SGLT2i compared to other non-insulin antidiabetics in Primary Care patients with earlier diabetes. Cohort profile and sample characteristics of DIAB-CV-AP study

Abstract

Aim

To describe the methodology and baseline characteristics of the sample of the DIAB-CV-AP study which, as a general objective of the cohort, aims to analyse the possible protective effect of iSGLT2 associated with metformin in patients with T2DM.

Methods

The National Health System's BIFAP Primary Care prescription database was used, which includes information on 20,831,855 patients. A total of 52,739 patients were identified who started a second non-insulin-based antidiabetic (NIAD) combined with metformin. After propensity score matching, 11,746 patients were obtained, 5873 in each group.

Results

The most used molecules as SGLT2i were dapagliflozin (51.3%) and empagliflozin (34.9%). The prevalence of CV risk factors was dyslipidemia (63.8%) and hypertension (62.9%), and CV disease was present in 7.7% of the patients. CKD prevalence was 1.8%. Patients in the SGLT2i group was younger (53.9 [10.3] vs 60.6 [13.1], p < 0.001) with had higher body mass index (32.6 [5.6] vs 31.6 [5.9]; p < 0.001), HbA1c (p < 0.001), better kidney function (p < 0.001) and lower albuminuria (p < 0.001). No differences were observed in the cholesterol levels (p = 0.519).

Conclusions

The findings of the DIAB-CV-AP study will expand our knowledge of SGLT2 inhibitors in the early stages of type 2 diabetes mellitus. The recruited cohort, with a lower baseline cardiovascular risk profile than other clinical trials and real-life studies, provides a solid basis for future analyses that will assess whether the protective effect at the cardiovascular and renal levels is reproduced in this clinical context.

Resumen

Objetivo

Describir la metodología y características basales de la muestra del estudio DIAB-CV-AP que, como objetivo general de la cohorte, plantea analizar el posible efecto protector de iSGLT2 asociado a metformina en pacientes con diabetes mellitus tipo 2 (DM2).

Métodos

Se utilizó la base de datos BIFAP de prescripciones en Atención Primaria del Sistema Nacional de Salud, que incluye información de 20.831.855 pacientes. Se identificaron 52.739 pacientes que iniciaron un segundo antidiabético no insulínico (ADNI) asociado a metformina. Tras el emparejamiento por puntuación de propensión, se obtuvieron 11.746 pacientes, 5.873 en cada grupo.

Resultados

Las moléculas más utilizadas como iSGLT2 fueron dapagliflozina (51,3%) y empagliflozina (34,9%). La prevalencia de factores de riesgo cardiovascular fue la dislipidemia (63,8%) y la hipertensión (62,9%), y la enfermedad cardiovascular se presentó en el 7,7% de los pacientes. La prevalencia de enfermedad renal crónica (ERC) fue del 1,8%. Los pacientes del grupo de iSGLT2 eran más jóvenes (53,9 [10,3] frente a 60,6 [13,1]; p < 0,001), con mayor índice de masa corporal (32,6 [5,6] frente a 31,6 [5,9]; p < 0,001), HbA1c (p < 0,001), mejor función renal (p < 0,001) y menor albuminuria (p < 0,001). No se observaron diferencias en los niveles de colesterol (p = 0,519).

Conclusiones

Los hallazgos del estudio DIAB-CV-AP ampliarán el conocimiento sobre los iSGLT2 en las primeras etapas de la DM2. La cohorte reclutada, con un perfil de riesgo cardiovascular basal inferior al de otros ensayos clínicos y estudios de vida real, proporciona una base sólida para futuras análisis que evaluarán si el efecto protector a nivel cardiovascular y renal se reproduce en este contexto clínico.

