Cohort description

The CARDIANA cohort is a population-based cohort from Navarra, an autonomous community located in a northern region of Spain with approximately 650 000 inhabitants and with a public health coverage (including both public and mixed coverage) over 99%.8 It was designed by a multidisciplinary team involving methodologists, primary care specialists, endocrinologists, healthcare policy makers and clinical and social science researchers, many of them with expertise in the design of strategies for the management of patients with diabetes. The creation of this cohort was used as a ‘case study’ in the development of BARDENA, the Results Analysis Database of Navarra that is being constructed under the adoption of the Observational Medical Outcomes Partnership (OMOP) Common Data Model, which aims at harmonising electronic medical records to facilitate participation on international distributed research.

The cohort includes all users of the Public Health Service of Navarra who, as of 1 January 2012, had active codes of T1D or T2D (T89 and T90 of the International Classification of Primary Care, version 2, ICPC-2, respectively) in the Primary Care Electronic Medical Record System of Navarra (ATENEA) records. Patients with descriptions of diabetes different from T1D or T2D were excluded, as well as when severe inconsistencies in the dates of diagnosis, birth or death were found. Patients were also excluded if no registry of contact with the public health system was found either before the inclusion date and/or in the follow-up period. No other exclusion criteria were applied, and patients of all ages and conditions were considered, including patients with T2D with onset during childhood and patients with T1D with late onset during adulthood. Causes of early termination of the patient data extraction were death or change of community/country.

Figure 1 shows the flow chart of the creation of the cohorts. The classification of patients into T1D or T2D took into account that the validity of the T2D diagnosis had been assessed in a previous study,9 but not that of the T1D diagnosis. Hence, the ICPC-2 codes and the descriptive field that goes with the code were first used, and after that, the classification procedure was complemented with the regional registry of T1D diabetes, which was legally approved by formal order 37/2014, on 16 April10 that includes all patients with T1D with an onset date after 1989. More precisely, we first included all patients with ICPC-2 code T89 and T90 with active Individual Health Card in ATENEA. Second, we excluded patients with incompatible or incongruent information, such as to have died before 2012, to have unrealistic birth or onset dates, or to have diagnostic literals ‘gestational diabetes’ or ‘pancreoprivic diabetes’. Third, using the regional registry of T1D, we maintained in the T2D cohort only patients who were not in the T1D registry, and passed all patients with T2D code to the T1D cohort if they were included in the T1D registry. Fourth, we excluded from the T1D cohort all patients that had onset date >1990 but were not in the T1D registry, and also all patients with onset date <1990 that were not treated with insulin. Finally, we excluded patients that had no contact with the health system before 2012, and patients who had no contact with the health system from 2012 to 2016. Combining the information from the health electronic records and administrative population datasets, two cohorts of prevalent patients with diabetes were created: the T1D CARDIANA cohort, with 1067 patients and the T2D CARDIANA cohort, with 33 842 patients. During the follow-up of the first wave, 33 (3.1%) patients with T1D and 455 (1.3%) patients with T2D were lost to follow-up because of having moved to another region, and information for these patients was censored at this date accordingly.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Flow chart for creation of the type 1 and type 2 prevalent diabetes (T1D, T2D) cohorts.

The actual follow-up period of the cohort is 5 years, from 1 January 2012 to 31 December 2016. The next data extraction process that will update longitudinal data and principal cardiovascular events is being conducted in late 2022 and early 2023, covering the period from 1 January 2017 to 31 December 2021, and further extractions are planned in 5 year waves.

Variables, databases and integration process

Sociodemographic and clinical variables of the defined CARDIANA T1D and CARDIANA T2D cohorts came from eight clinical and administrative databases: ATENEA, LAKORA-TIS, LAMIA, HCI, HIS-LEIRE (including the Minimum Basic Data Set at hospital discharge—MBDS), the population registry, the mortality registry and the T1D registry. For future updates of the cohort, the Results Analysis Database of Navarra (BARDENA) will be used. A brief description of the original databases, which were extensively described elsewere,11 is given in online supplemental table S1, whereas a summary of the variables considered is given in online supplemental table S2. In all, the information collected consists of all relevant structured data available from these sources generated during each contact of the patient with the health system. One set of variables was collected once and was considered fixed during the follow-up. These include the date of entry and/or exit from the health system, demographic and socioeconomic data such as the study level, lifestyle information such as tobacco use or physical activity level, the basic health zone the patients belongs to, coinsurance status,12 baseline comorbidities and a history of cardiovascular history using the ICPC-2 codification system, among others. Some new variables were created from these previous variables, such as the Charlson weighted score13 or the GMA14 15 comorbidity score. The other set of variables was collected longitudinally using a time-dependent structure and included all analytical results that occurred during follow-up as well as pharmacologic treatments, health service use and fatal and nonfatal cardiovascular events. For these time-dependent variables, the date on which they occurred was also included. Cardiovascular events were considered to occur during the follow-up when CVD diagnostic codes were recorded in the mortality or the MBDS dataset, as defined in the recent SCORE2 study16 (see the list of codes considered for fatal and non-fatal CVD in online supplemental table S3).

