Stroke Following Percutaneous Coronary Intervention
Stroke Following Percutaneous Coronary Intervention
The British Cardiovascular Intervention Society (BCIS) collects data on all PCI procedures in the UK. The data collection is coordinated by the National Institute of Cardiovascular Outcomes Research (www.ucl.ac.uk/nicor) via the Central Cardiac Audit Database. In 2011, this dataset collected information on 99% of all PCI procedures performed in National Health Service Hospitals in England and Wales.
The BCIS-NICOR database contains a total of 113 variables, which includes information on clinical variables, procedural parameters, and patient outcomes. Mortality tracking is undertaken by the Medical Research Information Service (MRIS) using patients' NHS number that provides a unique identifier for any person registered with the NHS in England and Wales. Mortality tracking was not possible in this study for patients from Scotland or Northern Ireland.
We analysed all patients who underwent PCI in the UK between 1 January 2007 and 31 December 2012. Patients were classified into three groups: (i) patients with no in-hospital stroke complications, (ii) ischaemic stroke or transient ischaemic attack (TIA) in-hospital complication and (iii) haemorrhagic stroke complication. The main outcomes that were examined were 30-day mortality and in-hospital major adverse cardiovascular events. Major adverse cardiovascular events were defined as the composite of in-hospital mortality, myocardial infarction or re-infarction and revascularization (emergency coronary artery bypass graft or re-intervention PCI).
Additional data were collected on baseline variables including year of procedure, age, gender, smoking status, comorbidities (diabetes, hypertension, hyperlipidaemia, previous myocardial infarction, previous stroke, peripheral vascular disease, chronic renal disease, and previous valve disease), previous PCI, previous coronary artery bypass graft, access site, cardiogenic shock, use of circulatory support, thrombus aspiration, use of ventilatory support, revascularization of the left main stem, indication/diagnosis (stable angina, NSTEMI, STEMI), and medications received (any glycoprotein IIb/IIIa inhibitor, aspirin, clopidogrel, prasugrel, ticagrelor, ticlopidine, and warfarin use).
Statistical analyses were performed using Stata/MP version 13 (Stata Corp., College Station, TX, USA) and StatsDirect version 3 (StatsDirect Ltd, Cheshire, UK).
Summary statistics were expressed as means and standard deviations for continuous variables, and percentages for categorical variables. Comparisons of means were made using t-tests and one-way analyses of variance, with appropriate consideration of multiple testing for pairwise contrasts of the three main groups. Proportions were compared using Fisher's method for four-fold tables and χ (with simulated exact P-values) linear trend tests for ordinal series. We excluded patients with missing values for age, mortality at 30 days, MACE, and stroke complications. A flow diagram graphically describes how the final cohort was derived (Figure 1).
(Enlarge Image)
Figure 1.
Flow chart of participant inclusion.
We present descriptive statistics of baseline variables by complication group in Table 1 . We graphically examined the rates of ischaemic stroke or TIA and haemorrhagic stroke over time (Figure 2). Simple logistic regressions were used to investigate the effect of each baseline variable as a potential predictor of ischaemic stroke or TIA and haemorrhagic stroke. Multiple logistic regressions, in which all baseline variables and year of PCI were included in the two models, were used to identify predictors of: (i) ischaemic stroke or TIA and (ii) haemorrhagic stroke.
(Enlarge Image)
Figure 2.
Rates of stroke/transient ischaemic attack and haemorrhagic stroke complications according to year of PCI. Changes in the rates (per 1000 patients) of ischaemic stroke/TIA complications and haemorrhagic stroke complications over time (2007–12). Error bars represent 95% confidence intervals.
The primary outcomes were 30-day mortality and in-hospital MACE. Multiple logistic regressions were used to investigate the effect of each neurological complication on 30-day mortality and in-hospital MACE rates, using both unadjusted and adjusted for baseline variable models. The final adjusted models were controlled for all collected baseline variables and year of PCI. Chained equations multiple imputations (using mi impute chained in Stata) were used to generate 10 complete datasets with imputed data for missing baseline variables. Additional sensitivity analyses were ran in which we tried to better control for covariates. Under a similar multiple imputation setting, we calculated the propensity scores for membership on each group in the comparisons we investigated, using all available covariates. Using the aggregate patient propensity score across all 10 multiple imputation datasets, we then matched each 'case' (e.g. TIA only) to up to 10 'controls' (e.g. no TIA) with scores within 10.
Cases with missing data that were excluded from the analyses were compared with cases that were included in the final analysis, for baseline variables and outcomes (when available) (see Supplementary material online, Table S1http://eurheartj.oxfordjournals.org/content/suppl/2015/04/19/ehv113.DC1). Results are presented as the main effect with 95% confidence interval (95% CI) unless otherwise stated.
