Citations

e15509 Background: Large language models (LLMs) are increasingly explored for oncology decision support, yet comparative evidence evaluating medical-specific versus general-purpose LLMs using expert-reviewed clinical cases is limited. We compared a medical-specific LLM (Google MedGemma 27B) with a general-purpose LLM (ChatGPT-5) using gastrointestinal (GI) and genitourinary (GU) oncology vignettes. Methods: This cross-sectional study evaluated 32 de-identified oncology vignettes (26 GI, 6 GU) created by a U.S. board-certified medical oncologist and informed by real-world cases. Vignettes included text-only information on the clinical context, description of pathology, tumor staging, biomarkers, and relevant comorbidities. Both LLMs were prompted to provide National Comprehensive Cancer Network(NCCN) guideline-based treatment recommendations, monitoring strategies, maintenance therapy considerations, and references, along with a self-reported confidence score (0–100). Board-certified U.S. medical oncologists independently reviewed outputs using a structured instrument; two disease specific expert oncologists evaluated each case. Outcomes included correctness and comprehensiveness (1–10 scales), hallucinations (factually fabricated or unsupported statements), NCCN traceability, potential patient harm, and clinician trust. Associations between accuracy and comprehensiveness were assessed using correlation coefficients. Inter-rater reliability was assessed using a two-way random-effects intraclass correlation coefficient (ICC). Results: ChatGPT-5 demonstrated higher mean correctness (8.65 vs 4.68) and comprehensiveness (8.68 vs 5.22) scores compared with MedGemma 27B. Accuracy and comprehensiveness were strongly correlated overall (Pearson r = 0.93, p < 0.001), with significant model-specific correlations (ChatGPT-5 r = 0.86, p < 0.001; MedGemma r = 0.77, p < 0.001). Hallucinations occurred less frequently with ChatGPT-5 (12.3% vs 65%), and NCCN-traceable recommendations were more common (75.4% vs 36.7%). Responses judged as potentially harmful occurred in 18.5% of ChatGPT-5 outputs versus 70% of MedGemma outputs. Clinician trust favored ChatGPT-5 (84.6% vs 15%). Inter-rater reliability improved with rater averaging (ICC(A,k): correctness 0.72; comprehensiveness 0.64). Conclusions: In expert-reviewed GI and GU oncology cases, ChatGPT-5 demonstrated higher clinical accuracy, comprehensiveness, and guideline alignment than a medical-specific LLM; however, a substantial rate of hallucinations and potentially harmful recommendations persisted across models. These findings underscore that, despite performance differences, current LLMs require rigorous clinician oversight and should not be used as standalone decision-support tools in oncology.

9030 Background: AI use is ubiquitous among HOT and healthcare workers, but with variable base knowledge and comfort level. Its indiscriminate use may be problematic. To address this educational gap, we developed a WS for HOT to assess their use of AI and educate them on how to employ it in MedEd and patient-centered CP based on LSP. Methods: A faculty member developed a WS on how to use AI in MedEd and CP based on LSP of analogy, contrasting cases, elaboration, generation, and question driven learning. The WS was for HOT, but faculty could join. It had 3 sections: introduction to basic concepts of AI; principles of the Health Insurance Portability and Accountability Act and protected health information; AI in MedEd; and AI in CP. Sections had real use case examples of AI, for participants to interact, ask questions, and share their personal use cases. HOT were invited to compare and generate examples during the WS, based on their educational and clinical experience and needs. Participants’ AI use and self-perceived and objective knowledge about it were assessed in a pre activity survey, a post WS survey, and a post 8 week survey to evaluate long term learning. Results: A 2 hour WS with 9 HOT and 2 faculty was completed successfully. 10 participants completed the pre WS survey, and 11 completed the post WS and the post 8 week surveys. 100% of them used generative AI tools before the WS, while 10% had prior training in AI. 90% used AI in personal life and MedEd, 70% used it in CP, and 30% in academic research. Table 1 shows participants’ self-assessed knowledge of AI, comfort level in using AI for MedEd and CP, self-assessed preparedness to implement AI in MedEd and CP, and objective knowledge. 8 weeks post WS, 91% of participants reported increased AI use in CP, leading anywhere from slight to significant decreases in time spent on documentation tasks for 91% of them. In MedEd, 73% reported increased AI use, and 82% described slight to significant improvement in the quality of time dedicated to it. 82% described the WS as extremely relevant to their daily responsibilities, and 91% reported confidence navigating ethical and privacy risks of AI in HemOnc. All participants were interested in having a similar WS in the future. Conclusions: An LSP based WS about AI increased the self-perceived and objective knowledge of HOT about using this tool in MedEd and CP. It also increased HOT comfort level, preparedness, and efficiency using AI. With the ubiquitous use of AI by HOT, ongoing education about it focused on supporting high-quality patient care should be a part of fellowship curriculum. Pre WS % Post WS % Post 8 week % Moderately/Very knowledgeable about AI 50 91 100 Comfortable using AI in MedEd 60 82 82 Prepared to implement AI in MedEd 30 91 73 Comfortable using AI in CP 40 73 82 Prepared to implement AI in CP 0 91 73 Correct answers on objective assessment 53 68 71

