Bansal, Radhika

8549 Background: First-line treatment of early stage non-small cell lung cancer (NSCLC) is surgical resection, but with a 5-year survival of only 54%, rates of future relapse are high. Identifying patients at high risk of relapse can help guide adjuvant treatment decisions. Deep convolutional neural network (DCNN) AI models trained on tumor histology have shown incredible flexibility as potential biomarkers, both in lung cancer and more generally across many malignancies. While DCNN models can obtain extremely accurate results when used for routine purposes such as diagnosis, subtyping, and grading of malignancies, most models trained for prognostication or treatment response prediction do not reach performance sufficient for clinical application. Uncertainty quantification (UQ) – a family of techniques that give DCNN models the ability to report confidence alongside predictions – is an underexplored avenue in cancer AI that may help further improve performance and clinical application of models designed to provide clinicians with estimations of risk. Methods: To explore the potential use of UQ in clinically-oriented DCNN models, we trained models on a single-institution, retrospective digital tumor histology slide cohort from patients with Stage I-III NSCLC who underwent surgical resection to predict risk of future relapse. Estimation of uncertainty was performed using dropout as a Bayesian approximation, and uncertainty thresholds were calculated from training sets using a novel method to identify and remove low-confidence predictions. For comparison, a separate multivariate logistic regression was trained in cross-validation using known clinical risk factors. Results: We trained DCNN models on slides from 198 patients (40 relapsed, 158 without relapse). In this cohort, 130 patients had stage I disease (65.6%), 42 had stage II (21.2%), and 26 had stage III (13.2%). The average age was 69 years, 85% were current or previous smokers, and 38 received guideline-concordant adjuvant chemotherapy. Without UQ, a DCNN model predicted risk of future relapse with an average area under receiver operator curve (AUROC) of 0.74 across three-fold cross-validation. Using UQ estimation, 63% of slides were reported with high confidence. Relapse prediction was significantly improved in the high-confidence cohort, with an average AUROC of 0.83 in cross-validation. With a specificity of 70%, this corresponds to an average sensitivity of 86.8% across the three cross-folds (79.1%, 85.5%, and 95.7%). In comparison, a clinical-only multivariate regression model predicted relapse with a cross-validated AUROC of 0.67. Conclusions: This method of uncertainty quantification appears to be a powerful tool to predict lung cancer recurrence risk from digital histopathology while simultaneously providing clinicians with a measurement of algorithm trustworthiness.

8022 Background: Bone marrow (BM) assessment of minimal residual disease (MRD) is being considered as a surrogate endpoint in clinical trials and is prognostic for survival in multiple myeloma (MM). Timing of BM assessment is variable across Chimeric Antigen Receptor T cell (CART) therapy trials and differs from standard of care practice. BM myeloma cell clearance can be detected by month 1 post CART, even before serum immunofixation becomes negative. BM is still hypocellular at month 1, thus prognostic value of MRD negative (MRDneg) at this timepoint is unclear. We examined impact of Day 30 MRD status in patients (pts) who received CART at Mayo Clinic. Methods: Medical records were reviewed retrospectively for MM pts who received CART between 8/2016 and 6/2021. PFS and OS were plotted by Kaplan-Meier method. Results: Sixty MM pts received CART and had BM biopsy at month 1. Median age was 62 yrs, 53% were male, and 78% were BM MRDneg by flow cytometry. Baseline demographics were similar between MRDneg and MRD+ (Table). Overall, 85% (40/47) who were month 1 BM MRDneg had i/u FLC<normal. Patients who achieved CR/sCR had higher rates of BM MRDneg (100% vs 61%, p<0.001) and i/u FLC< normal (89% vs 58%, p<0.001). At month 1, 24/60 (40%) pts had hypocellular BM. Serial BM samples at month 3 (n=35), 6 (n=28) and 12 (n=23) showed MRDneg rate of 93% (25/27), 56% (9/16) 58% (7/12), respectively.. Rate of hypocellularity was 54% (19/35), 32% (9/28) and 30% (7/23), respectively. Among the MRDneg/hypocellular pts at month 1, hypocellular BM was seen in 8/11 (73%) pts at month 3 and 2/4 (50%) pts at month 6 and 12. Compared to MRD+, pts who had BM MRDneg at months1 had longer PFS (Table). PFS was not statistically significantly different between pts who had BM MRDneg and were either hypocellular or not. MRDneg pts with i/u FLC<normal at months1 had better median PFS compared to those who did not. (MRD+:2.9 months (1.2-NR). MRDneg/FLC>normal: 4.9 months (2.3-NR). vs MRDneg/FLC<normal:17.9 months (11.8-NR), p<0.0001). Conclusions: Hypocellular BM is common in the 3 months post CART. Regardless of BM cellularity, BM MRDneg at month 1 correlates with deep response and prolonged PFS. Majority of BM MRDneg pts at month 1 also had FLC<normal. BM MRDneg status and FLC normalization were associated with longer survival. Our data support the continued evaluation of BM early post CART infusion as a prognostic tool. [Table: see text]

