Health-Related Quality-of-Life Outcomes
shows sample sizes, mean scores, and SDs by study arm at all time points for each of the three primary health-related quality-of-life outcomes; higher scores reflect better health-related quality of life. In both treatment arms, the Mental Health score increased and the Physical Functioning and Vitality scores decreased over time. However, there were no statistically significant differences in the mean scores for any SF–36 scale between the finasteride and placebo groups at any time point. We, therefore, highlight model results rather than these descriptive results because the models incorporate the effects of covariates.
Quality-of-life scale scores by treatment arm and assessment time for participants with usable scores*
Mixed-Effects Model Results: Physical Function The full results of the mixed-effects model analysis for physical function are presented in . The Physical Functioning score at enrollment was a strong predictor of later physical function (estimate = 0.50, 95% confidence interval [CI] = 0.48 to 0.51, P < .001; time interaction estimate = −0.009, 95% CI = −0.012 to 0.005, P < .001). Neither finasteride as a main effect (estimate = 0.07, 95% CI = −0.28 to 0.42, P = .71) nor the finasteride interaction with time (estimate = −0.08, 95% CI = −0.19 to 0.01, P = .08) had a statistically significant effect on Physical Functioning scores. The interaction of finasteride with the enrollment Physical Functioning score was statistically significant (estimate = 0.03, 95% CI = 0.01 to 0.05, P = .004). This finding indicates that although the intercept of each participant’s curve over time was higher, on average, in the finasteride group than the placebo group, the difference depended on the enrollment Physical Functioning score. The finasteride main effect, finasteride interaction with enrollment Physical Functioning, and finasteride effect over time (interaction) were combined to calculate an overall finasteride effect on Physical Functioning, which was small (see “Discussion” below).
Mixed model results of treatment, covariate, and individual effects on report of physical function during 7 years of follow-up (N = 16 077)*
Although most of the regression coefficients were statistically significantly different from zero, the magnitudes of most effect estimates were small. The following covariates were associated with lower Physical Functioning scores and had larger effect estimates relative to other estimates such as the finasteride main effect: diabetes (estimate = −1.31, 95% CI = −2.04 to −0.57, P < .001; time interaction estimate = −1.04, 95% CI = −1.24 to −0.83, P < .001); hypertension (estimate = −0.73, 95% CI = −1.11 to −0.35, P < .001; time interaction estimate = −0.23, 95% CI = −0.34 to −0.13, P < .001); congestive heart failure (estimate = −5.64, 95% CI = −7.96 to −3.32, P < .001; time interaction estimate = 0.25, 95% CI = −0.44 to 0.94, P = .48); leg pain (estimate = −2.57, 95% CI = −3.04 to −2.10, P < .001; time interaction estimate = −0.15, 95% CI = −0.28 to −0.02, P =.03); and current smoking status at enrollment (estimate = −2.34, 95% CI = −2.97 to −1.71, P < .001; time interaction estimate = −0.49, 95% CI = −0.67 to −0.31, P < .001). In addition, participant time on study was also negatively associated with Physical Functioning scores (estimate = −1.20, 95% CI = −1.36 to −1.03, P < .001).
Although the mixed-effects model results for mental health and vitality were generally similar to those for physical function, some differences were observed (data not shown). Finasteride was not associated with change in the Mental Health score over time (estimate = −0.28, 95% CI = −0.48 to 0.07, P = .14; time interaction estimate = 0.02, 95% CI = −0.04 to 0.08, P = .43). The strongest positive predictors of Mental Health over time were enrollment marital status (estimate = 1.28, 95% CI = 0.86 to 1.70, P < .001), the Mental Health score at enrollment (estimate = 0.43, 95% CI = 0.41 to 0.44, P < .001; time interaction estimate = −0.01, 95% CI = −0.02 to −0.01, P < .001), Bodily Pain scores at enrollment where higher scores reflect less pain (estimate = 0.06, 95% CI = 0.05 to 0.07, P < .001), enrollment Vitality score (estimate = 0.04, 95% CI = 0.03 to 0.05, P < .001), and enrollment Physical Functioning score (estimate = 0.03, 95% CI = 0.02 to 0.04, P < .001). The strongest negative predictors of Mental Health scores over time were current smoking at enrollment (estimate = −1.42, 95% CI = −1.98 to −0.86, P < .001), diabetes (estimate = −0.78, 95% CI = −1.39 to −0.17, P = .012; time interaction estimate = −0.29, 95% CI = −0.42 to −0.16, P < .001), the AUA Symptom Index (estimate = −0.14, 95% CI = −0.17 to −0.11, P < .001), and the Sexual Activity Score (estimate = −0.037, 95% CI = −0.045 to −0.030, P < .001).