Abstract Background and Aims Truncating variants in the TTN gene (TTNtv) are the most common genetic cause of dilated cardiomyopathy (DCM) but also occur as incidental findings in the general population. This study investigated factors associated with the clinical manifestation of TTNtv. Methods An international multicentre retrospective observational study was performed in families with TTNtv-related DCM. Shared frailty models were used to estimate associations of variant characteristics with lifetime risk of DCM, and logistic regression to estimate odds ratios (ORs) for individual-level clinical risk factor profiles (cardiac conditions, cardiovascular comorbidities, lifestyle) and DCM. Results A total of 3158 subjects in 1043 families with TTNtv-related DCM were studied. TTNtv-positive subjects were 21-fold more likely to develop DCM [OR, 21.21; 95% confidence interval (CI), 14.80–30.39]. Disease onset was earlier in males, but was similar for TTNtv of different types and locations. The presence of clinical risk factors was associated with earlier DCM onset (OR, 3.41; 95% CI, 2.06–5.64), with a prior history of atrial fibrillation having a two-fold increased odds of DCM (OR, 2.05; 95% CI, 1.27–3.32). The prevalence of clinical risk factors increased with age; however, the strength of the DCM association was greatest for young-onset (&lt;30 years) disease (OR, 4.75; 95% CI, 2.35–9.60). Administration of beta-adrenergic receptor or renin-angiotensin system-blocking drugs prior to overt DCM was associated with 87% reduced odds of DCM (OR, .13; 95% CI, .08–.23). Conclusions Disease onset in TTNtv-associated familial DCM is dependent on individual patient context and is potentially modifiable by risk factor management and prophylactic therapeutic intervention.

Abstract The retrosigmoid approach (RSA) is considered the standard surgical choice to address a neurovascular conflict involving the trigeminal nerve. The subtemporal transtentorial approach (STA) represents the main alternative. We describe here the mini-invasive anterior petrosal intertentorial approach with a view to reduce complications associated with the standard STA. Mini-invasive anterior petrosal intertentorial approach (M-APIA) was described for a case of refractory trigeminal neuralgia. This approach includes a very limited amount of drilling of the petrous ridge to expose the posterior fossa dura and the superior petrosal sinus to facilitate the subsequent tentorial peeling. Tentorial peeling enables to develop an intertentorial plane between the temporal tentorial leaf (TTL) and the posterior fossa tentorial leaf (PFTL). A linear incision over the posterior fossa dura and the PFTL offers a wide exposure of the neurovascular conflict. M-APIA enables to expose the root entry zone (REZ) of the trigeminal nerve without exposing the temporal lobe or the cerebellum. M-APIA can be considered as a safe alternative to the RSA or STA for trigeminal microvascular decompression (MVD), especially for select cases. Additional studies are needed for a better understanding of its advantages and disadvantages.

