diabetes research and clinical practice 171 (2021) 108616 Contents available at ScienceDirect Diabetes Research and Clinical Practice journal homepage: www.elsevier.com/locat e/dia bre s Analysis of the effectiveness of second oral glucose-lowering therapy in routine clinical practice from the mediterranean area: A retrospective cohort study Bogdan Vlacho a,b,1, Xavier Mundet-Tudurı́ a,c,1, Manel Mata-Cases a,d, Joan Antoni Vallès-Callol a, Jordi Real a,d, Magı́ Farré b,e, Francesc Xavier Cos a,c,f, Kamlesh Khunti g, Dı́dac Mauricio a,d,h,i,*, Josep Franch-Nadal a,d,* a DAP-Cat Group, Unitat de Suport a la Recerca Barcelona, Fundació Institut Universitari per a la recerca a l’Atenció Primària de Salut Jordi Gol i Gurina (IDIAPJGol), Barcelona, Spain b Universitat Autònoma de Barcelona (UAB, Pharmacology Department), Cerdanyola del Vallès, and Fundació Institut d’Investigació en Ciències de la Salut Germans Trias i Pujol (IGTP), Badalona, Spain. c Departament of Medicine, Universitat Autònoma de Barcelona, Bellaterra, Spain d CIBER of Diabetes and Associated Metabolic Diseases (CIBERDEM), Instituto de Salud Carlos III (ISCIII), Madrid, Spain e Clinical Pharmacology, Hospital Universitari Germans Trias i Pujol, Badalona, Universitat Autònoma de Barcelona, Bellaterra, Spain f Innovation office at Institut Català de la Salut, Barcelona, Spain g Diabetes Research Centre, University of Leicester, Leicester, UK h Department of Endocrinology and Nutrition, Hospital Universitari de la Santa Creu i Sant Pau, Universitat Autònoma de Barcelona, Bellaterra, Spain i Departament of Medicine, University of Vic – Central University of Catalonia, Vic, Barcelona, Spain A R T I C L E I N F O A B S T R A C T Article history: Aim: To compare the changes in HbA1c, the effect on body weight or both combined after Received 14 October 2020 the addition of a DPP-4i, SGLT-2i, or sulfonylureas (SU) to metformin in real-world condi- Received in revised form tion. 3 December 2020 Methods: We used a primary care SIDIAP database. The included subjects were matched by Accepted 7 December 2020 propensity score according to baseline age, sex, HbA1c, weight, inclusion date, diabetes Available online 10 December 2020 duration, and kidney function. Results: Mean absolute HbA1c reduction was: 1.28% for DPP4i, 1.29% for SGLT2i and 1.26% for SU. Mean weight reduction was: 1.21 kg for DPP4i, 3.47 kg for SGLT2i and 0.04 kg for Keywords: SU. The proportion of patients who achieved combined target HbA1c (0.5%) and weight DPP-IV inhibitor (3%) reductions after the addition of DPP-4i, SGLT-2i or SU, was: 24.2%, 41.3%, and Glycaemic control 15.2%, respectively. Small differences in systolic blood pressure reduction (1.07, 3.10 and SGLT2 inhibitor 0.96 mmHg, respectively) were observed in favour of SGLT-2i. Concerning the lipids, we Sulphonylureas Type 2 diabetes Weight control * Corresponding authors at: Centre d’Atenció Primària Raval Sud, Av. Drassanes, 17-21, 08001cBarcelona, Spain (J. Franch-Nadal). Hospital de la Santa Creu i Sant Pau, Sant Quintı́, 89, 08041 Barcelona, Spain (D. Mauricio). E-mail addresses: DMauricio@santpau.cat (D. Mauricio), jfranch.bcn.ics@gencat.cat (J. Franch-Nadal). 1 B.Vlacho and X. Mundet contributed equally to this work. https://doi.org/10.1016/j.diabres.2020.108616 0168-8227/Ó 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 2 diabetes research and clinical practice 171 (2021) 108616 observed small differences, with an HDL-cholesterol increase with SGLT-2i. Conclusion: Our real-world study showed that the addition of SGLT-2i to metformin was associated with greater reductions in weight and the combination target of weight- HbA1c compared to SU and DPP4 inhibitors. However, similar hypoglycaemic effectiveness was observed among the three-drug classes. Ó 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY- NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 1. Introduction 2. Material and methods Type 2 diabetes mellitus (T2DM) is a heterogeneous condition 2.1. Study design and setting characterized by progressive deterioration of glycaemic con- trol that with time usually requires the combination of sev- We designed a matched retrospective cohort study with a eral antidiabetic drugs with different mechanisms of action. follow-up period of up to 24-months. Data were obtained for Additionally, T2DM carries a high risk for cardiovascular dis- the period between January 1st, 2010, and December 31st, eases (CVD) due to cardiovascular risk factors including over- 2017. We used the primary care SIDIAP (The Information Sys- weight, hypertension, smoking, family history of premature tem for the development of Primary Care Research) database coronary disease, chronic kidney disease, albuminuria and [11], which contains anonymized data from electronic medi- dyslipidaemia [1]. cal records of the people attending 279 Primary Health Care Most current clinical guidelines recommend metformin as Centres (Spain) of the Institut Català de la Salut (ICS). SIDIAP the drug of choice for the initial pharmacological treatment of database has been extensively used for national and interna- T2DM. However, the recommendations for second-line ther- tional epidemiologic and pharmacoepidemiological studies apy in case of contraindications or intolerance to metformin and has been validated in primary care [12,13]. are less specific, and the selection of a drug class is usually based on patient-specific treatment goals and preferences, 2.2. Inclusion and exclusion criteria presence of comorbidities, drug efficacy, and safety profile [1–3]. In clinical practice, there