Relative Effects of Statin Therapy on Stroke and Cardiovascular Events in and Secondary Analysis of the Stroke Prevention by Aggressive Reduction in Cholesterol Levels (SPARCL) Study Larry B. Goldstein, MD; Pierre Amarenco, MD; Marian LaMonte, MD; Steven Gilbert, PhD; Michael Messig, PhD; Alfred Callahan, MD; Michael Hennerici, MD, PhD; Henrik Sillesen, MD, MSc; K. Michael A. Welch, MB, ChB; on behalf of the SPARCL Investigators Background and Purpose In SPARCL, treatment with atorvastatin 80 mg daily reduced stroke risk in patients with recent stroke or TIA and no known coronary heart disease by 16% versus placebo over 4.9 years of follow-up. The purpose of this secondary analysis was to determine whether men and women similarly benefited from randomization to statin treatment. Methods The effect of sex on treatment-related reductions in stroke and other cardiovascular outcomes were analyzed with Cox regression modeling testing for sex by treatment interactions. Results (n 1908) constituted 40% of the SPARCL study population. At baseline, men (n 2823) were younger (62.0 0.21versus 63.9 0.27 years), had lower systolic BPs (138.1 0.35 versus 139.5 0.47 mm Hg), higher diastolic BPs (82.2 0.20 versus 81.0 0.25 mm Hg), more frequently had a history of smoking (73% versus 38%), and had lower total cholesterol (207.0 0.54 versus 218.9 0.67 mg/dl) and LDL-C levels (132 0.45 versus 134 0.57 mg/dl) than women. Use of antithrombotics and antihypertensives were similar. After prespecified adjustment for region, entry event, time since event, and age, there were no sex by treatment interactions for the combined risk of nonfatal and fatal stroke (treatment Hazard Ratio, HR 0.84, 95% CI 0.68, 1.02 in men versus HR 0.84, 95% CI 0.63, 1.11 in women; treatment sex interaction P 0.99), major cardiac events (HR 0.61, 95% CI 0.42, 0.87 in men versus HR 0.76, 95% CI 0.48, 1.21 in women; P 0.45), major cardiovascular events (HR 0.78, 95% CI 0.65, 0.93 in men versus HR 0.84, 95% CI 0.65, 1.07 in women; P 0.63), revascularization procedures (HR 0.50, 95% CI 0.37, 0.67 in men versus HR 0.76, 95% CI 0.46, 1.24 in women; P 0.17), or any CHD event (HR 0.54, 95% CI 0.41, 0.72 in men versus 0.67 95% CI 0.46, 0.98 in women; P 0.40). Conclusion Stroke and other cardiovascular events are similarly reduced with atorvastatin 80 mg/d in men and women with recent stroke or TIA. (Stroke. 2008;39:2444-2448.) Key Words: stroke TIA prevention lipids statins sex Meta-analysis of 14 trials of statins including more than 90 000 subjects, most of whom had coronary heart disease (CHD) or major CHD risk factors, found men and women had similar treatment-associated reductions in both major coronary events and major vascular events. 1 Consistent with these results, the Prospective Pravastatin Pooling Project and the Heart Protection Study found statin treatment was associated with similar reductions in the occurrence of predominately first strokes regardless of sex. 2,3 A secondary analysis of subjects with CHD enrolled in 3 clinical trials of oral glycoprotein IIb/IIIa inhibitors found no significant difference in statin-associated stroke reductions between men and women. 4 Whether statin treatment is equally effective for secondary stroke prevention in men and women has not been addressed. The Stroke Prevention with Aggressive Reductions in Cholesterol Levels (SPARCL) trial showed that treating patients with recent stroke or TIA and no known coronary heart disease with a statin (atorvastatin 80 mg per day) reduced the combined risk of fatal and nonfatal stroke as well as other cardiovascular events. 