Mortality and persistence in therapy in elderly subjects with type 2 diabetes receiving gliflozins or incretins

Abstract

Purpose

Evidence supporting the use of sodium-glucose cotransporter 2 inhibitors (SGLT2i) and glucagon-like peptide-1 receptor agonists (GLP1-RA) in individuals aged ≥ 80 years remains limited, particularly regarding treatment persistence and real-world outcomes.

Methods

In this retrospective real-world study, 292 individuals with type 2 diabetes (T2D) aged ≥ 80 years who started SGLT2i (n = 155) or GLP1-RA (n = 137) after the age of 75 were stratified by age at treatment initiation (75–80 vs. > 80). Primary outcomes were all-cause mortality and treatment persistence; longitudinal renal function was also assessed. Survival analyses were performed using Cox models adjusted for confounders.

Results

Over a median follow-up of 40 (37–43) months, 47 deaths (16.1%) occurred. The incidence rate of mortality was 37.5 and 68.6 events/1000 person-years in the GLP1-RA and SGLT2i groups, respectively (adjusted HR = 0.58, 95% CI [0.28–1.21]; p = 0.148). Stratifying participants by age at first prescription, those starting either GLP1-RA or SGLT2i at age 75–80 showed similar mortality. In contrast, subjects starting GLP1-RA vs. SGLT2i after 80 years showed a higher survival rate (adjusted HR = 0.33, 95% CI [0.12–0.91]; p = 0.034); significance disappeared after further adjustment for cardiovascular burden. Treatment discontinuation was 14% with GLP1-RA and 20% with SGLT2i (HR 0.53, 95% CI [0.30–0.95]; p = 0.032), due to a better persistence in GLP1-RA of more elderly subjects. Longitudinal eGFR trajectories were similar between treatments.

Conclusion

GLP1-RA and SGLT2i were associated with similar all-cause mortality risk when started in very late age. Kidney function over time was comparable. Persistence on SGLT2i was lower when therapy was started after the age of 80. These findings highlight the need for individualized treatment decisions in older, frail patients and for prospective studies on glucose-lowering therapies in very old populations.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s40618-026-02994-1.

Introduction

Type 2 diabetes (T2D) is highly prevalent in older adults and represents a major contributor to cardiovascular morbidity, renal disease and all-cause mortality in this population. As life expectancy increases, the number of very old individuals living with T2D continues to rise, posing important challenges for individualized therapeutic decision-making [1, 2]. In this setting, treatment strategies should aim not only at glycaemic control but also at improving clinically meaningful outcomes while maintaining long-term tolerability and treatment persistence, both needing to be improved [3].

Sodium-glucose cotransporter-2 inhibitors (SGLT2i) and glucagon-like peptide-1 receptor agonists (GLP1-RA) have consistently demonstrated cardiovascular and renal benefits in people with T2D, often associated with improved survival, and these effects have been confirmed across a broad range of patient profiles in randomized controlled trials and real-world studies [4–7]. However, individuals of very advanced age, particularly those aged ≥ 80 years, have been largely underrepresented in pivotal trials, limiting the generalizability of these findings to the oldest segments of the T2D population.

The clinical management of T2D in very old adults is complicated by age-related physiological changes, a high burden of comorbidities, polypharmacy, and increased susceptibility to adverse events [8]. Furthermore, adherence and persistence to glucose-lowering therapies may substantially influence long-term outcomes in this age group, yet comparative data on treatment discontinuation between SGLT2i and GLP1-RA remain scarce [9, 10], especially in very elderly individuals. Whether the balance between benefits, tolerability and persistence differs between these two drug classes according to age at treatment initiation is still unclear.

Real-world evidence may provide valuable insights into the effectiveness and tolerability of glucose-lowering therapies in populations that are poorly represented in clinical trials. In particular, the evaluation of outcomes in relation to age at first prescription may help to better inform therapeutic choices in the oldest patients with T2D. Therefore, the aim of this retrospective real-world study was to compare clinical outcomes, longitudinal renal function and treatment persistence in individuals with T2D aged over 80 years who initiated therapy with either a SGLT2i or a GLP1-RA after the age of 75 years. We additionally explored whether age at treatment initiation was associated with differential effects on mortality and treatment discontinuation between the two therapeutic strategies.

