Erectile dysfunction (ED) affects approximately 18 million men age 20 years or older in the U.S. (1). Projections from U.S. prevalence data indicate that by 2025, over 300 million men worldwide will have ED (2). The relationship between ED and cardiovascular disease (CVD) has received substantial attention. The prevailing notion is that ED may serve as a sentinel marker for CVD (3- 13). This is based largely on shared pathophysiological mechanisms (e.g., endothelial dysfunction, arterial occlusion, systemic inflammation) (3,6,9,14- 19) and risk factors (6,20- 25), the high coprevalence of both conditions (8,10,26- 28), and the reasonable premise that progressive occlusive disease should manifest earlier in the microvasculature than in larger vessels (9,29). Prospective studies have shown that ED predicts the development of CVD (30- 33) and CVD mortality (34). Of particular interest is the observation that the risk of CVD associated with ED is in the range of risk associated with traditional CVD risk factors (31- 32,34), such as current smoking, hypertension, or a family history of myocardial infarction. However, it is not known whether ED improves the prediction of CVD beyond traditional risk factors. We sought to test the hypothesis that ED improves CVD risk prediction. Confirmation of this hypothesis would have major clinical and public health implications in light of the observation that sudden death may be the first manifestation of CVD (35- 37).

The MMAS (Massachusetts Male Aging Study) is a prospective, observational cohort study of aging, health, and endocrine and sexual function in a population-based random sample of men between the ages 40 and 70 years (38). A total of 1,709 respondents (52% of 3,258 eligible) completed the baseline (1987 to 1989) protocol. The MMAS subjects were observed again from 1995 to 1997 (n = 1,156, 77% response rate) and 2002 to 2004 (n = 853, 65% response rate). These response rates were expected given the requirements for early morning phlebotomy and extensive in-person interviews. Subjects received no financial incentive at baseline, and $50 and $75 remunerations at the first and second follow-ups, respectively.

Extensive details on MMAS have been published elsewhere (38). The core field protocol for MMAS remained the same over time. A trained field technician/phlebotomist visited each subject at home, administered a health questionnaire, and obtained 2 nonfasting blood samples. Anthropometrics (height, weight, hip and waist circumference) and blood pressure were directly measured according to standard protocols developed for large-scale fieldwork (39). Two nonfasting blood samples were drawn and serum was pooled for analysis. High-density lipoprotein cholesterol (HDL-C) was measured at a Centers for Disease Control and Prevention-certified lipid laboratory (Miriam Hospital, Providence, Rhode Island). The following information was collected via interviewer-administered questionnaire: demographics, psychosocial factors, history of chronic disease, self-assessed general health status, tobacco and alcohol use, nutritional intake, and physical activity/energy expenditure during the past 7 days. MMAS received institutional review board approval and all subjects gave written informed consent.

Established CVD risk factors were used to control for confounding. The following were input as continuous variables: age, body mass index, HDL-C, and total cholesterol. In addition, we adjusted for current smoking (yes/no) and hypertension categorized according to blood pressure readings by the Joint National Committee on Detection, Evaluation, and Treatment of High Blood Pressure definition (40). Also, we constructed the Framingham risk score, which gives the 10-year estimated probability of a coronary heart disease (CHD) event according to Adult Treatment Panel III guidelines (41).

At the end of the interview, the subject was given a 23-item questionnaire on sexual activity to be completed in private and returned in a sealed envelope (42). The questionnaire included 13 items related to ED, such as, “During the last 6 months have you ever had trouble getting an erection before intercourse begins?” The 13 items were combined in a discriminant-analytic formula to assign a degree of erectile function to each subject (43). The same discriminant formula was used at both baseline and follow-up.

Calibration data for the discriminant formula were taken from an additional single-question, subjective self-assessment of ED, included in the follow-up questionnaire in response to recommendations of the National Institutes of Health Consensus Panel (44). Impotence was defined as “being unable to get and keep an erection that is rigid enough for satisfactory sexual activity.” The subject rated himself as completely impotent (“never able to get and keep an erection …”), moderately impotent (“sometimes able …”), minimally impotent (“usually able …”), or not impotent (“always able …”). In random subsets of the follow-up samples the self-assessment was validated (45) against 2 established ED measures, the International Index of Erectile Function (46) (r = 0.71, n = 254) and the Brief Male Sexual Function Inventory (47) (r = 0.78, n = 251), as well as an independent urologic assessment (48). As we have done in previous analyses (23,26,49), we analyzed both the 4-category ED status variable and also a binary ED status variable (absence/presence) that was defined as moderate or complete ED.