is often a need for the addition Subjects were included if they were 18 years and older, had of second-line therapies to improve glycaemic control and to been diagnosed with T2DM (ICD-10 codes E11), and had reduce risks from macrovascular and microvascular compli- HbA1c values  7% (53 mmol/mol). The inclusion date for cations. Three classes of oral antidiabetics, namely dipeptidyl each cohort was defined by the day of the addition of peptidase-4 inhibitors (DPP-4i), sodium-glucose co- second-line therapy for the first time to metformin. We transporter 2 inhibitors (SGLT-2i) and sulfonylureas (SU) are excluded subjects with a code for other types of diabetes the most widely prescribed second-line oral therapy options (such as diabetes mellitus type 1, gestational or secondary; as add-on therapy to metformin [1,4]. ICD-10 codes: E8, E9, E10, O24, E13), as well as those with A recently published meta-analysis, including 36,746 missing baseline values of HbA1c or weight. The users in each adults in 68 randomized clinical trials, concluded that all cohort were followed for 24 months or up to premature dis- classes of oral antidiabetic drugs result in similar reductions continuation of the study, defined as death, treatment switch, in HbA1c levels when added to metformin, although SGLT-2i the addition of a third antidiabetic treatment, transfer to non- showed the additional benefit on body weight reduction [5]. ICS centres or end date 31/12/2017. No minimum follow-up However, subjects included in randomized clinical trials are time was established. not representative of the population heterogeneity and the real healthcare conditions (real-world data evidence) due to 2.3. Study variables the strict eligibility criteria [6,7]. Moreover, health policy vari- ations across countries (e.g., national reimbursement strate- Baseline clinical and medication data were extracted for the gies), local professional expertise, physician’s and patient’s inclusion date or, failing this, for the nearest date prior to personal choice, and study settings may be associated with the inclusion (maximum of 12 months apart). We collected different patterns of prescription of second-line treatment information on (1) demographic characteristics, including [8–10]. In view of these limitations, routine data from primary age, gender, and smoking habit; (2) the presence of comorbidi- healthcare databases could provide relevant information on ties, including heart failure, peripheral vascular disease, the effectiveness of intensification with a second antidiabetic ischemic heart disease, hypertension, hyperlipidaemia, drug. acute/chronic pancreatitis, chronic kidney disease, relevant The primary objective of the current study was to compare liver diseases, chronic obstructive pulmonary disease (COPD), the changes in glycated haemoglobin (HbA1c) and the effect diabetic neuropathy, diabetic retinopathy, and relevant men- on body weight following the addition of DPP-4i, SGLT-2i, or tal disorders; (3) variables related to T2DM, including diabetes SU as second-line therapies to metformin in a primary care duration and HbA1c value; and (4) additional clinical variables population with T2DM. such as body weight, BMI, blood pressure (BP), lipid profile, diabetes research and clinical practice 171 (2021) 108616 3 estimated glomerular filtration rate (eGFR; CDK-EPI equation), 2.5.3. Missing data and sensitivity analysis and albumin/creatinine ratio. Incomplete cases during the follow-up period were handled using multiple imputation cases analysis (MICA). For this, 2.4. Study outcomes we draw 10 resamples with the ‘‘mice-package” for the R3.6.0 statistical software [16]. The estimates of the parame- For each cohort (i.e., addition of DPP-4i, SGLT-2i, or SU), we ters for each imputed data set were combined using Rubin’s defined the primary study outcomes as the percentage of rules [15]. We then conducted a sensitivity analysis compar- patients with a reduction of HbA1c of at least 0.5% (5.5 mmol/- ing the results obtained with only subjects who had all vari- mol), a weight reduction of at least 3%, or both (main compos- ables observed (complete cases analysis [CCA]) vs those ite binary outcome). The reductions in the HbA1c value and obtained with the MICA approach. body weight were calculated as change between the baseline and after three months following the addition of second-line 2.5.4. Effect size estimate therapy (with a maximum of 24 months of follow-up); we cal- To assess the magnitude of the effect of each treatment on culated percentage of users reaching target of HbA1c below the outcomes, we calculated the effect size through the stan- 7% (53 mmol/mol). Secondary outcomes included changes dardized mean differences between drug classes. This analy- in systolic and diastolic blood pressure (SBP and DBP) and sis was conducted with ‘‘parameters”[17] and ‘‘effect size”[18] lipid profile. for the R3.6.1 statistical software. 