Category: 3. Business

Quincey to Continue as Executive Chairman of the Board

ATLANTA–(BUSINESS WIRE)–
The Coca-Cola Company today announced that its board of directors has elected Executive Vice President and Chief Operating Officer Henrique Braun as CEO, effective March 31, 2026. Braun will succeed James Quincey, who will transition to Executive Chairman after serving as CEO for nine years.

The board also plans to nominate Braun, 57, to stand for election as a director at the company’s 2026 Annual Meeting of Shareowners.

Leadership transition

Quincey, 60, will step down as CEO after a highly successful tenure. He has led the transformation of the business as a total beverage company, driven by a focus on staying closely connected to consumers. Under his leadership, the company has added more than 10 additional billion-dollar brands.

Quincey has reshaped the company’s strategy and operating model to create a more agile, networked company, including a focus on digital transformation and modernized marketing. He also led the company through the COVID-19 pandemic.

As CEO, Braun will focus on opportunities to build on this strong foundation. His priorities include seeking the best growth opportunities worldwide; driving the company to get even closer to consumer needs; and leveraging technology as an enabler of business performance and growth.

“James Quincey is a transformative leader,” said David Weinberg, Coca-Cola’s lead independent director. “James set and executed a strategy that has built Coca-Cola’s status as a global leader. James will continue to be very active in the business through his role as Executive Chairman. We are confident that Henrique Braun will build on the company’s existing strengths to unlock more growth opportunities and increase the power of the incredible Coca-Cola system.”

About Henrique Braun

Braun has served as EVP and COO since Jan. 1, 2025, overseeing all the company’s operating units worldwide. He has served as EVP since 2024. From 2023 to 2024, Braun served as Senior Vice President and President, International Development, overseeing seven of the company’s nine operating units.

Prior to that, Braun served as President of the Latin America operating unit from 2020 to 2022 and as President of the Brazil business unit from 2016 to 2020. From 2013 to 2016, Braun was President for Greater China & South Korea.

Braun joined Coca‑Cola in 1996 in Atlanta and progressed through roles of increasing responsibilities in North America, Europe, Latin America and Asia. Those positions included supply chain, new business development, marketing, innovation, general management and bottling operations.

He holds a bachelor’s degree in agricultural engineering from the University Federal of Rio de Janeiro, a Master of Science degree from Michigan State University and an MBA from Georgia State University. Braun is an American citizen who was born in California and raised in Brazil.

“I’m honored to take on this new role and have tremendous appreciation for everything James has done to lead the company,” Braun said. “I will focus on continuing the momentum we’ve built with our system. We’ll work to unlock future growth in partnership with our bottlers. I’m excited about the future of our business and see huge opportunities in a fast-changing global market.”

About James Quincey

Quincey became CEO in 2017 and Chairman of the board in 2019. He joined the company in 1996 and has held leadership roles around the world.

Before becoming CEO, Quincey served as COO from 2015 to 2017 and as President from 2015 to 2018. From 2013 to 2015, he was President of the company’s Europe Group. Under his leadership, the group expanded its brand portfolio and improved market share. Quincey also played a key role in the creation of Coca‑Cola Europacific Partners, one of the largest independent Coca-Cola bottlers in the world. Quincey served as President of the Northwest Europe and Nordics business unit from 2008 to 2012.

Quincey joined the company in Atlanta in 1996 as director of learning strategy for the Latin America Group. He went on to serve in a series of operational roles in Latin America, eventually leading to his appointment as President of the South Latin division in 2003.

He was President of the company’s Mexico division from 2005 to 2008. Prior to joining Coca-Cola, Quincey was a partner in strategy consulting at The Kalchas Group, a spinoff of Bain & Company and McKinsey.

Quincey is a director of Pfizer Inc. and a board member of The Consumer Goods Forum. He is a founding member of the New York Stock Exchange Board Advisory Council.

Quincey received a bachelor’s degree in electronic engineering from the University of Liverpool. He is a native of Britain.

“I’m stepping down as CEO after a 30-year career with the company, and I have an appreciation of what a privilege it has been to serve this great and enduring business,” Quincey said. “Henrique is a trusted and highly experienced business partner, and he’s the right leader to steer the company and the Coca-Cola system for future growth and success.”

Weinberg said the company is looking forward to a seamless transition in management.

“On behalf of the board, I thank James for his outstanding leadership,” Weinberg said. “James has done what a strong CEO should do – he has focused on the future and developing and empowering the next set of leaders who will take Coke forward. Henrique has shown that he is the right leader for the future of Coca-Cola.”

About The Coca-Cola Company

The Coca-Cola Company (NYSE: KO) is a total beverage company with products sold in more than 200 countries and territories. Our company’s purpose is to refresh the world and make a difference. We sell multiple billion-dollar brands across several beverage categories worldwide. Our portfolio of sparkling soft drink brands includes Coca-Cola, Sprite and Fanta. Our water, sports, coffee and tea brands include Dasani, smartwater, vitaminwater, Topo Chico, BODYARMOR, Powerade, Costa, Georgia, Fuze Tea, Gold Peak and Ayataka. Our juice, value-added dairy and plant-based beverage brands include Minute Maid, Simply, innocent, Del Valle, fairlife and AdeS. We’re constantly transforming our portfolio, from reducing sugar in our drinks to bringing innovative new products to market. We seek to positively impact people’s lives, communities and the planet through water replenishment, packaging recycling, sustainable sourcing practices and carbon emissions reductions across our value chain. Together with our bottling partners, we employ more than 700,000 people, helping bring economic opportunity to local communities worldwide. Learn more at www.coca-colacompany.com and follow us on Instagram, Facebook and LinkedIn.

Forward-Looking Statements

This press release may contain statements, estimates or projections that constitute “forward-looking statements” as defined under U.S. federal securities laws. Generally, the words “believe,” “opportunity,” “ahead,” “expect,” “intend,” “estimate,” “anticipate,” “project,” “will” and similar expressions identify forward-looking statements, which generally are not historical in nature. Forward-looking statements are subject to certain risks and uncertainties that could cause The Coca-Cola Company’s actual results to differ materially from its historical experience and our present expectations or projections. These risks include, but are not limited to, unfavorable economic and geopolitical conditions, including the direct or indirect negative impacts of the conflict between Russia and Ukraine and conflicts in the Middle East; increased competition; an inability to be successful in our innovation activities; changes in the retail landscape or the loss of key retail or foodservice customers; an inability to expand our business in emerging and developing markets; an inability to successfully manage the potential negative consequences of our productivity initiatives; an inability to attract or retain specialized or top talent with perspectives, experiences and backgrounds that reflect the broad range of consumers and markets we serve around the world; disruption of our supply chain, including increased commodity, raw material, packaging, energy, transportation and other input costs; an inability to successfully integrate and manage our acquired businesses, brands or bottling operations or an inability to realize a significant portion of the anticipated benefits of our joint ventures or strategic relationships; failure by our third-party service providers and business partners to satisfactorily fulfill their commitments and responsibilities; an inability to renew collective bargaining agreements on satisfactory terms, or we or our bottling partners experience strikes, work stoppages, labor shortages or labor unrest; obesity and other health-related concerns; evolving consumer product and shopping preferences; product safety and quality concerns; perceived negative health consequences of processing and of certain ingredients, such as non-nutritive sweeteners, color additives and biotechnology-derived substances, and of other substances present in our beverage products or packaging materials; failure to digitalize the Coca-Cola system; damage to our brand image, corporate reputation and social license to operate from negative publicity, whether or not warranted, concerning product safety or quality, workplace and human rights, obesity or other issues; an inability to successfully manage new product launches; an inability to maintain good relationships with our bottling partners; deterioration in our bottling partners’ financial condition; an inability to successfully manage our refranchising activities; increases in income tax rates, changes in income tax laws or the unfavorable resolution of tax matters, including the outcome of our ongoing tax dispute or any related disputes with the U.S. Internal Revenue Service (“IRS”); the possibility that the assumptions used to calculate our estimated aggregate incremental tax and interest liability related to the potential unfavorable outcome of the ongoing tax dispute with the IRS could significantly change; increased or new indirect taxes; changes in laws and regulations relating to beverage containers and packaging; significant additional labeling or warning requirements or limitations on the marketing or sale of our products; litigation or legal proceedings; conducting business in markets with high-risk legal compliance environments; failure to adequately protect, or disputes relating to, trademarks, formulas and other intellectual property rights; changes in, or failure to comply with, the laws and regulations applicable to our products or our business operations; fluctuations in foreign currency exchange rates; interest rate increases; an inability to achieve our overall long-term growth objectives; default by or failure of one or more of our counterparty financial institutions; impairment charges; an inability to protect our information systems against service interruption, misappropriation of data or cybersecurity incidents; failure to comply with privacy and data protection laws; evolving sustainability regulatory requirements and expectations; increasing concerns about the environmental impact of plastic bottles and other packaging materials; water scarcity and poor quality; increased demand for food products, decreased agricultural productivity and increased regulation of ingredient sourcing due diligence; climate change and legal or regulatory responses thereto; adverse weather conditions; and other risks discussed in our filings with the Securities and Exchange Commission (“SEC”), including our Annual Report on Form 10-K for the year ended December 31, 2024, and subsequently filed Quarterly Report on Form 10-Q, which are available from the SEC. You should not place undue reliance on forward-looking statements, which speak only as of the date they are made. We undertake no obligation to publicly update or revise any forward-looking statements.

