Statins for Primary Prevention of Cardiovascular Events and Mortality in Older Men

Ariela R. Orkaby, MD, MPH; J. Michael Gaziano, MD, MPH; Luc Djousse, MD, ScD; Jane A. Driver, MD, MPH


J Am Geriatr Soc. 2017;65(11):2362-2368. 

In This Article


Study Population

This study used data from Physicians' Health Study (PHS) participants. PHS methods have been described in detail.[22] Briefly, PHS I began in 1982 as a randomized, double-blind, placebo-controlled trial of aspirin and beta-carotene in 22,071 US male physicians aged 40 to 84 years, with no history of myocardial infarction (MI), stroke, transient ischemic attack, or cancer at randomization. PHS II[23] began in 1997 and was a randomized trial of efficacy of beta-carotene, vitamin C, vitamin E, and a multivitamin on CVD and cancer risk in 7,641 PHS I physicians who agreed to participate in the second trial and 7,000 newly recruited male physicians. All PHS subjects have been followed prospectively, using annual mailed health questionnaires to collect self-reported data, including new CVD diagnoses. Each participant signed an informed consent and the institutional review board at Brigham & Women's Hospital approved the study.

This analysis focused on all physicians ≥70 in the PHS cohort who completed annual questionnaires from 1999, the year a specific question regarding statin use was added. Of the 9,988 PHS participants (≥70 years in 1999), 2,670 participants were excluded because of prevalent CVD (MI, stroke, or peripheral vascular disease) and an additional 105 were excluded due to missing information on statin use at baseline.


The primary outcome is a composite of self-reported and subsequently validated major CV events, MI, stroke, and coronary revascularization (PTCA or CABG). Outcomes were assessed annually by questionnaires; ascertainment of events has been described previously.[24] All-cause mortality was the co-primary outcome, confirmed by an endpoints committee after review of medical records, death certificates, and family report. Details on mortality validation in PHS have been published.[25] Secondary outcomes were coronary heart disease (CHD) alone, and stroke alone. All outcomes were updated through 2012.

Other Variables

Data on demographics, including age and race (white or other); anthropometrics, including age and body mass index (BMI); comorbidities, such as congestive heart failure (CHF), hyperlipidemia (HL), hypertension (HTN), diabetes, kidney disease, and dementia; and lifestyle factors, including smoking, alcohol consumption, and activities of daily living; and concurrent medications, such as aspirin and anti-hypertensive use were assessed by annual questionnaires. Alcohol consumption was classified as none, 1–3 drinks per month, 1–6 drinks per week, and ≥7 drinks per week. Smoking was classified as never, past, or current. Diagnosis of diabetes was self-reported and validated in a subsample.[26] HL was defined by self-report, elevated measured or self-reported cholesterol value, or history of anti-lipemic medication use. HTN was defined by self-report, report of blood pressure >140/90 mm Hg, or use of antihypertensive medications. Functional status was assessed with questions regarding difficulty with activities of daily living, such as bathing, grocery shopping, walking several blocks, or climbing stairs.

Statin Use

Beginning in 1999, statin use was assessed by an annual questionnaire with the following question: "Over the past twelve months, on approximately how many DAYS did you take the following [medication]? Statins: (e.g., atorvastatin, cerivastatin, lovastatin, pravastatin, simvastatin, Zocor, etc.)." Respondents could choose varying amounts of use over a 12-month period: 0 (none), 1–13 days, 14–30 days, 31–60 days, 61–90 days, 91–120 days, 121–180 days, 181+ days. Because <4% of participants reported using statins for >0 and <180 days, statin exposure was dichotomized as users for those participants reporting ≥181 days and non-users for <180 days.

Statistical Analyses

Participant characteristics were summarized using descriptive statistics–the t-test for continuous variables and chi-square for binary and categorical variables. Cox proportional hazard models were run to estimate the hazard ratio (HR) and 95% confidence interval (CI) of having a major CV event. Proportional hazards assumptions were tested using product terms of variables and log-transformed person-time, and assumptions were met (all P > .05).

Due to significant differences between statin users and non-users, presumably due to the fact that statins are prescription drugs most often used for hypercholesterolemia and for those at increased risk of CV events due to co-morbidities, such as diabetes, we used propensity score (PS) analysis to attempt to reduce confounding by indication.[27,28] We developed a multivariable logistic regression model to create a PS to assess the probability of being a statin user verses a statin non-user. The majority of variables were missing <5% of responses and <10% of participants were missing >10% of the variables for the PS. We therefore used indicator variables to account for the small amount of missing covariates.

Elevated cholesterol is the most important driver of statin prescription; therefore we excluded those without information on cholesterol. Additionally, questions on functional status were only sent to physicians from the original PHS cohort; therefore only physicians from the original cohort were included in the PS analysis. The final PS model included 27 predictors of statin use, including comorbidities, markers of functional status, quadratic equation for age, and an interaction term for cholesterol and age. The propensity score is typically used in one of four ways: stratifying, matching, inverse probability of treatment weighting, or as a covariate.[28,29] Due to a considerable lack of overlap of the propensity scores amongst statin users and non-users (see Supplementary Appendix S1) we used the matching method.[29] Statin users and non-users were matched 1:1 using the ("greedy") nearest neighbor with caliper matching algorithm.[30] The paired-matches were maintained in all analyses. Person-time of follow up was calculated from 1999–2012 until the primary event occurred (MI, stroke, revascularization, or death).

Using Cox proportional hazards, we estimated the time to event HR and 95% CI of the risk of major CV events in the matched cohort. In a sensitivity analysis, we used age-at-event as the time scale. Subgroup analyses examined age, functional status, and total cholesterol. To ensure matched pairs were not broken, the PS was rerun within each subgroup. To examine function we created a dichotomous variable reflecting functional status as a proxy for frailty. Participants who reported any difficulty with either moderate activity, walking one block, climbing one flight of stairs, grocery shopping, bathing, or bending were considered to be functionally impaired.

Analysis was done using SAS® 9.3 and 9.4 (SAS Institute Inc.).