The prevalence of the CAD-T2DM without prior MI or stroke and THEMIS-like populations, estimated with sex-age standardization for the European adult population, was 6.17 and 1.53 per 1,000 persons, respectively, increasing with age and with a higher prevalence for men. Thus, the THEMIS-like population represented a quarter of all CAD-T2DM without prior MI or stroke, as well as for men, women and each age-class. The 2-year cumulative incidence of main outcome was lower for the THEMIS-like population than for all CAD-T2DM without prior MI or stroke patients, 1.3% and 1.7% for MI, 1.5% and 1.7% for stroke, 3.1% and 4.8% for major bleeding, 9.7% and 13.6% for all-cause death, 12.0% and 16.2% for the composite of MI, stroke and all-cause death, respectively.
Our results showed that the THEMIS-like population have same mean age, sex-ratio, T2DM and CAD history duration than all CAD-T2DM without prior MI or stroke patients, but a lower rate of comorbidities and diabetic complications, than can be explained by THEMIS exclusion criteria, especially renal failure requiring dialysis and liver cancer or cirrhosis, and probably also anticoagulant treatment or antiplatelet agent prescribed for other comorbidities.
Compared to the patients in the THEMIS RCT placebo arms, our THEMIS-like population was 6 years older (72 vs. 66 years) with same sex-ratio and duration of diabetes, a little more peripheral arterial disease (11.4% vs 9.0%) and diabetes complications (31.6% vs 25.3%), but less coronary arterial revascularization 18.6% within the 5-year history versus 79.9% any time with a median of 4.1 years for the most recent. The rate of hypertension and dyslipidemia was also lower (75.5% vs. 92.4% and 16.6% vs. 87.1%, respectively), but probably related to poor recording, which contrasts with 71.4% use of statins, and 72.0% use of ACE inhibitors or angiotensin receptor blockers, 33.1% use of calcium channel blockers within the year before index date. One patient out of 6 (17.0%) had cancer, but this was not specified for THEMIS trial.
In the THEMIS RCT, outcome incidences were also estimated using the Kaplan-Meier estimate but reported after 3 years of follow-up. We estimate the cumulative incidence at 2-years as two-thirds of the 3-year result, considering constant increase during time according to the figures of the publications (15, 16). The 2-year cumulative incidence of the composite of MI, any stroke, or all-cause death was 12.0% for the THEMIS-like population, the double than for the same outcome of the placebo arm of the THEMIS RCT (first Exploratory outcomes). The difference is related to all-cause death with a three times higher 2-year cumulative incidence (9.7% vs 3.2%), whereas this incidence was closer for ischemic stroke (1.5% vs. 1.2%) and lower for MI (1.3% vs 2.2%). The 6-year difference of age between the two studies could explained a part of the death difference, but a significant proportion should correspond to patient selection in RCT, as described by P.G. Steg with the same ratio of 1 to 3 (17). For major bleeding, the 2-year incidence is also much higher, about 4 times higher, 3.1% compared to 0.7% at 2 years of follow-up (1.0 at 3 years). The major bleeding TIMI classification includes intracranial bleeding, fatal bleeding, and hemorrhage associated with a drop in hemoglobin of ≥ 5 g/dL or a ≥ 15%. Since no lab test result was available in the SNDS database, for this last criterion we used the proxy of hospitalization with bleeding main diagnosis and transfusion during hospital stay, considering that such drop of hemoglobin is an indication for a transfusion and few transfusions should be prescribed for a bleeding with a lower hemoglobin decrease. The large difference of bleeding could be partially link to other drugs used in real-life setting. If we excluded patients who took P2Y12 inhibitor antiplatelet agents or anticoagulants 2 months before or after index date, 11.5% of included patients used P2Y12 inhibitor antiplatelet agent before during the year before index date, and then several patients start P2Y12 inhibitor antiplatelet agent or anticoagulant during the follow-up, while patients treatment was not authorized in the THEMIS RCT. Whatever, we had found similar distortion between patients included in the PEGASUS-TIMI 54 trial and a similar real-life population, in which higher all-cause bleeding rates and death were also found in real-life than in the clinical trial (4).
The SNDS is a national healthcare claims database linked to the national hospital discharge summaries database that covers about 99% of the French population. It provides a unique opportunity to identify all CAD-T2DM patients, with exhaustive information about reimbursed treatments out of hospital and use of reimbursed healthcare resources, as well as all hospitalizations. Furthermore, patient records from an existing database are not impacted by the study. The SNDS has already been used to study the type of patients in this study, including for secondary prevention post MI (15, 20), or to describe long-term patient outcomes after MI (5).
The main limitation of this claims and hospitalization database is that it was built for administrative and reimbursement purposes with little clinical data and no biological results, including severity or stage of the disease or some risk factors such as diet, environmental exposure, obesity, alcohol, family history, smoking status, and no information about drug adherence, though the full history of drug dispensing may be consistent with adherence, even though medicines that are prescribed are not always taken. This is common in all studies of drug utilization and effects, even with direct access to the patients. The results of lab tests can often be inferred by subsequent medical prescriptions or tests (e.g. measure of transaminases followed by hospital admission for hepatic injury).
Since all patients identified are extracted from a national database, there is no study selection bias, nor attrition bias; patients are included in the database from birth or immigration to death or emigration, irrespective of employer, income, age or social status. (18) Since deaths are recorded in the database using the national death registry, there is no information bias for this outcome. Causes of death were not available at the time of the study in the SNDS, and the primary outcome of THEMIS RCT cannot be estimated in our study, but we used the THEMIS exploratory outcome of MI, any stroke and death for any cause for comparison between both studies. The ICD-10 coding of hospital discharge summaries is regularly audited by the national health insurance system and there is cross-validation of coding between hospitals. In addition, a number of diagnostic codes and strategies such as for MI, heart failure or stroke have been verified with positive predictive values over 90%, improving over time (27). The same outcomes have been used in previous studies in the same database (4, 21, 23, 26). All information in the database is fully independent from the study and there is no reason that any potential miscoding will be different between drugs or populations. Some T2DM patients could be misclassified as type 1 because of insulin therapy. To prevent this classification bias, a 5-year history period was used to investigate prior treatment sequences with sufficient delay to select appropriate T2DM patients.