Investors and Analysts: Robin Halpern, koinvestorrelations@coca-cola.com

Media: Scott Leith, sleith@coca-cola.com

Source: The Coca-Cola Company

Released December 10, 2025

Emerging technology research dominates the Energy Department’s coming scientific agenda, with agency leadership moving in sync with the Trump administration’s larger goal of ensuring the U.S. wins the global race to artificial intelligence dominance.

Darío Gil, Energy’s undersecretary for science, testified before the House Science, Space and Technology Committee on the agency’s technological priorities — most notably the new Genesis Mission unveiled by President Doanld Trump last month. 

“This mission, which we envision as our generation’s Manhattan or Apollo scale project, will multiply the return on taxpayer investment and solidify America’s global technological and strategic leadership,” Gil said. 

In addition to clarifying the Genesis Mission’s role as an “integrated discovery platform” that will marry national laboratories, academia and industry partners to deliver new AI applications, Gil unveiled the first investment installment of $320 million for the American Science Cloud and the Transformational Model Consortia. 

These two initiatives, which were both created from the One Big Beautiful Bill Act, will further help disperse funding to Energy’s national laboratory network to begin curating data sets to train new scientific AI models.  

“At the core, we’re doing two things: we’re building a platform, and we are constructing a portfolio of scientific and engineering challenges for energy, national security and discovery science,” Gil said. 

Although Genesis is primarily focused on developing new and experimental AI applications and running them on the nation’s network of high-performance supercomputers, Gil also shared other emerging tech sectors that Energy is prioritizing and aiming to incorporate into the Genesis infrastructure.

Quantum computing is among the top fields to be eventually incorporated into the AI and supercomputing infrastructure Energy is scaling. Gil, who previously oversaw IBM’s scientific research portfolio that included multiple quantum information science and technology assets, testified that the convergence of AI and quantum computing will serve as “a scientific and technological revolution.”

This perspective will provide an opportunity to merge more advanced AI models with relatively immature quantum computing systems.

“We recognize that AI and quantum are not just tools, but foundational elements of new supercomputing platforms,” Gil said. “AI and quantum computing offer a fundamentally new paradigm for understanding the natural world. They are the new scientific instruments of our time. And just like telescopes and microscopes transform how we see the very large, the very far and the very small, AI and quantum supercomputers are going to transform how we make sense of the very complex.”

Gil also vocalized support for the National Quantum Initiative Act Reauthorization, a landmark appropriations bill specifically for quantum information sciences that has stalled in Congress since its lapse in 2023. 

Beyond quantum computing and AI, Gil said accelerating scalability of nuclear and fusion development is another top policy priority for Energy. He added that following Energy’s release of a formal roadmap to developing commercial fusion power in October, the department is working with private sector and academic stakeholders on a holistic look at what the U.S. needs to do to scale fusion power plants in the early 2030s.

Emerging technology research dominates the Energy Department’s coming scientific agenda, with agency leadership moving in sync with the Trump administration’s larger goal of ensuring the U.S. wins the global race to artificial intelligence dominance.

Darío Gil, Energy’s undersecretary for science, testified before the House Science, Space and Technology Committee on the agency’s technological priorities — most notably the new Genesis Mission unveiled by President Doanld Trump last month. 

“This mission, which we envision as our generation’s Manhattan or Apollo scale project, will multiply the return on taxpayer investment and solidify America’s global technological and strategic leadership,” Gil said. 

In addition to clarifying the Genesis Mission’s role as an “integrated discovery platform” that will marry national laboratories, academia and industry partners to deliver new AI applications, Gil unveiled the first investment installment of $320 million for the American Science Cloud and the Transformational Model Consortia. 

These two initiatives, which were both created from the One Big Beautiful Bill Act, will further help disperse funding to Energy’s national laboratory network to begin curating data sets to train new scientific AI models.  

“At the core, we’re doing two things: we’re building a platform, and we are constructing a portfolio of scientific and engineering challenges for energy, national security and discovery science,” Gil said. 

Although Genesis is primarily focused on developing new and experimental AI applications and running them on the nation’s network of high-performance supercomputers, Gil also shared other emerging tech sectors that Energy is prioritizing and aiming to incorporate into the Genesis infrastructure.

Quantum computing is among the top fields to be eventually incorporated into the AI and supercomputing infrastructure Energy is scaling. Gil, who previously oversaw IBM’s scientific research portfolio that included multiple quantum information science and technology assets, testified that the convergence of AI and quantum computing will serve as “a scientific and technological revolution.”

This perspective will provide an opportunity to merge more advanced AI models with relatively immature quantum computing systems.

“We recognize that AI and quantum are not just tools, but foundational elements of new supercomputing platforms,” Gil said. “AI and quantum computing offer a fundamentally new paradigm for understanding the natural world. They are the new scientific instruments of our time. And just like telescopes and microscopes transform how we see the very large, the very far and the very small, AI and quantum supercomputers are going to transform how we make sense of the very complex.”

Gil also vocalized support for the National Quantum Initiative Act Reauthorization, a landmark appropriations bill specifically for quantum information sciences that has stalled in Congress since its lapse in 2023. 

Beyond quantum computing and AI, Gil said accelerating scalability of nuclear and fusion development is another top policy priority for Energy. He added that following Energy’s release of a formal roadmap to developing commercial fusion power in October, the department is working with private sector and academic stakeholders on a holistic look at what the U.S. needs to do to scale fusion power plants in the early 2030s.

Introduction

Health care systems are facing growing pressures due to an aging population, multiple long-term conditions, and rising levels of frailty []. To address these challenges, innovative solutions that enhance productivity are essential [-]. In the United Kingdom, a paradigm shift is being promoted [,], emphasizing “predict and prevent” care, transitioning from “analog to digital” and delivering care closer to home [-,]. One potential solution is hospital-at-home (HaH) services, sometimes called virtual wards [], which deliver hospital-level care in a patient’s place of residence. These services aim to avoid hospital admissions or facilitate early discharge, demonstrably improving patient satisfaction, improving outcomes, and reducing admissions to hospital and residential care [-]. One area of use is for patients with frailty, who are at increased risk of functional decline and mortality [,]. The dynamic nature of frailty [,] underscores the need for timely detection, personalized care, and home-based interventions to support recovery.

In the United Kingdom, national guidance advocates the integration of digital technology to optimize service delivery []. National priorities have promoted significant investment in a ”tilt towards technology“ [], with the UK public sector spending £26 billion annually on digital technology [,]. The revolution of Healthcare 4.0 [-] highlights the transformative potential of advanced technologies in a digitally connected world, especially in home-based care [-]. These technologies can empower patient self-monitoring; enhance clinical decision-making; and optimize resources, costs, and safety [].

Reviews exploring technology in HaH care categorize these as low-intensity (eg, telephone or teleconferencing), high-intensity (eg, apps or wearables) [], manual, or automated remote monitoring []. Evidence on how to fully harness these technologies to optimize clinical decision-making, enhance risk prediction, and personalize care is limited.

An emerging approach in health care is digital twin (DT) technology, which creates a dynamic virtual representation of a patient or health care system leveraging real-time data. DT aims to closely mirror the real-world counterpart, analyzing its parameters using sophisticated methods like machine learning (ML) for dynamic simulation and predictive capabilities []. These methods can aid timely decision-making. With a patient DT, clinicians can visualize current states and forecast future scenarios, such as deterioration, early risk identification, timely intervention, optimization of care, and resource allocation. This supports proactive rather than reactive health care. To develop a DT, collaboration is required across disciplines, including engineering, medicine, computing, and data science, to understand the required data and enabling technologies within the 5 DT architectural layers: sensing (capturing information), communication, storage, analytics, and visualization [-].

