical variables c
BMI—mean (SD)	30.77 (6.8)	29.55 (6.5)	.002b
HbA1C—mean (SD)	7.25 (1.5)	6.87 (1.6)	<.001b
Cancer type—no. (%)			.978d
Renal cell carcinoma (RCC)	145 (25.3)	157 (27.4)
Small-cell lung cancer (SCLC)	113 (19.7)	121 (21.1)
Non-small-cell lung cancer (NSCLC)	104 (18.2)	108 (18.8)
Hepatocellular carcinoma (HCC)	65 (11.3)	60 (12.5)
Esophageal cancer	53 (9.2)	44 (7.7)
Gastric cancer	39 (6.8)	37 (6.5)
Melanoma	18 (3.1)	15 (2.6)
Merkel cell carcinoma	17 (3.0)	15 (2.6)
Urothelial cell carcinoma	<11	<11
Endometrial cancer	<11	<11
Comorbidity type—no. (%)			.874a
Type II diabetes mellitus (T2DM)	376 (65.6)	384 (67.0)
Congestive heart failure (CHF)	46 (8.0)	43 (7.5)
Both (CHF and T2DM)	151 (26.4)	146 (25.5)
First ICI type—no. (%)			.334d
Pembrolizumab	213 (37.2)	216 (37.7)
Atezolizumab	142 (24.8)	136 (23.7)
Nivolumab	109 (19.0)	128 (22.3)
Ipilimumab + nivolumab	64 (11.2)	61 (10.6)
Durvalumab	24 (4.2)	17 (3.0)
Durvalumab + tremelimumab	12 (2.1)	<11
Avelumab	<11	<11
Nivolumab-relatlimab	<11	<11
Ipilimumab	<11	<11
Tremelimumab	<11	<11
First SGLT2i type—no. (%)
Empagliflozin	378 (66.0)	–
Dapagliflozin	170 (29.7)	–
Canagliflozin	25 (4.4)	–

^aBased on χ² test.

^bBased on T-test.

^cBased on the most recent value before immunotherapy began.

^dBased on Fisher’s exact test, categories with 0 patients in either the exposure or control cohort were excluded.

Association between SGLT2i use and overall survival

Among all 1146 patients included in the analysis, 517 were censored with an overall censoring rate of 45.1%. The censoring rate was 51.3% (294 out of 573) in the exposure group and 38.9% (223 out of 573) in the control group. The median overall survival (mOS) was 27.4 months (95% CI: 22.6-32.7) in the exposure group compared to 17.4 months (95% CI: 15.0-20.7) in the control group (Figure 1A). The difference in survival between these 2 groups was statistically significant (HR 0.68, 95% CI: 0.58-0.80, log rank test, P < .001), indicating a 32% lower risk of death in patients receiving SGLT2i + ICI. Given that metformin and GLP-1 receptor agonist use were significantly more prevalent in the exposure group and have known or emerging associations with outcomes in patients treated with ICIs, Cox proportional hazards modeling was additionally adjusted for these medications.^29,53 After adjustment, the survival association remained statistically significant (HR 0.83, 95% CI: 0.71-0.98, P = .026).

Figure 1.

Kaplan-Meier survival curves demonstrating improved overall survival with SGLT2i use in patients receiving immune checkpoint inhibitors. Kaplan-Meier curves displaying overall survival stratified by exposure (SGLT2i) for the treatment group (SGLT2i + ICI) and control group (ICI alone) for the entire cohort of patients (A), patients with non-small-cell lung cancer (B), renal cell carcinoma (C), and esophageal cancer (D).

