Stroke and CVD Prevention With Lipid Lowering Drugs
Stroke and CVD Prevention With Lipid Lowering Drugs
The Three-City study is a prospective study aiming to assess the association between vascular diseases and risk of dementia. The detailed study protocol has been previously described. The Three-City cohort is composed of non-institutionalised people aged 65 years and over, randomly selected from electoral rolls of three cities in France (Bordeaux (south west), Dijon (north east), and Montpellier (south east), who agreed to participate in the study and signed an informed consent form.
Between March 1999 and March 2001, 9294 people were enrolled. A total of 1439 participants were not eligible for the study described here, as they reported a history of coronary heart disease (n=1017), stroke (n=330), or both vascular events (n=92) at baseline. Participants treated with lipid lowering drugs other than statins or fibrates (for example, bile acid sequestrants) (n=113) were also excluded. Among the 7742 remaining participants, 258 (3.3%) were lost to follow-up, leaving a study sample of 7484 participants.
Face-to-face examinations took place every two years during follow-up. Trained nurses and psychologists conducted interviews and made physical and cognitive measurements at the participant's home and at the study centre. Data collection included sociodemographic characteristics (education, occupation, and income), lifestyle (smoking, drinking, and food frequency questionnaire), assessment of disability (Instrumental Activities of Daily Living scale), global cognitive functioning (Mini Mental State examination) and depression (Centre for Epidemiologic Studies-Depression scale), and recording of height and weight. Past history of cardiovascular disease included a history of coronary heart disease, stroke, arrhythmia, and peripheral artery disease. Blood pressure was measured twice after five minutes' rest in a seated position with an electronic device (OMRON M6; OMRON Healthcare, Kyoto, Japan). Hypertension was defined as a systolic blood pressure 140 mm Hg or above, a diastolic blood pressure 90 mm Hg or above, or the use of antihypertensive treatment. At baseline, blood was collected after overnight fasting. Lipid concentrations (total cholesterol, high density lipoprotein cholesterol, low density lipoprotein cholesterol, and triglycerides) and glycaemia were measured centrally. Diabetes was defined as use of antidiabetic drugs or a fasting blood glucose of 7 mmol/L or above.
At each follow-up examination, at the participants' home, interviewers collected information on all drugs used during the preceding month. Participants were asked to show their prescriptions and drug packages. Drug names were coded according to the Anatomic Therapeutic Chemical classification of the World Health Organization. Lipid lowering drugs (code C10A) included hydroxymethyl glutaryl coenzyme A reductase inhibitors (C10AA), fibrates (C10AB), and other drugs such as bile acid sequestrants or nicotinic acid derivatives (not considered in this study). Use of blood pressure lowering or antithrombotic drugs was also registered.
At each follow-up visit, participants or informants for deceased participants were systematically questioned about the occurrence of any severe medical event or hospital admission since the last contact. For those reporting a possible coronary heart disease or stroke event, all available clinical information was collected from hospital records, and interviews were conducted with the participant's physician, nursing home staff (for participants admitted to a nursing home during follow-up), or family. Expert panels reviewed all available clinical information and classified each event according to ICD-10 (international classification of diseases, 10th revision). Cardiac events included hospital admission for definite angina, definite myocardial infarction, definite cardiovascular death, coronary balloon dilatation, or coronary artery bypass. Brain imaging data were available for more than 80% of validated stroke cases (computerised tomography 82%; magnetic resonance imaging 15%) and Doppler ultrasound for 62%. When no brain imaging was available, the diagnosis was based on signs and symptoms. Stroke was confirmed if the participant had a new focal neurological deficit of sudden onset attributable to a cerebrovascular event that persisted for more than 24 hours. The panel classified stroke as ischaemic stroke, intracerebral haemorrhage, or of unspecified type.
We described the characteristics of the cohort according to lipid lowering drug use at baseline. We compared users and non-users of lipid lowering drugs by using analysis of variance and χ tests adjusted for age, sex, and centre. We also compared users of statins and users of fibrates, and we assessed the associations between vascular events and classic vascular risk factors. We estimated the risk of vascular events related to lipid lowering drug use by using a Cox proportional hazards model with age as the timescale, and we calculated hazard ratios and their 95% confidence intervals with participants not taking any lipid lowering drug as the reference. Modelling included testing of the proportional hazard assumption. We estimated hazard ratios for any major fatal or non-fatal vascular event (coronary heart disease or stroke) and for each type of event separately, using successively use of any lipid lowering drug, statins, and fibrates as independent variables. If a participant had multiple cardiovascular events during follow-up, we considered only the date of the first in estimating the overall risk of vascular event. We first used a simple model adjusted for sex and study centre, with age used as the timescale (model 1). The multivariable model (model 2) was further adjusted for potential confounding factors: diabetes (yes, no), body mass index (<25, 25–29, ≥30), smoking (never, past, current), drinking alcohol (never, past, current), hypertension (yes, no), arrhythmia (yes, no), antithrombotic drugs (yes, no), triglyceride concentration (thirds), and low density lipoprotein to high density lipoprotein cholesterol ratio (thirds).