The potential of DT in health care is increasingly recognized—leveraging artificial intelligence (AI), Internet of Things (IoT), big data, and predictive analytics—to anticipate health risks, support early disease detection, and enhance operational efficiency []. DTs can simulate disease progression, enabling proactive, personalized treatment and outcome prediction [,]. Examples include electrocardiogram (ECG)-based heart rhythm classifiers to detect heart problems [], drug interaction modeling to predict medication responses [], and dynamic glucose monitoring for medication adjustment in diabetes [,]. DTs have shown promise in personalized medicine, hospital management, and surgery [,,].

However, challenges remain, with existing literature presenting various definitions of DT architectures [,,,]. Although this gives us valuable insight, the lack of clarity and consistency in frameworks for DT health care applications can make it difficult for implementation and evaluation []. This review aimed to address the knowledge gap linking DT architecture with HaH care for patients with frailty, considering the complexity and acuity of this patient cohort. To our knowledge, no literature currently addresses the application of DTs in this context.

This review aimed to identify existing evidence supporting the development of DTs for HaH care for patients with frailty, addressing the following research questions:

1. What DT-enabling tools have been applied in home-based care for patients with frailty, and how can they be systematically categorized?
2. How are data collected, recorded, and transmitted (sensing, communication, and storage)?
3. What outcomes are evaluated, and how are the results being used to inform care (analytics/decision-making, visualization)?
4. Are the tools evaluated for effectiveness and, if so, by what methods?
5. What are the challenges and opportunities with adopting DT-enabling tools for HaH care?

Methods

Review Principles and Protocol

This scoping review followed the principles in the Joanna Briggs Institute (JBI) methodology for scoping reviews [] and is reported in line with the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) [] () and PRISMA-S checklist (). This review followed a published protocol [] and the 5 key methodological stages of scoping reviews []: (1) identifying the research question; (2) identifying relevant studies; (3) selection of relevant studies; (4) charting the data; and (5) collating, summarizing, and reporting the results.

A scoping review was identified as the most appropriate systematic, rigorous approach to knowledge synthesis to capture the required information, as no previous systematic reviews nor evidence exists to answer the research question []. Additionally, many primary studies and evidence in practice are heterogenous in nature, making a scoping review a useful approach to capture the breadth of information and map key concepts required for this work.

Screening and Eligibility

Identifying Relevant Studies

Screening of eligible studies was conducted in line with a predefined protocol [], and the search strategy was supported by a medical research librarian. We searched 6 electronic databases (Embase, MEDLINE, Cochrane CENTRAL, CINAHL, Web of Science, and Scopus; ), and all identified studies were uploaded to a Rayyan database for screening and deduplication []. Gray literature and local evaluation reports relevant to HaH in the United Kingdom were retrieved from National Health Service England, Department of Health and Social Care, and HaH Society websites; the FutureNHS Platform; and related links. Experts in the field and authors were contacted to identify any additional sources. Forward citation searching from the included papers was also conducted.

Primary studies published in the English language were included. To obtain the most up-to-date and relevant information, only studies published from 2019 were included. Initial searches were run until September 11, 2024, then re-run and updated up to September 16, 2025 (). We also recognized that the COVID-19 pandemic may have vastly impacted patient care models and digital technology use; therefore, we wished to capture the most advanced information. Included studies had to report on the monitoring or management of patients with frailty within their own home environment and had to include the reporting of outcomes related to effectiveness, usability, acceptability, and safety. Review articles, protocols, and conference abstracts were excluded as they did not include the primary data required to meet the review objectives.

Selection of Sources of Evidence

In line with JBI guidelines [], a pilot stage was conducted in which 2 authors reviewed a sample of 25 title and abstracts independently and any conflicts were resolved through discussion, allowing screening to continue. The primary researcher (FY) was responsible for full-text screening, with support from a second independent reviewer (MC, HN) to validate the appropriateness of inclusion and exclusion. Discrepancies were resolved through discussion among all 3 reviewers. During the screening phase, reviewers agreed that articles were excluded if the population did not meet the inclusion criteria, did not explicitly mention frailty or meet the definition of frailty as per the British Geriatric Society definition stated in the protocol [], included older adults only or prefrailty only, or set out to assess frailty as an objective or outcome measure. Studies were also excluded if the setting was not a patient’s usual place of residence (ie, a nursing home).

Data Charting

A data extraction form was developed in a format that answered the objectives of the research question covering evidence for each of the 5 architectural layers, as well as country, population, concept, context, applications of the tool, evidence of effectiveness reported, and barriers or facilitators. This was independently piloted with different types of evidence sources (eg, from databases or gray literature) by 3 reviewers (FY, HN, and MC) then collaboratively discussed to highlight and clarify any discrepancies []. It was highlighted that data extraction should focus on the tools used rather than the study or model of care itself. It was also apparent that information obtained from the evaluation reports could often be less structured and more ambiguous, making data extraction more challenging. Data extraction and charting from the included sources of evidence were done initially by the primary researcher (FY) with independent confirmation from the other researchers (MC, WS, MK, HN) to validate the findings. Any uncertainties were clarified via discussion after reviewing the original evidence source. If the required information regarding the tool was not available in the included source, the cited protocol or paper was retrieved for further detail to answer the research objectives.

Critical Appraisal

Critical appraisal was not required in line with JBI guidelines for scoping reviews.

Ethical Considerations

Ethical approval was not required for a scoping literature review.

Collating, Summarizing, and Reporting the Results

Data were collated then summarized to identify themes and key concepts within each category in the data collection form. This was mainly (1) to map the extent and nature of the included studies and their characteristics and (2) to answer the research objectives (ie, the sensing technologies; the communication technologies; information storage, analytics, and visualization methods; ). Finally, the data collection sought to extract any evidence of effectiveness reported in the studies as well as any reported challenges and opportunities of the tools used to support our analysis and interpretation.

Results

Architectural Layers

The results of this review found no existing DT in place for the management of patients with frailty at home nor HaH models. Therefore, in line with the objectives of this review, evidence on DT-enabling tools that could inform any of the 5 architectural layers (ie, sensing, communication, storage, analytics, and visualization) of an HaH DT was sought. These are visualized in the proposed conceptual model in and guided the categorization of the DT-enabling tools in this review based on their functionalities.

‎

Study Characteristics

We included 62 reports from electronic databases and citation searches and 7 gray literature documents that focused on HaH care in the United Kingdom. As displayed in , this resulted in a total of 69 reports (65 studies). Included studies spanned 19 countries across all continents: Europe (n=41), North America (n=17), Asia (n=9), Australia (n=1), and South America (n=1). Over one-half (37/69, 54%) of all included reports used quantitative approaches, followed by 27% (19/69) using mixed methods and 19% (13/69) using qualitative approaches. A large proportion (25/69, 36%) were pilot or feasibility studies testing the tools or intervention. Study designs were described as randomized controlled trials (n=16), cohort studies (n=5), longitudinal studies (n=5), observational studies (n=6), case studies (n=6), evaluations (n=7), feasibility studies (n=10), experimental studies (n=5), quasiexperimental studies (n=1), cross-sectional studies (n=3), or exploratory studies (n=5). Individual study characteristics can be viewed in .

‎

Data Types and Applications of Existing Tools

The types of data captured by the tools () enable an understanding of health care management of patients at home, demonstrating that not only patient data but also environmental-related data and equipment use–related data can be collected. In addition to clinical monitoring, this can provide alerts about early changes in the patients’ usual patterns or behaviors such as the use of appliances in the home (ie, kettles), electricity usage, or use of medication (ie, in digital pill boxes). Many reports presented different tools with different applications and often a combination of functions. As displayed in , a large proportion (36/69, 52%) of reports was focused on measuring frailty markers and the physical function of the patient. Similarly, tools were aimed at improving patient outcomes (n=27) or used for clinical monitoring and review (n=21). The alerting of early deterioration was also a function described in 14 reports; care communication and coordination (n=11) and monitoring of the patient environment (n=10) were also described. Only 6 reports described the use of tools for the education and training of patients (for example, exercise videos, self-care, or nutrition advice).

‎

Sensing Layer

The sensing layer is the fundamental layer within the DT architecture. It is crucial to understand how data are collected for the purpose of informing a DT. Data can be dynamic (changing over time) or static and collected using one or more data-collecting devices [,].

