Effects of SGLT-2 inhibitors on thyroid-stimulating hormone and systemic immune-inflammation index in type 2 diabetes: a retrospective study

Abstract

Background/Aim

This retrospective study examined whether long-term SGLT-2i use is associated with differences in thyroid-stimulating hormone (TSH) and the systemic immune-inflammation index (SII) in patients with type 2 diabetes mellitus (T2DM) without known thyroid disease. The neutrophil-to-lymphocyte ratio (NLR), renal function (eGFR), diabetes duration, and metformin use were also evaluated as potential covariates.

Methods

This was a single-center, retrospective observational study. Adults with T2DM who had used empagliflozin or dapagliflozin for at least six months were compared with an age- and sex-matched control group of patients with T2DM not using SGLT-2i, drawn from the same outpatient clinics. Key exclusion criteria included known thyroid disease, known positivity for thyroid autoantibodies, active inflammatory or autoimmune processes, advanced organ failure, and immunomodulatory treatment. TSH, SII, and NLR were measured. Statistical analyses included intergroup comparisons, correlation, multivariate linear regression, post-hoc power calculation, and multicollinearity diagnostics (variance inflation factor, VIF).

Results

The study included 330 patients across the empagliflozin (n = 139), dapagliflozin (n = 85), and control (n = 106) groups. Median TSH (1.6, 1.8, and 1.9 µIU/mL; p = 0.338) and SII (481, 531, and 468.5; p = 0.509) did not differ significantly between groups. SII and NLR were significantly correlated (r = 0.467, p < 0.001). In multivariate regression, female sex (B = 0.161, p = 0.015) and eGFR (B=–0.0043, p = 0.037) were independent predictors of log-transformed TSH; SGLT-2i use, SII, NLR, age, metformin use, and diabetes duration were not. The study had 80% power to detect a Cohen’s f of 0.17 or greater; observed effect sizes were substantially below this threshold (f = 0.02 for TSH, f ≈ 0.00 for SII).

Conclusion

Long-term SGLT-2i use was not associated with significant changes in TSH or SII in patients with T2DM without known thyroid disease, and the achieved effect sizes suggest that any true effect, if present, is likely small. Renal function emerged as an independent correlate of TSH, suggesting that the relationship between the hypothalamic-pituitary-thyroid axis and renal physiology merits further attention in this population, independent of SGLT-2i exposure. Prospective, multicenter studies incorporating free thyroxine and triiodothyronine measurements are warranted to further clarify these relationships.

Clinical trial number

This study is retrospective and does not involve a clinical trial registration.

Introduction

Diabetes mellitus (DM) and thyroid dysfunction are among the most common metabolic disorders. According to the International Diabetes Federation, DM is projected to affect more than 629 million people by 2045 [1]. The prevalence of thyroid dysfunction among adults in the United States is 6.6% [2]. Type 2 diabetes mellitus (T2DM) and thyroid dysfunction frequently coexist [3], a coexistence associated with shared metabolic and inflammatory pathways as well as potential drug interactions. Thyroid hormones influence glucose metabolism and energy balance, and diabetes and thyroid dysfunction have similarly been reported to co-occur and associate with one another [4]. Consequently, the clinical importance of screening for thyroid dysfunction and correctly interpreting thyroid function tests in diabetic patients has been emphasized in the literature [5].

Chronic low-grade systemic inflammation plays a central role in the pathogenesis of T2DM and has been shown to adversely affect thyroid function and exacerbate insulin resistance [6]. A significant association has been identified between the Systemic Immune-Inflammation Index (SII) and DM [7], and the SII may serve as a useful biomarker for evaluating the relationship between thyroid dysfunction and diabetes [8]. Although the relationship between SII and TSH has not been examined in detail, inflammation may affect the hypothalamic-pituitary-thyroid (HPT) axis and could influence deiodinase activity and the peripheral conversion of thyroxine (T4) to triiodothyronine (T3) [9].

Antidiabetic agents may themselves affect thyroid function. Metformin has been repeatedly shown to lower TSH levels in patients with T2DM, an effect attributed to a direct central action on TSH secretion rather than to changes in peripheral thyroid hormone levels [10]; this thyrotropin-suppressing effect has been confirmed independently [11].