Association between SGLT2i use and overall survival by malignancy type

mOS was estimated using the KM method for each individual malignancy type and results are summarized in Table 2. For patients with NSCLC, the mOS was 32.7 months (95% CI: 15.7-48.6) in the exposure group vs 10.5 months (95% CI: 8.5-17.9) in the control group (log rank test, P = .002) (Figure 1B). For patients with RCC, the mOS was 53.5 months (95% CI: 45.2-not achieved (NA)) in the exposure group vs. 32.7 months (95% CI: 28.4-41.0) in the control group (log rank test, P = .006; Figure 1C). For patients with esophageal cancer, the mOS was 15.6 months (95% CI: 10.4-NA) in the exposure group vs. 11.7 months (95% CI: 7.7-19.3) in the control group (log rank test, P = .046). However, this finding should be interpreted with caution given the small sample size and wide CI of the Cox model estimate approaching the null. No statistically significant differences in mOS were observed between the exposure and control groups for patients with SCLC, HCC, melanoma, MCC, gastric cancer, or other cancer types.

Table 2.

Median overall survival by malignancy type.

Malignancy	Control		Exposure		P-valuea
	Median (months)	95% CI	Median (months)	95% CI
Renal cell carcinoma	32.7	28.4-41.0	53.5	45.2-NAb	.006
Small-cell lung cancer	10.0	8.3-15.6	12.5	9.9-15.5	.53
Non-small-cell lung cancer	10.5	8.5-17.9	32.7	15.7-48.6	.002
Hepatocellular carcinoma	13.7	12.2-28.7	20.7	13.6-27.6	.41
Esophageal cancer	11.7	7.7-19.3	15.6	10.4-NA	.046
Gastric cancer	11.5	8.5-19.7	20.9	12.1-NA	.11
Melanoma	54.7	14.8-NA	NA	NA-NA	.11
Merkel cell carcinoma	15.0	5.4-NA	47.1	28.4-NA	.23
Other cancer typesc	47.8	28.3-NA	NA	43.1-NA	.55

^aBased on log-rank test.

^bNA indicates that the value could not be estimated because more than 50% of patients were still considered alive (censored) at the end of the study period.

^cOther cancer types include endometrial cancer and urothelial cell carcinoma.

Cox proportional hazards modeling was performed for each malignancy type. The HRs supported the findings observed in the KM analysis: for patients with NSCLC, the HR was 0.58 (95% CI: 0.41-0.82, P = .002) and for patients with RCC, the HR was 0.60 (95% CI: 0.42-0.86, P = .006). For patients with esophageal cancer, the HR was 0.59 (95% CI: 0.34-0.996, P = .048).

Sensitivity analyses accounting for timing of SGLT2i initiation

To account for time-related biases inherent to observational data, particularly immortal time bias when exposures occur after cohort entry, sensitivity analyses were performed to evaluate the impact of timing of SGLT2i initiation relative to ICI on the observed survival association. Of the 573 patients in the exposure group, 420 (73.3%) initiated SGLT2i at or prior to ICI (concurrent initiators) and 153 (26.7%) initiated SGLT2i after ICI (late initiators). Median time from metastatic diagnosis to ICI initiation was 23 days in the control group (IQR, 9-64) and 21 days in the exposure group (IQR, 9-53) (Wilcoxon signed-rank test, P = .41). Cox proportional hazards models were run separately for the concurrent and late initiator groups using ICI initiation as the time of origin.

Among the concurrent initiators, adjusted Cox proportional hazards models demonstrated a 17% decrease in the risk of death compared to matched controls (HR 0.83, 95% CI: 0.70-0.98, P = .033; Table S4). Unadjusted models showed a similar decreased risk of death that was not statistically significant (HR 0.86, CI: 0.73-1.02, P = .087). Among the late initiators, adjusted Cox proportional hazards models demonstrated a 64% decrease in the risk of death compared to matched controls (HR 0.36, 95% CI: 0.27-0.47, P < .001; Table S5). Unadjusted models showed a similar, statistically significant decrease in the risk of death (HR 0.35, 95% CI: 0.26-0.47, P < .001). Adjusted and unadjusted time-dependent covariate Cox proportional hazards analyses of the late initiator cohort using ICI initiation as the time of origin subsequently revealed a non-significant decrease in the risk of death (adjusted HR 0.82, 95% CI: 0.61-1.10, P = .20; unadjusted HR 0.86, 95% CI: 0.64-1.16, P = .30).

