Session Item

Despite the high survival rates of localized prostate cancer (>98%), newly diagnosed patients face the challenge of choosing the most appropriate treatment from the main options as each option has different side effects. Previous studies have shown that irrespective of type of treatment, erectile dysfunction (ED) is a common side effect of prostate cancer. Many studies have explored the factors affecting ED after prostate cancer diagnosis in the population. However, there is limited individual-based research on predicting the risk of developing ED before the start of treatment. Understanding the risk of side effects before treatment, may improve health-related quality of life (HRQOL) outcomes and reduce decisional regret. Providing personalized prediction models will aid to identify patients who are at risk of developing ED and eventually facilitate informed decision making. In this study, we aimed to predict ED at 1-year and 2-year post-diagnosis. 

We used a subset of the ProZIB dataset which was collected by the Netherlands Comprehensive Cancer Organization (Integraal Kankercentrum Nederland; IKNL) from 69 Dutch hospitals. The subset contained information of 964 localized prostate cancer cases for model training and external validation, where the external validation set contained patients from a subset of hospitals (a split by location) included in the dataset. We developed two models with a logistic regression algorithm coupled with Recursive Feature Elimination (RFE) for two-time points: 1-year and 2-year post-diagnosis. These models were based on patient demographics, clinical data, and patient-reported outcomes (PROMs) measured at diagnosis. Both models were validated using a TRIPOD type 3 external validation methodology.

As illustrated in Figure 1, 46% and 47% of patients reported not being able to have an erection 1 year and 2 years post-diagnosis respectively. The remaining (54% and 53%) could have an erection to some extent. The 1-year model predicted ED based on 10 variables with 77.7% accuracy, 89.9% sensitivity and 66.8% specificity. The second model predicted the 2-year outcome based on 9 variables and with 73.9% accuracy, 86.4% sensitivity and 62.3% specificity.  The AUC values in the training set were slightly higher than models in the validation set. The training set AUCs were 0.86 and 0.84 for 1 year and 2 years post-diagnosis respectively, and the validation AUCs were 0.84 and 0.81 for two timepoints (Figure 2).

We identified treatment group, pretreatment frequency and quality of erections, and ISUP group, as the most important predictors of posttreatment ED. Several variables overlapped between the two models, but some were different.

We successfully developed and validated two models to predict ED in localized prostate cancer. To tailor the treatment and support informed decision-making, these models have been incorporated in a personalized patient decision aid (PDA) that is currently being clinically evaluated for effectiveness.