Publications by Year: 2026

Continual learning (CL) enables animals to learn new tasks without erasing prior knowledge. CL in artificial neural networks (NNs) is challenging due to catastrophic forgetting, where new learning degrades performance on older tasks. While various techniques exist to mitigate forgetting, theoretical insights into when and why CL fails in NNs are lacking. Here, we present a statistical-mechanics theory of CL in deep, wide NNs, which characterizes the network's input-output mapping as it learns a sequence of tasks. It gives rise to order parameters (OPs) that capture how task relations and network architecture influence forgetting and anterograde interference, as verified by numerical evaluations. For networks with a shared readout for all tasks (single-head CL), the relevant-feature and rule similarity between tasks, respectively measured by two OPs, are sufficient to predict a wide range of CL behaviors on classic benchmark tasks. In addition, the theory predicts that increasing the network depth can effectively reduce interference between tasks, thereby lowering forgetting. For networks with task-specific readouts (multihead CL), the theory identifies a phase transition where CL performance shifts dramatically as tasks become less similar, as measured by another task-similarity OP. While forgetting is relatively mild compared to single-head CL across all tasks, sufficiently low similarity leads to catastrophic anterograde interference, where the network retains old tasks and interpolates new training data perfectly but completely fails to generalize new learning. Our results delineate important factors affecting CL performance and offer theoretical insights into common heuristics for mitigation of forgetting.

Brainstem white matter (WM) bundles are essential conduits for neural signals that modulate homeostasis and consciousness. Their architecture forms the anatomic basis for brainstem connectomics, subcortical circuit models, and deep brain navigation tools. However, their small size and complex morphology, compared to cerebral WM, makes mapping and segmentation challenging in neuroimaging. As a result, fundamental questions about brainstem modulation of human homeostasis and consciousness remain unanswered. We leverage diffusion MRI tractography to create BrainStem Bundle Tool (BSBT), which automatically segments eight WM bundles in the rostral brainstem. BSBT performs segmentation on a custom probabilistic fiber map using a convolutional neural network architecture tailored to detect small anatomic structures. We demonstrate BSBT's robustness across diffusion MRI acquisition protocols with in vivo scans of healthy subjects and ex vivo scans of human brain specimens with corresponding histology. BSBT also detected distinct brainstem bundle alterations in patients with Alzheimer's disease, Parkinson's disease, multiple sclerosis, and traumatic brain injury through tract-based analysis and classification tasks. Finally, we provide proof-of-principle evidence for the prognostic utility of BSBT in a longitudinal analysis of traumatic coma recovery. BSBT creates opportunities for scalable mapping of brainstem WM bundles and investigation of their role in a broad spectrum of neurological disorders.

Esophageal adenocarcinoma is a major global health concern, primarily arising from gastroesophageal reflux disease, with Barrett's esophagus being its main precursor. Although proton pump inhibitors are commonly used to manage the symptoms from gastroesophageal reflux disease, their role in preventing esophageal adenocarcinoma progression remains uncertain. The aim of this review is to summarize the current advances in the diagnosis of Barrett's esophagus and its progression, as well as to critically evaluate and compare the impact of antireflux surgery on Barrett's esophagus and its potential role in preventing its progression to esophageal adenocarcinoma. In conclusion, surgical intervention, particularly antireflux surgery, has been associated with reduced esophageal adenocarcinoma risk in some studies, offering better long-term symptom control and possibly preventing cancer progression. However, other authors suggest that the cancer risk does not decrease significantly with surgery, highlighting the need for further investigation into its long-term preventive benefits. Several novel strategies have been established over the last few years that will facilitate an early diagnosis of Barrett's esophagus in the future.

IMPORTANCE: Physician emigration from low- and middle-income countries to high-income countries is a major driver of inequitable distribution of health care and health outcomes across the world. World Health Organization (WHO) signatory countries unanimously signed the voluntary Global Code of Practice on the International Recruitment of Health Personnel (WHO Code) in 2010. The goal was to reduce health care workforce emigration by discouraging active recruitment of physicians from WHO-designated shortage countries and by promoting investment in the physician workforce in those countries. This study adds to the literature by providing evidence about whether the goal has been achieved 10 years after worldwide implementation of the WHO Code.

OBJECTIVE: To evaluate whether the WHO Code was associated with changes in physician emigration and physician density in WHO-designated shortage countries after 2010.

DESIGN, SETTING, AND PARTICIPANTS: A difference-in-differences design was used to examine trends in physician supply before and after 2010. The data (from 2000 through 2021) were collected by the Organization for Economic Co-operation and Development (OECD) and were used to examine physician outflow from 56 WHO-designated shortage countries vs 116 nonshortage countries. The data analysis took place October 2024 to September 2025.

EXPOSURES: Worldwide adoption of the 2010 WHO Code.

MAIN OUTCOMES AND MEASURES: The primary outcome was annual physician migration to OECD countries and the secondary outcome was physician density by country and year (per 1000 population using World Bank data) in the origin countries.

