The Relationship Between Erection and Satisfaction in ED
The Relationship Between Erection and Satisfaction in ED
The data set came from a multicenter (United States), parallel-group, randomized (1:1) 10-week double-blind, placebo-controlled (DBPC) study of sildenafil with a 6-week open-label sildenafil extension; details of the patients and methods have been published previously. Written informed consent was obtained from all study participants, and the study complied with the Declaration of Helsinki, International Conference on Harmonisation Good Clinical Practice Guidelines and all local regulatory requirements.
Included were men aged ≥18 years with ED (defined as a score of ≤25 out of 30 on the erectile function domain of the International Index of Erectile Function) who had been in a stable sexual relationship for ≥6 months and had previously used ≤6 doses in total of sildenafil or any other PDE5 inhibitor and no dose within the previous 6 months. If patients were previously taking any other commercially available drug or non-drug treatment for ED (that is, intraurethral agents, prostheses, injection therapy, topical applications, herbal or alternative medications, or vacuum-assisted erection devices), such treatments must have been terminated at or before the screening visit and must not have been taken at any time during the study.
Flexible-dose (50 or 100 mg) sildenafil or placebo (DBPC phase) or sildenafil (open-label phase) was taken as needed for sexual activity but not more than once daily, and sexual activity was to be attempted at least once weekly. At the beginning of each phase, the dose for all men was 50 mg. After 2 weeks, the dose could be adjusted up to 100 mg based on tolerability and efficacy.
At baseline, week 4, week 10 (end of DBPC treatment) and week 16 (end of open-label treatment), each patient completed the SEX-Q, which consists of the erection domain (six items assessing erection quality (hardness, duration), worry/confidence about achieving an erection and satisfaction with the resulting erection), individual satisfaction domain (three items assessing satisfaction with sexual desire, satisfaction with overall sexual activity and pleasure from sexual activity) and couple satisfaction domain (three items assessing confidence in ability to satisfy partner, frequency of mutual satisfaction and satisfaction with ability to control ejaculation timing). The range of scores on each domain was from 0 (least favorable) to 100 (most favorable).
All analyses were performed in SAS (Version 9.2. SAS Institute, Cary, NC, USA) with the CALIS procedure, which estimates parameters and tests the appropriateness of a structural equation model using covariance analysis. Analyses were based on SEX-Q data and were conducted on data from the modified intent-to-treat population, defined as men who took at least one dose of study medication and provided sufficient efficacy data for at least one efficacy analysis.
A mediation model was used to determine the contribution of the direct effect of treatment vs the indirect effect of treatment via erection (SEX-Q erection domain) on individual satisfaction and couple satisfaction (SEX-Q individual satisfaction and couple satisfaction domains). Use of a standard mediation model on these data would look at the direct and indirect effects of treatment (sildenafil vs placebo) on erection, individual satisfaction and couple satisfaction at one assessment time or for the averaged data over the double-blind phase of the study.
In contrast, the unified model constructed for this study used repeated components, modeling the direct and indirect effects of treatment (sildenafil vs placebo) simultaneously at two separate assessment times during the double-blind phase of the study, but also included data from the open-label phase of the study (Figure 1). Thus, the unified model was intended to incorporate and unite all available information over both phases.
(Enlarge Image)
Figure 1.
Hypothesized unified longitudinal mediation model for determining the relationship between the independent variable (treatment: sildenafil vs placebo), the mediator variable (erection (Sexual Experience Questionnaire (SEX-Q) erection domain)), and the dependent variables (individual satisfaction (SEX-Q individual satisfaction domain) and couple satisfaction (SEX-Q couple satisfaction domain)). The uncorrected covariance matrix is augmented by an additional constant variable with the value 1 (constant intercept) in order to model the open-label phase of the trial, when all patients were on sildenafil. In order to make the figure more clear, pairwise covariances between disturbances are not shown for each domain over time as represented by d11, d21 and d31 for individual satisfaction; d12, d22 and d32 for couple satisfaction; and d13, d23 and d33 for erection.
