Journal of the National Cancer Institute Advance Access originally published online on June 27, 2007
JNCI Journal of the National Cancer Institute 2007 99(13):1025-1035; doi:10.1093/jnci/djm023
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
© The Author 2007. Published by Oxford University Press.
ARTICLES |
Longitudinal Analysis of Sexual Function Reported by Men in the Prostate Cancer Prevention Trial
Affiliations of authors: Southwest Oncology Group Statistical Center (CMM, AKD) and Division of Public Health Sciences (DLP), Fred Hutchinson Cancer Research Center, Seattle, WA; Pain Research Center, Department of Anesthesiology, University of Utah, Salt Lake City, UT (GWD); Department of Urology, University of Texas Health Sciences Center at San Antonio, San Antonio, TX (IMT); Department of Urology, Wilford Hall Medical Center, Lackland Air Force Base, TX (CL); Institute for Medical Informatics, Biometry and Epidemiology, University of Munich, Munich, Germany (DPA); Department of Health Services, University of Washington, Seattle, WA (DLP); QualityMetric Incorporated, Lincoln, RI (JEW); Health Assessment Lab, Waltham, MA (JEW); Schools of Medicine and Public Health and Jonsson Comprehensive Cancer Center, University of California at Los Angeles, CA (PAG); Department of Public Health Sciences, Wake Forest University School of Medicine, Winston-Salem, NC (SAS); Departments of Clinical Cancer Prevention and Thoracic/Head and Neck Medical Oncology, The University of Texas M. D. Anderson Cancer Center, Houston, TX (SML); Cancer Control and Prevention, Southwest Oncology Group, Operations Office, San Antonio, TX (CAC)
Correspondence to: Carol M. Moinpour, PhD, Southwest Oncology Group Statistical Center, Fred Hutchinson Cancer Research Center/M3-C102, 1100 Fairview Ave North, Box 19024, Seattle, WA 98109-1024 (e-mail: cmoinpou{at}fhcrc.org).
 |
ABSTRACT |
|---|
 Top Abstract Context and CaveatsSubjects and MethodsResultsDiscussionFundingReferencesNotes |
|---|
Background: The Prostate Cancer Prevention Trial (PCPT) was a randomized, double-blind, placebo-controlled study of the efficacy of finasteride in preventing prostate cancer in 18882 men aged 55 years or older. The PCPT offered an opportunity to prospectively study the effects of finasteride and other covariates on sexual dysfunction.
Methods: We assessed sexual dysfunction in 17313 PCPT participants during a 7-year period. A battery of questionnaires assessed sexual dysfunction (Sexual Activity Scale score); age; race; SF-36 Mental Health Inventory-5, Physical Function, and Vitality scores; body mass index; smoking status; and the presence of diabetes and hypertension. Assessments began at month 6 after random assignment and included the Sexual Activity Scale score at randomization as a covariate. Two-sided general t tests, with a cutoff of P value less than .05, were used to determine the statistical significance for mixed model effects with correlated random time slopes and intercepts. The changing impact of covariates on sexual dysfunction was also assessed at 6 months, 3.5 years, and 6.5 years after randomization.
Results: Finasteride increased sexual dysfunction only slightly and its impact diminished over time; the increase in the Sexual Activity Scale score relative to placebo of 3.21 points (95% confidence interval [CI] = 2.83 to 3.59 points; P<.001) at the first assessment decreased to 2.11 points (95% CI = 1.44 to 2.81 points; P<.001) at the end of study. These Sexual Activity score values were small on a scale of 0–100, the range observed in the study, and in comparison with individual variation. After adjustment for all covariates, mean sexual dysfunction increased in both arms from baseline (6 months after randomization) by 1.26 Sexual Activity points (95% CI = 1.16 to 1.36 points; P<.001) per year, corresponding to a cumulative increase of 8.22 points (95% CI = 7.52 to 8.92 points; P<.001) over the study period.
Conclusions: The effect of finasteride on sexual functioning is minimal for most men and should not impact the decision to prescribe or take finasteride.
Prior knowledge
The use of the 5 Study design PCTP participants completed questionnaires regularly throughout the 7 years of the study to quantify the effects of finasteride and other factors (e.g., age, medical conditions, smoking status) on sexual dysfunction over time. Contribution Finasteride caused a small average increase in sexual dysfunction that decreased over time. The effects of finasteride treatment on sexual dysfunction were clinically less relevant than natural sources of variability in the study population. Implications Finasteride would cause little or no sexual dysfunction for most men who might take it. Limitations Patient-reported outcomes were not collected from participants who temporarily or permanently discontinued taking the assigned study drug, and sexual dysfunction side effects were often cited as the reason for discontinuing. The scale used to measure sexual dysfunction was not an established psychometric measure before its use in the PCPT.