RESULTS: A total of 135 888 physicians emigrated from WHO-designated shortage countries during 2000 to 2021 and 516 030 physicians emigrated from nonshortage countries. Compared with nonshortage countries, there was a decrease in physician outflow by 47.03 physicians (95% CI, -92.29 to -1.76 physicians) per country per year after 2010 in WHO-designated shortage countries and the WHO Code was associated with a reduction of nearly 30% in the average annual outflow of physicians from these countries. However, there was a slight decrease in physician density in the WHO-designated shortage countries after 2010 (-0.22 [95% CI, -0.33 to -0.11] physicians per 1000 population) compared with the nonshortage countries.

CONCLUSIONS AND RELEVANCE: This study found that voluntary implementation of the WHO Code was associated with lower physician outflow from WHO-designated shortage countries without improvement in physician density in those countries.

IMPORTANCE: With the availability of multiple classes of advanced therapies for the treatment of Crohn disease (CD), understanding the comparative safety of different therapies can inform treatment positioning.

OBJECTIVE: To compare the risk of serious infections, venous thromboembolism (VTE), and major adverse cardiovascular events (MACE) with different advanced therapies in patients with CD.

DESIGN, SETTING, AND PARTICIPANTS: This retrospective comparative effectiveness research study was conducted between January 1, 2016, and December 31, 2022, with a mean (SD) follow-up of 26.9 (2.4) months until July 1, 2024. Using an administrative claims database (OptumLabs Data Warehouse), commercially insured patients with CD, who initiated treatment with tumor necrosis factor-α (TNF) antagonists, anti-integrin agents (vedolizumab), interleukin (IL)-12/23p40 antagonists (ustekinumab), IL-23p19 antagonists (primarily risankizumab), or Janus kinase inhibitors (upadacitinib) between 2016 and 2022 and had follow-up for at least 1 year before and after treatment initiation, were included.

EXPOSURE: TNF antagonists vs anti-integrin agents (vedolizumab) vs IL-12/23p40 antagonists (ustekinumab) vs IL-23p19 antagonists (risankizumab) vs Janus kinase inhibitors (upadacitinib).

MAIN OUTCOMES AND MEASURES: The risk of serious infections, VTE, and MACE was compared with various advanced therapies through multinomial propensity score-based inverse probability treatment weighting, with propensity scores estimated through generalized boosted models, accounting for disease characteristics, health care utilization, comorbidities, and prior and concomitant medications, and through competing risk of mortality. Cause-specific hazard ratios (HRs) and 95% CIs for multiple treatment comparisons were calculated.

RESULTS: This study included 12 245 patients with CD (mean [SD] age, 46.5 [17.5] years; 6642 females [54.2%]), who were treated with TNF antagonists (n = 5274), vedolizumab (n = 2716), ustekinumab (n = 3544), risankizumab (n = 559), or upadacitinib (n = 152). Serious infection incidence rates ranged from 5.46 (95% CI, 4.86-6.07) to 9.02 (95% CI, 6.38-11.89) per 100 person-years across therapies. After adjusting for confounding variables, there were no statistically significant differences in the risk of serious infections across different agents, including between risankizumab and ustekinumab (HR, 1.14 [95% CI, 0.78-1.67]), risankizumab and TNF antagonists (HR, 1.00 [95% CI, 0.68-1.47]), or ustekinumab and TNF antagonists (HR, 0.88 [95% CI, 0.74-1.04]). The incidence of VTE (incidence rate, 0.90 [95% CI, 0.71-1.10] to 2.33 [95% CI, 1.06-3.82] per 100 person-years) and MACE (0.68 [95% CI, 0.51-0.85] to 1.49 [95% CI, 0.43-2.76] per 100 person-years) was low, without any significant differences across agents.

CONCLUSIONS AND RELEVANCE: In this comparative effectiveness research study of patients with CD, no significant differences in the risks of serious infections, VTE, or MACE across various advanced therapies were found. These findings support clinical decision-making on choice of advanced therapies for most individual patients with CD to be driven primarily by comparative treatment effectiveness rather than driven by concerns of serious adverse events.

BACKGROUND: Neurodevelopmental models regard impulsivity as a central risk factor for adolescent substance use. However, the practical utility of impulsivity in predicting substance use is complicated by variability among measures that encompass multiple methods and theoretical domains. Prior research has been constrained by cross-sectional designs, small sample sizes, and/or the use of a narrow subset of impulsivity measures.

METHOD: Leveraging the ABCD dataset (n = 11,868), we identified and replicated correlations among impulsivity measures and assessed their prospective longitudinal and concurrent predictive utility regarding adolescent substance use outcomes before 15 years old. We then used simulation to inform how associations between impulsivity and substance use vary across sampling strategies (population vs. high-risk cohorts) and sample sizes.

FINDINGS: Correlations between questionnaire and behavioral measures of impulsivity were small, and questionnaires significantly outperformed behavioral measures in predicting substance use initiation, largely due to the contribution of the CBCL externalizing scale. Predictions of substance use based on impulsivity were statistically detectable but small according to clinical standards (AUCs 0.6-0.76), exhibiting sensitivity to sample size and base rate of substance use, and thus, poor absolute predictive performance. Large samples (n > 1,000) were needed to achieve adequate power for impulsivity measures to predict substance use initiation.

