Prediction of Erectile Function Following Treatment for Prostate Cancer. Part 3
Statistical Analysis
Having functional erections suitable for intercourse was defined as the patient selecting the response option of “firm enough for intercourse” to the EPIC-26 question, “How would you describe the usual quality of your erections during the last 4 weeks?” (other responses indicated erectile dysfunction). Erectile function 2 years after treatment was modeled separately, according to planned treatment, using logistic regression. The pretreatment patient and disease characteristics as well as planned treatment details considered are summarized in eTable 1.
Multivariable model development used a backward elimination selection procedure with 2-sided α = .05. Model selection was internally validated using bootstrap resampling (500 resamples), and bootstrap estimates of parameter estimates, standard errors, pointwise 95% confidence intervals for model-predicted probabilities, and area under the receiver operating characteristic curve (AUC) were also obtained for each final model. Individual predicted probabilities of functional erections at 2 years were calculated using the inverse logistic function {exp[X′β]/[1 + exp(X′β)]}, where X′β is the sum with X representing individual characteristics observed and β representing the associated log odds ratios for the individual characteristics estimated from the model.
The omission of 2-year nonrespondents from model development assumes data are missing completely at random. A sensitivity analysis assessed the effect of this assumption by refitting each final model using a model weighted for inverse probability of response; the probability of response was estimated using multivariable logistic regression including factors associated with nonresponse (education level, number of comorbid conditions, race, and pretreatment sexual functioning). This approach had little effect on the estimates, and results were not reported.
Canadian pharmacy viagra super force
Use of medications or devices to assist erection function as measured by patient self-report at 2 years was summarized overall and in detail among the subset of 694 men who were potent (ie, reported functional erections) before treatment, excluding patients with implanted erectile aid devices.
All analyses were performed using SAS version 9.1 (SAS Institute Inc, Cary, North Carolina).
External Validation
The community-based Cancer of the Prostate Strategic Urologic Research Endeavor the CaPSURE cohort registry served as an external validation cohort for the developed models. Men in the CaPSURE cohort reported HRQOL at baseline and every 6 months in follow-up; sexual function and bother (severity and impact of patient-reported erectile dysfunction) were determined from the UCLA Prostate Cancer Index (UCLA-PCI), the instrument from which the EPIC-26 was previously derived and from which it retained 6 items. Characteristics of the CaPSURE cohort have been previously described.
Of the 1913 CaPSURE patients who completed pretreatment and 2 years posttreatment evaluation of sexual HRQOL using the UCLA-PCI, 1655 had data for all available model covariates available for validation. The PROSTQA model–predicted probability of 2-year erectile function was computed for each CaPSURE patient (from the inverse logistic function of the final model equations) and compared with his reported (actual) 2-year erectile function (the definition of erectile function used in the CaPSURE validation was the same as that in the PROSTQA cohort). Validation was assessed by AUC from fitting univariable logistic regression of reported 2-year erectile function on model-predicted probability, and calibration was assessed by examining the average model-predicted probability vs observed proportion of men reporting functional erections at 2 years, which is summarized overall and according to quintiles of the distribution of model-predicted probabilities.