Predictors of efficacy of Sodium‐GLucose Transporter‐2 inhibitors and Glucagon‐Like Peptide 1 receptor agonists: A retrospective cohort study

The aim of the present study was to verify predictors of HbA1c reduction with Sodium‐GLucose Transporter‐2 (SGLT2) inhibitors and Glucagon‐Like Peptide 1 (GLP1) receptor agonists in routine clinical practice.


| INTRODUCTION
Sodium-GLucose Transporter-2 (SGLT2) inhibitors and Glucagon-Like Peptide 1 (GLP1) receptor agonists are widely used as second line or first-line 1,2 drugs in the treatment of type 2 diabetes.2][3] In clinical practice, the preference for one class over the other in individual patients is mainly based on the profile of action of different drugs, including the effects on parameters different from hyperglycemia (e.g., nephroprotection, cardiovascular risk reduction, weight loss, reduction of blood pressure, etc.), matched with patients' characteristics.For example, the American Diabetes Association/European Association for the Study of Diabetes Consensus documents recommend SGLT2 inhibitors in patients with predominant renal disease, and GLP1-receptor agonists in those with predominant cardiovascular disease. 1 The effect of each drug on glucose control differs in individual patients; for instance, GLP1 receptor agonists, which are on average very effective on HbA1c, have been reported to fail to determine relevant improvement of this parameter approximately in 30% of patients. 4,5veral attempts have been made to discriminate the heterogeneity of patients with type 2 diabetes, which are possibly useful for the prediction of treatment response to different drugs.On the basis of some clinical parameters, a cluster approach allowed the identification of subgroups of patients at higher risk of needing insulin treatment 6 ; to our knowledge, however, such clusters have never been used to predict the response to specific non-insulin treatments for type 2 diabetes.The effect of different drugs on blood glucose is moderated by some genetic factors: for example, specific alleles of some genes are associated with a greater response to metformin, sulfonylureas, pioglitazone, Dipeptidyl peptidase 4 inhibitors/GLP1-receptor agonists. 7However, only a small fraction of the variability of response to treatment is predicted by genetic markers known to date 8 ; as a consequence, genetic mapping cannot be considered, at present, a feasible strategy for the clinical characterisation of patients aimed at selecting the most appropriate drugs.The most straightforward approach for the identification of predictors of HbA1c response to a drug is the analysis of subgroups of patients enroled in clinical trials.Such analyses allowed, for example, the identification of impaired renal function as a predictor of poor response to SGLT2 inhibitors. 9On the other hand, subgroup analyses are often missing in reports of phase III trials.The attempt at meta-analysing subgroups of patients from different clinical trials to gain some insight into predictors of reduction of HbA1c and body weight with GLP1 receptor agonists provided only partial results. 10e enrolment in randomized trials of selected patients, with many exclusion criteria and an accurate follow-up procedure, could reduce the variability of response, 11 thus preventing the observation of relevant predictors.For this reason, observational studies may add relevant information.
The aim of the present study was to verify predictors of HbA1c reduction with SGLT2 inhibitors and GLP1 receptor agonists in routine clinical practice.

| Inclusion criteria
A retrospective cohort study was performed, enrolling patients with type 2 diabetes aged ≥18 years, referring to the Diabetes Outpatient Clinic of Careggi Hospital, Florence, between October 1 st and 31 October 2020, who had received a prescription of an SGLT2 inhibitor or a long-acting GLP1 receptor agonist any time before 1 October 2018, with at least 6 months of persistence in therapy after the first prescription, and who provided their informed consent, with no other exclusion criteria.Persistence was defined as the maintenance of the same therapy according to current therapy recorded in clinical records based on follow-up visit interview.

| Retrieved data and outcomes
Demographic and clinical characteristics of patients at the date of prescription (age, gender, duration of diabetes, pharmacological treatments, comorbidities, diabetic complications, body mass index (BMI), blood pressure, HbA1c, lipid profile, creatinine, microalbuminuria) were retrieved from patients' clinical records; estimated Glomerular Filtration Rate (eGFR) was calculated using the CKD-EPI algorithm. 12HbA1c levels after 6 months from the initiation of therapy were recorded for each patient.Treatment was considered as a success, with respect to glycaemic control, in case of either HbA1c reduction of at least 10 mmol/mmol, or when achieving HbA1c goals recommended by Italian Society of Diabetology/Medical Diabetologists Association (SID/AMD) standards of care. 13Those goals were: 1. in patients not treated with antidiabetics capable of causing hypoglycemia (insulin, sulphonylureas/glinides), HbA1c ≤ 48 mmol/ mol if aged ≤75 years, HbA1c ≤ 53 mmol/mol if aged >75 years or with multiple comorbidities; 2. in patients using drugs potentially causing hypoglycemia, HbA1c between 48 and 58 mmol/mol if aged ≤75 years, HbA1c between 53 and 64 mmol/mol if aged >75 years or with multiple comorbidities.

