Citations

1673 Background: Survivorship care plans (SCPs) summarize cancer treatments and provide evidence-based recommendations for surveillance, screening, and management of treatment-related complications. Despite the availability of guidelines and templates, SCPs remain underutilized in routine practice due to the time-intensive and error-prone nature of manual creation. Empowering cancer survivors to use publicly available large language models (LLM)–based chatbots, such as ChatGPT, may offer a feasible solution to this problem. Methods: Standardized patient profiles were developed for six common cancers: breast, colorectal, non–small cell lung, small cell lung, prostate cancer, and diffuse large B-cell lymphoma. Profiles were reviewed by a primary care physician and an oncologist to ensure clinical accuracy and representativeness of common survivorship scenarios. Using lay language to simulate real-world patient input, ChatGPT (version 5.2, publicly available at the time of the study) was prompted with each profile using the question, “What follow-up care do I need?” Common survivorship symptoms for each cancer were also tested. Responses were evaluated across five domains: medical accuracy (concordance with current ASCO or NCCN guidelines), thoroughness, patient safety and risk framing, clarity and patient comprehension, and actionability. Five-point Likert scales were used (1 = not at all; 5 = very much). Scores were averaged across two independent physician raters. Readability was assessed using the Flesch Reading Ease Score. Results: ChatGPT-generated SCPs demonstrated high medical accuracy (mean score 4.25/5), with no factual errors or hallucinations identified on manual review. Patient safety and risk framing scored 3.75/5; symptom red flags and care escalation guidance were appropriately stated in most scenarios. Thoroughness was moderate (3.50/5), with key survivorship elements, such as smoking cessation counseling or genetic risk considerations; sometimes omitted unless explicitly prompted. Readability was limited, with a mean Flesch Reading Ease Score of 40.5, corresponding to a college reading level and exceeding recommended readability standards for patient education materials. Clarity (3.38/5) and actionability (3.38/5) were similarly constrained due to dense medical language. Conclusions: ChatGPT demonstrated high medical accuracy in generating survivorship care plans across six common cancers, supporting the feasibility of using publicly available LLMs to guide survivorship care. However, improvements in comprehensiveness, readability, and patient-centered actionability are necessary before such tools can be safely integrated into clinical survivorship workflows.

1633 Background: CAR-T cell therapy has transformed outcomes of hematologic malignancies and is poised to expand into solid malignancies. However, broader use of CAR-T remains constrained by the need for timely cytokine release syndrome (CRS) detection. Current episodic monitoring may miss early physiologic changes. Wearable digital health technologies (DHTs), offering continuous monitoring and real-time alerts, could enable earlier CRS detection, further support outpatient delivery, and expand access to CAR-T and potentially other immunotherapies associated with CRS. Methods: This single-center observational study (NCT05018208) enrolled patients receiving FDA-approved CAR-T products in the outpatient setting at Mayo Clinic, Rochester MN. Management of post-CAR-T complications (e.g. CRS) followed the institution’s standard of care. Patients were monitored from day -5 to day +30 post infusion with intermittent remote patient monitoring (SoC) and an arm-worn wearable device for continuous monitoring (DHT). Patients were included in this analysis if DHT data were available for ≥ 5h of data/day or ≥ 2h of data/night on ≥2 days or nights for baseline and ≥1 day or night within 72h prior to CRS, or at least 2h within 4h of CRS onset, or if DHT was available for ≥7 days/nights post infusion with no CRS event. CRS detection time by DHT was defined as the onset of pyrexia or threshold-based deviation from an individual’s baseline pulse distributions. Results: 32 patients were included in this report (median age 64 years, gender 12 F/20 M, diagnoses 18 MM/14 NHL). 25 patients experienced a CRS event, with maximum CRS grades ranging from 1-3 (16 G1, 8 G2, 1 G3). 24/25 patients with CRS required tocilizumab/steroid. Median CRS onset time was 3.4 days (range 0.4-18.4). 12/32 patients had sufficient DHT data (7 CRS, 5 no CRS) for analysis. 5 patients with CRS had a DHT detection time in advance of the SoC by a median of 13h (range 2.8 - 69.2). The DHT prediction of CRS by baseline HR change was earlier than the first episodic fever by a median of 14.5h (range 13-77.5). The remaining 2/7 patients who did not have CRS detected by the DHT had missing DHT data for more than 45h before a treated CRS event. DHT detected CRS in 2/5 patients without CRS according to SoC. One patient had fever caught with intermittent monitoring. The other was treated for neutropenia fever with suspicion of CRS and/or infection. Conclusions: These results offer evidence that continuous DHT can identify CRS events earlier than intermittent monitoring (SoC). Despite limitations related to compliance with wearable device use and data missingness, and variability in clinical management of CRS, we highlight the usability of DHT and its potential for future studies with optimized strategies to address these limitations. Clinical trial information: NCT05018208 .