|
-reductase inhibitor finasteride has been linked to sexual dysfunction in men. However, results from previous studies including the Prostate Cancer Prevention Trial (PCPT) have suggested that evaluations of the impact of finasteride on sexual dysfunction must take into account the amount of time on study and participant age, as well as individual variation among men.Sexual dysfunction in men has been associated with many factors, including older age; poor general health; comorbidities and associated medications; lower levels of high-density lipoprotein cholesterol and of dehydroepiandrosterone; psychologic factors, such as depression, anxiety, anger, relationship/partner concerns, and dominance; and lifestyle factors, such as smoking and diet (1–7). Sexual dysfunction in men has also been linked to the use of the 5-alpha-reductase inhibitor finasteride. For example, in one double-blind study of men with benign prostatic hyperplasia, sexual dysfunction (including decreased libido, impotence, and ejaculation disorders) at 12 months was reported by 12.5% (of n = 297) of those who received 5 mg of finasteride compared with 4.7% (of n = 300) of those who received placebo (8). In a similar trial with longer follow-up, sexual dysfunction at 12 months was reported by 10.8% (of n = 543) of the men in the 5 mg finasteride arm compared with 5.2% (of n = 555) of the men in the placebo arm. Placebo arm follow-up ceased after 12 months; men on finasteride were followed for 36 months. By 36 months, 62% of adverse sexual events had resolved in the men receiving 5 mg of finasteride (9). The Prostate Cancer Prevention Trial (PCPT), with its very large sample and a follow-up period of 7 years, offered an excellent opportunity to evaluate effects of finasteride on sexual dysfunction for longer than 3 years; sexual dysfunction was therefore a prespecified secondary endpoint for the PCPT.
The PCPT was a 7-year randomized, double-blind, placebo-controlled trial of the efficacy finasteride for the prevention of prostate cancer (10–15). Trial results indicated a 24.8% reduction in the prevalence of prostate cancer among men who took finasteride. Finasteride use was also associated with fewer urinary symptoms but more sexual dysfunction side effects than placebo (14). During the trial, for purposes of participant education, the Data Safety and Monitoring Committee approved an analysis of placebo-arm PCPT participants; this analysis showed that men in the placebo arm reported increased sexual dysfunction during the first 2 years of the trial even though they were not taking finasteride (15). This finding was consistent with the results of studies such as the Massachusetts Male Aging Study (1). Together, these results indicate the need to incorporate both the time on study and participant age in any evaluation of the impact of finasteride on sexual dysfunction.
In this study, we quantified the effect of finasteride on sexual dysfunction in PCPT participants for the 7 years of the trial and compared the magnitude of its effect with that of naturally occurring influences, with that of relevant covariates associated with male sexual dysfunction, and with individual variation both within and between treatment arms.
|
Subjects and Methods |
|---|
|
TopAbstractContext and CaveatsSubjects and MethodsResultsDiscussionFundingReferencesNotes |
|---|
Prostate Cancer Prevention Trial Participants
The PCPT enrolled 24 482 men between October 13, 1993, and December 6, 1996. After a 3-month run-in period, 18 882 men were randomly assigned to receive finasteride or placebo. Men had to be 55 years old or older to be eligible (14). Consent forms and the PCPT protocol were approved by the institutional review boards at the participating institutions; all men provided written informed consent. To be included in this analysis, participants were required to have completed questionnaires that included the Sexual Activities Scale and questions about other covariates at the time of randomization and the Sexual Activity Scale at least two times after randomization. A total of 17313 PCPT participants were included in this analysis (Fig. 1).
|
Sexual Dysfunction Outcome
Assessment of sexual dysfunction in the PCPT included two participant-completed questionnaires [i.e., the Sexual Problems Scale (16) and the Sexual Activity Scale] and sexual toxicity ratings that were completed by clinical research associates. Self-reported patient measures were obtained at study enrollment, at randomization, at 6 months after randomization, and annually for 7 years; assessments by clinical research associates were made every 6 months beginning at study entry. We used the Sexual Activity Scale rather than the Sexual Problems Scale because 44% of PCPT participants reported Sexual Problems Scale scores of zero at randomization and because a substantial number of zero scores were maintained over time (17). These data indicate that the Sexual Problems Scale is not well suited for investigation of subtle longitudinal change in this population. The large number of zero scores on the Sexual Problems Scale constitutes a "floor effect" that is impossible to remove with any transformation usually employed to reduce skew and improve distributions. By contrast, the Sexual Activity Scale elicited very few zero scores (<1%) and the responses approximated a normal distribution.
The four-item Sexual Activity Scale was developed for the PCPT (by D. L. Patrick and C. M. Moinpour) based on items used previously in benign prostate hyperplasia research (18): ability to have an erection when desired (five response levels), degree of participant's satisfaction with his sexual activities (four response levels), change in sexual performance (seven response levels), and frequency of sexual activities (seven response levels). Each item is transformed to a 0–100 scale, and the overall Sexual Activity Scale score is computed as the mean score for the four items. Scores can range from 0 to 100, with higher scores reflecting more sexual dysfunction (less positive sexual activity).
