Citations

Abstract Generative artificial intelligence (AI) holds immense potential for medical applications, but the lack of a comprehensive evaluation framework and methodological deficiencies in existing studies hinder its effective implementation. Standardized assessment guidelines are crucial for ensuring reliable and consistent evaluation of generative AI in healthcare. Our objective is to develop robust, standardized guidelines tailored for evaluating generative AI performance in medical contexts. Through a rigorous literature review utilizing the Web of Sciences, Cochrane Library, PubMed, and Google Scholar, we focused on research testing generative AI capabilities in medicine. Our multidisciplinary team of experts conducted discussion sessions to develop a comprehensive 32‐item checklist. This checklist encompasses critical evaluation aspects of generative AI in medical applications, addressing key dimensions such as question collection, querying methodologies, and assessment techniques. The checklist and its broader assessment framework provide a holistic evaluation of AI systems, delineating a clear pathway from question gathering to result assessment. It guides researchers through potential challenges and pitfalls, enhancing research quality and reporting and aiding the evolution of generative AI in medicine and life sciences. Our framework furnishes a standardized, systematic approach for testing generative AI's applicability in medicine. For a concise checklist, please refer to Table S or visit GenAIMed.org .

Predictive tools provide a unique opportunity to explain the observed differences in outcome between patients of the COVID-19 pandemic. The aim of this study was to associate individual demographic and clinical characteristics with disease severity in COVID-19 patients and to highlight the importance of machine learning (ML) in disease prognosis. The study enrolled 344 unvaccinated patients with confirmed SARS-CoV-2 infection. Data collected by integrating questionnaires and medical records were imported into various classification machine learning algorithms, and the algorithm and the hyperparameters with the greatest predictive ability were selected for use in a disease outcome prediction web tool. Of 111 independent features, age, sex, hypertension, obesity, and cancer comorbidity were found to be associated with severe COVID-19. Our prognostic tool can contribute to a successful therapeutic approach via personalized treatment. Although at the present time vaccination is not considered mandatory, this algorithm could encourage vulnerable groups to be vaccinated.

e19049 Background: Acala and zanu are 2nd generation Bruton tyrosine kinase inhibitors (BTKis) that have demonstrated improved safety profiles vs 1st generation BTKi ibrutinib in clinical trials. While head-to-head comparison of zanu vs acala is lacking, a recent meta-analysis of clinical trials provided a comprehensive, indirect comparison of the AE profiles in B-cell malignancies (Hwang, Hemasphere. 2023;7(S3):1134). This study aims to evaluate costs and impacts on quality of life (QoL) for zanu vs acala using their AE profiles. Methods: A health economic model was developed to determine cost and QoL associated with AE management for zanu and acala. The incidence rates ( IRs) of all grade and grade ≥3 AEs of interest (n=21, including bleeding events, hypertension, atrial fibrillation, cytopenias, infections, headache, arthralgia, diarrhea) were taken from the meta-analysis. Additional model inputs, such as disutility, average duration of AE, and unit cost of each AE were sourced from published literature and publicly available cost databases in consultation with clinical experts. The model was developed from the US perspective, considering Medicare costs, inflated to 2023 USD. Model outcomes were AE management cost (= IR * unit cost) and quality-adjusted life years (QALYs) lost due to AE (= IR * Disutility * Duration/365.25). Results: In the base-case scenario (considering all AEs), treating a hypothetical cohort of 1000 patients (pts) with zanu instead of acala yielded cost savings of $124K and 3.7 QALY gains (i.e., 3.7 years extra in full health). Subgroup analysis