5 The purpose of this post hoc analysis of data from the SPARCL trial was to determine whether men Received December 31, 2007; accepted January 28, 2008. From Duke University and VA Medical Centers (L.B.G.), Durham, NC; Denis Diderot University (P.A.), Paris; the University of Maryland (M.L.), Baltimore, Md; Rho Inc (S.G.), Newton, Mass; Pfizer, (M.M..) New York; Neurologic Consultants (A.C.), Nashville, Tenn; the University of Heidelberg (M.H.), Mannheim, Germany; the University of Copenhagen (H.S.), Denmark; and Rosalind Franklin University of Medicine and Science (K.M.A.W.), North Chicago, Ill. Correspondence to Larry B. Goldstein, Box 3651, Duke University Medical Center, Durham, NC 27710. E-mail golds004@mc.duke.edu 2008 American Heart Association, Inc. Stroke is available at http://stroke.ahajournals.org DOI: 10.1161/STROKEAHA.107.513747 2444
Goldstein et al Gender and Statin Effects in SPARCL 2445 and women similarly benefited from randomization to statin treatment. Methods The methods of the SPARCL study have been described in detail previously. 5,6 The local research ethics committee or institutional review board at each participating study center approved the study protocol, and all patients gave written informed consent. The primary hypothesis of the SPARCL trial was that treatment with 80 mg of atorvastatin per day would reduce the combined risk of fatal and nonfatal stroke among patients with a history of stroke or TIA. Eligible patients were men and women over 18 years of age who had had an ischemic or hemorrhagic stroke or a TIA (diagnosed by a neurologist within 30 days after the event) 1 to 6 months before randomization. Stroke was defined by focal clinical signs of central nervous system dysfunction of vascular origin that lasted for at least 24 hours; TIA was defined by the loss of cerebral or ocular function for less than 24 hours. Patients had to be ambulatory, with a modified Rankin score of no more than 3 (scores can range from 0 to 5, with higher scores indicating more severe disability), and to have an LDL cholesterol level of at least 100 mg per deciliter (2.6 mmol per liter) and no more than 190 mg per deciliter (4.9 mmol per liter; 15 of 205 centers excluded otherwise suitable patients with an LDL cholesterol level above 160 mg per deciliter [4.1 mmol per liter], as required by their institutional review boards). Excluded patients otherwise included those with atrial fibrillation, mechanical prosthetic heart valves, severe mitral valve stenosis, or subarachnoid hemorrhage. Patients with hemorrhagic stroke (2% of the study population) could be included if they were deemed by the investigator to be at risk for ischemic stroke or coronary heart disease. Subjects were enrolled between September 1998 and March 2001. The primary outcome for the SPARCL trial was the time from randomization to a first nonfatal or fatal stroke. There were 7 prespecified secondary composite outcomes: stroke or TIA; major coronary event (cardiac death, nonfatal myocardial infarction [MI] or resuscitated cardiac arrest); major cardiovascular event (stroke plus any major coronary event); any CHD event (major coronary event, plus coronary revascularization procedure, unstable angina or angina/ischemia requiring emergent hospitalization); revascularization procedure (coronary, carotid or peripheral); and any cardiovascular event (any of the former plus clinically significant peripheral vascular disease). An independent end point committee adjudicated all potential end points without knowledge of the patients treatment status or cholesterol levels. For this post hoc analyses, we evaluated the relative effects of intention to treat with atorvastatin 80 mg per day in men and women on the primary and each secondary SPARCL outcome