We documented the psychometric properties of the Sexual Activity Scale in several ways. The internal consistency reliability (19) of the Sexual Activity Scale at study enrollment was 0.66 (7). Generally, reliability coefficients of 0.70 or more are considered to be acceptable for group-level comparisons (20). However, in a longitudinal study, a more appropriate estimate of scale reliability is the composite reliability for a longitudinal outcome, as measured by the intraclass correlation coefficient over multiple assessments (21); for four assessments, the composite reliability was 0.89, and for the eight PCPT assessments (beginning 6 months after randomization), it was 0.94. Known groups validity was documented (20), with the Sexual Activity Scale differentiating between groups of participants in the expected direction: for example, older versus younger participants reported more sexual dysfunction as did participants with diabetes, high blood pressure, and depression and those who were current smokers (7). The Sexual Activity Scale is also highly correlated with the Sexual Problems Scale, a previously validated measure (16); correlation coefficients ranged from .71 to .78 over the nine assessment points beginning with randomization (Student's t tests, P<.001 at all nine time points). These correlation coefficients are as large as mathematically possible given the reliabilities of the scales; hence, the two scales represent alternative measures of the same construct (20) of sexual dysfunction.
Covariates
Health-related quality of life was a secondary endpoint in the PCPT to ensure that giving healthy men a chemopreventive agent would not cause undue harm; this outcome was a covariate in the analyses presented here. Participants completed three questionnaires: the Health Survey Short Form-36 (SF-36), which was completed at study enrollment, 6 months after randomization, and annually for 7 years; the Participant Report of Symptoms (including the Sexual Activity Scale), which was completed at enrollment, randomization, 6 months after randomization, and annually for 7 years; and the Participant Lifestyles survey, which was completed at enrollment only (7). In addition, participants were monitored by clinical research associates for side effects, complications, and other clinical endpoints four times a year. Three of the SF-36 health-related quality of life scales were treated as covariates in this analysis: the Physical Functioning, the Mental Health Inventory-5, and the Vitality scales (22).
The Participant Lifestyles survey assessed four health behaviors: smoking status, physical activity, diet (not examined in this analysis), and alcohol use. Physical activity was reported in metabolic equivalents (METs) (23–27); this variable was not a statistically significant covariate in the final models. The number of drinks consumed per day was calculated as the sum of the responses to the health behaviors questionnaire items on beer, wine, and hard liquor consumption. Smoking status at enrollment was categorized as current smoker, former smoker, or nonsmoker. We assessed the dose effect of smoking on sexual dysfunction in the subset of participants who were currently smoking. To measure smoking dose, we examined the number of years the participant had smoked (duration), one of the two components of pack-years (number of packs per day and duration of smoking) rather than the composite variable, pack-years; the composite variable is considered to be problematic because both components are independent predictors (Gritz ER: personal communication).
Self-reported race was coded as white (non-Hispanic), black (regardless of ethnicity), or other. Body mass index (BMI) was also calculated.
Statistical Methods
Submission Rates. If participants were missing responses for one of the four items, the missing item was ignored in the computation of the mean overall score. Participants who were missing responses for two or more items were classified as missing that assessment on the Sexual Activity Scale. Questionnaire submission rates were calculated by treatment arm by dividing the number of questionnaires submitted by the number of participants who were expected to complete the questionnaire. Participants who stopped taking study drug (a "drug holiday") were not required to complete health-related quality of life forms until they returned to on-treatment status (active drug or placebo).
Model Selection. We considered two classes of models for this analysis. A discontinuous regression model (which generates "hockey-stick" trajectory patterns) (28) included the Sexual Activity Scale score at randomization as one of the longitudinal assessment points but introduced a special onset term for the initial gain in sexual dysfunction following randomization. An analysis of covariance model, by contrast, treated the Sexual Activity Scale score at randomization as a covariate for predicting the longitudinal assessment of sexual dysfunction following randomization. Hierarchical models (29) without covariates other than a time variable and the treatment effect (and, when appropriate, a variable marking treatment onset) were run using the mixed model procedure in SAS statistical software (version 9.0) (28) to compare these two classes of models. Plots of the residuals, model fit, and random effects for the two classes of models were compared and, on the basis of these comparisons, we selected the analysis of covariance model, which treated the Sexual Activity Scale score at randomization as a covariate in the analysis of posttreatment measures. Compared with the discontinuous regression model, the analysis of covariance model gave a substantially better fit to the data and produced residual and random effect plots that more closely matched a normal distribution.
Covariate Selection. Our initial selection of other possible covariates was based on a review of the literature (e.g., 1–7,17,30); preliminary models reduced this larger set of variables described above (see "Covariates") to a minimal set of covariates that included age (as a continuous variable), smoking status at enrollment (current versus former smoker; reference category, nonsmoker), diabetes (presence versus absence), hypertension (presence versus absence), and the SF-36 scales, Physical Functioning, Mental Health Inventory-5, and Vitality scores (as continuous variables). Each variable was included as a single covariate as well as with a time interaction in the longitudinal analyses, which began at 6 months after randomization. The remaining variables were compared in a subset of participants with no missing covariates by using the Bayesian Information Criterion (28). On the basis of this comparison, the best model added only BMI and the BMI-by-time interaction to the minimal set of predictors. The number of years smoked by current smokers was not statistically significant, and this dose variable was dropped from further consideration.