Introduction

Cardiovascular disease (CVD) is one of the main causes of morbidity and mortality worldwide,^1,2 and it is the second leading cause of death in our country.³ Its prevalence and incidence increase when other risk factors coexist, such as diabetes mellitus (T2DM) and chronic kidney disease (CKD), which represent groups at special cardiovascular risk (CVR).⁴

In other hand, a close relationship between CVD and CKD has been described, with reciprocal interaction between both pathologies.⁵ The mechanisms described that justify this association are alterations in sodium homeostasis, increased blood pressure, and the promotion of a chronic inflammatory state. This linkage between pathologies has been termed cardio-renal-metabolic syndrome (CRM) since the presence of one negatively influences the other organ, worsening its prognosis, and the presence of either of the two pathologies has a common metabolic basis related to obesity, sedentary lifestyle, and T2DM.⁵

Sodium-glucose cotransporter type 2 inhibitors (SGLT2i) have emerged as a revolutionary therapeutic class for reducing both CVR in patients with T2DM and in the prevention and treatment of CKD and heart failure. The first clinical trials that showed favourable results in reducing CVR were EMPA-REG OUTCOME⁶ and DECLARE-TIMI58,⁷ with empagliflozin and dapagliflozin respectively, in which a reduction in cardiovascular (CV) events was described with empagliflozin, as well as a reduction in complications and mortality associated with renal impairment and HF with both drugs.

These results were extended to patients with heart failure and renal disease with a protective effect of SGLT2i in mortality, CV events^8–10 and kidney progression.^11,12 These results have been the basis for SGLT2i to be included in clinical practice guidelines as a first-line therapeutic option for reducing CVR in patients with T2DM,¹³ but also for the treatment of HF¹⁴ and CKD,¹⁵ regardless of its presence.

Two real-world studies, as CVD-REAL, which included more than 300,000 patients treated with SGLT2 inhibitors, mostly dapagliflozin,¹⁶ and EMPRISE, which included 83,946 patients treated with empagliflozin,¹⁷ confirmed the prevention CV and CKD of SGLT2i in patients with T2DM. These studies have the limitation that none of them provide results in our country, nor do they reflect the treatment of patients in Primary Care (PC), because all of them included patients with two, three and more diabetic drugs. It is necessary to analyse these results in this population group to assess whether these benefits are replicated.

Given all the above, DIAB-CV-AP study was designed to analyse the protective effect in patients with T2DM when the SGLT2i is associated to metformin as second line of treatment, when this is not enough in the glycaemic control in a sample of PC. In this manuscript, the primary objective is to describe the methodology, cohort construction process, and baseline sample characteristics. Subsequent publications will report on comparative effectiveness and safety results.

Materials and methods

Patient cohort and study design

A retrospective cohort study was designed with patients recorded in BIFAP (Spain's Pharmacoepidemiologic Research Database for Public Health Systems) database.

The cohort was constructed with patients with a prior diagnosis of T2DM, who were already receiving treatment with metformin and were prescribed a second non-insulin antidiabetic drug (NIAD) between January 15, 2014, and December 5, 2019.

The patients were classified into two cohorts according the second NIAD: those starting SGLT2i (empagliflozin, dapagliflozin, or canagliflozin), analysis cohort, or any other NIAD, control cohort.

Patients had to be 18 years or older at the start of follow-up, with those previously treated with other second-line drugs, or suffering from other forms of diabetes (T1DM, gestational diabetes, or polycystic ovary syndrome), or diseases that significantly limited the use of non-insulin antidiabetic drugs (end-stage renal disease, palliative care, or having been or being on the waiting list for any organ transplant) being excluded.

A patient was considered included in the cohort when they had consumed the drug under study for at least 12 months, Fig. 1. Patient follow-up began on the index date (the date of the first prescription of the second NIAD) and continued until the first occurrence of one of the following censoring events: discontinuation of the index drug class (defined as a gap of >90 days in prescription records), switch to a drug in the opposite cohort class, death, or the administrative end of the study period (December 5, 2019). Patients who discontinued or switched treatment were censored at the date of the last recorded prescription. This as-treated approach was chosen to estimate the treatment effect while patients were actively receiving treatment.

Figure 1.

Design of the DIAB-CV-AP study.