This model is advanced in comparison with standard mediation modeling by the inclusion of not only data to model the repeated nature of measurements, but also data from the second treatment phase (the open-label phase). If the data were limited to those from the two DBPC assessment times, the treatment variable would affect both DBPC segments of the model in the same way; covariances between disturbance terms for erection, individual satisfaction and couple satisfaction could be used to account for the repeated nature of the data. For example, the disturbance term (or residual) for individual satisfaction at week 4 represents the portion of the total variance in individual satisfaction at week 4, which is not explained by the postulated model. Because data for a patient were collected repeatedly, it cannot be assumed that those disturbances are independent. As a result, disturbance or error terms of the same variable at different times are expected to covary with each other (hence the phrase 'covariance between disturbance terms'). For instance, the disturbance term for individual satisfaction at week 4 covaries with the disturbance term for individual satisfaction at week 10 and the disturbance term for individual satisfaction at week 16.
However, to introduce the constant factor that is needed to model the open-label phase (because this phase lacks a placebo arm), we used a mean structure model (as opposed to the covariance structure model routinely used in mediation modeling), along with an uncorrected (for the mean) covariance matrix of the data set, augmented by an additional constant (intercept) variable with the value 1. Although this approach allowed inclusion of all available data and accounted for the longitudinal interrelationship between all-time points as a part of one unified longitudinal mediation model, mediation effects per se at week 16 were not estimable because of the lack of a comparator to the active treatment arm.
The Bentler Comparative Fit index was used to assess the fit of the mediation model. A value >0.9 indicates an adequate fit. To assess statistical significance of the results for the estimated percentages of total, direct and indirect effects, bootstrap simulations were performed to assess inferences. For this method, repeated sampling with replacement is undertaken until the same sample size is obtained as in the original data set, and this process is executed a total of 50 000 times.
Using the mediation model, percentages for direct and indirect effects of treatment on the individual satisfaction and couple satisfaction domains were calculated. That is, of the total effect that sildenafil treatment (vs placebo) has on the individual satisfaction and couple satisfaction domains, the percentage (or proportion) that is indirect through the erection domain and the percentage that is direct through other factors was calculated. Corresponding 95% confidence intervals (CIs) were estimated by using bootstrap simulations.
Patients and Methods
Data Set
The data set came from a multicenter (United States), parallel-group, randomized (1:1) 10-week double-blind, placebo-controlled (DBPC) study of sildenafil with a 6-week open-label sildenafil extension; details of the patients and methods have been published previously. Written informed consent was obtained from all study participants, and the study complied with the Declaration of Helsinki, International Conference on Harmonisation Good Clinical Practice Guidelines and all local regulatory requirements.
Included were men aged ≥18 years with ED (defined as a score of ≤25 out of 30 on the erectile function domain of the International Index of Erectile Function) who had been in a stable sexual relationship for ≥6 months and had previously used ≤6 doses in total of sildenafil or any other PDE5 inhibitor and no dose within the previous 6 months. If patients were previously taking any other commercially available drug or non-drug treatment for ED (that is, intraurethral agents, prostheses, injection therapy, topical applications, herbal or alternative medications, or vacuum-assisted erection devices), such treatments must have been terminated at or before the screening visit and must not have been taken at any time during the study.
Flexible-dose (50 or 100 mg) sildenafil or placebo (DBPC phase) or sildenafil (open-label phase) was taken as needed for sexual activity but not more than once daily, and sexual activity was to be attempted at least once weekly. At the beginning of each phase, the dose for all men was 50 mg. After 2 weeks, the dose could be adjusted up to 100 mg based on tolerability and efficacy.
At baseline, week 4, week 10 (end of DBPC treatment) and week 16 (end of open-label treatment), each patient completed the SEX-Q, which consists of the erection domain (six items assessing erection quality (hardness, duration), worry/confidence about achieving an erection and satisfaction with the resulting erection), individual satisfaction domain (three items assessing satisfaction with sexual desire, satisfaction with overall sexual activity and pleasure from sexual activity) and couple satisfaction domain (three items assessing confidence in ability to satisfy partner, frequency of mutual satisfaction and satisfaction with ability to control ejaculation timing). The range of scores on each domain was from 0 (least favorable) to 100 (most favorable).