for severe (grade ≥3) and non-severe AEs (grade 1-2) demonstrated similar trends. A sensitivity analysis limited to AEs significantly different between zanu and acala (n=13) yielded consistent results. Additionally, a probabilistic sensitivity analysis, conducted with 1000 iterations to account for uncertainty in model parameters, confirmed robustness of the model, indicating stable conclusions across a wide range of parameter uncertainties. Conclusions: This analysis suggested that zanu is cost-saving and associated with added health benefits compared to acala in terms of AE management in pts with B-cell malignancies. Should these results derived from meta-analysis of clinical trial data be applicable to real-world pt populations across different indications, the potential cost savings could be considerable. These results could be further improved if the differences in efficacy of zanu vs acala are also incorporated. [Table: see text]

3020 Background: Radioligand therapy (RLT) use has been increasing with the approval of 177Lu-Dotatate for metastatic neuroendocrine tumors (NET) & 177Lu-PSMA-617 for metastatic castration resistant prostate cancer (mCRPC). Therapy-related myeloid neoplasms [tMN, including myelodysplastic syndrome (MDS) & acute myeloid leukemia, (AML)] have been reported after 177Lutetium (177Lu) for NET with rates of 2-20% & median time from first 177Lu to tMN of 2.8yrs [4-12]. Clonal hematopoiesis (CH) is a risk factor for tMN, with the spectrum of CH depending on selection pressures. To describe the prevalence of therapy selected CH & tMN, we performed a retrospective analysis of patients (pts) who had a bone marrow biopsy (BMB) post-177Lu. Methods: After IRB approval, we queried the electronic health record for pts with BMB post-177Lu. Pt demographics, treatment history, labs & BMB results were collected. Results: Fifty-seven pts met criteria, 41 (72%) with NET & 16 (28%) with mCRPC. . Median age at BMB was 68yrs (range, 28-85). Of the 57pts, prior treatments included: radiation (RT) in 21 (37%), temozolomide in 14 (25%), taxane in 15 (26%), & platinum agents in 10 (18%). The median number of 177Lu cycles was 4 (range, 1-11).Cytogenetics were available in 84% (n = 48) & were normal in 60% (n = 29). Of those with abnormal cytogenetics (n = 18, 38%), the most common abnormalities involved chromosome (chr) 7 (n = 12, 26%), chr 5 (n = 7, 15%), chr 20 (n = 5, 11%) and complex karyotypes in 7 (15%), with 3 (17%) of these patients having no mutations. Diagnoses were assessable in 49 (86%) pts: 15 (31%) with undefined cytopeniasand 16 (33%) with tMDS. Metastatic carcinoma was identified in 17 Fifty-one percent (n = 29), at a median of 0.5yrs (range, 0.04-2.6) after hematologic diagnosis. Conclusions: The enrichment of CH involving the DDR pathway in recipients of 177Lu underscores the impact of RLT on bone marrow progenitors.

TPS2700 Background: Though immunotherapy with anti-PD-1 immune checkpoint inhibitors (ICIs) has led to improvements in clinical outcomes in patients with cancer, when used as monotherapy in the advanced setting, only 18-25% of patients respond to ICI. Notably, ICI frequently lacks anti-tumor activity in immune-cold cancers where T cell priming is missing. Recent breakthroughs in identifying personal neoantigens via a comprehensive analysis of cancer sequencing data have brought increased attention to neoantigen cancer vaccines. Personalized cancer vaccines targeting neoantigens have shown early promise. While neoantigens have recently been investigated in some cancer types, the current neoantigen prediction algorithms have often focused on the MHC class-I subtype, single nucleotide mutations (SNM), small insertions and deletions (INDEL). We recently developed an informatics workflow, REAL-neo, for the identification, quality control (QC), and prioritization of both class-I and class-II human leukocyte antigen (HLA) bound neoantigens that arise from somatic SNM, INDEL, aberrant RNA splicing, and gene fusions, generating much more potent neoantigen candidates. Furthermore, we demonstrated robust T-cell response and prolonged survival by combining ICI and cancer vaccines in preclinical models. This single-arm phase