Shahar, Shelly

ABSTRACT Background Long‐term health effects of thymectomy in adults remain uncertain. Previous reports suggested associations with increased mortality, cancer incidence, and autoimmune disease, although these findings may reflect differences in patient selection, surgical indication, or follow‐up duration. We aimed to examine long‐term outcomes following adult thymectomy. Methods We performed a propensity score‐matched retrospective study of patients who underwent thymectomy or comparator cardiothoracic procedures, with follow‐up extending to 10 years. Primary outcomes were all‐cause mortality, new‐onset malignancy, and autoimmune disease. Findings were compared with those reported in a recent study that has raised caution regarding thymic excision. Results We identified 425 patients who underwent thymectomy and 5401 controls who underwent coronary artery bypass or valve surgery. After matching, 398 patients remained in each group. Thymectomy was not associated with increased mortality at 5 years ( HR = 0.63; 95% CI , 0.38–1.04) or 10 years ( HR = 0.78; 95% CI , 0.54–1.13). Excluding reclassified thymic cancers, 5‐year extrathymic malignancy incidence was 2.3% versus 1.8% ( p = 0.54); no significant differences were observed in immune‐related disease. Mortality was higher among patients with thymic malignancy. Among MG patients, only one event occurred, precluding reliable estimation. Conclusions Thymectomy was not associated with excess mortality or malignancy at 10 years compared with matched cardiothoracic controls. Indication‐stratified analyses within the thymectomy cohort showed higher mortality for thymic malignancy, with insufficient events to assess outcomes in MG patients. These findings do not support a generalized survival penalty from thymectomy, though caution is warranted for incidental thymic removal.

Abstract Most clinical large language model (LLM) benchmarks rely on clean, concise vignettes that do not reflect the noisy, long-form documentation typical of real clinical records. How LLM performance degrades under realistic chart conditions remains poorly characterised. Here we test whether structured retrieval workflows protect National Institutes of Health Stroke Scale (NIHSS) scoring accuracy under systematic context stress. Using 100 de-identified acute stroke cases and a fully crossed 4 x 4 x 3 x 3 condition matrix (144 conditions per case), we vary context acquisition method, document length, distractor load and critical-information position across four Gemini models (57,047 retained runs). Structured retrieval reduces mean absolute error (MAE) from 4.58 to 2.96 points relative to non-agentic baselines (mean gain 1.62 MAE points; 95% CI 1.57–1.67; 35% relative reduction), with consistent gains across all 36 stress combinations. Lower-cost models show disproportionately larger gains (2.76 versus 0.45 MAE points). Tool-retrieved pipelines outperform retrieval-augmented generation in 33 of 36 combinations. These findings indicate that retrieval architecture, rather than model scale alone, is a tractable lever for robust, equitable clinical LLM deployment.

ABSTRACT Background and Aims Chronic inflammatory demyelinating polyradiculoneuropathy ( CIDP ) is a common, treatable, autoimmune, neuropathy frequently misdiagnosed. The complex electrodiagnostic ( EDX ) criteria required for diagnosis are challenging to apply in routine practice. This study evaluates the effectiveness of a machine‐assisted EDX tool in real‐time identification of demyelinating nerve conduction studies ( NCS ) stipulated by 2021‐ EAN / PNS CIDP criteria in conjunction with an online CIDP clinical probability calculator. Methods We prospectively evaluated 100 consecutive patients with referral or EDX diagnosis of polyradiculoneuropathy beginning January 1, 2024. Fifty patients were included, each from Jacksonville and Rochester Mayo Clinics. Routine electromyography ( EMG ) reports were compared to a machine‐assisted support tool applying 2021‐ EAN / PNS CIDP EDX criteria to NCS assessing demyelination. The EDX ‐assistant performance was compared to the final clinical diagnosis and our previously reported online CIDP clinical prediction tool ( https://news.mayocliniclabs.com/cidp‐calculator/ ). Results Among 100 patients (60% male; median age 64), 30 had a final clinical CIDP diagnosis with 29 having intravenous immunoglobulin follow‐up; 27 (93%) improved and 2 (7%) stabilized. Routine EMGs reported demyelination in 14 of 30 (47%), with 46% sensitivity and 79% specificity. The EDX ‐assistant identified CIDP demyelination in 28 of 30 (93%), with 93% sensitivity and 57% specificity. The clinical calculator detected CIDP in 30 of 30 (100%) with 86% specificity, increasing to 94% after excluding false positives lacking demyelinating features on the EDX tool. Interpretation A real‐time machine‐assisted EDX support tool, used with a clinical web‐based calculator, streamlines CIDP assessment by improving standardized identification of 2021‐EAN/PNS CIDP demyelinating NCS criteria and aiding clinical decisions.