Sodium-glucose cotransporter-2 inhibitors (SGLT-2i) lower blood glucose by inhibiting renal glucose reabsorption and have demonstrated beneficial effects on body weight, blood pressure, and cardiovascular and renal outcomes [12, 13]. Recent research suggests that SGLT-2i may possess anti-inflammatory properties, and a limited number of clinical and preclinical studies suggest that SGLT-2i may affect the conversion of T4 to T3, altering the free T3/free T4 ratio, with changes in TSH levels observed in some studies [14–16]. Current evidence regarding the effects of SGLT-2i on thyroid function remains insufficient and contradictory, underscoring the need for evaluation following long-term use.

Beyond thyroid-immune interactions, renal function itself has an established, bidirectional relationship with the HPT axis. In patients with T2DM, lower estimated glomerular filtration rate (eGFR) has been independently and negatively associated with TSH, even among patients without overt thyroid dysfunction [17], and this inverse relationship between TSH and eGFR has also been demonstrated in broader chronic kidney disease populations [18]. Because SGLT-2i directly alter renal hemodynamics and glucose handling, and because diabetes duration is itself associated with cumulative microvascular injury, both eGFR and diabetes duration were considered biologically plausible covariates of TSH in this cohort and were incorporated into the present analysis together with metformin use.

In parallel with SII, the neutrophil-to-lymphocyte ratio (NLR) has emerged as a complementary marker of subclinical systemic inflammation, with recent evidence linking NLR to early vascular injury and cardiometabolic risk [19]. Because SII and NLR share neutrophil and lymphocyte counts as common inputs, they would be expected to correlate with one another if both meaningfully capture the same underlying inflammatory dimension; this internal consistency was therefore examined directly in the present study, and NLR was added as an additional inflammatory index alongside SII.

In contrast to the predominantly short-term, single-agent studies reported in the literature, the present study used a minimum treatment duration of six months to allow a more consistent evaluation of potential effects on the HPT axis, and compared two distinct SGLT-2i agents with a control group to assess potential class-versus-agent-specific effects. By incorporating SII and NLR measurements, together with renal function, diabetes duration, and metformin use as covariates, this study aims to provide a more complete account of whether the thyroid-related associations of SGLT-2i are direct, inflammation-mediated, or confounded by other clinically relevant factors.

Materials and methods

Study design and setting

This single-center, retrospective observational study was conducted in the Endocrinology and Metabolism Clinic at Ankara Etlik City Hospital. Ethical approval was obtained from the Ankara Etlik City Hospital Ethics Committee (Approval No: AEŞH-BADEK2-2025-220), in line with the principles of the Declaration of Helsinki. Owing to the retrospective design, data were obtained from the hospital information system and anonymized prior to analysis; the requirement for individual informed consent was therefore waived by the Ethics Committee.

Study population and control group selection

Adults aged ≥ 18 years who presented to the outpatient clinic between January 1, 2024, and April 1, 2024, and were diagnosed with T2DM according to American Diabetes Association (ADA) criteria, were identified through the hospital information system and retrospectively screened.

Diagnostic classification of T2DM followed the ADA Standards of Care in Diabetes—2025 [20], which was the most current guideline available at the time data were analyzed and the manuscript prepared. We note that the core biochemical thresholds used to diagnose T2DM (fasting plasma glucose, 2-hour plasma glucose on oral glucose tolerance testing, and HbA1c) have remained materially unchanged across consecutive ADA Standards of Care updates; consequently, applying the 2025 criteria to data collected in the first quarter of 2024 does not alter case ascertainment in this cohort.

The SGLT-2i cohort comprised patients with continuous SGLT-2i use (empagliflozin or dapagliflozin) for at least 6 months before the index thyroid evaluation, as documented in the electronic medical record. The control cohort was drawn from the same outpatient clinics and the same recruitment window, and consisted of consecutively presenting patients with T2DM who were not receiving an SGLT-2i at the time of evaluation; no separate 1:1 matching algorithm was applied, and the control group was not matched preferentially to either the empagliflozin or the dapagliflozin subgroup specifically. Age- and sex-comparability across all three groups was instead verified statistically post hoc (Table 1); the same inclusion and exclusion criteria, detailed below, were applied uniformly to all three groups.

Table 1.

