Citations

Background Graft-versus-host disease (GVHD) remains a major determinant of morbidity and mortality following allogeneic hematopoietic stem cell transplantation (allo-HSCT). Existing GVHD prediction models demonstrate modest discrimination and limited generalizability, and calibration drift across external populations is rarely characterized despite its essential role in the clinical interpretability of predicted probabilities. Objectives To develop and externally validate an explainable machine learning framework for predicting acute and chronic GVHD and associated overall survival in patients with acute myeloid leukemia (AML), acute lymphoblastic leukemia (ALL), and myelodysplastic syndromes (MDS) undergoing allo-HSCT, and to systematically characterize calibration across heterogeneous external validation cohorts to inform deployment requirements. Study Design The model was developed on three publicly available registry-derived datasets (N = 2,509) and externally validated across six independent cohorts (N = 14,788) comprising adult and pediatric allo-HSCT recipients, including a regional Middle Eastern cohort (UAE and Jordan). A standardized preprocessing pipeline harmonized heterogeneous datasets. Gradient boosting models (CatBoost) were used for binary GVHD prediction; exploratory overall survival analysis used a Cox proportional hazards model with predicted acute GVHD risk as a covariate. Discrimination (AUROC with bootstrap 95% CI), calibration (logistic recalibration intercept and slope with analytical 95% CI), and feature importance (SHapley Additive exPlanations, SHAP) were assessed in training out-of-fold and all external cohorts. Results In internal validation, AUROC was 0.63 (95% CI 0.61–0.65) for acute GVHD and 0.72 (95% CI 0.70–0.74) for chronic GVHD. External validation demonstrated AUROC ranges of 0.51–0.57 (acute) and 0.54–0.64 (chronic), with consistent performance across disease subgroups despite substantial heterogeneity in transplant practices and feature availability. In exploratory survival analysis, the acute-GVHD-informed Cox model achieved a training-cohort C-index of 0.679 (95% CI 0.658–0.697); external C-indices ranged from 0.47–0.53. Calibration analysis identified systematic external risk overestimation (negative calibration intercept in 10 of 11 evaluable external cohort-target combinations) with heterogeneous slope drift requiring cohort-specific recalibration. Key predictors included recipient age, graft source, conditioning intensity, GVHD prophylaxis, and HLA match ratio. Conclusions An explainable, externally validated GVHD prediction framework was developed using heterogeneous registry-derived datasets, with systematic characterization of calibration drift across multiple external cohorts, an analysis rarely reported in prior GVHD prediction literature. Predictive performance was modest for acute GVHD and moderate for chronic GVHD, constrained by missing immunobiological variables and incomplete HLA characterization. Per-cohort recalibration is required before clinical deployment, with prospective validation and benchmarking against established GVHD risk scores identified as priority next steps.

Abstract Rationale Long COVID (LC) has affected approximately 400 million people worldwide, with about 20 million in the US. LC is characterized by a host of new symptoms, including neuropsychiatric disorders including cognitive impairment. Disruption of the gut microbiome has been correlated with symptoms of LC. This study examined whether gut microbiome modulation via intake of probiotic fermented foods mitigated psychiatric and cognitive impairments. Methods This was a pilot trial performed at a single academic medical center that enrolled and randomized patients with LC 1:1 into the treatment vs. control arm. Those enrolled into the treatment arm received teaching on the gut microbiome, its relationship to LC, and were encouraged to eat a diet high in probiotic fermented foods for 12 weeks. Those in the control arm did not receive the teaching and were encouraged to eat their normal diet. Diets were tracked using the Mayo Clinic Diet application on a smart phone at least three days per week, and those in the treatment arm received routine calls to discuss protocol adherence and gather subjective feedback on the trial. Neuropsychiatric questionnaires (HADS, IES-R, EQ-5D-5L, and MoCA Blind) were completed at enrollment and upon completion of the trial. Questionnaire scores were evaluated for changes using a paired t-test, with statistical significance defined as a p-value <0.05. Correlation between the fermented food intake and improvements in questionnaires was completed using the Spearman test. Results 43 patients were enrolled; 18 patients in the treatment arm and 20 patients in the control arm completed the trial. Groups did not differ by age, race, education, or baseline neuropsychiatric testing. There was a significant increase in fermented food intake in the treatment arm over the duration of the study. Those in the treatment arm had improvements in questionnaires related to anxiety, PTSD, and quality of life. Those in the control arm only had improvements in reported overall quality of life. Intra-group changes can be seen in Table 1. Correlations between fermented food intake and change in questionnaire scoring showed numerical but not statistically significant improvements in many of the questionnaires, except for the MOCA questionnaire and EQ-5D-5L Self-Care question. Conclusion Implementing a diet high in fermented foods in patients with LC appears to have some signal for improvement in neuropsychiatric symptoms. Completing a larger trial would help clarify if changes would be statistically significant. This abstract is funded by: Mayo Clinic Critical Care Research Committee