We did sensitivity analyses stratified by age (< 75, ≥75), sex, triglyceride concentration (<1.6 g/L, ≥1.6 g/L), body mass index (<27, ≥27), hypertension (yes, no), and systolic blood pressure (<145 mm Hg, ≥145 mm Hg) for coronary heart disease and stroke separately. We built a high dimensional propensity score for lipid lowering drug use with a logistic regression model including all the adjustment variables of the primary multivariable model plus other available variables (education, income, systolic blood pressure, diastolic blood pressure, use of antihypertensive drug, self reported diabetes, glycaemia ≥7 mm/L or use of antidiabetic drug, incapacities (Instrumental Activities of Daily Living and Rosow-Breslau functional health scales), cognitive functioning (Mini Mental State score), Center for Epidemiologic Studies-Depression score, APOE genotype). We used this score in different ways. We included it as an adjustment variable in the Cox model. We also calculated hazard ratios within strata defined by thirds of the propensity score distribution. Lastly, we did a matched propensity score analysis. Finally, we estimated hazard ratios in participants who did not report any change in their use of lipid lowering drug (that is, remained a user or non-user) during the first seven years of follow-up.
We did additional analyses to explore complex selection biases. We calculated hazard ratios for death in lipid lowering drug users compared with non-users (all causes of death and deaths from any vascular disease (ICD-10 codes I00-I99)). We estimated hazard ratios for stroke in participants with no previous cardiac event during follow-up, and vice versa. We also estimated the total risk of vascular event (first ever or recurrent) related to lipid lowering drugs during follow-up in the whole Three-City cohort (that is, without excluding participants reporting a history of vascular events at baseline).
We used SAS 9.1 for statistical analyses. We considered a two tailed P value below 0.05 to be statistically significant.
Methods
Study Population
The Three-City study is a prospective study aiming to assess the association between vascular diseases and risk of dementia. The detailed study protocol has been previously described. The Three-City cohort is composed of non-institutionalised people aged 65 years and over, randomly selected from electoral rolls of three cities in France (Bordeaux (south west), Dijon (north east), and Montpellier (south east), who agreed to participate in the study and signed an informed consent form.
Between March 1999 and March 2001, 9294 people were enrolled. A total of 1439 participants were not eligible for the study described here, as they reported a history of coronary heart disease (n=1017), stroke (n=330), or both vascular events (n=92) at baseline. Participants treated with lipid lowering drugs other than statins or fibrates (for example, bile acid sequestrants) (n=113) were also excluded. Among the 7742 remaining participants, 258 (3.3%) were lost to follow-up, leaving a study sample of 7484 participants.
Face-to-face examinations took place every two years during follow-up. Trained nurses and psychologists conducted interviews and made physical and cognitive measurements at the participant's home and at the study centre. Data collection included sociodemographic characteristics (education, occupation, and income), lifestyle (smoking, drinking, and food frequency questionnaire), assessment of disability (Instrumental Activities of Daily Living scale), global cognitive functioning (Mini Mental State examination) and depression (Centre for Epidemiologic Studies-Depression scale), and recording of height and weight. Past history of cardiovascular disease included a history of coronary heart disease, stroke, arrhythmia, and peripheral artery disease. Blood pressure was measured twice after five minutes' rest in a seated position with an electronic device (OMRON M6; OMRON Healthcare, Kyoto, Japan). Hypertension was defined as a systolic blood pressure 140 mm Hg or above, a diastolic blood pressure 90 mm Hg or above, or the use of antihypertensive treatment. At baseline, blood was collected after overnight fasting. Lipid concentrations (total cholesterol, high density lipoprotein cholesterol, low density lipoprotein cholesterol, and triglycerides) and glycaemia were measured centrally. Diabetes was defined as use of antidiabetic drugs or a fasting blood glucose of 7 mmol/L or above.