2.5. Statistical analysis 2.6. Ethical review 2.5.1. Propensity score matching The study was approved by the Ethics Committee of the Pri- The basal characteristics of the three groups were assessed mary Health Care University Research Institute (IDIAP) Jordi for homogeneity and, based on the observed differences; Gol, Barcelona (P17/205). they were matched to ensure the balance of covariates. Baseline weight, HbA1c, sex, age, diabetes duration, year 3. Results of inclusion, and kidney function were used as the match- ing variables. Since the SGLT-2i group was the least repre- The flow chart of the study is shown in Supplementary Fig. 1. sented cohort, for each user in this group we selected A total of 75,808 patients with T2DM initiating a second treat- participants from the other two cohorts. This was done in ment in addition to metformin during the study period met two steps, firstly matching against the DPP-4i group, and the study inclusion criteria: 27,878 (36.7%) initiated a DPP-4i, thereafter against the SU group. Matching was done by 2,198 (2.89%) an SGLT-2i and 45,732 (60.3%) a SU. After the the ‘‘Nearest Neighbor algorithm” (caliper = 0.01), using matching procedure for baseline variables, the three treat- the ‘‘MatchIt” library of the R (v3.6.1) statistical package ment groups were well balanced (Supplementary Fig. 2), with [14,15]. 6,310 subjects available overall for comparisons and follow- up: 2,124 for DPP-4i, 2,124 for SGLT-2i and 2,062 for SU. Table 1 2.5.2. Main analysis provides the baseline characteristics of the study population. Mean, median, and standard deviation for continuous vari- Across treatment groups, the mean age ranged between 60.5 ables, frequency and percentage for categorical variables, and 61.2 years, the mean T2DM duration between 7.28 and were used to describe the baseline characteristics of the 7.89 years, and the mean HbA1c before treatment addition cohorts. We calculated the average changes from baseline varied between 8.75% and 8.78%. The average length of as well as the proportion of patients that achieved the follow-up was similar between treatment groups, with a med- reduction of HbA1c  0.5% or weight reduction  3%. The ian of 350, 472, and 381 days for DPP-4i users, SGLT-2i users, composite outcome (reduction in both HbA1c  0.5% and and SU users, respectively (Supplementary Table 1). In combi- weight  3%) was analysed using logistic regression and nation to metformin, the most frequently prescribed DPP-4i the findings summarized as absolute percentage risk differ- was sitagliptin (45.0%), and for SGLT-2i it was dapagliflozin ences and odds ratios (ORs), with confidence intervals (CI). (25.4%), while for SU it was gliclazide (83.2%) (Supplementary As a sensitivity analysis, adjusted estimates were calculated Table 5). with multivariate models (multivariate logistic for the binary outcome and linear regression models for continuous 3.1. Changes from baseline in HbA1c value and body response) including the following baseline variables: weight, weight HbA1c, age, diabetes duration, year of inclusion, eGFR and number of comorbidities. All pairwise comparisons were The composite main outcome (HbA1c reduction of at least conducted between groups (i.e., no reference group), where 0.5%, weight reduction of at least 3%, or both) was achieved the significance individual level was prefixed at a = 0.017 more frequently by SGLT-2i users (41.3%) than by DPP-4i users (familywise significance level = 5%) and individual confi- (24.2%) and SU users (15.2%) (Fig. 1 and Supplementary dence level at 0.98 (familywise confidence level = 95%). The Table 2). When comparing the drug cohorts, we observed that statistical analyses were performed using R3.6.1 (https:// the likelihood of achieving this composite outcome was 2.2- www.r-project.org/). fold higher among SGLT-2i users compared to DPP-4i users 4 diabetes research and clinical practice 171 (2021) 108616 Table 1 – Baseline characteristics of the study population for users initiating second add-on treatment to metformin with DPP-4i, an SGLT-2i, or an SU. Characteristic MET + DPP-4i (n = 2124) MET + SGLT-2i (n = 2124) MET + SU (n = 2062) Age, mean (SD), years 61.2 (12.1) 60.5 (11.1) 60.6 (12.0) Gender, n (%) Female 892 (42.0) 924 (43.5) 868 (42.1) Male 1232 (58.0) 1200 (56.5) 1194 (57.9) Smoking habit, n (%) Non-smoker 1010 (47.9) 957 (45.3) 996 (48.5) Ex-smoker 700 (33.2) 758 (35.9) 616 (30.0) Smoker 398 (18.9) 398 (18.8) 440 (21.4) Clinical variables Diabetes duration, mean (SD), years 7.64 (7.12) 7.89 (6.67) 7.28 (5.89) HbA1c, mean (SD), % 8.78 (1.48) 8.77 (1.49) 8.75 (1.39) HbA1c, mean (SD), (mmol/mol) 72.3 (15.9) 72.5 (16.2) 72.4 (16.2) Weight, mean (SD), kg 89.2 (18.5) 91.4 (17.5) 89.5 (18.2) BMI, mean (SD), kg/m2 32.8 (5.72) 33.9 (5.80) 32.8 (5.87) Blood pressure, mean (SD), mmHg SBP 134 (14.4) 133 (13.8) 134 (13.5) DBP 78.8 (10.2) 78.0 (9.55) 79.4 (9.46) Renal function Albumin / creatinine ratio, mean (SD) 57.1 (217) 72.0 (254) 43.4 (145) eGFR, mean (SD), mL/min/1.73 m2 81.2 (13.4) 81.0 (13.0) 82.2 (12.9) eGFR groups (mL/min/1.73 m2), n (%) <30 5 (0.24) 6 (0.28) 10 (0.48) 30–44 31 (1.46) 36 (1.69) 34 (1.65) 45–59 182 (8.57) 155 (7.30) 140 (6.79) 60 1871 (88.1) 1892 (89.1) 1845 (89.5) Lipid profile, mean (SD), mg/dL Total Cholesterol 191 (45.2) 186 (44.2) 193 (41.4) HDL-cholesterol 46.7 (11.3) 46.2 (12.1) 46.9 (11.5) LDL-cholesterol 106 (33.3) 102 (34.1) 108 (33.0) Triglycerides 207 (152) 216 (180) 207 (167) Number of comorbidities, mean (SD) 1.86 (1.22) 2.05 (1.25) 1.76 (1.16) Comorbidities, n (%) Heart failure 89 (4.19) 104 (4.90) 73 (3.54) Peripheral vascular disease 79 (3.72) 116 (5.46) 77 (3.73) Stroke 125 (5.89) 156 (7.34) 99 (4.80) Ischemic heart disease 205 (9.65) 334 (15.7) 164 (7.95) Hypertension 1426 (67.1) 1527 (71.9) 1363 (66.1) Hyperlipidaemia 1220 (57.4) 1318 (62.1) 1193 (57.9) Diabetic neuropathy 77 (3.63) 155 (7.30) 42 (2.04) Diabetic retinopathy 179 (8.43) 350 (16.5) 116 (5.63) Acute/chronic pancreatitis 29 (1.37) 53 (2.50) 19 (0.92) COPD 160 (7.53) 167 (7.86) 136 (6.60) Renal failure 159 (7.49) 140 (6.59) 124 (6.01) Relevant liver disease (excluding steatosis) 75 (3.53) 86 (4.05) 47 (2.28) Major Mental disorders 431 (20.3) 422 (19.9) 375 (18.2) BMI, body mass index; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate (CDK-EPI formula); HbA1c, glycated hae- moglobin A1c; SBP, systolic blood pressure; SD, standard deviation; HDL, high-density lipoprotein; LDL, low-density lipoprotein; COPD, chronic obstructive pulmonary disease. (OR2.23, 98% CI: 1.7, 2.9) and 3.9-fold higher than in SU users (Fig. 2A; Supplementary Table 2). As such, we did not observe (OR0.25, 98% CI 0.4, 0.8), and it was 1.7 -fold higher in DPP-4i significant differences between drug classes when we com- users than in SU users (OR0.56, 98% CI: 0.2, 0.3) (Fig. 1). pared the likelihood of a reduction of  0.5% in HbA1c from The proportion of patients achieving a reduction in baseline or for users reaching target of HbA1c < 7% (Fig. 1; HbA1c  0.5%, considered as a single outcome, was very sim- Supplementary Table 3). ilar across drug classes (71.6% for DPP-4i and 72.5% for SGLT- Regarding body weight as a secondary outcome, SGLT-2i 2i, and 74.1% for SU; Fig. 1 and Supplementary Table 2). Sim- was the drug class that was associated with the greatest pro- ilar tendency across the cohorts was observed for users portion of patients achieving a  3% reduction (53.6%; Fig. 1; reaching target of HbA1c < 7%. The mean in HbA1c values Supplementary Table 2) and also the drug associated with were similar across the cohorts with a reduction of 1.28%, the greatest decrease from baseline ( 3.47 kg) (Fig. 2B). In con- 1.29%, and 1.26% for DPP-4i, SGLT-2i, and SU, respectively trast, the mean change in weight with DPP-4i was 1.21 kg diabetes research and clinical practice 171 (2021) 108616 5 Fig. 1 – Percentage of patients that achieved the main outcomes for each treatment group and adjusted comparisons (odds ratio [OR] and 98% CI) between drug classes. *Statistically significant p-value after multiple testing correction (<0.017) HbA1c, glycated haemoglobin; IDPP4, inhibitors dipeptidyl peptidase-4; MET, metformin; SGLT-2i, inhibitors sodium/glucose cotransporter 2; SU, sulphonylureas. Fig. 2 – Adjusted mean changes from baseline by treatment group and differences in the reduction (DR and 98% CI) between drug classes in HbA1c values (A) and body weight (B). *Statistically significant p-value after multiple testing correction (<0.017). HbA1c, glycated haemoglobin; IDPP4, inhibitors dipeptidyl peptidase-4; MET, metformin; SGLT-2i, inhibitors sodium/glucose cotransporter 2; SU, sulphonylureas. (29.7% of patients achieving  3% reduction) and only (98% CI: 2.9, 3.9) weight loss compared to those initiated on 0.04 kg with SU (18.6% of patients with  3% weight loss). an SU (Fig. 2 B; Supplementary Table 4). Patients initiated on When comparing the drug cohorts, patients who initiated DPP-4i users had a mean of 1.26 kg (98% CI: 0.8, 1.8) reduction add-on therapy with SGLT-2i were 2.7 times more likely to in weight compared to SU users. achieve weight loss  3% compared to DPP-4i users (OR2.66, The median time to achieve the reduction in HbA1c and 98% CI 2.1, 3.3) and 5.0 times more likely than SU users weight was shorter for DPP-4i users (277 days and 295 days, (OR0.2, 98% CI 0.2,0.3) (Fig. 1). Overall there was a 2.11 kg respectively) compared to SGLT-2i users (343 days and (98% CI: 1.6, 2.6) reduction in weight in those initiated on 348 days, respectively) and SU users (286.5 and 301 days, SGLT-2i compared to those initiated on DPP-4i and a 3.36 kg respectively) (Supplementary Table 1). 6 diabetes research and clinical practice 171 (2021) 108616 3.2. Reduction of blood pressure and lipid parameters are shown in Fig. 3 and summarized in Table 2. The effect of SGLT-2i compared to DPP-4i was moderate regarding the com- Regarding SBP, There was a mean reduction in SBP of posite reduction of HbA1c  0.5% and weight  3%, but it was 3.10 mmHg (±15.0), 1.07 mmHg (±15.5), and 0.96 mmHg large compared to SU. Moreover, SGLT-2i had a large effect on (±15.2) in those initiated on SGLT-2i, DPP-4i and SU respec- weight reduction of  3% compared to both DPP-4i and SU. As tively (Supplementary Table 2). Comparable changes were per blood pressure, there was a small effect favouring SGLT-2i observed for DBP, with reductions of 1.75 mmHg (±9.94), compared to DPP-4i and SU. The only non-trivial differences 0.63 mmHg (±9.86), and 0.53 mmHg (±9.41) for SGLT-2i, DPP- in the lipid profile were a small benefit of SGLT-2i over SU in 4i, and SU, respectively. total cholesterol decrease, and a moderate effect of SGLT-2i Regarding the effects in the lipid profile (Supplementary in HDL increase compared to both DPP-4i and SU. Table 2), SU users had the greatest average reductions in total cholesterol 10.4 mg/dL (±36.2) triglycerides 29.7 mg/dL (±159) 3.4. Sensitivity analysis and LDL cholesterol 5.68 mg/dL (±27.5). In contrast, those ini- tiated on an SGLT-2i had a mean increase for HDL cholesterol The magnitude of the results of sensitivity analysis with the 2.36 mg/dL (±7.42). This effect on HDL was confirmed when multiple missing imputations approach was similar to those we compared the differences between the SGLT-2i and the from the analyses only considering complete cases. SGLT-2i DPP-4i cohort, as SGLT-2i users experienced an average HDL remained as the drug class with the highest percentage of additional increase of 2.33 mg/dL (98% CI: 3.14,-1.52) (Sup- patients achieving the composite outcome (i.e., HbA1c reduc- plementary Table 4). tion of at least 0.5% and weight reduction of at least 3%) (Sup- plementary Table 2) and the drug with the highest probability 3.3. Effect size analysis of such achievements compared to the two other treatments (Supplementary Table 3). In