Design
This is a population-based retrospective observational cohort study.
Population
The study population were all individuals older than 40 years with a new prescription of cilostazol or pentoxifylline between 2009 and 2011, from 274 PHC teams from the Catalan Health Institute (Institut Català de la Salut, ICS), which is the main health provider in Catalonia, with a reference population of 5,835,000 patients (80% of the Catalan population).
We excluded patients with less than two visits to the PHC centre during the year before the inclusion and patients with only one dispensing of the drugs of interest during the study period.
All patients were followed-up from the cohort entry date up to 31st December 2013, death or lost to follow-up.
Data source
The main data source is SIDIAP (Information System for Research in Primary Care), [17] which contains anonymized clinical information of all PHC centres of ICS. This information emerges from ECAP™, electronic health records in PHC of the ICS, and it includes socio-demographic characteristics, health conditions registered as ICD10 codes, clinical parameters, toxic habits, laboratory data, and General Practitioners’ prescriptions with their corresponding pharmacy invoice data. SIDIAP may be linked with CMBD-HA (“minimum set of data at hospital discharge”), [18] which contains diagnoses coded with ICD9 at hospital discharge from all hospitals in Catalonia, to obtain the data for comorbidities and for the endpoints of the study.
Variables collected at baseline
The following variables were collected from SIDIAP database: socio-demographic characteristics, smoking status, body mass index (BMI, kg/m2), laboratory data (total cholesterol and LDL-cholesterol determinations, creatinine and estimated glomerular filtration calculated by MDRD), blood pressure (BP) determinations, ankle-brachial pressure index (ABPI) measures, diagnosis of PAD, and other comorbidities of interest (hypertension, type 2 DM, dyslipidemia). The following variables were extracted from SIDIAP and CMBD-HA: previous history of haemorrhages (total and specific gastrointestinal and cerebral haemorrhages), stroke, CAD, and arrhythmias.
Exposure to drugs of interest (cilostazol and pentoxifylline) and to comedications (diuretics, β-blockers, calcium channel antagonists, angiotensin converter enzyme inhibitors (ACEI), angiotensin receptor blockers (ARB), nitrates and other vasodilators, lipid-modifying agents, antidiabetic drugs and insulins, proton pump inhibitors, non-steroidal anti-inflammatory drugs (NSAID), oral anticoagulants (OAC), and antiplatelets) were obtained from the pharmacy invoice registry, which contains all information on pharmaceutical products dispensed by community pharmacies with ICS prescriptions, by ATC codes. All diagnosis codes as ICD9 and/or ICD10 and ATC drug codes may be found at (Additional file 1: Table S1).
Outcomes of interest
We collected the following events from CMBD-HA during the study period: haemorrhages (total and specific gastrointestinal and cerebral haemorrhages), stroke, CAD, and arrhythmias.
Sample, matching process, and statistical power
The two cohorts were matched in order to balance socio-demographic characteristics, comorbidity and comedications. The method used was the “Nearest Neighbour”, which is based on Propensity Score (PS) Link logit with “MatchIt” library from R (v3.0.1).
The variables used to build this PS were: sex, age, BMI, smoking status, comorbidities, BP control, estimated glomerular filtration as per MDRD, ABPI and co-medications at baseline.
The final matched sample included 5810 individuals, 2905 per group. After the matching process, a 36.4% of the sample (n = 3319) was removed and the potential bias between the two samples (overall vs matched) was reduced in an 83%. Assuming that a cohort of 5810 patients with a 5-years follow-up period had an incidence of a cardiovascular event of 4.3% in one of the groups (incidence data of symptomatic patients from ARTPER study [3]), and between 1% (HR = 1.23) to 2% (HR = 1.46) of events attributable to cilostazol, the statistical power would be 53–96%. This approximation has been carried out with a Log-Rank test with an alpha-level of 5% in a bilateral contrast.
Statistical analysis
Descriptive statistics were used to summarize overall information. In order to compare the baseline characteristics between the treatment groups, Chi-square test was used for categorical variables and Student’s t-test for quantitative variables.
Conditional Cox regression models were used to estimate incidence rates and hazard ratios (HR) and the person/time value was used as offset. Risk functions and HR were estimated with their 95% confidence intervals (CI) to compare the two groups. 95% CI and p-values were calculated by robust standard errors (by clusters). Goodness of fit and proportional hazards assumption of Cox models were assessed through Schoenfeld residuals method.
These analyses were conducted in the population of 5810 patients and in three subgroups of patients: > 65 years-old, patients diagnosed with type 2 DM and those co-treated with antiplatelet agents.
All statistical tests were two-sided at the 5% significance level. The analyses were performed using SPSS-IBM PC v.18 and Stata v.11 (Stata Corp., Collage Station, TX).