Lower Risk of Death and Cardiovascular Events in Unselected Type 2 Diabetes Patients Initiated on GLP-1 Receptor Agonists and SGLT-2 Inhibitors in Two Large Italian Cohorts

Background: GLP-1 receptor agonists (G LP-1 RA) and SGLT-2 inhibitors (SGLT-2i) are recommended in type 2 diabetes subjects with established cardiovascular (CV) disease or at high CV risk. The ecacy and safety of GLP-1 RA and SGLT-2i compared with other anti-hyperglycemic agents (AHAs) was examined in large unselected populations of Lombardy and Apulia regions in Italy. Methods: An observational cohort study of new users of GLP-1 RA, SGLT-2i and other AHAs was conducted from 2010 to 2018. Death and cardiovascular events were evaluated using conditional Cox models in propensity-score matched populations. Adjusted hazard ratios (95%, CI) were calculated for each region and as meta-analysis for pooled risks. Apulia), and heart failure (HR 0.56, CI 0.46-0.70, Lombardy, HR 0.57, CI 0.42-0.77, Apulia). In the pooled cohorts, a reduction in heart failure was also observed with GLP-1 RA (HR 0.89, 95% CI 0.82-0.97). Serious adverse events were quite low in frequency. (IR) the number of incident events divided by the total number of person-years at per 100 person-years with 95% condence (CI). of in each calculated the number of incident events divided by the total number of persons regression on time to rst used hazard ratios (HRs) and 95% CI for each outcome, comparing the treatment effect of GLP-1 RA and SGLT-2i versus (reference group). HRs were adjusted for all comorbidities reported at baseline. by intention-to-treat. The frequency serious by total


Introduction
Patients with diabetes are at high risk for adverse outcomes from atherosclerotic cardiovascular (CV) disease [1][2][3][4] , heart failure (HF), and renal disease. Intensive and early control of hyperglycemia in type 2 diabetes (T2D) subjects may reduce the incidence of non-fatal myocardial infarction and coronary artery disease, while it does not affect the incidence of stroke and mortality [5][6][7][8] . Until recently, however, antihyperglycemic therapy did not conclusively prove to reduce overall macro-vascular events in T2D, and there was even concern that some medications could cause CV harms and increase mortality 9,10 .
However, a series of large CV outcomes trials (CVOTs) in T2D patients, principally designed to meet regulatory requirements for CV safety, recently assessed the effects of new anti-hyperglycemic agents (AHAs), such as glucagon-like peptide-1 receptor agonists (GLP-1 RA) and sodium glucose cotransporter-2 inhibitors (SGLT-2i), in addition to standard of care. In most cases, the results of these trials showed a signi cant reduction of major CV outcomes with the investigational drug compared with placebo (i.e., standard of care) 11,12 . Speci cally, GLP-1 RA showed a prevalent CV bene t in terms of reduced incidence of atherosclerotic events, such as non-fatal myocardial infarction (MI) and stroke, while SGLT-2i consistently reduced the risk of hospitalization for heart failure (HF). The use of both drug classes was also associated with reduction in the risk of renal endpoints, SGLT2-i acting both on the decline of glomerular ltration rate and albuminuria, GLP-1 RAs mainly on the latter [13][14][15][16][17][18][19][20] . Based on these results, in patients with established CV disease or at high CV risk, established kidney disease, or HF, treatment with SGLT-2i or GLP-1 RA is currently recommended as part of the glucose-lowering regimen [21][22] .
The clinical trials showed that treatment with GLP-1 RA and SGLT-2i has been also associated with speci c adverse events, such as increased risk for gastrointestinal side effects, genitourinary tract infections, and fractures 19,23,24 . This may limit the inclusion of these drugs in the therapeutic regimen, as well as adherence and persistence to these therapies in the long-term.
Some real-world studies have also investigated the e cacy and mortality outcomes of these new drugs, showing that initiation of SGLT-2i versus other AHAs was associated with a pertinent lower incidence of hospitalization for HF and death; on the other hand, the safety outcomes were not investigated or only in a limited sample size in this setting [25][26][27][28][29] . In addition, these observational studies have largely focused on the comparison between SGLT-2i and other AHAs, while information on population-based cohort studies comparing the e cacy and safety of GLP-1 RA versus other AHAs is more limited 30,31 To assess whether the results of randomized clinical trials with the new AHAs are generalizable to the whole diabetes population is important to analyze cohorts of patients treated in daily practice. The current availability of large administrative databases allows verifying the expected clinical bene ts and risks in routine clinical settings. In this study, we used administrative data from two highly populated Italian regions, Lombardy and Apulia. We aimed to describe the clinical characteristics of new users of GLP-1 RA and SGLT-2i in comparison with those receiving other AHAs and to compare the risks of death and major CV events as well the safety in these large unselected populations.

