Category: 3. Business

NEW YORK, Jan. 22, 2026 /PRNewswire/ — The Bank of New York Mellon Corporation (“BNY”) (NYSE: BK), a global financial services company, today announced that Dermot McDonogh, Chief Financial Officer, will speak at the BofA Securities Financial Services Conference in Miami, Florida at 12:10 p.m. ET on Wednesday, February 11, 2026. The discussion may include forward-looking statements and other material information.

A live webcast of the audio portion of the conference will be available on the BNY website (www.bny.com/investorrelations). An archived version of the audio portion will be available on the BNY website approximately 24 hours after the live webcast and will remain available until March 11, 2026. 

About BNY
BNY is a global financial services platforms company at the heart of the world’s capital markets. For more than 240 years BNY has partnered alongside clients, using its expertise and platforms to help them operate more efficiently and accelerate growth. Today BNY serves over 90% of Fortune 100 companies and nearly all the top 100 banks globally. BNY supports governments in funding local projects and works with over 90% of the top 100 pension plans to safeguard investments for millions of individuals. As of December 31, 2025, BNY oversees $59.3 trillion in assets under custody and/or administration and $2.2 trillion in assets under management.

BNY is the corporate brand of The Bank of New York Mellon Corporation (NYSE: BK). Headquartered in New York City, BNY has been named among Fortune’s World’s Most Admired Companies and Fast Company’s Best Workplaces for Innovators. Additional information is available on www.bny.com. Follow on LinkedIn or visit the BNY Newsroom for the latest company news.

Contacts:

Investors
Marius Merz
+1 212 298 1480
marius.merz@bny.com

Media
Anneliese Diedrichs
+1 646 468 6026
anneliese.diedrichs@bny.com

SOURCE BNY

Introduction

Extracranial carotid plaques are biomarkers of coronary artery disease and cerebral ischemic events, including ischemic heart disease and stroke. The global prevalence of carotid plaques among individuals aged 30‐79 years is estimated at 21.1% (n=815.76 million) in 2020. This high prevalence reflects a growing global burden of cardiovascular and cerebrovascular diseases, posing a significant challenge to public health systems []. Therefore, early detection and management of carotid plaque can potentially reduce the risk of stroke and cardiovascular events [-], and thus, effective detection and classification technologies need to be prioritized.

Imaging methods for carotid plaque imaging, such as ultrasound, computed tomography angiography (CTA), magnetic resonance imaging (MRI), and digital subtraction angiography, facilitate detection, stenosis assessment, and plaque composition analysis []. Conventional ultrasound is the first-line screening method []. Studies show that periapical radiographs (PRs) can serve as a supplementary screening tool, demonstrating a 50% concordance with ultrasound or CTA [-]. Current imaging primarily identifies high-risk features, such as plaque neovascularity, lipid-rich necrotic cores, thin fibrous caps, and intraplaque hemorrhage plaque ulceration [,]. Among them, the contrast-enhanced ultrasound or superb microvascular imaging can accurately quantify neovascularization and correlates well with histopathology [-], offering rapid, noninvasive, and reliable quantification []. It is proficient in vascular imaging and ulcer detection [], as well as stenosis assessment [], but it faces challenges with small lipid cores and thin fibrous caps []. MRI remains the gold standard for assessing plaque composition, particularly for identifying lipid cores and intraplaque hemorrhage []. While digital subtraction angiography is the reference standard, its invasive nature limits its application. Notably, the accuracy of these diagnostic techniques largely relies on the expertise of imaging or clinical physicians, which causes inconsistencies in the assessment results of carotid atherosclerotic plaques—particularly in measuring carotid intima-media thickness, characterizing intraplaque components, and evaluating fibrous cap integrity.

The radiomics algorithms and deep learning (DL) models have demonstrated significant potential in medical image analysis []. Radiomics is a quantitative medical imaging analysis approach that aims to transform high-dimensional image features (such as texture heterogeneity, spatial topological relationships, and intensity distribution) into quantifiable digital biomarkers, thereby providing objective evidence to guide clinical decision-making. However, the characteristic dimensionality of radiomics data often far exceeds sample sizes, which renders the traditional statistical methods inadequate []. Machine learning (ML), with the potential to process large-scale, high-dimensional data and uncover deep correlations among these complex features []. Combining radiomics with ML to develop an ML model using radiomics can enhance the diagnostic performance of AI in large and complex datasets, exceeding the performance of models constructed through traditional statistical methods.

DL is also one of the important subbranches of artificial intelligence, which can automatically learn and layer from raw data without manual design of features, ultimately generating predictions via an output layer []. DL-driven image generation techniques have demonstrated remarkable effectiveness in cross-modality imaging and synthesis tasks across various sequences within the same modality. With the rapid development of computer technology, ML models based on radiomics and DL models based on radiomics have become important tools for cardiovascular disease research. Current evidence suggests that these methods can significantly improve the quantitative assessment accuracy of atherosclerotic plaque progression and enhance the diagnostic and predictive power of major adverse cardiovascular events [-]. In recent years, research on the application of these methods in the fields of plaque diagnosis, stability assessment, and symptomatic plaque identification has increased significantly. Although these advancements have significantly improved the diagnosis of carotid plaques, variations in data dependency and imaging configurations among different models create inconsistencies in diagnostic accuracy. Moreover, these models may become overly specialized in common imaging configurations, even when using radiomics data from identical sources. Currently, systematic evaluations of its clinical validity remain limited.

Therefore, this systematic review comprehensively assesses the applications of ML models based on radiomics algorithms and DL models in carotid plaques, while highlighting gray areas in the available literature.

Methods

Study Registration

The study was performed in line with the PRISMA-DTA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses of Diagnostic Test Accuracy Studies) guidelines [] and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) standards [,] and was registered on the International Prospective Register of Systematic Reviews (PROSPERO CRD42025638492).

Data Sources and Search Strategy

Relevant articles were searched on PubMed, Embase, Web of Science, Cochrane Library, and Institute of Electrical and Electronics Engineers (IEEE) databases, focusing on English-language articles published up to September 24, 2025. The literature search was based on the PIO (population, intervention, and outcomes) principles: “P” represents carotid artery disease, carotid plaques, or atherosclerosis populations; “I” represents radiomics or DL as interventions; and “O” represents the outcomes of diagnosis and their subordinates and other keywords. Furthermore, we manually analyzed the reference lists of all included articles to identify additional relevant publications. The complete search strategy is outlined in Table S1 in . The EndNote 20 software (Clarivate Analytics) was used to manage the included studies.

Eligibility Criteria

Inclusion Criteria

The inclusion criteria included:

1. Studies on patients with extracranial carotid plaques that aimed to detect or distinguish between unstable and symptomatic plaques, among other factors.
2. Studies using radiomics algorithms or DL models based on medical imaging techniques, such as ultrasound, CTA, or MRI, to diagnose carotid plaques.
3. Studies reported the diagnostic performance metrics, including confusion matrix, 2×2 diagnostic tables, accuracy, sensitivity, specificity, receiver operating characteristic (ROC) curves, F₁-score, precision, recall, etc.
4. Those that adopted the following designs: prospective or retrospective cohorts, diagnostic accuracy trials, model development or validation studies, and comparative studies (eg, AI models vs AI models combined with clinical features).
5. Only studies published in English and with extractable quantitative data were deemed eligible.

Exclusion Criteria

The exclusion criteria excluded:

1. Studies involving nonhuman subjects (animal experiments or in vitro models), those that explored intracranial or coronary plaques, enrolled pediatric populations (<18 years), or reported only generalized atherosclerosis without plaque-specific criteria (focal intima-media thickness ≥1.5 mm) or specific diagnostic metrics;
2. Those that did not adopt well-defined deep learning models or radiomics algorithms, focused only on image segmentation or texture analysis without diagnostic validation, or reported predictive models without providing a clear diagnostic relevance.
3. Studies that lacked a validated reference standard.
4. Studies that did not report diagnostic performance.
5. Informal publication types (eg, reviews, letters to the editor, editorials, and conference abstracts).
6. Studies that did not report validation or test sets.