Subjects and methods

This retrospective, monocentric real-world observation includes all T2D individuals aged at least 80 years consecutively attending the Diabetes Outpatient Clinic of Internal Medicine section of the University Hospital in Pisa between January 2014 and March 2025, who received a prescription of either SGLT2i or GLP1-RA after the age of 75 years based on clinical judgment, either for poor metabolic control, cardiorenal protection and/or reduction of body weight. Exclusion criteria were history of previous discontinuation of SGLT2i or incretins and denied informed consent to participate in this observation. Based on the above reported inclusion criteria, selected participants were stratified by age at first prescription: 75–80 years (SGLT2i_75−80y and GLP1-RA_75−80y) or ≥ 80 years (SGLT2i_>80y and GLP1-RA_>80y).

Medical history, clinical information and blood laboratory test were collected at baseline (T0) and during follow-up visits performed in routine clinical practice. Follow-up assessments were not scheduled at fixed time intervals but occurred according to real-world care (every 6 ± 2 months). For survival and treatment persistence analyses, participants were followed from therapy initiation until death, treatment discontinuation, or administrative censoring at 60 months, whichever occurred first.

Persistence in therapy over the observational period was assessed during each visit, which reported home therapy and the possible cause of therapeutic changes.

Vital status for all the participants was retrieved at July 2025 from the official platform of the National Health System. The study was approved by the local Ethics Committee (CET25/2024); participants provided their informed consent.

Biochemistry

Routine blood laboratory tests included fasting glucose, glycated haemoglobin, lipid profile and serum creatine; all these determinations were performed at the certified and centralized laboratories of the Tuscany Health System. Estimated glomerular filtration rate (eGFR) was calculated by the 2021 Epidemiology Collaboration (CKD-EPI) equation [11]. Albumin to creatinine ratio (ACR) was measured on a spot morning urine sample.

Statistical analysis

Continuous variables are presented as mean ± standard deviation or median with interquartile range according to distribution normality assessed using the Shapiro–Wilk test. Categorical variables are reported as numbers and percentages and compared using the χ² test or Fisher’s exact test, as appropriate. Between-group comparisons of continuous variables were performed using Student’s t-test or Mann–Whitney U test, as appropriate.

Time-to-event analyses for all-cause mortality and treatment discontinuation were conducted using Kaplan-Meier survival curves and Cox proportional hazards regression models. Hazard ratios (HRs) with 95% confidence intervals were estimated. Multivariable models were adjusted for age, sex, BMI. A sensitivity analysis was also performed including baseline cardiovascular disease in the models. The proportional hazards assumption was verified using Schoenfeld residuals.

Patients were censored at 60 months of follow-up or at last available observation. Median follow-up time was calculated by the reverse Kaplan-Meier method. Univariable and multivariable Cox regression analyses adjusted for age and sex were performed to identify clinical predictors of outcome.

Longitudinal changes in renal function were analyzed using linear mixed-effects models with random intercepts for subjects, including time, treatment group and their interaction as fixed effects, with additional adjustment for age, sex and BMI.

All statistical tests were two-tailed, and a p value < 0.05 was considered statistically significant. Analyses were performed using JMP pro version 18 (SAS Institute, Cary, NC, USA).

Results

Based on the above-reported selection criteria, 292 T2D individuals were included in the analysis; 155 were treated with SGLT2i and 137 received GLP1-RA therapy. Clinical characteristics of the participants at baseline (T0) are shown in Table 1. Age at first prescription was slightly but significantly lower in the GLP1-RA group (p = 0.0002), while sex and diabetes duration did not differ between the two groups. Subjects on GLP1-RA had higher triglycerides (p = 0.0005) and BMI (p < 0.0001) compared to SGLT2i, but showed, in their medical history, a lower rate of cardiovascular disease (23.4 vs. 45.8%, p < 0.0001). No difference emerged in fasting glucose, HbA1c, total cholesterol, LDL cholesterol and eGFR.

Table 1.

