Citations

Abstract Objective Data extraction from the published literature is the most laborious step in conducting living systematic reviews (LSRs). We aim to build a generalizable, automated data extraction workflow leveraging large language models (LLMs) that mimics the real-world 2-reviewer process. Materials and Methods A dataset of 10 trials (22 publications) from a published LSR was used, focusing on 23 variables related to trial, population, and outcomes data. The dataset was split into prompt development (n = 5) and held-out test sets (n = 17). GPT-4-turbo and Claude-3-Opus were used for data extraction. Responses from the 2 LLMs were considered concordant if they were the same for a given variable. The discordant responses from each LLM were provided to the other LLM for cross-critique. Accuracy, ie, the total number of correct responses divided by the total number of responses, was computed to assess performance. Results In the prompt development set, 110 (96%) responses were concordant, achieving an accuracy of 0.99 against the gold standard. In the test set, 342 (87%) responses were concordant. The accuracy of the concordant responses was 0.94. The accuracy of the discordant responses was 0.41 for GPT-4-turbo and 0.50 for Claude-3-Opus. Of the 49 discordant responses, 25 (51%) became concordant after cross-critique, increasing accuracy to 0.76. Discussion Concordant responses by the LLMs are likely to be accurate. In instances of discordant responses, cross-critique can further increase the accuracy. Conclusion Large language models, when simulated in a collaborative, 2-reviewer workflow, can extract data with reasonable performance, enabling truly “living” systematic reviews.

AbstractBackgroundAlzheimer’s disease (AD) is a devastating neurodegenerative disorder with few therapies to treat, mitigate or prevent its onset. Understanding of this disease is predominantly based on research in non‐Hispanic Whites (NHW) although AD disproportionately affects African Americans (AA) and Latin Americans (LA), underrepresented in AD research. To address this knowledge gap, the Accelerating Medicine Partnership for Alzheimer’s Disease (AMP‐AD) Diversity Working Group was launched to generate multi‐omics data from post‐mortem brain tissue from donors of predominantly AA and LA descent. Here, we focus on genomic and transcriptomics measures, made available through the AD Knowledge Portal (ADKP).MethodBrain tissue from dorsolateral prefrontal cortex (DLPFC), caudate nucleus (CN), superior temporal gyrus (STG) and temporal pole (TP) regions of AA (N = 299), LA (N = 344) and NHW (N = 532) donors was obtained through Mayo Clinic, Mt. Sinai School of Medicine, Emory, Columbia, Rush Alzheimer’s Disease Center, Banner Sun Health, and 1Florida. DNA extracted from DLPFC tissue samples (n = 641) was utilized for whole genome sequencing (WGS) at New York Genome Center (NYGC). RNAseq of 2,224 samples was conducted either at Mayo, Rush or NYGC. Sequencing reads were processed through GenomeGPS and MAPRseq pipelines for WGS and RNAseq data, respectively. Following quality control (QC), RNAseq data was normalized and assessed for sources of variation (SOV). Integration of WGS and RNAseq data to perform comparative eQTL analysis across multi‐ethnic populations and brain regions is ongoing.ResultWGS data for 621 donors, and RNAseq data generated at Mayo (n = 1,136 samples), Rush (n = 627), and NYGC (n = 779) sequencing centers, were deposited to the ADKP. Consensus processing of RNAseq data revealed site, sex, race/ethnicity, RIN, AD diagnosis, age at death and/or library batch as SOV across various sites and brain regions. Principal component analysis of gene expression measures revealed no major stratification by disease or brain region. Accompanying clinical metadata of donors have been distributed through ADKP.ConclusionThese multi‐omics data made available to the research community is expected to be an initial step towards bridging our data and knowledge gap to elucidate conserved and distinct AD related pathways across these underrepresented at‐risk populations.

Abstract The interactions between transcription factors (TFs) and the target genes could provide a basis for constructing gene regulatory networks (GRNs) for mechanistic understanding of various biological complex processes. From gene expression data, particularly single-cell transcriptomic data containing rich cell-to-cell variations, it is highly desirable to infer TF–gene interactions (TGIs) using deep learning technologies. Numerous models or software including deep learning–based algorithms have been designed to identify transcriptional regulatory relationships between TFs and the downstream genes. However, these methods do not significantly improve predictions of TGIs due to some limitations regarding constructing underlying interactive structures linking regulatory components. In this study, we introduce a deep learning framework, DeepTGI, that encodes gene expression profiles from single-cell and/or bulk transcriptomic data and predicts TGIs with high accuracy. Our approach could fuse the features extracted from Auto-encoder with self-attention mechanism and other networks and could transform multihead attention modules to define representative features. By comparing it with other models or methods, DeepTGI exhibits its superiority to identify more potential TGIs and to reconstruct the GRNs and, therefore, could provide broader perspectives for discovery of more biological meaningful TGIs and for understanding transcriptional gene regulatory mechanisms.