Abstract This retrospective analysis aimed to provide evidence-based risk stratification of TP53-mutated (TP53mut) myeloid neoplasms (MNs). Of 580 MNs harboring TP53mut with variant allele frequency (VAF) ≥2%, 219 (37.8%), 194 (33.4%), 92 (15.9%), and 75 (12.9%) were classified as acute myeloid leukemia (AML), myelodysplastic syndrome (MDS) with low blasts (MDS-LB), MDS with excess blasts (MDS-EB)-2, and -EB1 according to the revised fourth edition of the World Health Organization (WHO) classification, respectively. Hierarchical analysis identified the following 4 risk groups with distinct survival: (1) MDS-LB, (2) MDS-EB1/EB2/AML VAF &lt;10%, (3) MDS-EB1/EB2 VAF ≥10%, and (4) AML VAF ≥10%. We next evaluated the impact of allelic status, VAF, and complex karyotype (CK). In our cohort, the significance of biallelic status was limited to MDS with &lt;5% blasts and not for blasts 5% to 9%, as proposed by the International Consensus Classification (ICC), or 5% to 19%, as proposed by the fifth edition of the WHO (WHO-5). MDS-EB1 and -EB2 with VAF ≥10% had comparable survival (9.6 vs 7.2 months; P = .12), regardless of allelic status. Contrary to the ICC proposal, MDS-EB1/EB2 with VAF &lt;10% and CK had poor survival compared with those without CK, comparable to MDS-EB1/EB2 with VAF ≥10% (5.6 vs 26.2 vs 6.3 months; P = .003). Survival of TP53mut AML was poor (median 3.9 months) regardless of allelic/CK status. Thus, using ICC or WHO-5 may underestimate prognosis of MDS with blasts 5% to 19% and 5% to 9%, respectively. Collectively, the hierarchical model acknowledges poor survival of 91.9% TP53mut MDS and AML compared with 36.5% and 80.7% by WHO-5 and ICC, respectively.