in separate Cox regression models with prespecified adjustment for region (7 dummy variables), entry event, time since event, and age (each requiring 1 degree of freedom). Adjustment was made in the regression models for potential confounding by inclusion of covariates. A second set of analyses to control for additional possible confounders was performed to validate the primary analysis by using a logistic regression model to reduce the dimensionality of the data (a logistic regression predicting sex was fitted to baseline covariates including age, days since qualifying event, type of qualifying event, geographic region baseline lipid levels, systolic and diastolic blood pressure, body mass index, smoking and hypertension history). 7 Because of sample size limitations, all potential covariates could not be directly added to the Cox regression models. 8 Because exploratory analyses revealed possible nonlinear relationships, the model was also allowed to fit the squared and square root transformed versions of the listed continuous variables based on a forward selection procedure. The resulting model had a c-statistic of 0.821. The linear predictor from this model was then saved and grouped by quintile. The data were then stratified by sex and quintile of the linear predictor from the previous logistic regression. Patients who could not be included in the logistic regression model because of missing data were assigned to their own strata. This resulted in a total Table 1. Baseline Characteristics by Sex (n 2823) (n 1908) Age, y 62.0 0.21 63.9 0.27 Systolic blood pressure, mm Hg 138.1 0.35 139.5 0.47 Diastolic blood pressure, mm Hg 82.2 0.20 81.0 0.25 Body-mass index, kg/m 2 27.3 0.07 27.7 0.12 Entry Event Stroke 1997 (70.7) 1271 (66.6) Hemorrhagic 60 (2.1) 33 (1.7) Ischemic: small vessel 841 (29.8) 568 (29.8) Ischemic: large vessel 479 (17.0) 270 (14.2) Unknown 617 (21.9) 400 (21.0) TIA 824 (29.2) 636 (33.3) Unknown/missing data 2 (0.07) 1 (0.05) Time since entry event, days 85.1 0.89 86.7 1.08 Risk factors Smoking status Current smoker 640 (22.7) 268 (14.1) Former smoker 1426 (50.5) 455 (23.9) History of hypertension 1662 (58.9) 1266 (66.4) History of diabetes mellitus 485 (17.2) 309 (16.2) Any prior statin therapy 71 (2.5) 49 (2.6) Concomitant therapy Aspirin or other anti-platelet drug 2465 (87.3) 1665 (87.3) ACE Inhibitor 812 (28.8) 538 (28.2) Calcium channel antagonist 414 (14.7) 295 (15.5) Beta blocker 422 (15.0) 414 (21.7) Angiotensin II antagonist 120 (4.3) 92 (4.8) Vitamin K antagonist, including warfarin 183 (6.5) 110 (5.8) Baseline Lipids, mean SE LDL-C, mg/dl 132.4 0.44 134.4 0.57 HDL-C, mg/dl 46.1 0.21 55.8 0.34 Total Cholesterol, mg/dl 207.0 0.54 218.9 0.67 Triglyceride, mg/dl 143.6 1.72 143.9 1.52 Apolipoprotein A1, mg/dl 140.8 0.44 162.0 0.67 Apolipoprotein B, mg/dl 133.7 0.41 133.4 0.51 *Continuous variables are summarized as means SE. **Categorical variables are summarized as count (percentage). of 12 strata. The data for all patients were then analyzed using a stratified Cox regression model with covariate terms for age, days since qualifying event, qualifying event, and geographic region. The SPARCL steering committee developed the study protocol with the sponsor and takes responsibility for the data and data analyses. Medpace (Cincinnati) managed all data. Medpace, Charles River Laboratories Clinical Services (Brussels), and the sponsor provided site monitoring throughout the study. A data and safety monitoring board with independent statistical support performed interim monitoring analyses for safety and efficacy. Results Table 1 compares baseline characteristics between men (n 2823) and women (n 1908). were younger, had lower systolic BP, higher diastolic BP, less frequently had a history of hypertension, and more frequently had a history of smoking.
