Citations

ImportanceThere exists substantial heterogeneity in outcomes within T stages for patients with cutaneous squamous cell carcinoma (cSCC).ObjectiveTo determine whether a customized generative pretrained transformer model, trained on a comprehensive dataset with more than 1 trillion parameters and equipped with relevant focused context and retrieval augmented generation (RAG), could excel in aggregating and interpreting vast quantities of data to develop a novel class-based risk stratification system that outperforms the current standards.Design, Setting, and ParticipantsTo build the RAG knowledge base, a systematic review of the literature was conducted that addressed risk factors for poor outcomes in cSCC. Using the RAG-enabled generative pretrained transformer (GPT) model, we developed a novel class-based risk stratification system that assigned point values for risk factors, culminating in a GPT-based prognostication system called the artificial intelligence–derived risk score (AIRIS). The system’s performance was validated on a combined prospective and retrospective cohort of 2379 primary cSCC tumors (1996-2023) with at least 36 months of follow-up, against Brigham and Women’s Hospital (BWH) and American Joint Committee on Cancer Staging Manual, eighth edition (AJCC8) systems in stratifying risk for locoregional recurrence (LR), nodal metastasis (NM), distant metastasis (DM), and disease-specific death (DSD).Main Outcomes and MeasuresPerformance metrics evaluated included distinctiveness, homogeneity, and monotonicity, as defined by the AJCC8, as well as sensitivity, specificity, positive predictive value, negative predictive value, accuracy, the area under the receiver operating characteristic curve, and concordance.ResultsThe median age at diagnosis was 73 (IQR, 64-81) years, with 38.5% female patients and 61.5% male patients. The AIRIS prognostication system demonstrated superior sensitivity across all outcomes (LR, 49.1%; NM, 73.7%; DM, 82.5%; and DSD, 72.2%) and the highest area under the receiver operating characteristic curve values (LR, 0.69; NM, 0.81; DM, 0.85; and DSD, 0.80), indicating significantly enhanced discriminative capability compared with the BWH and AJCC8 systems. While all systems were comparably distinctive, the AIRIS prognostication system consistently demonstrated the lowest proportion of tumors exhibiting poor outcomes in low-risk categories, suggesting its improved homogeneity and monotonicity.Conclusions and RelevanceThe results of this diagnostic study suggest that the AIRIS system outperforms the existing BWH and AJCC8 prognostication systems, potentially providing a more effective tool for predicting poor outcomes in cSCC. This study illustrates the potential of large language models in refining prognostic tools, offering implications for treating patients with cancer.

Abstract Background Real-world first-line maintenance (1LM) treatment data are limited for advanced/metastatic non-small cell lung cancer (a/mNSCLC). Materials and Methods In this electronic health record-derived, deidentified database study, eligible patients (≥18 years; diagnosed with stage III/IV non-small cell lung cancer [June 1, 2017-September 30, 2021]) initiated pembrolizumab-based 1LM after 4-6 cycles of first-line (1L) platinum-based chemotherapy-pembrolizumab ± pemetrexed. Study outcomes were real-world time to next treatment or death (rwTTNTD), overall survival (rwOS), and predictors of outcomes. Results Of 1944 patients analyzed (median follow-up, 12.2 months), 51.9% received 1LM pembrolizumab-pemetrexed and 48.1% pembrolizumab monotherapy. Median rwTTNTD and rwOS were 9.2 (95% CI: 8.5-9.8) and 18.7 (95% CI: 17.7-20.3) months, respectively. In multivariable analyses, factors significantly associated with shorter rwTTNTD included 5%-<10% (hazard ratio [HR], 1.34; 95% CI: 1.17-1.54) or ≥10% (HR, 1.69; 95% CI: 1.42-2.01) weight loss during 1L versus 0% or <5% weight loss. Programmed death-ligand 1 (PD-L1) expression 1%–49% (HR, 0.81; 95% CI: 0.71-0.93) or ≥50% (HR, 0.55; 95% CI: 0.47-0.64) and female sex (HR, 0.85; 95% CI: 0.75-0.95) were significantly associated with longer rwTTNTD. These variables were also significantly associated with shorter (weight loss 5%-<10%: HR, 1.52; 95% CI: 1.30-1.77; ≥10% HR, 2.06; 95% CI: 1.71-2.48) and longer rwOS (PD-L1 expression: 1%-49% HR, 0.84; 95% CI: 0.72-0.98; ≥ 50% HR, 0.57; 95% CI: 0.48-0.68; female sex: HR, 0.81; 95% CI: 0.71-0.92). Conclusions Predictors of real-world clinical outcomes included 1L treatment, weight loss, PD-L1 status, and sex. Poor outcomes persisted despite immunotherapy-based 1LM availability, revealing an unmet need in this population.