Abstract Importance Decline in olfactory function may be used as a predictor of cognitive decline, to enhance early detection models, improve risk stratification, and enable early intervention. Objective To assess the longitudinal association between olfactory decline, cognitive decline, and postmortem neuropathology. Design, setting and participants Retrospective longitudinal analysis with clinicopathological correlations of a prospective population-based cohort study using data from the Arizona Study of Aging and Neurodegenerative Disorders (AZSAND) and its Brain and Body Donation Program. Participants included cognitively unimpaired individuals without parkinsonism that converted to mild cognitive impairment (MCI) and/or dementia or remained cognitively stable. Main Outcomes and Measures longitudinal change in olfaction, neuropsychiatric symptoms, motor function and memory, conversion to MCI/dementia, postmortem neuropathology Results Over a mean follow-up period of 7.7 ± 5.4 years, out of 922 participants who were cognitively unimpaired at the first cognitive conference, 643 remained cognitively unimpaired, 279 converted to MCI, and 82 developed dementia. Of these, 633 individuals had at least 2 olfactory tests. Converters showed reduced olfactory function (t=-12.6, p <0.0001), faster progression in neuropsychiatric symptom burden (t=3.42, p < 0.001), and faster decline in memory (t= -7.33, p <0.0001) prior to conversion while no significant differences were observed in motor scores between converters and non-converters. Using ROC analysis, olfactory decline, increased neuropsychiatric symptom burden, as well as motor and memory decline predicted conversion to MCI with a consistent accuracy of ∼ 70% up to 5 years before conversion, while UPSIT alone had an accuracy of ∼ 60%. Longitudinal decline in olfaction was associated with a higher burden of a-synuclein (t= -8.21, p <0.0005), tau tangle (t= -2.66, p < 0.01) and amyloid plaque burden (t= -2.85, p < 0.005) and a faster decline over time was associated with a higher burden of tau (t=5.66, p<0.0001). Conclusions and Relevance A reduction in olfactory identification ability is observed up to a decade prior to conversion to MCI and is associated with underlying burden of neuropathology markers, underscoring the value of incorporating olfactory testing in cognitively unimpaired individuals to identify those at-risk of future cognitive decline.

Abstract Introduction: Underlying tumor biology in metastatic castration-resistant prostate cancer (mCRPC) is highly complex and heterogeneous. To date, mCRPC state biomarker-enriched clinical trial designs incorporating either prognostic or predictive biomarkers are limited to alterations in homologous recombination repair pathway genes. In this study, we use an integrated clinico-molecular prognostic score in mCRPC that combines plasma prognostic proteins with a plasma-based copy number alteration (CNA) risk algorithm (PMID: 40036789; PMID: 36230636) to simulate a biomarker-enriched clinical trial design. Methods: A plasma cell-free DNA (cfDNA) prognostic risk algorithm from eight CNAs (gains in AR, MYC, PIK3CA, PIK3CB; deletions in NKX3-1, TMPRSS2, TP53, ZBTB16) is combined with clinical prognostic proteins (alkaline phosphatase, lactate dehydrogenase) to generate a prognostic risk-score (RS) in three independent treatment-naïve mCRPC cohorts. Using the RS classifier, we estimate the number of high-RS patients (pts) required to be screened from an unselected mCRPC population and the number needed for enrollment to achieve adequate statistical power in a randomized trial for a predefined effect size (Hazard Ratio- 0.70; 80% power). The trial design aims to enrich High-RS pts for targeting lethal mCRPC. Trade-offs are evaluated between number of pts screened and enrolled under varying RS enrichment thresholds and degrees of enrichment. Results: The range of RS distribution in 561 mCRPC pts from three pooled datasets was 0.362-11.737. The median overall survival (OS) (months; 95% Confidence Intervals) based on risk stratification for Low-RS (≤0.8) is 46.2 (38.8-55.4); “Intermediate-risk” (RS: 0.8-2.6) is 27.1(24.3-31.5); High-Risk (RS: &gt;2.6) is 14.9 (11.9-17.6). For a 2-year OS mCRPC endpoint, a biomarker-enriched trial restricted to pts with RS ≥2, which corresponds to the 65th percentile of the RS distribution, would require enrollment of 398 pts, after screening 1,101 mCRPC pts to achieve 80% power to detect a 0.7 treatment vs. control hazard ratio (HR). In contrast, a traditional non-enrichment clinical trial (no molecular classifier) would require 574 pts to achieve to achieve 80% power and still not achieve adequate power to detect strong treatment by risk interactions. As a pragmatic alternative, we evaluated a hybrid enrichment strategy enrolling 50% of pts from the general mCRPC population and 50% of pts with RS ≥1 (41st percentile). This biomarker enriched design which takes 50% pts with a RS of 1.0 (or more) in a trial with 24-month follow-up will require a sample size 525 patients for which 676 pts will be screened to achieve an 80% power to detect a 0.7 treatment vs. control hazard ratio. Conclusion: A plasma-based molecular classifier can be used to design an enrichment trial in biologically defined lethal mCRPC, incorporating 50% of pts with High-RS and yet balance the required sample size and number of pts to be screened enabling an adequately powered molecular classifier-based biomarker-enriched clinical trial. Citation Format: Manish Kohli, Liang Wang, Aik C. Tan, Muhammad Zaki Fadlullah, Anushka Shankar, Sumithra Mandrekar, Benjamin Haaland. Clinico-Genomic Risk Score-Based Biomarker Enrichment Clinical Trial Designs in Poor Prognosis Metastatic Castrate Resistant Prostate Cancer [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Innovations in Prostate Cancer Research and Treatment; 2026 Jan 20-22; Philadelphia PA. Philadelphia (PA): AACR; Cancer Res 2026;86(2_Suppl):Abstract nr B039.

