Citations

Purpose of review Ophthalmology has emerged as the proving ground for medical artificial intelligence (AI) because of its imaging-centric workflows, standardized data acquisition, and well defined clinical endpoints. Recently, the field has started transitioning from narrowly trained, task-specific algorithms toward multimodal models that are trained on large datasets and can be adapted to many tasks. These models may integrate information from images, text, and clinical data and are increasingly trialed in real-world settings. This review examines recent advances and ongoing controversies in the use of AI as a basic science, translational, and clinical research tool in ophthalmology. Recent findings Recent literature highlights the growing role of self-supervised and foundation models for scalable phenotyping, biomarker discovery, and cross-task generalization, particularly in retinal imaging and optical coherence tomography (OCT). In parallel, large-scale clinical deployments, including cloud-based OCT platforms and autonomous diabetic retinopathy screening, demonstrate clinical impact while revealing persistent challenges related to workflow integration and patient generalizability. Regulatory and ethics-focused studies underscore increasing attention to validation rigor, bias, and postdeployment monitoring, exposing heterogeneity in research standards and physician practices. Summary Ophthalmic AI is maturing toward real-world clinical implementation. Future progress will depend less on performance gains alone and more on clinical alignment, rigorous validation, equity-aware evaluation, and postdeployment surveillance, alongside continued investment in basic science to advance disease understanding and inform robust, generalizable model development.

Although complications of autologous bone flap cranioplasty are well described, data on long-term bone flap survival and the comorbidities associated with secondary reconstruction are limited. This study aimed to quantify the need for secondary alloplastic cranioplasty within a standardized patient cohort. Patients who underwent delayed cranioplasty following craniectomy with banked, frozen autologous bone between 2012 and 2025 were identified and analyzed using Cox regression, spline analysis, and Kaplan–Meier survival analysis. A total of 367 patients with an average age of 39.2±19.7 years (range: 0.3–84.4 years) at the time of cranioplasty were included. A secondary cranioplasty was required in 13.6% of patients (n=50 of 367), with resorption (52%) and infection (42%) as the most common indications. In adjusted analyses, patients who underwent cranioplasty following a traumatic injury (HR: 2.37; 95% CI: 1.07–5.26; P =0.03) and patients younger than 18 years (HR: 2.99; 95% CI: 1.35–6.62; P =0.01) were more likely to require a secondary cranioplasty due to bone flap resorption. Increasing defect size was also associated with a higher risk of secondary cranioplasty due to bone flap resorption, becoming significant at roughly 95 cm² (HR: 2.54; 95% CI: 1.04–6.20; P =0.04). The estimated cumulative incidence of secondary cranioplasty was 37% at ten years after autologous reconstruction. These results indicate that although autologous bone flap cranioplasty is largely effective, select patients face a substantially higher risk of secondary cranioplasty, and alternative reconstructive options may be warranted.

Objectives: Identify the frequency of abnormal neurological examinations among patients presenting in person for headache consultation. Determine what proportion of patients with abnormal examinations have accompanying red flags in their histories and/or undergo diagnostic workup. Methods: We conducted a retrospective review of patients presenting in person for headache consultation at a tertiary/quaternary headache clinic between June 2022 and May 2023. We excluded examination abnormalities consistent with known comorbidity, functional neurological disorder, and/or non-headache disease. We then calculated the proportion of patients with an abnormal neurological examination. For patients with abnormal examinations, the electronic medical record was further reviewed for red-flag components in the history, diagnostic workup ordered, and the resulting diagnosis. Results: Of 1281 patients presenting for in-person headache consultation, 218 (17%) had abnormal neurological examinations that were not consistent with known comorbidity, functional neurological disorder, or non-headache disease. Of these, 203 (93%) had a red flag in their history. The remaining 15 patients with abnormal examinations and no historical red flags comprised 1.2% of all 1281 patients. Regarding workup, 160 (73%) had neuro-imaging and 61 (28%) had laboratory studies ordered and/or requested for review, of which 13% and 8% were concerning for pathology, respectively. Conclusions: A small but notable proportion of patients presenting for in-person headache consultation had abnormalities on neurological examination. The majority of these patients received imaging, with a diagnostic yield higher than previously reported. However, the vast majority of patients with abnormal examinations also had red flags in their history to prompt diagnostic workup.

