Study design and overview
In our retrospective cohort study, cases of T2DM patients (n = 152,678) starting antidiabetic therapy between 1 January, 2010 and 31 December, 2013 were retrospectively extracted from the database of the National Health Insurance Fund as an anonymized, aggregated patient data, and were considered as T2DM patients if having antidiabetic treatment (ATC A10) but not matching previously detailed and published criteria for type 1, insulin dependent diabetes and not having polycystic ovarium syndrome (ICD 10 E282) [
11].
Our main goal was to investigate the risk of cardiovascular morbidity and all-cause mortality since the diagnosis of diabetes. For this reason, we needed to identify the date of diagnosis of diabetes, this way only incident population met our criteria. Onset of diabetes was defined as the first occurrences of diabetes ICD code (E10–E14) in the database for a patient or the first antidiabetic treatment, whichever occurred first. Combined records of ICD E10–14 and ATC A10 treatment (with the exclusion of PCOS diagnosis) together with specified algorithm for excluding type 1 diabetes ensures the inclusion of all T2DM patients in Hungary who are involved into national diabetes care.
The control cohort (n = 305,356) was randomly selected from the total population recorded in NHIF database by matching the age at diabetes onset, gender and zip code of residence with the T2DM cohort resulting in two controls for each DM patient, matching 1:2. In the case of controls no diabetes related ICD codes or antidiabetic treatment were recorded.
The data source included information on mortality for any causes, cardiovascular complications of diabetes like incidence of myocardial infarction (ICD-10 I21–24) and ischemic and hemorrhagic stroke (ICD-10 I61–63, G4630, G4640, G4580, G4590) in in- and out-patient records. The International Classification of Diseases (ICD) codes of 9th and 10th Revisions, were used to define acute myocardial infarction, stroke diagnoses from 1 January, 2010 onward. Dates of death were also retrieved from the National Health Insurance Fund database, however this database does not include the cause of death, therefore we used only the cases of death from any type of causes (all-cause mortality) in our analysis.
All-cause mortality and cardiovascular morbidity as myocardial infarction and stroke were assessed in a period of 58 months, from 1 January, 2010 till 31 October, 2014, by comparing T2DM with the matched control population. In both T2DM and control patients, subgroups were defined according to decades of age resulting in subgroups of > 70, 61–70, 51–60, 41–50, 31–40 and < 31. Patients were also evaluated according to gender.
Limitation of the NHIF data investigation is that allows only detection of use of reimbursed drugs. Statin therapy during the investigated period was reimbursed, but blood pressure lowering and antiplatelet therapies are not completely reimbursed in Hungary, consequently these information do not exist in the NHIF database.
This study was approved by the Regional Research Ethics Committee of the Medical Center, University of Pécs, Medical School, Hungary (Study License Number: 6962/2017) and run without commercial sponsorship. The study protocol was also reviewed and confirmed by the National Health Insurance Fund (NHIF) (Identification Number: S04/161/2016).
Statistical analysis
Survival analyses were performed by Cox regression and age and gender matched groups were used. Our model runs separately for the sexes and the individual age groups, consequently the proportionality criterion for cox does not deteriorate if the same parentage is not the same for different gender or age groups). The final result was the adequately weighted average of the results in the different groups. This approach ensured correction for gender and age group. As the risk of death was higher in the first 4 months after the diagnosis of diabetes, and the risk of stroke and myocardial infarction was also higher in the 1st month, time dependent factor was used to separate this short-term effect from the long-term effect. Simultaneous confidence intervals including ratio of hazard ratios were calculated using contrast matrix.
Mean of age and follow-up time were compared using Welch’s two-sample test. Proportion of statin treatment, prior stroke and myocardial infarction were compared using Chi square test.
All analyses were performed by using of R Software, version 3.4.2 (2017-09-28), applying survival, survminer and multcomp packages [
12].