addition, SGLT-2i was the drug The clinical relevance of the observed differences between class that was associated with the greatest mean weight loss treatments was assessed through the effect sizes. The results from baseline, although the difference in reduction was only Fig. 3 – Forest plot of the standardized mean differences (effect sizes) between the different drug classes. HbA1c, glycated haemoglobin; IDPP4, inhibitors dipeptidyl peptidase-4; MET, metformin; SGLT-2i, inhibitors sodium/glucose cotransporter 2; SU, sulphonylureas. diabetes research and clinical practice 171 (2021) 108616 7 Table 2 – Magnitude of the effect sizes between treatments for the different outcomes. HbA1c, glycated haemoglobin; DPP4i, inhibitors dipeptidyl peptidase-4; MET, metformin; SGLT-2i, inhibitors sodium/glucose cotransporter 2; SU, sulphonylureas. Magnitude of effect size by Cohen´s f: Small < 0.10; Medium 0.10–0.25; Large > 0.4. significant when compared to SU users (Supplementary we found a lack of treatment intensification in a relevant pro- Table 4). Finally, the magnitude and direction of the effect portion of subjects type 2 diabetes with HbA1c values over 8% sizes did not change when assessed with the multiple impu- at baseline [26]. Indeed, the NICE guidelines suggest adding tation approach (Fig. 3). the second antidiabetic drug when HbA1c is above 6.5% [3]. The recent data suggest that early combination glucose- 4. Discussion lowering therapy provides greater and durable long term ben- efits [27]. Furthermore, this has recently been also recom- This real-world study based on a large representative cohort mended in primary care position statement [27]. of people with type 2 diabetes from Spain suggests that Similar retrospective studies have been recently published SGLT2i, DPP4i, and SU had similar effectiveness in terms of using data collected from electronic medical records from the HbA1c reduction, but significant differences for weight reduc- UK, US, Denmark, and Germany on patients who initiate add- tions and the composite outcome of HbA1c and weight reduc- on therapy to metformin [23,24,28–31], where the prescription tion in favour of SGLT-2i. patterns for these three antidiabetic combinations are differ- The important reduction in HbA1c, when compared to the ent. The data in our study are quite similar to those observed one usually found in clinical trials (1.29% greater in our in the UK, where the most common add-on treatment to met- study), may be related to the higher baseline HbA1c values formin were sulfonylureas, then DPP-4i and SGLT-2i [23,28,29]. when second-line therapies are initiated in clinical practice Conversely, the addition of SU was rarely chosen in Germany [19]. This is in line with the findings of a meta-analysis by (4.4%) [28]. The use of SU in our study was superior to the rest DeFronzo et al., which reported a positive relationship of drugs probably due to the in-force CatSalut local current between baseline HbA1c values and the magnitude of the guidelines during that period that recommend the use of SU HbA1c change across 10 categories of glucose-lowering thera- as the second line therapy of choice [32]. pies, irrespective of the class or mode of action [20]. In our Comparing the baseline characteristics of our study with study, treatment intensification with second-line therapy others, there were similarities in terms of age, gender, and occurred at a mean HbA1c level of 8.8% (72.3 mmol/mol), baseline HbA1c at the time of intensification with the which is far from the HbA1c > 7% threshold recommended second-line therapy. For instance, baseline HbA1c was higher in current guidelines, thus confirming the delay in intensifica- in the UK study (9.0%) [23] than the US cohort (8.4%) [24], tion observed in previous similar real-world studies [21–24]. while in our study the average baseline HbA1c values The high HbA1c level at the time of treatment intensification between the cohorts were slightly different, i.e. 8.8% for all suggests the need to tackle therapeutic inertia, in particular drug classes investigated. These HbA1c values are too high in early disease, with the addition of a second antidiabetic and far from those recommended by current guidelines as drug [25]. We have recently addressed the issue of clinical the level for timely intensification of treatment in daily clini- inertia in a previous study using the same database in which cal practice [1–4]. 8 diabetes research and clinical practice 171 (2021) 108616 In general, and before the propensity score matching pro- HDL-cholesterol with SGLT-2i, a greater decrease in total and cedure, the SGLT-2i group was younger, more obese, and with LDL-cholesterol with DPP4i and SU, and a greater decrease a longer duration of diabetes compared to the rest of the in triglycerides with SGLT-2i and SU. We only observed mod- groups. Moreover, these patients had the worst comorbidity erate beneficial effect of SGLT-2i in the increase of HDL- profile, notably mainly cardiovascular disease. Similar results cholesterol compared to the other 2 drug classes. were observed in the US study where users intensified with The main strength of our study is a large number of sub- SGLT2i were younger, more likely to be female, and obese, jects included and a population based representative (SIDIAP as compared with those who initiated dual therapy with information comes from ICS, which manages >75% of the DPP-4i, or SU [24]. The higher presence of the CVD comorbid- Catalan population) sample from primary care. However, ity among these users could be explained by the recommen- there