Abstract Introduction Inflatable penile prosthesis (IPP) surgery is a well-established treatment for erectile dysfunction (ED). There is little information in the literature regarding the utilization of perioperative antimicrobial prophylaxis in primary and revision IPP surgeries. Objective We investigated the patterns of perioperative antimicrobial prophylaxis among a large multinational cohort of primary and revision IPP surgeries. Methods We conducted a retrospective analysis of 5400 patients (4187 primary IPP patients and 849 revision IPP patients) from our Prosthetic Urology Multi-Institutional Partnership (PUMP) institutions. Usage data for intravenous (IV), preoperative oral, and postoperative oral antimicrobials were analyzed. We specifically assessed patient profiles, antimicrobial selection, and antimicrobial duration. Multivariate statistical analyses were conducted to identify significant variations between primary and revision surgeries. Results For patients undergoing primary IPP placement, 81.9% received IV antibiotics for <24 hours whereas 18.1% received IV antibiotics for >24 hours. Preoperative oral antimicrobials were used in 59.4% of patients and 99.4% received postoperative oral antimicrobials. The overall infection rate for primary implants was 2%. Notably, quinolone administration shows a higher postoperative infection rate (11.4%) compared to gentamicin (1.9%), cephalosporin (0%), and vancomycin (1.0%). Multivariable regression analysis identifies factors such as a history of infection, diabetes, and priapism as predictors for postoperative infections. Furthermore, the study finds that perioperative antifungal use is associated with a decreased risk of infection. Conclusions This study provides valuable insights into the patterns of perioperative antimicrobial prophylaxis. Duration, type, and route of antibiotics given preoperatively or postoperatively do not affect postoperative infection rates among high volume prosthetic urologists. Giving IV antibiotics for more than 24 hours does not affect postoperative infection. Perioperative antifungals are a protective factor. Disclosure Any of the authors act as a consultant, employee or shareholder of an industry for: Coloplast, Clarus Therapeutics, Antares Pharma, Acerus.

OBJECTIVE This study aimed to evaluate local control (LC) of tumors, patient overall survival (OS), and the safety of stereotactic radiosurgery (SRS) for esophageal cancer brain metastases (EBMs). METHODS This retrospective cohort study used data from 15 International Radiosurgery Research Foundation facilities encompassing 67 patients with 185 EBMs managed using SRS between January 2000 and May 2022. The median patient age was 63 years, with a male predominance (92.5%). Most patients (64.2%) had a single brain metastasis, while 7.5% had more than 5 metastases. The median tumor volume was 0.9 cm 3 , and the median margin dose delivered to the tumor was 20 Gy. RESULTS The median OS post-SRS was 15.2 months, with 1- and 2-year OS rates of 65.7% and 32.3%, respectively. A significant inverse correlation was found between the number of EBMs and OS in the univariable analysis. LC rates at 1 and 2 years were 89% and 76%, respectively. Adverse radiation effects (AREs) were observed in 17.9% of patients, with 13.4% being mild and transient and 4.5% severely symptomatic (Common Terminology Criteria for Adverse Events grade 3). New intracranial disease developed in 58.2% of patients, with 1- and 2-year rates of 58% and 73%, respectively. CONCLUSIONS SRS for EBMs demonstrated high survival rates and effective tumor control, with a low incidence of severe AREs. These findings highlight the potential role of SRS in the multidisciplinary multimodality management paradigm of EBM.

