Haddad, Tufia C.

Abstract Background The ability of large language models (LLMs) to work collaboratively and screen studies in a systematic review (SR) is under-explored. Hence, we aimed to evaluate the effectiveness of LLMs in automating the process of screening in systematic reviews. Methods This is an observational study which included labeled data (title and abstracts) for five SRs. Originally, two reviewers screened the citations independently for eligibility. A third reviewer cross-checked each citation for quality assurance. GPT-4, Claude-3-Sonnet, and Gemini-Pro-1.0 were used using zero-shot chain-of-thought prompting. Collaborative approaches included (i): conflict resolution using benefit of the doubt, (ii) majority voting using an independent third LLM and (iii) conflict resolution using an informed third LLM. Performance was assessed using accuracy, precision for exclusion, and recall for inclusion. Work saved over samples (WSS) was computed to estimate the reduction in manual human effort. Results A total of 11300 articles were included in this study. The individual models, GPT-4, Claude-3-Sonnet, and Gemini-Pro-1.0 exhibited a high precision for exclusion, achieving 99.7%, 99.7%, and 99.2% and high recall for inclusion achieving 95.5%, 96.6% and 85.7%, respectively. However, the collaborative approach utilizing the two best-performing models (GPT-4 and Claude-3S) achieved an average precision of 99.9% and a recall of 98.5% (across all collaborative approaches). Furthermore, the proposed collaborative approach resulted in an average WSS of 63.5%, compared to the average WSS of 45.2% for individual models. Conversational LLM interactions showed a consistent pattern of results. Limitations This study was limited due to reliance on proprietary models, and evaluation on oncology datasets. Conclusion Evidence shows that collaborative LLMs enable efficient, high-performing screening in systematic reviews, supporting continuous evidence updates. Primary funding source NIH (U24CA265879-01-1) and Carolyn-Ann-Kennedy-Bacon Fund.

Abstract Background Chimeric-antigen-receptor (CAR) T-cell therapy has transformed outcomes for relapsed or refractory hematologic malignancies, yet early identification of patients at highest risk for death remains elusive. A prognostic model that leverages only the information routinely available at the time of CAR-T evaluationcould guide pre-infusion counseling, intensity of follow-up, and resource allocation. Methods We analyzed a prospectively maintained cohort of CAR-T recipients treated at Mayo Clinic to develop and test our model. For each patient, data (vital signs, lab results, cardiac assessments [ECG and echocardiography], and PET/CT images) from initial CAR-T evaluation (up to 60 days prior to CAR-T infusion) was considered. Vital signs and labs were left un-imputed, while ECG, echocardiographic parameters, and PET/CT features were forward-filled from the preceding 60 days. Each patient-day was labeled as “outcome present” or “no outcome,” enabling supervised training of gradient-boosting models to predict overall survival (OS) at 6 months post CAR-T infusion. Model performance was evaluated using the area under the ROC curve (AUC) with 10,000-fold bootstrapping with data shuffling between each iteration to derive confidence intervals for stability. To enhance interpretability, we also developed a simplified model using the top 19 features (determined by Shapley value-based importance), aiming to preserve predictive accuracy with a smaller feature set. Results A total of 570 CAR-T cell treated patients were included in this study (476 patients in the training cohort and 94 patients in the validation cohort). The median age was 64 years and 61% males. The primary diagnoses were B-cell lymphoma (324 patients) and multiple myeloma (246 patients). A total of 4,851 individual vital-sign readings, 11,781 laboratory measurements were included. PET/CT images were available in 318 patients. The ML model demonstrated high discriminative ability for predicting OS at 6 months post CAR-T infusion. Vital signs alone produced a mean AUC of 0.63; radiomics features alone produced a mean AUC of 0.71, while laboratory tests alone produced a mean AUC of 0.84. The features most contributory to each single-modality model were then combined into a simplified multimodal model with a total of 19 features, which produced a mean AUC of 0.87, with a 95% CI of 0.73 - 0.97. The 19 features included in the simplified model were: LDH, ferritin, uric acid, eosinophils, fibrinogen, D-dimer, heart rate, mean arterial blood pressure, QTc, QT interval, and the radiomics features of sphericity, first-order Kurtosis, first-order mean, first-order energy, first-order total energy, shape maximum2D diameter (sagittal plane), SUV CoV, SUVp90, shape flatness. This multimodal predictive model was then applied to the validation cohort (uninvolved with model training or testing), which produced an AUC of 0.93, with a 95% CI of 0.84 - 0.98. Conclusions Our ML-driven approach successfully predicts overall survival 6 months post CAR-T infusion, highlighting those individuals at most risk of poor outcomes at the time of CAR-T evaluation. The strong performance of the multimodal version of the model suggests that each data domain contributes complementary resolving power, and by including novel data types, such as PET CT radiomics features, the most accurate predictions could be realized. By providing accurate, near-real-time risk stratification for poor outcomes within 6 months of CAR-T infusion, the model could facilitate proactive patient management. Prospective studies of the integration of this ML model into CAR-T treatment workflows will be crucial to confirm its clinical benefits and to facilitate its adoption in practice.