Screening of Articles and Data Extraction

In the initial screening, duplicates were excluded followed by reading of full texts, and data were entered into a predefined extraction table, which included surnames of authors, source of data, publication year, algorithm architecture, type of internal validation, availability of open access data, external verification status, reference standard, transfer learning application, number of cases for training, test, internal, or external validation, study design, sample size, mean or median age, inclusion criteria, and model evaluation metrics. The contingency tables are derived from the models explicitly identified by the original authors as the best-performing ones. Data from external validation sets were prioritized. If there were no external validation set in the original studies, data from internal validation sets were used. If neither was available, the contingency tables corresponding to the test sets were selected. This process was performed by two researchers (LJ and YG), working independently, and any differences were resolved through discussion with a third researcher (HG).

Quality Assessment

Two blinded investigators (LJ and YG) systematically assessed the quality of studies using the Quality Assessment of Diagnostic Accuracy Studies for Artificial Intelligence (QUADAS-AI) tool. Specifically, they evaluated the risk of bias and applicability concerns across 4 domains: flow and timing, reference standard, index test, and participant selection. Although the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) is extensively applied to assess the quality of diagnostic accuracy studies [], it does not address the specific methodological choices, result analyses, and measurements related to diagnostic studies using AI. To address this gap, QUADAS-AI was developed as a consensus-based tool to aid readers in systematically examining the risk of bias and the usability of AI-related diagnostic accuracy studies (Table S6 in ) [], thereby improving the quality assessment process [,]. Any evaluation discrepancies were resolved by a third investigator (HG).

Statistical Analysis

A meta-analysis was performed using STATA/MP software (version 17.0; Stata Corporation) with a bivariate random-effects model. For meta-analyses of the diagnostic accuracy of AI-based models, bivariate mixed-effects models can account for both within-study variability (random effects) and between-study heterogeneity (fixed effects), ensuring the robustness of the pooled estimates []. A contingency table was generated using data from the included literature, and then we calculated metrics such as the number of cases, the Youden index, sensitivity, specificity, and recall. The diagnostic efficacy of radiomics algorithms and DL models in evaluating carotid plaque was determined using a summary receiver operating characteristic (SROC) curve and area under the curve (AUC; 0.7≤AUC<0.8 fair; 0.8≤AUC<0.9 good; and AUC≥0.9 excellent). Publication bias was explored using Deeks funnel plot asymmetry test. The Fagan nomogram was developed to determine clinically pertinent posttest probabilities (P-post) and likelihood ratios (LRs). LRs were determined by comparing the probability of test results between diseased and nondiseased groups. The pretest probability was subsequently adjusted based on test results and LRs to obtain P-post []. The Cochran Q (P≤.05) and I² statistic were used to explore heterogeneity among the included studies, and regression analysis was conducted to assess sources of heterogeneity. I²≤50% indicated mild heterogeneity, 50%<I²<75% reflected moderate heterogeneity, and I²≥75% indicated high heterogeneity.

The subgroup analysis encompassed the following factors: (1) model type (DL or ML model), (2) medical imaging modalities (PRs, ultrasound, MRI, or CTA), (3) application of transfer learning, (4) characteristics of carotid plaques (presence vs absence, stable vs vulnerable, and symptomatic vs asymptomatic), (5) comparison of the most effective ML model based on radiomics algorithm and DL models using the same dataset and clinicians’ diagnoses, (6) different types of datasets (testing and validation), (7) low and high or unclear risk of bias studies, (8) different sample sizes of model, and (9) models with different research designs (multicenter studies and single-center studies). To identify the sources of heterogeneity associated with nonthreshold effects, meta-regression was performed using the above-mentioned covariates.

Sensitivity analysis was performed to assess the stability of the results by several steps: (1) excluding specific articles one by one to determine the stability of the results, (2) excluding studies with extremely large sample sizes (N≥500; n=7 studies), (3) excluding studies with extremely small sample sizes (N≤50; n=4 studies), and (4) excluding studies with extreme effect sizes (sensitivity or specificity>0.95 or <0.7; n=11 studies).

Results

Study Selection

We obtained 5834 studies in the initial analysis, of which 1233 were excluded for duplication or redundancy. After screening titles and abstracts, 4507 publications were eliminated. After the full texts of the 94 articles were read, 40 studies were eligible for meta-analysis. The PRISMA flow diagram of the study showing the selection process is presented in .

‎

Study Characteristics

Among the 40 studies that fulfilled the systematic review’s inclusion criteria, 34 provided sufficient quantitative data (contingency tables from validation or test sets) eligible for incorporation into the meta-analysis. The detailed characteristics of all 40 eligible studies are summarized in Tables S3 and S4 in , while all subsequent quantitative analyses were conducted based on the 34 studies with available quantitative data. Overall, 34 studies were included [-], among which 9 were multicenter studies [,,,,,-,], 3 used public databases [,,], 13 provided open access to the data [,,,-,,,,-]. A total of 12 studies conducted internal validation [,,,,,,,,,,,] to confirm the reproducibility of the model development process and prevent overfitting. In addition, 7 studies conducted external validation [,,,,,,] to assess the model’s transportability and generalizability using unused datasets. Only 1 study conducted a comparative analysis of the diagnostic performance of DL models with that of clinicians []. The medical imaging modalities included PRs (n=5), ultrasound (n=16), MRI (n=5), and CTA (n=8). The core features of the 34 studies are presented in and , with further details provided in Tables S2 and S3 in .

Meta-Analysis of Diagnostic Performance

Synthesized Results

The meta-analysis revealed pooled sensitivity, specificity, and an area under the SROC curve (SROC AUC) of 0.88 (95% CI 0.85‐0.91; I²=93.58%; P<.001; in [-]), 0.89 (95% CI 0.85‐0.92; I²=91.38%; P<.001; in [-]), and 0.95 (95% CI 0.92‐0.96) for all 34 studies (); 0.88 (95% CI 0.84‐0.92; I²=93.70%; P<.001; [-]), 0.91 (95% CI 0.86‐0.94; I²=95.55%; P<.001; [-]), and 0.95 (95% CI 0.93‐0.97) for all DL models (); 0.89 (95% CI 0.82‐0.93; I²=90.20%; P<.001; [-]), 0.83 (95% CI 0.76‐0.88; I²=78.92%; P<.001; [-]), and 0.92 (95% CI 0.89‐0.94) for all ML models based on radiomics algorithms (), respectively. Notably, some studies used multiple diagnostic models; however, the diagnostic accuracy of certain models was not thoroughly assessed.

‎

Subgroup Analysis

Medical Imaging Modalities

The pooled sensitivity, specificity, and SROC AUC were 0.91 (95% CI 0.80‐0.96), 0.93 (95% CI 0.84‐0.97), and 0.97 (95% CI 0.95‐0.98) for the 5 studies using PRs (P<.001; with 5 contingency tables; ); 0.89 (95% CI 0.84‐0.93), 0.90 (95% CI 0.84‐0.94), and 0.95 (95% CI 0.93‐0.97) for the 16 studies using ultrasound images (P<.001with 16 contingency tables; ); 0.87 (95% CI 0.87‐0.92), 0.87 (95% CI 0.76‐0.93), and 0.93 (95% CI 0.91‐0.95) for the 5 studies using MRI images (P<.001; with 5 contingency tables; ); 0.83 (95% CI 0.76‐0.88), 0.83 (95% CI 0.75‐0.89), and 0.90 (95% CI 0.87‐0.92) for the 8 studies using CTA images (P<.001; with 8 contingency tables; ), respectively. In addition, we conducted subgroup analyses using the same imaging modality based on differentiation. However, only subgroups of identifying the presence and stability of plaque had sufficient data for the ultrasound modality to perform statistical analyses and obtain pooled diagnostic performance metrics (Table S5 in ). The pooled sensitivity, specificity, and SROC AUC were 0.88 (95% CI 0.72‐0.96), 0.91 (95% CI 0.80‐0.96), and 0.95 (95% CI 0.93‐0.97) for determining the presence of plaques (P<.001; with 5 contingency tables; ), 0.90 (95% CI 0.84‐0.94), 0.92 (95% CI 0.83‐0.96), and 0.96 (95% CI 0.94‐0.97) for distinguishing the stability of plaques (P<.001; with 8 contingency tables; ).

‎

Use of Transfer Learning

The pooled sensitivity, specificity, and SROC AUC were 0.92 (95% CI 0.87‐0.95), 0.93 (95% CI 0.88‐0.96), and 0.97 (95% CI 0.95‐0.96) for the 10 studies using transfer learning (P<.001; with 10 contingency tables; ) and 0.86 (95% CI 0.82‐0.90), 0.86 (95% CI 0.81‐0.90), and 0.93 (95% CI 0.90‐0.95) for the 24 studies without transfer learning (P<.001; with 24 contingency tables; ), respectively.