1615 Background: Immune checkpoint inhibitors (ICIs) have revolutionized lung cancer treatment but cause adverse events, including immune-related colitis (ir-colitis). Accurate identification of ir-colitis is critical for improving patient outcomes and clinical research. However, current methods relying on ICD codes have substantial limitations: non-specific categories that fail to capture the cases or distinguish cases from infectious or ischemic etiologies, as well as incomplete coding that leads to missed cases. Large language models (LLMs), with advanced natural language understanding capabilities, offer a scalable approach for automated phenotyping using unstructured clinical notes. We aimed to develop and evaluate an LLM-based approach for identifying ir-colitis in patients with lung cancer receiving ICIs and compare it with ICD code based method. Methods: We identified 5,278 lung cancer patients with ICI administration at Mayo Clinic (2012-2025). To investigate ir-colitis occurrence within two years after ICI initiation, patients were stratified into three ICD-based cohorts: (1) IR-Colitis, with clinician-assigned ir-colitis-specific diagnosis codes; (2) Non-IR Colitis, with colitis diagnosis codes excluding ir-colitis; and (3) No-Colitis, absent of any colitis-related codes. We used three open-source LLMs to process clinical notes : Llama4, MedGemma, and Qwen. Prompts were designed to extract both gold standard evidence (explicit ir-colitis terminology, such as "immune-mediated colitis" and "checkpoint inhibitor colitis") and supportive evidence (such as pathology and lab results). Results: Llama4 demonstrated superior accuracy among three LLMs in manual evaluation and was thus selected for all cohort analyses. Among ICD-based IR-Colitis patients (n = 339), 82.0% had confirmatory evidence while 18% lacked evidence, suggesting potential false-positive coding. In the Non-IR Colitis cohort (n = 297), LLM-based extraction identified 43.8% patients in fact had ir-colitis, demonstrating substantial misclassification. In the No-colitis cohort (n = 4,642, random sample of 996 for analysis), the LLM identified 10.7% with documented ir-colitis, revealing substantial missing by administrative coding. Conclusions: LLM-based phenotyping demonstrates superior sensitivity for identifying ir-colitis compared to ICD codes alone, successfully uncovering cases missed by administrative coding. This scalable approach enables more comprehensive case identification for immunotherapy safety research and clinical surveillance. ICD-Based Cohorts LLM-Identified ir-Colitis Gold Standard Evidence Supportive Evidence Any Evidence ir-Colitis (n = 339) 277/339 (81.7%) 275/339 (81.1%) 278/339 (82.0%) Non-ir-Colitis (n = 297) 129/297 (43.4%) 126/297 (42.4%) 130/297 (43.8%) No Colitis (n = 996) 106/996 (10.6%) 104/996 (10.4%) 107/996 (10.7%)

1617 Background: Large language models (LLMs) are increasingly evaluated for oncology decision support, yet guideline concordance and hallucinations remain variable across cancer types and tasks. Retrieval-augmented generation (RAG) may improve safety by constraining the evidence base used during generation. Prior oncology LLM studies report discordance with guideline-based care, but few describe a reproducible health-system “vetting mechanism” that couples guideline-only retrieval with clinician-scored safety rubrics across multiple oncology domains. Methods: We assembled 316 de-identified oncology cases in a tumor-board template: medical oncology (breast, GI, GU, leukemia; n=216), gynecologic oncology (n=50), and neuro-oncology/CNS metastatic cancer (n=50). Each case was run on three systems: baseline GPT-4 (O1), NCCN-guideline anchored RAG (O2; guideline-constrained retrieval), and a literature-anchored system (OpenEvidence; O3). Oncologists independently scored outputs using a modified Generative Performance Score (mGPS; −1 to +1) combining guideline concordance with hallucination penalties; readability/rationality was scored on a 1–5 Likert scale. Inter-rater reliability was assessed (κ =.63). This workflow was designed as a pre-deployment model credentialing step for health-system governance (source policy → case bank → clinician-scored acceptance criteria). Results: Across all 316 cases (case-averaged), O2 achieved the highest mean total mGPS (0.75) compared with O1 (0.43) and O3 (0.59). In the medical oncology cohort (n=216), O2 outperformed O1 and O3 (mean mGPS 0.72 vs 0.35 and 0.54), with higher guideline concordance (0.85 vs 0.68 and 0.75) and