I clinical trial is in progress to assess the safety, feasibility, and immunogenicity of personalized neoantigen vaccines in combination with pembrolizumab in patients with advanced solid cancers. Methods: The vaccine consists of up to 20 peptides (15-30 mer) comprising patient-specific neoantigens identified from tumor DNA and mRNA sequencing data using REAL-neo and delivered with GM-CSF as an adjuvant. The first cohort of three patients will be treated at dose level 1 consisting of 4 or 5 peptides at 300 mcg/peptide and GM-CSF 125 mcg per injection site in each of four limbs. After Cohort 1 is deemed safe, the study will expand to cohort 2, when patients will be enrolled at the same dose level for vaccine plus Pembrolizumab 200 mg i.v. Neoantigen vaccine will be given via subcutaneous injection on days 1, 4, 8, 15, and 21 (cohorts 1 and 2) and weeks 5 and 8 (booster dose for cohort 2 only). Cohort 1 has been completed without dose-limiting toxicity (DLT). Enrollment in Cohort 2 began in January 2024. AEs are assessed according to CTCAE v5. and safety findings are reviewed by data safety medical board (DSMB). Key eligibility criteria include 1) histologically confirmed locally advanced or metastatic solid malignancies and 2) cancer progression after at least one line of the standard-of-care (SOC) systemic treatment. For cohort 2, patients must be eligible to receive pembrolizumab per SOC or the treating physician’s judgment. Clinical trial information: NCT05269381 .

TPS2699 Background: Despite the success in hematologic cancers, chimeric antigen receptor (CAR) T-cell therapies are challenging to implement in solid tumors owing to a lack of tumor-specific targets that discriminate cancer from normal cells. MSLN expression normally is limited to the mesothelium of major body cavities but can be upregulated in diverse solid tumor types (TCGA 2022), making it a potential target for cancer therapy. MSLN-targeted cell approaches, including CAR T-cell and T-cell receptor fusion therapies, have shown promising clinical activity; however, on-target, off-tumor toxicity including fatal events have occurred (1-3). A2B694 is an autologous logic-gated, MSLN-targeted Tmod CAR T-cell therapy that addresses the challenges of on-target, off-tumor toxicity by combining 2 CARs: an activating and blocking receptor. The activator recognizes MSLN present on the surface of both tumor and normal cells; the blocker binds HLA-A*02 and prevents CAR T-cell activity. Thus, in patients with both germline HLA-A*02 and tumor-associated HLA-A*02 LOH, the blocker prevents on-target, off-tumor toxicity on normal cells owing to retained HLA-A*02 expression (4). Through this unique discriminatory mechanism, A2B694 may provide a therapeutic window to treat patients with MSLN-expressing solid tumors exhibiting HLA-A*02 LOH. Methods: EVEREST-2 (NCT06051695) is a first-in-human, phase 1/2, open-label, nonrandomized study to evaluate the safety and efficacy of A2B694 in adults with recurrent unresectable, locally advanced, or metastatic cancers with MSLN expression, including non-small cell lung cancer, colorectal cancer, pancreatic cancer, ovarian cancer, mesothelioma, or other solid tumors with MSLN expression. Eligible patients should have received ≥1 line of prior therapy such as checkpoint inhibitor, molecular targeted, or chemotherapy. Enrollment to EVEREST-2 occurs through the prescreening study BASECAMP-1 (NCT04981119), which identifies patients with tumor-associated HLA-A*02 LOH via next-generation sequencing (Tempus AI, Inc.). Eligible patients enroll in the BASECAMP-1 study and undergo leukapheresis. A2B694 is manufactured from cryopreserved T cells when clinically appropriate for patients. The primary objective of phase 1 is to evaluate the safety and tolerability of A2B694 and identify the recommended phase 2 dose (RP2D). The dose-expansion phase will confirm RP2D and collect biomarker data to further characterize A2B694. Phase 2 will assess overall response rate per RECIST v1.1. 1. Beatty, et al. Gastroenterology. 2018. 2. Haas, et al. Mol Ther. 2023. 3. Hong, et al. ESMO 2021. Abstract 9590. 4. Hamburger, et al. Mol Immunol. 2020. Clinical trial information: NCT06051695 .

