BACKGROUND:Guidelines endorse specific life expectancy (LE) cutoffs for triage of definitive local treatment of prostate cancer, but the lack of validated, prostate cancer-specific LE prediction tools limits incorporation of LE in management decisions. OBJECTIVE:We sought to provide long-term LE predictions in older men using the Prostate Cancer Comorbidity Index (PCCI), a validated tool for prediction of other-cause mortality based on age and presence and severity of major comorbidities, in nationally representative cohorts of US prostate cancer patients. DESIGN, SETTING, SUBJECTS:We performed an observational study of 916,890 men in the SEER-Medicare database and 243,928 men in the VA diagnosed with clinically localized prostate cancer between 2000 and 2019. METHODS:PCCI scores were calculated using ICD-9 and ICD-10 codes. Kaplan Meier and multivariable Cox proportional hazards analysis were used to measure overall survival by age-adjusted PCCI score groups. RESULTS:Median follow up was 11 years in both cohorts. In SEER-Medicare, men with PCCI scores of 1-2, 3-4, 5-6, 7-9, and 10+ had median estimated LE (95%CI) of 16.6 (16.5-16.7), 12.2 (12.1-12.2), 10.7 (10.6-10.7), 9.3 (9.2-9.3), and 5.4 (5.3-5.4) years, respectively. In the VA, men with the same PCCI scores had estimated median LE (95%CI) of 16.1 (16.0-16.3), 11.2 (11.1-11.4), 9.1 (8.9-9.2), 7.5 (7.4-7.7), and 5.2 (5.0-5.3) years, respectively. CONCLUSIONS:The PCCI robustly predicts long-term longevity in SEER-Medicare and VA populations, with similar LE estimates in each. LE prediction tables with their accompanying variability estimates can help clinicians implement guidelines endorsed LE cutoffs for management of older men with prostate cancer.

ObjectivesProstate cancer is the most common cancer among men in the United States. This study examines factors associated with active surveillance (AS) uptake, timing of treatment decision making, and whether timing affects quality of life.MethodsWe used data from a population-based observational study of 512 patients aged 40–79 diagnosed with low-risk prostate cancer from 2016 to 2022. Factors associated with AS receipt and treatment decision-making trajectories were assessed using robust Poisson regression models. Patient-reported satisfaction with treatment and PROMIS measures were analyzed using Poisson and linear regression.ResultsAS uptake was 70.9% in the overall sample, higher among non-Hispanic (NH) White (76.6%) and NH Asian American/Pacific Islander (75.0%) patients than among NH Black (64.1%) and Hispanic (57.3%) patients. In adjusted models, higher education was associated with greater AS uptake (prevalence ratio (PR): 1.35; 95% CI: 1.02–1.78), and Hispanic patients were less likely to choose AS (PR: 0.80; 95% CI: 0.65–0.99). Older patients were more likely to delay disease management decisions (60–69 years: PR: 1.47; 95% CI: 1.09–1.98; 70 + years: PR: 1.67; 95% CI: 1.17–2.37), while partnered patients were less likely to delay (PR: 0.69; 95% CI: 0.54–0.88). Late deciders reported lower satisfaction with treatment decisions (PR: 0.92; 95% CI: 0.85–0.99) and higher anxiety scores at follow-up (coef: 1.86; 95% CI: 0.15–3.57).ConclusionsHispanic patients, those who completed less education, and older patients face barriers to timely decisions about disease management. Partner involvement supports earlier decisions, while later decision was linked to lower satisfaction and higher anxiety.