Clinical characteristics of the study participants at baseline. Data are expressed as mean ± SD, median [IQ range] or number (percentage), as appropriate

	SGLT2i n = 155	GLP1-RA n = 137	p
Age, years	82.4 ± 2.9	81.1 ± 1.8	< 0.001
Women, n (%)	55 (35.5)	63 (46)	0.068
Diabetes duration, years	18.9 ± 10.9	20.3 ± 11.5	0.277
Cardiovascular disease, %	45.8	23.4	< 0.001
BMI, Kg/m2	27.8 ± 4.9	31.0 ± 5.7	< 0.001
SBP, mmHg	149 ± 19	153 ± 20	0.181
DBP, mmHg	78 ± 10	76 ± 12	0.147
Fasting glucose, mg/dL	149 ± 41	159 ± 49	0.061
HbA1c, mmol/mol	58.1 ± 13.4	61.2 ± 15.0	0.077
Serum creatinine, mg/ml	1.1 ± 0.4	1.1 ± 0.4	0.500
eGFR, mL/min/1.73 m2	64.0 ± 18.2	61.4 ± 19.7	0.278
ACR, mg/g	54 [8-162]	20 [8–68]	0.191
Total cholesterol, mg/dL	155 ± 35	156 ± 31	0.913
HDL cholesterol, mg/dL	50 ± 14	48 ± 13	0.216
LDL cholesterol, mg/dL	78 ± 32	75 ± 27	0.149
Triglycerides, mg/dL	99 [75–140]	142 [93–194]	< 0.001

BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate; ACR, albumin-to-creatinine ratio

After stratification by age at first prescription (Table 2), no significant differences were observed between GLP1-RA_75−80y vs. SGLT2i_75−80y, except for BMI (p = 0.0007) and lipid parameters, with higher triglyceride levels and lower LDL cholesterol in the GLP1-RA group (p = 0.0046 and p = 0.0458, respectively). Other baseline clinical and metabolic variables, including prevalence of cardiovascular disease, glycemic control, and renal function, were comparable between groups. At odd, among individuals aged > 80 years, the GLP1-RA group exhibited a higher proportion of women (46.9% vs. 32.3%, p = 0.047), a lower prevalence of cardiovascular disease (18.5% vs. 49%, p < 0.0001) and a higher BMI (p = 0.0010). No significant differences were observed in plasma glucose, blood pressure, lipid profile or renal function.

Table 2.

Clinical characteristics of the study participants at baseline according to age. Data are expressed as mean ± SD, median [IQ range] or number (percentage), as appropriate

	SGLT2i 75–80 y n = 59	GLP1-RA 75–80 y n = 56	p	SGLT2i > 80 y n = 96	GLP1-RA > 80 y n = 81	p
Age (> 80 y), years	80	80	-	83.9 ± 2.8	81.8 ± 2	< 0.001
Women, n (%)	24 (40.7)	25 (44.6)	0.667	31 (32.3)	38 (46.9)	0.047
Diabetes duration, years	20.3 ± 11.2	20.7 ± 10.6	0.498	17.9 ± 10.6	20 ± 12.1	0.234
Therapy duration, months	26.4 ± 14.4	27.7 ± 13.8	0.633	0	0	-
Cardiovascular disease, %	40.7	30.4	0.248	49	18.5	< 0.001
BMI, Kg/m2	28.3 ± 4.5	31.6 ± 5	0.001	27.6 ± 5	30.6 ± 6.1	0.001
SBP, mmHg	148 ± 17	154 ± 23	0.213	150 ± 21	152 ± 18	0.476
DBP, mmHg	74 ± 8	77 ± 12	0.877	79 ± 11	76 ± 12	0.063
Fasting glucose, mg/dL	149 ± 34	168 ± 60	0.057	148 ± 44	152 ± 37	0.583
HbA1c, mmol/mol	56.3 ± 10.2	60.7 ± 16.2	0.099	59.1 ± 15.1	61.5 ± 14.1	0.314
Serum creatinine, mg/ml	1.06 ± 0.33	1.15 ± 0.43	0.228	1.13 ± 0.4	1.13 ± 0.44	0.972
eGFR, mL/min/1.73 m2	66.2 ± 18.4	62.0 ± 18.8	0.265	62.6 ± 17.9	60.9 ± 20.3	0.589
ACR, mg/g	22 [7–71]	21 [10–100]	0.823	62 [8-296]	14 [5–58]	0.071
Total cholesterol, mg/dL	161 ± 33	160 ± 31	0.504	151 ± 35	155 ± 32	0.576
HDL cholesterol, mg/dL	52 ± 13	46 ± 12	0.034	50 ± 14	50 ± 14	0.915
LDL cholesterol, mg/dL	86 ± 30	74 ± 25	0.046	77 ± 29	75 ± 29	0.799
Triglycerides, mg/dL	94 [75–128]	159 [109–221]	0.004	105 [76–144]	122 [85–189]	0.103

BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate; ACR, albumin-to-creatinine ratio

All-cause mortality by treatment group

Over a median follow-up of 40 (95% CI 37–43) months, 47 deaths (16.1%) occurred (53.5 events/1,000 person-years). The incidence rate of mortality was 37.5 and 68.6 events/1,000 person-years in the GLP1-RA and SGLT2i groups, respectively.

Data for survival rates are shown in Fig. 1. When comparing the two treatments (Fig. 1a), the GLP1-RA group showed a strong trend to a better survival compared with SGLT2i (crude HR = 0.55, 95% CI [0.30–1.00]; p = 0.052); such trend disappeared after adjustment for age, sex and BMI (adj HR = 0.58, 95% CI [0.28–1.21]; p = 0.148).

Fig. 1.

Kaplan–Meier curves for all-cause mortality according to treatment group. (A) Overall population; (B) participants aged 75–80 years; (C) participants aged > 80 years. Blue lines indicate GLP1-RA and red lines SGLT2i. Tick marks indicate censored observations. Hazard ratios (HR) with 95% confidence intervals (CI) are shown. P values refer to the log-rank test

When participants were stratified by age at first prescription (Fig. 1b), individuals starting either GLP1-RA or SGLT2i between 75 and 80 years of age showed similar mortality, with incidence rates of 42.2 and 29.1 events/1,000 person-years respectively (crude HR 1.41, 95% CI 0.45–4.45; p = 0.558). In contrast, subjects starting GLP1-RA after 80 years showed a higher survival rate than those receiving SGLT2i, with incidence rates of 34.5 and 89.6 events/1,000 person-years, respectively (crude HR 0.37, 95% CI 0.18–0.80; p = 0.011) (Fig. 1c). Such difference persisted after adjustment for age, sex and BMI (adj HR = 0.33, 95% CI [0.12–0.91]; p = 0.034), but not when analysis was further adjusted for cardiovascular burden (adj HR = 0.71, 95% CI [0.39–1.31]; p = 0.121).

Treatment persistence and discontinuation

We evaluated whether persistence on therapy might have played a role in determining the observed difference in all-cause mortality. Fifty subjects suspended treatment during the observation; overall, treatment discontinuation was lower with GLP1-RA (14%) than with SGLT2i (20%) (crude HR 0.53, 95% CI [0.30–0.95]; p = 0.032) (Fig. 2a). The analysis by prescription age group showed that persistence was similar in GLP1-RA_75−80y vs. SGLT2i_75−80y: crude HR 1.21, 95% CI [0.48–3.08]; p = 0.683 (Fig. 2b), whereas it was greater in GLP1-RA_>80y than SGLT2i_>80y: crude HR 0.32, 95% CI [0.15–0.68]; p = 0.004 (Fig. 2c). Suppl Table 1 shows distribution and dosage of the different GLP1-RA and SGLT2i at baseline and at the last visit: no major variations in these parameters were observed.

Fig. 2.

Kaplan–Meier curves for treatment persistence according to treatment group. (A) Overall population; (B) participants aged 75–80 years; (C) participants aged > 80 years. Blue lines indicate GLP1-RA and red lines SGLT2i. Tick marks indicate censored observations. Hazard ratios (HR) with 95% confidence intervals (CI) are shown. P values refer to the log-rank test