Baseline demographic and clinical characteristics of the study population

Variable	Dapagliflozin (n = 85)	Empagliflozin (n = 139)	Control (n = 106)	p-value
Age, years, median [IQR]	61.0 [56.0–65.0]	59.0 [52.0–63.5]	59.0 [48.2–64.0]	0.069
Female sex, n (%)	55 (64.7%)	72 (51.8%)	57 (53.8%)	0.149
Diabetes duration, years, median [IQR]*	15.0 [8.2–20.0]	12.0 [7.0–18.0]	10.0 [4.0–12.0]	< 0.001
eGFR, mL/min/1.73 m², median [IQR]	89.0 [65.0–100.0]	97.0 [82.0–104.0]	94.5 [79.2–105.0]	0.003
Metformin use, n (%)	61 (71.8%)	118 (84.9%)	77 (72.6%)	0.025
Retinopathy, n (%)	16 (18.8%)	21 (15.1%)	8 (7.5%)	0.063
Nephropathy, n (%)	14 (16.5%)	15 (10.8%)	11 (10.4%)	0.360
Neuropathy, n (%)	35 (41.2%)	42 (30.2%)	33 (31.1%)	0.203
CAD, n (%)	23 (27.1%)	24 (17.3%)	19 (17.9%)	0.167
CVE, n (%)	0 (0.0%)	0 (0.0%)	1 (0.9%)	0.347
PAD, n (%)	2 (2.4%)	3 (2.2%)	3 (2.8%)	0.943

Abbreviations: IQR, interquartile range; eGFR, estimated glomerular filtration rate; CAD, coronary artery disease; CVE, cerebrovascular event; PAD, peripheral arterial disease. *Diabetes duration available in 277/330 patients (83.9%); percentages and median/IQR for this variable are calculated on available cases

All patients in the SGLT-2i cohorts were receiving these agents at the maximum recommended daily dose (empagliflozin 25 mg/day; dapagliflozin 10 mg/day) at the time of the index evaluation.

Inclusion and exclusion criteria

Only cases with thyroid function tests (TSH ± free T4/T3) and thyroid ultrasonography performed during the same clinical timeframe were included in the analysis. Demographics, comorbidities, glycemic indices, and concomitant medications were recorded using a standardized data form.

Exclusion criteria were: known thyroid disease (hypothyroidism, hyperthyroidism, Graves’ disease, Hashimoto’s disease, or prior thyroid surgery or ablation); known positivity for thyroid autoantibodies (TPOAb and/or TgAb); pregnancy; type 1 diabetes; known hypothalamic–pituitary disorders; and use of medications known to markedly affect thyroid function (e.g., amiodarone, lithium, high-dose glucocorticoids). Additionally excluded were patients with acute infections in the preceding 2–4 weeks, active autoimmune or inflammatory disease exacerbations, recent surgical intervention, trauma or blood transfusion history, active malignancy or antineoplastic treatment, hematological disease (leukemia, lymphoma, myeloproliferative disease, severe thrombocytopenia/thrombocytosis), use of immunomodulatory drugs or systemic corticosteroids, severe chronic organ failure (decompensated liver disease, cirrhosis, advanced renal or heart failure), a history of hypersplenism or splenectomy, and severe alcohol or substance use disorder. The study flowchart is shown in Fig. 1.

Fig. 1.

Patient selection and creation of the study population

Free T4 and/or free T3 values were available only in a subset of patients, reflecting the retrospective nature of data collection and variability in ordering practice; for this reason, these values were not analyzed as a continuous outcome. Where free T4 was available in patients with TSH values in the subclinical hypothyroid range, free T4 was within the normal reference range, consistent with the clinical impression of euthyroidism at the time of testing; however, because this observation was not systematic across the full cohort, it is reported narratively rather than as a quantitative result. Consequently, the present study should be understood as an evaluation of TSH specifically, rather than a comprehensive assessment of thyroid hormone status.

Variables and definitions

The following variables were evaluated: age, sex, SGLT-2i use and agent type, metformin use, diabetes duration, estimated glomerular filtration rate (eGFR), comorbid conditions, thyroid ultrasonography, TSH, glycated hemoglobin (HbA1c), neutrophil and lymphocyte counts, the neutrophil-to-lymphocyte ratio (NLR), the systemic immune-inflammation index (SII), microvascular complications (retinopathy, nephropathy, neuropathy), and macrovascular complications (coronary artery disease [CAD], cerebrovascular event [CVE], peripheral arterial disease [PAD]).

SII was calculated using the standard formula:

SII was calculated for each patient from complete blood count results using the standard formula, consistent with its original description [21].

eGFR was obtained from the hospital laboratory information system as a pre-calculated value derived from serum creatinine at the time of measurement.

Diabetes duration was obtained from clinical records; this variable was unavailable for 53 of 330 patients (16.1%), and an available-case approach was applied, with the resulting reduction in analytic sample size reported alongside each relevant result.