Abstract BACKGROUND Cognitive symptoms are common among cancer patients treated with intracranial radiation but inadequately assessed and suboptimally treated, owing in part to the time-consuming and resource-intensive nature of formal neuropsychological testing. Through remote administration of brief cognitive assessments and symptom-directed interventions, decentralization uniquely addresses this problem. METHODS In this single-site phase 2 fully decentralized clinical trial (NCT06377696), patients treated with radiation therapy (RT) for primary or metastatic brain tumors undergo cognitive evaluations with the application, Mayo Test Drive, and the Short Test of Mental Status (STMS; via telehealth or in person), at baseline and every three months for one year. Participants are provided wearable devices for health promotion. Half the participants are randomized to metformin to evaluate this agent’s efficacy in treating RT-related cognitive symptoms. Target enrollment is 100 participants. RESULTS Seventy-three patients (median age 45.5, 46% female) with brain tumors who received RT were enrolled May 2024 to May 2025. In this unplanned interim analysis, we evaluated the impact of decentralization on study participation. Thirty-six participants (49%) reported a primary residence greater than 150 miles from our institution. Seven patients were consented remotely, the remainder in person. Of 380 study-related visits (201 STMS, 179 treatment toxicity checks) 216 (57%) were completed by telehealth. Thirty-two of 336 (9.5%) study-related lab draws were completed at locations outside our institution. Wearable device data was continuously collected, with median 3022 steps tracked per patient per day on study. 190/250 (76%) quality-of-life surveys sent digitally were completed to date. CONCLUSION Decentralization implemented as part of this clinical trial enabled participation despite significant distance from our institution and facilitated study-related assessments through digital solutions, eliminating the need for travel. Our findings demonstrate the feasibility of conducting a fully decentralized supportive care trial in neuro-oncology, which can increase participation and accelerate evaluation of novel therapeutics.

Abstract BACKGROUND Access to in-person evaluations is challenging for patients with glioma due to disease-related disability and distance from specialized neuro-oncology centers. Telehealth is an alternative to in-person evaluations, but acceptability of and therapy compliance with this model has not been prospectively evaluated. METHODS In this single-arm non-randomized pragmatic clinical trial, we compared patient satisfaction with telehealth and in-person assessments among patients with glioma undergoing chemotherapy with temozolomide. Participants acted as their own controls, alternating between telehealth and in-person visits. Patient satisfaction with care delivered was evaluated by Press-Ganey surveys with Likert response scales (1-5, 1=very poor, 5=very good). A digital pill diary (DPD) was implemented to track chemotherapy compliance. RESULTS We accrued 30 patients (median age 59, 38% female) between October 2024 and May 2025. Half of the patients (n=15) reported a primary residence greater than 150 miles from our institution. Surveys from 25 participants were reviewed, with 17/25 completing one or more surveys following an in-person visit, and 21/25 completing one or more surveys following a telehealth visit. Overall average Press-Ganey scores were similar between the two visit modalities with high scores (telehealth=4.8; in-person=4.7). Median scores for specific modalities were similarly highly rated: likelihood of recommending practice to others (telehealth=5.0; in-person=5.0), concern the care provider showed (telehealth=5.0, in-person=4.7), provider’s discussion of treatment options (telehealth=5.0, in-person=5.0), provider effort to include patients in decision-making (telehealth=5.0, in-person=5.0). DPDs used to monitor treatment compliance indicated adherence to all prescribed doses of temozolomide in 75/79 cycles; the other 4 (of 79) treatment cycles were halted per provider instructions. CONCLUSION This interim assessment demonstrates feasibility of treating patients with similar patient satisfaction of care delivered between telehealth and in-person visits. The DPD was effective in tracking chemotherapy administration. Additional results including visit type preference and acute care utilization data are maturing.