Clinical predictors of outcomes

Table 3 shows the clinical predictors of the study outcomes (all-cause mortality and treatment discontinuation). In univariable Cox analyses, age was significantly associated with an increased risk of all-cause mortality (HR 1.12, 95% CI [1.03–1.21]; p = 0.008). Among the other variables, higher baseline eGFR was associated with reduced mortality risk (HR per 10 mL/min/1.73 m² increase = 0.84, 95% CI [0.71–0.99]; p = 0.032), and this association remained significant after adjustment for age and sex (HR 0.83, 95% CI [0.71–0.98]; p = 0.031). Higher BMI was also associated with lower mortality risk in univariable analysis (HR per 5 kg/m² increase = 0.74, 95% CI [0.54–0.99]; p = 0.049), although this association was not retained in the multivariable model (p = 0.147). No significant associations were observed for the other biochemical variables. Regarding treatment discontinuation (Table 3B), no clinical or biochemical variables were significantly associated with the risk of therapy discontinuation in univariable Cox analyses, although age (HR 1.09, 95% CI [1.00–1.20]; p = 0.056) showed a borderline association. In the multivariable analysis adjusted for age and sex, no independent predictors of discontinuation emerged; however, triglycerides showed a borderline association with increased discontinuation risk (HR 1.93, 95% CI [0.98–3.78]; p = 0.056).

Table 3.

Clinical predictors of mortality (A) and treatment discontinuation (B): univariable and multivariable Cox regression analyses. Multivariable models were adjusted for age and sex

Factors	Univariate analysis		Multivariable analysis
	Crude HR (95% CI)	p value	Adjusted HR (95% CI)	p value
(A)
Age, per year	1.12 (1.03–1.21)	0.008	-	-
Sex, female	0.59 (0.32–1.12)	0.108	-	-
BMI, per 5 kg/m2	0.74 (0.54–0.99)	0.049	0.79 (0.57–1.08)	0.147
Fasting glucose, per 10 mg/dL	1.03 (0.95–1.08)	0.476	1.02 (0.94–1.09)	0.487
HbA1c, per 10 mmol/mol	0.94 (0.75–1.18)	0.640	0.99 (0.79–1.23)	0.921
eGFR, per 10 ml/min/1.73 m2	0.84 (0.71–0.99)	0.032	0.83 (0.71–0.98)	0.031
LDL cholesterol, per 10 mg/dL	0.94 (0.83–1.07)	0.386	0.95 (0.83–1.07)	0.394
Triglycerides, log transformed	1.49 (0.74–2.99)	0.264	1.67 (0.84–3.13)	0.149
(B)
Age, per year	1.09 (1.00–1.20)	0.056	-	-
Sex, female	0.76 (0.42–1.35)	0.357	-	-
BMI, per 5 kg/m2	0.77 (0.57–1.03)	0.083	0.81 (0.59–1.10)	0.181
Fasting glucose, per 10 mg/dL	1.01 (0.94–1.07)	0.770	1.02 (0.94–1.07)	0.652
HbA1c, per 10 mmol/mol	0.96 (0.77–1.17)	0.654	0.99 (0.80–1.20)	0.922
eGFR, per 10 ml/min/1.73 m2	0.89 (0.75–1.05)	0.151	0.90 (0.76–1.06)	0.190
LDL cholesterol, per 10 mg/dL	1.01 (0.90–1.14)	0.827	1.01 (0.89–1.14)	0.821
Triglycerides, log transformed	1.74 (0.89–3.39)	0.105	1.93 (0.98–3.78)	0.056

eGFR, estimated glomerular filtration rate; HR, hazard ratio; CI, confidence interval

Longitudinal trajectories of eGFR

Given the established cardio-renal effects of both drug classes, longitudinal changes in renal function were also evaluated. Longitudinal eGFR trajectories were similar between treatments, with no significant difference in slope over time in linear mixed-effects models, both unadjusted and after adjustment for age, sex, and BMI (treatment x time interaction p = 0.302).

Discussion

In this real-world cohort of very elderly individuals with T2D, we compared survival and renal effects in subjects initiating SGLT2i and GLP1-RA after the age of 75 years, paying attention to the role of therapeutic compliance.

Large trials and meta-analyses have demonstrated that SGLT2i and GLP1-RA significantly reduce major adverse cardiovascular events and, in pooled analyses, all-cause mortality [4, 7]. Although both drug classes are recommended within a cardio-renal framework independently of HbA1c, very elderly individuals remain underrepresented in randomized trials and clinical studies [12, 13], limiting direct extrapolation to this population. In fact, subjects aged ≥ 75 years account for a small fraction of participants, and subgroup analyses in very old individuals are often underpowered [14, 15]. This uncertainty is clinically relevant because frailty, multimorbidity, polypharmacy and competing mortality risks increasingly modify treatment effects in advanced age [8]. Our findings offer valuable insights into the real-world effectiveness of SGLT2i and GLP1-RA in this vulnerable, often underrepresented, patient population.