Model Fit Assessments. Mixed-effects models (29) with correlated random time slopes and intercepts were used to fit the analysis models. As in standard multiple regression, each model incorporated the fixed effects of finasteride treatment, time on study, demographic and comorbidity covariates, the Sexual Activity Scale score at randomization, and covariate interactions with time on study. In addition to the fixed effects, which estimate population relationships, the mixed-effects models also provided variance estimates ("random effects") that accounted for differences among participants in their level of sexual dysfunction 6 months after randomization and their rate of change in sexual dysfunction over the course of the study. The analytic approach we used treats these individual differences as distinct from measurement and other random errors. Preliminary comparisons of nested models indicated that it was not necessary to include an interaction term between treatment arm and the Sexual Activity Scale score at randomization.
Clinically Significant Predictors. All continuous covariates (i.e., the Sexual Activity Scale score at randomization; Physical Functioning, Mental Health Inventory-5, and Vitality scores; age; and BMI) were adjusted by subtracting the population median from individual scores. This adjustment made it possible for the intercept term to stand for the Sexual Activity Scale score at 6 months after randomization for the "median man" (i.e., a man in placebo arm of the PCPT who did not smoke, was not diabetic or hypertensive, and had median scores on all linear covariates at 6 months after randomization). Two-sided general t tests (31) with a cutoff of P value less than .05 were used to determine statistical significance for mixed model effects: 95% confidence intervals (CIs) are provided.
|
Results |
|---|
|
TopAbstractContext and CaveatsSubjects and MethodsResultsDiscussionFundingReferencesNotes |
|---|
Participant Characteristics and Questionnaire Submission Rates
Demographic and clinical characteristics of study participants were similar between the two arms (Table 1). Questionnaire submission rates for usable Sexual Activity Scale forms (i.e., those that could be used to calculate a total sexual activity score) were high in both study arms, ranging from 98% in each arm at study enrollment to 92% in each arm at 7 years after randomization (finasteride [enrollment, randomization, 6 months, 1 year, 2 years, 3 years, 4 years, 5 years, 6 years, 7 years]: 98%, 98%, 93%, 95%, 94%, 94%, 93%,93%, 92%, 92%; placebo [enrollment, randomization, 6 months, 1 year, 2 years, 3 years, 4 years, 5 years, 6 years, 7 years]: 98%, 98%, 93%, 95%, 94%, 94%, 94%, 93%, 92%) . However, although the questionnaire completion rates were high among participants taking the assigned study drug (i.e., those expected to submit questionnaires), the number of men who stopped taking their assigned study drug (i.e., went off treatment) between randomization and 6 months differed statistically significantly (P<.001, chi-square test) between the finasteride and placebo arms. Consequently, the number of men who had Sexual Activity Scale scores at study randomization and at 6 months after randomization also differed statistically significantly between the two arms; this pattern continued throughout the 7-year study period (P<.001, chi-square test). In addition, because the primary reason for going off treatment was sexual dysfunction, our selection of analysis methods for these data must be based on the actual number of available Sexual Activity Scale forms and whether or not there is any bias associated with the reason for missing forms. To evaluate whether the declining rates might have induced some systematic difference in the composition of the treatment arm participants over time, we conducted a simple sensitivity analysis. We compared the full dataset model for the Sexual Activity Scale–dependent variable with models run on the subsets of participants with the fewest and the most Sexual Activity Scale assessments (data not shown). Because these three models were nearly identical in both the size and the direction of key coefficients, we treated the missing data as data missing at random (32) in the mixed model analyses reported below.
|
Sexual Dysfunction Outcomes
The descriptive data in Table 2 show sample sizes and mean Sexual Activity Scale scores and standard deviations for the two study arms at all time points. Participants who received finasteride reported slightly higher scores (i.e., more sexual dysfunction) than those who received placebo beginning 6 months after randomization. By the end of the study (i.e., at 7 years after randomization), the sample size decreased to 56% of the men at study entry in the placebo arm (n = 5169) and 52% of the men at study entry in the finasteride arm (n = 4777).
|
Table 3 presents the full formal coefficient estimates for the effects described above under "Model Selection" and "Covariate Selection." Although the P values for most estimates were highly statistically significant, the magnitudes of the estimates were often small in terms of practical clinical impact, as we describe below. In particular, there was a statistically significant difference between the two treatment arms on the Sexual Activity Scale score after statistical adjustment for prerandomization differences: participants who received finasteride reported Sexual Activity Scale scores that were, on average, 3.21 points higher (95% CI = 2.83 to 3.59 points; P<.001) than those reported by participants who received placebo.
|
Model complexity and unfamiliar metrics can obscure interpretation of the key statistical results shown in Table 3. Therefore, in Table 4, we have re-expressed the estimates from Table 3 in terms of equivalent coefficients that are easier to interpret. These re-expressions take three forms. One type of re-expression provides a more natural interpretation for the complicated time interactions in Table 3. The second type of re-expression transforms the arbitrary units of the SF-36 scales into more natural ones. The third type of re-expression provides effect sizes that scale impact relative to the expected variation in Sexual Activity scores among "similar men" who are otherwise identical (see below for more precise definition).
|
The estimates in Table 4 show how the impact of covariates assessed at baseline changed at representative time points: baseline (i.e., 6 months after randomization), 3.5 years after baseline (i.e., 4 years after randomization), and 6.5 years after baseline (i.e., 7 years after randomization). The estimates tabulated at each of the three time points denote the effect of the described covariate at that time. For example, the finasteride effect at baseline is 3.21, indicating that the mean Sexual Activity Scale score for the finasteride group exceeds the mean Sexual Activity Scale score for the placebo group by 3.21 Sexual Activity Scale points (adjusted for all other covariates in the model) at 6 months after randomization. The comparable mean difference 6.5 years later was 2.11 Sexual Activity Scale points. Despite the high levels of statistical significance reported for finasteride in Table 3, these mean effects, and the changes in the mean effects, are small when considered relative to the 0–100 range of the Sexual Activity Scale.
