Associations With Anticoagulation: Stroke Survivors With AF
Associations With Anticoagulation: Stroke Survivors With AF
We conducted a cross-sectional analysis of a city-wide primary care data resource. Conduct and reporting of our analysis is in accordance with the STROBE guidelines for cross-sectional studies.
Greater Glasgow and Clyde Health board provides services for a population of around 1.2 million people in the Glasgow city area. Annual hospital admissions for stroke are around 3000, with 15 312 people registered by primary care practices (all general practitioner (GP) practices) as having previously had a TIA or stroke, for the year 2010. Glasgow is broadly typical of urban, UK settings although with a high level of cardiovascular disease burden and socioeconomic deprivation.
We used the Glasgow LES database, limited to the last available year with full data input (2010). LES is a contractual arrangement with primary care services designed to augment the basic patient-level data collection required through the General Medical Services (GMS) Quality and Outcome Framework (QoF) specification. LES facilitates effective monitoring and clinical audit. In total, 209 out of 213 GP practices in Glasgow participated in the LES initiative. There are currently several active LES covering key disease areas, including coronary heart disease, stroke and heart failure. LES offers financial incentives to encourage proactive case finding, annual nurse-led reviews and centralised data storage. To ensure data quality, the LES initiative provides annual practice nurse training to ensure consistency of data collection and central data analysis and quality control. By linking LES to practice-level prescribing and diagnostic/referral registers, the system offers robust data on medication and comorbidity.
We identified all stroke survivor patients from the LES Stroke Database. We excluded care-home residents or housebound subjects and limited data by ischaemic aetiology and presence of AF using LES-specific read-codes. We limited our search to the most recent year of LES with full dataset available and collated clinical, demographic and prescribing data by predefined variables described below.
Presence of AF is assessed at annual LES review using medical record review and manual pulse check, supplemented if required by 12-lead electrocardiograph. This process of AF case-finding has been shown to be sensitive and has been employed in clinical trials. Under the rubric 'AF' we included persistent and paroxysmal AF and also atrial flutter. Data on anticoagulant treatment were taken from patient-level prescribing data and were defined as at least one prescription for warfarin or other VKA within the 1-year period of interest. Using the same method we also collated data on any antiplatelet agents prescribed. The novel oral anticoagulants (NOACs) were not prescribed in primary care for stroke prevention during the period of the analysis and so were not considered.
We collated data on age; sex; race; systolic blood pressure (using standard sphygmomanometer, mm Hg); glycosylated haemoglobin (HbA1c, %); total cholesterol (mmol/L) and body mass index (kg/m). We described rates of excessive alcohol intake (defined at practice level); smoking (defined as any current use of cigarettes or other related products); any major bleeding episode in last year (defined as requiring hospitalisation); AF duration of greater than 10 years and substantial disability (defined as requiring external assistance with mobility and transfers).
Socioeconomic status was described using the Scottish Index of Multiple Deprivation (SIMD). The SIMD is assigned on the basis of the datazone (using postcode data) of residence and contains various domains, which carry different weighting: income (28%), employment (28%), health (14%), education (14%), geographic access to services (9%), crime (5%) and housing (2%). The data from each domain are combined into an overall index to rank relative multiple deprivation. We used postcode data within LES to assign SIMD and then described data as quintiles, with quintile 1 representing the most-deprived area.
Risk of AF-related stroke was described using CHADS2 (input covariates: cardiac failure; hypertension; age; diabetes; previous stroke) and CHA2DS2-VASC (input covariates as before with additional scoring by age and presence of vascular disease) scores, both scored using conventional criteria. As all included patients had history of ischaemic stroke, minimum possible score was 2 for both tools. We did not have access to all variables that comprise HAS-BLED, and so we used a modified HAS-BLED (mHAS-BLED) (input covariates; hypertension, age, excessive alcohol, previous stroke or bleeding episode (total possible five points)). To complement this analysis, we also described bleeding risk using criteria derived from the National Institute of Health and Care Excellence (NICE) Guideline 36, assigning one point each for age (≥75 years); concomitant antiplatelet use; bleeding history; comorbidity (≥3 other active medical conditions), suboptimal diabetes control (HbA1c≥7.5%) and suboptimal blood pressure control (BP≥160 mm Hg systolic).