Abstract Background Preparing tumor board patient summaries is time intensive. Large-language-model based systems may automate summarization but require real-world evaluation prior to clinical use. We performed an exploratory retrospective evaluation of the Microsoft Healthcare Agent Orchestrator (HAO), deployed in a Mayo Clinic–controlled staged environment, to generate tumor board–style patient summaries from retrospective Electronic Health Record (EHR) notes. Methods HAO generated summaries for breast, hepatobiliary, and neuro-oncology tumor board cases using up to the most recent 1,000 clinical notes. Clinician reviewers evaluated outputs via REDCap surveys across perceived factuality, completeness, clarity/conciseness, temporal cohesion, comparative performance, safety, and clinical utility (0–4 Likert scale). Reviewers were permitted to query the HAO chat interface to address missing details. Automated factuality was assessed using TBFact (bidirectional entailment), reporting precision and recall against available reference summaries. Results Among 57 survey responses from 5 different physicians, mean scores exceeded 2.8 across domains, with medians of 3 for most axes. In an exploratory comparison, oncology fellows required less time to review HAO-generated summaries than to manually generate patient summaries (mean difference 13.57 minutes per patient, p<0.001), although this difference may be influenced by prior familiarity with the same cases; 96% of survey responses indicated that HAO would save time. TBFact evaluations showed higher recall than precision across domains, consistent with broad capture of reference content alongside additional content that was not present in gold-standard summaries. Attribution was viewed favorably but showed issues with primary-source specificity and link reliability. Conclusions In a controlled Mayo environment, HAO demonstrated moderate performance and was associated with reduced review time for tumor board preparation. These findings are promising but preliminary and do not establish clinical safety, noninferiority to manual review, or readiness for routine clinical use. Limitations, including verbosity, specialty-specific content gaps, and inconsistent attribution, highlight the need for iterative refinement and further evaluation. Graphical Abstract Created using ChatGPT

Abstract Background Sepsis is a life-threatening condition in which delayed recognition and treatment are associated with increased mortality. While predictive models such as Epic’s Early Detection of Sepsis Model (ESM) were developed to support early intervention, their real-world impact after integration into clinical workflows remains difficult to evaluate. Objectives To evaluate the real-world impact of ESM integrated into clinical workflow on clinical outcomes, antibiotic use, and harm-benefit tradeoffs. Methods We conducted a quasi-experimental study in a single healthcare system using encounter-level data from inpatient settings. Inpatient mortality, prolonged hospitalization, antibiotic use, and sepsis prevalence were compared between the pre-implementation period (3 June 2023 to 20 August 2024) and the online period (21 August 2024 to 26 December 2024) when the model became visible to clinicians. We also applied a counterfactual framework using models trained on pre-implementation data to estimate expected outcomes without ESM and to quantify harms related to overtreatment and delayed treatment. Results Among 101,138 encounters, 86,884 occurred during the pre-implementation period and 14,254 during the online period. In unadjusted analyses, the online period had lower inpatient mortality, prolonged hospitalization, antibiotic use, and sepsis prevalence (all p≤0.002). In the counterfactual analyses, observed outcomes were lower than expected without ESM for mortality (1.21% vs 1.82%; p<0.001), prolonged hospitalization (5.56% vs 7.95%; p<0.001), and antibiotic use (43.52% vs 47.04%; p<0.001). False positive harm (37.72% vs 41.68%; p<0.001) was also lower than expected. Conclusions Integration of ESM into clinical workflow was associated with improved patient outcomes, reduced antibiotic use, and decreased harm from overtreatment, without evidence of increased harm from delayed treatment, supporting a positive net clinical benefit and the safety and effectiveness of ESM under Software as a Medical Device principles.