are some limitations in our study, typically inherent to dations issued in guidelines that SGLT2i should be used in observational studies using real-world databases such as patients with established CVD or at high cardiovascular risk the non-availability of some data. Despite this potential [1]. Moreover, previous clinical trials have demonstrated source of confounding bias, the sensitivity analysis showed cardio-renal benefits of these drugs in patients with T2DM that the results were comparable to the ones obtained using and established cardiovascular disease [33–35] or high cardio- only complete cases. Another limitation is the possibility of vascular risk [34] that seems independent of the reduction on indication bias or confounding by severity of disease. Using the HbA1c. propensity matching, we could establish well-balanced The composite primary outcome of HbA1c and weight groups and we would have minimized this possibility of this reduction was achieved in 41.3%% of subjects in the SGLT-2i bias but not completely eliminated this. The time of follow- group, followed by lower target achievement among DPP-4i up was different in the cohorts, and this is mainly because (24.2%), and SU (15.2%) groups. Moreover, SGLT-2i users were the SGLT-2i were commercialized later in the study period more likely to achieve this outcome than DPP-4i and SU users, in comparison with the other groups, but we may assume and the magnitude of this effect was estimated as large. The that there is no differential bias since the cohorts were results observed in our study are in concordance with a sim- matched for year of inclusion. We did not use BMI in the ilar study performed in the UK with an 18-month follow-up propensity matching due to quality of the variable and pri- period, where the composite outcome was HbA1c reductions mary objective of our study. Despite this we obtained well bal- of  0.5% (5.5 mmol/mol), and body weight loss  2 kg; 36.5% anced cohort groups for weight and BMI. This is an of users intensified with SGLT-2i achieved this composite out- observational study; therefore, it shows associations but does come, while this percentage was lower for DPP-4i (17.1%) and not allow establishing a causal relationship that is exclusive SU (9.6%) users [23]. of randomized clinical trials. Regarding outcomes individually, we did not find signifi- In summary, the present study showed similar effective- cant differences in terms of HbA1c reduction between the ness on HbA1c reduction among the different drug classes therapy cohorts, independently of the type of analysis (com- as second-line antidiabetic drug following metformin failure. plete cases and imputation case analysis). However, we did However, SGLT-2i were associated with significantly greater observe significant differences in terms of weight changes reductions in weight or the combination target of weight- among the cohorts ( 1.21 kg for DPP-4i, 3.47 kg for SGLT- HbA1c compared to SU and DPP4 inhibitors. Moreover, our 2i, and 0.04 kg for SU). As such, the effect size of the HbA1c findings confirm the existence of far from optimal treatment reduction (and reduction  5%) was trivial between drug intensification practice and suggest the need to overcome classes, while the effect on weight for SGLT-2i was large com- therapeutic inertia also in the early stages of treatment pro- pared to both DPP-4i and SU regarding weight reduction  3%. gression. There is a need for more studies on effectiveness Our real-world findings are consistent with a recent random- in real-world conditions as they can help in better selection ized clinical trials meta-analysis, in which all investigated of treatment for poorly controlled T2DM patients. glucose-lowering drugs lowered HbA1c to a similar extent Disclosures when combined with metformin, but only SGLT-2i showed M. M.-C. has received an advisory and or speaking fees superiority in weight reduction [5,36]. from Astra-Zeneca, Bayer, Boehringer Ingelheim, GSK, Lilly, Regarding changes in systolic blood pressure between the MSD, Novartis, Novo Nordisk, and Sanofi; he has received three therapy groups, results were in favour of the use of research grants to the institution from Astra-Zeneca, GSK, SGLT-2i compared to DPP4i or SUs (reductions of 3.10, 1.07, Lilly, MSD, Novartis, Novo Nordisk, and Sanofi. J. F.-N. has and 0.96 mmHg, respectively). When comparisons among received advisory and or speaking fees from Astra-Zeneca, the different cohorts were made, there were statistically sig- Ascensia, Boehringer Ingelheim, GSK, Lilly, MSD, Novartis, nificant differences in favour of SGLT-2i compared to the Novo Nordisk, and Sanofi; he has received research grants other 2 drug classes, although the magnitude of this effect to the institution from Astra-Zeneca, GSK, Lilly, MSD, Novar- was small. These findings are similar with those of a recently tis, Novo Nordisk, Sanofi, and Boehringer. D.M. has received published study in a cohort study in UK primary care by advisory and/or speaking fees from Astra-Zeneca, Ascensia, Wilkinson et al, where the mean difference (mmHg) over Boehringer Ingelheim, GSK, Lilly, MSD, Novartis, Novo Nor- 96 weeks of follow-up for SGLT2i users was 1.82 (95% CI disk, and Sanofi; he has received research grants to the insti- 3.18, 0.45) compared to DPP4i and 3.06 (95% CI 4.43, tution from Astra-Zeneca, GSK, Lilly, MSD, Novartis, Novo 1.68) compared to SU [31]. Nordisk, Sanofi, and Boehringer. KK has acted as a consultant, Small numerical, although statistically significant differ- speaker or received grants for investigator-initiated