Data sources and ethical criteria

BIFAP is a database of medical records of the National Health System, in PC level, financed and managed by the Spanish Agency of Medicines and Medical Devices (Agencia Española de Medicamentos y Productos Sanitarios, AEMPS). It includes information recorded by PC physicians in their routine clinical practice. Information about pharmaceutical dispensations, diagnoses, hospitalizations, and mortality is captured retrospectively.

The 2019 study database includes information on 20,831,855 patients from PC citizens, that covers near than 50% of the population.¹⁸

Deidentified health records were obtained from this database, crossing them with the causes of mortality from statistic national institute (Instituto Nacional de Estadística, INE), to ensure the mortality causes.

The study was conducted in accordance with the protocol approved by the BIFAP scientific committee on December 3, 2020 (protocol 14/2020), and by the Galicia drug research ethics committee (CEIm-G) on February 25, 2021 (registration code: 2021/057). Given the nature of the patient data, coded and without identifying information from an administrative database, informed consent was not required.

Registered variables

A database was created with the epidemiological data of the patients (age and sex), as well as the index date, treatment start date, and final follow-up date. Additionally, physical examination variables (weight, height, blood pressure) and analytical variables (renal function, blood glucose, HbA1c, and LDL cholesterol) were recorded prior to follow-up, during the follow-up, and prior to the end of the observation period. The medications used simultaneously were also recorded, including antithrombotics, antihypertensives, and hypolipidemics.

Although detailed coding of outcome variables is provided in the supplementary material, they are briefly defined here for clarity. The composite cardiovascular event included nonfatal myocardial infarction, nonfatal stroke, nonfatal peripheral arterial disease, and nonfatal heart failure and cardiovascular death. The composite renal event was defined as presence of estimated glomerular filtration rate (eGFR) ≤60 ml/min (in patients with previously normal values), albumin/creatinine ratio (ACR) ≥30 mg/g (in patients with previously normal values), a sustained reduction in eGFR ≥40%, increase in ACR ≥30%, end-stage renal disease (eGFR <15 ml/min), or renal death.

Total mortality was recorded based on deaths from any cause throughout the follow-up period. In the supplementary material the codes used to define each of the variables are specified.

Statistical analysis

The baseline characteristics of the sample were analysed using descriptive parameters such as mean or standard deviation (SD) and median or interquartile range (IQR) for quantitative variables, and proportion calculations for qualitative variables. For the bivariate analysis, t-Student and χ² were used, or their non-parametric counterparts, respectively.

After analysing the initial database, treatment with the different NIAD could be determined by certain sociodemographic and clinical characteristics of the patients that defined a prescription pattern, Tables S1 and S2. To avoid confounding by indication and facilitate the comparability of both treatment groups, a propensity score (PS) matching analysis was used. A multivariate logistic regression model was used to estimate PS for receiving SGLT2i or any other NIAD. Covariates were selected based on their known association with both the choice of antidiabetic medication and the cardiovascular and renal outcomes of interest, according to established guidelines for PSM analysis, including demographic factors, comorbidities, and concomitant medications (age, sex; age, sex, personal history of hypertension, dyslipidemia, obesity, smoking, heart failure, stroke, myocardial infarction, kidney disease, peripheral vascular disease; and the dose of each of the comparator molecules). We performed 1:1 nearest neighbour matching with replacement, using a calliper of 0.005 of the standard deviation of the estimated PS. While the literature often recommends wider callipers (usually, 0.2 of the SD of the logit of the PS) to balance bias and variance, the large size of the initial BIFAP cohort allowed us to employ a more stringent matching algorithm. This approach was chosen to prioritize minimizing bias, ensuring a very close PS match, at the cost of an acceptable reduction in the final sample size, which remained adequately powered for future outcome analyses. To assess the accuracy of the matching, we compared the covariates before and after PS matching using standardized differences, absolute standardized differences ≤0.1 being considered to denote negligible imbalances between the two groups. After the 1:1 matching, two comparable branches with the same number of patients were obtained, Fig. 1.