‎

Carotid Plaque Type

The pooled sensitivity, specificity, and AUC were 0.89 (95% CI 0.81‐0.94), 0.91 (95% CI 0.86‐0.95), and 0.96 (95% CI 0.94‐0.97) for the 11 studies identifying the presence or absence of carotid plaques (P<.001; with 11 contingency tables; ); 0.90 (95% CI 0.85‐0.94), 0.91 (95% CI 0.85‐0.95), and 0.96 (95% CI 0.94‐0.97) for the 12 studies identifying stable or vulnerable carotid plaques (P<.001; with 12 contingency tables), respectively (); and 0.86 (95% CI 0.78‐0.91), 0.81 (95% CI 0.74‐0.87), and 0.90 (95% CI 0.87‐0.92) for the 10 studies identifying symptomatic or asymptomatic plaques (P<.001; with 10 contingency tables; ), respectively.

‎

Pure Artificial Intelligence Models Versus Models Constructed by Combining Clinical Features

The pooled sensitivity, specificity, and SROC AUC were 0.82 (95% CI 0.74‐0.88), 0.74 (95% CI 0.69‐0.79), and 0.77 (95% CI 0.73‐0.80) for the 7 studies involving pure artificial intelligence models meeting the inclusion criteria (P<.001; with 7 contingency tables; ) and 0.85 (95% CI 0.76‐0.92), 0.75 (95% CI 0.70‐0.80), and 0.77 (95% CI 0.73‐0.81) for models constructed by combining clinical features (P<.001; with 7 contingency tables; ), respectively.

‎

Different Sets of Datasets

The pooled sensitivity, specificity, and AUC were 0.90 (95% CI 0.87‐0.93), 0.91 (95% CI 0.87‐0.93), and 0.96 (95% CI 0.94‐0.97) for testing sets (P<.001; with 27 contingency tables; ); 0.78 (95% CI 0.71‐0.83), 0.80 (95% CI 0.73‐0.86), and 0.86 (95% CI 0.82‐0.88) for external validation sets (P<.001; with 7 contingency tables; ), respectively.

‎

Low and High or Unclear Risk of Bias Studies

The pooled sensitivity, specificity, and AUC were 0.80 (95% CI 0.73‐0.85), 0.80 (95% CI 0.71‐0.87), and 0.86 (95% CI 0.83‐0.89) for studies with a low risk of bias (P<.001; with 5 contingency tables; ), and 0.89 (95% CI 0.86‐0.92), 0.90 (95% CI 0.86‐0.93), and 0.95 (95% CI 0.93‐0.97) for studies with a high or unclear risk of bias (P<.001; with 29 contingency tables; ), respectively.

‎

Different Sample Sizes of Model

The pooled sensitivity, specificity, and AUC were 0.91 (95% CI 0.86‐0.94), 0.92 (95% CI 0.87‐0.95), and 0.97 (95% CI 0.95‐0.98) for sample size≥200 (P<.001; with 14 contingency tables) (), and 0.85 (95% CI 0.80‐0.88), 0.86 (95% CI 0.80‐0.90), and 0.91 (95% CI 0.89‐0.94) for sample size<200 (P<.001; with 20 contingency tables; ), respectively.

‎

Models With Different Research Designs (Multicenter Studies and Single-Center Studies)

The pooled sensitivity, specificity, and AUC were 0.84 (95% CI 0.77‐0.89), 0.87 (95% CI 0.81‐0.91), and 0.92 (95% CI 0.90‐0.94) for multicenter studies (P<.001; with 9 contingency tables; ), and 0.89 (95% CI 0.84‐0.92), 0.89 (95% CI 0.84‐0.93), and 0.95 (95% CI 0.93‐0.97) for single-center studies (P<.001; with 22 contingency tables; ), respectively.

‎

Heterogeneity Analysis and Meta-Regression Analysis

The Cochran Q test was used to indicate the presence of heterogeneity among subgroups (significance level P≤.05) []. The I² index was used to assess the extent of heterogeneity among studies [], revealing high sensitivity (I²=93.58%) and specificity (I²=91.38%; ). The Deek funnel plot asymmetry test, with P=.21, indicated no apparent publication bias (). Subgroup analyses were performed using the random-effects models to identify the potential sources of heterogeneity, particularly when I² exceeded 50% []. Results were as follows:

1. AI model for carotid plaques: Both ML models based on radiomics algorithms and DL models exhibited high sensitivity, with an I² of 90.20% and 93.70%, and high specificity, with an I² of 78.92% and 95.55%, suggesting high performance and significant heterogeneity ( [-]).
2. Medical imaging modalities: the sensitivity and specificity for PRs (sensitivity I²=82.28%; specificity I²=79.16%; [-]) and ultrasound (sensitivity I²=96.92%; specificity I²=94.98%; [-]). The sensitivity and specificity for MRI (sensitivity I²=71.57%; specificity I²=73.21%; [-]) and the sensitivity for CTA (I²=56.80%) displayed moderate heterogeneity ( [-]). The specificity of CTA (I²=83.79%) was high ( [-]). In the ultrasound modality, the sensitivity and specificity for determining the presence of plaques (sensitivity I²=96.78%; specificity I²=97.97%; [-]) and distinguishing the stability of plaques (sensitivity I²=97.01%; sensitivity I²=94.43%; [-]) were high.
3. Use of transfer learning: the specificity for models using transfer learning (specificity I²=74.85%; [-]) displayed moderate heterogeneity. The sensitivity for models using transfer learning (sensitivity I²=79.84%; [-]) and the sensitivity and specificity for the models without transfer learning (sensitivity I²=94.12%; specificity I²=87.35%; [-]) were high.
4. Carotid plaque type: all plaque types showed higher sensitivity and specificity; presence or absence of plaques (sensitivity I²=94.08%; specificity I²=97.60%; part A in [-]), stable or vulnerable plaques with (sensitivity I²=95.19%; specificity I²=91.29%; part B in [-]), and symptomatic or asymptomatic plaques (sensitivity I²=93.28%; specificity I²=84.67%; part C in [-]).
5. Both pure AI models and combined clinical features models did not exhibit high heterogeneity for AI models (sensitivity I²=62.97%; specificity I²=2.41%; part B in [ ,,,,,,]) and combined models (sensitivity I²=69.77%; specificity I²=40.08%) for combined models (part A in [ ,,,,,,]).
6. Different sets of datasets: both testing (sensitivity I²=94.23%; specificity I²=93.45%; part A in [-]) and external validation (specificity I²=84.42%; part B in [-]) were high heterogeneity, except the sensitivity for external validation (I²=66.67%; part B in [-]).
7. Different risk of bias studies: the sensitivity and specificity for high or unclear risk of bias studies (sensitivity I²=94.61%; specificity I²=92.59%; part B in [-]) and the specificity for low risk of bias studies (I²=87.10%) were high (part A in [-]). The sensitivity for low risk of bias studies (I²=62.20%) was moderate (part A in [-]).
8. Different sample sizes of model: The sensitivity and specificity for sample size ≥200 (sensitivity I²=97.91%; specificity I²=97.40%; part A in [-]) and the specificity for sample size <200 (I²=78.02%; part B in [-]) were high. The sensitivity for sample size <200 (I²=60.64%) was moderate (part B in [-]).
9. Models with different research designs: The sensitivity and specificity for multicenter studies (sensitivity I²=81.36%; specificity I²=80.24%; part A in [,,,-,-]) and single-center studies (sensitivity I²=95.07 %; specificity I²=90.63%) were high (part B in [,,,-,-]).

The meta-regression did not explore the factors contributing to heterogeneity (parts A-I in [-]). The results of all subgroups are depicted in Table S4 in . The Fagan nomogram was used to evaluate the diagnostic performance of ML models based on radiomics algorithms and DL models for carotid plaques. The results showed a P-post of 89% and 12% for the positive and negative tests, respectively ().

Sensitivity Analysis

Excluding the specific studies did not significantly change our research results (Table S7-S8 in ).

