Multimodal Alzheimer’s disease (AD) cohorts capture cognition, function, neuroimaging, and fluid biomarkers, yet overall disease severity remains difficult to summarize on a single clinically meaningful scale. The apolipoprotein E ε4 (APOE ε4) allele is the strongest common genetic risk factor for late-onset AD, but its association with “progression” has been inconsistent because earlier placement along the disease continuum is often conflated with faster within-stage decline. Using data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), we analyzed an amyloid-positive baseline cohort (N = 1058) and a longitudinal subset (N = 932; ≥ 2 visits and ≥ 4 of 13 measures per visit). Measures included standardized cognitive and functional assessments, structural and functional neuroimaging, cerebrospinal fluid biomarkers of amyloid‑beta and tau pathology, and plasma neurofilament light protein as a marker of neuroaxonal injury. Magnetic resonance imaging (MRI) volumes were adjusted using amyloid-negative cognitively normal controls with quadratic age and intracranial volume terms. Probabilistic principal component analysis (PPCA) was used to derive a latent severity coordinate, defined as the first principal component (PC1). Hierarchical Bayesian random-intercept and random-slope models were used to estimate individual trajectories, partition APOE ε4 effects into baseline severity and within-stage rate, and generate genotype-stratified ages at prespecified severity landmarks. Axis stability was assessed with 100 bootstrap refits, and predictive performance was assessed with participant-level fivefold cross-validation. The PC1 explained 38.7
Recent work shows that text-only reinforcement learning with verifiable rewards (RLVR) can match or outperform image-text RLVR on multimodal medical VQA benchmarks, suggesting current evaluation protocols may fail to measure causal visual dependence. We introduce a counterfactual evaluation framework using real, blank, and shuffled images across four medical VQA benchmarks: PathVQA, PMC-VQA, SLAKE, and VQA-RAD. Beyond accuracy, we measure Visual Reliance Score (VRS), Image Sensitivity (IS), and introduce Hallucinated Visual Reasoning Rate (HVRR) to detect cases where models generate visual claims despite producing image-invariant answers. Our findings reveal that RLVR improves accuracy while degrading visual grounding: text-only RLVR achieves negative VRS on PathVQA (-0.09), performing better with mismatched images, while image-text RLVR reduces image sensitivity to 39.8
Large vision language models (VLMs) report strong accuracy on medical question-answering, yet it remains unclear whether they reason from visual evidence or exploit textual shortcuts. We introduce a counterfactual evaluation framework that decouples visual and textual contributions by substituting input images with controlled surrogates blank, pixel-shuffled, image-absent, and CLIP-retrieved hard negatives and derive a suite of grounding metrics including the Visual Reliance Score (VRS) and Visual Hallucination Rate (VHR). We further introduce CORAL (COntrastive Retrieval-Augmented Learning), a 7B-parameter LoRA fine-tune of Qwen2.5-VL-7B trained with a Contrastive Grounding Objective (CGO) that penalises answer invariance under hard-negative image swaps. On a paired controlled evaluation across four closed-form medical VQA benchmarks (PathVQA, PMC-VQA, SLAKE, VQA-RAD; n=400 total), CORAL improves macro accuracy by +6.7 pp (P(Delta>0)=0.988) and reduces VHR by 8.0 pp (P<0.001) over the matched Qwen2.5-VL-7B base; neither MedVLThinker RL variant achieves a significant gain on either metric. Cross-domain diagnostics further reveal that image substitution costs only <=6.5 pp on medical benchmarks versus 48-61 pp on general-domain tasks, situating the grounding gap that CGO targets. We discuss evaluation limitations openly including train/eval benchmark overlap and underpowered secondary metrics and release our framework, training code, and model weights to support reproducible grounding audits of medical VLMs.
PURPOSE:Angiosarcoma and epithelioid hemangioendothelioma (EHE) are two rare vascular sarcomas with limited therapeutic options. Prior reports have shown sensitivity to microtubule-targeting agents in these histologies. We report the efficacy and safety of eribulin in these vascular sarcomas in a pooled analysis of two parallel phase 2 studies. PATIENTS AND METHODS:Patients more than age 18 years with metastatic or recurrent angiosarcoma or EHE were treated with eribulin (1.4 mg/m2 on days 1 and 8 of a 21-day cycle) until progression or unacceptable toxicity. The primary endpoint was objective response rate (ORR) by RECIST 1.1. RESULTS:Twenty-nine patients were accrued to the study, with 25 (85%) having had prior taxane exposure. We observed an ORR of 17% for angiosarcomas, with 6 of 23 (26%) patients achieving disease stability for greater than 6 months, and an ORR of 33% (2/6) for EHE, with two of six continuing treatment for over 12 months. Five patients experienced a >1.3-fold time to progression ratio (TTP2/TTP1) on eribulin compared with the immediately prior therapy. Eribulin tolerability was consistent with published data. CONCLUSIONS:Eribulin showed clinical activity in this largely taxane-pretreated population. Future studies will be needed to confirm activity. See related commentary by Chen, p. 2130.
Alzheimer’s Disease (AD) is the most common neurodegenerative disorder and the leading cause of dementia characterised by the accumulation of beta amyloid (Aβ) plaques and neurofibrillary tangles (NFT). Several monoclonal antibodies against amyloid in early AD have shown the utility of reducing brain amyloid in large phase 3 studies, resulting in modest clinical benefit. To further slow, or even halt disease progression, targeting additional pathobiological pathways is likely to be necessary. In this review, we aim to appraise the scientific rationale for targeting tau in AD. The burden of tau pathology is correlated with disease severity and its role in AD progression is complex, involving synapse dysfunction, neuronal loss, neuroinflammation and autophagy impairment. Evidence suggests that hypersecretion, post-translational modifications and aggregation propensity, dependent and independent of amyloid, are also linked to neurodegeneration. We review the therapeutic agents in development including tau synthesis modifiers, active and passive immunotherapies, post-translational modifiers and aggregation inhibitors. Finally, we consider the available biomarker tools for patient selection and drug effectiveness evaluation, and we identify key knowledge gaps that future novel biomarkers might address to make clinical trials of tau therapies more likely to succeed.