lower hallucination penalties (−0.09 vs −0.25 and −0.19); readability was also highest for O2 (4.17 vs 3.22 and 3.86). In gynecologic oncology (n=50), O2 again performed best (mGPS 0.827) relative to O3 (0.697) and O1 (0.647). In the CNS cohort (n=50), mean composite mGPS favored O2 (0.82) over O3 (0.68) and O1 (0.55), while readability remained stable across models (3.55–3.73). Severe hallucination/unsafe recommendation rate: O1 [26%], O2 [5%], O3 [17%] (pre-specified definition; e.g., major fabricated citation or guideline-discordant harmful recommendation). Conclusions: In a 316-case, multi-domain clinician-scored evaluation, limiting retrieval to oncology guideline sources improved safety-aligned performance relative to unconstrained generation or literature-only anchoring. A guideline-constrained RAG configuration paired with clinician-scored rubrics and tail-risk reporting can serve as a practical health-system vetting mechanism for governance, re-validation, and change control of LLM decision support, including in resource-limited settings where reliable guideline-consistent recommendations are essential.

1591 Background: The 21-gene OncotypeDX (ODX) recurrence score (RS) is the standard for chemotherapy decision-making in HR+/HER2- early breast cancer (EBC), but cost limits accessibility in resource-constrained settings. The Magee equations (MEs) use routine pathology features (Nottingham grade and IHC; ER/PR/HER2/Ki-67) to estimate RS. We performed a meta-analysis to validate MEs performance as a triage tool for ODX testing and a qualitative synthesis to identify sources of discordance, highlighting the transition toward objective digital pathology AI models. Methods: A systematic review identified validation studies comparing MEs with RS in HR+ EBC. Studies with sufficient data to construct a 2X2 contingency table were included; neoadjuvant trials were excluded. Primary outcomes were Negative Predictive Value (NPV) for identifying low-risk cases (RS <26 for the current TAILORx cutoff and RS<31 for the historical cutoff) and ODX test sparing rate (TSR). Data were pooled using random-effects models. Qualitative themes regarding interpretation variability were extracted from study results and discussion sections. Results: Thirteen studies met the inclusion criteria, representing 5396 patients in global cohorts from the USA, Mexico, Colombia, Jordan, France, Belgium, and Canada. For RS >26, Magee <18 achieved pooled NPV of 0.96 (95% CI: 0.94-0.97) and TSR 61% (95% CI:51%-69%). For RS >31, NPV increased to 0.99 (95% CI: 0.97-1.00) and TSR 52% (95% CI: 44%-59%). Pooled diagnostic performance for Magee <18 predicting RS <26 demonstrated 83% sensitivity and 71% specificity (AUC 0.81, LR+:2.53, LR-:0.28). Qualitative analysis of discordance identified: (1) Inter-observer variability: Nottingham mitotic counts and Ki-67 scoring were the primary drivers of score fluctuations. (2) IHC quantification: Variations in H-score calculation and Allred-to-H-score conversion introduced heterogeneity (3) Pre-analytical and biological factors: Inflammation and stromal proliferation occasionally inflated genomic RS; selection of non-representative blocks and intratumor heterogeneity further contributed to discordance (4) Technological solutions: Digital pathology and AI models emerge as solutions to standardize these variables. Conclusions: Magee-based triage provides a safe and cost-effective strategy for HR+ EBC risk stratification, with NPV up to 99% for excluding high-risk disease and sparing up to 61% of ODX testing. These results confirm a robust histopathological signal. Transitioning to objective digital pathomics provides a pathway to reduce manual subjectivity and democratize precision oncology globally. Pooled diagnostic accuracy of Magee equations by cutoff. Magee score and RS cutoff Number of studies Total patients Pooled NPV (95% CI) ODX test sparing rate (95% CI) Magee <18/ RS>26 8 4153 0.96 (0.94-0.97) 61% (51%-69%) Magee <18/ RS>31 5 1243 0.99 (0.97-1.00) 52% (44%-59%)