Data source
Our study used linkable administrative health databases of Lombardy and Apulia regions in Italy, which include population registries with demographic data of all residents and detailed information on drug prescriptions and hospital records. Data are available for about 10 and 4 million inhabitants of Lombardy and Apulia, respectively, from 2000 to 2018. Access to data is allowed within the agreement between the Istituto di Ricerche Farmacologiche Mario Negri (IRFMN) and Regional Health Ministry of Lombardy and, between the IRFMN and the Regional Healthcare Agency of Apulia Healthcare in Italy is publicly funded for all residents, irrespective of social class or employment, and everyone is assigned a personal identi cation number kept in the National Civil Registration System. All residents are assisted by general practitioners (GPs) under the National Health System (NHS). The pharmacy prescription database contains the medication name and anatomic therapeutic chemical (ATC) classi cation code, quantity, and date of dispensation of drugs reimbursed by the NHS. No information is available on drugs dispensed in hospital. The hospital databases contain information on date of admission, discharge, death, primary diagnosis, and up to ve co-existing clinical conditions and procedures received.
The diagnoses, uniformly coded according to the 9th International Code of Diseases (ICD-9-CM) and standardized in all Italian hospitals, are compiled by the hospital specialists directly in charge of the patients and are validated by hospitals against detailed clinical-instrumental data, as they determine reimbursement from the NHS. A unique identi cation code allows linkage of all databases. To ensure privacy, each identi cation code was automatically converted into an anonymous code before we received the dataset. In Italy, studies using retrospective aggregated-anonymous data from administrative databases do not require Ethics Committee/IRB approval nor noti cation.

Study cohorts and follow-up
We conducted a cohort study using the two administrative health databases from Lombardy and Apulia.
Subjects 50 years and older with chronic exposure to AHAs (at least two packages in the year -ATC code A10*-) from January 1, 2010, through December 31, 2018, were included in the analysis. Subjects were split into three groups according to the rst exposure (new users) to one of the following drug classes: GLP-1 RA, SGLT-2i or other AHAs including metformin, sulfonylureas, glinides, thiazolidinediones, acarbose, dipeptidyl-peptidase-4 inhibitors (DPP-4i). To avoid potential bias and compare relatively similar populations, subjects who were new users for insulin (as chronic treatment) were not included in the other AHAs group. Indeed, insulin therapy may be a marker of more severe and/or advanced disease and we could not adjust for disease severity due to the lack of availability of full clinical data of these cohorts. However, a percentage of subjects within the AHAs group may have had previous insulin exposure in previous years before entering cohort, but not as a chronic treatment according to selection criteria used in this analysis. New users were de ned as subject whose rst exposure (index date) to one of the anti-hyperglycemic drug classes occurred in the index year with no prior exposure to any of medications belonging to the same class in the previous ve years before entering the cohort. Subjects initiating GLP-1 RA and SGLT-2i were included in the study cohort from 2010 and 2015, respectively, according to the availability of these drugs in the Italian market.
The propensity score matching (PSM) was used to reduce confounding due to imbalance in study covariates. A systematic approach to selection of variables was used to create balanced cohorts, attempting to exclude as few subjects as possible. The PMS matching was used to match GLP-1 RA and SGLT-2i cohorts with other AHAs in a 1:1 ratio using the following variables: age classes ( ve years intervals), sex, index year, prior exposure to insulin, duration of diabetes, and the Drug Derived Complexity Index (DDCI) as a proxy of comorbidities. The DDCI is a predictive score derived from drug prescriptions able to stratify the general population according to the risk of one-year and long-term mortality, as well as the risk of unplanned hospitalization and hospital readmission 32 . Subjects were followed-up from drug initiation until the rst occurrence of i.) outcomes of interest or ii.) migration, admission to a nursing home or up to the end of follow-up (December 31, 2018). A sensitivity analysis having new users for DDP4-i as a comparator cohort was also carried out. This comparison allowed to assess the effects of GLP-1 RA and SGLT-2i versus a drug class, i.e. DPP4-i, known to exhibit neutral effects on major CV outcomes in T2D subjects [33][34][35][36] ; moreover, GLP-1 RA, SGLT-2i and DPP-4i are only prescribed by diabetes specialists in Italy.