| Statistical analysis
Patients with missing data were excluded from each analysis.Success and failures were compared using two-tailed t tests and chi square tests for continuous and categorial variables, respectively.In addition, in order to adjust for confounders, a multivariate logistic regression analysis was performed, with success as the outcome, and age, and patients' characteristics significantly (p < 0.05) associated with success at univariate analysis as covariates.Statistical significance was set at p < 0.05.Statistical analysis was performed using sample of 206 patients would have been sufficient to detect the effect of an evenly distributed categorial parameter with a relative risk of 1.66, with a and ß 0.05 and 0.80, respectively.

| Ethics and funding
The study protocol was approved by the local Ethical Board of Florence (Ref.18526_oss/2020) and was performed as part of the institutional activity of the investigators with no specific funding.The study was reported following STROBE 14 recommendations (Table S1 of Supplementary Material).

| Baseline characteristics
The characteristics of the samples enroled are summarised in Table 1, whereas the proportion with missing data for each variable is reported in Supplementary Table S2.Twelve patients were excluded because they did not provide their informed consent.None of the included patients received a combined treatment with SGLT2 inhibitors and GLP1 agonists, which had not yet been approved by regulatory authorities at the time of enrolment.The mean age, duration of diabetes and baseline HbA1c levels were similar in the two cohorts.The mean BMI was above 30 kg/m 2 in both the SGLT2 inhibitor and GLP1 receptor agonist cohorts.

| DISCUSSION
Many different drugs are currently available for the treatment of type 2 diabetes.The choice of the most suitable drug (or combination of drugs) for each patient is often mainly based on the effects of treatment other than the reduction of HbA1c, which may be particularly beneficial in some individuals, for example, reduction of major cardiovascular events in patients with high cardiovascular risk, nephroprotection in those with incipient nephropathy, weight loss in obese subjects, etc. 1 On the other hand, clinical trials show that the effects of drugs on HbA1c, which is still a major therapeutic target, 13 are heterogeneous.Although the initial glycaemic response to a drug could be a predictor of its longer-term efficacy on glucose control, 15 it would be clinically useful to identify predictors of response even before initiating a treatment.
Retrospective observational studies on cohorts of patients receiving different treatments could be a useful alternative source of information collected in routine clinical practice.In the present study, we identified some potential predictors of therapeutic success with respect to HbA1c for SGLT2 inhibitors and GLP1 agonists.
The higher risk of failure to SGLT2 inhibitors in patients with impaired renal function has already been observed in clinical trials. 16,17In the present study, such an association was not significant at multivariate analysis after adjusting for confounders.However, SGLT2 inhibitors, which are now indicated in the treatment of chronic renal disease regardless of diabetes, were not approved or reimbursed for renally impaired patients at the time of enrolment; as a consequence, the whole sample was composed of individuals with relatively preserved renal function.This study cannot therefore provide reliable information on the effect of renal impairment on the efficacy of SGLT2 inhibitors.Interestingly, the proportion of success (with respect to the attainment of HbA1c targets) was smaller in older individuals.This association of a smaller reduction of HbA1c with SGLT2 inhibitors in older patients has not been previously described.Interestingly, the proportion of failures in the elderly retains statistical significance even after adjusting for estimated glomerular filtration; however, the estimation of glomerular filtration rate on the basis of serum creatinine is problematic in advanced age. 18It should be noted that a greater proportion of failures on HbA1c in elderly patients does not necessarily imply that these drugs have a smaller nephroprotective effect in this population 19 ; however, SCOCCIMARRO ET AL.
[21] Notably, in contrast with previous reports, 22,23 the reduction of HbA1c with SGLT2 inhibitors was not significantly associated with the duration of diabetes, despite a nonsignificant trend towards lower efficacy in patients with a longer duration of disease; the limited sample size might have prevented the detection of a T A B L E 1 Baseline characteristics of patients treated successfully and not with Sodium-GLucose Transporter-2 (SGLT2)-inhibitors, and total sample.