1587 Background: Rehabilitation services (physical medicine and rehabilitation [PM&R], physical therapy [PT], occupational therapy [OT], and speech-language pathology [SLP]) improve physical function, symptom burden, and quality of life for cancer survivors, yet utilization remains low due to persistent barriers. The collaborative care model–based Enhanced Electronic Health Record–Facilitated Cancer Symptom Control (E2C2) pragmatic trial improved symptom burden and palliative care utilization. Using post hoc exploratory analyses of population-level trial data, we evaluated whether this EHR-facilitated approach also increased rehabilitation service use. Methods: E2C2 was conducted across Mayo Clinic Midwest medical oncology practices using an unblinded, stepped-wedge, cluster-randomized cohort design. The EHR-embedded intervention targeted SPPADE symptoms (sleep interference, pain, impaired physical function, anxiety, depression, and fatigue). Symptoms and physical function were assessed using 11-point numerical rating scales linked to oncology encounters. Patients received self-management materials for moderate symptoms (4–6/10) and care manager support for severe symptoms (7–10/10). All outpatient rehabilitation encounters across the Mayo Clinic Enterprise were abstracted. Mixed-effects Poisson models estimated intervention effects on rehabilitation visits during the first 365 days of exposure. Results: Between March 28, 2019, and January 31, 2023, 50,207 patients were enrolled. Intervention and control conditions included 31,719 and 28,938 patients, respectively, with 10,450 contributing data to both periods. Cross-sectionally, 25% of participants reported moderate and 9% severe physical function impairment. Of 91,220 rehabilitation encounters scheduled, 57,669 (63.3%) were completed. Completed encounters were predominantly PT (68.3%), followed by OT (16.4%), physician visits (9.4%), and SLP (4.2%). The intervention was associated with a more than fourfold increase in completed rehabilitation encounters during the six weeks following index symptom assessment (adjusted mean 2.22 [95% CI, 0.14–4.29] vs 0.54 [0.40–0.68]; incidence rate ratio 4.11 [1.43–11.82]; P =0.009). Scheduled encounters also favored the intervention but did not reach statistical significance. Conclusions: An EHR-embedded symptom surveillance intervention increased short-term rehabilitation service completion, but overall utilization remained low relative to the prevalence of physical function impairment, highlighting the need for system-level strategies to translate symptom identification into sustained rehabilitation engagement.

1557 Background: Immunotherapy is now a standard-of-care treatment for lung cancer across most stages. Access to immunotherapy and molecular diagnostic tests in conflict-affected countries such as Syria remains poorly described. As the principal cancer center in Syria, Al-Bairouni University Hospital (ABUH) provides care for nearly three-quarters of all cancer cases nationally. We conducted this study to evaluate access to immunotherapy and molecular diagnostic services at ABUH. Methods: A mixed-methods study was performed between 2022 and 2024 at ABUH. This study included a retrospective cohort of all patients diagnosed with lung adenocarcinoma. Quantitative data included disease stage at diagnosis, molecular testing rates, and immunotherapy utilization. In addition, a qualitative patient survey was thematically analyzed to explore barriers to treatment access in a separate cohort of patients with mixed tumor types. Results: Among 2,920 patients diagnosed with lung cancer, 1,393 (47.7%) had lung adenocarcinoma. The median age at diagnosis was 61 years; 79.7% of patients were male. Smoking history was as follows: Current 792 (56.9%), Former 268 (19.2%), never smoked 318 (22.8%) At presentation, 3 patients (0.2%) had stage I disease, 40 (2.9%) stage II, 174 (12.5%) stage III, and 778 (55.8%) stage IV disease, staging was unknown for the rest of the patients. EGFR mutation testing was performed by ABUH for 226/1393 patients (16.2%) only, with positive results identified in 124 cases (8.9%). PD-L1 testing was conducted by ABUH in only 32/1393 patients (2.3%) and was positive in 23 (1.65%). Of all patients with lung adenocarcinoma, only 5 patients (0.6%) with stage IV disease received immunotherapy. All were treated with nivolumab. Thematic analysis of qualitative data (35 patients) revealed that high treatment costs and limited drug availability are the primary barriers to accessing immunotherapy. Conclusions: This study represents the first evaluation of access to immunotherapy for patients with lung cancer in Syria. Even at ABUH, the country's main cancer center, access to immunotherapy and molecular testing remains markedly limited. International collaboration is necessary to expand diagnostic infrastructure, standardize data reporting, and enhance financial protection mechanisms. These are critical steps toward achieving better cancer care in Syria.