4554 Background: Renal cell carcinoma (RCC) remains a significant cause of cancer mortality in the United States, with poor outcomes for advanced-stage disease and limited tools for early detection. Tumor-specific antibodies, known to develop early in other solid tumors, offer biomarker opportunities for early RCC detection. This study aimed to leverage Serum Epitope Repertoire Analysis (SERA), an advanced platform for profiling B cell epitopes, to develop a machine learning-based classifier capable of distinguishing patients with RCC from those with benign renal masses and from individuals without known renal neoplasms. Methods: We obtained 564 serum or plasma samples from 1) 260 patients with pathologically confirmed RCC, spanning all stages; 2) 21 patients with benign renal masses (predominantly oncocytoma and angiomyolipoma); and 3) 283 age-matched non-RCC controls (self-reported healthy donors). The SERA platform uses a library of 8 billion unique 12-mer peptides, each expressed on a DNA-barcoded E. coli strain. Number and type of peptides bound by an antibody are identified by next-generation sequencing, enabling comprehensive profiling of B cell epitopes. Using machine learning, a classifier was trained on a subset of 178 samples (88 RCC and 90 healthy controls) to predict the presence of RCC in the validation cohort of 386 samples (172 RCC, 21 benign renal masses, and 193 healthy controls). The area under the receiver operating characteristic curve (AUC) served to evaluate the classifier's performance, overall and stratified by RCC stage. Results: Using the SERA platform, 26.4 million potential amino acid motifs were scored based on enrichment in RCC versus controls, yielding 7,244 motifs that met the predefined thresholds for inclusion in the classifier. These features were used to train a 10,000-tree random classification forest. In validation, the model achieved an AUC of 0.76 (95% confidence interval [CI]: 0.72 - 0.81), and scores were not significantly different (Mann-Whitney U test, alpha = 0.05) in the benign renal lesion control samples vs. healthy controls. Performance was consistent across both early- and late-stage RCC, with an AUC of 0.78 (95% CI: 0.70–0.85) for stage 1, 0.72 (95% CI: 0.49–0.95) for stage 2, 0.81 (95% CI: 0.70–0.92) for stage 3, and 0.75 (95% CI: 0.68–0.81) for stage 4 RCC, each compared to controls, demonstrating robust detection across all disease stages. Conclusions: Our findings suggest that a non-invasive SERA-based classifier can distinguish RCC from benign renal masses and healthy controls, with consistent performance across all stages of RCC. The robust detection of early-stage RCC underscores the potential of this approach to enhance early diagnosis of RCC and to guide clinical management while obviating the need for renal mass biopsy. Future studies will focus on refining the classifier and validating its performance in larger, multi-institutional cohorts.

Developing accurate clinical prediction models is often bottlenecked by the difficulty of deriving meaningful structured features from unstructured EHR notes, a process that traditionally requires manual, unscalable clinical abstraction. In this study, we first established a rigorous patient-level Clinician Feature Generation (CFG) protocol, in which domain experts manually reviewed notes to define and extract nuanced features for a cohort of 147 patients with prostate cancer. As a high-fidelity ground truth, this labor-intensive process provided the blueprint for SNOW (Scalable Note-to-Outcome Workflow), a transparent multi-agent large language model (LLM) system designed to autonomously mimic the iterative reasoning and validation workflow of clinical experts. On 5-year cancer recurrence prediction, SNOW (AUC-ROC 0.767) achieved performance comparable to manual CFG (0.762) and outperformed structured baselines, clinician-guided LLM extraction, and six representational feature generation (RFG) approaches. Once configured, SNOW produced the full patient-level feature table in 12 hours with 5 hours of clinician oversight, reducing human expert effort by approximately 48-fold versus manual CFG. To test scalability where manual CFG is infeasible, we deployed SNOW on an external heart failure with preserved ejection fraction (HFpEF) cohort from MIMIC-IV (n=2,084); without task-specific tuning, SNOW generated prognostic features that outperformed baseline and RFG methods for 30-day (SNOW: 0.851) and 1-year (SNOW: 0.763) mortality prediction. These results demonstrate that a modular LLM agent-based system can scale expert-level feature generation from clinical notes, while enabling interpretable use of unstructured EHR text in outcome prediction and preserving generalizability across a variety of settings and conditions.