Introduction: Recent efforts have shed light on the clinical and laboratory factors associated with toxicity and response to BCMA-directed CAR-T therapy in patients with relapsed/refractory multiple myeloma (RRMM). However, a major limitation of current scoring systems lies in their assessment at time of lymphodepletion. Because they are not determined before apheresis, they are ill-equipped to guide patient selection, anticipate level-of-care, or plan pre-emptive toxicity management strategies. We therefore aimed to characterize early predictors of CAR-T toxicity and response. Study Design: In this multicenter observational study, we examined demographic, clinical, and laboratory data across three distinct timepoints: indication (e.g., treatment decision), apheresis, and pre-lymphodepletion (LD). CRS and ICANS were graded according to ASTCT criteria. Hematotoxicity (ICAHT) was graded using EHA/EBMT criteria. Early infections (day 0-100) were defined based on microbiologic data or as a clinical infection syndrome, and graded per Hill et al, Blood 2018. Kaplan-Meier estimates of progression-free (PFS) and overall survival (OS) were compared by log-rank test. Cox proportional hazard models were applied to explore associations between pre-apheresis risk factors and key toxicity and survival outcomes. The CAR-HEMATOTOX (HT) score - encompassing markers of inflammation (e.g., CRP, ferritin) and marrow reserve (e.g., hemoglobin, ANC, platelet count) - was determined at each timepoint (Rejeski et al, Blood 2021). Results: We included 530 RRMM patients treated with standard-of-care cilta-cel (n=184) or ide-cel (n=346) across nine centers. Median age was 65 (range 31-88), median ECOG was 1 (IQR 0-1), and patients received a median of 5 prior lines of therapy (IQR 4-7). The median time from indication to apheresis was 16 days (IQR 7-52), while the median time from apheresis to infusion was 56 days (IQR 46-69). The proportion of high-risk patients by HT score did not change between the pre-apheresis and LD timepoints (19% vs. 20%, p=0.82). Most patients stayed either low-risk (n=388, 73.2%) or high-risk (n=68, 12.8%). However, 35 patients (6.6%) shifted from high-to-low risk, while 39 patients (7.4%) progressed from low-to-high risk - reflecting their response to bridging therapies. We noted considerable differences in morbidity and mortality based on the pre-apheresis HT score. Compared to their low-risk counterparts (score 0-2, n=427), high-risk patients (score ≥3, n=103) were hospitalized longer (12 vs. 15 days, p=0.009) and required more frequent ICU admissions (6% vs. 12%, p=0.05). They exhibited more severe early cytopenias (G3+: 8% vs. 25%, p<0.001) and late cytopenias (G3+: 7% vs. 21%, p<0.001), resulting in increased use of TPO agonists and stem cell boosts. Grade 3 or higher ICANS was also more common in HThigh patients (3% vs. 11%, p=0.006), as were severe infections (14% vs. 27%, p=0.007). We found increased non-relapsed mortality in HThigh patients (1-yr NRM 4.6% vs. 16.1%, p<0.001), but no significant difference in the occurrence of secondary malignancies (p>0.9). With a median follow-up of 12.4 months, HThigh patients showed inferior PFS (1-yr PFS 29% vs. 63%, p<0.0001) and OS (1-yr OS 49% vs. 83%, p<0.0001). Transitioning from a low-to-high score was linked to poor survival (1-yr PFS 26%), while the shift from high-to-low risk was associated with favorable outcomes (1-yr PFS 51%). In a multivariable Cox regression model of pre-apheresis features, a high HT score (aHR 2.52, p<0.0001), prior anti-BCMA therapy (aHR 1.55, p=0.04), history of extramedullary disease (aHR 1.42, p=0.02), and penta-refractoriness (aHR 1.45, p=0.02) were independently associated with PFS. Notably, the presence of more than 3 of these high-risk features defined an ultra-high-risk cohort of 28 patients (5.3%) with particularly dismal survival (1-year PFS 15%). Conclusions: Leveraging granular pre-apheresis data from a large international cohort, we identify early markers of toxicity and response in CAR-T-treated RRMM patients. The HT score - reflecting baseline inflammation and hematopoietic reserve - identified high-risk patients likely to benefit from pre-emptive toxicity mitigation strategies such as prophylactic stem cell collection. Most importantly, we characterize a small yet ultra-high-risk group of RRMM patients in need of more effective bridging and therapeutic options.