While age at first prescription was slightly, but significantly, lower in the GLP1-RA group, other crucial baseline characteristics such as sex, diabetes duration, persistence in therapy, survival time, and the number of deaths were comparable between the two groups. Notably, GLP1-RA users exhibited higher triglycerides and, as anticipated, a higher BMI, consistent with the established metabolic profiles often targeted by these drug classes [16]. The absence of significant differences in fasting glucose, HbA1c, total cholesterol, LDL cholesterol, and eGFR at baseline across the groups suggests a relatively balanced comparison regarding key metabolic and renal parameters. This may reflect prescribing patterns where GLP1-RA are more often chosen for patients with higher BMI due to their ability to promote satiety and significant weight loss [17], while SGLT2i are likely preferred in subject with lower BMI or less severe hyperglycaemia.

With regard to all-cause mortality, the analysis of the whole study population showed no significant difference between subjects treated with SGLT2i and GLP1-RA. In real-world clinical practice, SGLT2i are preferentially prescribed in subjects with heart failure and established CKD, in accordance with guideline-supported indications, resulting in a higher-risk clinical profile at treatment initiation [18, 19]. Although our results apparently might raise doubts on the clinical efficacy of SGLT2i in ensuring a prolonged survival, the high cardiovascular risk of this subgroup (despite similar control of other risk factors like hypertension and dyslipidaemia) overcomes the efficacy of the treatment. We should, however, consider that all-cause mortality is not higher than that observed in GLP1-RA subjects, who showed a 50% lower prevalence of CVD at baseline.

Age-stratified analyses further revealed a survival advantage with GLP1-RA initiation beyond 80 years of age. However, when baseline cardiovascular burden was incorporated into the model, the association between treatment class and mortality was no longer statistically significant. This finding highlights the dominant role of comorbidity burden in shaping prognosis in very elderly individuals [20, 21] and suggests that the observed mortality difference between treatment groups is largely driven by underlying cardiovascular vulnerability, rather than by pharmacological effects alone. Again, our results reflect real-life prescribing patterns [22] and underscore that, when pharmacologic treatments are started in very elderly subjects, baseline disease burden strongly modifies observed outcomes. Other explanations might be hypothesized: very advanced age may represent an important modifier of the effect of different drug classes on survival outcomes, and it is plausible that, in individuals over 80 years of age, the side effects seldom associated with SGLT2i in relatively young subjects, such as dehydration, hypotension, genitourinary infections and worsening nutritional status [23, 24], may have a greater clinical impact than in younger populations, translating into an increased risk of adverse events and mortality. Conversely, GLP1-RA may be prescribed more selectively in relatively healthy, although very elderly individuals, a population often excluded from pivotal clinical trials, resulting in an apparent prognostic advantage. Subjects receiving GLP1-RA had also higher BMI; the modestly reduced mortality in this group may partly reflect the well-known obesity paradox, whereby individuals with overweight or obesity and established CV disease exhibit better survival outcomes than their normal-weight counterparts [25].

Another relevant observation is the substantial overlap of the longitudinal trajectories of renal function between the two treatments. Despite the strong evidence demonstrating a nephroprotective benefit of SGLT2i in younger populations and in subjects with chronic kidney disease [26, 27], no significant differences in eGFR decline between SGLT2i and GLP1-RA emerged in our sample, even after adjusting for confounders. This data suggests that, in very elderly subjects, the effect of hypoglycemic therapies on renal function may be attenuated or masked by factors related to aging, comorbidities and frailty, which play a predominant role in the progression of renal damage [28, 29]. In addition, competitive mortality and clinical heterogeneity of this population could limit the ability to detect differences in eGFR slope over time.