Time-dependent covariate Cox proportional hazards analysis of the full matched cohort using ICI initiation as the time of origin was also performed and demonstrated no statistically significant difference in the hazard of death between groups (HR 1.00, 95% CI: 0.85-1.17, P > .90, Table S6).

Association between SGLT2i use and overall survival by ICI and SGLT2i type

mOS was estimated using the KM method for each ICI and SGLT2i type in the primary propensity-matched cohort (Figure 2). Due to substantial overlap in the survival curves across groups, the survival differences among the ICI types may not be statistically significant (Figure 2A). The KM survival curves among the different SGLT2i types also exhibited substantial overlap with no statistically significant difference in mOS across the 3 SGLT2i types (log-rank test, P = .74; Figure 2B). To further estimate the treatment effect of SGLT2i, Cox Proportional Hazards modeling was performed for each ICI type. Among the ICI types, pembrolizumab and ipilimumab + nivolumab demonstrated a significant survival improvement for patients who received SGLT2i. Pembrolizumab had an HR of 0.56 (95% CI, 0.43-0.74, P < .001) and ipilimumab + nivolumab had an HR of 0.52 (95% CI, 0.28-0.94, P = .032). No statistically significant improvements in survival were observed amongst the other ICIs.

Figure 2.

Kaplan-Meier survival curves displaying overall survival stratified by ICI and SGLT2i types. Kaplan-Meier curves for the treatment group (SGLT2i + ICI) stratified by ICI type (A) and SGLT2i type (B).

Association between SGLT2i use and time on immunotherapy

Due to a lack of data regarding radiographic and clinical progression in this data set, TOI was evaluated as an exploratory clinical endpoint. KM curves for TOI stratified by SGLT2i exposure are shown in Figure 3. Median TOI in the exposure group was 6.2 months (95% CI: 5.5-6.9) compared to 4.1 months (95% CI: 3.4-4.5) in the control group. The difference in TOI between the 2 groups was statistically significant (log-rank test, P < .001); patients in the exposure group had a 22% lower risk of discontinuing ICI compared to the control group (HR = 0.78, 95% CI: 0.69-0.88).

Figure 3.

Kaplan-Meier curves demonstrating longer time on immunotherapy with SGLT2i use in patients receiving immune checkpoint inhibitors. Kaplan-Meier curves displaying time on immunotherapy stratified by exposure (SGLT2i) for the treatment group (SGLT2i + ICI) and control group (ICI alone) for the entire cohort of patients (A), patients with non-small-cell lung cancer (B), renal cell carcinoma (C), and hepatocellular carcinoma (D).

Median TOI was also calculated for patients stratified by malignancy type. Significantly longer durations of TOI were observed in the exposure group in several malignancy types, including NSCLC (5.5 vs 3.4 months, log-rank test, P = .021), SCLC (4.8 vs 4.1 months, P = .024), HCC (6.2 vs 2.9 months, P = .013), and gastric cancer (6.9 vs 2.8 months, P = .012; Figure 3).

Treatment-related adverse events

Information pertaining to TRAEs was not available in the Epic Cosmos database. 51 patients from our single institution cohort who received SGLT2i + ICI were included in the exploratory, descriptive TRAE analysis; these patients were excluded from the survival analysis. Regarding ICI-related toxicities, 13 irAEs occurred in 12 (23.5%) patients; 1 patient experienced 2 separate irAEs. The most common irAE observed was colitis (n = 5, 38.5%), followed by hepatitis (n = 2, 15.4%), myositis, (n = 1, 7.7%), hypophysitis (n = 1, 7.7%), mucositis (n = 1, 7.7%), dermatitis (n = 1, 7.7%), diabetes mellitus (n = 1, 7.7%), and choroidal inflammation (n = 1, 7.7%). Most irAEs were grade 1 (n = 3, 23.1%) or grade 2 (n = 4, 30.8%); however, grade 3 (n = 4, 30.8%) and grade 4 (n = 2, 15.4%) events were also observed. No renal or cardiac irAEs were observed. The observed rates of irAEs and their severities were similar to previously reported clinical trials and real-world cohorts.⁵⁴ Regarding SGLT2i-related toxicities, euglycemic DKA was observed in 1 patient (2.0%) and genitourinary infections were observed in 3 patients (5.9%), consistent with previously reported rates.^31,55

