Citations

Abstract Backgrounds Isolated REM sleep behaviour disorder (iRBD) is a prodromal marker of α-synucleinopathies. Although cognitive impairment is consistently reported in iRBD, its assessment varies considerably across studies. This study characterized mild cognitive impairment (MCI) patterns in iRBD using machine learning and identified the minimal neuropsychological test battery needed for accurate diagnosis. Methods Comprehensive neuropsychological data from 216 assessments (162 with MCI, 54 without) in 110 iRBD patients were analysed. Four supervised classifiers (logistic regression, support vector machine, random forest, gradient boosting) were trained across three frameworks (raw scores, raw scores plus demographics, binarized scores) to distinguish MCI status. A multi-objective hyperparameter optimization approach then identified the minimal test subset preserving diagnostic accuracy. Results Of MCI cases, 30% were single-domain and 70% multi-domain, with attention/executive functions and memory most frequently impaired. Binarized neuropsychological scores yielded the best classification performance, with support vector machine achieving 94.7% accuracy. Feature importance analyses consistently highlighted episodic memory, executive, and visuospatial measures. The optimization framework identified a mean of 8 informative tests spanning these domains, achieving comparable diagnostic accuracy to the full battery. Conclusions Cognitive impairment is common in iRBD and predominantly multidomain. Machine learning approaches applied to neuropsychological data can accurately classify MCI status and identify a concise, clinically feasible set of tests with high diagnostic value. These findings support comprehensive, multidomain neuropsychological assessment in iRBD and may inform optimized risk stratification in prodromal α-synucleinopathies.

Objective To characterize regional brain atrophy patterns, evaluate their association with regional leucine‐rich glioma‐inactivated‐1 ( LGI1 ) gene expression, and identify clinical predictors of hippocampal atrophy in LGI1 autoimmune encephalitis (AE). Methods This study compared regional brain volumes in LGI1‐AE patients versus age‐ and sex‐matched healthy controls (HC) and Alzheimer's disease (AD) patients. LGI1 gene expression was derived from the Allen Human Brain Atlas. Multivariable linear regression and generalized linear models were utilized. Results A total of 55 LGI1‐AE patients were included. Median age was 68 years (IQR 62.5–74.2 years) at volumetric analysis, after 19 months (IQR 11–45 months) from symptom onset. Compared with HC ( n = 55) and AD patients ( n = 55), LGI1‐AE patients had greater regional atrophy of the hippocampus, insula, caudate, and putamen ( p < 0.002). Longitudinal analyses demonstrated hippocampal and pallidal atrophy after covariate adjustment. Among those who did not fulfill limbic encephalitis criteria and/or had non‐lesional magnetic resonance imaging, limbic and striatal atrophy were identified compared with HC ( p < 0.01). Regions with the greatest differential atrophy corresponded to higher LGI1 gene expression (Spearman ρ = 0.7, p = 0.005). Severe cognitive deficits at diagnosis ( β = −0.56; p = 0.002) and longer disease duration ( β = −0.01; p = 0.001) independently predicted total hippocampal volume ( R 2 = 0.31). Interpretation LGI1‐AE demonstrates a distinct pattern of limbic and striatal atrophy that differs from AD and aligns with regional LGI1 expression, supporting a model of selective network vulnerability. Greater baseline cognitive impairment and longer disease duration independently associated with hippocampal atrophy, underscoring the potential importance of early diagnosis and treatment to limit structural brain injury. ANN NEUROL 2026

This Perspective discusses the misalignment physicians face between their professionalism-based values and the conditions of practice that are influenced by pressures from market, managerial, and platform logics, and offers suggestions to align practice conditions and professional identity.

Abstract Context. Thyroid nodule ultrasound findings are crucial in guiding management decisions for patients with thyroid nodules. However, the frequency with which ultrasound images are discussed and reviewed with patients during clinical visits is not well understood. Objective. To assess how often thyroid ultrasound images are shown and discussed with patients during consultations, and to identify predictors of image review during clinical visits. Design and setting. Secondary analysis of a multicenter pilot randomized clinical trial conducted at two academic medical centers in the United States. Participants. Adults undergoing evaluation for thyroid nodules. Intervention. Participants were randomized to receive either usual care or care enhanced with a conversation aid. Main Outcome Measures. The primary outcome was the frequency of thyroid nodule imaging review evaluated as a binary outcome. Secondary outcomes included evaluation of predictors of image discussion. Results. Eighty-five patients were evaluated. Most participants were women (80%), mean age (60 years, SD 16), and White (86%), undergoing new patient visits (84%). Thyroid nodule ultrasound images were reviewed with patients in 53% of visits (45/85), and in all cases, clinicians initiated the imaging discussion. Predictors of image review included type of consultation (new patient visit) (OR 10.6, 95% CI: 1.79, 62.3). Age, sex, study arm, thyroid nodule size or cancer risk and plan of care (e.g., completing a biopsy) were not statistically associated with review of clinical images. Conclusions. Review of thyroid ultrasound images during clinical visits is relatively common, with consultation type serving as significant predictor.