Background/Objectives: To evaluate risk factors for postoperative neurological deficits following cervical endoscopic unilateral laminotomy for bilateral decompression (CE-ULBD) and to determine whether intraoperative neurophysiological monitoring (IONM) can predict neurological compromise. Methods: A multicenter retrospective review was performed on 42 CE-ULBD procedures conducted between 2016 and 2024; 33 cases met the inclusion criteria with available imaging and electromyography data. Demographic, operative, and neurophysiological variables were analyzed. Preoperative stenosis severity was graded using the Kang MRI system. Intraoperative IONM data, including electromyography firing and motor evoked potential (MEP) changes, were correlated with new postoperative weakness. Results: The cohort (69.1% male, mean age 70.2 ± 1.7 years, mean BMI 29.6 ± 1.1) included 56 decompressed levels. The most common operative levels were C3-4 (37%) and C4-5 (24%). Postoperative weakness occurred in four patients (12.1%), all of whom had severe (Grade 3) preoperative stenosis. Among these, 50% exhibited preoperative weakness. Neuromonitoring changes correlated significantly with postoperative weakness (Fisher’s Exact, p < 0.001); 100% of patients with new post-operative weakness had sustained MEP decrease at the time of closure. Conclusions: Patients with severe cervical stenosis and preoperative weakness are at heightened risk of postoperative neurological deficits following CE-ULBD. Elevated epidural pressure from continuous irrigation in a constricted canal may exacerbate cord compression, particularly in those with preexisting myelopathy. IONM changes strongly correlate with new deficits and may exacerbate cord compression, particularly in those with preexisting myelopathy, and may serve as an early warning system for impending neurological injury. Surgeons should exercise caution and maintain low irrigation pressures in patients with severe stenosis undergoing endoscopic cervical decompression.

Multiplex immunofluorescence and immunohistochemistry (mIF/IHC) are increasingly employed antibody-based technologies that use tissue sparingly and facilitate the detection of co-localized or neighboring biomarkers. Specifically, these platforms enable spatial analyses of the tumor microenvironment as well as extended applications, for example, describing normal tissue anatomy, autoimmunity, infectious diseases, etc. mIF/IHC has greatly enhanced biomarker discovery efforts, and a growing number of studies suggest superiority to traditional IHC. Standardization of staining approaches, reporting of image analysis strategies and resultant data is critical for facilitating cross-study comparisons, validation, deployment, and generalization of findings. To address this challenge, The Society for Immunotherapy of Cancer (SITC) previously published two articles providing best practice guidelines for mIF/IHC staining, image analysis, and data sharing. Here, SITC convened stakeholders to develop the third article in the series, a consensus checklist for scientific reporting of mIF/IHC data to support and complement the best practice guidelines. The checklist includes critical components of mIF/IHC applications to be defined within publications such as detailed descriptions of analytical validation; image acquisition, selection, and registration methods; and cell clustering and spatial analysis strategies, amongst others. Such information will help with data reproducibility and comparison across studies towards future drug and assay development.

