Publications by Year: 2026

OBJECTIVE: We aim to evaluate the effectiveness of the novel real-time continuous glucose monitoring (rtCGM) system "Glunovo" in improving glycemic control and patient outcomes in individuals with poorly controlled type 2 diabetes (T2D).

RESEARCH DESIGN AND METHODS: This prospective, open-label, randomized controlled trial included 172 patients with T2D from the Fatebenefratelli-Sacco Hospital in Milan. Participants were randomized into 2 groups: 86 patients received the Glunovo rtCGM system (case group), whereas 86 continued standard self-monitoring blood glucose with glucometers (control group). The primary outcome was the change in hemoglobin A1c levels after 6 months. Secondary outcomes included glucose metrics and patient well-being assessed by the World Health Organization-5 Well-Being Scale.

RESULTS: After 6 months, the Glunovo group showed a significantly higher reduction in hemoglobin A1c levels (Δ = -1.4%) compared to the control group (Δ = -0.6%). Time in range significantly increased in the rtCGM group (Δ = +18.4%). Time above range and glucose management indicator showed a greater reduction in the rtCGM group, with no changes in the time below range. Patient satisfaction increased significantly over the study period with the rtCGM system.

CONCLUSION: The use of the Glunovo rtCGM system significantly improved glycemic control and patient satisfaction compared to self-monitoring blood glucose. These findings suggest that the Glunovo rtCGM is an effective tool for managing poorly controlled T2D.

CLINICAL TRIAL REGISTRATION: NCT07089979.

BACKGROUND: Large language models (LLMs) like ChatGPT are increasingly being recognized as credible tools for use across diverse healthcare settings. While artificial intelligence (AI) use has previously been evaluated in emergency medicine, its use in subspecialty care - particularly spine surgery - remains underexplored. This study evaluates the clinical accuracy, management appropriateness, completeness, helpfulness, and overall quality of ChatGPT responses compared to those of board-certified, spine surgeons in response to common emergency department (ED) consultations.

METHODS: A 7-part questionnaire was developed based on common ED spine consultations (eg, Cauda Equina Syndrome, compression fracture in elderly patients, purulent drainage from surgical wound, acute lumbar disc herniation, incomplete spinal cord injury, epidural abscess, and metastatic spine disease). Each case included 3-4 questions pertaining to examination, diagnosis, management, and counseling. Responses from ChatGPT and 7 board-certified spine surgeons were restricted to 3-4 sentences per question. Three emergency medicine physicians rated each de-identified questionnaire response using a 5-point Likert scale. Statistical analysis was conducted using a 2-sample T-test with unequal variance. Inter-rater reliability was assessed using pairwise weighted Cohen's kappa coefficient (κ).

RESULTS: When comparing AI responses versus spine surgeon responses to proposed ED consultations, AI responses were rated to be superior across all 5 metrics of clinical accuracy, management appropriateness, completeness, helpfulness, and overall quality (p<.05). Inter-rater reliability was assessed using the average pairwise weighted Cohen's kappa coefficient which showed substantial agreement (κ=0.76).

CONCLUSIONS: ChatGPT responses to emergency department spine consultations were rated as significantly higher compared to board-certified spine surgeons by emergency medicine providers. Though further improvement and validation is warranted, these findings suggest that ChatGPT can be a useful clinical adjunct for spine-related emergency department consultations.

BACKGROUND: Early diagnosis of Alzheimer's disease and related dementias (AD/ADRD) is critical but often constrained by limited access to fluid and imaging biomarkers, particularly in low-resource settings.

OBJECTIVE: To develop and evaluate a predictive model for cognitive decline using survey-based data, with attention to model interpretability and fairness.

METHODS: Using data from the Mexican Health and Aging Study (MHAS), a nationally representative longitudinal survey of adults aged 50 and older (N = 4095), we developed a machine learning model to predict future cognitive scores. The model was trained on survey data from 2003 to 2012, encompassing demographic, lifestyle, and social determinants of health (SDoH) variables. A stacked ensemble approach combined five base models-Random Forest, LightGBM, XGBoost, Lasso, and K-Nearest Neighbors-with a Ridge regression meta-model.