This scoping review included not only technological sensing devices but also any methods that were used to capture data that are used for patient care. Most of the studies relied on manual (human-assisted) methods, wearable sensors, or a combination of both to collect the required data. Manual data collection with equipment, via practitioner observations, or by patient self-reporting (ie, with a questionnaire or diary) was the most prevalent way of capturing data [-]. An array of wearable sensors existed to support with real-time data collection [,,,-], vital sign monitoring [,,,-,-], and assessing functional status [,,,,,-,,,-,,,-] to assist with improving health outcomes [,,-] (see and ).

displays the applications of the tools in relation to the included reports; however, provides a detailed mapping exercise of the applications used to the types of data needed to be collected and the technologies that could be considered within individual architectural layers.

Other types of sensors (listed in ), categorized as physiological measurement sensors (), are nonwearables or nonobtrusive and capture data by activation. Examples include smart weight scales [,,,,,,], “grip balls” to measure grip strength [], or devices that are used to measure gait speed [,,,]. Other nonwearables included ambient sensors that measured motion and activity. Examples included visual (ie, cameras, webcams) [,,] or motion-sensing (ie, motion sensors, mat sensors for sedentary behavior) [,,,,] devices. Some devices had inbuilt hardware that processed the raw data from accelerometers, such as Arduino Nano boards [] used for local data processing, eliminating the need to transmit unprocessed data externally. Similarly, a custom smart speaker built on a Raspberry Pi platform included a 6-microphone audio board, allowing it to capture and process audio data locally [].

Studies also reported on ambient sensors that monitored the environmental surroundings (eg, air quality, water flow, infrared, sound sensors, or GPS [,,,,,]). These would enable the monitoring of any change to usual patterns of behavior in the home. An additional category of sensing included assistive or safety technologies that would be related to patient welfare. This may include detecting any unusual entry or exit to the home by window or door sensors and detecting any hazards in the kitchen with stove guards or gas circuit breakers [,,,-]. Detecting the use of electrical appliances in the home may involve sensors on kitchen appliances (ie, 3Rings smart plug) [] to detect any unusual patterns alerting to deteriorating condition of the patient [,]. Digital calendars and digital medicine dispensers can provide reminders and detect medication administration []. Most studies used a combination of sensors to build a picture of the environment and the patient as part of their intervention.

‎

Communication Layer

The communication layer, sometimes referred to in the literature as the transmission layer or network layer, is responsible for receiving the raw data from the sensing layer and transferring it to the storage layer to be processed and visualized [,]. This layer acts as the intermediary connection or “the bridge” sharing information between the physical layer and the other components in the virtual layer [,,]. In this review, communication methods are any methods used to communicate the information or transmit data, whether this is technology or nontechnology-based. This also included verbal, written, and audiovisual or multimedia as categories of data transmission.

However, digital tools have been shown to transfer data via 3 main categories: (1) mobile or digital messaging, (2) Health Information Exchange or cloud-based communication, or (3) wireless and network-based communication (see , ).

Bluetooth technology was used for the communication layer in 12 reports [,,,,,,,,,,,]. Bluetooth Low Energy (BLE) protocols were also used and are known to be more energy efficient than traditional Bluetooth []. Wireless networks such as WiFi were popular networks of communication in many cases, with some reports emphasizing that a local internet connection was not always required [,]. Nonetheless, studies also reported a privacy-compliant mobile network was needed [,]. Integrating a communication layer into the software or providing 2-way messaging channels were seen as useful for connecting clinicians and patients or caregivers [-,,]. A communication protocol is a key component of the communication layer that governs how data are transmitted between the system components, ensuring efficient and secure transfer across the communication network [,,]. Different protocols were reported in the reviewed literature, including GATT (Generic Attribute Profile) protocols [], RESTful protocols [], or use of an application programming interface gateway []. Some sophisticated technologies used a cloud-based Amazon Lex Chatbot in smart speakers to manage conversations with users using automatic speech recognition in natural language [].

Storage Layer

The storage layer is responsible for hosting the sensed raw data from different sources as well as other integrated data such as historical data that can be used for forecasting future analysis []. Other studies included this as part of the network layer [] or encapsulated it with the communication or analytics layer. We have presented this as its own layer for a more granular examination, to identify how data are stored, and to clarify what potentially needs to be considered to develop an HaH DT prototype.

As represented in , the storage layer was the least described in the included reports (n=18). However, we categorized these as cloud-based storage solutions, server-based storage solutions, databases, or external storage solutions. The most used storage options described were cloud-based platforms [,,,,,,,,,], mobile storage (such as apps) [,,,,,], or remote servers []. Secure servers were described on either desktops or mobile devices []. Few studies described using secure Amazon Web Services as a storage method [,]. NoSQL databases [] and MySQL databases [,] were also used, representing the complexity of the types of data in patient care at home.

Analytics Layer

This layer is responsible for data preparation, processing, and translation to produce meaningful insights and can sometimes be referred to as the computing layer [,]. In the majority of the literature, this comprised all the data-driven models []. In the analytics layer, data may go through data modeling, data mining, or data fusion to consider the current physical state and manage big data to enable and represent future predictions []. Prior to analysis, the data may go through a stage of data aggregation. Data aggregation was mentioned in 7 reports [,,,,,,], including measuring changes in clinical scales or multiparametric data analysis [] to understand and collate the data retrieved from the sensors, although this would not constitute technical analysis itself.

In the analytics layer, methods could fall under any of 4 types: descriptive (What has happened?), diagnostic (Why have we seen this result?), predictive (What might happen?), or prescriptive (What should we do?). However, the included reports described methods corresponding to only 3 of the 4 analytics descriptions: descriptive, predictive, and less so prescriptive methods.

Descriptive analytics could encompass pattern analysis or a variety of statistical analyses as represented in . For example, tests like Cohen kappa were used for categorical data from motion sensors, Bland Altman plots were used for continuous data (ie, sedentary time and stair climbing time), and bivariate correlation analyses were used for weight data []. In addition, principal component analysis has been used for dimensionality reduction with large datasets [,].

Predictive analysis allows forecasting of future possibilities and was described in 8 reports [,,,,,,,], primarily using ML algorithms. This could either be supervised ML (which uses labeled data to train algorithms) or unsupervised ML (which uses clustering on unlabeled data) []. Examples of supervised ML included methods of regression (ie, using logistic regression analysis) [,] or classification (ie, using tree-based classifiers [,,] or support vector machines [,]). Reports also described the use of artificial neural networks, deep learning (multilayer perceptron neural network or with an artificial neural network architecture) [], and neuro-fuzzy systems or adaptive neuro-fuzzy inference systems [].

On the other hand, unsupervised ML would, for example, encompass neural networks or hierarchical cluster analysis like dynamic time warping distance matrix []. This can be a method of time-series analysis, which has been demonstrated to distinguish between ascension and descension on a stairway by measuring changes in speed and frequency and force on a handrail biaxial sensor [].

One report [] used universal sequence mapping as a method to model human frailty through smart home sensor readings. Alternative methods, namely approximate entropy and sample entropy, were compared for sensitivity using Monte Carlo simulation. Markov-Chain analysis was then used to understand which method had higher precision to understand behavioral complexity as a potential biomarker for frailty [].

Prescriptive analysis evidence demonstrated this can support decision-making using methods such as modeling (ie, 3D digital simulation or virtual patient modeling []). One study also used social media analysis in their platforms to assess mental frailty or personality trait shifts (ie, with language analysis) to process the users’ typed text, social interactions, or questionnaires [].

Visualization Layer

This layer is a way of presenting the analyzed data and generating insights to enable informed decision-making by clinicians, patients, or caregivers. Some studies combined this with analytics as a “data analytics and visualization layer“ []; however, for the purposes of this study, we realized that these layers are primarily distinct. In this review, we also separated the visualization layer from the feedback mechanisms in a DT.

This layer was categorized in 3 ways: (1) graphical and dashboard visualizations, (2) 3D visualization, and (3) software or hardware.

The graphical and dashboard visualizations are represented in 2D form. Common methods involved digital dashboards either on a computer or mobile device [,] or remote telemedicine center []. One study also described the use of universal sequence mapping both in analytics and as a way of visualizing patterns similar to chaos game representations [].

A 3D visualization, such as with point clouds, was used to represent 3D skeletal reconstruction by combining IMU data, RGB images, and depth measurements processed by neural networks to understand functional activity []. Similarly, visualization of joint coordinates using visualization tools such as Master Active Gestures ID can help assess sitting or standing behavior []. Alternatively, visualization of a virtual living room through an interface was possible [].