1584 Background: Decades of evidence demonstrate that patients across the cancer continuum may benefit from validated rehabilitation services, yet delivery remains inconsistent. Prior reports of underuse have largely relied on small, single-site cohorts and have lacked population-level estimates of service need, referral, and visit completion, contributing to the assumption that provider reluctance to refer is a primary driver. Using post hoc exploratory analyses of population-level trial data, we evaluated patient- and system-level factors associated with non-use of rehabilitation services. Methods: This secondary analysis used data from the Enhanced Electronic Health Record–Facilitated Cancer Symptom Control (E2C2) pragmatic trial, an unblinded, stepped-wedge, cluster-randomized cohort study comparing an EHR-embedded symptom surveillance and management system with usual care. All patients receiving care in Mayo Clinic Midwest medical oncology practices were enrolled regardless of cancer stage or phase. Symptoms and physical function were assessed using 11-point numerical rating scales linked to oncology encounters. Sociodemographic and clinical characteristics and all outpatient rehabilitation encounters across the Mayo Clinic Enterprise were extracted from the EHR. Rates and types of completed, canceled, and no-show rehabilitation encounters were examined. Results: Between March 28, 2019, and January 31, 2023, 50,207 patients were enrolled; 34% reported moderate or severe physical function impairment during follow-up. Despite this, only 8,818 patients (17.6%) completed at least one rehabilitation encounter. Of 91,220 encounters scheduled, 57,669 (63.3%) were completed (mean 6.5; median 3 per participant), leaving a substantial proportion canceled or no-showed (n=33,523, 36.7%). Completed encounters were predominantly physical therapy (68.3%), with fewer occupational therapy (16.4%), physician (9.4%), and speech therapy (4.2%) visits. Encounters were concentrated among patients with breast, hematologic, gastrointestinal, and genitourinary cancers. Completion was more likely among patients who were older, female, partnered, retired, and federally insured, whereas cancellations and no-shows were more common among younger, racial or ethnic minority, urban, unpartnered, employed, college-educated, and commercially insured patients (all P <0.001). Virtual visits and encounters in community (vs academic) settings were more likely to be completed (all P <0.001). Conclusions: A substantial gap persists between cancer survivors’ rehabilitation needs and service receipt, varying by cancer type and sociodemographic factors. High cancellation and no-show rates suggest that underuse extends beyond referral and underscore the need for redesigned, more accessible rehabilitation delivery models.

1564 Background: Adolescent and young adult (AYA) females with early-stage ER-negative and/or HER2-positive breast cancer commonly receive chemotherapy to reduce risk of recurrence. Fertility concerns, counseling, and preservation may influence chemotherapy decisions and time to chemotherapy (TTC) initiation. Methods: We surveyed AYA survivors of stage I–III premenopausal ER-negative and/or HER2-positive breast cancer treated at Kaiser Permanente Northern and Southern California (2013-2022). Fertility concerns were defined as not having completed desired family size or being unsure about having completed desired family size at diagnosis. We summarized receipt of fertility counseling, receipt of chemotherapy, and use of oocyte/embryo cryopreservation for fertility preservation (FP). Among fertility-concerned patients who received chemotherapy, TTC (days from diagnosis) was assessed separately for neoadjuvant chemotherapy (NAC) vs for adjuvant chemotherapy (AdjC). Time to NAC was natural log–transformed and modeled with multivariable log-linear regression to estimate relative differences in TTC, adjusting for stage, study site, ER, HER2, age, and diagnosis year. Results: Of 363 respondents, 197 (54%) had fertility concerns at diagnosis; 183/197 (93%) received chemotherapy, similar to 153/166 (92%) among those without concerns (p=0.7). Of the concerned chemotherapy recipients, 45 (25%) saw a fertility specialist but did not pursue FP while 55 (30%) underwent