The second form of re-expression in Table 4 provides a practical clinical interpretation of the SF-36 scales, psychometric variables with arbitrary units. The original estimates (in Table 3) defined the expected change in Sexual Activity based on a change of one unit in an SF-36 scale. Because a one-unit change in an SF-36 scale is too small to have clinical meaning, we rescaled the original SF-36 effect estimates by multiplying each estimate by 100 (the full range of the scale), as shown in Table 4. The rescaled estimates in Table 4 are statistically equivalent to the estimates in Table 3 but are easier to interpret: that is, the estimates in Table 4 for the SF-36 scales represent the expected difference in Sexual Activity Scale scores between two otherwise identical men having the maximum possible difference on a single SF-36 scale. For example, consider two men, one with a score of 100 (best possible score) and the other with a score of 0 (the worst possible score) on the Physical Functioning scale at 6 months after randomization. The Physical Function estimate of –5.54 [–0.055 x 100] (95% CI = –6.88 to 4.20; P<.001) reflects an expected advantage of 5.54 Sexual Activity Scale points for the man with the score of 100 on Physical Functioning. A difference of 100 points is, of course, unlikely in two otherwise identical men, but nonetheless provides an interpretable standard. In this example, even the maximum possible difference on the Physical Functioning scale would correspond to a difference of only 5.54 points—i.e., a small difference—on the 0–100 Sexual Activity Scale; smaller, more realistic differences in Physical Functioning estimates would correspond to proportionally smaller differences in Sexual Activity Scale points.
The third form of reexpression in Table 4 compares the size of the effect estimate relative to the natural variation among similar men. By "similar men," we refer to men in the same treatment arm who were similar to each other with respect to all demographic and medical characteristics coded in the analysis. Table 4 presents effect ratios for those covariates, such as finasteride and hypertension, which have natural interpretations as the presence or absence of a condition. These effect ratios were computed as the ratio of the effect estimate (the mean Sexual Activity Scale score for those having the condition minus the mean score for those not having the condition) to the standard deviation in the Sexual Activity Scale among similar men who were similarly measured (i.e., control for measurement error). Because the standard deviation describes a typical deviation for a population, the effect ratio compares the typical difference in Sexual Activity Scale scores between similar men in two different conditions (covariate of interest in the numerator) to the typical deviation expected within similar men (the denominator). The denominator for the effect ratios was estimated as the square root of the between-person variance of the intercept (i.e., the Sexual Activity Scale score when time = 0) from the mixed-effects analysis: the typical deviation among similar men, similarly measured was approximately 10.85 Sexual Activity Scale points. This value indicates excellent sensitivity for the scale, just as large variabilities for measures such as weight and height indicate the underlying scales are sensitive and reliable. A related index of scale sensitivity is the typical difference between two randomly selected similar men, similarly measured. This expected person-to-person difference (after controlling for measurement error) is 15.34 points (i.e., 10.85 x
2). Even the larger effect estimates in Table 4 were dwarfed by this intrinsic variability. The largest effect ratio in Table 4, which was associated with current smoking at 6.5 years, was 0.32; the effect ratio for finasteride at 6.5 years was 0.20.
In general, the practical impact of finasteride and the covariates on the Sexual Activity Scale scores was small. The largest effect of finasteride on sexual function was a mean difference of 3.21 points at 6 months after randomization, which was associated with an effect ratio of only 0.30. Diabetes, hypertension, and smoking also did not have large impacts on Sexual Activity Scale scores. The effects of the SF-36 functioning measures were even smaller. As already noted, a difference of the full 100-point range of each SF-36 scale would generate only small differences in the Sexual Activity Scale score, which also has a range of 100 points (–5.54, –1.63, and –7.32 for the SF-36 Physical Functioning, Mental Health Inventory-5, and Vitality scales, respectively). In a more realistic clinical context, a 10-point difference in an SF-36 scale score would yield expected adjusted Sexual Activity Scale score differences at 6 months after randomization of –0.55 (–5.54/10), –0.16 (–1.63/10), and –0.73 (–7.32/10) points for the SF-36 Physical Functioning, Mental Health Inventory-5, and Vitality scales, respectively; mean differences of less than one point on the 0–100 Sexual Activity Scale are very small in real terms. The effects of BMI, though statistically significantly different from zero, were also extremely small; 1-point differences in BMI (equivalent to a weight increase of about 7 pounds for a man with a median height of 70 inches) yielded expected differences of much less than 1 point in sexual dysfunction; the negative effect of BMI on Sexual Activity Scale scores increased slightly over time.