Laboratory methods

Plasma glucose was measured using the hexokinase-based enzymatic reference method (Roche Cobas platform). Serum creatinine was measured using the Jaffe method (kinetic colorimetric analysis, Roche Cobas). Liver enzymes (ALT, AST) were measured using photometric reaction rate analysis (Roche Cobas). Serum TSH was measured using electrochemiluminescence immunoassay (ECLIA, Roche Cobas). Hemoglobin was measured using the sodium lauryl sulfate (SLS) method, and leukocyte, neutrophil, lymphocyte, and platelet counts were obtained by flow cytometry (Sysmex XN series). HbA1c was measured by high-performance liquid chromatography (HPLC, Lifotronic analyzer).

Statistical analysis

All statistical analyses were performed using IBM SPSS Statistics, version 27 (IBM Corp., Armonk, NY, USA). Continuous variables were tested for normality using the Shapiro–Wilk test and, owing to non-normal distributions, reported as median (interquartile range, IQR). Categorical variables were summarized as counts and percentages. Group comparisons for continuous variables across the three groups (empagliflozin, dapagliflozin, control) were performed using the Kruskal–Wallis test. Categorical variables were compared using the chi-square test. Correlations were assessed using Spearman’s rank correlation coefficient (rho), including a correlation between SII and NLR to assess the internal consistency of the two inflammatory indices.

Because serum TSH concentrations were not normally distributed, values were log-transformed before regression modeling. Univariate linear regression analyses were conducted to identify factors associated with log-transformed TSH. Variables that were statistically significant in univariate analysis (p < 0.05) or considered clinically relevant based on prior literature (sex, age, SGLT-2i use, SII) were entered into a multivariate linear regression model. Cerebrovascular event (CVE) and peripheral arterial disease (PAD) were excluded from multivariate modeling owing to their very low event counts in this cohort (1 and 8 patients, respectively), which violate conventional events-per-variable (EPV ≥ 10) recommendations for stable regression estimation; these variables are instead reported descriptively in Table 1. Multicollinearity among multivariate model predictors was assessed using the variance inflation factor (VIF), with VIF > 5 considered indicative of a relevant collinearity concern.

Post-hoc statistical power was calculated for the primary group comparisons (TSH and SII across the three exposure groups) based on the observed Kruskal–Wallis test statistic, converted to an eta-squared effect size estimate and Cohen’s f, using the noncentral F-distribution (df1 = 2, df2 = 327, α = 0.05). The minimum effect size detectable with 80% power under the same design was also calculated for reference. Results were presented as unstandardized coefficients (B), standard errors (SE), standardized coefficients (β), 95% confidence intervals, t-values, and corresponding p-values. A two-tailed p < 0.05 was considered statistically significant throughout.

This study did not involve a clinical trial registration.

Results

The study included 330 patients in the empagliflozin (n = 139), dapagliflozin (n = 85), and control (n = 106) groups. Baseline demographic and clinical characteristics are presented in Table 1. The three groups were comparable with respect to age, sex distribution, TSH, HbA1c, SII, NLR, neutrophil count, and the prevalence of most diabetic complications (nephropathy, neuropathy, CAD, CVE, PAD), with retinopathy showing a borderline trend toward imbalance (p = 0.063). However, diabetes duration, eGFR, lymphocyte count, and metformin use differed significantly across groups: diabetes duration was longer in the SGLT-2i-treated groups than in controls (dapagliflozin 15.0 years, empagliflozin 12.0 years, control 10.0 years; p < 0.001).

Thyroid function, systemic inflammation, and glycemic control parameters are summarized in Table 2. Median TSH levels were similar across the dapagliflozin, empagliflozin, and control groups (1.8 [1.2–2.7], 1.6 [1.2–2.6], and 1.9 [1.3–2.8] µIU/mL, respectively; p = 0.338). SII (p = 0.509) and NLR (p = 0.110) likewise did not differ significantly between groups. Median HbA1c was numerically higher in the dapagliflozin group than in the empagliflozin and control groups, but this difference was not statistically significant (p = 0.123). Lymphocyte count differed significantly between groups (p < 0.001), while neutrophil count did not (p = 0.631).

Table 2.