Comorbidities and pharmacological treatments
Comorbidities were gathered in the ve years before the index date using hospital records according to ICD 9-CM as primary diagnosis and up to ve co-existing conditions. Previous exposure to any AHAs class, hospital admissions and DDCI-index were collected in the previous ve years, while information on the other medications of interest was retrieved in the previous twelve months. Information on duration of diabetes was collected from 2000 to 2018 (Supplementary materials).

Study Outcomes
Outcomes included the following events: death from any cause, hospital admission for cerebrovascular disease, CV disease, ischemic stroke, acute coronary syndrome, heart failure, peripheral vascular disease, lower limb complications, as primary diagnosis. Serious adverse events including hospital admission for hypoglycemia, ketoacidosis, diabetes with coma, amputation, acute renal failure, syncope, and fracture as primary diagnosis were also analyzed. All clinical events were collected using hospital admission according to the ICD 9-CM codes (Supplementary Material). Renal outcomes were not analyzed since the initiation of SGLT-2i therapy is indicated only in individuals with eGFR > 60 ml/m 2 /min; this may likely select a patient cohort with better renal health compared to that treated with other AHAs, representing a bias for analyzing renal outcomes.

Statistical analysis
Baseline characteristics of patients in each groups of treatment were evaluated by using descriptive statistics. Categorical variables were described by frequencies and percentages and compared using Chi-squared test, continuous variables by using mean ± standard deviation (SD), and compared using Student's t-test. Drug Derived Complexity Index (DDCI), previous hospital admission, history of diabetes and follow-up times were expressed as median and interquartile range [IQR].
Participants were matched on the logit of the propensity score using calipers of width equal to 0.1 of the standard deviation of the logit of the estimated propensity score 37 . Speci cally, based on PSM, patients receiving GLP-1 RA or SGLT-2i were matched 1:1 with those receiving other AHAs. The adequacy (congruency) of PSM was assessed by standardized differences of post-matching patients' characteristics. To evaluate the balance between groups after matching, we calculated the standardized mean difference (SMD); good balance is conventionally set at SMD < 0.10 38 .
Longitudinal analyses were performed in matched populations. Outcomes were calculated as crude incidence rate (IR) as the number of incident events divided by the total number of person-years at risk and expressed per 100 person-years with 95% con dence interval (CI). The percentage of events in each group was calculated as the number of incident events divided by the total number of persons at risk. Cox proportional hazard regression model based on time to rst event was used to estimate hazard ratios (HRs) and 95% CI for each outcome, comparing the treatment effect of GLP-1 RA and SGLT-2i versus other AHAs (reference group). HRs were adjusted for all comorbidities reported at baseline. Outcomes were analyzed by intention-to-treat. The frequency of serious adverse events was calculated as the number of events divided by total population in each group of treatment.
Results are presented for each region separately, since the Italian Privacy Policy on data protection does not allow to export pooled data from multiple health administrative databases of distinct Italian regions.
Then, pooled risks from the two regions were presented as meta-analysis for an overall summary. Isquare was used to calculate heterogeneity among the studies. A probability value of I-square ≥ 50% indicated the presence of signi cant heterogeneity. The xed effects model was used in the presence of no signi cant interstudy heterogeneity; otherwise, the random effects model was used. The log-rank test, strati ed by region, was used for comparisons, and the hazard ratios with 95% CI of events were calculated. A p-value < 0.05 was considered statistically signi cant. A sensitivity analyses was computed comparing GLP-1 RA and SGLT-2i with DPP-4i. All analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA).