To analyse the possible protective effect of iSGLT2 associated with metformin compared to other NIAD, a Cox regression model will be implemented for each event. The state variable will be the incidence of the event. The time variable will be the time to event. And the independent variable will be the group, taking the control group as a reference. Given that both groups are balanced with respect to the variables that could condition the treatments, introducing other covariates into the model is not considered in principle. Thus, for each event, the hazard ratio will be calculated with a 95% confidence interval. Additionally, the survival of both groups will be graphically represented for each event using Kaplan–Meier plots, comparing the groups using the Log Rank (Mantel–Cox) test.

Sensitivity analysis by subgroups will be performed for sex, age ≥70, body mass index (BMI) ≥30, and history of CVD yes/no. For each of these subgroups, the same Cox model described above will be applied to each of its categories separately. This will yield a hazard ratio for each category of the subgroup. These HRs will be compared using their ratio (ratio of hazard ratios). A single Cox model will be used to calculate the statistical significance of this comparison. The state variable will be the incidence of the event. The time variable will be the time to event. And the independent variable will be the group, taking the control group as a reference. In addition, the subgroup and a group × subgroup interaction term will be included as covariates. The statistical significance of the comparison of the hazard ratios will be the significance of the interaction term. For example, the model for BMI ≥30 would be: (1) The state variable will be the incidence of the event; (2) The time variable will be the time to event; (3) Independent variable will be the group (iSGLT2 vs NIAD); (4) Covariates: BMI and interaction group × BMI.

No intention-to-treat analyses were conducted because there were no losses to follow-up, as in patients who changed cohorts, the follow-up was concluded upon making the change. To verify the stability of the results, the PS was adjusted for multiple variables that could influence the prescription profile, as previously mentioned, and additionally, for patients who underwent treatment changes resulting in a cohort modification, the observation was limited to the first treatment, being assigned to this group.

For the processing and analysis of the data, the statistical package SPSS 22.0 for Windows was used.

Results

Characteristics of the samples

A total of 52,739 patients were identified (46,864 in the other NIAD group and 5875 in the SGLT2i group) from the BIFAP database with initiations of antidiabetic treatment added to metformin, Fig. 2.

Figure 2.

Flowchart of overall study population.

The initiation of SGLT2i prescriptions was recorded starting from January 15, 2014, so 21,281 patients with initiation prior to this date were excluded: obtaining a final sample of 31,458 patients, 5875 patients in the SGLT2i group and 25,583 patients in the other NIAD group.

Prior to propensity matching, patients with SGLT2i were younger (p < 0.001) and had a higher proportion of males (p < 0.001). No differences were observed regarding lifestyle habits (smoking or alcohol consumption). The prevalences of diseases such as hypertension, stroke, or heart failure were more frequent in patients with NIAD (p < 0.001 for each) and dyslipidemia and ischemic heart disease were more frequent in the SGLT2i group (p < 0.001 for each), Table 1. The use of statins alone and in combination with ezetimibe was higher in patients who initiated SGLT2i, while the use of other drugs was lower compared to patients who initiated other NIAD, Table S1.

Table 1.

Baseline characteristics of patients prior to propensity matching.

	SGLT2i group		NIAD group		p
	N	%	N	%
	5,875		25,583
Women	2,285	38.9%	12,410	48.5%	<0.001
Age (mean [SD])	53.9 [10.3]		60.6 [13.1]
<50 years	1,972	33.6%	5,312	20.8%	<0.001
50–59 years	2,213	37.7%	6,954	27.2%	<0.001
60–69 years	1,291	22.0%	6,528	25.5%	<0.001
≥70 years	398	6.8%	6,788	26.5%	<0.001
AHT	3,489	59.4%	16,594	64.9%	<0.001
Time of evolution of diabetes (mean [SD])	4.4 [3.4]		3.8 [3.4]		<0.001
Dyslipidemia	3,376	57.5%	13,875	54.2%	<0.001
Tobacco	3,064	52.2%	13,246	51.8%	0.602
Alcohol	4,394	74.8%	19,352	75.6%	0.171
CVD	453	7.7%	2,088	8.2%	0.253
CVD con AF	637	10.8%	3,486	13.6%	<0.001
Ischaemic heart disease	230	3.9%	694	2.7%	<0.001
PAD	99	1.7%	394	1.5%	0.420
Stroke	89	1.5%	516	2.0%	0.012
AF	249	4.2%	1,956	7.6%	<0.001
HF	79	1.3%	727	2.8%	<0.001
CKD	108	1.8%	2,376	9.3%	<0.001