Data on CVD were obtained from 3 sources: self-reports, linkage of the MMAS database with the National Death Index (NDI) (50), and medical records. Self-reports included a wide range of major CVD end points (e.g., myocardial infarction, atherosclerosis, stroke, coronary artery bypass graft surgery, congestive heart failure). Subjects who gave positive endorsement of any of these were considered to have CVD. Based on medical records (primary discharge diagnosis) and the NDI (underlying cause), CVD was determined according to the International Classification of Diseases (ICD). Before 1999, events/deaths were coded according to the ICD-9th Revision and subsequently, according to the ICD-10th Revision. Subjects with the following codes were considered to have developed CVD: ICD-9/ICD-10 codes 390 to 459/I00 to I99, which include coronary heart disease, heart failure, peripheral vascular disease, cerebrovascular disease, and other vascular diseases (51).

Person-years were accumulated from each subject's baseline visit to date of last observation or event date. We computed incidence rates (cases/person-years) in each ED category, with 95% confidence intervals (CIs) estimated under the assumption that incidence rates followed a Poisson distribution (52). A Kaplan-Meier survival curve was used to illustrate the association between ED and CVD. Hazard ratios (HRs) were calculated using the Cox proportional hazards regression model (53); men with no ED served as the reference group for the 4-category ED variable and men with no or minimal ED served as the reference group for the binary ED variable. Tests for linear trend across the 4-category ED variable were performed by creating linear contrasts.

In order to address the question of whether ED contributes to the prediction of CVD, we conducted 3 sets of analyses. First, we fit multivariate Cox proportional hazards regression models to examine the independent influence of ED. Second, we evaluated the discriminatory capability of ED and traditional risk factors using C statistics, which is an extension of the traditional receiver-operator characteristic curve analysis to survival analysis (54- 55). Finally, we assessed the reclassification of CVD risk associated with ED using methods developed by Pencina et al. (56) that estimated the net reclassification improvement. This methodology involved the fitting of 2 statistical models, the first including age and Framingham risk score, and a second that added ED. Based on this, we evaluated changes in Framingham risk category reclassification (57) separately for CVD cases and noncases that occurred during the first 10 years of follow-up. The net reclassification improvement was computed by summing the following quantities: 1) the difference in proportions of individuals reclassified into a higher risk category and the proportion reclassified into a lower risk category among men who developed events; and 2) the difference in the proportion of individuals reclassified into a lower risk category and the proportion reclassified into a higher risk category among those who did not develop events. The significance of the net reclassification improvement was assessed with an asymptotic test (56). We also calculated an alternative index of discrimination that does not rely on category cut points, the integrated discrimination improvement, which can be viewed as a difference between improvement in average sensitivity and any potential increase in average 1 – specificity (56). We used SAS version 9.2 (SAS Institute, Cary, North Carolina) for all analyses. Significance was considered present when p < 0.05.

The dataset included men with complete baseline risk factor data who were free of CVD and diabetes (a CHD risk equivalent) at baseline. Of these 1,057 men, 261 (25.0%) developed CVD. Of the 261 CVD cases, 200 were confirmed by either NDI or medical record and the remaining 61 were obtained by self-report only. Of 261 CVD events, 71 (27.2%) were fatal CVD events. Men without ED at baseline (n = 879) were followed for an average of 12.0 years and men with ED (n = 178) for 10.3 years.

(Table 1) shows baseline characteristics of men according to ED status. Men with ED were older on average (age 59 ± 8 years) than men without ED (age 53 ± 8 years). Among men with ED, prevalence of hypertension and smoking was higher. Men with ED also had slightly higher body mass index, lower total and HDL-C, higher systolic blood pressure, and higher Framingham risk score. Overall, 37% of men with ED were in the highest risk category for Framingham risk score, compared with 17% of men without ED.