Background: Sudden cardiac death (SCD) is a leading cause of mortality in the U.S., with emerging evidence pointing to racial disparities and the impact of social vulnerability. Yet, geographic patterns integrating race and county-level vulnerability remain underexplored. Objective: To identify U.S. counties where socially vulnerable communities, stratified by race/ethnicity, experience disproportionate burdens of SCD-related mortality. Methods: We analyzed county-level data (1999-2020), integrating age-adjusted mortality rates (AAMR) for SCD with the 2020 CDC Social Vulnerability Index (SVI). Counties were categorized into SVI quartiles (Q1: Low to Q4: High Social Vulnerability). Analyses were stratified by race/ethnicity: Black or African American, Hispanic/Latino, White, American Indian/Alaska Native (AI/AN), and Asian/Pacific Islander (API). We used RStudio (v2023.06.1) for data cleaning, ANOVA testing, and descriptive analysis. “Unreliable” mortality values were handled as missing or set to zero, with only valid AAMR > 0 included. Hotspot counties were identified by selecting the top decile of AAMR SCD within Q4 SVI counties. Results: Significant disparities emerged across SVI quartiles for Black (p = 0.0003), Hispanic (p < 0.0001), and White (p < 0.0001) groups. Mean AAMR more than doubled from low to high SVI counties in Black communities (17.0 to 39.4 per 100,000). White and Hispanic groups showed smaller but consistent increases. API (p = 0.60) or AI/AN (p = 0.63) groups showed no significant trends. Madison Parish, LA appeared in the top-10 highest-burden counties across all groups. Black populations faced extreme rates in some SVI counties (e.g., Madison Parish, LA: 351.4 per 100,000), while API groups had low AAMR even in vulnerable areas. Data misclassification limited interpretation in the Hispanic group. Conclusions: Black, Hispanic, and White populations showed significant differences in SCD mortality tied to social vulnerability. Black populations in the most vulnerable areas experienced more than twice the rate of mortality compared to those in less vulnerable areas, suggesting a strong correlation between social vulnerability and SCD mortality in Black communities. API and AI/AN groups showed no significant associations, likely due to small sample sizes. Madison Parish, LA reflects the compounded impact of structural disadvantage. Findings support targeted interventions in socially and racially vulnerable counties to reduce preventable cardiac deaths.

Abstract Foramen of Winslow hernia is a rare subtype of internal hernias, accounting for ⁓8% of all internal hernia cases. Its nonspecific clinical presentation often delays diagnosis, with ˂10% identified preoperatively. With advances in minimally invasive techniques, robotic-assisted surgery has emerged as a safe and precise method for managing complex intra-abdominal pathology. A 70-year-old male presented with an acute abdomen. He was hemodynamically stable, and computed tomography imaging suggested an internal hernia. Robotic-assisted abdominal exploration revealed a segment of colon herniated through the foramen of Winslow. The colon was carefully reduced without complications. A piece of omentum was placed into the defect, and the retracted gall bladder naturally positioned itself within the foramen, potentially minimizing the risk of recurrence. The patient recovered uneventfully and discharged on postoperative day one. To our knowledge, this is the first documented case of foramen of Winslow hernia repair using a robotic-assisted approach.

AbstractNeuroblastoma (NB) is an aggressive malignant tumor arising from a primitive neural crest origin. While the significance of tumor location in survival outcomes is recognized, it remains inadequately explored. This study provides a comprehensive analysis of the survival and characteristics of pediatric adrenal and nonadrenal NBs in the United States.A retrospective analysis of pediatric primary NB between 1975 and 2016 was conducted using the Surveillance, Epidemiology, and End Results (SEER) database. Univariate and multivariate regression analyses were used to determine prognostic variables.A total of 4,554 patients were included, comprising 52% males (n = 2,385) and 79.2% Caucasians (n = 3,569). The median age of the patients was one year (range: 0–19 years). In all, 44.4% (n = 1,996) of the patients had adrenal NB and 55.6% (n = 2,496) patients had nonadrenal NB. Adrenal NB was significantly more prevalent among males and those presenting at a younger age (p < 0.001). Adrenal NB was more likely to have a higher grade and distant metastasis at diagnosis (p < 0.001). Nonadrenal NB, female sex, surgical resection, and later year of diagnosis were associated with improved survival (p < 0.001).This study highlights important factors that are helpful for prognostication of NB patients in the United States. Tailored approaches considering tumor site are crucial for effective management of NB.