The LES data input defaults to 'not present' unless data are entered, thus we do not have specific data on missing variables for categories included in the LES database. Clinical diagnoses recorded in LES are linked to hospital discharge records and primary care registers and so should be robust. Mechanisms for practice level and central data quality control are routinely employed for LES. Prescribing data, postcode (SIMD) and comorbidity were all linked for patient-level practice records. As further internal validity checks, our local stroke Managed Clinical Network (MCN) clinical lead reviewed our collated data to 'sense check' the face validity.
Descriptive statistics were recorded for stroke survivors with AF, assessing the complete cohort and comparing those who were VKA-treated and VKA-untreated. We described mean (SD) or median (IQR) for continuous variables and count (percentage) for categorical variables. Patients were categorised by stroke risk using CHADS2 and CHA2DS2-VASC and by bleeding risk using mHAS-BLED score and NICE criteria.
Unadjusted comparisons of VKA-treated and VKA-untreated groups were conducted using two-sample t test, Mann–Whitney U test, two proportions test or the χ test depending on the distribution and nature of the data. As an internal quality control measure, we recorded 'significance' on univariate at the conventional level (p<0.05) and using sequentially rejective Bonferroni method analyses to correct for multiple analysis. Under this correction, 'significance' was defined as p<0.002 (significant level/number of variables, 0.05/25=0.002). Factors to include in the multivariate analysis were chosen on the basis of clinical and scientific validity as well as (unadjusted) significance. Input covariates were age; socioeconomic deprivation (SIMD); systolic BP; body mass index; smoking; history of bleeding; duration of AF >10 years; depression; obesity; diabetes; disability; heart failure; use of antiplatelet; a combined comorbidity score; and the risk scores of CHADS2, CHA2DS2-VASC, mHASBLED and NICE criteria.
We calculated OR and corresponding 95% CI to express the odds of VKA treatment in univariate analysis. Our multivariate analysis adjusted for clinically important or specific covariates using a binary logistic regression against a dichotomised outcome measure of VKA-treated/VKA-untreated. All analyses were undertaken using SAS V.9.2 (SAS Institute, Inc, Cary, North Carolina, USA).
Methods
We conducted a cross-sectional analysis of a city-wide primary care data resource. Conduct and reporting of our analysis is in accordance with the STROBE guidelines for cross-sectional studies.
Setting
Greater Glasgow and Clyde Health board provides services for a population of around 1.2 million people in the Glasgow city area. Annual hospital admissions for stroke are around 3000, with 15 312 people registered by primary care practices (all general practitioner (GP) practices) as having previously had a TIA or stroke, for the year 2010. Glasgow is broadly typical of urban, UK settings although with a high level of cardiovascular disease burden and socioeconomic deprivation.
Data Source
We used the Glasgow LES database, limited to the last available year with full data input (2010). LES is a contractual arrangement with primary care services designed to augment the basic patient-level data collection required through the General Medical Services (GMS) Quality and Outcome Framework (QoF) specification. LES facilitates effective monitoring and clinical audit. In total, 209 out of 213 GP practices in Glasgow participated in the LES initiative. There are currently several active LES covering key disease areas, including coronary heart disease, stroke and heart failure. LES offers financial incentives to encourage proactive case finding, annual nurse-led reviews and centralised data storage. To ensure data quality, the LES initiative provides annual practice nurse training to ensure consistency of data collection and central data analysis and quality control. By linking LES to practice-level prescribing and diagnostic/referral registers, the system offers robust data on medication and comorbidity.
Participants
We identified all stroke survivor patients from the LES Stroke Database. We excluded care-home residents or housebound subjects and limited data by ischaemic aetiology and presence of AF using LES-specific read-codes. We limited our search to the most recent year of LES with full dataset available and collated clinical, demographic and prescribing data by predefined variables described below.
Variables
Presence of AF is assessed at annual LES review using medical record review and manual pulse check, supplemented if required by 12-lead electrocardiograph. This process of AF case-finding has been shown to be sensitive and has been employed in clinical trials. Under the rubric 'AF' we included persistent and paroxysmal AF and also atrial flutter. Data on anticoagulant treatment were taken from patient-level prescribing data and were defined as at least one prescription for warfarin or other VKA within the 1-year period of interest. Using the same method we also collated data on any antiplatelet agents prescribed. The novel oral anticoagulants (NOACs) were not prescribed in primary care for stroke prevention during the period of the analysis and so were not considered.