TPS2698 Background: Chimeric antigen receptor (CAR) T-cell therapies for treating solid tumors are challenging due to a lack of tumor-specific targets that discriminate cancer from normal cells. Previous studies using CEA T-cell receptors and T-cell engagers have resulted in dose-limiting, on-target, off-tumor toxicities (1,2). A2B530 is an autologous logic-gated, CEA-targeted Tmod CAR T-cell therapy that addresses the challenges of on-target, off-tumor toxicity by combining a CAR-activating receptor with a blocking receptor to discriminate tumor from normal cells (3). The activator recognizes CEA on the surface of both tumor and normal cells while specificity for tumor cells is provided by a blocker that binds HLA-A*02. In patients with both germline HLA-A*02 and tumor-associated HLA-A*02 LOH, the blocker prevents on-target, off-tumor toxicity on normal cells owing to retained HLA-A*02 expression (4). HLA-A*02 LOH can be detected using next-generation sequencing (Tempus AI, Inc.). With this definitive discriminator target, A2B530 can potentially provide a therapeutic window to treat patients with CEA-expressing solid tumors exhibiting HLA LOH. Methods: EVEREST-1 (NCT05736731) is a seamless, phase 1/2, open-label, nonrandomized study to evaluate the safety and efficacy of A2B530 in adult patients. Patients are enrolled through BASECAMP-1 (NCT04981119), a master prescreening study that identifies patients with HLA LOH at any time in the course of their disease. BASECAMP-1–eligible patients undergo leukapheresis and, when clinically appropriate, their banked T cells are manufactured for the EVEREST-1 study. The key inclusion criteria include histologically confirmed recurrent unresectable, locally advanced, or metastatic cancers associated with CEA expression: non-small cell lung, colorectal, or pancreatic cancers. Patients should have received ≥1 line of prior therapy such as checkpoint inhibitor, molecular targeted, or chemotherapy. The primary objective of phase 1 is to evaluate the safety and tolerability of A2B530 and determine the recommended phase 2 dose (RP2D). The dose-escalation portion is based on the Bayesian optimal interval design. The dose-expansion phase will confirm RP2D and collect biomarker data to further characterize A2B530. Phase 2 will assess overall response rate per RECIST v1.1. As of January 29, 2024, 8 patients have been enrolled on EVEREST-1. A2B530 was successfully manufactured for all patients, and all patients have received A2B530 infusion, with the first patient dosed in May 2023. Dose escalation is ongoing. 1. Parkhurst, et al. Mol Ther. 2022. 2. Tabernero, et al. J Clin Oncol. 2017. 3. Hamburger, et al. Mol Immunol. 2020. 4. Sandberg, et al. Sci Transl Med. 2022. Clinical trial information: NCT05736731 .