Finasteride was not associated with change in the Vitality score over time (estimate = −0.05, 95% CI = −0.38 to 0.28, P = .807; time interaction estimate = 0.01, 95% CI = −0.07 to 0.09, P = .083). Positive predictors of change in the Vitality score were black race (estimate = 1.86, 95% CI = 0.96 to 2.76, P < .001), marital status (estimate = 1.11, 95% CI = 0.60 to 1.63, P < .001; interaction with time estimate = 0.23, 95% CI = 0.11 to 0.35, P < .001), the Vitality score at enrollment (estimate = 0.57, 95% CI = 0.56 to 0.59, P < .001; interaction with time estimate = −0.007, 95% CI = −0.010 to 0.003, P < .001), enrollment BMI (estimate = −0.09, 95% CI = −0.13 to −0.05, P < .001; interaction with time estimate = −0.02, 95% CI = −0.03 to −0.01, P < .001), enrollment Physical Function score (estimate = 0.07, 95% CI = 0.06 to 0.08, P < .001), enrollment Bodily Pain score (estimate = 0.06, 95% CI = 0.05 to 0.07, P < .001), MET-hours of walking at enrollment (estimate = 0.04, 95% CI = 0.03 to 0.05, P < .001). Negative predictors of change in the Vitality score were current smoking at enrollment (estimate = −1.81, 95% CI = −2.45 to −1.18, P < .001; time interaction estimate = −0.27, 95% CI = −0.42 to −0.12, P < .001); enrollment leg pain (estimate = −1.61, 95% CI = −2.08 to −1.13, P < .001); diabetes at enrollment (estimate = −1.58, 95% CI = −2.32 to −0.83, P < .001; time interaction estimate = −0.63, 95% CI = −0.81 to −0.46, P < .001); hypertension (estimate = −0.82, 95% CI = −1.21 to −0.44, P < .001); AUA Symptom Index score (estimate = −0.24, 95% CI = −0.27 to −0.20, P < .001); enrollment Mental Health score (estimate = −0.07, 95% CI = −0.09 to −0.06, P < .001); and enrollment Sexual Activity Scale score (estimate = −0.04, 95% CI = −0.05 to −0.03, P < .001).
Cross-sectional Perspective: Physical Function
. Because the unfamiliar metrics and model complexity make it difficult to understand the practical meaning of estimates from mixed-effect models, in we provide additional interpretations of these data. To understand how interactions with time modified the effects of covariates, we chose three time points: T
= 0 (ie, 6 months after randomization), T
= 3.5 years after baseline (ie, 4 years after randomization), and T
= 6.5 years after baseline (ie, 7 years after randomization). At each time point, we combined the respective covariate and interaction effects to yield a cross-sectional perspective. We calculated our estimates to reflect clinically meaningful differences. For example, a clinically meaningful difference is presence or absence of a comorbidity (ie, diabetes, hypertension, or congestive heart failure). For some linear covariates (eg, BMI or age), a 1-unit difference is clinically meaningful. However, a 1-point difference on an SF–36 scale is clinically negligible. A 10-point difference has been previously documented as clinically meaningful for 0–100 scales (24
). Therefore, estimates in for the SF–36 scales are multiplied by 10 to display the effect of a 10-point difference.
Cross-sectional perspective: SF–36 Physical Function outcome (N = 16 077)*
Finasteride was not statistically significantly associated with the Physical Functioning score at any of the three time points (). Having diabetes at enrollment was associated with an 8.06-point decrease in the Physical Functioning score at 6.5 years after randomization (95% CI = −9.37 to −6.75, P < .001). Being on study for 7 years was associated with 7.77-point decrease in the Physical Functioning score by the end of the study (95% CI = −8.87 to −6.67, P < .001). Being a current smoker at enrollment was associated with a 5.53-point decrease in the Physical Functioning score at 6.5 years after randomization (95% CI = −6.67 to −0.53, P < .001). Congestive heart failure, leg pain, and hypertension at enrollment were associated with 4.01-point (95% CI = −8.40 to 0.38, P = .073), 3.54-point (95% CI = −4.36 to −2.72, P < .001), and 2.26-point (95% CI = −2.93 to −1.59, P < .001) decreases, respectively, in Physical Functioning scores at 6.5 years after randomization.
Variability among individual participants in Physical Functioning score was high: the SD of individual intercepts (estimated by maximum likelihood) was approximately 8.92. This is the amount by which a typical man’s expected Physical Functioning score would differ from the average Physical Functioning score for all similar men (those sharing the same treatment assignment, enrollment, Physical Functioning level, and other covariate values). The estimated difference between any two randomly chosen men with the same covariate profile is 12.62 points in Physical Functioning score. Even some of the larger effects (such as the −5.53-point effect at 6.5 years after randomization of being a current smoker at enrollment) are small compared with this systematic inter-individual variability. To expand on the importance of considering inter-individual variation in clinical trial results (25
) and to provide context for interpreting the clinical importance of the covariates associated with self-reported physical function, we include , which compares the relative expected impacts of covariates on Physical Functioning scores at 6 months after randomization (T
= 0) with that at 7 years after randomization (T
= 6.5 years).
Figure 2. Relative levels of treatment, covariate, and individual effects on change in Physical Functioning scores. This figure illustrates the expected impact, in Physical Functioning points (adjusted for Physical Functioning level at T = 0 [6 months after randomization]), (more ...)
In addition, in we also present the semistandardized effect sizes for dichotomous covariates as a ratio of the estimate to the SD of the random intercept term (this SD scales systematic variability across individuals who share the same predictor profile). This effect ratio is an estimate of how large the average cross-condition effect is compared to the amount a typical patient systematically differs from the within-condition (those who shared a predictor profile) average. The magnitudes of the ratios for the effect of finasteride on Physical Functioning scores over time were very small (effect ratios of 0.01, −0.02, and −0.05 at the three time points), reflecting the clinically unimportant effect of finasteride on Physical Functioning scores. By contrast, 6.5 years after randomization, the effect ratio for having diabetes at study enrollment was −0.90, which is a clinically large effect on Physical Functioning scores (26
). Current smoking at enrollment had a moderate effect on Physical Functioning scores, with an effect ratio of −0.62 at 6.5 years after randomization. Both of these variables were more important over time than having other comorbidities at study enrollment with moderate to small effect sizes over time, such as leg pain (effect ratio = −0.45).