Introduction: Large language models (LLMs) match or exceed physician accuracy on benchmark diagnostic tasks. Whether higher accuracy reflects human-like diagnostic behavior, which bears on how safely clinicians can supervise these systems, is rarely assessed. Objectives: To determine whether diagnostic accuracy and alignment with human reader responses move together across widely accessible LLMs under default, point-of-care conditions. Methods: This cross-sectional evaluation submitted the 29 most recent New England Journal of Medicine (NEJM) case challenges to 5 publicly accessible LLMs (GPT-5, Claude, Google Gemini, DeepSeek, and Grok) through each model’s web interface (November 5-8, 2025) at default settings, one query per case, no system prompt. Each case offered a clinical vignette and 6 candidate diagnoses; published human reader response distributions (4386-17 714 per case; median, 7732) were the comparator. Diagnostic accuracy against the published gold-standard diagnosis, compared with the human reader majority by the exact McNemar test; agreement by Cohen κ; alignment by a human-error similarity index (the share of model errors on cases the reader majority also missed) and by the Jensen-Shannon divergence (JSD) between each model’s response and the reader distribution; and ordinary least squares regression of per-case JSD on normalized reader-disagreement entropy. Results: Reader majority accuracy was 48.3% (95% CI, 31.4-65.6). All 5 reached or exceeded this value, from 51.7% for DeepSeek (95% CI, 34.4-68.6) to 82.8% for Grok (95% CI, 65.5-92.4); Grok (P = .02), GPT-5 (P = .004), and Gemini (P = .04) significantly exceeded the reader majority, whereas Claude and DeepSeek did not. Grok, the most accurate model, showed the weakest alignment (similarity index, 0.40; mean JSD, 0.50 bits), with 3 of 5 errors on cases readers answered correctly. GPT-5 showed the strongest alignment (index, 1.000), failing only on cases that also challenged readers. For 4 of 5 models, divergence rose with reader-disagreement entropy; DeepSeek was the exception. Conclusions and Relevance: In this evaluation, diagnostic accuracy and alignment with human reader responses diverged across LLMs. Because a model can outperform clinicians in aggregate yet fail where oversight is least likely to detect it, evaluation of clinical AI should weigh error structure alongside accuracy.

ABSTRACT Introduction/Aims Phrenic nerve conduction studies (NCS) are regularly used in the evaluation of neuromuscular respiratory dysfunction; however, the impact of body mass index (BMI) on phrenic compound muscle action potential (CMAP) amplitudes remains poorly characterized. We evaluated the relationship between BMI and phrenic CMAP amplitudes in patients without evidence of neuromuscular diaphragmatic dysfunction. Methods We performed a retrospective secondary analysis of patients derived from two previously published cohorts from Mayo Clinic (Rochester, Minnesota) and the University of Alberta (Edmonton, Alberta) evaluating diaphragm ultrasound in suspected neuromuscular respiratory insufficiency. Patients without evidence of neuromuscular disease affecting the diaphragm who underwent bilateral phrenic NCS and diaphragm ultrasound were included. Results Seventy‐nine patients met inclusion criteria, including 36 with BMI ≥ 30 kg/m 2 . Median phrenic CMAP amplitudes were significantly lower in patients with BMI ≥ 30 kg/m 2 compared to BMI < 30 kg/m 2 (0.25 mV [ IQR 0.10–0.38] vs. 0.50 mV [0.30–0.65], p < 0.001). Low phrenic CMAP amplitudes (< 0.3 mV ) were more common in patients with obesity (58% vs. 21%, p < 0.001). BMI demonstrated a moderate negative correlation with phrenic CMAP amplitude ( ρ = −0.50, p < 0.001). Diaphragm thickness at functional residual capacity did not differ significantly between BMI groups, while diaphragm thickening ratio was modestly lower in obesity. Discussion Elevated BMI is associated with lower phrenic CMAP amplitudes in patients without evidence of neuromuscular diaphragmatic dysfunction. Reduced phrenic CMAP amplitudes in obesity may reflect technical recording limitations rather than clinically meaningful diaphragmatic dysfunction. Low‐amplitude phrenic responses in patients with obesity should be interpreted cautiously and alongside adjunctive assessments such as diaphragm ultrasound.