2446 Stroke September 2008 Table 2. Baseline, End of Treatment, and Percent Change From Baseline Lipid Levels by Sex (n 922) also had lower total cholesterol, LDL-C, HDL-C, and Apolipoprotein A1 levels than women. Use of antithrombotics and antihypertensives were similar. Table 2 gives the mean percent change from baseline to the last study visit for each lipid measure for atorvastatin and placebo treated men and women. There were no sex-based treatment differences in lipid profiles between men and women (LDL-cholesterol, P 0.986; total cholesterol, P 0.452; HDL-cholesterol, P 0.126; triglycerides, P 0.8437; Apo A1, P 0.135; Apo B, P 0.955). Table 3 gives adjusted hazard ratios, Wald probability values, and treatment by sex interaction probability values for each of the SPARCL trial primary and secondary outcomes. The separate results for men and women were generated by removing the sex term from the model (since each model includes either all men or all women). and women had similar treatment-associated benefits for each study end point (ie, there were no significant treatment by sex interactions for (n 943) (n 1397) (n 1379) LDL cholesterol, mg/dl Baseline, mean SE 134.1 0.80 134.6 0.82 131.8 0.64 133.0 0.63 End of treatment, mean SE 84.6 1.19 125.7 1.05 77.9 0.88 118.8 0.84 mean % change from baseline 35.3 4.4 39.8 9.1 Total cholesterol, mg/dl Baseline, mean SE 219.1 0.96 218.7 0.95 206.3 0.77 207.7 0.75 End of treatment, mean SE 166.5 1.40 210.6 1.16 149.8 1.05 194.1 0.97 mean % change from baseline 23.2 2.6 26.7 5.7 HDL cholesterol, mg/dl Baseline, mean SE 55.7 0.49 56.0 0.49 46.3 0.30 45.8 0.30 End of treatment, mean SE 57.4 0.51 56.9 0.50 49.0 0.36 47.2 0.34 mean % change from baseline 4.6 3.2 7.0 4.2 P value, mean % change 0.1428 0.0001 Triglycerides, mg/dl Baseline, mean SE 147.0 2.29 141.1 2.06 141.9 2.85 145.2 1.97 End of treatment, mean SE 123.2 2.15 141.7 2.47 116.3 1.88 141.2 2.25 mean % change from baseline 10.5 5.6 11.2 3.2 Apo A1, mg/dl Baseline, mean SE 161.7 0.96 162.5 0.98 141.0 0.62 140.8 0.63 End of treatment, mean SE 164.6 0.98 167.0 0.97 146.3 0.71 146.5 0.70 mean % change from baseline 2.8 4.1 4.8 5.0 P value, mean % change 0.0345 0.8244 Apo B, mg/dl Baseline, mean SE 133.6 0.73 133.1 0.73 132.8 0.59 134.7 0.59 End of treatment, mean SE 90.3 0.98 121.9 0.87 85.9 0.76 118.3 0.73 mean % change from baseline 31.4 7.1 34.5 11.0 *Baseline values summarize only those patients who also had at least one postbaseline measurement. **P values are based on an analysis of covariance with terms for baseline value, treatment, sex and treatment by sex interaction. the combined risk of nonfatal and fatal stroke, nonfatal stroke, fatal stroke, stroke or TIA, major cardiac events, major cardiovascular events, revascularization procedures, or any CHD event). The treatment by sex interaction term for all cause mortality is marginally significant (P 0.06). The second set of analyses using Cox regression models with stratification to control for additional potential confounders provided almost identical results (data not shown). Table 4 gives safety data by treatment and sex. 80 mg per day was similarly tolerated in men and women. Discussion This analysis of data from the SPARCL trial found similar statin-related reductions in stroke and other cardiovascular events in men and women who had a recent stroke or TIA and no known coronary heart disease at the time of study enrollment. To reduce the chances of spurious associations in