Results
Overall, during the study period, 29,634 new users of GLP-1 RA, 25,141 new users of SGLT-2i and 228,274 new users of other AHAs were identi ed (baseline patients characteristics before matching are reported in Supplementary material). The median follow-up time for GLP-1 RA, SGLT-2i and other AHAs, was 4.0 ± 2.7, 1.7 ± 1.0 and 5.2 ± 2.8 years in Lombardy, and 3.5 ± 2.7, 1.7 ± 0.9 and 5.3 ± 2.8 years in Apulia, respectively.
After PSM, the Lombardy study cohort included 18,716 pairs for the GLP-1 RA group and 11,683 pairs for the SGLT-2i group, while the Apulia study cohort comprised 9,772 and 6,046 for the GLP-1 RA and SGLT-2i groups, respectively (Table 1, A and B). After matching, the variables included in the PMS were wellbalanced (all standardized differences were < 0.1). In general, in both regions, subjects newly prescribed a GLP-1 RA or a SGLT-2i had similar comorbidities as controls, although patients initiating SGLT-2i were more likely to have CV disease in comparison with new users of other AHAs. In the GLP-1 RA and SGLT-2i groups, 4.0-7.0% and 12-16% of patients had established cerebrovascular and CV diseases, respectively, in both Lombardy and Apulia regions ( Table 1, A and B).
Patients belonging to the GLP-1 RA or SGLT-2i groups presented with slightly higher rates of background anti-hyperglycemic treatment and received slightly more antihypertensive and lipid-lowering medications as compared to the other AHAs cohort ( Table 1, A and 1B).
Differences between the GLP-1 RA and SGLT-2i groups in both regions were observed in mean age (67-68 years in SGLT-2i group versus 64-65 years in GLP-1 RA group), history of diabetes (a higher percentage of subjects in the SGLT-2i group had a duration of T2DM ≥ 10 years compared to the GLP-1 RA group), and previous prescription of insulin (34% in SGLT-2i group versus 29% in GLP-1 RA group).
Risks of death and clinical outcomes in propensity score-adjusted populations by treatment status for Lombardy and Apulia regions are reported in Fig. 1, A  When results from the two cohorts were pooled, a small but signi cant reduction in the risk of hospitalization for heart failure (HR 0.89, 95% CI 0.82-0.97) with GLP-1 RA compared with other AHAs was also apparent (Fig. 1S Supplementary Material).
During follow-up, the rate of serious adverse events was quite low in each region. In general, fractures were documented more frequently; however, the rate of this event was slightly lower in the SGLT-2i group (around 1%) than in the GLP-1 RA and AHAs groups (2.5%) ( Table 2).
Results of the preplanned sensitivity analysis comparing GLP-1 RA or SGLT-2i with DPP-4i for each region are reported in the Supplementary material, showing baseline characteristics of matched and unmatched populations as well as risks for all considered outcomes. After PSM, the population of the two cohorts were well-matched for multiple clinical variables (all standardized differences were < 0.1, except for renal disease and heart failure in some of the comparisons; Supplementary Material). In comparison with subjects who received DPP-4i, those initiating GLP-1 RA showed statistically signi cant risk reductions for death, cerebrovascular disease, peripheral vascular disease, lower limb complications (Lombardy cohort, Table S9), and death and lower limb complications (Apulia cohort, Table S11), while those initiating SGLT-2i had risk reductions for death, cerebrovascular disease and heart failure (Lombardy cohort, Table S10), and death and heart failure (Apulia cohort, Table S12). These results were similar to those observed in comparison with other AHAs.