TPS2682 Background: Myeloid cells are actively recruited to the solid tumor microenvironment (TME) and have the potential to mediate tumor control via phagocytosis, TME remodeling, and T cell activation. We previously developed human chimeric antigen receptor macrophages (CAR-M) and have shown potent anti-tumor activity in pre-clinical solid tumor models. The anti-HER2 CAR-M cell therapy product, CT-0508, is currently being evaluated in a Phase I trial as a monotherapy and in combination with pembrolizumab. Early clinical data have shown feasibility, safety, and validated the mechanism of action. We have developed a next-generation CAR monocyte (CAR-Mono) platform to increase the dose, improve tumor trafficking and engraftment, and shorten the manufacturing and vein-to-vein time as compared to CAR-M therapy. CT-0525 is an autologous anti-HER2 CAR-Mono cell therapy based on CD14+ monocytes engineered with an Ad5f35 adenoviral vector to express an anti-HER2 CAR. Pre-clinical studies have demonstrated the feasibility, phenotype, pharmacokinetics, durable CAR expression, cellular fate, antigen specificity, and anti-tumor activity of CT-0525. Pre-clinical studies have shown that CT-0525 differentiated into pro-inflammatory CAR-M in vivo and controlled tumor growth. The CT-0525 manufacturing process takes one day and enables the production of up to 10 billion cells from a single apheresis. CT-0525 is being investigated in a first-in-human, open-label, multi-center, Phase I study in patients with HER2 overexpressing solid tumors. Methods: This Phase 1, first-in-human study evaluates the feasibility, safety, tolerability, trafficking, TME activation, and preliminary evidence of efficacy of the investigational CAR-Mono product CT-0525 in 6 participants (pts) with locally advanced unresectable/metastatic solid tumors overexpressing HER2. Pts previously treated with anti-HER2 therapies are eligible. Filgrastim mobilized autologous CD14+ monocytes are collected by apheresis, followed by manufacturing and cryopreservation. The 1st cohort of pts (n=3) will receive 3 x 109 CT-0525 CAR positive monocytes administered IV in one infusion. If tolerated as per the modified toxicity probability interval algorithm (mTPI), the 2nd cohort of pts (n=3) will receive up to 10 x 109 CT-0525 CAR positive monocytes in one infusion. CT-0525 will be administered without conditioning chemotherapy. Primary endpoints include assessment of safety and tolerability, as well as manufacture feasibility. Correlative assessments include pre- and post-treatment biopsies and blood samples for safety, immunogenicity, pharmacokinetics, tumor trafficking, TME modulation, epitope spreading, and other translational biomarkers. Clinical trial information: NCT06254807 .

TPS2675 Background: Checkpoint inhibitor therapy is the standard first-line treatment for advanced/recurrent HCC, but there is a significant unmet therapeutic need for patients (pts) who progress after first-line treatment. Adoptive cellular immunotherapy may provide a novel treatment option. Glypican-3 (GPC3), a heparan sulfate proteoglycan, is overexpressed by tumor cells in HCC and virtually absent in healthy tissues, making it an ideal target for CAR-T therapy. AZD5851 is an autologous CAR-T product that expresses a CAR specific for GPC3 and a dominant negative (dn)TGFβRII as an armoring strategy. Expression of dnTGFβRII can block TGFβ signaling, protecting CAR T-cells from the immunosuppressive effects of this cytokine. Early clinical data with C-CAR031, a CAR-T product with the same GPC3-specific CAR and dnTGFβRII-armoring transgene as AZD5851, showed promising tolerability and antitumor activity in pts with HCC [1]. ATHENA (NCT06084884) is a first-in-human, single-arm, open-label, multicenter phase 1/2 study to evaluate the safety, tolerability, and antitumor activity of AZD5851 in pts with GPC3+ advanced/recurrent HCC, for whom ≥1 prior line of standard therapy failed or was not tolerable. Methods: Eligible pts are aged ≥18 years with histologically confirmed advanced/recurrent or metastatic and/or unresectable HCC, GPC3+ tumor sample as determined by immunohistochemistry, ≥1 prior line of standard systemic therapy, ≥1 measurable lesion per RECIST 1.1, Child-Pugh A, ECOG PS 0/1, and adequate organ and bone marrow function. Approximately 24 pts will be enrolled in Part A (Phase I, dose escalation) and 50 in Part B (Phase II dose expansion). Pts will undergo apheresis to collect peripheral blood mononuclear cells for AZD5851 production, after which they may receive approved bridging therapy for disease control. Upon successful generation of AZD5851 product, pts will undergo lymphodepletion before receiving a single dose of IV AZD5851. Subsequent follow-up is divided into 3 stages: Stage 1, active follow-up of all pts through 6 months post AZD5851 infusion; Stage 2, further evaluation of pts who have not progressed or started a new anticancer regimen during Stage 1; Stage 3, long-term follow-up for all pts who have received AZD5851. The primary endpoints are to assess the safety/tolerability of AZD5851 (both study parts), and to determine the recommended dose(s) of AZD5851 for expansion and phase 2 trials (Part A). Secondary endpoints include efficacy (objective response rate, best overall response, duration of response, disease control rate, durable response rate, time to response, progression-free survival, and overall survival) and pharmacokinetics. Exploratory endpoints include pharmacodynamic biomarkers and immunogenicity. Enrollment began Nov 2023. 1. Zhang et al. AACR 2023. Abstract CT097. Clinical trial information: NCT06084884 .