FP. For the 112 fertility-concerned NAC recipients, FP was associated with longer TTC than no fertility specialist counseling (mean=41 vs. 30 days; adjusted ratio=1.39; 95% CI 1.15–1.69), whereas fertility specialist counseling without FP was not (mean= 31 vs. 30 days, adjusted ratio=1.04; 95% CI 0.86–1.25). For the 71 fertility-concerned AdjC recipients, TTC was similar for patients who did not see a fertility specialist (mean=63 days), those who froze oocytes/embryos (mean=65 days), and those who saw a fertility specialist but did not pursue FP (mean=69 days). Conclusions: In this AYA breast cancer cohort, there was no evidence that fertility concerns led patients to decline chemotherapy. FP was associated with a modest delay in NAC initiation (<2 weeks on average). Time to chemotherapy (TTC) among 183 AYA patients with ER− and/or HER2+ breast cancer and fertility concerns. Fertility-focused care N (n=183) Neoadjuvant chemotherapy (n=112) Mean TTC, days Relative difference in TTC, adjusted ratio (95%CI) Adjuvant chemotherapy (n=71) Mean TTC, days No fertility counseling from a specialist 83 (45%) 55 (49%) 29.6 Reference 28 (39%) 62.6 Received fertility counseling from a specialist No fertility preservation 45 (25%) 28 (25%) 30.7 1.04 (0.86,1.25) 17 (24%) 68.6 Fertility preservation 55 (30%) 29 (26%) 41.2 1.39 (1.15,1.69) 26 (37%) 64.6

1503 Background: CAR T-cell therapy has revolutionized treatment outcomes in hematological malignancies, but has traditionally required inpatient post-infusion monitoring, which is associated with significant cost and time toxicity. Outpatient CAR T infusion and monitoring has been the standard approach at Mayo Clinic site in Rochester, MN, with demonstrated feasibility and safety (Bansal et al., ASCO 2023). We present data comparing healthcare utilization between CAR T recipients who underwent outpatient remote monitoring and those with traditional inpatient monitoring. Methods: Electronic medical records were retrospectively analyzed for 293 patients who received CAR T-cell therapy for a hematological malignancy at any of the 3 Mayo Clinic sites (MN, AZ, FL) between 2020 and 2024 . Healthcare utilization was compared between patients who received outpatient CAR T with remote patient monitoring in Rochester, MN (N = 125) and those who received inpatient CAR T in the Arizona (n = 90) and Florida (n = 78) sites. Per protocol, patients in the inpatient group underwent mandatory inpatient monitoring for 7 days, with extension per provider discretion. Data was analyzed through Chi-Square and Wilcoxon tests. P values < 0.05 were considered statistically significant. Results: There was no significant difference in baseline sex, age, ethnicity, LDH, platelets, or neutrophils among the 293 patients. Diagnoses included lymphoma (n = 175), multiple myeloma (n = 106), and B-ALL (n = 12). The most common CAR T-cell products used were axicabtagene ciloleucel (n = 138) idecabtagene vicleucel (n = 56), and ciltacabtagene autoleucel (n = 50). Median hospital days were lower in the outpatient group vs the inpatient group at 30 days (4.4 vs 13.3, P < 0.001) and 90 days (4.6 vs 13.6, P < 0.001). ICU admission occurred in 1 outpatient patient ( < 1%) vs 18 inpatient patients (11%) in the first 30 days (P < 0.001). In the first 30 days, 22% of patients in the outpatient group did not require hospitalization, while 56%, 15%, and 6% required 1, 2, and 3 hospitalizations, respectively. Similarly, only 11 (8.8%) patients in the outpatient group had ≥1 ED visit in the first 30 days compared with 32 (20%) pts in the inpatient group (P = 0.009). At 30 days, there was no difference in patient portal use, with 58% of patients using the portal at least once and a median of 1 message in both groups. Within the first 30 days, 53% and 2% of patients in the outpatient group required 1 and 2 outpatient visits, respectively, vs 29% and 2% in the inpatient group (P < 0.001). There was no difference in 30-day mortality (1.6% outpatient vs 2.4% in inpatient) between the 2 groups (P = 1.00) Conclusions: Outpatient CAR T-cell therapy monitoring was associated with significantly fewer hospital days without increased ED visits, ICU stays, portal use, or 30-day mortality. This demonstrates lower healthcare utilization for outpatient CAR-T therapy.