The coefficient of the age variable (chronological age at randomization) and the coefficient of the interaction of age and time were among the larger effects on Sexual Activity Scale scores in Tables 3 and 4. At baseline, the estimated age coefficient of 0.49 describes an expected increase of 0.49 Sexual Activity Scale points per year of age. For example, a 70-year-old man 6 months after randomization might expect to have a Sexual Activity Scale score that was 4.9 points higher (worse) than a 60-year-old man 6 months after randomization. Longitudinal age changes occurring in both treatment arms are somewhat more potent than age differences. The time effect estimate of 1.26 Sexual Activity Scale points per year from Table 4 translates to average decrements of 4.43 points (95% CI = 4.05 to 4.81 points; P<.001) and 8.22 points (95% CI = 7.52 to 8.92 points; P<.001) by 3.5 years and 6.5 years on study, respectively.
Figure 2 depicts the relative expected impacts of covariates on Sexual Activity Scale scores in terms of number of Sexual Activity Scale points associated with selected study covariates, as well as the typical difference between the sexual dysfunction of any two "similar" men and the typical deviation from the average "similar" man. This figure illustrates the clinical importance of these covariates and the degree of intrinsic, systematic interindividual variation in sexual dysfunction that remains even after accounting for all the measured covariates, including treatment assignment.
|
|
Discussion |
|---|
|
TopAbstractContext and CaveatsSubjects and MethodsResultsDiscussionFundingReferencesNotes |
|---|
Sexual dysfunction was of particular interest in the PCPT because finasteride had been reported to cause erectile dysfunction (8,13,14). The large sample size and 7-year follow-up of the PCPT provided an opportunity to characterize the relationship between finasteride and broader sexual dysfunction using a health-related quality of life assessment. Our results indicate that, on average, finasteride caused a small increase in sexual dysfunction by the time of the first assessment after randomization. This small effect decreased over time, a finding that was generally consistent with the results of an earlier study (9) and had the added value of 4 additional years of follow-up. Other covariates yielded similarly small effects, with the exception of age. Older age at randomization predicted greater sexual dysfunction at the rate of 0.49 points per year, or approximately 1 point for every 2 years of age. Chronological aging, indexed with the time variables in this longitudinal design, had a greater effect on sexual dysfunction than finasteride, with sexual dysfunction increasing on average 1.26 points per year. The average longitudinal (time) increase in the Sexual Activity Scale score over the course of the study (6.5 years) was 8.22 points. This increase is presumably the result of normal aging in this cohort, although standard longitudinal designs cannot logically disentangle true aging from shared study influences occurring in the two treatment arms.
Our study has several limitations. One limitation is that patient-reported outcomes were not collected for participants who had temporarily or permanently discontinued taking the assigned study drug, and sexual dysfunction side effects were often cited as the reason for dropping out. In fact, as noted above, there were statistically significantly more men with missing questionnaire data at 6 months after randomization in the finasteride arm than in the placebo arm and participant dropout was greater in the finasteride arm during the 7 years of the trial. However, our finding that model coefficients for participants with all assessments did not differ from those with missing assessments during the course of the 7-year study supports our use of analysis techniques that assume data were missing at random (32). Therefore, we believe that the attrition over 7 years, although greater in the finasteride arm than in the placebo arm, did not compromise our evaluation of the impact of finasteride and other covariates on sexual dysfunction.
Another limitation is the fact that the Sexual Activity Scale had not been used in research before its use in the PCPT, which could cast doubt on its validity as a sound psychometric measure of sexual dysfunction. However, in this study, we have documented the reliability, validity, and sensitivity to change of the Sexual Activity Scale, supporting the scientific generalizability of findings obtained with it and its use in future research. In addition, the statistical methods we used incorporated repeated assessments optimally so that effect estimates were obtained in terms of a theoretical reliable variable that was completely separate from measurement error, a key feature of the theory of multilevel, structural equations and of mixed-effects models, such as the one we used (29,33). As noted in the "Results" and below, the Sexual Activity Scale was sensitive to individual variation both between and within participants. Figure 2 shows that the Sexual Activity Scale captures a large degree of interindividual variation in men's self-reported sexual dysfunction as well as the smaller effect of different covariates, indicating that the dependent measure is sensitive to systematic differences at a wide range of scales.
At a more practical level, our results clearly show that the Sexual Activity Scale was able to detect a variety of other covariate effects besides the finasteride effect and to compare the degree of impact of these other variables with that of the finasteride. For example, the effect of aging (time on study) was 8.22 points for the duration of 6.5 years on study, and the typical true difference between two similar men was 15.34 points. Relative to other covariate effects and individual variation, the finasteride effect of 3.21 points was small.