Software or hardware, such as gaming systems, was used to visualize progress or weekly changes [], for example with exercise or walking speed, and can incentivize physical activity with digital rewards or ranking points. Such systems may include augmented reality glasses [] or interactive platforms. Interfaces for this software can be accessed through the web or mobile apps. Some integrated platforms such as in-built interfaces on a wall or television in the patient’s home can also include a virtual embodied agent (Robin the Robot) to support and tailor advice to users [].

Feedback Mechanisms

In addition to the visualizations described in the previous section, methods for feedback to the users (clinicians, patients, caregivers) were used and categorized into 3 groups: (1) health alerts and notifications, (2) health feedback and reports, and (3) mobile or digital platforms.

Health alerts and notifications include automatic alert triggers when a risk is identified. One example is alert-triggered health interventions, which raise an automatic call to the clinician’s hub if a high risk is detected according to its risk categorization []. Calls may also be actioned directly to the caregiver [,]. Alternative forms are notifications via email or text message or to the emergency services [].

Health feedback and reports involve email reports or letters with health information or location [,,,]. Visual reports with videos or photos are also common [,,]. Some reports have also shown voice or audio feedback [] or used specific devices like mobility feedback devices for tailored information [].

Mobile or digital platforms are also commonly used for feedback. Some are integrated with other platforms such as electronic health systems and some are via web interfaces or applications [,] (ie, via a smartwatch []).

As described in , 26 of 69 reports explored the user experience including acceptability and usability, especially as these tools were in infancy of development and implementation. However, only 4 reports explored the impact on care delivery, the system, or clinical outcomes. Where reports demonstrated positive findings, there was a predominant use of manual sensing measures and using mobile apps and self-reporting (). Nonetheless, there is an emergence of wearable technologies, including wearable accelerometers that can be wrist-worn, lumbar-worn, thigh-worn, worn in a smartvest, or worn as a pendant sensor. Devices for smart weight and blood pressure monitoring as well as devices that measure gait speed, sit-to-stand transitions, and movement in the environment were also prominent.

Challenges and Opportunities

Studies reported many challenges with the use of the tools that would need to be considered in future work, but they also described many opportunities. Challenges in these studies were categorized into 4 themes: (1) patient or carer factors, (2) device factors, (3) external factors, and (4) organizational or administrative factors.

Patient or Carer Factors

Most studies identified various patient-related factors impacting successful adoption of digital tools. These included varying levels of digital literacy [,,,,], previous experience with technology [], technophobia [], and a general preference for face-to-face interactions [], all affecting acceptability. Perceived benefit [,], and willingness [,,,,,] to engage are critical yet were often influenced by poor understanding of the technology’s role, unclear instructions, or complex processes. Where self-reporting was needed, it was recognized that this can lead to underestimating or overestimating measures [,]. Physical or visual impairments can affect usability, including device readability or difficulties with touchscreens, multiple-tapping, and managing pop-up notifications [,,,]. Other medical conditions (eg, psychiatric or cognitive disorders, dementia) may affect memory or concentration, hindering device use [,,]. Furthermore, articulation issues were found to hinder voice recognition by assistive robots and led to inadequate conversation [].

Some patients found wearables uncomfortable and inconvenient [,,], disrupting daily activities (eg, washing), or experienced disturbances from LED lights or main leads [], potentially attenuating psychological and physical burden (ie, stress, sleep disturbance, anxiety, depression) [,]. Variabilities in-home routines can also affect the ability to individualize care []. Carers reported burden where patients required support and had difficulties navigating technology independently. They sometimes felt they had to be “on-call,” leading to increased expectations on their care, which reduced their autonomy []. Furthermore, notifications to caregivers can also cause frustration, affecting well-being [,].

Device Factors

Beyond device complexity for certain patient groups (eg, frail older adults), this evidence highlighted uncertainties with device accuracy [,] and potential for false readings [,,]. Some studies reported difficulties with devices distinguishing between subtasks or detecting visitors, leading to noisy or biased spatial measurements []. There were also concerns that the accuracy may not have matched hospital devices (eg, ECGs) [] or varied between standardized clinic or home settings (eg, the 6-minute walk test) []. Current devices also lacked validation for specific populations (eg, those with frailty), leading to potential inaccuracies in algorithms like step counts []. Short or inconsistent data sequences, storage limitations [], and missing values [] further compromised the quality of results. Concerns also included corrupt data, device failure, and malfunctions [,]. Frequent recharging was also reported as burdensome [,], and incompatibility of devices to synchronize with platforms used by health care professionals can limit practical use [,].

External Factors

External factors outside of the home, such as internet signal or power outages, can also affect notifications and data quality [,,]. Factors in the surroundings such as the size, shape, and geometry of a patient’s home environment or furniture can affect the practical application of sensors or standard clinical tests such as walking speed between two points [,]. This can result in artifacts in the data, adding a layer of complexity, especially if visitors are present, making data evaluation difficult. The environment cannot always be anticipated (eg, availability of sockets or flat surfaces), making locations for the devices challenging []. There is also a higher degree of variability in the home, dependent on the patients’ motivation or daily occupational demands, which cannot be predicted. Some patients may also find sensors on their furniture visually unpleasant, such as iAQ sensors on the dining table [].

Organizational Factors

The use of devices often requires efficient and secure communication and coordination between services. Challenges were observed with variable adoption and understanding when remote monitoring was suitable []. A degree of model reorganization was needed to ensure prompt responses as well as access to clinicians out of hours [,]. Factors such as staff capacity, planning, and managing administration (if paper was used) were some of the challenges encountered []. Interoperability with clinical systems, data quality, and coding were also problematic [,,].

Opportunities

Studies reported that the tools offered positive motivation for users [,,], improving physical activity [,,,,] and self-management [,,,,]. Furthermore, they provided a sense of reassurance and safety for patients and carers [,,,]. Opportunities for autonomous and continuous monitoring [,,] as well as integrating data from multiple sensors can increase the accuracy of information [,] and personalization [,,,,,,] and provide the potential to evaluate trends to enhance decision-support systems []. At a service level, these can support early identification and referral [,], support cost-effectiveness of services by reducing the number of clinical visits [,], decrease waiting times [], and support scalability of services []. Furthermore, information sharing between services, clinicians, and for research [] means that data can be used as a proxy for frailty markers [,], to create frailty assessment tools (ie, web applications) [], and to support integrated and collaborative care (ie, supporting staff handovers) [,,,,]. Ways to overcome practical challenges with devices were considered in some reports; these included avoiding recharging by using induction charging [,] or hiding devices in socially acceptable objects [].

Discussion

Principal Findings

This review aimed to explore the existing evidence for the potential DT architectural components that are tools or functionalities currently used in the management of patients at home with frailty and presents a novel conceptual model of an HaH DT ().

The model is supported by previous definitions of DTs expressing that data acquisition and feedback should be dynamic, be timely [], and have bidirectional communication [,]. Prior studies describing DTs have typically focused on specific aspects such as body organs, drug manufacturing, or device and facility management rather than holistic management []. Emerging evidence shows great promise of DTs in home care; however, a lack of in-depth studies exist, making implementation difficult [].

The developed blueprint (conceptual model) presented in this review provides detailed information about the tools that can be used in an HaH DT and can be used to inform the development of a prototype for user testing. An example of a use application such as the monitoring of functional activity as a marker of frailty derived from the blueprint of an HaH DT detailed in is shown in .

‎

The results improve understanding so researchers, practitioners, and solution providers can better navigate the complexity of an HaH DT. This will help identify which data needs to be collected and how the data are managed in each layer, in relation to the specific applications.

This review recognizes that management as part of an HaH DT can support either at the service level (to support workflow management) or the patient level (ie, remote monitoring). Additionally, development of a DT in this context can be understood as a system of systems (ie, comprised of a DT of the home environment and a DT of the patient).

A taxonomy of existing sensing technologies has been identified that can be used both to monitor the patient and the environment, building a comprehensive picture of patient status and support decision-making. These offer the opportunity for collecting active and passive sensor data [] (including the use of unobtrusive sensing technologies [,]), which can benefit the implementation of DTs for patients with frailty. This builds on current evidence around remote monitoring in HaH that aims for early prediction of patient outcomes [,]. In this review, communication technologies currently focus on Bluetooth and network-based wireless methods; however, the emerging use of advanced IoT technologies using 5G and AI may warrant further exploration.