Quality Assessment

The quality of the 34 studies was evaluated using the QUADAS-AI tool (). The QUADAS-AI specifically evaluates bias risk and applicability concerns in AI studies. Here, we observed that most studies had significant bias or applicability concerns, particularly regarding the selection of patients and index test. In the “patient selection” domain, 20 studies were classified as either high-risk or indeterminate due to reliance on closed-access data or failure to present the rationale and breakdown of its training, validation, and test sets. Only 7 externally validated studies were classified as low-risk in the “index test” category, while others showed elevated risks due to a lack of validation. In the “reference standard” assessment, the reference standard of all studies could be used to classify the target condition correctly. For the “flow and timing” assessment, 10 studies showed indeterminate risks due to insufficient justification for the timing between index and reference tests. Additionally, 20 studies presented significant concerns regarding applicability in the “patient selection” domain, receiving unclear ratings. In the “index test” domain, 7 studies were rated as having low applicability, while all studies received low applicability ratings in the “Reference Standard” domain.

Discussion

Principal Findings

This study represents the first systematic evaluation of ML models based on radiomics and DL models for the characterization of extracranial carotid plaques. Both approaches demonstrated robust diagnostic performance, with high SROC values of 0.95 and 0.92, respectively, highlighting their promising potential for clinical application in plaque detection and risk stratification.

Initially, the SP and SROC AUC of DL models were improved compared to ML models based on radiomics (0.91 vs 0.83; 0.95 vs 0.92), while their sensitivity was similar to that of ML (0.88). Moreover, we observed that radiomics and DL models used to identify the presence of plaques and stable plaques had similar diagnostic capabilities (SROC 0.96, 95% CI 0.94‐0.97), and both were effective in identifying symptomatic plaques (SROC 0.90, 95% CI 0.87‐0.92). Notably, these differences may not be simply due to model performance, but could result from a combination of different clinical objectives (simple exclusion diagnosis or differentiation of specific cases), imaging variations, and model techniques. By using knowledge gained from previous tasks, transfer learning enhances model performance on new datasets and minimizes data requirements. It has been successfully applied in various areas of cardiovascular disease to boost the performance of models [,,]. In subgroup analyses, transfer learning significantly enhances model performance in data-limited scenarios and prevents overfitting. Large sample sizes can minimize sampling bias, decrease overfitting, and enhance the stability and reproducibility of the models. Moreover, we performed more detailed subgroup analyses based on the same imaging modality. Only the type of plaques in the ultrasound modality had sufficient data to perform statistical analysis and obtain summary diagnostic efficacy indicators. Results showed that ultrasound-based models have demonstrated excellent and similar performance in detecting the presence of plaques and assessing their stability. Considering the differences in equipment characteristics, patient demographics, and study design, these findings should be interpreted with caution. Nevertheless, these results provide valuable insights into the efficacy of radiomics algorithms and DL models in the diagnosis of carotid plaque.

Analysis of the Main Aspects

This meta-analysis demonstrates that radiomics-based models and DL models can diagnose extracranial carotid plaque, but the advantages of DL models in specificity and SROC should be interpreted with caution. A review of the included studies revealed that, among the 24 investigations using DL models, 20 primarily focused on plaque characterization (11 on the detection of plaques and 9 on plaque stability). Of these, 13 studies used ultrasound imaging to identify plaque-specific features such as echogenicity, morphology, and composition. In contrast, among the 10 studies using radiomics-based ML models, 6 were dedicated to identifying symptomatic plaques, predominantly using MRI (n=2) and CTA (n=3). The accuracy of symptomatic plaque identification was influenced not only by intrinsic imaging characteristics but also by clinical indicators, including plaque rupture, thrombus formation, and the occurrence of cerebral hypoperfusion. The tasks were more complex, and model training seemed to focus on reducing false negatives to lower the risk of adverse outcomes such as stroke. In addition, traditional ML algorithms may rely on manual preprocessing and struggle to capture other subtle differences (such as the presence of tiny thrombi or fibrous cap thickness), which may introduce variability and additional costs. In contrast, the DL models (particularly convolutional neural networks) do not rely on artificially designed features; instead, they can directly process raw medical images, automatically filter noise, and automatically extract more meaningful image features (eg, slight echo attenuation behind plaques, differences in vascular wall elasticity, etc) []. It can also analyze the preset artificial extraction features, conduct independent learning, and uncover potential rules, thereby addressing the aforementioned challenges [,]. It is worth noting that a mismatch in the number of studies may also affect the interpretation of the results. Therefore, these differences may not be simply due to model performance, but could also be caused by multiple factors, which need to be further investigated.

Besides, the “black box” nature of AI algorithms, particularly DL models, raises concerns about the transparency and reliability of decision-making. Of the 34 studies reviewed, only 2 used explainable DL models, achieving an accuracy of 98.2% [,]. The explainable AI (XAI) approach leverages visualization techniques, feature attribution analysis, and both global and local explanations to clarify how models derive predictions from input data. By enhancing transparency, XAI fosters greater trust among medical professionals, strengthens model reliability and accountability, and helps mitigate concerns related to opaque decision-making []. The integration of XAI in medicine not only represents a technological advancement but also ensures safe, efficient, and robust medical decision-making, which needs to be further investigated. To realize this potential, a clinically oriented XAI implementation framework needs to be developed. First, the reporting criteria for interpretable techniques (including clinical applicability evaluation and operational guidelines) should be standardized to lower the threshold for physician use. Second, the design of algorithms should be optimized through collaborative efforts of medical professionals and engineers to improve the specificity of feature attribution methods based on real clinical needs. Further clinical validation studies are needed to evaluate the practical utility of XAI across diverse diagnostic settings—such as varying regions, hospital levels, and clinician experience—and to determine its true value in supporting clinical decision-making beyond algorithmic performance []. Furthermore, incomplete disclosure of model development processes in reports, selective presentation of results by investigators, and heterogeneity in diagnostic standard implementation across practitioners with different levels of experience may decrease the reliability and generalizability of findings. Therefore, we recommend the formulation of standardized imaging protocols, reporting procedures, and quality control measures for carotid plaque assessment and advocate for the establishment of specialized AI reporting guidelines for cardiovascular diseases.

Advances in imaging technology have now largely met the diagnostic requirements of current clinical practice, and current guidelines place heavy reliance on imaging tests for carotid plaque assessment. Among the 34 included studies, 27 constructed diagnostic models based only on imaging data. However, this should not be interpreted as rendering other clinical parameters irrelevant. Multidimensional diagnostic models combined with clinical features have been shown to achieve good diagnostic performance in identifying various diseases, such as pancreatic ductal adenocarcinoma [], HCC recurrence after liver transplantation [], hemorrhagic brain metastases [], malignant BI-RADS 4 breast masses [], and others. In our study, the diagnostic performance of combined models did not slightly improve, which may be due to the small sample size or some features could not provide more diagnostic information (for example, Hu et al [] constructed a model relying only on indirect perivascular adipose tissue radiomic features and clinical features to identify symptomatic plaques, lacking direct imaging features). Considering this evidence, we strongly recommend that future research should aim to not only systematically incorporate laboratory tests, medical history, and other clinical parameters to develop multidimensional diagnostic models, but also to summarize the most meaningful features for specific types of plaques. This could address the limitations in current studies regarding single imaging modalities. This will also improve the precise classification of carotid plaques and personalized risk assessment.

This meta-analysis identified significant heterogeneity, while meta-regression and subgroup regression analysis did not identify the source, primarily attributable to the intrinsic challenges in regulating all potential confounding factors. Different imaging techniques can affect model performance based on the type of images used (static images vs dynamic videos), the equipment, and the operators. Guang et al [] used a contrast-enhanced ultrasound video-based DL model to evaluate the diagnostic efficacy of a new carotid network structure for assessing carotid plaques, whereas other ultrasound studies consistently used static images. The sequence of MRI scans also influences diagnostic outcomes. Zhang et al [] reported that a model incorporating a combination of T1-weighted, T2-weighted, dynamic contrast-enhanced, and postcontrast (POST) MRI sequences achieved a higher AUC for identifying high-risk carotid plaques compared to models using individual sequences or partial combinations. This enhanced performance is attributed to the complementary nature of these imaging sequences, each capturing distinct pathophysiological characteristics of the plaque, thereby improving diagnostic accuracy when used in combination. PRs have limited resolution, only detecting calcified components of carotid plaques and missing features such as lipid-rich necrotic cores or thin or ruptured fibrous caps. There are also notable differences in model architecture. Yoo et al [] found performance variations among different convolutional neural network architectures within the CACSNet framework on the same dataset. Gui et al [] compared multiple DL models (eg, 3D-DenseNet, 3D-SE-DenseNet) with 9 ML algorithms (including Decision Tree, Random Forest, SVM, etc) using identical datasets. They found that DL models generally performed better across key metrics like AUC and accuracy, with significant performance differences between and within the two model types. These suggest that scanning parameters, model architectures, image segmentation, and algorithms may explain the heterogeneity in the research results. However, the small number of studies limits our ability to perform comprehensive subgroup analyses, which need to be further investigated.