566 Background: Video telemedicine visits (VTVs) have the potential to overcome disparities in access to cancer care for many socially, geographically, and economically disadvantaged populations. Internet connectivity, digital literacy, and socioeconomic factors are traditionally cited as influencing VTV utilization. However, digital diffusion has occurred rapidly across several sectors, including healthcare, since the onset of the COVID-19 pandemic, and less is known about how these factors have evolved to influence VTV utilization behavior over this time. Through a qualitative approach, we aimed to: (1) explore the lived experiences of cancer patients who transitioned to VTV use as compared to those who persistently did not, and (2) identify similarities and differences between these two groups that could inform targeted interventions to support access and patient acceptance of VTVs in the setting of cancer care. Methods: During June 2024, we conducted semi-structured interviews with 24 patients (Mean age 71.3 years, range 42-88; female 50%) receiving ongoing, longitudinal cancer care at our academic medical center since July 2021 who had not used VTVs through at least October 2022. Using purposeful sampling, patients were equally divided according to VTV utilization behaviors defined as “transitioned users” (TU; i.e. those who transitioned to VTV utilization after October 2022) and “persistent non-users” (PNU). Transcripts were analyzed using a combined deductive–inductive thematic approach and categorized into themes of barriers and facilitators across three domains consistent with the COM-B model: capabilities (C), opportunities (O), and motivations (M). Results: TU patients frequently cited a reduction in travel and logistical considerations as a primary reason for uptake of VTVs. PNU patients frequently reported digital literacy barriers, particularly among older patients, as primary reasons for lack of VTV utilization. TU patients reported more experience and comfort in technical skills as compared to PNU participants. PNU patients also reported less frequent offering of VTV as an option by their care teams over time, implying they may have transitioned to VTV utilization if it had been offered with more regularity. Both study groups desired improved internet access and technical support to enhance (or, in the case of PNU patients, enable) VTV utilization. Conclusions: These findings suggest that multi-modal care delivery may best serve current patient populations, not requiring, yet offering VTV services as an adjunct to in-person care. Longer term, individual barriers to VTV use will also need to be addressed to further optimize access for patients who desire to incorporate this care modality.

560 Background: Early diagnosis remains a critical challenge in oncology, particularly for the five deadliest cancers: lung, breast, prostate, colorectal, and pancreatic. These cancers often present with nonspecific symptoms, leading to repeated healthcare visits without a timely suspicion of malignancy. Artificial intelligence (AI) offers an opportunity to augment clinical decision-making and support earlier identification of cancer. C the Signs , an AI cancer prediction platform, has been widely adopted across the UK with substantial evidence supporting its ability to identify cancer at earlier stages. Methods: This retrospective study assessed the performance of C the Signs in intercepting cancer earlier than physician diagnosis. De-identified electronic medical records were accessed via the Mayo Data Platform, spanning a 20-year period from January 1, 2002 to December 31, 2021. The study cohort included 55,375 patients diagnosed with lung, breast, prostate, colorectal or pancreatic cancer. The primary outcome was the proportion of cases identified earlier by the AI model, measured in annual increments up to five years prior to physician diagnosis. Additional analyses were conducted across the five target cancer types. Results: The C the Signs platform identified 14,528 patients (26.2%) with cancer earlier than the original physician diagnosis. Early identification occurred across multiple time points: 8.9% in year 1, 4.7% in year 2, 4.4% in year 3, 4.3% in year 4, and 3.9% in year 5. Across cancer types, the highest early interception rate was observed in prostate cancer (31.7%), while breast cancer showed the lowest impact (21.0%). Conclusions: Use of the C the Signs AI cancer prediction platform demonstrated meaningful improvements in early cancer interception, particularly for lung and prostate cancers. These findings highlight the potential of AI tools to enhance early diagnosis and support timely clinical interventions, offering a scalable solution to improve cancer outcomes across populations.