Abstract Purpose Large language models (LLMs) have proven performance for certain diagnostic tasks; however, limited studies have evaluated their consistency in recommending appropriate medication regimens for a given diagnosis. Medication management is a complex task that requires synthesis of drug formulation and complete order instructions for safe use. Here, the performance of GPT-4o, an LLM available with OpenAI’s ChatGPT, was tested on 3 medication management tasks. Methods GPT-4o performance was tested on 3 medication tasks: identifying available formulations for a given generic drug name, identifying drug-drug interactions (DDIs) for a given medication regimen, and preparing a medication order for a given generic drug name. For each experiment, the model’s raw text response was captured exactly as returned and evaluated using clinician evaluation in addition to standard LLM metrics, including Term Frequency-Inverse Document Frequency (TF-IDF) vectors, normalized Levenshtein similarity, and Recall-Oriented Understudy for Gisting Evaluation (ROUGE-1/ROUGE-L) F1 score between each response and its reference string. Results For the first task of drug-formulation matching, GPT-4o had 49% accuracy for generic medications being matched to all available formulations, with an average of 1.23 omissions per medication and 1.14 hallucinations per medication. For the second task of drug-drug interaction identification, the accuracy was 54.7% for identifying the DDI pair. For the third task, GPT-4o generated order sentences containing no medication or abbreviation errors in 65.8% of the cases. Conclusion Model performance for basic medication tasks was consistently poor. This evaluation highlights the need for domain-specific training through clinician-annotated datasets and a comprehensive evaluation framework for benchmarking performance.

Purpose: Hereditary transthyretin amyloidosis (ATTRv) is a progressive multisystem disease in which median mononeuropathy at the wrist (MNW) is common and often precedes systemic involvement. Nerve conduction studies (NCS) have historically been the gold standard for MNW, but neuromuscular ultrasound (NMUS) is complementary. The aim of this study was to compare NMUS with electrodiagnostic (EDX) testing for the diagnosis of MNW in patients with symptomatic ATTRv. Methods: A retrospective review was performed of patients 18 years and older with ATTRv confirmed by genetic testing who underwent both EDX testing and NMUS of the median nerve between 2019 and 2025. Demographic, NMUS, and EDX data were collected. Results: Data from 15 upper limbs in 10 patients were analyzed. Median nerve cross‐sectional area at the wrist was significantly greater than reference values ( P = 0.041). MNW was identified in 80% of wrists by NMUS and 100% by NCS, with no significant difference between modalities. Median nerve cross‐sectional area at the wrist did not significantly correlate with MNW severity determined by NCS ( r = −0.018; P = 0.941) or with any NCS parameters (all P > 0.080). Concurrent polyneuropathy was diagnosed in 60% of patients, but there was no association between polyneuropathy and MNW detection by either modality ( P > 0.197). Conclusions: NMUS and NCS demonstrate comparable diagnostic accuracy for MNW in ATTRv. Given its noninvasive nature and efficiency, NMUS may serve as an adjunct to classic EDX testing in this population.

Abstract Purpose: Inhibitors of the Ataxia telangiectasia and Rad3-related (ATR) protein kinase, a critical component of the DNA damage repair response, have synergistic anti-cancer activity with platinum compounds in preclinical models. We therefore conducted a Phase 2 study of the ATR inhibitor berzosertib+carboplatin vs. docetaxel+carboplatin in metastatic castration-resistant prostate cancer. Patients and Methods: Patients previously treated with at least one androgen receptor pathway inhibitor and taxane underwent mandatory biopsy and were randomized 1:1 to receive Arm A (docetaxel 60 mg/m2+carboplatin AUC4 day 1; or carboplatin AUC5 if not a docetaxel candidate) or Arm B (berzosertib 90 mg/m2 days 2,9+carboplatin AUC5 day 1) every 21 days. The primary endpoint was overall response rate (ORR; ≥50% PSA decline or radiographic response). Results: Of 73 randomized patients, 65 received protocol treatment: 34 on Arm A (26 docetaxel+carboplatin; 8 carboplatin alone) and 31 on Arm B. Thirteen patients (38%) in Arm A and 20 (65%) in Arm B had ≥Grade 3 treatment-related adverse events (TrAEs). ORR was 15% in Arm A (5/34; 5/26[19%] with docetaxel+carboplatin) and 0% in Arm B (0/31). At interim analysis after 65 of planned 130 patients were treated, enrollment was halted due to futility. Homologous recombination repair deficiency status from tissue and blood did not correlate with clinical outcomes, but ATM-deficiency or high levels of phoshpo-KAP1 detected in tissue was associated with clinical benefit across both arms of the trial. Conclusions: Berzosertib+carboplatin demonstrated lower ORR and more frequent ≥grade 3 TrAEs compared to docetaxel+carboplatin in this heavily pre-treated, biomarker-unselected population. NCT03517969