Characteristics
Total significant difference.On the other hand, the reduction of HbA1c during treatment with GLP1 receptor agonists appears to be smaller in those with a long duration of diabetes, even after adjusting for confounders.This is not surprising, since the therapeutic action of these drugs is largely mediated by the stimulation of glucosedependent insulin secretion 24 ; the response to GLP1 and its analogues depends therefore on a sufficient ß cell functional mass, which declines with the duration of diabetes. 25Conversely, the reduction of HbA1c with GLP1 receptor agonists was unaffected by patients' age.
In addition, there is no moderating effect of BMI on HbA1c reduction with GLP1 receptor agonists, which appears to be similar in obese and lean patients.This latter result is in line with reports from randomized clinical trials 26 and cohort studies. 27The effect of initial HbA1c on the incidence of treatment failure depends on the definition of success: if success is defined as the attainment of pre-defined target HbA1c levels, a higher HbA1c level should be expected to be associated with a higher proportion of failures; on the other hand, if success is defined as a pre-determined absolute reduction of HbA1c levels, such a target could be easier to reach in patients with higher baseline HbA1c.In the present study, with a combined definition of success, including both the attainment of treatment goals and/or a reduction of HbA1c of at least 10 mmol/mol, no significant association of treatment failure with baseline HbA1c levels was detected.
The present study is intended to be a preliminary exploration of for some of the variables considered, in particular blood pressure.In adopted for the two classes.Additionally, some parameters potentially affecting drug response (e.g., glucagon secretion for patients receiving SGLT2 inhibitors 28 ) were not assessed.Finally, differences between reported and actually administered drugs could not be detected by the present study, in which current treatment was based on patients' interviews during follow-up visits; therefore, this study cannot incorporate data on compliance and/or adherence to the prescribed treatment.
In conclusion, differences in predictors of HbA1c changes across different classes of drugs could be useful for the identification of the most suitable drug in individual patients.SGLT2 inhibitors seem to be associated with a greater reduction of HbA1c in younger subjects, and GLP1 receptor agonists in those with a shorter duration of diabetes.Further studies on larger samples are needed to obtain greater details on possible predictors of therapeutic response to those two classes of drugs.
factors associated with HbA1c response to SGLT2 inhibitors and GLP1 receptor agonists.An observational study is inevitably affected by an allocation bias, which cannot be eliminated by statistical adjustments; therefore, the observed associations between changes in HbA1c levels after beginning therapy and clinical parameters could have been determined, at least partly, by undetected or unadjusted confounders.In fact, other characteristics of patients different from HbA1c, and most notably cardiovascular risk and renal impairment, can differentially affect prescription patterns for the two classes, irrespective of glucose control.The relatively small sample size could have prevented the observation of the relevant association.The size of the sample did not allow a specific analysis of the effects of changes in doses of the same drug.Changes in the doses of concurrent therapies could also have interfered with the results.Furthermore, since the data were retrospectively extracted from routine clinical records, the proportion of missing data was relevant F I G U R E 1 Panel A. Multivariate regression analysis for predictors of success with Sodium-GLucose Transporter-2 (SGLT2)-inhibitors.Panel B. Multivariate regression analysis for predictors of success with Glucagon-Like Peptide 1 (GLP1) receptor agonists.Success is intended as HbA1c reduction of at least 10 mmol/mol or achieving HbA1c goals.OR, Odds Ratio; eGFR, estimated Glomerular Filtration Rate.
Baseline characteristics of patients treated successfully and not with GLP1-RA, and total sample.Success is intended as HbA1c reduction of at least 10 mmol/mol or achieving HbA1c goals.Data are expressed as mean � standard deviation or number (%).
addition, the sample studied, which was enroled in a Diabetes Outpatient Clinic of a University Hospital, cannot be considered representative of the whole population of patients with diabetes receiving a treatment with either SGLT2 inhibitors or GLP1 receptor agonists.Furthermore, potential differences in average efficacy on HbA1c between SGLT2 inhibitors and GLP1 receptor agonists could have affected results, since the same definition of success was T A B L E 2 a Transient ischaemic attack.b estimated Glomerular Filtration Rate.