Goldstein et al Gender and Statin Effects in SPARCL 2447 Table 3. Stroke and Cardiovascular Outcomes and All-Cause Mortality by Sex Event Sex Subgroup n/n (%) (%) n/n (%) Adjusted Hazard Ratio (Wald 95% CI) Wald P Value Treatment X Sex Interaction P Value Any stroke All subjects 265/2365 (11.2) 311/2366 (13.1) 0.84 (0.71,0.99) 0.03 0.99 176/1427 (12.3) 204/1396 (14.6) 0.84 (0.68,1.02) 0.08 89/938 (9.5) 107/970 (11.0) 0.84 (0.63,1.11) 0.21 Fatal stroke All subjects 24/2365 (1.0) 41/2366 (1.7) 0.57 (0.34,0.95) 0.03 0.23 18/1427 (1.3) 24/1396 (1.7) 0.71 (0.38,1.31) 0.27 6/938 (0.6) 17/970 (1.8) 0.37 (0.14,0.93) 0.03 Nonfatal stroke All subjects 247/2365 (10.4) 280/2366 (11.8) 0.87 (0.73,1.03) 0.11 0.77 163/1427 (11.4) 186/1396 (13.3) 0.85 (0.69,1.05) 0.13 84/938 (9.0) 94/970 (9.7) 0.90 (0.67,1.21) 0.47 Stroke or TIA All subjects 375/2365 (15.9) 476/2366 (20.1) 0.77 (0.67,0.88) 0.001 0.64 232/1427 (16.3) 298/1396 (21.3) 0.75 (0.63,0.89) 0.001 143/938 (15.2) 178/970 (18.4) 0.81 (0.65,1.00) 0.05 MCE All subjects 81/2365 (3.4) 120/2366 (5.1) 0.65 (0.49,0.87) 0.003 0.45 49/1427 (3.4) 78/1396 (5.6) 0.61 (0.42,0.87) 0.006 32/938 (3.4) 42/970 (4.3) 0.76 (0.48,1.21) 0.25 MCVE All subjects 334/2365 (14.1) 407/2366 (17.2) 0.80 (0.69,0.92) 0.002 0.63 216/1427 (15.1) 266/1396 (19.1) 0.78 (0.65,0.93) 0.006 118/938 (12.6) 141/970 (14.5) 0.84 (0.65,1.07) 0.15 Revascularization All subjects 94/2365 (4.0) 163/2366 (6.9) 0.55 (0.43,0.72) 0.001 0.17 66/1427 (4.6) 126/1396 (9.0) 0.50 (0.37,0.67) 0.001 28/938 (3.0) 37/970 (3.8) 0.76 (0.46,1.24) 0.27 Any CHD Event All subjects 123/2365 (5.2) 204/2366 (8.6) 0.58 (0.46,0.73) 0.001 0.40 78/1427 (5.5) 137/1396 (9.8) 0.54 (0.41,0.72) 0.001 45/938 (4.8) 67/970 (6.9) 0.67 (0.46,0.98) 0.04 All-cause mortality All subjects 216/2365 (9.1) 211/2366 (8.9) 1.00 (0.82,1.21) 0.98 0.06 146/1427 (10.2) 124/1396 (8.9) 1.14 (0.90,1.45) 0.28 70/938 (7.5) 87/970 (9.0) 0.78 (0.57,1.07) 0.12 Results reflect adjustment for region, entry event, time since entry event, and age. MCE indicates major cardiac event; MCVE, major cardiovascular event; CHD, coronary heart disease (see methods for description). this post hoc analysis, we tested treatment by sex interactions for the SPARCL primary end point (first occurrence of a nonfatal or fatal stroke) as well as for all of the study s prespecified secondary endpoints. This analysis is a combination of a randomized trial and an observational study (ie, randomized treatment effects and nonrandom subgroups defined by sex). In Table 3, the adjusted Table 4. SPARCL Safety by Sex hazard ratios by sex were obtained by running separate analyses on men and women controlling for age, qualifying event, days since qualifying event, and geographic region. Because treatment was randomized, these provide unbiased estimates of treatment effects for men and women. Other baseline variables such as age, baseline lipids, and blood pressure, however, differed between men and women (Table 1). To assure that any (n 938) (n 970) (n 1427) (n 1396) ALT/AST 3 ULN*, n (%) 13 (1.39%) 1 (0.10%) 8 (0.56%) 2 (0.14%) CPK 10 ULN*, n (%) 1 (0.11%) 0 1 (0.07%) 0 Musculoskeletal AEs Myalgia, n (%) 63 (6.7%) 72 (7.4%) 66 (4.6%) 69 (4.9%) Myopathy, n (%) 3 (0.3%) 2 (0.2%) 4 (0.3%) 5 (2.0%) Rhabomyolysis, n (%) 1 (0.1%) 2 (0.2%) 1 (0.1%) 1 (0.1%) *Persistent elevations: 2 consecutive elevations within 4 to 10 days. ALT indicates Alanine aminotransferase; ULN, upper limits of normal; CPK, creatine phosphokinase.