Comparison of thyroid function, ınflammatory ındices, and glycemic control among study groups

Variable	Dapagliflozin (n = 85)	Empagliflozin (n = 139)	Control (n = 106)	p-value
TSH, µIU/mL, median [IQR]	1.8 [1.2–2.7]	1.6 [1.2–2.6]	1.9 [1.3–2.8]	0.338
SII, median [IQR]	531.0 [364.0–697.0]	481.0 [356.0–638.5]	468.5 [363.8–679.5]	0.509
NLR, median [IQR]	1.9 [1.6–3.0]	2.0 [1.6–2.4]	1.9 [1.4–2.3]	0.110
HbA1c, %, median [IQR]	8.3 [7.2–9.9]	7.8 [6.8–9.2]	7.9 [6.4–9.6]	0.123
Neutrophil, ×10³/µL, median [IQR]	4.5 [3.5–5.8]	4.7 [3.8–6.0]	4.7 [3.6–5.8]	0.631
Lymphocyte, ×10³/µL, median [IQR]	2.1 [1.7–2.6]	2.4 [2.0–2.9]	2.5 [2.1–3.2]	< 0.001

Abbreviations: TSH, thyroid-stimulating hormone; SII, systemic immune-inflammation index; NLR, neutrophil-to-lymphocyte ratio; HbA1c, hemoglobin A1c; IQR, interquartile range

Correlation analyses between TSH and selected clinical and laboratory parameters are shown in Table 3. TSH was not significantly correlated with SII, age, HbA1c, diabetes duration, NLR, neutrophil count, or lymphocyte count (all p > 0.05). A significant negative correlation was observed between TSH and eGFR (r=–0.120, p = 0.029). SII and NLR were significantly and positively correlated with one another (r = 0.467, p < 0.001).

Table 3.

Correlation analysis between serum TSH levels and clinical/laboratory parameters (Spearman’s rho)

Variable	Correlation coefficient (r)	p-value	n
TSH ↔ SII	–0.032	0.568	330
TSH ↔ Age	0.045	0.411	330
TSH ↔ HbA1c	–0.008	0.892	330
TSH ↔ eGFR	–0.120	0.029	330
TSH ↔ Diabetes duration	–0.033	0.582	277
TSH ↔ NLR	–0.096	0.083	330
TSH ↔ Neutrophil count	–0.085	0.125	330
TSH ↔ Lymphocyte count	0.057	0.302	330
SII ↔ NLR (internal consistency)	0.467	< 0.001	330

Abbreviations: TSH, thyroid-stimulating hormone; SII, systemic immune-inflammation index; eGFR, estimated glomerular filtration rate; NLR, neutrophil-to-lymphocyte ratio

Because serum TSH concentrations were not normally distributed, values were log-transformed before regression analysis. As shown in Table 4, univariate linear regression indicated that female sex (B = 0.133, p = 0.026), NLR (B=–0.065, p = 0.027), and eGFR (B=–0.003, p = 0.019) were significantly associated with log-transformed TSH. No significant univariate associations were observed for age, SGLT-2i use, HbA1c, SII, diabetes duration, metformin use, or the evaluated complications (retinopathy, nephropathy, neuropathy, CAD).

Table 4.

Univariate linear regression analysis of factors associated with log-transformed TSH

Variable	B	SE	β	p-value	n
Gender (female vs. male)	0.133	0.059	0.122	0.026	330
Age (years)	0.002	0.003	0.043	0.442	330
SGLT-2 inhibitor use	–0.065	0.064	–0.056	0.309	330
HbA1c (%)	–0.001	0.015	–0.004	0.946	330
SII	< 0.001	< 0.001	–0.078	0.159	330
NLR	–0.065	0.029	–0.122	0.027	330
eGFR (mL/min/1.73 m²)	–0.003	0.002	–0.129	0.019	330
Diabetes duration (years)	–0.002	0.004	–0.023	0.703	277
Metformin use	–0.042	0.071	–0.033	0.553	330
Retinopathy (yes vs. no)	–0.040	0.087	–0.025	0.649	330
Nephropathy (yes vs. no)	0.042	0.091	0.026	0.642	330
Neuropathy (yes vs. no)	–0.065	0.063	–0.057	0.301	330
CAD (yes vs. no)	–0.038	0.074	–0.028	0.610	330

Abbreviations: B, unstandardized regression coefficient; SE, standard error; β, standardized regression coefficient; SGLT-2, sodium-glucose cotransporter-2; HbA1c, hemoglobin A1c; SII, systemic immune-inflammation index; NLR, neutrophil-to-lymphocyte ratio; eGFR, estimated glomerular filtration rate; CAD, coronary artery disease. CVE and PAD were excluded from regression modeling owing to extremely low event counts (see Statistical Analysis)

Variables that were statistically significant in univariate analysis, together with variables considered clinically relevant a priori (age, SGLT-2i use, SII, metformin use, diabetes duration), were entered into the multivariate model (Table 5; n = 277 owing to missing diabetes duration data). After adjustment, female sex remained an independent predictor of higher log-transformed TSH (B = 0.161, β = 0.150, p = 0.015). eGFR also emerged as an independent predictor (B=–0.0043, β=–0.161, p = 0.037), indicating that lower renal function was independently associated with higher TSH even after accounting for sex, age, SGLT-2i use, SII, NLR, metformin use, and diabetes duration. NLR showed a trend toward an independent association that did not reach statistical significance (p = 0.093). SGLT-2i use, age, SII, metformin use, and diabetes duration were not independently associated with log-TSH. The overall model explained a modest proportion of variance (R²=0.067, adjusted R²=0.039).