RESULTS: The model achieved a root-mean-square error (RMSE) of 39.25 (95 % CI: 38.12-40.52), representing 10.2 % of the cognitive score range, on a 20 % held-out test set. Features influencing predictions, included education level, age, reading behavior, floor material, mother's education level, social activity frequency, the interaction between the number of living children and age, and overall engagement in activities. Fairness analyses revealed model biases in underrepresented subgroups within the dataset, such as individuals with 7-9 years of education.

DISCUSSION: These findings highlight the potential of using accessible, low-cost SDoH survey data for predicting risk of cognitive decline in aging populations. They also underscore the importance of incorporating fairness metrics into predictive modeling pipelines to ensure equitable performance across diverse groups.

Cancers of the endometrium, cervix, and anus are frequently treated with radiation therapy (RT), which carries a risk for secondary malignancies. RT causes vaginal atrophy, dryness, and increased mucosal friability, predisposing to ulcers. Here, we describe "atypical vulvovaginal postradiation vascular lesion" (AVPRVL), a benign post-RT vascular lesion that histologically mimics angiosarcoma. Fifteen cases were retrospectively identified in patients aged 44 to 77 years (median: 70 y) who presented 3 to 23 years post-RT (median: 14 y), usually with vaginal bleeding. Histologically, AVPRVL demonstrated papillary endothelial hyperplasia (Masson change), mild-to-moderate nuclear atypia, and direct juxtaposition to squamous epithelium. Atypical vessels were located in fibrin and were not distributed in the native lamina propria. MYC immunohistochemistry demonstrated only scattered, weak expression in 5 tested lesions. Fluorescence in situ hybridization was negative for MYC amplification in 11 tested lesions. Five sequenced lesions lacked recurrent pathogenic alterations or copy number changes. Clinical follow-up was available for 11 patients (73%; median length: 3.3 y; range: 1 to 21.7 y). None developed metastases. Six patients (55%) experienced persistent or recurrent lesions, although sometimes it was not possible to distinguish true recurrence from new lesions developing in the RT field. At the most recent follow-up, 9 patients were alive with no evidence of disease, and 2 were alive with recurrent lesions. Ultimately, most lesions resolved with excision or cautery. Unlike RT-associated angiosarcoma, AVPRVL lacked MYC amplification, showed indolent clinical behavior, and usually resolved with conservative management. Unlike cutaneous atypical postradiation vascular proliferation, AVPRVL did not show vessels distributed within native subepithelial tissue. Instead, it more closely resembled a Masson change with nuclear atypia. We conclude that AVPRVL represents a distinctive, benign, possibly reactive vascular proliferation of the vulvovaginal mucosa occurring years after pelvic RT. The distinction between AVPRVL and angiosarcoma is critical to avoid overtreatment.

Fifty years ago, at the remarkably young age of 37, David Baltimore received the Nobel Prize (with Howard Temin and Renato Dulbecco) for "discoveries concerning the interaction between tumor viruses and the genetic material of the cell." David was a prolific scientist whose work spanned many topics, but he was first and foremost a virologist. His recent passing invites us to reflect on a remarkable intellectual trajectory that began with seminal discoveries in virology, broadened to encompass major advances in cancer biology and immunology, and culminated in a legacy-sustained by the many scientists he trained-that will continue to shape modern biomedicine for years to come.

OBJECTIVE: To address previous inconsistencies in reports of differential adherence to diabetes medications, we examined medication adherence and evaluated treatment group differences in a substudy of participants in the Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE).

RESEARCH DESIGN AND METHODS: GRADE participants (type 2 diabetes duration <10 years, HbA1c 6.8%-8.5%, on metformin alone) were randomly assigned to add insulin glargine, glimepiride, liraglutide, or sitagliptin. Adherence was measured semiannually for 3 years using a validated three-item scale (0-100, lowest to highest adherence) in a substudy (N = 1,739). Analyses included evaluation of adherence over time and testing treatment group differences in adherence and in the association between adherence and primary (HbA1c ≥7.0%) and secondary (HbA1c >7.5%) glycemic outcomes.

RESULTS: Overall mean ± SD adherence (average of participant-level mean ± SD) was high over 3 years of follow-up at 88.7 ± 10.01, on a scale of 0-100, and decreased slightly by 3 years relative to baseline (-2.0 ± 14.7; P < 0.0001). No intergroup differences were observed until 3 years, when adherence was 5% and 3% higher for the glimepiride and sitagliptin groups, respectively, than for liraglutide (both P < 0.05). Over follow-up and across groups, a 10-point decrease in adherence was associated with 15% and 19% increased risk of reaching primary (HbA1c ≥7.0%) and secondary (HbA1c >7.5%) glycemic outcomes (both P < 0.0001). Lower adherence was somewhat more predictive of the secondary outcome for those assigned to glargine or liraglutide, compared with glimepiride or sitagliptin (each P < 0.05). No other comparisons were significant.