Discussion
In this analysis, we have examined large cohorts of T2D initiating treatment with GLP-1 RA or SGLT-2i compared with other anti-hyperglycemic therapies, including DPP-4i, in routine clinical settings. The available information refers to data from two distinct Italian regions, Lombardy and Apulia, which allows an assessment of the epidemiology of diseases unbiased by patient selection. In Italy, all subjects are covered by the NHS, according to the distinct reimbursement policies, with a high level of completeness regarding drug prescriptions, diagnosis, and length of observation. Administrative databases have been increasingly recognized as a reliable tool to prospectively describe the pharmaco-epidemiology and outcomes of large patients cohorts representing the real clinical care, collecting data over time in a standardized way and at low cost 39,40 . We found that initiation of GLP-1 RA therapy was associated with consistent risk reductions in all-cause death and hospitalization for peripheral vascular disease and lower limb complications, with additional risk reductions for cerebrovascular disease and ischemic stroke that were evident in the Lombardy cohort. Initiation of SGLT-2i was associated with risk reductions of allcause death and hospitalization for cerebrovascular disease, and heart failure. These risk reductions occurred in comparison with the initiation of other AHAs (except insulin), and most differences were also observed after comparing the new users of GLP-1 RA or SGLT-2i with patients initiating DPP-4i.
As of today, results from seven CVOTs with GLP-1 RA have been disclosed. The de nition of secondary prevention cohorts according to previous CV disease of the enrolled population varied among these trials, and so was the proportion of such patients 11 . Nevertheless, ELIXA and HARMONY Outcome involved only patients with recent acute coronary syndrome or any CV disease, respectively; by contrast, REWIND assessed a population with 70% of individuals without prior CV disease and with the lowest proportion (only 8%) affected from congestive HF. Large observational studies assessing CV outcomes with GLP-1 RA are not available at present. The CVOTs with SGLT-2i enrolled patients at high CV risk (with percentages of subjects with atherosclerotic CV disease ranging from 41% in DECLARE-TIMI 58 to 65% in CANVAS and 100% in EMPA-REG). In these trials, at baseline, subjects with HF ranged from 10-14% of the population, while between 6.5% and 23.3% had a stroke 12 . In the real-world observational studies assessing CV outcomes with SGLT-2i, the CV risk level of the examined populations is very different. CVD-REAL and CVD-REAL-2 included 13% and 26% of patients with T2D and established CV disease, respectively 25,29 , while in CVD-REAL Nordic, this proportion was 25% 41 . In the CVD-REAL program, the proportion of patients with HF at baseline ranged from 3-6.8%. By contrast, EASEL, involved only patients with T2D and established CV disease 42 . In our study, the proportion of individuals with CV disease was between 12-16% and that with HF between 3-6% (Table 1). Moreover, these proportions did not appear to differ between Lombardy and Apulia. Therefore, the population examined here had a level of CV risk largely lower than in the CVOTs with GLP-1 RA or SGLT-2i and somewhat similar to that in the observational study with SGLT-2i CVD-REAL.
In our analysis, rst use of GLP-1 RA was associated with 37-39% reduced risk of all-cause death and reductions of peripheral vascular disease and lower limb complications. Meta-analysis of the GLP-1 RA CVOTs also showed an overall 12% reduction in the risk of all-cause mortality 14 , while the effects on peripheral vascular disease and lower limb complications were not considered as primary or secondary endpoints in those trials. The risk reductions for cerebrovascular disease and ischemic stroke observed in the Lombardy cohort is of interest, given that in some GLP-1 RA CVOTs, such as REWIND 16 and SUSTAIN-6 43 the risk of stroke was also reduced with the investigational GLP-1 RA. In a recent study on a relatively smaller cohort from North-East Italy, including ~ 2,800 propensity score matched initiators of GLP-1RA or DPP-4i, 15% with pre-existing cardiovascular disease, reduced rates of a composite of allcause death, myocardial infarction, or stroke (HR: 0.67; 95% C.I. 0.53-0.86) were found in patients treated with GLP-1RA compared to DPP4i 31 .
In our study, rst use of SGLT-2i was associated with large risk reductions of all-cause death (by 53-57%) and HF (by 43-44%), as well as of cerebrovascular disease (by 25-28%). The risk reduction in HF hospitalization has been consistently observed in all CVOTs and observational studies with SGLT-2i, while all-cause death was reduced in EMPA-REG and in the observational studies. Of note, the effect size of those reductions resembles the one found in the present analysis. Results from Lombardy and Apulia also consistently show a reduction in the risk of cerebrovascular disease. While the risk of fatal or nonfatal stroke was not changed in the three major SGLT-2i CVOTs, the observational study CVD-REAL 2 (conducted in Australia, Canada, Israel, Japan, Singapore, and South Korea on 235,064 patients with T2DM) also showed an association of SGLT-2i use with a signi cantly reduced risk of stroke (HR: 0.68; 95% CI: 0.55, 0.84) 29 , in line with this analysis.
The results of the meta-analysis between Lombardy and Apulia study cohorts largely con rmed the results obtained from the main analysis of the individual regions but also provided further information.
Pooling the data, initiation of GLP-1 RA was not associated with a lower risk of cerebrovascular disease and ischemic stroke, as observed in the Lombardy region. However, GLP-1 RA initiation was associated with a small but signi cant lower risk of HF. This observation deserves further investigation with dedicated studies. Similar risk reductions were found for all-cause death and other CV outcomes comparing initiation of GLP-1 RA with initiation of DPP-4i, while patients initiating SGLT-2i had consistent risk reductions for death and HF in both regional cohorts. This data is of interest, since initiation of GLP-1 RA and SGLT-2i is being compared here with a drug class, DPP4-i, that also does not cause hypoglycemia or weight gain.
Moreover, the two treatment strategies were equally positioned in the Italian treatment algorithm and could be prescribed only by diabetes specialists. Such direct comparisons have not been addressed in the CVOTs, while a single additional analysis from CVD-REAL Nordic revealed that the SGLT-2i dapagli ozin was also associated with signi cantly lower incidence of hospitalization for HF, all-cause mortality and major adverse CV events in comparison with DPP-4i 44 ; moreover, an interim analysis from the ongoing EMPRISE observational study including 224,528 patients with T2D with and without established CV disease reported that initiation of SGLT-2i was associated with a 48% lower rate of hospitalization for HF in comparison with DPP-4i 27 .
In general, our analysis shows a low occurrence of adverse effects that could be captured using the administrative database, including ketoacidosis, amputations, renal failure, syncope, and fractures, which represent the most worrisome adverse events associated with SGLT-2i use emerging from randomized controlled trials 45 .