Our findings, as well as results of ongoing work, may ultimately contribute to clarifying the risk–benefit profile of finasteride for prostate cancer prevention (14). Our data show that the effects of finasteride treatment are clinically far less relevant than natural sources of variability in this heterogeneous population and suggest that finasteride would cause little or no sexual dysfunction for most men who might take it.
|
Funding |
|---|
|
TopAbstractContext and CaveatsSubjects and MethodsResultsDiscussionFundingReferencesNotes |
|---|
Supported by Public Health Service grants No. 5 U10 CA37429 and 2 U10 CA37429-09 from the National Cancer Institute, National Institutes of Health, Department of Health and Human Services, Bethesda, MD. The study agents (finasteride and placebo) were provided by Merck, Inc. (Merck, Inc research staff was allowed a courtesy review of the manuscript before its submission for publication.) Merck, Inc, and the National Cancer Institute, Division of Cancer Prevention also provided small grants to produce videos and to support projects to enhance trial recruitment and adherence. The study sponsor, Division of Cancer Prevention/National Cancer Institute, provided input regarding study design but had no role in the following study activities: collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication.
|
NOTES |
|---|
|
TopAbstractContext and CaveatsSubjects and MethodsResultsDiscussionFundingReferencesNotes |
|---|
The authors extend appreciation to PCPT participants who provided health behaviors information at study entry and faithfully completed the sexual function and quality-of-life forms during the course of the trial. Study clinical research associates and nurses were also instrumental in our ability to achieve such impressive submission rates for the participant-completed forms. We would also like to thank Ms Shannon Hill for assistance with the literature review.
Drs G. W. Donaldson, P. A. Ganz, C. M. Moinpour, D. L. Patrick, S. A. Shumaker, and J. E. Ware were members of the Quality of Life Advisory Committee for the PCPT.
Address reprint requests to Southwest Oncology Group (S9217), Operations Office, 14980 Omicron Dr, San Antonio, TX 78245-3217.
|
REFERENCES |
|---|
|
TopAbstractContext and CaveatsSubjects and MethodsResultsDiscussionFundingReferencesNotes |
|---|
(1) Feldman HA, Goldstein I, Hatzichristou DG, Krane RJ, McKinlay JB. Impotence and its medical and psychosocial correlates: results of the Massachusetts male aging study. J Urol (1994) 151:54–61.[Web of Science][Medline]
(2) Drugs that cause sexual dysfunction: an update. Med Lett Drugs Ther (1992) 34:73–8.[Medline]
(3) Seidman SN. Exploring the relationship between depression and erectile dysfunction in aging men. J Clin Psychiatry (2002) 63(Suppl 5):5–12.
(4) Johannes CB, Araujo AB, Feldman JA, Derby CA, Kleinman KP, McKinlay JB. Incidence of erectile dysfunction in men 40 to 69 years old: longitudinal results from the Massachusetts Male Aging Study. J Urol (2000) 163:460–3.[CrossRef][Web of Science][Medline]
(5) O'Leary MP, Rhodes T, Girman CJ, Jacobson DJ, Roberts RO, Lieber MM, et al. Distribution of the brief male sexual inventory in community men. Int J Impot Res (2003) 15:185–91.[CrossRef][Web of Science][Medline]
(6) Bortz WM 2nd, Wallace DH, Wiley D. Sexual function in 1,202 aging males: differentiating aspects. J Gerontol A Biol Sci Med Sci (1999) 54:M237–41.[Abstract]
(7) Moinpour CM, Lovato LC, Thompson IM Jr, Ware JE Jr, Ganz PA, Patrick DL, et al. Profile of men randomized to the Prostate Cancer Prevention Trial: baseline health-related quality of life, urinary and sexual functioning, and health behaviors. J Clin Oncol (2000) 18:1942–53.
(8) Gormley GJ, Stoner E, Bruskewithz RC, Imperato-McGinley J, Walsh PC, McConnell JD, et al. The effect of finasteride in men with benign prostatic hyperplasia. N Engl J Med (1992) 327:1185–91.[Abstract]
(9) Stoner E. Members of the Finasteride Study Group. Three-year safety and efficacy data on the use of finasteride in the treatment of benign prostatic hyperplasia. Urology (1994) 43:284–94.[CrossRef][Web of Science][Medline]
(10) Thompson IM, Coltman CA. Screening for prostate cancer: opportunities for prevention. Semin Urol Oncol (1996) 14:4–11.[Medline]
(11) Brawley OW, Ford LG, Thompson IM. 5-alpha-reductase inhibition and prostate cancer prevention. Cancer Epidemiol Biomarkers Prev (1994) 86:177–82.
(12) Gormley GJ, Brawley O, Thompson IM. The potential application of finasteride for chemoprevention of prostate cancer. Ann N Y Acad Sci (1995) 768:163–9.[Web of Science][Medline]
(13) Thompson I, Feigl P, Coltman C. Chemoprevention of prostate cancer with finasteride. In: Important advances in oncology, chapter 5—DeVita VT, Hellman S, Rosenberg J, eds. (1995) Philadelphia (PA): Lippincott Company. 57–76.
(14) Thompson IM, Goodman PJ, Tangen CM, Lucia MS, Miller GJ, Ford LG, et al. The influence of finasteride on the development of prostate cancer. N Engl J Med (2003) 349:215–24.
(15) Coltman CA Jr, Thompson IM Jr, Feigl P. Prostate Cancer Prevention Trial (PCPT) update. Eur Urol (1999) 35:544–7.[CrossRef][Web of Science][Medline]
(16) Sherbourne CD. Social functioning: sexual problems measures. In: Measuring functioning and well-being: the medical outcomes study approach—Stewart AL, Ware JE Jr, eds. (1992) Durham (NC): Duke University Press. 194–204.