The use of AI has significantly promoted the diagnosis of carotid plaque; however, its application requires cautious evaluation. Only 9 studies were multicenter (most used external validation), with diagnostic performance lower than single-center studies. Most studies (n=29) had a high risk of bias due to a lack of open-source data and external validation and failure to present the rationale and breakdown of its sets, which led to overestimation of the research results and affected the reproducibility and generalizability of the findings. Similar issues have been noted in previous reports, highlighting a broader deficiency in rigorous research standards within the field [-]. Furthermore, the contingency tables mostly come from the testing sets. Although the testing set achieved the best diagnostic performance, it had higher data quality or similar data distribution to the training, or overfitting noise, resulting in inaccurate performance estimation, and strong regularization may also decrease its performance, ultimately undermining clinical confidence in these models.

This study has certain clinical significance. We conducted an in-depth literature review and methodological quality evaluation, presenting the most current and comprehensive systematic review of AI-based diagnostic approaches for assessing carotid plaque. The findings reveal that AI technology shows considerable potential for diagnosing carotid plaque, but the findings need to be further validated by conducting more rigorous external validation using large-scale, high-quality independent datasets.

Limitations

This study has several limitations. First, the heterogeneity in model architectures and validation methods across studies prevents definitive conclusions regarding the most effective AI approaches. Second, many studies lack multicenter external validation, leading to a high risk of bias. The model overfitting and clinical applicability need to be carefully evaluated. Third, meta-regression and subgroup analysis did not identify the sources of high heterogeneity that existed in most of the included studies. We hypothesize that this heterogeneity may be caused by scanning parameters, model architectures, image segmentation, and algorithms. However, the overly scattered distribution of subgroups due to the limited number of studies restricts more in-depth subgroup analyses. Finally, although the Deeks test did not show significant publication bias, the included studies may have intentionally unreported negative results and omitted potentially relevant non-English literature.

Future studies should use a more comprehensive analytical methodology based on the current model. Researchers should strictly follow regulatory norms and standardized operating procedures. Prospective and multicenter studies and additional external validation are warranted to enhance the robustness and generalizability of the existing models. In the future, researchers should perform independent systematic reviews on specific subtopics—such as imaging modalities, lesion types, or model architectures—to facilitate targeted evaluations of AI performance across distinct clinical scenarios. In addition, studies on imaging modalities such as CT and MRI are advocated to generate more data, conduct subgroup analyses, and clarify the optimal matching of modality, plaque type, and algorithm. Future efforts should focus on identifying more meaningful features and building and evaluating the diagnostic performance of multidimensional diagnostic models. In parallel, establishing clinically oriented, XAI frameworks will be essential for enhancing transparency.

Conclusions

Current findings indicate that radiomics algorithms and DL models can effectively diagnose extracranial carotid plaque. However, the irregularities in research design and the lack of multicenter studies and external validation limit the robustness of the present findings. Future research should aim to reduce bias risk and enhance the generalizability and clinical orientation of the models.

(Bloomberg) — The cooling of geopolitical tensions, a rally in big tech and solid economic data fueled gains in stocks, with the market remaining higher after an in-line inflation report. Short-dated bonds fell.

Equities rose around the world, with the S&P 500 up almost 1%. Tech megacaps rallied as comments by Nvidia Corp.’s chief Jensen Huang bolstered the artificial-intelligence trade. Small caps beat the US equity benchmark for a 14th straight session. JPMorgan Chase & Co. pared gains as President Donald Trump sued the lender and its head Jamie Dimon over alleged debanking.

Treasury two-year yields headed toward their highest since early December as strong economic data reinforced the argument for the Federal Reserve to keep rates on hold.

The US economy expanded in the third quarter by slightly more than initially reported, supported by stronger exports and smaller drag from inventories. Initial jobless claims steadied at 200,000 last week. And personal spending rose at a solid pace in November, underscoring consumer resilience.

“US consumers continue to underpin the economy,” said Lale Akoner at eToro. “Resilient spending lowers near-term recession risk and supports corporate revenues, particularly in consumer-facing sectors. However, steady demand also means interest rates are likely to stay higher for longer.”

Meantime, European Union lawmakers are expected to vote on ratifying the bloc’s trade deal with the US, restarting the process after Trump walked back his latest threat to impose tariffs on European allies that opposed his plans to annex Greenland. The island’s prime minister says he’s willing to go further in increasing defense, including agreeing on a permanent NATO mission.

“This episode once again highlights how headline-driven the market remains, and how quickly sentiment can flip when geopolitical risk is dialed back,” said Fawad Razaqzada at Forex.com.

The S&P 500 rose 0.8%. A gauge of the “Magnificent Seven” shares climbed 2.3%. The Russell 2000 index of small firms hit a fresh record. A key measure of stock volatility — the VIX — tumbled to around 15.

The yield on 10-year Treasuries was little changed at 4.25%. The dollar lost 0.3%. Oil sank as Ukrainian President Volodymyr Zelenskiy discussed plans for trilateral meetings with the US and Russia. Gold rose to all-time highs.

The latest data should reassure the Fed that the economy remains on a solid footing, despite a cooler labor market, said James McCann at Edward Jones.

“There looks to be little urgency to cut rates at next week’s meeting, and the central bank could stay on hold for longer should growth remain robust into 2026 and inflation continue to run at above target rates,” he added.

Inflation-adjusted gross domestic product increased at a revised 4.4% annualized rate, the fastest in two years. The Fed’s preferred measure of underlying inflation rose 0.2% in November from the prior month and 2.8% from a year earlier. The core personal consumption expenditure’s price index picked up slightly from October on an annual basis.

“This is likely going to keep the Fed on pause for a few a months, at least until we get a new Fed Chair who will likely push for renewed cuts,” said Sonu Varghese at Carson Group.

This set of new data reinforces the view that the US is experiencing stronger — not hotter — growth, according to Marco Casiraghi at Evercore.

“If macro conditions continue to evolve in this favorable manner, we think the Fed will keep rates on hold before delivering a cut in June – when the new Fed chair will take over,” and then cut two more times in the second half of 2026, he said.

Recent data support the Fed adopting a cautious approach to policy changes in the near term, according to Oscar Munoz and Gennadiy Goldberg at TD Securities.

“There is now a higher burden on the data to justify further easing,” they said.

The TD strategists expect policymakers to keep rates on hold at 3.50%-3.75%. While Fed Chair Jerome Powell is likely to sound noncommittal around near-term rate cuts, they expect him to remind markets that the median Fed official still looks for easing this year.

“Fundamentals are good and the Fed is likely to cut two or three times this year,” said Scott Helfstein at Global X. “That continues to set up a favorable backdrop even if the calm is occasionally disrupted by geopolitical volatility.”

Speculation that Europe could leverage US assets to retaliate against Trump’s bid for Greenland has been the chatter on trading floors and at the Davos gathering this week. Greenland’s pension fund is mulling whether it should continue investing in US stocks, in what its chief executive says would be a symbolic stand against the push to seize control of the island.

Trump vowed “big retaliation” if European countries sell US assets in response to his tariff threats related to Greenland, adding pressure on them to stick with an emerging deal over the future of the island.

“If they do, they do. But you know, if that would happen, there would be a big retaliation on our part,” Trump said during a Fox Business interview at the World Economic Forum in Davos. “And we have all the cards.”

Meantime, the dollar retained its supremacy in global trade despite the persistent uncertainty associated with Trump’s policies.

The greenback’s portion of international transactions rose to 50.5% in December, up from 46.8% a month earlier, according to the latest data compiled by global financial messaging service Swift, or the Society for Worldwide Interbank Financial Telecommunication. That’s the highest share since 2023 when the Belgium-headquartered consortium revised how it collects the transaction data.