Abstract Introduction: This ongoing study investigates the impact of vitamin D deficiency on the immune response to prostate cancer (CaP) in Black patients and its potential contribution to CaP disparities. The study also evaluates differences in immune function, including p53 and CaP associated antibody levels, in Black men with localized versus metastatic CaP. Methodology: Eligible patients with a history of localized or metastatic CaP are enrolled following informed consent (in person or remote). At baseline, blood is collected for calcium and vitamin D3 levels, and for immune testing. Patients who participate in the screening portion of the trial and are found to have vitamin D3 levels <30 ng/mL are prescribed oral vitamin D3 (2000 IU daily) and asked to record adherence in a medication diary. Follow-up includes a 4-week check-in and end-of-treatment labs at 8 weeks for calcium, vitamin D3 levels, and immune re-testing. Virtual consults occur after baseline and at 8 weeks. Quality of life is measured at baseline and EOT follow-up using the CDC HRQOL–4. The study began accruing patients in December 2021. In an attempt to accelerate patient accrual, the protocol was amended to become one of Mayo Clinic’s first fully decentralized clinical trials. Newly integrated components included at-home and community-based blood collection (outside of Mayo Clinic southern sites) and continued virtual engagement. These changes were designed to reduce barriers linked to social determinants of health (SDOH) and expand participation among medically underserved populations. This is a report on the impact of clinical trial DCT. Results: From December 2021 to September 2023 (pre-DCT), 24 patients were enrolled from 66 patients who participated in the screening portion for the trial (mean accrual: 1.09/month; SD: 1.15). Following the DCT amendment, enrollment rose to 102 patients from 213 screened (mean: 4.86/month; SD: 3.12) through June 2025. Of 126 enrolled patients, 119 (94.4%) returned medication diaries, with only 6 non-returns occurring post-DCT. QOL measures were completed by approximately 75% of participants. Importantly, 90 of 126 participants (71%) were enrolled from outside Jacksonville, FL or Scottsdale/Phoenix, AZ, indicating successful national expansion. A total of 111 patients failed the screening due to baseline vitamin D levels >30 ng/ml. The total number of protocol deviations increased post-DCT (60 from 10/23–6/25 vs. 14 from 12/21–9/23), there was no change in the number of protocol deviations per enrolled patients. Minor documentation and study timeline deviations were the most common in both periods. Conclusions: The successful implementation of a fully decentralized model was associated with a marked increase in the accrual of underserved patients and allows the completion of critical clinical trials that address important cancer care disparities without impacting protocol integrity. Clinical trial DCT can potentially reduce SDOH-related barriers and improve representation of underserved populations in research. Citation Format: Sandra Casanova Leyva, Kim M. Barbel Johnson, Tufia Haddad, Lydia Mercado, Trevanne Matthews Hew, Cassandra Moore, Gerardo Colon-Otero, Devina Joshi, Roselyn Heduvor. Feasibility of an interventional clinical trial addressing prostate cancer disparities in Black men by decentralization (DCT): A potential solution to disparities in clinical trial participation [abstract]. In: Proceedings of the 18th AACR Conference on the Science of Cancer Health Disparities; 2025 Sep 18-21; Baltimore, MD. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2025;34(9 Suppl):Abstract nr C064.