CONCLUSION: These results support a significant but small contribution of impulsivity in predicting the onset of early adolescent substance use, indicating that these factors alone are insufficient for clinically deployable prediction. In community samples, large sample sizes are needed for reproducible impulsivity prediction of adolescent substance use.

PURPOSE: To investigate the effect of high-iodine concentration contrast material for dual-energy CT angiography (DECTA) on arterial visualizations by comparing it with medium-iodine concentration.

METHOD: This prospective, single-blind, randomized controlled trial included 100 consecutive participants undergoing DECTA from November 2023 to February 2025. The participants were randomly assigned into two protocols: receiving high-iodine concentration contrast material of 370 mgI/mL (Group A, n = 51) and receiving medium-iodine concentration of 300 mgI/mL (Group B, n = 49). The axial, coronal, and volume-rendered (VR) images were reconstructed at 40 keV in both groups. Two radiologists reviewed three image types and assessed the arterial visualizations using a five-point scale. The primary outcome was the score for the iliolumbar artery on the VR images, whereas secondary outcomes were the scores for all others. The Wilcoxon rank sum test was conducted to compare the outcomes between the two groups.

RESULTS: No statistical significance in terms of the score for the iliolumbar artery on the VR images was found between the two groups; however, the median score was higher in Group A than in Group B (3.5 vs. 3.0; P = .05). The scores for secondary outcomes in Group A were equal to or greater than that in Group B, and significant differences were observed, especially in the small arteries, including the bronchial, internal thoracic, intercostal, left gastric, and inferior phrenic arteries (P < .05).

CONCLUSION: In small arteries, the protocol with high-iodine concentration contrast material exhibited better arterial visualizations compared with medium-iodine concentration in DECTA at 40 keV.

BACKGROUND: Reduced mismatch negativity (MMN) is a widely replicated schizophrenia biomarker. Time-frequency analyses suggest that deficient phase synchrony and/or power of electroencephalography (EEG) event-related oscillations, especially theta, contribute to MMN deficits in schizophrenia. Whether theta oscillations assessed from a passive auditory oddball paradigm are abnormal in clinical high-risk for psychosis (CHR-P) individuals and whether these oscillations predict CHR-P clinical outcomes remain unclear. These questions were addressed using data from NAPLS2 (North American Prodrome Longitudinal Study 2).

METHODS: EEG was recorded from 77 CHR-P individuals who converted to psychosis (CHR-Cs), 238 CHR-P nonconverters (CHR-NCs) who completed a 24-month follow-up, and 241 healthy control (HC) individuals. Theta oscillations elicited by standard and deviant tones were calculated. Theta (4-6 Hz) intertrial phase coherence (ITC) and total power were compared between groups and evaluated as predictors of time to psychosis conversion in the full CHR-P sample. Furthermore, analyses of covariance were used to assess whether theta deficits persisted while covarying for MMN.

RESULTS: The CHR-C group, relative to the HC and CHR-NC groups, had reduced theta ITC for standards and deviants (ps < .029) and reduced total power for standards and deviants relative to the HC group (p = .021). Reduced theta ITC for standards and deviants predicted earlier psychosis conversion. The previously reported deficit in MMN amplitude in the CHR-C group was no longer significant after accounting for theta ITC or total power averaged across stimuli (ps > .187), but this was not true for theta ITC or total power deviant-standard difference scores (ps < .039).

CONCLUSIONS: These results implicate abnormalities in microcircuit generators of theta oscillations in CHR-P individuals at highest risk.

BACKGROUND: Relapsed medulloblastoma remains a significant therapeutic challenge as it is near universally fatal. The tumor microenvironment of medulloblastoma plays a critical role in tumor progression, influencing tumor growth, immune evasion, and therapeutic resistance. We hypothesised that defining tumor-immune interactions in diagnostic and relapsed medulloblastoma may uncover mechanisms of immune evasion and identify novel therapeutic targets.

METHODS: We analysed paired primary and recurrent RNA-sequencing data from 140 medulloblastoma patients to profile immune cell composition and validate spatial relationships within the tumor microenvironment. To identify key tumor-immune interactions, we developed a novel algorithm to detect receptor-ligand pairs using single cell RNA-sequencing data. These interactions were validated across RNA and proteomic datasets. Their functional significance was empirically demonstrated in newly developed immunocompetent models of recurrent medulloblastoma that closely recapitulate the human disease.

RESULTS: We observed a shift in toward a heightened immunosuppressive tumor microenvironment at relapse. Using our algorithm, we identified biologically significant receptor-ligand interactions, most notably MIF-CD74, constitutively expressed at RNA and protein levels across medulloblastoma subgroups, at diagnosis and relapse. Disrupting MIF-CD74 interactions led to significant alterations in the tumor microenvironment, highlighting its functional significance.

CONCLUSIONS: Our multifaceted approach identified key tumor-immune interactions in medulloblastoma. Among these, MIF-CD74 was validated as a targetable interaction, demonstrating the utility of our integrative approach for identifying novel therapeutic targets across multiple tumor types.