studies ences were observed regarding lipid profile: an increase in for Astra Zeneca, Novartis, Novo Nordisk, Sanofi-Aventis, Lilly diabetes research and clinical practice 171 (2021) 108616 9 and Merck Sharp & Dohme, Boehringer Ingelheim, Bayer, network meta-analysis. Diabetes Ther 2018;9:1945–58. Berlin-Chemie AG / Menarini Group, Janssen, and Napp. J.R, https://doi.org/10.1007/s13300-018-0482-5. JA.V, X.M-T, M.F and B.V, have no conflict of interest to declare. [6] Boye KS, Riddle MC, Gerstein HC, Mody R, Garcia-Perez L, Karanikas CA, et al. Generalizability of glucagon-like peptide- Data sharing statement 1 receptor agonist cardiovascular outcome trials to the The data controller for SIDIAP does not allow the sharing overall type 2 diabetes population in the United States. of raw data. Statistical codes are available upon request from Diabetes, Obes Metab 2019;21:1299–304. https://doi.org/ the corresponding authors (J.F.-N or D.M). 10.1111/dom.13649. Author contributions [7] Birkeland KI, Bodegard J, Norhammar A, Kuiper JG, J.F.-N, M.M-C, J.R, D.M, X.M-T, JA.V, and B.V, conceived the Georgiado E, Beekman-Hendriks WL, et al. How representative of a general type 2 diabetes population are research and participated in its design. J.R performed statisti- patients included in cardiovascular outcome trials with cal analysis. B.V wrote the initial draft of the manuscript, SGLT2 inhibitors? A large European observational study. which J.F.-N, M.M-C J.R, D.M, X.M-T, JA.V, M.F and K.K edited. Diabetes, Obes Metab 2019;21:968–74. https://doi.org/ All authors approved the final manuscript. 10.1111/dom.13612. [8] Mata-Cases M, Franch-Nadal J, Real J, Mauricio D. Glycaemic control and antidiabetic treatment trends in primary care Declaration of Competing Interest centres in patients with type 2 diabetes mellitus during 2007– 2013 in Catalonia: a population-based study. BMJ Open The authors declare that they have no known competing 2016;6:e012463. https://doi.org/10.1136/bmjopen-2016- financial interests or personal relationships that could have 012463. appeared to influence the work reported in this paper. [9] Persson F, Bodegard J, Lahtela JT, Nyström T, Jørgensen ME, Jensen ML, et al. Different patterns of second-line treatment in type 2 diabetes after metformin monotherapy in Denmark, Acknowledgement Finland, Norway and Sweden (D360 Nordic): A multinational observational study. Endocrinol Diabetes Metab 2018;1: We thank Mònica Gratacòs for the editorial support of this e00036. https://doi.org/10.1002/edm2.36. manuscript. This study was supported by AstraZeneca, Spain [10] Nicolucci A, Charbonnel B, Gomes MB, Khunti K, Kosiborod [grant number ESR-16-12628]. Kamlesh Khunti is supported M, Shestakova MV, et al. Treatment patterns and associated by the National Institute for Health Research (NIHR) Applied factors in 14 668 people with type 2 diabetes initiating a second-line therapy: Results from the global DISCOVER study Research Collaboration East Midlands (ARC EM) and the NIHR programme. Diabetes, Obes Metab 2019;21:2474–85. https:// Leicester Biomedical Research Centre (BRC). Bogdan Vlacho is doi.org/10.1111/dom.13830. supported by the Instituto de Salud Carlos III (ISCIII, Fondo [11] Bolı́bar B, Fina Avilés F, Morros R, del Mar G-G, Hermosilla E, Investigación sanitaria-FIS-Fondo Europeo de Desarrollo Ramos R, et al. Base de datos SIDIAP: la historia clı́nica Regional-FEDER: Plataforma de Unidades de Investigación informatizada de Atención Primaria como fuente de Clı́nica y Ensayos Clı́nicos-SCREN. información para la investigación epidemiológica. Med Clin (Barc) 2012;138:617–21. https://doi.org/10.1016/j. medcli.2012.01.020. Appendix A. Supplementary material [12] Mata-Cases M, Franch-Nadal J, Mauricio D, Bolı́bar B. Investigar en diabetes desde una base de datos de atención Supplementary data to this article can be found online at primaria: la experiencia del Sistema de Información para el https://doi.org/10.1016/j.diabres.2020.108616. Desarrollo de la Investigación en Atención Primaria (SIDIAP). Av En Diabetol 2013;29:169–74. https://doi.org/10.1016/j. avdiab.2013.09.002. [13] Mata-Cases M, Mauricio D, Real J, Bolı́bar B, Franch-Nadal J. Is R E F E R E N C E S diabetes mellitus correctly registered and classified in primary care? A population-based study in Catalonia, Spain. Endocrinol y Nutr 2016;63:440–8. https://doi.org/10.1016/j. [1] Association AD. 9. Pharmacologic approaches to glycemic endonu.2016.07.004. treatment: standards of medical care in diabetes-2020. [14] Ho DE, Imai K, King G, Stuart EA. MatchIt : nonparametric Diabetes Care 2020;43:S98. preprocessing for parametric causal inference. J Stat Softw [2] Davies MJ, D’Alessio DA, Fradkin J, Kernan WN, Mathieu C, 2011;42.. https://doi.org/10.18637/jss.v042.i08. Mingrone G, et al. A Consensus Report by the American [15] Team RC. R: A language and environment for statistical Diabetes Association (ADA) and the European Association for computing 2013. the Study of Diabetes (EASD). Diabetes Care 2018;2018 [16] van Buuren S, Groothuis-Oudshoorn K. mice : multivariate (41):2669–701. https://doi.org/10.2337/dci18-0033. imputation by chained equations in R. J Stat Softw [3] NICE. Type 2 diabetes in adults: management. NG28. 