Table 5.

Multivariate linear regression analysis of factors associated with log-transformed TSH (n = 277)

Variable	B	SE	β	95% CI	p-value	VIF
Gender (female vs. male)	0.161	0.066	0.150	0.032, 0.290	0.015	1.08
Age (years)	0.0005	0.004	0.009	–0.008, 0.009	0.903	1.71
SGLT-2 inhibitor use	–0.017	0.069	–0.015	–0.153, 0.119	0.807	1.07
SII	< 0.001	< 0.001	–0.065	–0.0004, 0.0001	0.316	1.21
NLR	–0.060	0.035	–0.112	–0.129, 0.010	0.093	1.27
eGFR (mL/min/1.73 m²)	–0.0043	0.002	–0.161	–0.008, − 0.0003	0.037	1.71
Metformin use	–0.024	0.080	–0.019	–0.181, 0.132	0.762	1.16
Diabetes duration (years)	–0.0054	0.005	–0.079	–0.015, 0.004	0.243	1.31

Abbreviations: B, unstandardized regression coefficient; SE, standard error; β, standardized regression coefficient; CI, confidence interval; VIF, variance inflation factor; SII, systemic immune-inflammation index; NLR, neutrophil-to-lymphocyte ratio; eGFR, estimated glomerular filtration rate. Model: R²=0.067, adjusted R²=0.039. All VIF values were below 2, indicating no relevant multicollinearity among model predictors despite the correlation between SII and NLR

Post-hoc power analysis

Post-hoc power analysis based on the observed Kruskal–Wallis statistics indicated low achieved power for the primary intergroup comparisons: 6.3% for TSH (Cohen’s f = 0.02) and 5.0% for SII (Cohen’s f ≈ 0.00). Given the present sample size (N = 330) and three-group design, the study had 80% power to detect a Cohen’s f of 0.17 or greater (a small-to-medium effect by conventional standards). The observed effect sizes for TSH and SII were considerably smaller than this threshold, suggesting that if a true difference in TSH or SII exists between SGLT-2i users and non-users, its magnitude is likely to be very small and of limited clinical relevance, rather than reflecting a meaningfully under-powered design.

Discussion

This single-center, retrospective study found that long-term SGLT-2i use was not associated with significant differences in TSH or SII in patients with T2DM without known thyroid disease, and that the achieved statistical power for these comparisons was low, with observed effect sizes well below the threshold the study was designed to detect. Beyond this primary negative finding, two results merit closer interpretation: female sex and eGFR were independent predictors of TSH in multivariate analysis, and SII and NLR were significantly correlated with one another, supporting their combined use as complementary inflammatory indices. Rather than reading the absence of an SGLT-2i effect in isolation, we interpret these findings together as indicating that, within this cohort, variation in TSH is more plausibly explained by sex and renal function than by SGLT-2i exposure itself or by the inflammatory markers examined.

The absence of a significant SII difference between SGLT-2i users and non-users is consistent with a body of evidence indicating that the anti-inflammatory effects of this drug class are heterogeneous and marker-specific rather than global. A meta-analysis of randomized controlled trials found limited evidence that SGLT-2i improve classical inflammatory biomarkers such as C-reactive protein, despite favorable effects on adipokine profiles and insulin sensitivity [22], and a separate meta-analysis similarly found no significant reduction in CRP, IL-6, or TNF-α with SGLT-2i treatment [23]. By contrast, other meta-analyses have reported that SGLT-2i significantly reduce circulating IL-6 [24] and improve markers of oxidative stress, including malondialdehyde and superoxide dismutase [25]. Direct measurement of CRP, IL-6, or TNF-α was not available in the present retrospective dataset, and this represents a limitation when interpreting our SII-based findings against this more granular cytokine literature; we are therefore unable to determine whether SGLT-2i exerted effects on these specific mediators that were not captured by SII. Taken together with our results, this pattern suggests that any anti-inflammatory action of SGLT-2i, if present, is more likely to be confined to specific cytokine pathways than to manifest as a measurable shift in a composite hematological index such as SII over a minimum six-month exposure window.