N: total number of patients; %: percentage of the total group; SGLT2i: sodium-glucose cotransporter type 2 inhibitors; NIAD: non-insulin antidiabetics; HTA: arterial hypertension; CVD: cardiovascular disease; AF: atrial fibrillation; PAD: peripheral arterial disease; HF: heart failure; CKD: chronic kidney disease.

After the PS, a total of 11,746 patients were identified, of which 5873 were assigned to each group, Table S2 and Figs. S1 and S2. The baseline characteristics were well balanced between both groups after matching, Table 2 and Fig. S3, with standardized differences in all variables ≤4%. 38.9% were women, with an average age of 55.5 years, with the 50–59 age group being the largest (37.7%).

Table 2.

Baseline characteristics of the sample (post-match).

	SGLT2i group		NIAD group		Standardized difference
	N	%	N	%
	5,873		5,873
Women	2.285	38.9%	2.328	39.6%	−0.015
Age	53.9 [10.3]		54.5 [11.2]		0.001
<50 years	1.971	33.6%	1.988	33.8%	−0.0061
50–59 years	2.213	37.7%	2.041	34.8%	0.0610
60–69 years	1.291	22.0%	1.262	21.5%	0.0120
≥70 years	398	6.8%	582	9.9%	−0.1135
HTA	3.692	62.9%	3.711	63.2%	−0.021
Time of evolution of diabetes (mean [SD])	4.4 [3.4]		4.9 [3.8]		0.002
Dyslipidemia	3.748	63.8%	3.656	62.3%	−0.018
Tobacco	3.064	52.2%	3.181	54.2%	−0.040
Alcohol	4.392	74.8%	4.475	76.2%	−0.033
CVD	452	7.7%	440	7.5%	0.008
CVD con AF	636	10.8%	643	10.9%	−0.004
Ischaemic heart disease	230	3.9%	220	3.7%	0.009
PAD	99	1.7%	97	1.7%	0.003
Stroke	88	1.5%	92	1.6%	−0.006
AF	248	4.2%	292	5.0%	−0.036
HF	79	1.3%	90	1.5%	−0.016
CKD	108	1.8%	98	1.7%	0.013

N: total number of patients; %: percentage of the total group; SGLT2i: sodium-glucose cotransporter type 2 inhibitors; NIAD: non-insulin antidiabetics; HTA: arterial hypertension; CVD: cardiovascular disease; AF: atrial fibrillation; PAD: peripheral arterial disease; HF: heart failure; CKD: chronic kidney disease.

The prevalence of CV risk factors was present in almost two-thirds of the patients: dyslipidemia (63.8%) and hypertension (62.9%). CVD had been previously diagnosed in 7.7% of the patients, with ischemic heart disease being the most frequent (3.9%). 4.2% of the patients suffered from atrial fibrillation and 1.8% from CKD, Table 2 and Fig. S3. Patients in the SGLT2i group had higher body mass index (p < 0.001), HbA1c (p < 0.001), better kidney function (p < 0.001) and lower albuminuria (p < 0.001). No differences were observed in the cholesterol levels (p = 0.519), Table S3.

The most used molecules as SGLT2i were dapagliflozin (51.3%) and empagliflozin (34.9%); canagliflozin represented 13.8%, and only 4 patients received ertugliflozin during the observation period. As NIAD, the most used were iDPP4 (42.3%) followed by arGLP1 (25.9%) and sulfonylureas (15.3%). Overall, 57.7% of patients received statins, 64.9% antihypertensive medications and 56.3% ACEis/ARBs.