Discussion

In this large, propensity-matched retrospective analysis of patients with metastatic solid tumors, we observed an apparent association between concomitant SGLT2i use with ICIs and improved mOS in conventional analyses. However, the significant survival benefit observed in the primary analysis was attenuated after performing time-dependent covariate Cox proportional hazards analysis among the late initiators of SGLT2i, which likely reflects substantial immortal time bias in this cohort. Patients who initiated SGLT2i after ICI were required to survive and remain on ICI long enough to receive an SGLT2i, and this pre-SGLT2i survival time was incorrectly attributed to SGLT2i use in the primary conventional analyses, inflating the apparent treatment effect.⁵⁶

Among patients that initiated SGLT2i at or prior to ICI (73.3% of the cohort), however, a modest improvement in mOS persisted after accounting for SGLT2i exposure timing. Immortal time bias is minimized in this subgroup due to the temporal alignment of exposure with cohort entry. The observed survival benefit remained consistent in Cox proportional hazards models adjusted for sex, age at diagnosis, co-morbidities, malignancy type, and time from metastatic diagnosis to ICI initiation. This subgroup represents both the most biologically relevant population for future clinical trials and the most methodologically rigorous subset within this cohort.

Time-dependent covariate Cox proportional hazards analysis of the full matched cohort demonstrated no statistically significant association between SGLT2i use and overall survival (HR 1.00, 95% CI: 0.85-1.17, P > .9). However, this result should be interpreted with caution given an important structural limitation of the model in our dataset. As 420 patients (73.3%) initiated SGLT2i at or prior to ICI initiation and therefore contribute essentially zero unexposed person-time to the time-dependent model from ICI initiation, the model is driven predominantly by the 153 late initiators in terms of unexposed person-time estimation, limiting the reliability and generalizability of this estimate. For the concurrent initiator group, in whom immortal time bias is zero by construction, a standard Cox model from ICI initiation is the more appropriate and reliable analytical approach, demonstrating a modest but statistically significant survival association (HR 0.83, 95% CI: 0.70-0.98, P = .033).

To our knowledge, this study is the largest to date to evaluate the impact of SGLT2i use among a wide variety of cancers treated with ICIs, several of which have not been previously studied in this context. Our results are consistent with a small but growing body of retrospective data suggesting that SGLT2i use in patients with malignancy may be associated with improved outcomes.^57,58 Luo et al.⁵¹ recently reported significantly reduced mortality with SGLT2i use in patients with NSCLC and preexisting diabetes, although few patients on ICIs were included. Similarly, Flausino et al. demonstrated reductions in all-cause mortality, hospitalization rates, and adverse events in a study of patients with gastrointestinal malignancy treated with chemotherapy with or without radiotherapy in addition to SGLT2i.⁵⁹ Improvements in mortality have also been observed with SGLT2i use in patients with HCC and endometrial cancer.^52,60 Perelman et al. reported lower all-cause mortality in patients with cancer and T2DM prescribed SGLT2i and ICIs, although only 24 patients were included in the SGLT2i group.⁶¹ To our knowledge, prior retrospective studies evaluating SGLT2i in this context have not routinely used time-dependent exposure modeling, despite its relevance for addressing exposure timing bias. No prospective randomized clinical trials have investigated the anticancer effects of SGLT2i to date.