2448 Stroke September 2008 possible sex differences were not explained by differences in the baseline covariates, a second set of more complex analyses were performed with results not substantially different than those given in Table 3. It should be recognized, however, that the SPARCL trial was not designed to detect differences in event rates between men and women, and that statistical power of this analysis may be limited. As shown in Table 3, the treatment by sex interaction term for all cause mortality is marginally significant. Although a relative treatment-related reduction in all cause mortality in women as compared to men cannot be excluded, the individual treatment effects are not significant (HR 1.14, 95% CI 0.90 to 1.45, P 0.28 for men and HR 0.78, 95% CI 0.57 to 1.07, P 0.12 for women). The Heart Protection Study (HPS) compared the effects of simvastatin and placebo in a broad spectrum of patients that included 3280 subjects with prior cerebrovascular disease. 3 As in SPARCL, 5 the HPS found a reduction in major vascular events in patients with prior cerebrovascular disease treated with a statin, but in contrast to SPARCL, there was no reduction in recurrent stroke (and no analysis for a treatmentrelated difference in recurrent stroke based on sex). Some potential reasons for the difference between the trial results have been discussed previously and include later enrollment after the index event in HPS, smaller treatment-related reductions in lipid levels, and lack of power because of a small number of outcome strokes. 5 In HPS, there was not a specific analysis to determine whether the effects of the statin on the occurrence of major vascular events differed between men and women with prior stroke. 3 Several studies provide data assessing potential sex-related differences in the effects of statins for primary stroke prevention. The HPS found similar reductions in the frequencies of a first stroke with statin treatment in men and women. 3 At least 2 other randomized placebo-controlled statin trials have provided subgroup analyses of stroke outcomes stratified by sex. 4 The Cholesterol and Recurrent Events (CARE) study found a nonsignificant 35% reduction of stroke with statin treatment in women but no difference between men and women. 9 The Long-Term Intervention With Pravastatin in Ischemic Disease (LIPID) study found that more women in the statin group (4.4%) than in the placebo group (3.6%) had strokes with a beneficial effect in men (3.6% with statin treatment versus 4.7% with placebo). 10 When the data from these 2 trials was pooled with data from the West of Scotland Coronary Prevention Study (WOSCOPS), there was no statistical interaction between sex and statin treatment on the occurrence of stroke events. 2 Therefore, there is no evidence for a sex-related difference in the effects of statins in primary stroke prevention in the setting of known coronary heart disease or in other primary prevention populations. There are several stroke risk factors that are unique to women, and new areas of research offer the possibility of developing preventive therapies aimed at reducing a woman s chances of having a first or recurrent stroke. 11 As reflected in the baseline data from the SPARCL trial, even traditional stroke risk factor profiles for men and women can differ. Despite these differences, this secondary analysis of SPARCL trial data suggests that statin treatment similarly reduces the likelihood of stroke and other cardiovascular events in men and women with recent TIA or stroke. Sources of Funding The SPARCL trial was funded by Pfizer. Employees of Pfizer contributed to the design and conduct of the SPARCL trial, the collection, management, analysis, and interpretation of the data, and reviewed the manuscript. Disclosures Larry Goldstein has received honoraria from Pfizer during the course of this study. The honoraria did not exceed $10 000/yr. Pierre Amarenco has received grants from Pfizer for other