2011;45:2007. https://doi.org/10.18637/jss.v045.i03. December 2015 [17] Lüdecke D, Ben-Shachar M, Patil I, Makowski D. Extracting, [4] Garber AJ, Abrahamson MJ, Barzilay JI, Blonde L, Bloomgarden computing and exploring the parameters of statistical ZT, Bush MA, et al. Consensus statement by the american models using R. J Open Source Softw 2020;5:2445. https://doi. association of clinical endocrinologists and american college org/10.21105/joss.02445. of endocrinology on the comprehensive type 2 diabetes [18] Ben-Shachar M, Makowski D, Lüdecke D. Compute and management algorithm – 2016 executive summary. Endocr interpret indices of effect size. R package: CRAN; 2020. Pract 2016;22:84–113. https://doi.org/10.4158/EP151126.CS. https://github.com/easystats/effectsize. [5] Qian D, Zhang T, Zheng P, Liang Z, Wang S, Xie J, et al. [19] Type 2 Diabetes in Adults. Staged Diabetes Manag., Oxford, Comparison of oral antidiabetic drugs as add-on treatments UK: Wiley-Blackwell; 2011, p. 77–137. DOI: 10.1002/ in patients with type 2 diabetes uncontrolled on metformin: a 9781119950424.ch5. 10 diabetes research and clinical practice 171 (2021) 108616 [20] DeFronzo RA, Stonehouse AH, Han J, Wintle ME. Relationship Diabetes, Obes Metab 2018;20:389–99. https://doi.org/ of baseline HbA 1c and efficacy of current glucose-lowering 10.1111/dom.13083. therapies: a meta-analysis of randomized clinical trials. [29] Thomsen RW, Baggesen LM, Søgaard M, Pedersen L, Diabet Med 2010;27:309–17. https://doi.org/10.1111/j.1464- Nørrelund H, Buhl ES, et al. Early glycaemic control in 5491.2010.02941.x. metformin users receiving their first add-on therapy: a [21] Khunti K, Wolden ML, Thorsted BL, Andersen M, Davies MJ. population-based study of 4,734 people with type 2 diabetes. Clinical inertia in people with type 2 diabetes: a retrospective Diabetologia 2015;58:2247–53. https://doi.org/10.1007/s00125- cohort study of more than 80,000 people. Diabetes Care 015-3698-1. 2013;36:3411–7. https://doi.org/10.2337/dc13-0331. [30] Desai U, Kirson NY, Kim J, Khunti K, King S, Trieschman E, [22] Conthe P, Mata M, Orozco D, Pajuelo F, Barreto CS, Anaya SF, et al. Time to treatment intensification after monotherapy et al. Degree of control and delayed intensification of failure and its association with subsequent glycemic control antihyperglycaemic treatment in type 2 diabetes mellitus among 93,515 patients with type 2 diabetes. Diabetes Care patients in primary care in Spain. Diabetes Res Clin Pract 2018;41:2096–104. https://doi.org/10.2337/dc17-0662. 2011;91:108–14. https://doi.org/10.1016/j.diabres.2010.10.002. [31] Wilkinson S, Williamson E, Pokrajac A, Fogarty D, Stirnadel- [23] Wilding J, Godec T, Khunti K, Pocock S, Fox R, Smeeth L, et al. Farrant H, Smeeth L, et al. Comparative effects of Changes in HbA1c and weight, and treatment persistence, sulphonylureas, dipeptidyl peptidase-4 inhibitors and over the 18 months following initiation of second-line sodium-glucose co-transporter-2 inhibitors added to therapy in patients with type 2 diabetes: results from the metformin monotherapy: a propensity-score matched cohort United Kingdom Clinical Practice Research Datalink. BMC study in UK primary care. Diabetes, Obes Metab Med 2018;16:116. https://doi.org/10.1186/s12916-018-1085-8. 2020;22:847–56. https://doi.org/10.1111/dom.13970. [24] Montvida O, Shaw J, Atherton JJ, Stringer F, Paul SK. Long- [32] Fau E, Mata-Cases M, Morros R, Pellicer À, Ricart W, Vallès JA, term trends in antidiabetes drug usage in the U.S.: Real-world et al. Pautes per al tractament farmacològic de la diabetis evidence in patients newly diagnosed with type 2 diabetes mellitus tipus 2. Agència Qual i Avaluació Sanitàries 41:69–78. Diabetes Care 2018. https://doi.org/10.2337/dc17-1414. Catalunya 2013. http://scientiasalut.gencat.cat/handle/ [25] Mata-Cases M, Franch-Nadal J, Gratacòs M, Mauricio D. 11351/1403 (accessed November 28, 2020). Therapeutic inertia: still a long way to go that cannot be [33] Zinman B, Wanner C, Lachin JM, Fitchett D, Bluhmki E, Hantel postponed. Diabetes Spectr 2020;33:50–7. https://doi.org/ S, et al. Empagliflozin, cardiovascular outcomes, and 10.2337/ds19-0018. mortality in type 2 diabetes. N Engl J Med 2015;373:2117–28. [26] Mata-Cases M, Franch-Nadal J, Real J, Gratacòs M, López- https://doi.org/10.1056/NEJMoa1504720. Simarro F, Khunti K, et al. Therapeutic inertia in patients [34] Neal B, Perkovic V, Matthews DR, Mahaffey KW, Fulcher G, treated with two or more antidiabetics in primary care: Meininger G, et al. Rationale, design and baseline Factors predicting intensification of treatment. Diabetes, characteristics of the CANagliflozin cardioVascular Obes Metab 2018;20:103–12. https://doi.org/10.1111/dom.1304 Assessment Study-Renal (CANVAS-R): A randomized, hh5. placebo-controlled trial. Diabetes, Obes Metab [27] Seidu S, Cos X, Brunton S, Harris SB, Jansson SPO, Mata-Cases 2017;19:387–93. https://doi.org/10.1111/dom.12829. M, et al. A disease state approach to the pharmacological [35] Wiviott SD, Raz I, Bonaca MP, Mosenzon O, Kato ET, Cahn A, management of Type 2 diabetes in primary care: A position et al. Dapagliflozin and cardiovascular outcomes in type 2 statement by Primary Care Diabetes Europe. Prim Care diabetes. N Engl J Med 2019;380:347–57. https://doi.org/ Diabetes 2020. https://doi.org/10.1016/j.pcd.2020.05.004. 10.1056/NEJMoa1812389. [28] Khunti K, Godec TR, Medina J, Garcia-Alvarez L, Hiller J, [36] Palmer SC, Mavridis D, Nicolucci A, Johnson DW, Tonelli M, Gomes MB, et al. Patterns of glycaemic control in patients Craig JC, et al. Comparison of clinical outcomes and adverse with type 2 diabetes mellitus initiating second-line therapy events associated with glucose-lowering drugs in patients after metformin monotherapy: Retrospective data for 10 256 with type 2 diabetes a meta-analysis. JAMA - J Am Med Assoc individuals from the United Kingdom and Germany. 2016;316:313–24. https://doi.org/10.1001/jama.2016.9400.