Background: TP53-mutated (TP53mut) myeloid neoplasms (MN) have particularly poor survival. Discrepancies in defining TP53mut MN have impact on diagnostic work up, therapeutic decisions, and clinical trial enrolment. Methods: MN cases harboring TP53mut (variant allele frequency, VAF ≥2%) were classified based on bone marrow (BM) and peripheral blood (PB) blast % using the 4th revision of the WHO classification (WHO-4R). Biallelic loss was defined as (i) ≥2 mutations each with VAF ≥2%; (ii) single TP53mut VAF ≥2% plus 17p loss detected on karyotype, FISH, or SNP; or (iii) single TP53mut VAF ≥50%, whereas monoallelic loss was defined as single TP53mut VAF &lt;50% without 17p loss. Primary endpoint was overall survival (OS) from diagnosis. Conditional Inference Trees (CIT) analysis was used to determine the true effect of a predictor. Results: We identified 580 cases including WHO-4R defined acute myeloid leukemia (AML, BM/PB blasts ≥20%, n=219, 37.8%), myelodysplastic syndrome (MDS) with excess blasts-2 (EB-2, BM 10-19% or PB 5%-19%, n=92, 15.9%), MDS with excess blasts-1 (EB-1, BM 5-9% or PB 2-4%, n=75, 12.9%), and the remaining were MDS with low blasts (-LB, BM &lt;5% and PB ≤1%, n=194, 33.4%). Of the 578 cases evaluable for allelic status, 437 (75.6%) and 141 (24.4%) had biallelic and monoallelic TP53 loss, respectively. Higher frequency of biallelic loss was observed with increasing blasts. 63.2%, 73.3%, 78% and 86.3% of cases with MDS-LB, MDS-EB1, MDS-EB2, and AML had bi-allelic TP53 loss (P&lt;0.0001). In a multivariable analysis, WHO-4R categories, VAF ≥10%, and biallelic loss were independent predictors of poor survival. CIT analysis further stratified the cohort into 4 groups with distinct survival: (1) MDS-LB (n=194, 33.4%); (2) MDS-EB1/EB2 and AML with VAF &lt;10% (n=38, 6.6%); (3) MDS-EB1/EB2 with VAF ≥10% (n=149, 25.7%); and (4) AML with VAF ≥10% (n=199, 34.3%). MDS-LB cases could be further stratified based on allelic status and complex karyotype (CK): OS was poor for cases with biallelic TP53 loss and monoallelic TP53mut with CK compared to monoallelic TP53mut without CK, who had the most favorable survival (12.7 vs. 13 vs. 34.8 months, P&lt;0.0001).MDS-EB1/EB2/AML with VAF&lt;10% could also be stratified by CK: MDS-EB1/EB2/AML with VAF&lt;10% cases without CK had the favorable survival compared to their counterparts with CK and MDS-EB1/EB2 with VAF ≥10% cases (26.2 vs. 5.6 vs. 6.3 months, P=0.003).MDS-EB1/EB2 with VAF ≥10% was a homogenous subgroup: Biological characteristics including the proportion of cases with biallelic TP53 loss, CK (P&gt;0.9), monosomal karyotype (MK, P&gt;0.9), del 5q (P=0.6), del 7q (P=0.4), and del 17p (P=0.2) were comparable between MDS-EB1 and -EB2 with VAF ≥10%. Importantly, median OS of MDS-EB1 and EB2 with VAF ≥10% was comparable (9.6 vs. 7.2 months, P=0.12). OS in this group was comparable regardless of the blast %, allelic status, or presence of CK.AML with VAF ≥10% had extremely poor survival: Median OS was 3.9 months. Survival of AML was equally poor irrespective of the mono- or biallelic (3.6 vs. 3.9, P=0.89) and non-CK vs. CK (1.8 vs. 3.9 months, P=0.09). Combined, our model would acknowledge poor survival of 514 (89.1%) cases. In contrast, retrospective application of the 5th edition of the WHO classification (WHO-5) and International Consensus Classification (ICC) would classify only 211 (36.4%) and 468 (80.7%) cases as biallelic/multi-hit TP53mut MN, respectively. Finally, the Harrell's c-index of the proposed model, WHO-5, and ICC was 0.66, 0.571, and 0.617, respectively. Conclusion: We provide evidence for subsequent refinements in classifying MN harboring TP53mut. Our salient findings include: (i) acknowledgement of CK as ‘biallelic equivalent’ in MDS-LB; as well as EB1/EB2/AML with VAF 2 to &lt;10% cases; and (ii) demonstration of poor survival of MDS-EB1 and -EB2 with VAF ≥10% regardless of the allelic status. In conclusion, our proposed model acknowledged poor survival of a higher proportion of cases as TP53mut MN compared to the current models with a higher c- index.