There’s little sign of foreign investors shunning US equities and bonds amid tensions surrounding the Trump administration’s stance toward Greenland, according to JPMorgan strategists including Nikolaos Panigirtzoglou.

“Greenland is likely to stay in the headlines in the near term, and markets remain susceptible to fresh political or geopolitical developments,” said Ulrike Hoffmann-Burchardi at UBS Global Wealth Management. “But the latest stock rebound serves as a reminder that favorable fundamentals remain in the driver’s seat.”

She maintains the view that staying invested via a diversified portfolio remains the most effective way to manage market uncertainty.

Geopolitics only truly affect stock prices when they have direct effects on the factors that truly influence equity valuations, according to Steve Sosnick at Interactive Brokers.

“Unless one can draw a straight line between the global event and the revenues, earnings, or cash flows of a particular company or sector – the items that directly affect the value of a company – then geopolitics can be considered ‘background noise’ from a market viewpoint, no matter how newsworthy the events,” he said.

Companies from around the world remain focused on American markets, driven by the money they’re getting out of the US, Nasdaq Inc. Chief Executive Officer Adena Friedman said.

“The investment firms are obligated to find the best returns,” Friedman told Bloomberg Television in Davos. There’s been a $3 trillion increase in equity flows into the US from foreign investors in the past year, she said. “We just have to continue to drive those outsize returns within our economy to continue the flows coming in.”

BlackRock Inc. Chief Executive Officer Larry Fink said there is no bubble in artificial intelligence, emphasizing the volume of investment needed to develop the technology.

“I don’t think there is any uncertainty about AI,” Fink said in a Bloomberg Television interview on the sidelines of the World Economic Forum. “I sincerely believe there is no bubble in the AI space.”

With small-caps are outperforming large-caps yet again, there is a clear shift in leadership underway, noted Jonathan Krinsky at BTIG.

“While there will be pullbacks, we want to stick on the side of this new trend which is still in its early days, in our view,” he said.

Retail investors shelled out an “impressive” $12.9 billion on equities this week, according to JPMorgan strategist Arun Jain. On Tuesday, retail investors had responded to geopolitical developments by purchasing stocks, marking the third-largest single-day buying event in a year, he noted.

Meantime, trend-following funds are starting 2026 with fresh momentum, outperforming stocks and bonds after a year of false starts.

A Societe Generale index tracking major trend-following funds has climbed almost 4% in the opening weeks of the year, the second-strongest start on record in data going back to 2000. The performance follows a rally in metals, a weakening yen and resilient global equities, just the kind of sustained price moves these strategies need to deliver returns.

Despite the fact that growth stocks are negative year-to-date, momentum names are doing well, noted Louis Navellier at Navellier & Associates.

“The broadening of returns is a positive development, and the strength of smaller companies is a major vote of confidence in broad economic growth and reflects the expectation of lower interest rates, which is more meaningful for smaller companies,” he concluded.

Corporate Highlights:

Tesla Inc. will probably sell its Optimus robots to the public by the end of next year, according to Chief Executive Officer Elon Musk, who’s said the carmaker’s fortunes will be increasingly dependent on humanoid machines. SpaceX has lined up four banks to lead its initial public offering, according to people familiar with the matter, as Musk’s rocket and satellite firm moves forward with plans for the biggest-ever listing. Apple Inc. has expanded the job of hardware chief John Ternus to include design work, solidifying his status as a leading contender to eventually succeed Chief Executive Officer Tim Cook. Alphabet Inc.’s Google is rolling out a new option to personalize search results by tapping user data from the tech giant’s other applications, its latest bid to keep ahead of competition from the likes of OpenAI. Alibaba Group Holding Ltd. is preparing to list its chipmaking arm, tapping strong investor interest in the small circle of companies aspiring to compete with Nvidia Corp. in the hot AI accelerator business. Netflix Inc. co-Chief Executive Officer Ted Sarandos is planning to testify in February at a US Senate committee hearing looking into his company’s proposed $82.7 billion purchase of the streaming and studio operations of Warner Bros. Discovery Inc. Paramount Skydance Corp. again extended its tender offer for Warner Bros. Discovery Inc. shares and said it would ask investors to vote against a proposed sale to Netflix Inc. at a special meeting of Warner Bros. shareholders. Bank of America Corp. and Citigroup Inc. are exploring options they could offer up as an olive branch to satisfy President Trump’s demand to cap credit card interest rates at 10% for one year. US airlines are already announcing backup plans for passengers ahead of an expected winter storm this weekend that could be the biggest of the season and cause massive disruptions to air traffic nationwide. General Motors Co. plans to move production of its next-generation Buick Envision compact SUV, which is currently built in China, to a plant in Kansas in 2028, a sign of the pressure automakers are under to reshore output of vehicles sold in the US. Procter & Gamble Co.’s executives signaled sales are rebounding in the US and expressed confidence the company will meet its full-year guidance. General Electric Co.’s full-year outlook underwhelmed investors, a sign of high expectations on the jet-engine maker after a steep rise in the stock last year. Abbott Laboratories forecast a first-quarter profit that was lower than Wall Street expected and missed fourth-quarter sales estimates after its nutrition unit fell short of expectations, sending shares lower. Moderna Inc.’s chief executive officer said the company doesn’t plan to invest in new late-stage vaccine trials because of growing opposition to immunizations from US officials. Waymo will start offering its robotaxi service in Miami to the public Thursday, the first of around a dozen cities where the Alphabet Inc. company plans to launch this year. Freeport-McMoRan Inc. is making progress on a restart of its sprawling Indonesian copper mine, it said Thursday, after a deadly mudslide shuttered the operation that’s critical to global supply. Target Corp. is adding two retail veterans to its board as the beleaguered retailer seeks to reverse a sales slump under incoming Chief Executive Officer Michael Fiddelke. Lululemon Athletica Inc.’s “Get Low” leggings, derided for being see-through, are available for sale again online. Shoppers just have to make sure to read the disclaimers first. PayPal Holdings Inc. agreed to acquire Cymbio, a platform designed to help merchants sell products across AI chatbots. Terms weren’t disclosed. General Fusion Inc. has agreed to a merger with a blank-check company in a deal that’s expected to create one of the first publicly traded nuclear fusion technology developers. Some of the main moves in markets:

Stocks

The S&P 500 rose 0.8% as of 2:36 p.m. New York time The Nasdaq 100 rose 1% The Dow Jones Industrial Average rose 0.9% The MSCI World Index rose 0.9% Bloomberg Magnificent 7 Total Return Index rose 2.3% Philadelphia Stock Exchange Semiconductor Index rose 0.4% The Russell 2000 Index rose 1% KBW Bank Index rose 0.9% Currencies

The Bloomberg Dollar Spot Index fell 0.3% The euro rose 0.5% to $1.1743 The British pound rose 0.4% to $1.3489 The Japanese yen was little changed at 158.43 per dollar Cryptocurrencies

Bitcoin fell 0.9% to $89,413.96 Ether fell 2.8% to $2,944.25 Bonds

The yield on 10-year Treasuries was little changed at 4.25% Germany’s 10-year yield was little changed at 2.89% Britain’s 10-year yield advanced two basis points to 4.47% The yield on 2-year Treasuries advanced three basis points to 3.61% The yield on 30-year Treasuries declined two basis points to 4.85% Commodities

West Texas Intermediate crude fell 2.1% to $59.36 a barrel Spot gold rose 1.7% to $4,912.33 an ounce ©2026 Bloomberg L.P.

COPPA Rule

TAKE IT DOWN Act

Age verification

Giredestrant, a next-generation oral selective estrogen receptor degrader (SERD) and full antagonist, significantly improved invasive disease–free survival as adjuvant treatment for patients with estrogen receptor–positive, HER2-negative early breast cancer compared with standard-of-care endocrine therapy, according to data presented at the 2025 San Antonio Breast Cancer Symposium (SABCS).¹

Results from the global, randomized lidERA Breast Cancer trial position giredestrant as a potential new standard of care, marking the first phase III trial to demonstrate a benefit with an oral SERD in this setting.

The study met its primary endpoint of invasive disease–free survival, with a 30% reduction in the risk of invasive recurrence or death; this benefit appeared to be consistent across various subgroups. A favorable safety profile was reported, with a notably lower rate of treatment discontinuation compared with standard endocrine therapy.