e13119 Background: Breast cancer metastasis to the central nervous system (CNS) and intracranial area often leads to severe morbidity and mortality. While systemic treatments manage systemic tumors, their impact on CNS/intracranial metastases is limited. Cell based therapies like SV-BR-1-GM may overcome this limitation. SV-BR-1-GM is a GM-CSF secreting, antigen-presenting immortalized breast cancer cell line. We report its efficacy on CNS tumor regression. Methods: Post-hoc analysis of a subset of all advanced MBC patients with CNS/intracranial metastases from previous phase 1, 2, and an ongoing randomized phase 2, trials (n=7 subjects) of SV-BR-1-GM. Bria-IMT monotherapy (intradermal SV-BR-1-GM + cyclophosphamide 300 mg/m2, +interferon-alpha) was administered q3wks to 2 subjects; an immune checkpoint inhibitor (ICI) was administered with each cycle in the combination setting (n=5). Results: There were 7 patients with intracranial metastases at baseline of whom 5 had intracranial tumor responses. Of the 5 responders, median pt age = 66 (range 58-70), median prior lines = 5.5 (3-13), and median time from initial diagnosis = 7.5 years (3-10). 4/5 pts had ≥1 HLA class match with SV-BR-1-GM. 3 were HR + and 1 HER2 +. Pt A002 (ER+/PR+, HER2-) received 6/6 planned cycles, achieving an 83% reduction in breast lesions; after 3 months off therapy, imaging revealed multiple brain metastases and recurrence of right breast lesions. Following 3 additional cycles there was marked regression in both brain and breast lesions. Pt 02-003 (ER-/PR-, HER2 1+ (IHC)) received 5 cycles of SV-BR-1-GM and observed a 60% reduction (5mm -> 2mm) in a left parietal periventricular lesion, and complete resolution (CR) of a left parietal lobe. Pt 06-005 (ER+/PR-, HER2 1+) demonstrated a 56% (9mm -> 4mm) reduction in the thickness of a dural lesion overlying the left anterior temporal lobe after 6 cycles of SV-BR-1-GM + ICI. After 2 additional cycles of SV-BR-1-GM + ICI and 1 cycle of SV-BR-1-GM (ICI skipped on final cycle due to grade 2 AE), patient achieved CR of a left orbital mass observed at baseline (24mm -> 0m). Pt06-007’s (ER-/PR-, HER2 1+) CNS lesions showed a median reduction of 49% compared to baseline after 3 cycles. Pt 11-018 (ER+/PR-, HER2 +) showed standard progression (SD) at initial restaging; 14% increase in sum of diameters of intracranial metastasis. Subsequent imaging, after 3 cycles, showed significant orbital tumor reduction along with reported improvement in eye pain. Conclusions: This post-hoc analysis demonstrated promising efficacy of the Bria-IMT regimen either alone or in combination with ICIs in CNS/intracranial metastasis. The CNS/intracranial regression seen across all breast cancer subtypes among heavily pretreated patients highlights the potential of SV-BR-1-GM in managing CNS metastasis. Ongoing trials will further evaluate the efficacy of the Bria-IMT regimen in patients with CNS/intracranial metastasis. Clinical trial information: NCT03328026 .