Regarding thyroid function specifically, our finding of no significant TSH difference associated with SGLT-2i use is only partially concordant with the existing, still limited literature. One study reported that empagliflozin increased T3 levels and the T3/T4 ratio without significantly changing TSH [15], a pattern that a TSH-only analysis would not detect, and one that reinforces the importance of including peripheral thyroid hormone measurements in future studies. Another retrospective study reported a non-significant decrease in TSH in SGLT-2i-treated patients [16]. Differences in study design, population characteristics, treatment duration, and the specific agents evaluated likely contribute to this heterogeneity across studies, and our findings should be interpreted as one additional, TSH-focused data point within this still-developing literature rather than as a definitive resolution. Of note, metformin has been repeatedly shown to lower TSH in T2DM, and given that its use differed significantly across groups — empagliflozin-treated patients had a higher rate of metformin co-prescription (84.9%) than the dapagliflozin (71.8%) or control (72.6%) groups (p = 0.025) — metformin exposure represents an important potential confounder. This imbalance may partly reflect the clinical availability in Türkiye of fixed-dose empagliflozin–metformin combination preparations; it was addressed by incorporating metformin use as a covariate in the multivariate model, where it was not independently associated with TSH after adjustment.

The independent inverse association between eGFR and TSH observed in this cohort is, to our knowledge, a more substantively interpretable finding than the primary negative result, and is biologically well-grounded rather than incidental. A large cohort study in patients with T2DM similarly found that lower eGFR was independently and negatively associated with TSH, even in the absence of overt thyroid dysfunction [17], and a comparably inverse relationship between renal function and thyroid status has been demonstrated across the broader spectrum of chronic kidney disease severity [18]. Proposed mechanisms for this relationship include reduced renal clearance of iodine and thyroid hormone metabolites, altered peripheral deiodinase activity, and changes in thyroid-binding protein handling as glomerular filtration declines. Because both diabetes duration and SGLT-2i exposure are themselves correlated with renal trajectory, this relationship reinforces the importance of having incorporated eGFR as a covariate in the present analysis: without this adjustment, any unmeasured renal confounding could plausibly have been misattributed to SGLT-2i exposure itself. In this cohort, after such adjustment, SGLT-2i use was not independently associated with TSH, while renal function was — a result that, in our view, more directly addresses the clinical question of whether SGLT-2i affect the thyroid axis than the unadjusted group comparison alone.

SII is a composite, readily available hematological index, but it remains a relatively nonspecific surrogate for the compartmentalized, multi-pathway nature of inflammation in T2DM, which involves innate immune activation, cytokine signaling, oxidative stress, and tissue-specific responses that a single platelet-neutrophil-lymphocyte ratio cannot fully capture. Markers reflecting myeloid compartment activation, such as NLR, have been more specifically associated with early vascular injury and cardiometabolic risk independently of traditional inflammatory markers [19], which is the rationale for our inclusion of NLR alongside SII in this analysis rather than relying on SII in isolation. The significant positive correlation between SII and NLR observed here (r = 0.467, p < 0.001) confirms that the two indices capture a shared inflammatory dimension, as expected given their overlapping cellular components, while their distinct associations with TSH in univariate analysis (NLR significant, SII not) suggest that they are not fully redundant and may differ in sensitivity to specific physiological processes. Consequently, the absence of a between-group difference in SII does not, on its own, exclude a more targeted anti-inflammatory action of SGLT-2i that a broader cytokine panel or a more granular inflammatory index might detect; this is an explicit limitation of relying on composite hematological indices as the primary inflammatory readout, which we acknowledge here rather than overstating the mechanistic conclusiveness of a null SII finding.

Inflammation and oxidative stress are increasingly recognized as contributors to diabetic microvascular complications beyond their role in glycemic pathophysiology alone, with mechanistic work implicating these pathways in the development and progression of diabetic nephropathy [26], obesity-associated glomerular injury and albuminuria [27], and retinal vascular and inflammatory changes [28]. These mechanistic findings, while derived predominantly from preclinical models, support the broader rationale for evaluating inflammatory indices such as SII and NLR in a population with diabetic microvascular and macrovascular complications, as in the present cohort. With respect to SGLT-2i specifically, proposed renoprotective and anti-inflammatory mechanisms include hemodynamic effects on intraglomerular pressure and metabolic actions favoring reduced tubular workload [29], with renoprotective benefit increasingly recognized even in the absence of diabetes [30]. At the cellular level, SGLT-2i have been shown to activate AMPK signaling and promote autophagy in immune cells [31–33], to favorably modulate macrophage polarization from a pro-inflammatory M1 toward an anti-inflammatory M2 phenotype [34, 35], and to reduce M1-associated inflammatory burden in adipose tissue and liver in experimental models [13, 36]. The absence of an observable SII difference in our cohort does not contradict these mechanistic data; rather, it suggests that such cellular and tissue-level anti-inflammatory effects, even if biologically present, did not translate into a measurable shift in a circulating composite hematological index over the exposure window captured in this retrospective sample, and may instead be more evident in tissue-specific or cytokine-specific readouts than in SII.