Importance:Men with limited life expectancy (LE) have historically been overtreated for prostate cancer despite clear guideline recommendations. With increasing use of active surveillance, it is unclear if overtreatment of men with limited LE has persisted and how overtreatment varies by tumor risk and treatment type. Objective:To determine if rates of overtreatment of men with limited LE have persisted in the active surveillance era and whether overtreatment varies by tumor risk or treatment type. Design, Setting, and Participants:This cohort study included men with clinically localized prostate cancer in the Veterans Affairs health system who received a diagnosis between January 1, 2000, and December 31, 2019. Main Outcomes and Measures:LE was estimated using the validated age-adjusted Prostate Cancer Comorbidity Index (PCCI). Treatment trends among men with limited LE were assessed using a stratified linear and log-linear Poisson regression in aggregate and across PCCI and tumor risk subgroups. Results:The mean (SD) age for the study population of 243 928 men was 66.8 (8.0) years. A total of 50 045 (20.5%) and 11 366 (4.7%) men had an LE of less than 10 years and LE of less than 5 years based on PCCI scores of 5 or greater and 10 or greater, respectively. Among men with an LE of less than 10 years, the proportion of men treated with definitive treatment (surgery or radiotherapy) for low-risk disease decreased from 37.4% to 14.7% (absolute change, -22.7%; 95% CI, -30.0% to -15.4%) but increased for intermediate-risk disease from 37.6% to 59.8% (22.1%; 95% CI, 14.8%-29.4%) from 2000 to 2019, with increases observed for favorable (32.8%-57.8%) unfavorable intermediate-risk disease (46.1%-65.2%). Among men with an LE of less than 10 years who were receiving definitive therapy, the predominant treatment was radiotherapy (78%). Among men with an LE of less than 10 years, use of radiotherapy increased from 31.3% to 44.9% (13.6%; 95% CI, 8.5%-18.7%) for intermediate-risk disease from 2000 to 2019, with increases observed for favorable and unfavorable intermediate-risk disease. Among men with an LE of less than 5 years, the proportion of men treated with definitive treatment for high-risk disease increased from 17.3% to 46.5% (29.3%; 95% CI, 21.9%-36.6%) from 2000 to 2019. Among men with an LE of less than 5 years who were receiving definitive therapy, the predominant treatment was radiotherapy (85%). Among men with an LE of less than 5 years, use of radiotherapy increased from 16.3% to 39.0% (22.6%; 95% CI, 16.5%-28.8%) from 2000 to 2019. Conclusions and Relevance:The results of this cohort study suggest that, in the active surveillance era, overtreatment of men with limited LE and intermediate-risk and high-risk prostate cancer has increased in the VA, mainly with radiotherapy.

BACKGROUND:Inappropriate imaging to stage low-risk prostate cancer is considered low-value care. Determining the effectiveness of a theory-based intervention-Prostate Cancer Imaging Stewardship (PCIS)-to promote guideline-concordant imaging. METHODS:A stepped-wedge, cluster-randomized trial, PCIS, was conducted between March 2018 and March 2021 at 10 Veterans Health Administration medical centers (VAMCs) initially selected for prostate cancer volume, geographic diversity, and willingness to participate. Intervention initiations at sites were randomized in 3-month intervals. We enrolled 61 urology providers who treat prostate cancer at participating sites. Outcomes were assessed among 2302 patients with incident prostate cancer aged 18-85 years. PCIS combines 3 evidence-based provider-focused behavior change strategies: (1) Clinical Reminder Order Check triggered when a provider attempted to order imaging for a patient with prostate-specific antigen < 20 ng/mL, (2) VAMC-level academic detailing at initiation and every 3 months thereafter, and (3) Audit and Feedback for providers to improve their imaging performance. The main outcome was guideline-discordant nuclear medicine bone scan (NMBS) imaging for low-risk prostate cancer patients. RESULTS:NMBS imaging would be consistent with National Comprehensive Cancer Network guidelines in 878 patients (38%) and inconsistent in 1424 patients (62%). Among patients not requiring NMBS, 141/690 (20.4%) received guideline-discordant imaging (ie, NMBS ordered) during Control compared with 109/734 (14.9%) during Intervention (odds ratio [OR] = 0.54, P = .04). Among patients requiring a NMBS, 29 of 425 (6.8%) did not receive one (ie, guideline-discordant imaging) during Control compared with 25 of 453 (5.5%) during the Intervention (OR = 1.36, P = .36). CONCLUSION:PCIS significantly reduced low-value, guideline-discordant NMBS imaging among low-risk prostate cancer patients without negatively affecting necessary imaging for high-risk patients. CLINICAL TRIALS REGISTRATION:NCT03445559.