Data Modeling and Analyses
All analyses were performed in SAS (Version 9.2. SAS Institute, Cary, NC, USA) with the CALIS procedure, which estimates parameters and tests the appropriateness of a structural equation model using covariance analysis. Analyses were based on SEX-Q data and were conducted on data from the modified intent-to-treat population, defined as men who took at least one dose of study medication and provided sufficient efficacy data for at least one efficacy analysis.
A mediation model was used to determine the contribution of the direct effect of treatment vs the indirect effect of treatment via erection (SEX-Q erection domain) on individual satisfaction and couple satisfaction (SEX-Q individual satisfaction and couple satisfaction domains). Use of a standard mediation model on these data would look at the direct and indirect effects of treatment (sildenafil vs placebo) on erection, individual satisfaction and couple satisfaction at one assessment time or for the averaged data over the double-blind phase of the study.
In contrast, the unified model constructed for this study used repeated components, modeling the direct and indirect effects of treatment (sildenafil vs placebo) simultaneously at two separate assessment times during the double-blind phase of the study, but also included data from the open-label phase of the study (Figure 1). Thus, the unified model was intended to incorporate and unite all available information over both phases.
(Enlarge Image)
Figure 1.
Hypothesized unified longitudinal mediation model for determining the relationship between the independent variable (treatment: sildenafil vs placebo), the mediator variable (erection (Sexual Experience Questionnaire (SEX-Q) erection domain)), and the dependent variables (individual satisfaction (SEX-Q individual satisfaction domain) and couple satisfaction (SEX-Q couple satisfaction domain)). The uncorrected covariance matrix is augmented by an additional constant variable with the value 1 (constant intercept) in order to model the open-label phase of the trial, when all patients were on sildenafil. In order to make the figure more clear, pairwise covariances between disturbances are not shown for each domain over time as represented by d11, d21 and d31 for individual satisfaction; d12, d22 and d32 for couple satisfaction; and d13, d23 and d33 for erection.
This model is advanced in comparison with standard mediation modeling by the inclusion of not only data to model the repeated nature of measurements, but also data from the second treatment phase (the open-label phase). If the data were limited to those from the two DBPC assessment times, the treatment variable would affect both DBPC segments of the model in the same way; covariances between disturbance terms for erection, individual satisfaction and couple satisfaction could be used to account for the repeated nature of the data. For example, the disturbance term (or residual) for individual satisfaction at week 4 represents the portion of the total variance in individual satisfaction at week 4, which is not explained by the postulated model. Because data for a patient were collected repeatedly, it cannot be assumed that those disturbances are independent. As a result, disturbance or error terms of the same variable at different times are expected to covary with each other (hence the phrase 'covariance between disturbance terms'). For instance, the disturbance term for individual satisfaction at week 4 covaries with the disturbance term for individual satisfaction at week 10 and the disturbance term for individual satisfaction at week 16.
However, to introduce the constant factor that is needed to model the open-label phase (because this phase lacks a placebo arm), we used a mean structure model (as opposed to the covariance structure model routinely used in mediation modeling), along with an uncorrected (for the mean) covariance matrix of the data set, augmented by an additional constant (intercept) variable with the value 1. Although this approach allowed inclusion of all available data and accounted for the longitudinal interrelationship between all-time points as a part of one unified longitudinal mediation model, mediation effects per se at week 16 were not estimable because of the lack of a comparator to the active treatment arm.
The Bentler Comparative Fit index was used to assess the fit of the mediation model. A value >0.9 indicates an adequate fit. To assess statistical significance of the results for the estimated percentages of total, direct and indirect effects, bootstrap simulations were performed to assess inferences. For this method, repeated sampling with replacement is undertaken until the same sample size is obtained as in the original data set, and this process is executed a total of 50 000 times.
Main Outcome Measures
Using the mediation model, percentages for direct and indirect effects of treatment on the individual satisfaction and couple satisfaction domains were calculated. That is, of the total effect that sildenafil treatment (vs placebo) has on the individual satisfaction and couple satisfaction domains, the percentage (or proportion) that is indirect through the erection domain and the percentage that is direct through other factors was calculated. Corresponding 95% confidence intervals (CIs) were estimated by using bootstrap simulations.