Strengths And Limitations
To our knowledge, this is the rst study of real-world evidence that evaluated the effect of GLP-1 RA on mortality and major adverse CV events in comparison with other AHAs. The analysis was conducted on a large number of patients after PSM (30,399 in Lombardy and 15,818 in Apulia), initiating both GLP-1 RA and SGLT-2i. Results found with GLP-1 RA and SGLT-2i in comparison with other AHAs were largely con rmed when DPP-4i were used as a comparator. The length of observation was almost four years for GLP-1 RA and two years for SGLT-2i. Finally, Lombardy and Apulia are two representative regions of Northern and Southern Italy, respectively; the results of this analysis, therefore, could be potentially generalized to the whole Italian population.
Our study also has several limitations that are typical of all the studies on administrative databases. First on the basis of available information, we are unable to distinguish between type 1 and type 2 diabetes, so both are considered in this study even though more than 90% are estimated to have T2D 46 . Moreover, some speci c information on clinical variables or laboratory tests (such as HbA1C, BMI, eGFR) that deserves attention mainly referring to glycemic values, lifestyle habits, CV risk factors, or NYHA class is not available. Thus, we could not correct for these confounding factors, as well as to distinguish between primary versus secondary prevention.
Our results must be interpreted with caution since our study was not randomized, and some clinically important characteristics might be not have been taken into account. Finally, different length of follow-up was considered for GLP-1 RA and SGLT-2i (much shorter for SGLT-2i), and somewhat larger use of antihypertensive and lipid-lowering medications was found in the GLP-1 RA and SGLT-2i groups.
A potential issue with large pharmaco-epidemiological studies, such as CVD-REAL, EASEL, EMPRISE and the present analysis, is the possibility of 'immortal time bias', which could exaggerate the observed bene ts regarding rates of all-cause death. Immortal time bias could occur when two patient groups are formed within a time interval in a hierarchical manner 47,48 . While propensity matching of patients treated with SGLT-2i, GLP-1 RA or other AHAs minimizes the risk of such bias, residual confounding could still in uence results even following PSM.

Conclusions
Most ndings of reduction of all-cause death, vascular outcomes and HF were consistent in the two Italian regions, and the pattern of protection for each class seen in the CVOTs and, when available, observational studies from real-world evidence was largely con rmed by this analysis (e.g. hospitalization for HF with SGLT-2i, effects on CV disease and stroke with GLP-1 RA). The favorable effects of both GLP-1 RA and SGLT-2i on all-cause death and several CV endpoints, as well as the safety outcomes, over other commonly used glucose-lowering therapies and in a population with a low prevalence of CV complications suggests that GLP-1 RA and SGLT-2i should be preferentially used not only in high CV risk T2D patients.

Declarations
Funding This study was supported by grant from the Health Ministry of Lombardy and Apulia that was involved in data collection. Access to data is allowed within the Agreement between the Istituto di Ricerche Farmacologiche Mario Negri and Regional Health Ministry of Lombardy and, between the Istituto di Ricerche Farmacologiche Mario Negri and the Regional Healthcare Agency of Apulia.

Availability of data and materials
Access to row data is allowed only to the investigators within the Agreement. Programme code can be shared following the permission granted from the Regions involved in the Agreement with Istituto di Ricerche Farmacologiche Mario Negri.

Ethics approval and consent to participate
In Italy, studies using retrospective aggregated-anonymous data from administrative databases do not require Ethics Committee/IRBapproval nor noti cation.    Supplementarymaterial.pdf