PURPOSE Neoadjuvant immunotherapy (IO) has become the standard of care for early-stage triple-negative breast cancer (TNBC), but not yet for other subtypes. We previously developed a clinical-grade mRNA-based immune classifier (ImPrint) predicting response to IO that is now being used in I-SPY2.2 as part of the response predictive subtypes. We report the performance of ImPrint in hormone receptor–positive and human epidermal growth factor receptor 2–negative (HR+HER2–) patients from five IO arms. METHODS A total of 204 HR+HER2– (MammaPrint high-risk) patients from five IO arms (anti–PD-1, anti–PD-L1/poly [ADP-ribose] polymerase inhibitor combination, anti–PD-1/toll-like receptor 9 dual-IO combination, and anti–PD-1 ± lymphocyte activation gene 3 dual-IO combination) and 191 patients from the chemotherapy-only control arm were included in this analysis. Patients were classified as ImPrint+ (likely sensitive) versus ImPrint– (likely resistant), using pretreatment mRNA. Performance of ImPrint for predicting pathologic complete response (pCR) to IO-containing arms was characterized and compared with tumor grade (III), MammaPrint (ultra) High2 risk (MP2), and estrogen receptor (ER)-low (ER ≤ 10%). RESULTS Overall, the pCR rate across the five IO arms was 33%. 26% of HR+HER2– patients were ImPrint+, and pCR rates with IO were 75% in ImPrint+ versus 17% in ImPrint–, with the highest pCR rate >90% in a dual-IO arm. In the control arm, pCR rates were 33% in ImPrint+ and 8% in ImPrint–. Tumor grade (III), MP2, and ER-low showed pCR rates in IO of 45%, 56%, and 63%, respectively, with lower pCR odds ratios (OR < 7.5) compared with ImPrint (OR = 14.5). CONCLUSION Using an accurate selection strategy, HR+HER2– patients could achieve pCR rates similar to what is seen with best neoadjuvant therapies in TNBC and HER2+ (ie, pCR rate >65%-70%). ImPrint, an Food and Drug Administration IDE-enabled assay, may represent a way to identify HR+HER2– patients for IO that best balances likely benefit versus risk of serious immune-related adverse events.

Although innovation in cancer treatment has improved cure rates and survival, the costs have escalated beyond societies' ability to pay. Cancer care costs in the United States are expected to rise to 245 billion by 2030. In this health policy session, we focused on three cost-containment strategies, and their potential impact on cancer care delivery. Site Neutrality: Curbing the Cost of Cancer Care, but at What Risk? , explores legislation that would introduce payment parity across differing sites of care: elimination of facility fees in hospital owned practices and shift of billing for oncology infusions away from hospital owned practices. Unintended consequences could reduce access to care for vulnerable populations, shift costs to patients, and reduce safety. Post-Pandemic Digital Health Reimbursement: Impact on Access to Care , explores how the emergency waivers, licensure flexibilities, and parity reimbursement necessitated by COVID-19 shepherded innovation that outpaced the regulatory framework needed for long-term sustainability. Today, telehealth services, hospital at home, remote patient monitoring, and decentralized clinical trial enrollment face reintroduction of regulatory and payment barriers. Will we live to see efficacy-based pricing for cancer drugs? Learning from international models , explores how lack of drug price negotiation has led to higher prices in the United States compared with Canada and Europe. Comparative clinical effectiveness pricing systems compare the value of each drug to standard treatments. Comparative cost-effectiveness programs look at the incremental cost per additional unit of health gained above the standard of care. The United States could learn from these approaches used in comparable countries.

Abstract Background Anthracyclines (ACs) are a risk factor for heart failure (HF), but their impact on broader composite outcomes such as major adverse cardiovascular events (MACE) remain less well characterized. From the Rochester Epidemiology Project, we identified patients with breast cancer or lymphoma to assess for association between AC use and MACE in these patients with > 25 years of follow-up. Methods We identified patients with cancer, with or without an AC as first-line treatment, and a control group of individuals without cancer, matched by age, sex, and cardiovascular comorbid conditions to the cancer patients. Outcomes of interest were 3-MACE (myocardial infarction, stroke, and cardiovascular death), and 4-MACE (3-MACE plus HF). Cox regression analysis was used to identify variables associated with MACE at 5, 10, and 25 years of follow-up. Results Of 812 patients with cancer, 675 had AC treatment (AC group) and 137 did not (no-AC group); the control group comprised 1,384 matched patients without cancer. The risk of 3-MACE was similar among the 3 groups at all time points. Older age, male sex, and hypertension were associated with 3-MACE. The risk of 4-MACE was similar between the control and no-AC groups; the risk was increased for the AC group vs controls at all time points. Older age, male sex, smoking history, type 2 diabetes, and hypertension were associated with 4-MACE. Conclusions Cardiovascular risk factors are the main drivers of 3-MACE and ACs increase the risk of 4-MACE given their association with HF. This underscores the importance of implementing lifestyle changes among these patients.