Background: TP53 mutations (TP53mut) are highly prevalent in therapy-related myeloid neoplasms (t-MN) compared to de novo MN (35-40% vs 7-13%) and are associated with a poor outcome. Understanding the risk factors involved in the pathogenesis of TP53mut t-MN is critical to prevent its development. Methods: We characterized t-MN cases by examining primary cancer type vs autoimmune disease diagnosis vs solid organ transplantation, treatment modality, and latency (interval between primary cancer/disease and t-MN diagnosis) to identify patient and disease-related factors associated with TP53mut t-MN. Allelic loss of TP53mut MN was defined according to International Consensus Classification (ICC) criteria (Arber et al., Blood 2022). Results: Of 587 t-MN patients, 33% (n=194) had acute myeloid leukemia (t-AML) and 67% (n=393) had myelodysplastic syndrome (t-MDS, blasts&lt;20%). Primary diagnosis comprised of 243 (41.6%) solid cancers and 300 (51.4%) hematological malignancies, while 7% (n=41) of t-MN cases had autoimmune disease or underwent solid organ transplant. Overall, 94 (16.1%) patients received prior radiotherapy (RT) alone and 75.6% (n=444) received chemotherapy (CT) with/without RT. 109 patients (18.7%) underwent autologous stem cell transplant (ASCT). The median latency from primary diagnosis to t-MN was 83.2 months (IQR 41.3, 153.2), and there was a trend towards shorter latency in t-AML compared to t-MDS (70.5 vs 85.8 months; P=0.07). Of the t-MN patients, 280 (47.7%) harbored TP53mut, which was associated with significantly poorer survival compared to wild-type (TP53wt) t-MN (9.7 vs 21.9 months; P &lt;0.001). In contrast to TP53wt (59.0% vs 41.0%), the majority of TP53mut t-MN presented with t-MDS vs t-AML (75.7% vs 24.3%; P &lt;0.001). According to ICC criteria, multi-hit MDS (n=131; 47.8%) was the most common entity, followed by AML (n=68; 24.8%), MDS/AML (n=34; 12.4%), and single-hit MDS (n=8, 2.9%). Poor prognostic karyotype changes e.g., del 5q, del 7q, complex karyotype and monosomal karyotype, were more prevalent in TP53mut compared to TP53wt t-MN. Co-mutations in other genes were enriched in TP53wt t-MN. As TP53mut t-MN is associated with a dismal outcome, we assessed the factors associated with its development. There were no significant differences in median age at primary cancer/disease diagnosis (59.0 vs 59.9 years; P=0.4) or gender (females 43.3% vs 39.3%; P=0.3) between TP53mut and TP53wt t-MN. However, TP53mut patients were older at t-MN diagnosis (69 vs 67.4 years; P=0.05). Correlations between gene mutations and prior malignancies revealed a significant association between TP53mut and hematological primary cancers (odds ratio [OR] 1.6; P=0.006), especially multiple myeloma (MM, OR 4.9; P&lt;0.001) and lymphoproliferative disease (LPD, OR 2.0; P=0.013). Among solid cancers, head and neck malignancies were significantly associated with increased risk of developing TP53mut t-MN (OR 7.7; P=0.01). Compared to RT alone, ASCT (OR 2.2; P=0.004) and CT (OR 2.0; P=0.008) were associated with an increased risk of TP53mut t-MN. Recent studies indicate that CT/RT confer a selective advantage to preexisting TP53mut hematopoietic stem and progenitor cells, allowing clonal expansion/evolution and progression to t-MN. We hypothesized that TP53mut t-MN with high-risk disease features is a function of TP53mut clonal expansion/evolution rate and would predict latency between primary cancer and t-MN. Indeed, the latency between primary cancer and t-MN diagnosis was longer in TP53mut (87.8 months, IQR 45.5, 154.1) compared to TP53wt (74.7 months, IQR 37.5, 146.9 months) t-MN (P=0.05). Interestingly, 54% and 29% of TP53mut t-MN had a latency of ≥5 and ≥10 years, respectively. There were no correlations between latency and TP53 variant allele frequency (VAF) (r=0.38 and P=0.7) or between TP53mut t-MDS and t-AML (85.2 vs 109.3 months; P=0.9). Together, latency between primary cancer and t-MN was not associated with TP53mut t-MN phenotype, VAF or other poor-risk disease features. Conclusion: Hematological malignancies, especially MM and LPD treated with high-dose CT, were strongly associated with TP53mut t-MN development, highlighting the importance of careful and ongoing screening in this population. Also, the long latency of TP53mut t-MN provides an opportunity to devise effective surveillance strategies and preventative interventions long before TP53mut t-MN development.

Tubulin, one of the most abundant cytoskeletal building blocks, has numerous isotypes in metazoans encoded by different conserved genes. Whether these distinct isotypes form cell type– and context-specific microtubule structures is poorly understood. Based on a cohort of 12 patients with primary ciliary dyskinesia as well as mouse mutants, we identified and characterized variants in the TUBB4B isotype that specifically perturbed centriole and cilium biogenesis. Distinct TUBB4B variants differentially affected microtubule dynamics and cilia formation in a dominant-negative manner. Structure-function studies revealed that different TUBB4B variants disrupted distinct tubulin interfaces, thereby enabling stratification of patients into three classes of ciliopathic diseases. These findings show that specific tubulin isotypes have distinct and nonredundant subcellular functions and establish a link between tubulinopathies and ciliopathies.

Abstract High prevalence of IDH mutations in seronegative rheumatoid arthritis (RA) with myeloid neoplasm, elevated 2-hydroxyglutarate, dysregulated innate immunity, and proinflammatory microenvironment suggests causative association between IDH mutations and seronegative RA. Our findings merit investigation of IDH inhibitors as therapeutics for seronegative IDH-mutated RA.