This cross-sectional study evaluates the performance of the US Food and Drug Administration–cleared phosphorylated tau 217 (p-tau217) to β-amyloid 42 ratio cut points used to detect amyloid pathology associated with Alzheimer disease.

This case report describes a 65-year-old man who experienced complete remission of cavin-4 immunoglobulin G (IgG) rippling muscle disease after receiving rituximab therapy.

Abstract Purpose: Fibroblast activation protein (FAP) is an attractive target for radiopharmaceutical therapy. Phase I of the LuMIERE study (ClinicalTrials.gov, NCT04939610) investigated the safety of [177Lu]Lu-FAP-2286 (177Lu-FAP-2286) in heavily pretreated patients with advanced solid tumors and identified the recommended phase II dosage (RP2D). Patients and Methods: LuMIERE is a prospective, open-label, nonrandomized, phase I/II, multicenter study. Phase I followed a Bayesian optimal interval design evaluating four escalating activity levels of 177Lu-FAP-2286 (3.70, 5.55, 7.40, and 9.25 GBq). Patients were selected by positive [68Ga]Ga-FAP-2286 (68Ga-FAP-2286, also known as [68Ga]Ga-HKG301) PET/CT imaging on all target lesions [maximum standardized uptake value (SUVmax) ≥1.5× SUVmean of mediastinal blood pool]. 177Lu-FAP-2286 was administered intravenously every 6 weeks for ≤6 cycles. The primary endpoint was dosage-limiting toxicities (DLT) and treatment-emergent adverse events (TEAE). Results: Of 35 patients imaged with 68Ga-FAP-2286, 27 received 177Lu-FAP-2286 (3.70 GBq, n = 3; 5.55 GBq, n = 6; 7.40 GBq, n = 7; 9.25 GBq, n = 11). Two DLTs were observed: one patient who received 5.55 GBq of 177Lu-FAP-2286 experienced grade 4 lymphopenia, and one patient who received 9.25 GBq experienced grade 3 hemoptysis; these were the only treatment-related grade ≥3 TEAEs. All-cause grade ≥3 TEAEs were reported for 11 patients (40.7%). The RP2D for 177Lu-FAP-2286 monotherapy was determined to be 9.25 GBq. Preliminary Response Evaluation Criteria in Solid Tumors efficacy findings demonstrated a partial response in 1 patient and stable disease in 10 patients (maintained for at least two assessments in 6 patients). Conclusions: 177Lu-FAP-2286 was well tolerated in patients with advanced solid tumors. The safety and preliminary efficacy findings support its continued development in phase II.

Background The pathophysiology of CS is complex and is associated with increased inflammation and impaired vascular tone. Corticosteroids are recommended in septic shock and have been proposed as a potential treatment for other types of shock. Objectives We sought to evaluate the clinical outcomes associated with early corticosteroid use in patients with cardiogenic shock (CS). Methods Using a nationally representative database including over 1000 hospitals, we identified adults ≥18 years of age admitted from 2015–2023 with a diagnosis of CS. Patients with adrenal insufficiency, chronic rheumatologic conditions, COVID-19 infection and acute COPD exacerbation were excluded. Using inverse probability treatment weighting (IPTW), we assessed for the association of receiving early corticosteroids (within the first 2 days of admission) versus no early corticosteroids and in-hospital mortality. Results Of the 167,721 identified patients with CS, the mean (SD) age was 65.5 (±15.2) years and 35.0% were women. A total of 13.2% received any corticosteroid within the first 2 days of admission. The most common corticosteroid was hydrocortisone (73.9%). Mortality for those receiving and not receiving early corticosteroids was 48.8% and 29.6% ( p < 0.001), respectively. After IPTW, early corticosteroid use remained associated with a 3.1% (95% confidence interval [CI]: 2.4% to 3.8%, p < 0.001) higher mortality. Among patients with CS and concomitant sepsis, 27.6% received early corticosteroids, which was similarly associated with a higher mortality (weighted mean 5.8% [95% CI: 4.6% to 7.0%, p < 0.001]). Conclusions Approximately 1 in 7 patients with CS received corticosteroids early during their admission, which was associated with higher in-hospital mortality.