We collated data on age; sex; race; systolic blood pressure (using standard sphygmomanometer, mm Hg); glycosylated haemoglobin (HbA1c, %); total cholesterol (mmol/L) and body mass index (kg/m). We described rates of excessive alcohol intake (defined at practice level); smoking (defined as any current use of cigarettes or other related products); any major bleeding episode in last year (defined as requiring hospitalisation); AF duration of greater than 10 years and substantial disability (defined as requiring external assistance with mobility and transfers).
Socioeconomic status was described using the Scottish Index of Multiple Deprivation (SIMD). The SIMD is assigned on the basis of the datazone (using postcode data) of residence and contains various domains, which carry different weighting: income (28%), employment (28%), health (14%), education (14%), geographic access to services (9%), crime (5%) and housing (2%). The data from each domain are combined into an overall index to rank relative multiple deprivation. We used postcode data within LES to assign SIMD and then described data as quintiles, with quintile 1 representing the most-deprived area.
Risk of AF-related stroke was described using CHADS2 (input covariates: cardiac failure; hypertension; age; diabetes; previous stroke) and CHA2DS2-VASC (input covariates as before with additional scoring by age and presence of vascular disease) scores, both scored using conventional criteria. As all included patients had history of ischaemic stroke, minimum possible score was 2 for both tools. We did not have access to all variables that comprise HAS-BLED, and so we used a modified HAS-BLED (mHAS-BLED) (input covariates; hypertension, age, excessive alcohol, previous stroke or bleeding episode (total possible five points)). To complement this analysis, we also described bleeding risk using criteria derived from the National Institute of Health and Care Excellence (NICE) Guideline 36, assigning one point each for age (≥75 years); concomitant antiplatelet use; bleeding history; comorbidity (≥3 other active medical conditions), suboptimal diabetes control (HbA1c≥7.5%) and suboptimal blood pressure control (BP≥160 mm Hg systolic).
The LES data input defaults to 'not present' unless data are entered, thus we do not have specific data on missing variables for categories included in the LES database. Clinical diagnoses recorded in LES are linked to hospital discharge records and primary care registers and so should be robust. Mechanisms for practice level and central data quality control are routinely employed for LES. Prescribing data, postcode (SIMD) and comorbidity were all linked for patient-level practice records. As further internal validity checks, our local stroke Managed Clinical Network (MCN) clinical lead reviewed our collated data to 'sense check' the face validity.
Statistical Methods
Descriptive statistics were recorded for stroke survivors with AF, assessing the complete cohort and comparing those who were VKA-treated and VKA-untreated. We described mean (SD) or median (IQR) for continuous variables and count (percentage) for categorical variables. Patients were categorised by stroke risk using CHADS2 and CHA2DS2-VASC and by bleeding risk using mHAS-BLED score and NICE criteria.
Unadjusted comparisons of VKA-treated and VKA-untreated groups were conducted using two-sample t test, Mann–Whitney U test, two proportions test or the χ test depending on the distribution and nature of the data. As an internal quality control measure, we recorded 'significance' on univariate at the conventional level (p<0.05) and using sequentially rejective Bonferroni method analyses to correct for multiple analysis. Under this correction, 'significance' was defined as p<0.002 (significant level/number of variables, 0.05/25=0.002). Factors to include in the multivariate analysis were chosen on the basis of clinical and scientific validity as well as (unadjusted) significance. Input covariates were age; socioeconomic deprivation (SIMD); systolic BP; body mass index; smoking; history of bleeding; duration of AF >10 years; depression; obesity; diabetes; disability; heart failure; use of antiplatelet; a combined comorbidity score; and the risk scores of CHADS2, CHA2DS2-VASC, mHASBLED and NICE criteria.
We calculated OR and corresponding 95% CI to express the odds of VKA treatment in univariate analysis. Our multivariate analysis adjusted for clinically important or specific covariates using a binary logistic regression against a dichotomised outcome measure of VKA-treated/VKA-untreated. All analyses were undertaken using SAS V.9.2 (SAS Institute, Inc, Cary, North Carolina, USA).