Data Governance, Security, and Ethical Considerations

Knowledge on efficient and secure data management (storage, integration, interoperability) with current health care systems remains crucial. Previous work has identified the need for data repositories such as ”data lakes” or data warehouses to enable data aggregation from multiple technologies and allow for data to be structured and analyzed in DTs []. Although the included studies demonstrated examples of either edge computing or cloud computing, health care systems are increasingly encouraging a cloud-first approach to improve efficiency and organizational agility and to enable scalability []. These may include public cloud services such as Google Cloud, Amazon Web Services, or Microsoft Azure []. Understanding the different data types from this scoping review and recognizing that data can fall into one of 3 types (structured, semistructured, and unstructured []) demonstrates the complexity of health care at home and the need for advancing knowledge around enhanced data management structures. Proposed DT frameworks suggesting data-centric approaches recognize that multimodal data integrating information from various health records are likely and potential solutions to aid implementation may involve in-silico studies using synthetic data []. Furthermore, the challenges with data acquisition, accuracy, and quality have similarly been highlighted previously [,,,]. In-depth information on data security was lacking in the included reports; however, some of the literature discussed potential solutions such as blockchain technology []. Data security measures need to be considered from development and inception, abide by security protocols and regulations, and consider aspects like robust encryption, authorization, and multifactor authentication to mitigate any risks of data leakage [].

Other considerations such as fairness, accountability, transparency, and explainability when designing and implementing data-driven solutions and AI algorithms in a DT system should be integrated [] . Cybersecurity when using IoT devices in health care remains a concern that can have serious consequences for the patient and the health care organization []. However, the integration of security into digital health is increasingly emphasized as we use more digital technology solutions. Carboni et al [] discussed the balance needed between security and care practices, highlighting the perspective that health care systems in reality can be complex environments and solutions must involve end users rather than only the technology itself. Open conversations with users (both staff and patients) and participatory design approaches are methods to improve the adaptability and adoptability of advanced digital technology moving forward. Despite the potential of passive remote monitoring and DTs, previous work has likened remote monitoring to having “nanny cams” [], highlighting that evidence and data are still required to promote their ethical use [,,,].

Potential for Advanced Data Analytics

Most studies in this review described tools that used descriptive and predictive analytics, and some used prescriptive (clinical decision support). However, there is a need to explore the use of diagnostic analytics (ie, why is something happening?) by identifying anomalies in trends or performing root-cause analysis. In HaH, previous studies attempted to explore tools for risk prediction using predictive modeling, but these appear to primarily be case management tools and not necessarily dynamic nor real-time [,]. Emerging studies in health care have developed advanced risk prediction models to support decision-making in high-pressure, time-critical health care settings such as ambulance clinical transport decisions [] . Understanding such novel approaches and how they integrate with electronic patient records is highly valuable []. Other examples [,] that also use ML algorithms in health care are emerging to alert to a patient’s deterioration early; however, these are hospital-based and may not reflect nonhospital settings. Health care is evolving, with increasing use of “command centers” to support patient flow and decision-making [,]. Although these update a dashboard in real time, they are often centralized in hospitals and still require manual input of information by staff. They can support workflow management and efficiency, aiming to enhance safety []; however, they may not have simulation or prediction capabilities like DTs. Moreover, their ability to adapt to complex home environments remains underexplored.

This review advances findings from other literature recognizing the potential of DTs in the management of hospital processes [,] to enhance efficiency or optimize environments []. For patient management, previous studies [] have explored the potential of wearables as trigger warning alerts [,] that can be used in DT technology. In current practice in HaH, evidence for continuous vital sign monitoring using wearables shows benefit, and use is increasing in clinical practice; however, robust evidence is limited [,]. Adding to emerging evidence on DTs that are in early stages of development [], this review demonstrates how existing technologies can be utilized in DTs specifically for HaH, as well as identifying key architecture that warrants further understanding, such as information storage, analytics, and visualization.

Barriers to Adoption and Strategies for Integration

This review has provided some valuable learning regarding the barriers to adoption of digital technologies in the home for patients with frailty. These predominantly include digital literacy and understanding, both for patients and the workforce, followed by usability of the technology. Concerns around device accuracy, longevity, and interoperability and integration with current health care systems were also major challenges, which is recognized in some emerging studies related to remote monitoring [,]. Some potential strategies to overcome these would be to explore organizational readiness and ensure any prototype is co-designed with potential stakeholders (patients, carers, and health care organizations). Furthermore, improving awareness and understanding of the DT system to be deployed would be needed by factoring in training [], funding, and workforce capacity.

Strengths and Limitations

This review provides a systematic approach to reviewing current evidence on the tools used for the management of patients at home with frailty. As limited literature on HaH exists, this review was comprehensive enough to gather learning from the general management of patients at home. However, as the population in question was those with frailty, we may have excluded some studies that involved interventions that would have been useful to explore for nonfrail patients or those only at risk. During the search, many studies were correlation studies looking at the association of a parameter with frailty, and these were excluded as they did not meet the inclusion criteria. The search was comprehensive at trying to include all relevant search terms; however, we recognize that other search terms not included may have been helpful (ie, “Healthcare 4.0” or “5G”). Furthermore, the restricted inclusion of only those documents published in the past 5 years could have meant that helpful studies to answer our question may have been excluded.

Implications for Future Work

This review has highlighted many of the technological capabilities for DTs in HaH as a first step to understanding the potential of current technologies in use for managing patients at home with frailty. This review provides the foundations to enable stakeholders to advance research and development in areas where there are knowledge gaps and consider how an HaH DT can effectively operate within current health care systems to enable safer, personalized, and timely care. However, to advance this concept further, we need to understand its potential application in current practice and its acceptability before it is developed and piloted. Using frameworks such as the Non-adoption, Abandonment, Scale-up, Spread, Sustainability (NASSS) framework [] or the 6 categories in the DT consortium capabilities periodic table [] would be helpful as a basis for understanding how the generated conceptual model may be implemented in practice. Furthermore, design of a potential HaH DT prototype and demonstration as a “show home” would benefit from stakeholder involvement and co-design to ensure that any advancing technologies are user-friendly, have supportive infrastructures, and have the appropriate educational support in place []. For organizations that wish to take learning from this review forward, understanding these 5 core architectural layers provides the fundamental knowledge to build on this foundation. Knowledge about how these interconnect within the current health care systems needs to be understood, especially with regards to data management and transfer of data between the 5 layers. Before this can be deployed in real health care settings, further research or small-scale prototype testing will be required, with a focus on a specific clinical cohort of patients and to determine which data types are appropriate for collection that would usefully alert to early deterioration and methods for coordinating these alerting systems or feedback. Evidence shows that a wealth of sensing technologies already exist, but understanding how these can be adapted to fit in with current systems is necessary. Inclusive methods for communication that do not rely on a patient to have connectivity are important to avoid digital exclusion. It would be useful to further understand secure data storage in health care or linking in with secure data environments before prototype development and the accuracy of potential predictive analytics.

Conclusion

DTs in HaH offer a novel solution to prediction and prevention, reducing the burden on patients and health care systems and potentially improving clinical outcomes. Leveraging the use of technology-enabled care can enhance remote monitoring in real time to support clinicians with delivering care efficiently and safely. This review enabled us to understand the 5 architectural layers that can support stakeholders in advancing research and development of forward-thinking systems such as DTs, particularly within complex settings like HaH for patients with frailty. It also identified knowledge gaps in the existing evidence and allowed us to consider where health care systems should explore improvements to current structures to achieve successful adoption in a rapidly evolving landscape of digital solutions.

Introduction

Acute ischemic stroke (AIS) is a critical neurological emergency caused by sudden cerebral blood flow disruption, leading to rapid neuronal death and potential long-term disability. Timely reperfusion of the ischemic penumbra is essential for effective treatment and represents a crucial therapeutic window to improve outcomes. Intravenous thrombolysis (IVT), administered within 4.5 hours of symptom onset, remains the gold standard therapy for AIS, with its efficacy well established in randomized controlled trials and meta-analyses [-]. While IVT often yields neurological improvement or complete recovery, 8%‐28% of the patients with AIS experience early neurological deterioration (END) [-], defined as a ≥4-point increase in the National Institutes of Health Stroke Scale (NIHSS) score within 24 hours post treatment [,]. This complication has severe clinical consequences, as END is an independent predictor of poor functional recovery and increased 3-month mortality [,]. Given its substantial impact on long-term prognosis, there is an urgent need for reliable predictive models and early intervention strategies to maximize the therapeutic benefits of IVT in patients with AIS.