Introduction Symptom burden in myeloproliferative neoplasms (MPN) is quantified using total symptom scores (TSS) calculated from the patient-reported scores for clinically meaningful symptoms. The impact of demographic, clinical, and laboratory features on these symptoms remains unclear and warrants further investigation. Methods Patients with polycythemia vera (PV), essential thrombocythemia (ET), and myelofibrosis (MF) were identified from the retrospective chart review. TSS, scores of individual symptoms (fatigue, early satiety, abdominal discomfort, inactivity, concentration difficulties, fever, night sweats, itching, bone pain, weight loss), demographic characteristics (race, ethnicity, age, and gender), clinical features (time since diagnosis, depression status, obesity status), laboratory results and season at the time of visit when the first TSS assessment was performed were recorded. Multicollinearity and homoscedasticity were assessed using variance inflation factors and visual inspection of the data, respectively. A multivariable regression analysis was performed using a robust variance estimator. A p-value <0.05 indicated a statistically significant association with the symptom scores. Results The chart review identified 557 (PV: 201; ET: 191; MF: 165) patients. The median age of patients with PV, ET, and MF was 65 (range: 27-93), 65 (20-96), and 69 (27-96) years, respectively. Most patients were white (PV: 92%; ET: 90%; MF: 93%) and non-Hispanic or Latino (PV: 93%; ET: 96%; MF: 92%). Most patients with PV and ET were female (PV: 51%; ET: 73%), while most patients with MF were male (55%). Among the included patients 22% (PV: 21%; ET: 27%; MF: 18%) had the diagnosis of depression at the time of the clinical visit. TSS was highest in MF (mean (SD): 18.1±14.6), followed by PV (15.7±14.5) and ET (15.1±14.0) patients. Fatigue was the most reported symptom by the patients (PV: 81%; ET: 82%; MF: 84%), whereas weight loss was the least reported symptom (PV: 4.5%; ET: 6.3%; MF: 16%). With regards to the TSS multivariable regression analysis showed that age (beta coefficient: 0.2; p=0.01), depression (7.2; 0.008), and low hemoglobin (1.2; 0.003) in PV, depression (11.7; 0.001) in ET, and depression (9.5; 0.001), splenomegaly (6.9; 0.01), and WBC count (0.16; 0.048) in MF patients were significantly associated with TSS. In terms of individual symptoms in PV, depression was associated with fatigue (beta coefficient: 1.31; p-value: 0.01) and concentration difficulties (1.7; <0.001), high WBC count was associated with early satiety (0.09; 0.03), and non-white race was associated with fever (1.4; 0.007). In ET, depression was associated with fatigue (1.6; <0.001), abdominal discomfort (1.4; <0.001), inactivity (2.5; <0.001), concentration difficulties (2.4; <0.001) and night sweats (1.4; <0.001), anemia was associated with fatigue (1.2; 0.03), fewer months since diagnosis was associated with inactivity (0.01; 0.02) and young age was associated with bone pain (0.02; 0.03). In MF, depression was associated with fatigue (1.8; 0.001), concentration difficulties (1.3; 0.01), fever (1.1; 0.02), and night sweats (2.2; 0.03), low hemoglobin was associated with fatigue (0.46; <0,001) and early satiety (0.31; 0.003), high WBC count was associated with fatigue (0.03; 0.04) and early satiety (0.04; 0.02), low platelet count was associated with abdominal discomfort (0.002; 0.03), splenomegaly was associated with inactivity (2.27; 0.003) and female gender was associated with itching (0.57; 0.01). Conclusions This study demonstrates that depression in PV and ET and depression, low hemoglobin, and a high WBC count in MF contribute to multiple symptoms seen in these patients. Identifying demographic groups at risk of a higher symptom burden and addressing the comorbidities that drive symptoms, in addition to disease-directed therapy, is crucial to improving the quality of life in these patients.