Age-adjusted CVD incidence rates are shown in (Table 2). As expected, CVD incidence was strongly related to Framingham risk score. Data not shown provide no evidence to suggest variation in the association between ED and CVD according to age (interaction p values for ED by age [categorical or continuous] >0.5). Age-adjusted CVD incidence increased with ED severity in a nonmonotonic fashion (p = 0.08), with higher rates observed in men with moderate and complete ED compared with men who had no or minimal ED. For the binary ED variable, CVD incidence was 19.7 (95% CI: 17.3 to 22.5) per 1,000 person-years among men with none/minimal ED compared with 26.9 (95% CI: 20.9 to 34.7) per 1,000 person-years among men with moderate/complete ED (p = 0.02).

(Table 3) shows the relationship between ED and CVD in various multivariate models. The assumption of proportional hazards was met for these models. Adjusted for age, ED was significantly associated with CVD incidence (HR: 1.42, 95% CI: 1.05 to 1.90, p = 0.02). Further adjustment for body mass index, HDL-C, total cholesterol, current smoking, and Joint National Committee on Detection, Evaluation, and Treatment of High Blood Pressure hypertension categories decrease the HR slightly to 1.41 (95% CI: 1.05 to 1.90, p = 0.02). In addition, ED was significantly associated with CVD when adjusted for age and Framingham risk score (HR: 1.40, 95% CI: 1.04 to 1.88, p = 0.03). In multivariate sensitivity analyses in which we included only men with CVD that was confirmed by medical record or NDI (with self-reports considered nonevents) or fatal CVD events (with nonfatal CVD considered nonevents), the HRs associated with ED were 1.37 (95% CI: 0.98 to 1.90, p = 0.07) and 1.34 (95% CI: 0.79 to 2.28, p = 0.28), respectively.

(Table 4) shows the C statistics for CVD according to various multivariate models. The C statistic for the full multivariate model was 0.7068. Addition of ED to this model offered only a small improvement in the resulting C statistic to 0.7106. The same pattern was observed with the age and Framingham risk score model, where the addition of ED caused the C statistic to increase from 0.6910 to 0.6953.

Data on the number of subjects according to Framingham CVD risk category based on an age-adjusted regression model, with reclassification of risk category after inclusion of ED status in a multivariate statistical model are shown in (Table 5). Several noteworthy observations can be made. First, among 902 men who did not develop CVD within 10 years (non-CVD cases in this analysis), inclusion of ED resulted in reclassification of 56 men (6.2%, 95% CI: 4.6% to 7.8%); 39 of these men were reclassified into a lower risk category and 17 were reclassified into a higher risk category. Second, among 155 men who developed CVD within 10 years, inclusion of ED resulted in reclassification of 17 men (11.0%, 95% CI: 6.1% to 15.9%); 8 of these men were classified into a lower risk category and the remaining 9 into a higher risk category. Based on this information, the net reclassification improvement for ED was calculated as 3.1% (95% CI: –2.4% to 8.5%), which was not statistically significant (p = 0.27). An alternative measure of discrimination, the integrated discrimination improvement, was estimated at 0.003 (95% CI: –0.001 to 0.008), which was also not statistically significant (p = 0.13).

In this prospective study of 40- to 70-year-old men followed for 12 years, ED predicts the development of CVD, independent of age, traditional risk factors, and Framingham risk score. In models adjusted for established risk factors, men with ED have a 40% higher risk of developing CVD than men without ED. Contrary to our hypothesis, and in spite of the statistical significance of the association between ED and CVD, we are not able to confirm that ED improves the prediction of CVD incidence in middle-aged and older men beyond that offered by the Framingham risk score. This is perhaps expected given the strength of the association between traditional risk factors and CVD, the relative magnitude of the observed HR associated with ED, and that numerous studies have shown that the factors that comprise the Framingham risk score are associated with ED itself (6,20- 25).