INTRODUCTION:Good-quality care for patients with a serious illness often requires interdisciplinary expertise. In the urologic perioperative period, this can include urologists and Palliative Care (PC). Our objective is to understand how to improve perioperative coordination between urologists and PC providers in the context of urologic serious illness. MATERIALS AND METHODS:We interviewed 38 providers: urologists (13), PC physicians (12), and clinical team members (13) in phase I of this study. From these interviews, there were 96 examples of interdisciplinary communication that were analyzed using qualitative content analysis with dual review in phase II of this study. RESULTS:Two key themes emerged regarding communication between urology and PC teams. First, effective collaboration is often hindered by logistical challenges, such as surgeons' limited availability due to time spent in surgery and difficulties coordinating in-person meetings. Fostering bidirectional, timely communication through asynchronous communication and structured meetings improves alignment within the clinical team before patient interactions. Second, hierarchical structures within medical teams can discourage open dialogue, with nonsurgeons sometimes feeling hesitant to share input. Promoting mutual respect is essential to creating a more balanced and collaborative environment. Together, these themes highlight the need for systemic changes that support accessibility, respect, and communication in interdisciplinary care. CONCLUSIONS:Future directions include implementing an evidence-based intervention with structures and processes to improve interdisciplinary collaboration among urologists and PC.

INTRODUCTION:Consistent urologic oncology follow-up after radical cystectomy (RC) improves survival. However, there is scarce literature describing postoperative communication. We aimed to identify differences in postoperative communication patterns and healthcare utilization among English-speaking patients (ESPs) and patients with limited English proficiency (LEP) following RC. METHODS:We conducted a single-institution, retrospective cohort study, examining patients who underwent RC for bladder cancer. We used propensity score matching to match 50 ESPs and 50 patients with LEP on age and sex. We abstracted patient demographics, postoperative communication and healthcare utilization within 90 days of surgery. We fit multivariable linear regression to investigate factors associated with postoperative communication frequency. RESULTS:Postoperative communication was common, with 82% of patients placing ≥1 phone call/message. ESPs communicated more than patients with LEP (6.04 vs. 3.80 average calls/messages), though this difference was not statistically significant (P = 0.08). ESPs were more likely to initiate the communication themselves and have postoperative communication result in reassurance from the surgical team (P = 0.03), while patients with LEP were more likely to have a family member communicate on their behalf (P < 0.001) and have postoperative communication result in outpatient evaluation/treatment (P = 0.01). Patients with a neobladder reconstruction placed an increased number of phone calls/messages. There were no differences in postoperative healthcare utilization between the 2 groups. CONCLUSIONS:Postoperative communication is frequent following RC. ESPs communicated nearly twice as often as patients with LEP, suggesting a clinically relevant difference in patient communication following radical cystectomy. Primary language spoken is not associated with differences in postoperative healthcare utilization.

OBJECTIVE:To determine rates of urology follow-up and implementation of stone prevention measures after stone surgery and to assess variation in care delivery within a large, integrated healthcare system. MATERIALS AND METHODS:We used nationwide data from the United States Veterans Health Administration to identify patients who had stone surgery between 2016 and 2018 and who were at higher risk for recurrence. Our cohort included 13,444 Veterans across 90 facilities. We examined the proportion of patients who had a post-operative urology visit or who received a prevention measure (24-hour urine test, serum parathyroid hormone measurement, or prescription of a stone-related medication) within 6 months of stone surgery. We calculated the median odds ratio to quantify facility-level variation in urology care after stone surgery, adjusting for patient- and facility-level characteristics. RESULTS:Within 6 months of stone surgery, 94.2% Veterans had a urology visit, yet only 8.8% completed 24-hour urine testing, 8.4% had a parathyroid hormone measurement, and 31.0% were prescribed a stone-related medication. Implementation of prevention measures varied widely across facilities with the median odds ratio ranging between 1.18 for medication prescriptions and 1.77 for 24-hour urine testing. CONCLUSION:While most patients have a urology visit after stone surgery, stone prevention measures were implemented infrequently and inconsistently for patients at higher risk for recurrence, indicating an opportunity for quality improvement.