Methods
The British Cardiovascular Intervention Society Database
The British Cardiovascular Intervention Society (BCIS) collects data on all PCI procedures in the UK. The data collection is coordinated by the National Institute of Cardiovascular Outcomes Research (www.ucl.ac.uk/nicor) via the Central Cardiac Audit Database. In 2011, this dataset collected information on 99% of all PCI procedures performed in National Health Service Hospitals in England and Wales.
The BCIS-NICOR database contains a total of 113 variables, which includes information on clinical variables, procedural parameters, and patient outcomes. Mortality tracking is undertaken by the Medical Research Information Service (MRIS) using patients' NHS number that provides a unique identifier for any person registered with the NHS in England and Wales. Mortality tracking was not possible in this study for patients from Scotland or Northern Ireland.
Study Definitions
We analysed all patients who underwent PCI in the UK between 1 January 2007 and 31 December 2012. Patients were classified into three groups: (i) patients with no in-hospital stroke complications, (ii) ischaemic stroke or transient ischaemic attack (TIA) in-hospital complication and (iii) haemorrhagic stroke complication. The main outcomes that were examined were 30-day mortality and in-hospital major adverse cardiovascular events. Major adverse cardiovascular events were defined as the composite of in-hospital mortality, myocardial infarction or re-infarction and revascularization (emergency coronary artery bypass graft or re-intervention PCI).
Additional data were collected on baseline variables including year of procedure, age, gender, smoking status, comorbidities (diabetes, hypertension, hyperlipidaemia, previous myocardial infarction, previous stroke, peripheral vascular disease, chronic renal disease, and previous valve disease), previous PCI, previous coronary artery bypass graft, access site, cardiogenic shock, use of circulatory support, thrombus aspiration, use of ventilatory support, revascularization of the left main stem, indication/diagnosis (stable angina, NSTEMI, STEMI), and medications received (any glycoprotein IIb/IIIa inhibitor, aspirin, clopidogrel, prasugrel, ticagrelor, ticlopidine, and warfarin use).
Statistical Analysis
Statistical analyses were performed using Stata/MP version 13 (Stata Corp., College Station, TX, USA) and StatsDirect version 3 (StatsDirect Ltd, Cheshire, UK).
Summary statistics were expressed as means and standard deviations for continuous variables, and percentages for categorical variables. Comparisons of means were made using t-tests and one-way analyses of variance, with appropriate consideration of multiple testing for pairwise contrasts of the three main groups. Proportions were compared using Fisher's method for four-fold tables and χ (with simulated exact P-values) linear trend tests for ordinal series. We excluded patients with missing values for age, mortality at 30 days, MACE, and stroke complications. A flow diagram graphically describes how the final cohort was derived (Figure 1).
(Enlarge Image)
Figure 1.
Flow chart of participant inclusion.
We present descriptive statistics of baseline variables by complication group in Table 1 . We graphically examined the rates of ischaemic stroke or TIA and haemorrhagic stroke over time (Figure 2). Simple logistic regressions were used to investigate the effect of each baseline variable as a potential predictor of ischaemic stroke or TIA and haemorrhagic stroke. Multiple logistic regressions, in which all baseline variables and year of PCI were included in the two models, were used to identify predictors of: (i) ischaemic stroke or TIA and (ii) haemorrhagic stroke.
(Enlarge Image)
Figure 2.
Rates of stroke/transient ischaemic attack and haemorrhagic stroke complications according to year of PCI. Changes in the rates (per 1000 patients) of ischaemic stroke/TIA complications and haemorrhagic stroke complications over time (2007–12). Error bars represent 95% confidence intervals.
The primary outcomes were 30-day mortality and in-hospital MACE. Multiple logistic regressions were used to investigate the effect of each neurological complication on 30-day mortality and in-hospital MACE rates, using both unadjusted and adjusted for baseline variable models. The final adjusted models were controlled for all collected baseline variables and year of PCI. Chained equations multiple imputations (using mi impute chained in Stata) were used to generate 10 complete datasets with imputed data for missing baseline variables. Additional sensitivity analyses were ran in which we tried to better control for covariates. Under a similar multiple imputation setting, we calculated the propensity scores for membership on each group in the comparisons we investigated, using all available covariates. Using the aggregate patient propensity score across all 10 multiple imputation datasets, we then matched each 'case' (e.g. TIA only) to up to 10 'controls' (e.g. no TIA) with scores within 10.
Cases with missing data that were excluded from the analyses were compared with cases that were included in the final analysis, for baseline variables and outcomes (when available) (see Supplementary material online, Table S1http://eurheartj.oxfordjournals.org/content/suppl/2015/04/19/ehv113.DC1). Results are presented as the main effect with 95% confidence interval (95% CI) unless otherwise stated.