The average age in patients with SGLT2i was lower (53.9 [10.3] vs 60.6 [13.1], p < 0.001), but the BMI was slightly higher with statistically significant differences (32.6 [5.6] vs 31.6 [5.9]; <0.001).

Discussion

The DIAB-CV-AP study is based on the analysis of PC prescription records from BIFAP between 2014 and 2019, of low-risk diabetic patients with at least 12 months of treatment with SGLT2i in second line after metformin.

The studies that have provided results in CV and renal prevention to date, both clinical trials and observational studies, recruit patients in advanced stages of T2DM with a higher prevalence of CVD and cardiovascular risk factors. The DIAB-CV-AP study includes patients being followed in PC and in whom treatment with SGLT2 inhibitors is initiated at this level of care, which is where T2DM control and prevention of CV and renal complications should be achieved. In our view, after reviewing the extensively published evidence, the DIAB-CV-AP study will provide results of a therapeutic strategy – the addition of SGLT2 inhibitors to patients in whom metformin monotherapy does not achieve targets – which are increasingly used in PC and will demonstrate its efficacy in preventing CV and renal events in patients with T2DM.

Comparing our sample with real-life studies on SGLT2i (CVD-REAL) and empagliflozin (EMPRISE), the percentage of men and the average age were similar, around the 50s decade. Compared with benchmark real-life studies such as CVD-REAL, our cohort exhibited a 41% lower prevalence of established cardiovascular disease (7.7% vs. 13%) and a 28% lower prevalence of chronic kidney disease (1.8% vs. 2.5%).¹⁹ These differences underline the substantially lower risk profile of our population, representative of the earliest phases of T2DM management in PC. For its part, the EMPRISE study included patients with a higher prevalence of hypertension (74%), hypercholesterolemia (79%), but especially ischemic heart disease (15.7%), heart failure (4.2%), stroke (4.7%) or PAD (4.7%), and CKD (6.6%)²⁰; all much higher than in our sample with lower than 4% for ischaemic heart disease or 2% for the other CVD. This patient profile indicates that our sample has a lower CVR than those used in both studies. Nevertheless, these differences in CV risk are even more significant in clinical trials with SGLT2i,^21,22 making the results of our work of interest, as they expand on those already known in a lower-risk patient profile, corresponding to earlier stages of T2DM, more similar to those seen daily in PC.

Another illuminating aspect of the disease phase is the disease duration. While our study was conducted with a sample of subjects with a mean disease duration of less than 5 years, CVD-REAL presented mean durations of around 9 years and 57% of patients had durations greater than 10 years.¹⁹ In EMPRISE, more than 40% of patients were receiving two or more drugs at the time of starting empagliflozin, and a quarter of the sample were receiving insulin treatment,²⁰ reflecting a longer time since the onset of DM.

The AGORA study also enrolled patients with T2DM receiving dapagliflozin in PC; however, key methodological differences exist that highlight the unique contribution of our work. The AGORA-AP study included patients aged 18–75 years, a similar age range to ours, but excluded those with an estimated glomerular filtration rate (eGFR) <60 ml/min/1.73 m², and the was more than 10 years in the evolution.²³ In contrast, our DIAB-CV-AP cohort does include patients with established CKD (prevalence of 1.8%), allowing for a more representative analysis of the general population seen in PC, where renal comorbidity is common. Furthermore, the ambispective design of AGORA-AP differs from our retrospective, registry-based approach to a population-based database such as BIFAP. Therefore, DIAB-CV-AP complements the AGORA-AP findings by providing evidence from a broader population with a different renal risk profile.

The basal characteristics of our sample showed a lower prevalence of CVD and CKD than other PC samples as IBERICAN, that showed 13.9% of prevalence in CVD and 8.4% in CKD. In the other hand, hypertension and hypercholesterolemia showed higher prevalence in our sample compared with IBERICAN, with 48.0% and 50.3% respectively.²⁴ These characteristics demonstrate that the BIFAP sample used for this study has a similar characteristic as other samples in PC and at least showed a lower CVR indicating maybe a group of patients in the beginning of the T2DM because their lower prevalence of CVD and cardiovascular risk factors related to the stage of treatment, second step after metformin.