Use of Lipid Lowering and Other Drugs
At each follow-up examination, at the participants' home, interviewers collected information on all drugs used during the preceding month. Participants were asked to show their prescriptions and drug packages. Drug names were coded according to the Anatomic Therapeutic Chemical classification of the World Health Organization. Lipid lowering drugs (code C10A) included hydroxymethyl glutaryl coenzyme A reductase inhibitors (C10AA), fibrates (C10AB), and other drugs such as bile acid sequestrants or nicotinic acid derivatives (not considered in this study). Use of blood pressure lowering or antithrombotic drugs was also registered.
Ascertainment of Vascular Events During Follow-up
At each follow-up visit, participants or informants for deceased participants were systematically questioned about the occurrence of any severe medical event or hospital admission since the last contact. For those reporting a possible coronary heart disease or stroke event, all available clinical information was collected from hospital records, and interviews were conducted with the participant's physician, nursing home staff (for participants admitted to a nursing home during follow-up), or family. Expert panels reviewed all available clinical information and classified each event according to ICD-10 (international classification of diseases, 10th revision). Cardiac events included hospital admission for definite angina, definite myocardial infarction, definite cardiovascular death, coronary balloon dilatation, or coronary artery bypass. Brain imaging data were available for more than 80% of validated stroke cases (computerised tomography 82%; magnetic resonance imaging 15%) and Doppler ultrasound for 62%. When no brain imaging was available, the diagnosis was based on signs and symptoms. Stroke was confirmed if the participant had a new focal neurological deficit of sudden onset attributable to a cerebrovascular event that persisted for more than 24 hours. The panel classified stroke as ischaemic stroke, intracerebral haemorrhage, or of unspecified type.
Statistical Analysis
We described the characteristics of the cohort according to lipid lowering drug use at baseline. We compared users and non-users of lipid lowering drugs by using analysis of variance and χ tests adjusted for age, sex, and centre. We also compared users of statins and users of fibrates, and we assessed the associations between vascular events and classic vascular risk factors. We estimated the risk of vascular events related to lipid lowering drug use by using a Cox proportional hazards model with age as the timescale, and we calculated hazard ratios and their 95% confidence intervals with participants not taking any lipid lowering drug as the reference. Modelling included testing of the proportional hazard assumption. We estimated hazard ratios for any major fatal or non-fatal vascular event (coronary heart disease or stroke) and for each type of event separately, using successively use of any lipid lowering drug, statins, and fibrates as independent variables. If a participant had multiple cardiovascular events during follow-up, we considered only the date of the first in estimating the overall risk of vascular event. We first used a simple model adjusted for sex and study centre, with age used as the timescale (model 1). The multivariable model (model 2) was further adjusted for potential confounding factors: diabetes (yes, no), body mass index (<25, 25–29, ≥30), smoking (never, past, current), drinking alcohol (never, past, current), hypertension (yes, no), arrhythmia (yes, no), antithrombotic drugs (yes, no), triglyceride concentration (thirds), and low density lipoprotein to high density lipoprotein cholesterol ratio (thirds).
We did sensitivity analyses stratified by age (< 75, ≥75), sex, triglyceride concentration (<1.6 g/L, ≥1.6 g/L), body mass index (<27, ≥27), hypertension (yes, no), and systolic blood pressure (<145 mm Hg, ≥145 mm Hg) for coronary heart disease and stroke separately. We built a high dimensional propensity score for lipid lowering drug use with a logistic regression model including all the adjustment variables of the primary multivariable model plus other available variables (education, income, systolic blood pressure, diastolic blood pressure, use of antihypertensive drug, self reported diabetes, glycaemia ≥7 mm/L or use of antidiabetic drug, incapacities (Instrumental Activities of Daily Living and Rosow-Breslau functional health scales), cognitive functioning (Mini Mental State score), Center for Epidemiologic Studies-Depression score, APOE genotype). We used this score in different ways. We included it as an adjustment variable in the Cox model. We also calculated hazard ratios within strata defined by thirds of the propensity score distribution. Lastly, we did a matched propensity score analysis. Finally, we estimated hazard ratios in participants who did not report any change in their use of lipid lowering drug (that is, remained a user or non-user) during the first seven years of follow-up.
We did additional analyses to explore complex selection biases. We calculated hazard ratios for death in lipid lowering drug users compared with non-users (all causes of death and deaths from any vascular disease (ICD-10 codes I00-I99)). We estimated hazard ratios for stroke in participants with no previous cardiac event during follow-up, and vice versa. We also estimated the total risk of vascular event (first ever or recurrent) related to lipid lowering drugs during follow-up in the whole Three-City cohort (that is, without excluding participants reporting a history of vascular events at baseline).
We used SAS 9.1 for statistical analyses. We considered a two tailed P value below 0.05 to be statistically significant.