Over the past decades, numerous studies have investigated predictors of END following IVT in patients with AIS. For instance, Yu et al [] identified advanced age as a significant risk factor for END. A high-quality meta-analysis further demonstrated that hypertension substantially worsens functional outcomes in patients with AIS, nearly doubling the risk of END, early-onset epilepsy, poststroke epilepsy, and mortality []. Additionally, antiplatelet drug resistance in the Chinese population has been linked to both recurrent mild ischemic stroke and END [].

Despite these advances in understanding risk factors, current END prediction methods remain inadequate for clinical practice. Recent attempts to develop predictive tools [-] have faced multiple critical limitations that prevent their widespread adoption: (1) traditional statistical approaches fail to capture complex nonlinear interactions between risk factors, resulting in oversimplified models; (2) existing models demonstrate only moderate discriminative performance (area under the receiver operating characteristic [ROC] curve [AUC] typically <0.80), insufficient for high-stakes clinical decisions; (3) most models lack rigorous external validation, raising concerns about generalizability across diverse patient populations; and (4) many incorporate laboratory parameters unavailable within the critical therapeutic window, rendering them impractical for time-sensitive clinical decision-making. Perhaps most significantly, no existing model provides actionable risk stratification thresholds linked to specific management protocols—a crucial element for translating predictions into clinical benefit.

To address these specific gaps, we developed and validated ENDRAS (Early Neurological Deterioration Risk Assessment System), a machine learning–based predictive model, using a large, multicenter cohort. By integrating readily obtainable clinical variables with advanced computational algorithms, our system enables real-time, high-accuracy risk stratification prior to IVT administration. This work establishes a clinically actionable decision-support tool designed to identify high-risk candidates for intensified monitoring and preemptive therapeutic interventions, with the ultimate goal of improving postthrombolysis outcomes through personalized management strategies.

Methods

Study Design and Setting

We conducted this multicenter, prospective cohort study across 3 strategically selected medical centers in Lianyungang, China, representing distinct demographic populations: Lianyungang Clinical Medical College of Nanjing Medical University (urban core population), Guanyun County People’s Hospital (county or rural population), and Lianyungang Dongfang Hospital (eastern urban population).

Ethical Considerations

This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Lianyungang Clinical Medical College (KY-20240403001‐01). The study was registered with the China Clinical Trial Registry (ChiCTR2400085504) and National Medical Research Registration System (MR-32-24-016371). Written informed consent was waived for the retrospective cohort and obtained from all participants or their legal representatives in the prospective validation cohort. All data were deidentified to protect patient privacy, and no participants received financial compensation.

Recruitment

From January 2017 through April 2024, we conducted a multicenter retrospective study enrolling consecutive adult patients (≥18 y) with AIS who received IVT. The model development cohort (n=1361) comprised patients treated at Lianyungang Clinical Medical College of Nanjing Medical University (January 2017-August 2023), while the external validation cohort (n=566) included patients from Dongfang Hospital and Guanyun County People’s Hospital (September 2023-April 2024).

To ensure methodological rigor, we implemented strict temporal and statistical separation protocols. The chronologically nonoverlapping cohorts maintained complete temporal independence, with feature normalization coefficients derived exclusively from the development dataset and subsequently applied to the validation cohort without modification. This unidirectional preprocessing workflow prevented statistical contamination between phases, preserving analytical integrity and enabling the unbiased evaluation of model generalizability across different clinical settings and time periods.

The inclusion criteria were as follows: (1) age ≥18 years with AIS in the hyperacute phase (<4.5 h from symptom onset), (2) treatment with recombinant tissue plasminogen activator or tenecteplase within 4.5 hours of onset, (3) stroke severity assessed using the NIHSS, and (4) blood pressure controlled to systolic <180 mm Hg or diastolic <115 mm Hg before thrombolysis. The exclusion criteria were as follows: (1) patients who underwent endovascular bridging therapy following IVT and (2) cases with incomplete clinical data.

Data Collection and Outcome Assessment

Clinical data were prospectively collected using standardized electronic case report forms. Baseline characteristics included (1) demographics: age, gender, height, and weight; (2) risk factors: smoking, alcohol consumption, hypertension, diabetes mellitus, atrial fibrillation, valvular heart disease, and coronary artery disease; (3) medication history: antihypertensive, hypoglycemic, lipid-lowering, and anticoagulant therapies; (4) clinical parameters: NIHSS scores (baseline, 12 h, and 24 h post IVT) and blood pressure measurements; (5) laboratory tests: complete blood count, coagulation profile, and biochemical parameters; (6) imaging data: head computed tomography (CT), magnetic resonance imaging, and CT angiography (CTA); and (7) treatment metrics: onset-to-door time and door-to-needle time. The primary outcome was END, defined as an increase of ≥4 points in the NIHSS score within 24 hours post IVT compared to baseline. At least 2 independent neurologists classified stroke subtype according to the Trial of ORG 10172 in Acute Stroke Treatment (TOAST) criteria. In our study, all patients underwent CTA before IVT as part of the standardized acute stroke protocol. For quality control, 2 neuroradiologists and 2 stroke neurologists, blinded to outcomes, independently reviewed the CTA images, with consensus review for discordant readings.

To ensure temporal precedence and eliminate potential information leakage, all predictor variables were obtained prethrombolysis. Laboratory parameters, including red cell distribution width (RDW) and neutrophil count, were measured within 15-30 minutes of admission. The electronic medical record system was configured with timestamp restrictions to ensure that only pretreatment data were incorporated into the prediction model.

Model Development and Validation

For model development, we employed stratified random sampling to partition the development cohort into training (80%) and internal validation (20%) sets, preserving class distribution. We conducted a systematic evaluation of 11 machine learning algorithms, including (1) base classifiers: regularized logistic regression, support vector machine, decision tree, multilayer perceptron, and naive bayes; ensemble methods: voting, stacking, bagging, boosting, random forest, and XGBoost; (2) model performance was comprehensively assessed using multiple metrics: (1) AUC; (2) accuracy, precision, recall, and F₁-score; and (3) confusion matrices.

To ensure robust validation, we implemented a 3-phase approach: (1) internal validation: using the held-out 20% development cohort, (2) external validation: with an independent cohort (n=566) from 2 additional medical centers, and (3) prospective validation: in 20 consecutive thrombolysis-eligible patients.

Feature Importance Analysis

To identify the most influential predictors, we performed a comprehensive feature importance analysis on the optimal algorithm model, specifically employing permutation importance. This method quantifies each feature’s contribution by measuring the increase in prediction errors after randomly permuting the feature, thereby ensuring an unbiased assessment of relevance while accounting for feature correlations. Feature selection was optimized by balancing discriminative performance metrics (AUC, sensitivity or specificity, F₁-score) with clinical utility considerations, including indicator accessibility and cost-effectiveness.

Model Optimization and Selection

We constructed several candidate models using the top 15 features ranked by feature importance from the best algorithm model. We systematically tested models with feature counts ranging from 2 to 15 using the following methods: (1) 5-fold cross-validation (AUC, accuracy, precision, recall, F₁-score); (2) DeLong test for ROC curve comparisons; and (3) clinical feasibility assessment, encompassing data acquisition time and cost-effectiveness, resource availability across health care institutions, and interobserver reliability for subjective indicators. The final model integrated statistical performance, feasibility scores, and cost-benefit analysis. The optimized model was deployed as a web-based decision support system with functionalities such as automatic data validation, an intuitive user interface, and real-time visualization. Minimal input requirements enable seamless clinical integration.

Risk Stratification Analysis

Using the combined dataset (development cohort: n=1361; external validation cohort: n=566), we conducted a systematic risk stratification analysis as follows:

1. Risk prediction and threshold optimization: Individual risk probabilities were generated using the web-based ENDRAS model. The optimal stratification threshold was determined via the Youden index (maximizing [sensitivity+specificity–1]) and ROC curve analysis, evaluating multiple candidate cutoffs. Patients were classified into low-risk and high-risk groups based on the final threshold.
2. Stratum-specific validation: Predictive performance was assessed for each stratum using positive or negative predictive values, likelihood ratios, and diagnostic accuracy metrics.
3. Calibration assessment: Calibration plots visualized agreement between predicted and observed risks across all probabilities.