More broadly, inflammatory status in T2DM is increasingly interpreted within an integrated cardiometabolic framework that also incorporates direct vascular assessment. Early vascular alterations, such as arterial stiffness and carotid intima-media thickness, have been shown to correlate strongly with cardiovascular risk stratification and may identify subclinical damage that isolated biochemical markers do not [37]. Vascular imaging was not available in this retrospective cohort, and we therefore cannot determine whether the absence of an SII or NLR difference between SGLT-2i users and controls corresponds to an absence of vascular benefit or harm; this should be considered an important avenue for prospective study designs that combine inflammatory indices with direct vascular assessment.

Limitations

This study has some limitations. The retrospective, single-center design precludes causal inference, and the absence of pre-treatment baseline measurements prevents within-patient comparisons. Residual confounders including BMI, blood pressure, liver enzymes, statin use, smoking status, and insulin resistance severity were not consistently captured and could not be incorporated into the analysis. Free T4 and free T3 values were unavailable for the full cohort; patients with known thyroid autoantibody positivity (TPOAb and/or TgAb) were excluded by design, but testing was not performed in all patients. Diabetes duration data were missing in 16.1% of patients, and CVE and PAD were excluded from regression modeling owing to insufficient event counts. Finally, the single time-point measurement of SII and NLR limits the detection of subtle differences.

Conclusıon

This study contributes to a still-limited literature by demonstrating that long-term SGLT-2i use does not appear to exert a measurable effect on TSH or composite inflammatory indices in patients with T2DM, and by highlighting the independent contribution of renal function to TSH variation in this population — a relationship that warrants attention in its own right, independent of SGLT-2i exposure. The observed correlation between SII and NLR supports their complementary use as inflammatory markers in future research. Prospective, multicenter studies with standardized thyroid panels, treatment duration data, and direct inflammatory and vascular assessments are needed to further delineate these relationships.

Abbreviations

ADA

American Diabetes Association

AMPK

AMP-activated protein kinase

AST/ALT

Aspartate aminotransferase / Alanine aminotransferase

BMI

Body mass index

CAD

Coronary artery disease

CI

Confidence interval

CKD

Chronic kidney disease

CRP

C-reactive protein

CVE

Cerebrovascular event

DM

Diabetes mellitus

ECLIA

Electrochemiluminescence immunoassay

eGFR

Estimated glomerular filtration rate

FT3 / FT4

Free triiodothyronine / Free thyroxine

HbA1c

Glycated hemoglobin

HPLC

High-performance liquid chromatography

HPT axis

Hypothalamic-pituitary-thyroid axis

IL-6

Interleukin-6

IQR

Interquartile range

NLR

Neutrophil-to-lymphocyte ratio

PAD

Peripheral arterial disease

SE

Standard error

SGLT-2i

Sodium-glucose cotransporter-2 inhibitor(s)

SII

Systemic immune-inflammation index

T2DM

Type 2 diabetes mellitus

TgAb

Thyroglobulin antibody

TNF-α

Tumor necrosis factor-alpha

TPOAb

Thyroid peroxidase antibody

TSH

Thyroid-stimulating hormone

VIF

Variance inflation factor

Author contributions

All authors contributed to the study conception and design, and participated in data interpretation. All authors read and approved the final version of the manuscript.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data availability

The data that support the findings of this study were obtained from Ankara Etlik City Hospital under ethical approval. Due to the sensitive nature of patient data, they are not publicly available. However, anonymized data may be available from the corresponding author upon reasonable request and with permission from the institutional ethics committee.

Declarations

Ethics approval

This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Ankara Etlik City Hospital (Approval No: AEŞH-BADEK2-2025-220).

Consent to participate

Due to the retrospective nature of the study, data were obtained from the hospital information system and anonymized before analysis. Therefore, the requirement for individual informed consent was waived by the Ethics Committee.

Competing interests

The authors declare no competing interests.