“lidERA is a pivotal study…as we talk about these results in the adjuvant setting,” said Aditya Bardia, MD, MPH, FASCO, Program Director, Breast Medical Oncology, University of California at Los Angeles (UCLA) Jonsson Comprehensive Cancer Center.

As Dr. Bardia explained, estrogen receptor–positive breast cancer accounts for most breast cancers, with endocrine therapy being the mainstay of adjuvant management. Despite its efficacy, up to one-third of patients eventually experience recurrence, he stated.

Although advancements such as aromatase inhibitors in the early 2000s and, more recently, CDK4/6 inhibitors have improved outcomes, said Dr. Bardia, they have also introduced associated toxicities that lead many patients to early treatment discontinuation and increase their risk of recurrence. These limitations underscore the ongoing need for more effective and better tolerated adjuvant endocrine therapies.

Giredestrant is designed to induce full estrogen receptor antagonism and degradation, resulting in deep and sustained inhibition of estrogen receptor signaling across both ligand-dependent and ligand-independent pathways, which may offer a mechanistic advantage over aromatase inhibitors. Preclinical data and preliminary clinical trials (coopERA Breast Cancer, EMPRESS) have indicated that giredestrant exhibits increased potency and superior antiproliferative activity compared with other SERDs and standard endocrine therapies.

Study Design

The lidERA Breast Cancer study was a global, randomized phase III trial that enrolled 4,170 patients with stage I to III estrogen receptor–positive, HER2-negative early breast cancer. They were randomly assigned in a 1:1 ratio to receive either giredestrant (30 mg orally once daily, with concomitant luteinizing hormone-releasing hormone [LHRH] agonist therapy for pre- and perimenopausal women and for men) or standard-of-care endocrine therapy (tamoxifen or an aromatase inhibitor [ie, exemestane, letrozole, or anastrozole, with concomitant LHRH agonist therapy for pre- and perimenopausal women and for men]) for 5 years.

The primary endpoint was invasive disease–free survival, assessed in an intention-to-treat fashion. A prespecified efficacy interim analysis of invasive disease–free survival was performed after 336 events, at which time the first interim overall survival analysis also occurred per the hierarchical design.

The baseline demographics were found to be well balanced between the arms. The median age was 54.0 years, with 59.3% of patients being postmenopausal and approximately 40% premenopausal. About 50% of patients had stage II disease, 40% had stage III, and 10% had stage I. The majority of the study population had received chemotherapy prior to enrollment. In the standard-of-care endocrine therapy arm, 84% of patients were treated with an aromatase inhibitor, and 16% received tamoxifen.

At the data cutoff of August 8, 2025, with a median follow-up of 32.3 months, treatment had been discontinued in 347 patients receiving giredestrant and 520 receiving standard-of-care endocrine therapy.

Improvement in Invasive Disease-Free Survival

As Dr. Bardia reported, the study met its primary endpoint, demonstrating a statistically significant and clinically meaningful improvement in invasive disease–free survival with giredestrant vs standard-of-care endocrine therapy.

Giredestrant reduced the risk of invasive recurrence or death by 30%, corresponding to a hazard ratio (HR) of 0.70 (P = .0014). The 3-year invasive disease–free survival rate was 92.4% with giredestrant vs 89.6% with standard-of-care endocrine therapy, representing an absolute benefit of approximately 3%.

“The Kaplan-Meier curves for invasive disease–free survival separated early and remained separated over time,” said Dr. Bardia. “This superiority of giredestrant was consistent across all predefined subgroups, including region, menopausal status, risk, prior chemotherapy, and tumor stage [HR = 0.58 for stage II; HR = 0.74 for stage III].”

In addition, giredestrant demonstrated superiority in distant recurrence–free interval compared with standard-of-care endocrine therapy, representing a 31% reduction in the risk of developing distant metastatic disease (HR = 0.69), although absolute risks appeared to be low at this time point (96.1% vs 94.2%).

“Giredestrant also showed a trend for improvement in overall survival compared with standard endocrine therapy [HR = 0.79], although these results are immature at this interim analysis, and further follow-up is ongoing,” said Dr. Bardia.

The safety profile of giredestrant was found to be consistent with its known profile. The overall incidences of adverse events and grade 3 to 4 adverse events appeared comparable between the arms.

Of note, the discontinuation rate because of adverse events was lower with giredestrant (5.3%) compared with standard-of-care endocrine therapy (8.2%). This lower discontinuation rate was observed despite high compliance in both arms, with a mean dose intensity of over 99%. However, dose interruptions were more frequent with giredestrant (12.7% vs 6.6%).

Common adverse events in both arms included arthralgia (48.0% vs 47.1%), hot flush (27.4% vs 28.8%), and headache (15.3% vs 13.2%). Although arthralgias were common, said Dr. Bardia, those leading to discontinuation were lower with giredestrant (1.6% vs 3.7%).

Giredestrant was found to be associated with a higher incidence of bradycardia, occurring in approximately 10% of patients, but predominantly as asymptomatic grade 1 events. Grade 3 to 4 venous thromboembolism was more frequent in the standard-of-care endocrine therapy arm, likely reflecting tamoxifen use.

“These results support giredestrant as a potential new standard for patients with ER [estrogen receptor]-positive, HER2-negative early breast cancer,” Dr. Bardia concluded. 

DISCLOSURE: Dr. Bardia reported financial relationships with Pfizer, Novartis, Merck, Genentech, AstraZeneca/Daiichi Sankyo, Alyssum, Menarini, Gilead, Eli Lilly, and OnKure.

REFERENCE

1. Bardia A, Schmid P, Martin M, et al: Giredestrant vs standard-of-care endocrine therapy as adjuvant treatment for patients with estrogen receptor-positive, HER2-negative early breast cancer: Results from the global phase III lidERA Breast Cancer trial. SABCS 2025. Abstract GS1-10. Presented December 10, 2025.

 

EXPERT POINT OF VIEW

Invited discussant Lisa A. Carey, MD, ScM, FASCO, the L. Richardson and Marilyn Jacobs Preyer Distinguished Professor for Breast Cancer Research and Deputy Director of Clinical Sciences at the University of North Carolina (UNC) Lineberger Comprehensive Cancer Center, called the findings of the lidERA Breast Cancer trial a “pivotal moment” and the “first positive data for an oral SERD [selective estrogen receptor degrader] in the adjuvant setting.”

Dr. Carey began by tracing the history of estrogen receptor targeting in hormone receptor–positive disease. She highlighted that although advancements have been made in the metastatic setting with oral SERDs, the early breast cancer setting saw “not much in the last 20 years related to estrogen receptor targeting until now,” with the success of giredestrant.

Dr. Carey detailed key aspects of the lidERA Breast Cancer trial design, noting the inclusion of patients with high-risk stage I disease, the requirement for ovarian function suppression with giredestrant in premenopausal patients, and the allowance of short-term CDK4/6 inhibition before but not during the trial. She highlighted the trial’s predominantly high-risk patient population, with 70% falling into the high clinical risk category, and a significant proportion having node-positive disease. She pointed out that in the high-risk population treated in lidERA Breast Cancer, modern therapy for most would include a CDK4/6 inhibitor, which was not included in the study’s design.

Regarding the primary efficacy data, Dr. Carey affirmed that the invasive disease–free survival endpoint was met, with giredestrant reducing the risk of recurrence or death by 30% (hazard ratio [HR] = 0.70). Despite it being early (median follow-up = 32.3 months), she found the invasive disease–free survival and distant recurrence–free interval data “quite reassuring,” noting the early and sustained separation of the Kaplan-Meier curves. She estimated an absolute invasive disease–free survival difference of 2.8% at just under 3 years, which, while small, is “likely to grow with additional follow-up.”

Dr. Carey also praised the tolerability of giredestrant, noting a low rate of discontinuation (5.3% vs 8.2% for standard endocrine therapy), including fewer discontinuations because of musculoskeletal symptoms. She addressed the class effect of bradycardia, observing it was a “pretty minor issue” in lidERA Breast Cancer, with predominantly asymptomatic grade 1 events.

Crucially, Dr. Carey contextualized the findings against other adjuvant advances in hormone receptor–positive breast cancer, including those from the ATAC, monarchE, and NATALEE trials. She observed “similar differences, both proportional and absolute” (2%–3% absolute benefit and HRs of 0.7–0.8) at comparable follow-up times, suggesting the efficacy of giredestrant is generally consistent with other improvements seen in this setting.