Multiple mechanisms by which SGLT2i suppress, and slow tumor growth have been proposed. SGLT2 has been identified as a positive regulator of PD-L1 expression via colocalization with PD-L1 at the plasma membrane, protecting it against ubiquitination and proteasome-mediated degradation.⁴⁸ SGLT2i, such as canagliflozin, have been shown to enhance anti-tumor immune response by triggering downregulation and decreased expression of PD-L1, thereby leading to increased activity of antitumor cytotoxic CD8+ T cells. When examined in-vivo, use of the SGLT2i canagliflozin in conjunction with an anti-CTLA-4 mAb synergistically activated T cells in a tumor xenograft model and significantly reduced the size of tumors compared with either canagliflozin or anti-CTLA-4 mAb alone. Canagliflozin has been shown to increase infiltration of CD4+ and CD8+ T-cells leading to inhibition of tumor growth in sarcoma mouse models.⁶² The modest survival association observed among concurrent initiators in our study is consistent with these proposed immunologic mechanisms of SGLT2i. SGLT2i also inhibit the phosphoinositide 3-kinase/protein kinase B (PI3K/AKT) pathway, one of the most frequently dysregulated pathways in cancer, leading to reduced cellular proliferation and increased cell death.^63,64 Sodium–glucose cotransporters, including SGLT1 and SGLT2, have been shown to be expressed on the cell surface of multiple malignancies, including NSCLC, RCC, HCC, pancreatic, prostate, and breast cancer.^65–69 Use of SGLT2i may inhibit glucose uptake via these transporters, attenuating cancer cell growth. The preferential expression of SGLT2 in NSCLC and RCC may provide a biological basis for the histology-specific survival benefit observed. Finally, SGLT2 has also been shown to be expressed in microglia/macrophages surrounding tumor vasculature and in endothelial cells of the microvasculature in glioblastoma multiforme. While endothelial expression of SGLT2 has not been confirmed in other tumor types, these findings suggest that use of SGLT2i could affect endothelial proliferation, intratumoral angiogenesis, and alter the TME, though further studies are needed.⁷⁰

Historically, patients with T2DM and cancer have worse outcomes, including higher all-cause mortality and reduced treatment efficacy.^25,71,72 Our findings suggest that SGLT2i may offer oncologic advantages for these patients while also providing their established metabolic and cardiovascular benefits, though prospective validation is required. Given the potential differential effects of individual glucose-lowering medications on ICI outcomes, the choice of antihyperglycemic agents in patients with T2DM receiving ICI may warrant prospective investigation, with SGLT2i representing a candidate of particular interest given their metabolic, cardiovascular, and potentially immunologic properties.²⁹

Treatment-related adverse events occurring in the context of co-administration of SGLT2i and ICIs have not previously been reported. Given the mechanism of action of SGLT2i, concerns regarding increased renal irAEs, specifically acute tubulointerstitial nephritis, when used in conjunction with ICIs have been proposed.⁴⁴ We did not observe any renal irAEs in our exploratory analysis. Although our toxicity analysis was small and lacked a matched comparator group, these descriptive data suggest that co-administration of SGLT2i and ICI did not result in unexpected adverse events, supporting the clinical feasibility of future prospective trials. Overall, the rates of irAEs stemming from ICI use and euglycemic DKA and genitourinary infections resulting from SGLT2i use were like the landmark trials leading to the approval of these medications, suggesting no increased risk of TRAEs based on this retrospective data.^36,38,73

There are multiple limitations to this analysis. Although propensity score matching was applied to reduce confounding, key clinical variables such as performance status, metastatic disease burden, and line of therapy in which patients received ICIs were not available and therefore not controlled for. These unmeasured confounders may influence mOS and TOI, and the magnitude of the observed survival difference warrants careful interpretation considering these potential confounders. Time-dependent covariate Cox proportional hazards analysis among the late initiators of SGLT2i demonstrated significant attenuation of the apparent survival benefit, consistent with immortal time bias applying to this subgroup within the primary analysis. Although baseline clinical characteristics were well-matched, the exposure group had higher baseline BMI and HbA1c values than the control group, which may have influenced SGLT2i prescribing and other factors. While these values were significantly different statistically, we note that they are unlikely be significantly different clinically and represent similar patient populations with well-controlled T2DM overall.