AbstractEndoscopic sleeve gastroplasty (ESG) is an effective treatment for obesity but typically is not covered by insurance. It is not known whether patients with financial investment in their endoscopic procedure are more likely to achieve and/or maintain weight loss as compared with those who have no financial buy-in. We aimed to compare treatment adherence and outcomes between patients paying out-of-pocket (OOP) and those who underwent ESG as part of any clinical trial where costs were covered by a study protocol (no payment; NP).Data were collected via retrospective chart review. One hundred sixty-four patients who had an ESG with at least 6 months of follow-up were included. Repeated measures with generalized linear model were used to evaluate weight loss at different time points after ESG and labs values at baseline and 1-year follow-up to assess for comorbidity improvement between cohorts. Compliance was evaluated by comparing exercise adherence rates.The pattern of weight loss and change in laboratory values was not different over time between the OOP group (n = 139) and NP group (n = 25). Patients lost an average of 14% (12.2–15.9) and 12.9% (9.3–16.5) of total body weight over all time points, respectively, in both groups (6, 12 and 24 months). Treatment adherence also did not differ between the groups.Having “skin in the game” by paying for ESG OOP does not correlate with better outcomes or treatment adherence, which further supports broad insurance coverage for this procedure.

Palliative care (PC) has been shown to improve quality of life (QoL), reduce symptom burden, and align care with patient goals in those with serious illnesses. However, its integration into the emergency department (ED), where many patients with advanced illness present, remains underexplored and inconsistently applied. This narrative review evaluates several recent studies on ED-based palliative care, aiming to clarify whether the early benefits in symptom management and QoL are associated with reduced healthcare utilization and cost. A comprehensive literature search identified five new key studies with significant methodological differences, including variations in patient populations, outcome measures, and study designs. Some studies demonstrated short-term symptom relief, yet only one showed significant improvements in QoL at long-term follow-up. For healthcare utilization, still only one of these new studies reported substantial cost savings, underscoring the need for further research to validate financial benefits. We suggest how future work should be designed to explore both clinical and economic outcomes in tandem to establish a clear value proposition for ED-based palliative care. By addressing this lack of cohesion, we can determine the merit of ED-palliative care integrations with the goal of improving patient care and optimizing resource utilization.

5060 Background: Rucaparib significantly improved radiographic progression-free survival (rPFS) in men with BRCA-mutated chemotherapy-naïve metastatic castration-resistant prostate cancer (mCRPC) vs a control arm of physician’s choice of therapy (docetaxel or androgen-receptor pathway inhibitor [ARPI] therapy: abiraterone acetate or enzalutamide) in the randomized, multicenter, open-label, phase 3 TRITON3 (NCT02975934) study. Herein, we conducted an analysis to determine the impact of germline vs somatic mutation status on the efficacy of rucaparib. Methods: Patients were randomized 2:1 to receive rucaparib 600 mg BID or physician’s choice of docetaxel or ARPI following progression while on 1 prior second-generation ARPI in any setting. The primary endpoint was rPFS. Color Health did genetic testing. Treatment-emergent adverse events (TEAEs) were reported for the BRCA subgroup. Results: In the rucaparib arm, 201/270 patients had BRCA mutations, and in the physician’s choice arm, 101/135 patients had BRCA mutations. Of patients with BRCA mutations: in the rucaparib arm, 80/201 (40%) were germline and 116/201 (58%) were somatic with 5/201 (2%) BRCA mutation status unknown, while in the physician’s choice arm, 39/101 (39%) were germline and 48/101 (48%) were somatic with 14/101 (14%) BRCA mutation status unknown. rPFS was significantly improved with rucaparib treatment vs physician’s choice in both the germline and somatic mutation groups (germline: HR, 0.52 [95% CI, 0.32–0.84]; somatic: HR, 0.38 [95% CI, 0.25–0.59]). Incidence rates of TEAEs were similar overall between patients with germline and somatic BRCA mutations in both arms. Conclusions: Rucaparib improves progression-free survival for patients with mCRPC with either germline or somatic BRCA mutations with a manageable safety profile. These data support the use of rucaparib as a beneficial treatment option for patients with BRCA-mutated mCRPC with germline or somatic mutations. Clinical trial information: NCT02975934 .