Searching for other factors putatively able to influence the outcome, we tested the role of BMI and HbA1c at baseline. In univariable analyses, BMI showed an inverse association with mortality; however, this association was not confirmed after adjustment for age and sex, indicating that BMI does not represent an independent predictor in this population. This finding should be interpreted with caution considering the limited ability of BMI to accurately reflect adiposity in very old individuals [30, 31]. In this population, BMI does not distinguish between fat and lean mass, and a normal or mildly elevated BMI does not exclude sarcopenia or altered fat distribution, while a lower BMI may reflect reduced muscle mass, frailty or subclinical malnutrition rather than simply lower adiposity. This limitation, increasingly recognized in geriatric populations, may partly explain the apparent protective association of higher BMI observed in older adults [32, 33].

Also baseline HbA1c was not associated with mortality. This finding is consistent with the current “treat-to-benefit” paradigm for SGLT2i and GLP-1RA, whereby their cardiovascular and renal protective effects appear to be largely independent of baseline glycaemic control. In very elderly individuals this concept is particularly relevant, as treatment decisions are increasingly driven by comorbidity burden and organ protection rather than strict glycaemic targets.

Notably, baseline renal function emerged as the only independent predictor of mortality, outweighing traditional metabolic parameters. This finding highlight kidney function as a key integrative marker of biological aging, comorbidity burden, and overall vulnerability in very elderly individuals with T2D, extending observations from large population-based studies demonstrating a strong association between reduced eGFR and mortality [34, 35].

To explore other determinants, the persistence in therapy of the study participants was also studied, and a greater therapeutic persistence with GLP1-RA compared to SGLT2i emerged, particularly in the subgroup that started therapy after the age of 80. The probability of discontinuation of treatment was, in fact, significantly higher in subjects treated with SGLT2i, but no differences in persistence were observed in subjects who had started therapy between 75 and 80 years of age and were still taking the treatment after 80 years. This finding suggests that greater SGLT2i discontinuation occurs early after prescription, while discontinuation rate is comparable to GLP1-RA after the first months of treatment.

This different persistence, evident only in the very elderly, offers a compelling explanation that might partially contribute to the observed similar survival when drugs are started earlier.

Better persistence could lead to more consistent therapeutic exposure and, consequently, greater long-term benefits [36]. Factors such as ease of administration, side effect profiles, or patient perception of benefit could contribute to these differences in persistence, especially in an older population where polypharmacy and adherence can be challenging, being compromised by frailty and adverse events [37]. It is therefore possible that the lower persistence observed with SGLT2i in older subjects with a high prevalence of HF and CAD, might have partially blunted the protective effect of the drugs.

While our study provides valuable real-world data, the results must be interpreted in the light of important limitations. The retrospective and observational nature of the study exposes to the risk of indication bias and residual confounding and precludes definitive conclusions on causality. The sample size, particularly within the stratified age groups, could limit the generalizability of some findings. Baseline differences between groups, particularly for age, BMI, and lipid profile, suggest that treatment choice was influenced by clinical factors not fully captured in adjustment patterns, such as functional status, degree of frailty, presence of geriatric syndrome, or patient and caregiver preference. In addition, as detailed information on the specific adverse events leading to discontinuation of therapy is not available, it is not possible to accurately identify the mechanisms underlying the reduced persistence observed with SGLT2i.

In conclusion, our study provides real-world evidence that, in elderly T2D subjects, GLP1-RA and SGLT2i are associated with the same all-cause mortality risk, despite a lower persistence in therapy of subjects starting SGLT2i after the age of 80. Renal function trajectories over time were also comparable between the two treatment groups. These findings underscore the importance of an individualized risk–benefit assessment in the choice of hypoglycemic therapy in older and more frail patients and provide useful evidence to guide clinical decision-making in this expanding demographic, indicating the need for prospective studies dedicated to this age group to better define the optimal role of different pharmacological classes in the management of T2D in very old age.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Anna Solini, Diego Moriconi and Domenico Tricò contributed to the conception and design of the study. Alice Del Zoppo and Eleni Rebelos recruited participants and collected data. Alice Del Zoppo, Diego Moriconi and Domenico Tricò analyzed data. Anna Solini and Diego Moriconi wrote the paper. Eleni Rebelos and Domenico Tricò contributed to the editing and revision of the manuscript. All authors read, provided comments on previous versions, and approved the final manuscript.

Funding

The study was not funded by any specific grant.

Declarations

Competing interests

The Authors have nothing to disclose.