In both low (32- 33) and high (30,58- 59) cardiovascular risk populations, ED has been shown to predict a composite end point of various adverse cardiac events. Montorsi et al. (8), in a sample of 285 patients with coronary artery disease, showed that extent of coronary artery disease is related to severity of ED. In their study (8), ED generally preceded presentation of CAD by 2 to 3 years on average. Among men with type 2 diabetes who did not have clinically overt CVD, the presence of ED predicted CHD events (59). A historical cohort study based on medical records data (60) showed that ED significantly predicted CVD in the period before the introduction of sildenafil, but not afterward. Three large prospective cohort studies have shown that ED predicts CVD. In the PCPT (Prostate Cancer Prevention Trial) study (32), the multivariate-adjusted HR for ED was 1.45, which was independent of age and other CVD risk factors. Indeed, PCPT data show that ED was as strongly related to CVD as some traditional CVD risk factors. In the Krimpen Study (33), the age- and Framingham-adjusted HR was 1.6 (95% CI: 1.2 to 2.3) for reduced erectile rigidity and 2.6 (95% CI: 1.3 to 5.2) for severely reduced erectile rigidity. Using data from the Olmstead County Study, Inman et al. (31) have recently shown that ED was associated with an approximately 80% higher risk of subsequent coronary artery disease. The association of ED with coronary artery disease in that study was particularly strong among younger men; this is unlike the current study, in which the association between ED and CVD was consistent across age groups. Despite the fact that all 3 large prospective cohort studies observed a significant association of ED with CVD independent of risk factors, as observed in this report, none assessed whether ED improved the prediction of CVD using reclassification statistics.

The biological mechanisms linking ED and CVD are relatively well-established. Endothelial dysfunction, characterized by impaired nitric oxide bioavailability, precedes the development of atherosclerotic lesions and has been suggested as an important link between ED and CVD (3,6,8,14- 19,61). The penile corpora may be more susceptible to the consequences of reduced vasodilation and blood flow reserve than the heart or brain given the smaller diameter of the penile arteries (29). In addition, the peripheral cavernosal arteries are end arteries, and thus do not have the ability to form collaterals to compensate for decreased blood flow, as does the heart (62). Thus, loss of vasodilation may be recognized earlier in the microvascular penile bed than in coronary arteries.

Limitations to the current study should be acknowledged. Perhaps the most important limitation concerns the measurement of ED. The ED variable used in this report was derived from an ED self-assessment, widely considered the gold standard, performed during the second examination; it was not measured directly. Unfortunately, the self-assessment was not included at baseline. Also, we were not able to confirm 61 CVD self-reports with objective information. Nonetheless, in a sensitivity analysis where all unconfirmed self-reported CVD events were coded as noncases, the multivariate-adjusted HR associated with ED (1.37, 95% CI: 0.98 to 1.90) was similar in magnitude to the HR that included unconfirmed events, suggesting no bias in the estimate due to inclusion of self-reported CVD events. Another concern is that MMAS included mostly white men of higher socioeconomic status, so these results may not be generalizable to more diverse populations. However, MMAS was representative of the greater Boston, Massachusetts, male population at the time of sampling (63). Although the low (52%) response rate at baseline is cause for concern, a telephone survey of 206 nonrespondents to MMAS (42) showed that whereas nonrespondents were older, less likely to report cancer or heart disease, and more likely to report their health as fair or poor compared with the entire cohort, there were no differences in the prevalence of diabetes, high blood pressure, history of prostate surgery, or restriction in activity due to poor health. Furthermore, the crude CVD incidence rate observed in this cohort (21.0 per 1,000 person-years) is nearly identical to the CVD incidence rate among men age 55 to 64 years in the Framingham Heart Study (21.4 per 1,000 person-years) (64), suggesting that attrition and inclusion of self-reported CVD events did not bias our estimates.

These limitations must be considered in light of the strengths of this study. These include a random, population-based sample of generally healthy, well-characterized men from a defined geographic area, the ability to statistically adjust for a number of factors that could confound the association between ED and CVD, as well as the length of follow-up and the relatively sizable number of events. We also used novel statistical methods that were designed to assess the additional predictive utility of new markers for disease outcomes and which extend traditional reclassification estimates that ignore the direction of the reclassification.

The clinical implications of the current study are mixed. On the one hand, this study provides confirmatory evidence that ED is a sentinel for CVD, independent of established risk factors. On the other, we are unable to show that ED significantly improves the prediction of who will develop CVD. Nonetheless, any reclassification would be useful clinically given that the assessment of ED is associated with little cost and no risks. Thus, the threshold for demonstration of clinical utility for ED screening would need to be far lower than for more expensive screening tests, such as C-reactive protein or coronary calcium. Finally, the present findings emphasize the need for primary care physicians and other health care providers to pay particular attention to the cardiovascular risk profiles of their patients with ED, in keeping with current recommendations (4,65).