Benjamin R. Stockton

Evaluation of Prostate Cancer Risk Calculators for Shared Decision Making Across Diverse Urology Practices in Michigan

OBJECTIVE To compare the predictive performance of a logistic regression model developed with contemporary data from a diverse group of urology practices to that of the Prostate Cancer Prevention Trial (PCPT) Risk Calculator version 2.0. MATERIALS AND METHODS With data from all first-time prostate biopsies performed between January 2012 and March 2015 across the Michigan Urological Surgery Improvement Collaborative (MUSIC), we developed a multinomial logistic regression model to predict the likelihood of finding high-grade cancer (Gleason score ≥7), low-grade cancer (Gleason score ≤6), or no cancer on prostate biopsy. The performance of the MUSIC model was evaluated in out-of-sample data using 10-fold cross-validation. Discrimination and calibration statistics were used to compare the performance of the MUSIC model to that of the PCPT risk calculator in the MUSIC cohort. RESULTS Of the 11,809 biopsies included, 4289 (36.3%) revealed high-grade cancer; 2027 (17.2%) revealed low-grade cancer; and the remaining 5493 (46.5%) were negative. In the MUSIC model, prostate-specific antigen level, rectal examination findings, age, race, and family history of prostate cancer were significant predictors of finding high-grade cancer on biopsy. The 2 models, based on similar predictors, had comparable discrimination (multiclass area under the curve = 0.63 for the MUSIC model and 0.62 for the PCPT calculator). Calibration analyses demonstrated that the MUSIC model more accurately predicted observed outcomes, whereas the PCPT risk calculator substantively overestimated the likelihood of finding no cancer while underestimating the risk of high-grade cancer in this population. CONCLUSION The PCPT risk calculator may not be a good predictor of individual biopsy outcomes for patients seen in contemporary urology practices.

askMUSIC: Leveraging a Clinical Registry to Develop a New Machine Learning Model to Inform Patients of Prostate Cancer Treatments Chosen by Similar Men

BACKGROUND Clinical registries provide physicians with a means for making data-driven decisions but few opportunities exist for patients to interact with registry data to help make decisions. OBJECTIVE We sought to develop a web-based system that uses a prostate cancer (CaP) registry to provide newly diagnosed men with a platform to view predicted treatment decisions based on patients with similar characteristics. DESIGN, SETTING, AND PARTICIPANTS The Michigan Urological Surgery Improvement Collaborative (MUSIC) is a quality improvement consortium of urology practices that maintains a prospective registry of men with CaP. We used registry data from 45 MUSIC urology practices from 2015 to 2017 to develop and validate a random forest machine learning model. After fitting the random forest model to a derivation cohort consisting of a random two-thirds sample of patients after stratifying by practice location, we evaluated the model performance in a validation cohort consisting of the remaining one-third of patients using a multiclass area under the curve (AUC) measure and calibration plots. RESULTS AND LIMITATIONS We identified 7543 men diagnosed with CaP, of whom 45% underwent radical prostatectomy, 30% surveillance, 17% radiation therapy, 5.6% androgen deprivation, and 1.8% watchful waiting. The personalized prediction for patients in the validation cohort was highly accurate (AUC 0.81). CONCLUSIONS Using clinical registry data and machine learning methods, we created a web-based platform for patients that generates accurate predictions for most CaP treatments. PATIENT SUMMARY We have developed and tested a tool to help men newly diagnosed with prostate cancer to view predicted treatment decisions based on similar patients from our registry. We have made this tool available online for patients to use.