The integration procedure was conducted by the Statistic Institute of Navarra (NASTAT) and the Directorate-General for Informatics, Telecommunications and Innovation of the Health Department of Navarra, who supervised the data extraction and guaranteed fulfilment of the law in terms of personal data protection. Afterward, the anonymised databases were provided to the research team.

Patient and public involvement

None.

Findings to date

The T1D and T2D CARDIANA cohorts consisted of 1067 and 33 842 patients, respectively. Their sociodemographic characteristics are given in table 1. No adjustment has been included due to the descriptive nature of the objective, but information on both cohorts is presented in parallel. Patients in both cohorts were primarily men (57.4% and 55.7% in T1D and T2D cohorts, respectively), and only 5% were immigrants. Compared with patients in the T2D cohort, patients in the T1D cohort were much younger (mean age 36.9 years in T1D vs 69.4 years in T2D), had a higher probability of being part of the workforce (84.5% vs 26.6%), had a higher-income level (38.3% vs 27.5% had over €18 000 per year) and also higher educational attainment (17.8% vs 4.7% had university studies).

Health-related patients’ status at baseline, including lifestyle data, laboratory tests values, and office measured parameters and comorbidities are given in table 2. The mean duration of diabetes was 3 years higher in patients with T1D than in patients with T2D (11.0 vs 8.1 years), but their comorbidity indices were lower, with a weighted Charlson score equal to 1.2 vs 2.1 and a weighted Adjusted Morbidity Groups (AMG, a measure of morbidity used in our health service that clasify general population according to type and complexity of the diseases they have) equal to 6.0 vs 11.4, respectively. Similarly, patients with T1D have much lower prevalence of CVD history (4.9% vs 23.8%), were more frequently active (71.5% vs 55.9%) and alcohol abstinent (69.0% vs 66.5%) but had higher probability of being smokers (32.2% vs 17.7%).

Regarding clinical and laboratory test parameters, patients with T1D showed much better control of their body mass index (25.7 vs 30.6 kg/m²) but worse control of their glycosylated hemoglobin (HbA1c) levels (8.2% vs 7.1%). They also had better control of the other metabolic parameters considered, namely, high-density lipoprotein (62.0 vs 48.9 mg/dL), low-density lipoprotein (105.7 vs 111.4 mg/dL), triglycerides (84.2 vs 142.7 mg/dL) and albumin to creatinine ratio levels (13.4 vs 35.9).

The use of primary care services during the last year before baseline (table 3) was lower in patients with T1D than in patients with T2D for the total number of visits (17.1 vs 21.1 visits/year) and for all types of visits, except for emergency visits (0.6 vs 0.5 visits/year) and remote visits to nursing (4.1 vs 1.1 visits/year). Active prescriptions of antihypertensive medication were 41 points less frequent in patients with T1D (15.2% vs 56.2%), and their lipid-lowering treatment use was less than half (18.8% vs 49.3%). Similar differences were observed in antithrombotic treatment prescription (15.6% vs 38.2%), but baseline glucose-lowering treatment prescription was higher in patients with T1D (83.6% vs 67.2%).

Only nine patients (0.8%; 95% CI (0.4% to 1.6%)) in the T1D cohort developed a CVD event, five of which were fatal (0.5% of the total cohort). In the same follow-up period, 22 patients died from non-cardiovascular-related events. For the T2D cohort, 2602 (7.7%; 95% CI (7.4% to 8.0%)) had an event and 1268 of them were fatal (3.7% of the total cohort). During this follow-up, 5072 patients died from non-cardiovascular-related events.

For the T2D cohort, the occurrence of CVD events along the follow-up has been associated with physical activity, with estimated ORs after matching and adjusting for confounders equal to 0.84 (95% CI 0.66 to 1.07) for the partially active group and 0.71 (95% CI 0.56 to 0.91) for the active group, compared with patients in the non-active group.17 Note that, in this study, a slightly different CVD outcome was considered.

Strengths and limitations

The main strength of this study is that the CARDIANA cohorts integrate exhaustive clinical, socioeconomic and behavioural information from all available administrative and clinical data sources, providing a complete framework to assess the course of the disease in patients with diabetes and the factors that affect it. Especially in relation to socioeconomic variables, these cohorts have individual information on country of origin, working status, educational level and income, which are not frequently available, with most studies using area-level proxies. Another strength is that data have been subjected to quality control procedures before-and-after database integration.

The main limitation of this study is the possible presence of bias resulting from the use of existing electronic clinical records, which may affect different methodological aspects. First, patients with T2D without a diabetes code in the ATENEA records were not included. Although the validity of the code has been satisfactory asssessed,9 undiagnosed patients have not been included, which in Spain it has been estimated that could account for 4%–6% of the overall prevalence of 11%–14%.18 19 Second, data completeness can be low for some variables dependent on physicians’ idiosyncratic reporting procedures, such as tobacco use or physical activity and some variables that have been considered fixed may have changed along the follow-up. Although an effort has been made to complement variables with others that had a text format, information bias may be present, so imputation methods and sensitivity analyses will be required. Third, electronic prescriptions were only fully implemented in 2014, and at baseline it is estimated that 8%–10% of the total prescriptions have not been accounted for. Fourth, patients without any contact with the regional public health system because of using exclusively private health institutions were not included. Nevertheless, it is estimated that these patients account for less than 1% in the region.8