Introduction: We aimed to prove integration of advanced machine learning methods within a robust ensemble framework can enhance clinical decision-support for neurologists managing patients in the emergency department (ED). Methods: We engineered an ensemble framework leveraging the capabilities of the Gemini 1.5-pro-002 large language model (LLM). The model was enhanced using prompt engineering and retrieval-augmented generation (RAG). Predictive modeling achieved by combining eXtreme Gradient Boosting (XGBoost) and logistic regression for optimal accuracy in clinical decision-making. Key clinical outcomes, such as admission and mortality, were assessed. A random subset of 100 cases was reviewed by three senior neurologists to evaluate the alignment of the AI’s predictions with expert clinical judgment. Results: We retrospectively analyzed 1368 consecutive ED patients who underwent neurological consultations, assessing their clinical features, diagnostic tests, and admission outcomes. Patients admitted were typically older and had higher mortality rates, shorter intervals to neurological evaluation, and a higher incidence of acute stroke compared to those discharged. For the primary analysis (n = 250), the Neuro artificial intelligence (AI) model demonstrated significant performance metrics, achieving an area under the curve (AUC) of 0.88 for general admission predictions in comparison to actual outcomes, an AUC of 0.86 for neurological department admissions, 0.93 for long-term mortality risk, and 1 for 48 h mortality risk. Our Neuro AI model predictions showed a strong correlation with expert consensus (Pearson correlation 0.79, p < 0.001), indicating its ability to provide consistent support amid divergent clinical opinions. Conclusions: Our Neuro AI model accurately predicted hospital admissions (AUC = 0.88) and neurological department admissions (AUC = 0.86), demonstrating strong alignment with expert clinical judgment.

Abstract Chronic headache is a common complication after mild traumatic brain injury (mTBI), which affects close to 70 million individuals annually worldwide. This study aims to test the utility of a unique, early predictive magnetic resonance imaging (MRI)-based classification model using structural brain MRI scans, a rarely used approach to identify high-risk individuals for post-mTBI chronic pain. We recruited 227 patients with mTBI after a vehicle collision, between March 30, 2016 and December 30, 2019. T1-weighted brain MRI scans from 128 patients within 72 hours postinjury were included and served as input for a pretrained 3D ResNet-18 deep learning model. All patients had initial assessments within the first 72 hours after the injury and performed follow-ups for 1 year. Chronic pain was reported in 43% at 12 months postinjury; remaining 57% were assigned to the recovery group. The best results were achieved for the axial plane with an average accuracy of 0.59 and an average area under the curve (AUC) of 0.56. Across the model's 8 folds. The highest performance across folds reached an AUC of 0.78, accuracy of 0.69, and recall of 0.83. Saliency maps highlighted the right insula, bilateral ventromedial prefrontal cortex, and periaqueductal gray matter as key regions. Our study provides insights at the intersection of neurology, neuroimaging, and predictive modeling, demonstrating that early T1-weighted MRI scans may offer useful information for predicting chronic head and neck pain. Saliency maps may help identify brain regions linked to chronic pain, representing an initial step toward targeted rehabilitation and early intervention for patients with mTBI to enhance clinical outcomes.

Abstract Traumatic brain injury (TBI) annually impacts 69 million individuals worldwide. Mild TBI constitutes approximately 90% of all TBIs. Chronic pain post-mTBI occurs in 29% to 58% of patients. This study aims to introduce a predictive model for chronic pain development in individuals diagnosed with mild traumatic brain injury (mTBI) immediately postinjury. We included individuals who had sustained mTBI in motor vehicle accident (MVA). All patients had initial assessments within the first 72 hours (representing the subacute period) after the injury and performed follow-ups for 1 year. Machine learning model was applied to the integrated measures of clinical pain, pain-related psychological parameters, mTBI clinical signs, and sociodemographic information. This study included 203 patients experiencing acute head or neck pain attributable to mTBI post-MVA. We categorized these patients into 2 groups: patients who progressed to develop chronic head or neck pain (n = 89, 43.8%) and patients who recovered (low/mild pain) (n = 114, 56.2%). Severity of the subacute neck pain, number of painful body areas, and education years were identified as the most significant factors predicting chronic pain. The optimized predictive model demonstrated high efficacy, with an accuracy of 83%, a sensitivity of 92%, and an area under the receiver operating characteristic curve of 0.8. Our findings indicate feasibility in predicting chronic post-MVA pain within the critical 72-hour window postinjury using simple bedside metrics. This approach offers a promising avenue for the early detection of individuals at increased risk for chronic pain, enabling the implementation of targeted early interventions.