(17) Moinpour CM, Darke AK, Pauler DK, Donaldson GW, Patrick DL, Ware JE Jr, et al. Determining predictors of men's report of sexual problems: advantages of a two-part model for zero scores. Qual Life Res (2003) 12:748.
(18) Epstein RS, Deverka PA, Chute CG, Panser L, Oesterling JE, Lieber MM, et al. Validation of new quality of life questionnaire for benign prostatic hyperplasia. J Clin Epidemiol (1992) 45:1431–45.[CrossRef][Web of Science][Medline]
(19) Cronbach LJ. Coefficient alpha and the internal structure of tests. Psychometrika (1951) 16:297–334.[CrossRef][Web of Science]
(20) Nunnally JC. Psychometric theory (1978) New York (NY): McGraw-Hill Book Company.
(21) Lord FM, Novick MR. Statistical theories of mental test scores (1968) Reading (MA): Addison-Wesley.
(22) Ware JE Jr, Sherbourne CD. The MOS 36-item Short-Form Health Survey (SF-36). I. Conceptual framework and item selection. Med Care (1992) 30:473–83.[Web of Science][Medline]
(23) Stewart AL, Hays RD, Wells KB, Rogers WH. Development of measures of physical activity/exercise and tests of variations in physical activity levels within chronic disease groups in the medical outcomes study (1993) Santa Monica (CA): RAND Corporation. Report No.: P-7781.
(24) Stewart AL, Hays RD, Wells KB, Rogers WH, Spritzer KL, Greenfield S. Long-term functioning and well-being outcomes associated with physical activity and exercise in patients with chronic conditions in the medical outcomes study. J Clin Epidemiol (1994) 47:719–30.[CrossRef][Web of Science][Medline]
(25) Satia-Abouta J, Patterson RE, Schiller RN, Kristal AR. Energy from fat is associated with obesity in U.S. men: results from the Prostate Cancer Prevention Trial. Prev Med (2002) 34:493–501.[CrossRef][Web of Science][Medline]
(26) Ainsworth BE, Haskell WL, Leon AS, Jacobs DR Jr, Montoye HJ, Sallis JF, et al. Compendium of physical activities: classification of energy costs of human physical activities. Med Sci Sports Exerc (1993) 25:71–80.
(27) Ainsworth BE, Haskell WL, Whitt MC, Irwin ML, Swartz AM, Strath SJ, et al. Compendium of physical activities: an update of activity codes and MET intensities. Med Sci Sports Exerc (2000) 32:S498–516.
(28) Littell RC, Milliken GA, Stroup WW, Wolfinger RD. SAS System for mixed models (1996) Cary (NC): SAS Institute Inc.
(29) Laird NM, Ware JH. Random-effects models for longitudinal data. Biometrics (1982) 38:963–74.[CrossRef][Web of Science][Medline]
(30) Ware JE Jr, Snow KK, Kosinski M, Gandek B. SF-36 Health survey. In: Manual and interpretation guide (1993) Boston (MA): Nimrod Press.
(31) McLean RA, Sanders WL. Approximating degrees of freedom for standard errors in mixed linear models. Proceedings of the Statistical Computing Section (1988) New Orleans (LA): American Statistical Association. 50–9.
(32) Rubin DB. Inference and missing data. Biometrika (1976) 63:581–92.
(33) Singer JD, Willett JB. modeling change and event occurrence. In: Applied longitudinal data analysis (2003) New York (NY): Oxford University Press.
Manuscript received December 1, 2006; revised April 27, 2007; accepted May 22, 2007.
Related Article in JNCI
CiteULike 
Connotea 
Del.icio.us What's this?
J Natl Cancer Inst 2007 99: 981.
This article has been cited by other articles:
|
|
 |
|
  B. S. Kramer, K. L. Hagerty, S. Justman, M. R. Somerfield, P. C. Albertsen, W. J. Blot, H. Ballentine Carter, J. P. Costantino, J. I. Epstein, P. A. Godley, et al. Use of 5-{alpha}-Reductase Inhibitors for Prostate Cancer Chemoprevention: American Society of Clinical Oncology/American Urological Association 2008 Clinical Practice Guideline J. Clin. Oncol., March 20, 2009; 27(9): 1502 - 1516. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. A. Klein and I. M. Thompson Current Status of Chemoprevention of Prostate Cancer ASCO Educational Book, January 1, 2009; 2009(1): 276 - 281. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 O. Canguven and A. L. Burnett The Effect of 5 {alpha}-Reductase Inhibitors on Erectile Function J Androl, September 1, 2008; 29(5): 514 - 523. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. W. Redman, C. M. Tangen, P. J. Goodman, M. S. Lucia, C. A. Coltman Jr., and I. M. Thompson Finasteride Does Not Increase the Risk of High-Grade Prostate Cancer: A Bias-Adjusted Modeling Approach Cancer Prevention Research, August 1, 2008; 1(3): 174 - 181. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Preventing Prostate Cancer with Finasteride: Effect on Sexual Function Journal Watch Oncology and Hematology, July 16, 2007; 2007(716): 3 - 3. [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||