11114 Background: Advances in early detection and treatment of breast cancer (BC) have significantly improved survival outcomes. In this growing survivor population, addressing modifiable lifestyle factors such as alcohol intake (AI), a known risk factor for BC and BC recurrence, is critical to improving prognosis and health-related quality of life (HRQoL). This study examines AI patterns and their associations with mental and physical HRQoL in early BC survivorship. Methods: Adult patients newly diagnosed (≤1 year) with stage I–III BC at Mayo Clinic Rochester were invited to enroll in the Mayo Clinic Breast Disease Registry. Of the 3252 patients who consented, 1997 completed surveys approximately one-year post-diagnosis, capturing weekly AI, demographic factors, and PROMIS-10 scores. PROMIS-10 (a 10-item measure of health, well-being, and distress) physical and mental health T-scores (mean = 50, SD = 10) are standardized to the U.S. population. Weekly AI was categorized as minimal/none (<1 drink), mild (1–4 drinks), moderate (5–14 drinks), or high (≥15 drinks). Univariate analyses were performed with Monte Carlo-based Fisher exact tests. Multivariate multinomial logistic regression models using a glogit link function assessed AI as the outcome and PROMIS-10 scores as independent variables, adjusting for financial status, smoking, chemotherapy, and moderate exercise because those were associated with AI at year 1 in prior analyses. Results: Univariate analyses revealed statistically significant associations between AI and PROMIS-10 scores (Table 1). Adjusted multivariate analyses identified that patients with better physical health were more likely to drink 1–4 or 5–14 drinks/week and less likely to abstain (<1 drink/week) than those with poorer physical health (p=0.027). There was no statistically significant association between AI and mental health QoL after adjustment for covariates. Conclusions: Better physical HRQoL during early survivorship was associated with higher AI, suggesting that patients who are feeling physically unwell may be less likely to drink alcohol. Additional public health messaging about the relationship between AI and breast cancer risk and recurrence may be needed, especially for those who are feeling well enough to consume alcohol. Univariate analyses of year 1 PROMIS-10 scores with year 1 alcohol consumption using Monte Carlo-based Fisher exact test. Alcoholic drinks per week at year 1 < 1(N=1204) 1-4(N=600) 5-14(N=175) 15+(N=18) Total(N=1997) P-value PROMIS global mental health T score, Year 1, n (%) 0.008 < 50 466 (63.8%) 200 (27.4%) 54 (7.4%) 10 (1.4%) 730 (37.1%) 50+ 718 (57.9%) 394 (31.8%) 120 (9.7%) 7 (0.6%) 1239 (62.9%) Missing 20 6 1 1 28 PROMIS global physical health T score, Year 1, n (%) <0.001 < 50 440 (66.5%) 165 (24.9%) 50 (7.6%) 7 (1.1%) 662 (33.6%) 50+ 740 (56.7%) 432 (33.1%) 124 (9.5%) 10 (0.8%) 1306 (66.4%) Missing 24 3 1 1 29

9002 Background: The field of AI is rapidly evolving. With recent development of more user-friendly AI software, its integration into healthcare across different medical specialties, including HemOnc has emerged. Data detailing the perception and concerns of the HemOnc workforce about AI’s roles is lacking and is needed if AI is to be integrated in HemOnc CliPr and MedEd. Methods: Questionnaires about the perception and concerns regarding AI in CliPr and MedEd were created on REDCap and approved by IRB. We surveyed the entire HemOnc workforce at the 3 Mayo Clinic major sites between 11/07/24 and 01/20/25, including physicians, both faculty (FAC) and fellows (FEL), advanced practice providers (APP), and nurses (RN). Participation was voluntary. Simple statistical analyses were employed for the results. Results: 344 participants (PTP) responded to the survey, 118 physicians (41 FEL and 77 FAC), 49 APP, and 177 RN. 64% of PTP report having used AI but only 31% used it in HemOnc MedEd, and 28% in HemOnc CliPr, with 67% considering themselves to have little to no knowledge about AI. 94% of PTP believe AI will be integrated in HemOnc MedEd, with such integration being seen as beneficial by 90%. Among physicians, 85% of FEL and 92% of FAC report that fellowship programs should incorporate AI training into curricula. 95% of PTP believe AI will be incorporated into HemOnc CliPr. Meanwhile, 62% of PTP (50.8% of physicians, and 68% of APP and RN) are concerned about risks it may pose to CliPr. Nevertheless, 90% of PTP would embrace AI’s use in HemOnc. Table 1 details the perceptions and concerns about AI in HemOnc MedEd and CliPr. The main perceived risks are decreased time spent with patients (pts), and inaccuracies or worse pt care. Meanwhile, 33% believe AI would increase efficiency and quality of care, while 30% believe it would increase time spent with pts. 14% of PTP worry their role could be replaced by AI. Conclusions: To the best of our knowledge, this is the largest assessment of HemOnc workforce’s perception and concerns about AI. Based on our survey, the HemOnc workforce, in general, endorses AI use, but with some concerns raised. While incorporation in MedEd is perceived with excitement, many envision the use in CliPr involves risks and challenges. Proper systematic education of the workforce about AI, with well-designed CliPr integration methods are needed to mitigate the legitimate existing concerns. Perceptions & concerns about AI in HemOnc MedEd & CliPr. FEL % APP % RN % FAC % AI will assist documentation in CliPr 98 82 79 92 AI will be used in pt communication 73 80 78 87 AI will screen pts for clinical trial enrollment 85 86 76 96 AI will increase the efficiency and quality of CliPr 43 37 24 46 AI will increase time spent with pts 39 27 25 38 AI will decrease time spent with pts 29 41 48 23 AI will lead to inaccuracies or worse CliPr 12 20 25 20 AI may replace the HemOnc workforce 11 16 19 5

