This study is part of the EXAcerbations of COPD and their OutcomS on CardioVascular disease (EXACOS-CV) programme. The overall study design and analytical approaches were detailed previously [6].

This retrospective cohort study used data from the PHARMO Data Network from the Netherlands [7]. The source population included people who were registered as patients with participating GP’s (in total 20% of the Dutch population) and whose GP records could be linked to hospital and out-patient pharmacy records, irrespective of whether they were hospitalized or received medication between 1 January 2014 and 31 December 2018 (45% of the GP population). Individuals aged ≥ 40 years, with an incident, GP-reported diagnosis of COPD (International Classification of Primary Care (ICPC) R95 or R91.01) in this period were selected. The entire available lookback period was used to confirm the incident nature of COPD diagnosis. The date of COPD diagnosis was cohort entry date (CED). A medical diagnosis code for COPD diagnosis was required to be confirmed by at least one of the following additional variables: a record of spirometry measurement, a hospitalisation with a discharge diagnosis for COPD or exacerbation of COPD (International Classification of Disease (ICD)-10 J44) or GP-reported Global Initiative For Chronic Obstructive Lung Disease (GOLD) classification recorded in 3 years prior, or 3 years following the CED. Patients were included in the study population if they had at least 12 months of data available before CED. Patients with alpha-1 antitrypsin deficiency (ICD-10 E88.0) were excluded.

A moderate exacerbation event was defined as an outpatient visit to the GP for COPD in combination with a dispensation of an oral corticosteroid within the 5 days following the visit and for a duration of ≥ 5 days and ≤ 15 days. The date of the GP visit was used as the first date of exposure (Day 1) to a moderate exacerbation.

A severe exacerbation was defined as a hospitalisation with a primary discharge diagnosis code of COPD or exacerbation of COPD (ICD-10 J44.0 or J44.1), or a secondary discharge diagnosis code for exacerbation of COPD (ICD-10 J44.1). The start date of exposure to a severe exacerbation was the date of hospital admission.

Exposed time was set as the 365-day period starting on day 1 of an exacerbation event. Exposed time was divided into sub-periods of exposure, in which the hazard of the outcome could be assumed constant over time [8,9,10]. The following exposed time periods were used: 1–7 days, 8–14 days, 15–30 days, 31–180 days and 181–365 days following an exacerbation (Fig. 1). Upon the occurrence of a subsequent exacerbation, the exposed time periods started again at Day 1. As such, a patient could contribute data on a certain exposed time period more than once. Patients were considered unexposed during the time period prior to the first exacerbation and the time period post 365 days following an exacerbation. Patients who did not experience an exacerbation during follow-up were unexposed throughout the entire follow-up (Fig. 1).

Cohort design of the present study. ECOPD: Exacerbation of chronic obstructive pulmonary disease; CV: cardiovascular; ID: patient identifier

From the CED, each patient was followed until the earliest of (1) a first outcome of interest occurring after CED or (2) the end of data availability, or (3) end of study period on December 31st, 2019.

Outcomes of interest included all-cause death and four categories of non-fatal severe CV events, defined as events that required a hospital admission with at least one overnight stay, with a primary or secondary discharge code (ICD-10; all coded provided on Additional file 1: Table S1) for (1) acute coronary syndrome (acute myocardial infarction and unstable angina, ACS), (2) heart failure (HF) decompensation, (3) cerebral ischaemia (including transient ischaemic attack (TIA)), or (4) arrhythmias (atrial fibrillation (AF) or other arrhythmias). To ascertain HF decompensation, only primary discharge codes were used reflecting the acuteness and decompensation of a chronic disease. For AF, events among patients with a history of AF were not considered since hospitalisations may occur in a non-urgent context for the management of AF. The mid-point of the hospitalisation period was considered as the date of the CV event, since the CV event may have occurred during the admission and not on the admission date. All-cause death was derived from death information from records in the PHARMO Data Network, supplemented with all-cause death data from the Dutch death registry. Several endpoints were defined: (a) time to the first outcome of interest (composite, i.e., a non-fatal severe event or all-cause death); (b) time to the first non-fatal severe CV event (to examine contribution of all-cause death to the composite risk); (c) time to the first event of each non-fatal CV event and all-cause death.

Baseline and time-dependent covariates were pre-specified as potential confounders. Baseline covariates that were defined over 12-month lookback period preceding CED were age, sex, socio-economic status (low/middle/high/unknown), and obesity (ICPC T82, ICD-10 E66, or body mass index > 30 kg/m2). Baseline covariates that were defined based on the entire available history (hospital ICD-9 or ICD-10 diagnosis or GP-reported ICPC diagnoses) included diabetes mellitus type 2, disorders of lipoprotein metabolism and other lipidaemias, ischaemic heart diseases, hypertensive diseases, heart failure, pulmonary oedema, pulmonary hypertension, venous thromboembolism, cerebrovascular disease, arrhythmia, asthma, chronic kidney disease or renal failure, mental illness and/or anxiety disorder. Time-dependent covariates were updated annually from CED onward unless otherwise stated and included COPD-related variables (number of GP visits in the last 12 months, number of prior moderate or severe exacerbations, and comedication use in the last 12 months (at least one prescription for long-acting inhaled bronchodilators, inhaled corticosteroids and their combination, short-acting inhalers, roflumilast and/or slow-release theophylline, any cardiac medication, any metabolic medication (ATC code listed in Additional file 1: Table S2). Smoking wasn’t included due to a substantial amount (> 50% of patients) of missing data.

Baseline characteristics of the study population were described overall, and separately in patients who have, or have not exacerbated during study follow-up. Crude incidence rates (IR) of each outcome (composite, non-fatal CV events only, and individual categories) per 100 person-years were obtained, together with the corresponding 95% exact Poisson confidence intervals (CI).

Time-dependent Cox proportional hazards models were used with binary indicators of each exposed time period as time-dependent covariates, and the unexposed period as the reference. Separate models were fitted for (1) the association between exposure time periods following a moderate or severe exacerbation and each endpoint of interest, (2) for the separate association with moderate and severe exacerbations, and (3) for the association with a first, second and third exacerbation. All models were fitted with and without adjustment for all pre-specified fixed and time-varying confounders. For the analyses that examined the first, second and third exacerbation, exacerbations after the third were not evaluated. Analyses for the time to first severe CV event of each specific category, or all-cause death, were adjusted for the other “competing” CV events of interest, as time-varying confounders. All analyses of time to a first severe CV event were censored when patients died without experiencing the outcome of interest.

All data were analysed within Statistical Analysis Software (SAS) Enterprise Guide version 8.2 (SAS Institute Inc., Cary, NC, USA) and Windows using SAS version 9.4. Missing data were not imputed.