research or activities not reported in this research exceeding $10 000/yr and honoraria from Pfizer in excess of $10 000/yr during the course of this study. Marian LaMonte declares no conflicts aside from participation as a site investigator in the SPARCL trial. Steven Gilbert is employed by Rho Inc, a company that provides statistical consultation for Pfizer. Michael Messig is an employee of Pfizer and owns Pfizer stock. Alfred Callahan has received research support and honoraria from Pfizer in excess of $10 000/yr during the course of this study. Michael Hennerici has received grants from Pfizer for other research or activities not reported in this research/article and honoraria from Pfizer during the course of the study. Neither the grants nor the honoraria exceeded $10 000/yr. Henrik Sillesen has received grants from Pfizer for other research or activities not reported in this research/article in excess of $10 000/yr and honoraria exceeding $10 000/yr during the course of this study. K. Michael Welch has received honoraria from Pfizer during the course of the study in excess of $10 000/yr and research support less than $10 000/yr. References 1. Cholesterol Treatment Trialists (CTT) Collaborators. Efficacy and safety of cholesterol-lowering treatment: prospective meta-analysis of data from 90 056 participants in 14 randomised trials of statins. Lancet. 2005;366: 1267 1278. 2. Byington RP, Davis BR, Plehn JF, White HD, Baker J, Cobbe SM, Shepherd J. Reduction of stroke events with pravastatin: the Prospective Pravastatin Pooling (PPP) Project. Circulation. 2001;103:387 392. 3. Heart Protection Study Collaborative Group. Effects of cholesterollowering with simvastatin on stroke and other major vascular events in 20 536 people with cerebrovascular disease or other high-risk conditions. Lancet. 2004;363:757 767. 4. Bushnell CD, Griffin J, Newby LK, Goldstein LB, Mahaffey KW, Graffagnino CA, Harrington RA, White HD, Simes RJ, Califf RM, Topol EJ, Easton JD. Statin use and sex-specific stroke outcomes in patients with vascular disease. Stroke. 2006;37:1427 1431. 5. The Stroke Prevention by Aggressive Reduction in Cholesterol Levels (SPARCL) Investigators. High-dose atorvastatin after stroke or transient ischemic attack. N Engl J Med. 2006;355:549 559. 6. The SPARCL Investigators. Design and baseline characteristics of the Stroke Prevention by Aggressive Reduction in Cholesterol Levels (SPARCL) Study. Cerebrovascular Diseases. 2003;16:389 395. 7. Rosenbaum PR, Rubin DB. Reducing bias in observational studies using subclassification on the propensity score. J Am Stat Assoc. 1984;79:516 524. 8. Harrell F, Lee K, Mark D. Multivariate prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat in Med. 1996;15:361 387. 9. Plehn JF, Davis BR, Sacks FM, Rouleau JL, Pfeffer MA, Bernstein V, Cuddy TE, Moye LA, Piller LB, Rutherford J, Simpson LM, Braunwald E. Reduction of stroke incidence after myocardial infarction with pravastatin. The Cholesterol and Recurrent Events (CARE) study. Circulation. 1999;99:216 223. 10. LIPID Study Group. Prevention of cardiovascular events and death with pravastatin in patients with coronary heart disease and a broad range of initial cholesterol levels. The Long-term Intervention with Pavastatin in Ischaemic Disease (LIPID) study group. N Engl J Med. 1998;339:1349 1357. 11. Bushnell CD, Hurn P, Colton C, Miller VM, del Zoppo G, Elkind MSV, Stern B, Herrington D, Ford-Lynch G, P. G., James A, Brown CM, Choi E, Bray P, Newby LK, Goldstein LB, Simpkins J. Advancing the study of stroke in women. Summary and recommendations for future research from an NINDS-sponsored multidisciplinary working group. Stroke. 2006;37:2387 2399.