12083 Background: Fatigue and sleep disturbance are prevalent during and after cancer treatment; sleep-related symptoms affect 30-50% of patients with cancer, and sleep-wake disruptions persist beyond the immediate post-surgical period in breast cancer (BC) survivors. Identifying demographic- and treatment-related factors associated with sleep-related symptoms during survivorship is critical to inform interventions for those at highest risk of sleep disturbance. Methods: Patients who were seen at least once at Mayo Clinic Rochester following an initial diagnosis of BC at age 18 or older and provided informed consent were prospectively enrolled in the Mayo Clinic Breast Disease Registry. Surveys mailed to this cohort included two 11-point numeric rating scale questions regarding difficulty falling asleep and staying asleep in the past week, with values for each ranging from 0 (no problem) to 10 (as severe as you can imagine). Participants who provided sleep-related information on the Year 1 survey (answered approximately one year after cancer diagnosis) were included. Participants were excluded if they had clinical or pathologic metastatic disease and/or recurrence prior to the Year 1 survey. Associations of sleep-related symptoms with demographic and clinical characteristics were assessed using multivariate linear regression models, fitting the sleep difficulty rating scale as the outcome and clinical and demographic factors as exposures. Results: In total, 3,354 participants met inclusion criteria. The average age at BC diagnosis was 59, and most participants were female (99.3%), White (95.4%), non-Hispanic (96.2%), married (78.1%), had at least some college education (80.4%), did not report financial insecurity (76.2%), did not consume alcohol (60.6%) or smoke tobacco (93.2%), and reported at least some weekly exercise (76,6%). Sixty-two percent of participants had clinical stage 0 or I disease; 50.3% underwent lumpectomy (not mastectomy), 59.2% received radiotherapy, 34.1% received chemotherapy, and 67.2% received endocrine therapy. Overall, sleep-related symptoms were relatively low (mean rating for falling asleep = 2.1, mean rating for staying asleep = 3.1). In multivariate analyses, more difficulty falling asleep was associated with higher clinical stage (p < 0.001), more cigarettes smoked per day (p = 0.012), less moderate and/or strenuous exercise (p = 0.001), and more financial hardship (p < 0.001). More difficulty staying asleep was associated with higher clinical stage (0.003), older age (p < 0.001), more education (p = 0.011), and more financial hardship (p < 0.001). Conclusions: In our cohort of BC survivors, factors associated with both increased difficulty with falling and staying asleep at one year after diagnosis are higher clinical stage and more financial hardship. Future research should explore the course of these symptoms over time and across varied treatment trajectories.