Dr. Carey also discussed the potential implications for clinical practice, particularly regarding CDK4/6 inhibitors, which were not concurrently addressed in the trial. She proposed that giredestrant will be the favored endocrine therapy for patients not receiving a CDK4/6 inhibitor. For those in whom a CDK4/6 inhibitor is appropriate, she suggested an aromatase inhibitor plus CDK4/6 inhibition for the initial 2 years, followed by a potential switch to giredestrant, though she acknowledged this specific sequencing was not tested.

The cost implications of giredestrant, according to Dr. Carey, also warrant consideration. She noted that oral SERDs in the metastatic setting typically exceed $20,000 per month, emphasizing the “large potential impact on the national health-care systems” if giredestrant is priced similarly.

Dr. Carey concluded with a strong call for predictive biomarkers to tailor therapy beyond the estrogen receptor, as giredestrant has “very broad clinical implications,” and noted that models are needed to guide decision-making. 

DISCLOSURE: Dr. Carey reported no personal financial interests with any commercial entity.

First data protection regulatory report

In January 2026, and in the context of recent technological developments, industry announcements and market attention, the UK Information Commissioner’s Office (ICO) published a Tech Futures report on agentic AI. In doing so, it became the first data protection regulator to tackle the topic. The report does not constitute formal ICO guidance but does provide important insight to the regulator’s view on data protection implications, risks and how organisations may be able exploit the opportunities the technology offers. The report delivers on the action set out in the ICO’s AI and biometrics strategy to engage with industry to assess the data protection implications of agentic AI. It is intended to support the ICO’s ambition to encourage the responsible development and use of agentic AI.

This blog takes an initial look at the key messages that can be drawn from the ICO’s report and what organisations can learn at this stage of the regulator’s policy thinking.

The ICO’s report is to be welcomed in its assessment of evidence related to advancement, benefits and risk proliferation. Its nuanced approach provides a range of options as to how the technology may develop and be deployed. The report can therefore play a useful role in assisting organisations with initial risk assessments and planning for deployment.

The ICO has gathered significant evidence to inform the report and has transparently set out the methodology used and resources that can be considered alongside the report. 

Definitions

Terms such as agents and agentic AI are often used interchangeably and the ICO helpfully provides definitions to enable a common language for our discussions about implications of agentic AI in the data protection community.

The ICO explains that an agent is “software or a system that can carry out processes or tasks with varying levels of sophistication and automation.” It then explains that “when large language models (LLMs) or foundation models are integrated (‘scaffolded’) with other tools, including databases, memory, computer operating systems and ways of interacting with the world, they create what industry is agentic AI.” 

The ICO also notes that because agentic AI systems build on LLMs, some of the negative characteristic features of LLMs (such as hallucinations and bias) may be present. 

Agentic AI can take different forms, perhaps as a standalone agent or, when several agents are, combined, as a ‘multi-agent system’.

Likely evolution of agentic AI, capabilities and use cases

Four scenarios for the evolution of agentic AI are set out and explained in the ICO report:

1. Scarce, simple agents (low adoption, low agentic capability)
2. Just good enough to be everywhere (high adoption, low agentic capability)
3. Agents in waiting (low adoption, high agentic capability)
4. Ubiquitous agents (high adoption, high agentic capability)

It is important to understand the new capabilities that agentic AI may offer. The report highlights perception (i.e. working with a wide range of inputs), planning or reasoning-like actions (e.g. generating plans, dividing tasks, error checking), action (e.g. accessing tools, interacting with people or AI agents, running code), and learning and memory (i.e. adaptive decision making, correcting errors in future plans, learning preferences and from feedback) as capabilities likely to be demonstrated by agentic AI systems to some extent. The report also indicates that capabilities are being assessed by researchers based on autonomy, efficacy, goal complexity, generality and under-specification. The ICO’s report therefore focuses on how agents can autonomously pursue goals, adapt to new situations and contexts, and an exhibit some reasoning-like capacities.

Potential use cases covered in the report include research, coding, and planning, organising and executing transactions.   Use in agentic commerce, workplace applications, government services, automated cybersecurity applications, integrated personal assistants and the medical sector are also covered in the report, indicating a wide range of potential deployments. 

It is also important to note that the ICO’s assessment of the stakeholder evidence in the report indicates that the rate of improvement in LLM capabilities may slow or even stop in the short to medium term. The graph of agentic technology evolution may not be a linear one.

Looking further ahead, amongst other things, the ICO sees evidence that the following technical developments could emerge – truly multimodal agents, increasing agent autonomy and agent-to-agent communication and agentic AI embedded into a wider range of software and devices.   

Key data protection implications

Accountability and governance

A key message from the ICO is on accountability- organisations must continue to take responsibility for AI in the context of data protection:

“AI agency does not mean the removal of human, and therefore organisational, responsibility for data processing. Organisations must be clear on the expectations that still apply under data protection legislation.”

The ICO notes that currently, and for the foreseeable future, organisations can control factors such as the actions the agent is authorised to take and the information the agent can access. An important learning point for organisations planning agentic AI governance is to prioritise the effective risk assessment of these elements and ensure that relevant controls are in place.

On governance, the ICO highlights the importance of flexible and adaptable governance to move with changes in how agentic AI systems may operate. Whilst not a formal endorsement, the ICO highlights the relevance of the Safer Agentic AI Foundations, from the Agentic AI Safety Community of Practice. The ICO also indicates that organisations may need a separate, standalone monitoring system.

The report’s section on accountability is relatively short at half a page, and it will be helpful for organisations if greater coverage is given to this topic in future ICO publications. Such further output could include information on how existing AI governance models will need to evolve, including existing AI standards and risk management frameworks such as those in use from NIST and ISO. It will also be relevant to address how to scale governance, and the balance between centralisation and decentralisation. Organisations will also need to address the risks of shadow AI, which could proliferate further with use of unauthorised AI agents, and how agents are managed and ultimately removed from use when no longer needed.

Novel risks

Data protection issues that may arise in the context of AI more generally, and particularly generative AI, can also be seen (perhaps even to a greater extent) in the context of agentic AI. Novel agentic AI data protection risks are also highlighted in the report, including:

• issues in relation to determining controller and processor responsibilities through the agentic AI supply chain;
• rapid automation of increasingly complex tasks resulting in a larger amount of automated decision-making;
• purposes for agentic processing of personal information being set too broadly so as to allow for open-ended tasks and general-purpose agents;
• agentic AI systems processing personal information beyond that which is necessary to achieve instructions or aims;
• potential unintended use or inference of special category data;
• increased complexity impacting transparency and the ease with which people can exercise their information rights;
• new threats to cyber security resulting from, for example, the connected and autonomous nature of agentic AI;
• and the concentration of personal information to facilitate personal assistant agents.

The section of the report on automated decision making (ADM) sets out some general impacts that must be considered but is otherwise surprisingly short given the challenges that agentic AI may pose in this area. This appears to be because the ICO will set out its thinking on ADM and AI in more detail in the forthcoming code of practice.    

It will be important for the ICO to explore issues of human intervention in more detail in future publications, including the role of alternatives to human in the loop solutions, such as human on the loop.

The Tech Dispatch Report of the European Data Protection Supervisor is also helpful reading in this area.

Role of agentic AI in automating data protection compliance

Lastly, the report sets out how agentic AI could pose challenges for DPOs in maintaining oversight, but also how agentic AI could assist the process of data protection compliance itself. The report sets out the idea of ‘DPO agents’ i.e. systems that are integrated into data protection teams to scale and augment the role of human staff. The report also notes how privacy and personal information management agents could help people manage their own privacy settings and controls. 

The ICO also calls for innovation in methods for the practical evaluation of the compliance of agentic AI systems with data protection legislation.

Next steps

The ICO has committed to publish a new statutory code of practice on AI and data protection in 2026 and we can expect a consultation process in the coming months.   The code is expected to have a specific focus on automated decision making, which will be particularly relevant to the implementation of agentic AI.

We can also expect collaboration internationally with other DP regulators and with UK regulators via the Digital Regulation Cooperation Forum , workshops with stakeholders, and advice via the ICO’s innovation service and regulatory sandbox.