The DIAB-CV-AP study offer other advantages as an equitable distribution of other drugs, with preventive effect in the same disease, in both compared groups of treatment, as statins or ACEis/ARBs, after PS. Some of these drugs demonstrate an interaction positive effect in prevention as Zhao et al. meta-analysis that the combination therapy of ACEI/ARB and SGLT2 inhibitor in lowering MACE, CV death or HHF, and CKO compared to the ACEI/ARB monotherapy among T2DM patients.²⁵

Strengths and limitations

Our study presents several strengths that reinforce the validity of its findings. First of all, its design based on real-world data from the BIFAP database, which provides prescriptions in PC for more than half of the Spanish population. Of the real-world studies mentioned, only EMPRISE includes Spanish patients, providing 5865 patients in each comparison group after PS.²⁰ Our sample, although slightly smaller, includes patients from more than half of the Spanish population, after at least 12 months of treatment. On the other hand, we have been able to analyse the target variables in a more heterogeneous sample, representative of the usual clinical practice of PC in our country and, in light of the published data, at a lower risk than those recruited in clinical trials and observational studies, which increases the external validity of our results. Moreover, the use of PS to minimize biases in the comparison between therapeutic groups, along with the initiation of the observation period with the first prescription of each drug, which had to be carried out in the same period (between the years 2014 and 2019), adds methodological robustness, reducing the possibility of confusion due to baseline differences between patients and survival and observation time biases.

However, the study also has some limitations inherent to its observational design. Despite the statistical adjustment with PS, the presence of residual confounding factors cannot be completely ruled out, given that treatment assignment is not random and may be influenced by individual medical decisions that have not been controlled. Moreover, although clinical events were recorded with standardized criteria within the BIFAP database, the possibility of classification bias or underreporting of adverse events, especially in the identification of mild or transient episodes, remains an aspect to consider.

Despite statistical adjustment with PS, the presence of residual confounding cannot be completely ruled out, given that treatment assignment is not random. Unmeasured variables, such as the exact duration of T2DM, socioeconomic status, or previous treatment adherence, could influence the results. Similarly, channelling bias, whereby physicians may prescribe SGLT2i to patients with specific unobserved characteristics (e.g., higher motivation or weight concerns), may not be fully controlled by the covariates included in the propensity model. Although in view of Table 2 there does not appear to be any predictor variable that shows significant differences between the groups after matching, a subsequent adjustment of the comparisons is not considered necessary.

Another limitation is the lack of data on treatment adherence, which could influence the observed results, given that the effectiveness of drugs largely depends on their continued use. However, the analysis of patients who have collected treatment from pharmacies for at least 12 months allows us to approximate a situation of correct therapeutic adherence.

Conclusions

The findings of DIAB-CV-AP study will extend the knowledge in the SGLT2i in PC beginning the treatment in patients con T2DM in the first stages of the disease, after metformin. The recruited sample with lower risk as other clinical trials and real-life studies will show if the protective effect at CV and renal levels is the same as other studies described.

CRediT authorship contribution statement

JPO: study design, investigation, data collection, data analysis, review of analyzed data and results, manuscript writing and review. DRA: investigation, review of analyzed data and results, manuscript writing and review. ASF: study design, investigation, data collection, data analysis, review of analyzed data and results, manuscript writing and review. PMR: investigation, review of analyzed data and results, manuscript writing and review. MPR: investigation, review of analyzed data and results, manuscript writing and review. JRGJ: study design, investigation, data collection, data analysis, review of analyzed data and results, manuscript writing and review. SCS: study design, investigation, data collection, data analysis, review of analyzed data and results, manuscript writing and review.

Declaration of generative AI and AI-assisted technologies in the writing process

The authors declare that AI was not used in any stage in the development of this manuscript.

Funding

None declared.

Declaration of competing interests

The authors declare no conflicts of interest in relation to this article.

Data availability

Data may be provided upon justified request to the corresponding author.