Introduction Prognostication in myelofibrosis (MF) currently relies on manual assessment of clinical, pathological, and laboratory variables, which is inefficient and error-prone. The potential of large language models (LLMs) and rule-based programming to automate data extraction and prognostication in these patients remains unexplored and merits investigation. Methods Patients with MF were identified from a retrospective chart review. Data regarding clinical (age, constitutional symptoms, transfusion requirements), laboratory (hemoglobin, circulating white blood cell, platelet, and blast counts), and pathological (grade of bone marrow fibrosis, chromosomal analyses, cytogenetics) features was extracted from the latest free-text clinical notes and pathology reports using the generative pretrained transformer-4 (GPT-4) model in a zero-shot setting with the prompting parameters set to generate deterministic responses. The patients were randomly sampled into development (15%) and test (85%) sets, with the documents of patients in the development set used for iterative prompt engineering. The finalized prompts were used for data extraction in the test set. Rule-based programming was used for post-processing of the extracted data and ascertaining prognosis via Dynamic International Prognostic Scoring System (DIPSS), DIPSS-plus (DIPSS+), and Mutation-enhanced International Prognostic Scoring System-70 (MIPSS-70). Automatically extracted data and prognostic categories were compared with the manually generated human annotations. Mean accuracy, precision, and recall with 95% confidence intervals (CI) were computed. Detailed error and sensitivity analyses were conducted to assess performance. Results The study included 100 patients (development: 15; test: 85) with 100 clinical notes (development: 15; test: 85) and 90 pathology reports (development: 13; test: 77). The mean age of these patients was 70 (SD: 10.6); 55 (55%) patients were men, and 93 (93%) patients were White. In the development set, the mean accuracy, precision, and recall for data extraction from clinical notes were 94% (95% CI: 90-98%), 92% (84-99%), and 99% (97-100%). For data extraction from pathology reports, the mean accuracy, precision, and recall were 96% (93-100%), 98% (92-100%), and 83% (63-100%). For prognostication, the mean accuracy, precision, and recall were 87% (79-94%), 80% (73-87%), and 100% using DIPSS (accuracy: 80%; precision: 73%; recall: 100%), DIPSS+ (accuracy: 93%; precision: 83%; recall: 100%) and MIPSS-70 (accuracy: 87%, precision: 83%; recall: 100%). In the test set, the mean accuracy, precision, and recall for data extraction from clinical notes were 96% (95% CI: 93-98%), 91% (87-96%), and 98% (97-100%). For data extraction from pathology reports, the mean accuracy, precision, and recall were 95% (92-98%), 93% (87-99%), and 90% (80-100%). For prognostication the mean accuracy, precision, and recall were 95% (94-96%), 89% (88-90%) and 99% (96-100%) using DIPSS (accuracy: 94%; precision: 89%; recall: 100%), DIPSS+ (accuracy: 95%; precision: 88%; recall: 95%) and MIPSS-70 (accuracy: 94%, precision: 90%; recall: 100%). Error analyses identified incorrect extractions by GPT-4 were related to the circulating blasts and constitutional symptoms variables from the clinical notes. Sensitivity analyses excluding the associated variables improved overall accuracy from 94% to 97% and 96% to 97% in the development and test sets, respectively. Similarly, incorrect extractions from the pathology reports were related to the CALR gene mutation variables. Sensitivity analyses excluding the associated variables improved the overall accuracy from 96% to 98% and 95% to 97% in the development and test sets, respectively. Conclusion LLMs and rule-based programming can accurately extract data from the electronic health records and prognosticate MF patients via DIPSS, DIPSS+, and MIPSS-70. The automated approach could significantly streamline clinical and research workflows, providing an efficient and less resource-intensive alternative to manual prognostication.

Abstract INTRODUCTION Alzheimer's disease (AD) is the most prevalent neurodegenerative disease, yet our comprehension predominantly relies on studies within non‐Hispanic White (NHW) populations. Here we provide an extensive survey of the proteomic landscape of AD across diverse racial/ethnic groups. METHODS Two cortical regions, from multiple centers, were harmonized by uniform neuropathological diagnosis. Among 998 unique donors, 273 donors self‐identified as African American, 229 as Latino American, and 434 as NHW. RESULTS While amyloid precursor protein and the microtubule‐associated protein tau demonstrated higher abundance in AD brains, no significant race‐related differences were observed. Further proteome‐wide and focused analyses (specific amyloid beta [Aβ] species and the tau domains) supported the absence of racial differences in these AD pathologies within the brain proteome. DISCUSSION Our findings indicate that the racial differences in AD risk and clinical presentation are not underpinned by dramatically divergent patterns in the brain proteome, suggesting that other determinants account for these clinical disparities. Highlights We present a large‐scale proteome (∼10,000 proteins) of DLPFC (998) and STG (244) across AD cases. About 50% of samples were from racially and ethnically diverse brain donors. Key AD proteins (amyloid and tau) correlated with CERAD and Braak stages. No significant race‐related differences in amyloid and tau protein levels were observed in AD brains. AD‐associated protein changes showed a strong correlation between the brain proteomes of African American and White individuals. This dataset advances understanding of ethnoracial‐specific AD pathways and potential therapies.