INTRODUCTION:Renal ultrasound (US) offers less radiation exposure than computed tomography (CT) for kidney stone surveillance but has lower sensitivity and specificity for nephrolithiasis diagnosis. Additionally, US may overestimate stone size, leading to unnecessary surgical interventions. Evidence on US performance for kidney stone surveillance is variable, making its clinical utility unclear. We aimed to assess US accuracy against CT and identify factors influencing US performance. METHODS:We performed a retrospective review of patients with known nephrolithiasis seen in urology clinic at Stanford who underwent both renal US and CT within 90 days for surveillance from January to December 2022. Patients with spontaneous stone passage or interventions were excluded. Stone characteristics were recorded, and statistical analysis compared the diagnostic accuracy of US and CT. RESULTS:A total of 107 patients and 128 stones were included, with a mean time difference of 25.7 days between US and CT. US sensitivity was 77%, with a positive predictive value (PPV) of 75% for stone detection. The PPV was only 59% for stones >4 mm by CT. Mean stone size was 8.7 mm on US vs. 5.5 mm on CT (p=0.02), with more pronounced overestimation in smaller stones and higher body mass index (BMI) (p<0.05). No significant differences in US performance were found by stone location, laterality, or time between scans. Differences in stone detection (p=0.01) and size (p=0.03) were associated with the individual performing the ultrasound. CONCLUSIONS:US performance is limited compared to CT and is influenced by stone size, BMI, and sonographer. Overestimation by US may lead to unnecessary interventions in up to 40% of patients with stones >4 mm.

Over the past decade, the use of machine learning (ML) models in healthcare applications has rapidly increased. Despite high performance, modern ML models do not always capture patterns the end user requires. For example, a model may predict a non-monotonically decreasing relationship between cancer stage and survival, keeping all other features fixed. In this paper, we present a reproducible framework for investigating this misalignment between model behavior and clinical experiential learning, focusing on the effects of underspecification of modern ML pipelines. In a prostate cancer outcome prediction case study, we first identify and address these inconsistencies by incorporating clinical knowledge, collected by a survey, via constraints into the ML model, and subsequently analyze the impact on model performance and behavior across degrees of underspecification. The approach shows that aligning the ML model with clinical experiential learning is possible without compromising performance. Motivated by recent literature in generative AI, we further examine the feasibility of a feedback-driven alignment approach in non-generative AI clinical risk prediction models through a randomized experiment with clinicians. Our findings illustrate that, by eliciting clinicians' model preferences using our proposed methodology, the larger the difference in how the constrained and unconstrained models make predictions for a patient, the more apparent the difference is in clinical interpretation.

CONTEXT:Many urologic serious illnesses are treated with surgical procedures, which may put patients at a further risk of diminished quality of life. OBJECTIVE:To understand stakeholder perceptions on integrating perioperative Palliative Care (PC) for patients with serious urologic illness. METHODS:We conducted semi-structured interviews and team-based thematic analysis to consensus with a dual review. Purposefully sampled urologists, palliative care physicians, and clinical team members at fourteen geographically distributed Veteran Health Administration sites were interviewed. RESULTS:We identified one general overall theme, to "change culture" so that PC is not a "last resort," and three opportunities along the perioperative continuum for integrating urology and PC. Opportunity 1: Utilizing telehealth and team member role expansion when discussing the initial diagnosis, with surgery as a potential treatment option, allows for multiple conversations "so they're not rushed in 15 minutes to mentally deal with the new diagnosis." Opportunity 2: Creating a process to ensure goal of care conversations occur, since "urologic procedures can have complications that significantly impact quality of life," which "would require changing how our workflow is structured." Opportunity 3: During the preoperative visits, interdisciplinary input and evaluation of the patient prior to surgery allows the patient to "have a sort of joint meeting with us and the urologist." This represented the last point in time to de-escalate and offer nonsurgical options prior to surgery. CONCLUSIONS:The study informs future interventions to improve the quality of surgical care by integrating PC with urology in a unified workflow.