CONCLUSIONS: Medication adherence was consistently high in GRADE. Observed treatment group differences were small and of unclear clinical significance. Overall, lower adherence robustly predicted worsening glycemic control, highlighting the importance of ongoing assessment.

OBJECTIVE: Pain sketches help visualize neuropathic pain patterns in amputees and may predict surgical outcomes following targeted muscle reinnervation (TMR). Current manual interpretation introduces subjectivity and potential bias. Machine learning offers potential for automated, objective classification of these sketches. Therefore, we aimed to develop and evaluate a machine learning approach for automated classification of pain sketches from lower extremity amputees who underwent targeted muscle reinnervation (TMR).

METHODS: Here, 588 pain sketches from 206 lower extremity amputees (2021-2024) were analyzed. Convolutional neural networks were trained to perform binary classifications between pain categories (focal, radiating, diffuse, and no pain). Unsupervised hierarchical clustering was used to identify novel pain distribution patterns.

RESULTS: Binary classification models achieved the highest performance distinguishing no pain versus diffuse pain (AUROC: 0.799). Other models showed AUROCs between 0.587-0.760. Hierarchical clustering revealed distinct pain distribution patterns based on anatomical location and extent, providing insights beyond traditional classification schemes.

CONCLUSIONS: Machine learning can effectively automate pain sketch classification in lower extremity amputees, offering potential clinical utility for preoperative planning. This approach may help standardize interpretation and improve surgical decision-making for TMR procedures.

LEVEL OF EVIDENCE: IV-Therapeutic.

The stable reactive oxygen species (ROS) hydrogen peroxide (H2O2) acts as a key signaling molecule for many vital intracellular pathways. In diverse cell types, surface receptors control intracellular H2O2 levels by modulating the activity of NADPH oxidases (NOX), a family of enzymes responsible for ROS synthesis. Most NOX isoforms are regulated through the reversible assembly of protein subunits to form an active oligomeric complex. The NOX isoforms NOX2 and NOX4 are expressed in endothelial cells and generate H2O2 in response to activation of cell surface receptors. The GPCR agonist histamine activates NOX2 independently of NOX4, but the H2O2 response to activation of the receptor tyrosine kinase agonist vascular endothelial growth factor (VEGF) involves both NOX2 and NOX4 by unknown mechanisms [M. Waldeck-Weiermair et al., Redox Biol. 58, 102539 (2022); M. Waldeck-Weiermair et al., Redox Biol. 73, 103214 (2024)]. Here, we show that endothelial NOX4 is localized to the endoplasmic reticulum (ER). We define the redox states of various subcellular locales in the vascular endothelium and demonstrate that NOX2 is responsible for cytosolic H2O2 signaling, whereas NOX4 contributes to H2O2 generation in the endoplasmic reticulum. Using biochemical assays and structural modeling, we further identify a previously unrecognized regulatory interaction in which the NOX2 subunit p67 associates with NOX4. VEGF stimulation induces dynamic dissociation of p67 from NOX4, unveiling a "cross talk" between NOX isoforms that coordinates the activation of both NOX2 and NOX4 and thereby produces compartment-specific H2O2 signals. This mechanism underscores the pivotal roles of NOX2 and NOX4 subunit interactions in endothelial redox homeostasis controlling cell survival, proliferation, and migration.

Hearing loss is a risk factor for dementia, but dementia subtypes underlying this association and effect modifiers are unknown. Using data collected from 2000 Framingham Heart Study participants we found that hearing loss increases risk for Alzheimer's disease and vascular dementia in participants aged 60-70 years ("young-old") at time of hearing assessment (Alzheimer's disease: HR 1.46[CI 1.07-2.0] p = 0.017; vascular dementia: HR 2.08[CI 1.22-3.56] p = 0.007). Longer duration of hearing loss determines increased risk for Alzheimer's disease and vascular dementia, and screening and intervention for hearing loss from mid-life may help reduce dementia.