Additionally, several cancer subtypes included in the KM survival analyses had small sample sizes, limiting the survival estimates for those subgroups. Our study was not designed to determine if the observed survival benefit from SGLT2i use was due to treatment of the patient’s co-morbid CHF or T2DM versus potential antineoplastic effects of this class of medications, although longer observed TOI may support the latter. Furthermore, patients prescribed SGLT2i are likely to have greater interaction with the healthcare system leading to closer longitudinal follow-up and improved access to multidisciplinary care. These factors may independently influence survival and treatment continuation irrespective of any direct biologic effect of SGLT2i. Time on treatment has been used in some real-world oncology studies, it may be influenced by factors unrelated to disease progression, such as toxicity, patient choice, or logistical issues and must be interpreted cautiously considering these limitations. Finally, the full cohort time-dependent analysis should be interpreted with caution, as 73.3% of patients in the exposure group initiated SGLT2i before or at time of ICI and therefore contribute minimal unexposed person-time to the model, limiting the reliability of this estimate.

Conclusion

We observed an association between SGLT2i use and improved mOS in multiple metastatic solid tumor malignancies treated with ICI in this real-world, propensity matched analysis. The observed survival association persisted in sensitivity analyses for patients prescribed SGLT2i at or prior to ICI initiation. These findings warrant validation in prospective clinical trials and further clinical and preclinical mechanistic studies to better elucidate the underlying metabolic and immunologic pathways involved. This concept, linking metabolic modulation with ICI via SGLT2i, is supported by preclinical evidence that SGLT2i can modulate PD-L1 expression, enhance T cell activity, and influence tumor metabolism.

Author contributions

Sean C. Dougherty (Conceptualization, Project administration, Writing—original draft, Writing—review & editing), Sara Young (Conceptualization, Writing—original draft, Writing—review & editing), Hector Picon (Conceptualization, Writing—original draft, Writing—review & editing), Qingyi He (Data curation, Formal analysis, Investigation, Writing—original draft, Writing—review & editing), Asal Pilehvari (Data curation, Formal analysis, Writing—original draft, Writing—review & editing), Wen You (Data curation, Formal analysis, Writing—original draft, Writing—review & editing), Richard D. Hall (Conceptualization, Methodology, Project administration, Supervision, Writing—review & editing), and Matthew J. Reilley (Conceptualization, Methodology, Supervision, Writing—review & editing)

Supplementary material

Supplementary material is available at The Oncologist online.

Funding

Data analysis costs were supported by the University of Virginia Comprehensive Cancer Center National Cancer Institute Support Grant (P30CA44579), Farrow-Weiss Endowment Fund, and University of Virginia Department of Medicine Physicians Scientist Training Program. The Population Health and Cancer Outcomes Core at the University of Virginia Comprehensive Cancer Center provided research support for data analysis and interpretation.

Conflicts of interest

S.C.D. reports the following conflicts of interest: honoraria from OncLive; travel support to meetings from OncLive and International Association for the Study of Lung Cancer. R.D.H. reports the following conflicts of interest: research funding to the institution from Amgen, AstraZeneca, Alliance Foundation, Big Ten Research Consortium, Daiichi Sankyo, ECOG-ACRIN, Genentech, Hoosier Cancer Research Network, Lilly, Merck, Mirati, NCI, Regeneron; consulting fees from Bayer, EMD Serono, Jazz, Regeneron, Takeda. M.R. reports the following conflicts of interest: research funding/support from Astrazeneca, Amgen, Abbvie, Cardiff Oncology, Crinetics, Genfit, Macrogenics, Merck, Pfizer, RayzeBio, Surface Oncology, Xencor; consulting fees from Cardiff Oncology, Incyte, Natera, NEJM, Pfizer. All other authors have no disclosures.

Data availability

All data and material are available upon request. Please contact the corresponding author for further information.