e20021 Background: CAR T cell therapies are FDA approved for patients with triple refractory multiple myeloma (MM) and >4 lines of therapy. The median survival of this is group is less than 12 months so salvage therapies need to be implemented quickly. Real-world access to CAR T remains challenging due to supply chain limitations impacting manufacturing. The goal of this study was to evaluate how centers are handling the challenges of CAR T slot allocation. Methods: MM CAR T physician leaders at each CAR T treatment center certified for idecabtagene vicleucel across the US were surveyed. Results: We received responses from 15/20 centers. Summary of CAR T volumes and outcomes of patient on waitlists are shown in the table. The median time on the waiting list was 6 months with only 25% of patients receiving CAR T eventually. For patient selection, all centers reported using a committee of experienced CART physicians to ensure consistency. Selection committee included: primary MD (n=9), CAR T MD (n=12), social workers (n=3), CAR T RN (n=10), APP (n=2), pharmacists (n=2) and ethicists (n=1). To ensure transparency, centers have clear selection criteria (n=10), priority score (n=8) and selection timeline (n=11) . To ensure accountability, centers document priority scores (n=5) and have pre-specified criteria (n=9) for selection. Centers also reported using ethical principles for selection: a) equal treatment: time spent on waiting list (n=8); b) priority to the worst-off: limited therapeutic options (n=10), MM burden (n=7), ineligible for trials (n=2); c) maximize benefit: most likely to complete apheresis (n=8) or infusion (n=8) or achieve response (n=6) and d) social value: younger pts (n=2). Maximizing benefit was considered the most important criterion by 7 centers. Conclusions: More stringent GMP manufacturing requirements with FDA approved CAR T and real-world practice with broader patient demographics can present challenges to implementation of CAR T in standard of practice. Our study is the first attempt to evaluate and highlight existing issues with MM CAR T access and the variability and challenges in patient selection. Learning from other models of resource allocation (ex: UNOS) and sharing experience across centers can help providers optimize slot allocation to improve accrual. Increasing supplies of key reagents and novel manufacturing methods(eg: non-viral vectors and allogeneic CAR) could help overcome these limitations. [Table: see text]

e19049 Background: Prognosis of COVID-19 is poor in the setting of immunosuppression. Casirivimab/imdevimab (REGEN-COV), bamlanivimab, and sotrovimab are investigational monoclonal antibodies (MoAbs) authorized for treatment of mild/moderate COVID-19 for patients (pts) 12 years or older and who are at high-risk for progression to severe COVID-19. These neutralizing antibodies, against SARS-CoV-2 spike proteins, have been shown to decrease risk of progression to severe disease. Recipients of allogeneic stem cell transplants (allo-SCT) or chimeric antigen T cell therapy (CAR T cell) represent a high risk population. However, treatment outcomes with these MoAbs in these pts are not well described. Methods: This retrospective study included 33 consecutive adult pts who developed mild/moderate COVID-19 and received anti-spike SARS-CoV-2 MoAbs between December 2020 and November 2021. Allo-SCT (N=27) or CAR T cell (N=6) recipients were included, and outcomes were analyzed separately. Pts received REGEN-COV (N=19), bamlanivimab (N=11), or sotrovimab (N=1), missing (N=2). Results: In the allo-SCT cohort (N=27), median age at time of COVID-19 was 55 (23-76) years. Median time from allo-SCT to COVID-19 was 31 (22-64) months. Two pts received CAR T-cell therapy prior to allo-SCT. Diagnoses included leukemia or myeloid diseases (82%), lymphoma (11%), or myeloma (7%). Transplant characteristics are summarized (Table). Thirteen pts were vaccinated against SARS-CoV-2 prior to breakthrough COVID-19. Events considered included hospitalization due to COVID-19, disease progression, or death from any cause. The 6-month event-free and overall survivals were 81% and 91%, respectively. In the CAR T cell recipients cohort (N=6), 4 pts received axicabtagene ciloleucel for diffuse large B-cell or follicular lymphoma and 2 received brexucabtagene autoleucel for mantle cell lymphoma. The median follow-up was 8 (1-11) months. Two pts received autologous SCT prior to COVID-19. Median time from CAR T cell therapy to COVID-19 was 10 (3-24) months. Three pts were vaccinated prior to COVID-19. Only 1 pt was hospitalized due to severe COVID-19 requiring mechanical ventilation leading to death. Conclusions: These results show a potential benefit of MoAbs in high-risk pts, namely allo-SCT or CAR T cell recipients. Future studies should evaluate the role of prophylactic use MoAbs in these populations. A comparative analysis with a matched control cohort (who did not receive MoAbs) will be provided at the meeting. [Table: see text]