PD58-07 MUSIC OCTAVE – COMPOSITE MEASURES TO ASSESS SURGEON PERFORMANCE FOR ROBOTIC PROSTATECTOMY

evaluated the type and quantity of opioids prescribed, standardized to morphine milligram equivalents (MME). Finally, we quantified surgeonspecific variation in MMEs prescribed for surgeons with 10 or more patients in the cohort, and at least 5 filling an opioid prescription postoperatively. RESULTS: We identified 25,102 men who received a vasectomy during the study interval. Among this group, 10,442 (41.6%) patients filled an opioid prescription after surgery. Hydrocodone was the most common medication, comprising 66.7% of filled prescriptions. The median number of MMEs prescribed was 112.5 [IQR 82.5-150]; equivalent to twenty-three, 5 mg hydrocodone tablets per prescription [IQR 16.5-30 tablets/ prescription]. Across 360 surgeons meeting criteria for surgeon-specific analysis, the average number of MMEs prescribed after vasectomy varied substantially (range: 29.2-390 MMEs (p<0.001); corresponding to a range of six to seventy-eight, 5 mg hydrocodone tablets per prescription (Figure). CONCLUSIONS: Less than half of men fill an opioid prescription following vasectomy, indicating that non-opioid pain strategies may be sufficient for most patients. Nonetheless, surgeon-specific analyses revealed a 13-fold difference in the average quantity of opioids supplied. Because patient necessity is unlikely to entirely explain this variability, efforts to reduce excess opioid prescribing after vasectomy are warranted.

PNFBA-06 ASKMUSIC©: LEVERAGING A CLINICAL REGISTRY TO INFORM PATIENTS

INTRODUCTION AND OBJECTIVES: Clinical registries increasingly provide physicians with a means for making data-driven decisions; however, few opportunities exist for patients to interact with registry data to support their own decisions. Herein, we report a webbased system that uses a prostate cancer (CaP) registry to provide newly-diagnosed men with a platform to understand treatment decisions made by others with similar characteristics. METHODS: The Michigan Urological Surgery Improvement Collaborative (MUSIC) is a consortium of 43 diverse urology practices that maintains a prospective registry of men with CaP. We developed a patient-facing, web-based tool that uses self-reported information and registry data to generate a personalized prediction of the likelihood of receiving a given treatment for CaP (Figure 1). The treatment predictions rely on registry data from 1/2011 to 12/2015 and were generated using a random forest machine learning model derived in a 2/3 random sample of the data. Predictive performance was measured in this derivation cohort (using 10-fold cross validation) and verified in the remaining data using multinomial area-under-the-curve (AUC) and calibration plots. RESULTS: Between the included dates, 11,456 men were diagnosed with CaP and 44.7% underwent prostatectomy, 22.0% surveillance, 19.5% radiation (RT), 8.8% androgen deprivation, and 3.6% watchful waiting (WW). The predictive model demonstrated consistent discrimination between treatments in the derivation and validation cohorts (AUCs 0.762 and 0.744, respectively). The predicted likelihood of receiving a given treatment was accurate for the most common treatment types in the derivation and validation cohorts although the model overpredicted the likelihood of receiving WW in both cohorts and RT in the validation cohort (Figure 2). CONCLUSIONS: With MUSIC registry data and machine learning methods, we were able to create a tool, designed for patients, that generates accurate predictions for most CaP treatments. As a newly diagnosed man considers treatment options, this tool will provide insight into choices made by similar men. Source of Funding: Blue Cross and Blue Shield of Michigan and grant 1T32-CA180984 from the National Cancer Institute.

Drainage Characteristics and Differential Function of the Horseshoe Kidney: What Is Typical?

PURPOSE To describe the drainage and differential function of the horseshoe kidney. PATIENTS AND METHODS A retrospective review of mercaptoacetyltriglycine (MAG3) nuclear renograms from 1991 to 2007 was performed. Nineteen patients with horseshoe kidney who had not undergone previous renal surgery were identified. All studies were reviewed, and patient information was gathered. Drainage characteristics and differential function were determined. RESULTS Nineteen primary MAG3 studies were found, representing 38 evaluable renal units. For these units altogether, the median t1/2 was 11 minutes. A t1/2 of 10 minutes or less was found in 18 units (47%), while 12 units (32%) showed a t1/2 longer than 20 minutes. The relative function difference mean was 24.6%. The number of patients with a relative function difference greater than 10% was 11 (57%), and those with a relative function difference greater than 20% was 7 (37%). CONCLUSION Horseshoe kidney is associated with fairly high rates of poor renal drainage and elevated differential function.