12038 Background: CIPN, primarily associated with sensory neuropathy rather than motor or autonomic dysfunction, is a potentially long-term complication of cancer treatment including taxanes and platinums (T/Ps), and can negatively impact quality of life for breast cancer (BC) survivors. This project aims to quantify the severity of neuropathic symptoms at one and three years after diagnosis in BC survivors, comparing recipients of T/P to non-recipients. Methods: In the Mayo Clinic Breast Registry (MCBDR), a longitudinal cohort, surveys and medical record data from patients with stage 1-3 BC were used to understand the burden of neuropathic symptoms at 1- and 3-years post-diagnosis (denoted as Y1 and Y3). Y1 and Y3 raw scores from The Quality of Life Questionnaire-Chemotherapy-Induced Peripheral Neuropathy 20 (CIPN20) composite (CIPN20-C) and sensory subscale (CIPN-S) were converted to a 0-100 point scale, with lower scores corresponding to worse symptoms. Patients with BC recurrence prior to Y3 or incomplete surveys were excluded. We used two sample t-tests and multivariable linear regression modeling to compare recipients of T/P to non-recipients of T/P (with threshold for statistical significance p <0.05). Results: 786 patients were included, 112 of whom (14.2%) received T/P. T/P recipients were younger (p<0.001), more likely to have Stage II/III disease (p<0.001), and less likely to have received endocrine therapy (p<0.001). Univariate analyses revealed worse CIPN20-C score at Y1 (p=0.02) and worse CIPN20-S scores (p=0.004) at Y1 and Y3 in T/P recipients compared to non-recipients (Table). However, differences between the groups were no longer statistically significant after adjustment for age, stage and endocrine therapy. Conclusions: In this cohort, neuropathic symptom severity at Y1 and Y3 after a breast cancer diagnosis did not differ between recipients of taxane and/or platinum agents and nonrecipients after adjustment for age, stage, and endocrine therapy. These data may reassure patients and clinicians who are concerned about CIPN and considering use of these chemotherapies in this setting. Patient demographics and CIPN20 results. Clinical characteristic T/P recipients Non-recipients of T/P Age at diagnosis, mean (SD) 55.3 (11.6)* 59.8 (11.9) White race, N (%) 108 (96.4%) 660 (97.9%) Clinical stage II/III, N (%) 76 (68.5%)* 250 (37.5%) Endocrine therapy, N (%) 73 (65.2%)* 636 (94.4%) Diabetes mellitus, N (%) 9 (8.9%) 52 (8.6%) Y1 CIPN20-C score, mean (SD) 89.6 (10.9)* 92.0 (9.7) Y3 CIPN-C score, mean (SD) 89.4 (11.4) 91.1 (9.7) Y1 CIPN20- S score, mean (SD) 87.3 (15.1)* 91.3 (13.0) Y3 CIPN20-S score, mean (SD) 88.2 (14.1)* 91.1 (11.7) *Statistically significant (p <0.05) on univariate analysis.

e13042 Background: The conventional treatment paradigm for metastatic breast cancer (MBC) is to continue systemic therapy until disease progression. With improved systemic therapy options, some patients (pts) with MBC attain radiographic remission. Tumor-informed circulating tumor (ct)DNA detects molecular disease burden and may guide decision-making. We report on an institutional cohort of pts with MBC in which tumor-informed ctDNA testing was utilized prospectively to guide decisions on systemic therapy or surveillance imaging frequency. Methods: Utilizing an IRB-approved protocol, we retrospectively identified pts monitored with tumor-informed ctDNA using the Signatera™ mPCR-NGS assay (Natera, Inc) as part of routine clinical care in Mayo Clinic, Rochester. We then identified a subset with biopsy-proven MBC who attained radiographic remission of disease following systemic therapy, deemed by the treating medical oncologist as “exceptional responders”. Real-time ctDNA testing was used to inform a decision to either discontinue systemic therapy, treatment de-escalation (remove a component of multi-drug therapy), increase the time interval between restaging scans, or stay off therapy previously stopped due to toxicity . Data on breast cancer characteristics, imaging results, ctDNA test results, and outcomes were collected. Descriptive statistics were used to analyze the data. Results: 10 pts met our study criteria. The median age at MBC diagnosis was 42 years (range 28-61); 5 pts had TNBC, 3 pts had HER2+, and 2 pts with ER+/HER2- MBC. Most pts (60%) were on first or second line therapy (range 1-5). The median duration of the most recent regimen was 27.5 months (range 10-70 months). All pts were without radiographic evidence of active cancer at the time of ctDNA testing with negative results (0 MTM/mL) at the time of clinical decision-making. 4 pts stopped all systemic therapy and proceeded with observation (1 trastuzumab, 1 trastuzumab deruxtecan, 1 pembrolizumab, and 1 talazoparib). 2 pts discontinued a targeted therapy (1 olaparib, 1 abemaciclib) while continuing endocrine therapy. 1 patient stopped targeted therapy (abemaciclib) due to a blood clot, and ctDNA testing guided the decision to not start another CDK4/6 inhibitor while continuing endocrine therapy. In 3 pts the frequency of restaging scans was decreased to either annually or semi-annually. The median follow-up after a ctDNA-informed decision was 11 months (range 1-41 months). At this time, no patient has converted to ctDNA-positivity, experienced resurgence or progression of disease, needed to change systemic therapy or increase imaging frequency. Conclusions: Tumor-informed ctDNA may offer additional confidence to oncologists and pts to deescalate systemic therapy or decrease the interval of imaging frequency in exceptional responders with MBC. Additional data to support this approach is needed.