Statistical Analysis

Statistical analyses were performed using IBM SPSS Statistics 24 (IBM Corporation) and Python-based machine learning libraries. Categorical variables were expressed as numbers (percentages) and compared using χ² or Fisher exact tests. Continuous variables were presented as median (IQR) or mean (SD) and compared using the unpaired 2-tailed t test or Mann-Whitney U test, as appropriate.

Model discrimination was assessed using AUC values, and calibration was evaluated using calibration plots. Decision curve analysis was performed to assess the clinical utility of the ENDRAS model. The optimal cutoff value for risk stratification was determined using the Youden index. Statistical significance was set at P<.05 (2-tailed).

Results

Study Population Screening

From January 2017 to April 2024, we initially identified 2567 patients with first-ever AIS who received IVT within 4.5 hours of symptom onset. After applying the predefined inclusion or exclusion criteria, 1927 patients were enrolled in the final analysis (development cohort: 1361 patients [Lianyungang Clinical College of Nanjing Medical University, The First People’s Hospital of Lianyungang]; external validation cohort: 566 patients [Lianyungang Dongfang Hospital and Guanyun County People’s Hospital]). The patient selection process is detailed in .

‎

Comparative Analysis of Baseline Characteristics

Cohort Homogeneity: Development Cohort Versus External Validation Cohort

The development and external validation cohorts exhibited a high degree of comparability in terms of baseline characteristics (age: median 67, IQR 59-75 years versus median 68, IQR 59-74 years; gender [male]: 899/1361, 66.1% vs 373/566, 65.9%; stroke severity [NIHSS]: median 4, IQR 2-11 versus median 5, IQR 2-10). The prevalence of cardiovascular risk factors, including hypertension, diabetes, and atrial fibrillation, was comparable between the cohorts (see Table S1, ).

Pathophysiological Markers and Risk Factor Distribution

Patients experiencing END in the development cohort demonstrated a distinct pathophysiological profile (). Temporally, END was characterized by counterintuitive clinical metrics—shorter onset-to-door intervals yet prolonged door-to-needle times—alongside significantly elevated baseline NIHSS and modified Rankin scale scores.

Laboratory analyses revealed a comprehensive pattern of dysregulation across multiple systems: hemodynamic instability (elevated systolic blood pressure [SBP]/diastolic blood pressure), systemic inflammation (increased white blood cell, neutrophils, lymphocytes, neutrophil-to-lymphocyte ratio), hematological alterations (elevated platelets, RDW), metabolic derangement (higher glycosylated hemoglobin, homocysteine), coagulation abnormalities (increased international normalized ratio, fibrinogen), and renal dysfunction (elevated creatinine clearance rate).

END demonstrated significant associations with an elevated cardiovascular comorbidity profile (hypertension, diabetes mellitus, atrial fibrillation, valvular/coronary disease), prior antiplatelet therapy exposure, and specific stroke phenotypes—notably large-artery atherosclerosis (LAA) and cardioembolism etiologies, severe intracranial atherosclerotic stenosis (IAS, defined as ≥50% stenosis of the intracranial artery), and heightened 24-hour hemorrhagic transformation risk.

Model Development and Performance

Evaluation Data Preprocessing and Model Architecture

Data preprocessing was performed in Python 3.9.19 (Python Software Foundation) using a structured pipeline: (1) median imputation for missing values, (2) binary target encoding (END occurrence within 24 h post thrombolysis), (3) one-hot encoding for categorical variables, and (4) standardization of continuous features to zero mean and unit variance via sklearn.preprocessing.StandardScaler.

To maximize model performance, we employed an integrated optimization framework following a sequential approach: (1) recursive feature elimination with cross-validation for feature selection to reduce dimensionality, (2) synthetic minority oversampling technique (SMOTE) for class imbalance correction in the training set only, (3) L1 or L2 regularization to prevent overfitting, and (4) Gaussian process–based Bayesian optimization for hyperparameter tuning across all models.

After SMOTE resampling, the class distribution in the training set was balanced to 818:818 (positive:negative ratio). Hyperparameters for all 11 machine learning models were optimized using 5-fold stratified cross-validation on the resampled training dataset. The best-performing model (XGBoost) achieved a cross-validation AUC of 0.983 and maintained robust performance (AUC=0.94) when evaluated on the original, unbalanced validation set.

Class weighting was considered an alternative approach for handling class imbalance. However, SMOTE was ultimately selected based on the following evidence-based considerations: (1) it has consistently demonstrated superior performance in health care prediction tasks with limited sample sizes; (2) it generates synthetic minority class samples while maintaining the underlying feature space distribution; and (3) it provides additional training instances, which is particularly critical for complex ensemble methods that require ample data for robust learning and generalization.

Predictive Algorithm Benchmarking

We systematically compared the 11 machine learning algorithms for END prediction using a multidimensional assessment framework. Performance evaluation incorporated both discriminative metrics (AUC, accuracy, precision, recall, F₁-score) and computational efficiency parameters (training time, prediction time, central processing unit utilization, memory consumption). This comprehensive analytical approach facilitated robust algorithm benchmarking across all candidate models. Comparative performance metrics and the average computational efficiency of the models over different time periods are listed in and , respectively.

‎

XGBoost Demonstrates Superior Predictive Performance

XGBoost consistently outperformed all other algorithms, achieving 88% accuracy and favorable F₁-score for END prediction. The XGBoost model exhibited good discriminative capability across all datasets: training (AUC=1.0), internal validation (AUC=0.94, 95% CI 0.91‐0.97), and external validation (AUC=0.92, 95% CI 0.88‐0.94; ). In the pooled dataset analysis (n=1927), XGBoost maintained robust performance (AUC=0.98, 95% CI 0.97‐0.99; accuracy=94%; precision=87%; recall=90%; F₁-score=0.89), confirming its generalizability and predictive stability ( and ).

‎

‎

‎

Feature Selection

To identify the optimal feature subset, we evaluated models with varying numbers of predictor variables (2-15) using 10-fold cross-validation. XGBoost feature importance analysis, ranked in descending order, identified the top 8 key predictors: NIHSS score, SBP, red blood cell distribution width, internal carotid artery stenosis, homocysteine, neutrophil count, TOAST-LAA subtype, and time from onset to admission (). The 7-variable model (excluding onset-to-door time) achieved the best performance (accuracy=88%; AUC=0.927, 95% CI 0.90‐0.96); improvements beyond this threshold were negligible (6 variables: AUC=0.923; 8 variables: AUC=0.925; Table S2 in ).

To optimize clinical utility in time-sensitive acute stroke settings, we conducted accessibility and cost-effectiveness analyses of predictor variables. Removing homocysteine from the model resulted in minimal performance impact (AUC reduction of 0.003) while significantly enhancing practical implementation by eliminating a laboratory parameter typically unavailable within the critical IVT decision window.

‎

Shapley Additive Explanations Value Analysis

To quantify feature contributions, we performed Shapley additive explanations (SHAP) analysis on the optimized 6-variable XGBoost model (after removing homocysteine to enhance clinical applicability). The NIHSS exhibited the highest feature importance (|SHAP|: mean 2.19, SD 1.72), followed by neutrophil count and RDW (both mean 1.09, SD 0.76), SBP (mean 0.83, SD 0.61), IAS (mean 0.59, SD 0.41), and TOAST-LAA (mean 0.32, SD 0.29). Higher NIHSS scores, neutrophil counts, RDWs, and SBPs generally increased the predicted risk, while the binary variables (IAS, TOAST-LAA) showed more concentrated effects. The ranking differences between XGBoost feature selection and SHAP analysis stem from their complementary methodologies: XGBoost quantifies features based on splitting frequency and information gain across the population, while SHAP values measure each feature’s actual contribution to individual predictions. This methodological distinction enhances our model interpretation—XGBoost identifies globally relevant predictors, while SHAP reveals the personalized impact of features on individual risk assessments, providing crucial insights for precision medicine applications in stroke care.

Technical Implementation

The resulting 6-variable ENDRAS [] incorporates only readily available clinical and laboratory variables and has been implemented as a web-based calculator to facilitate point-of-care clinical decision support in acute stroke management.

ENDRAS was developed as a react.js-based responsive web app with bilingual (English or Chinese) capabilities and robust validation protocols. The architecture features (1) an interactive interface with real-time validation, (2) structured input forms with range verification, (3) color-coded risk stratification visualization, (4) dynamic probability calculation with immediate feedback, and (5) responsive cross-platform design.

Performance evaluation utilized a dataset (n=1927) segmented into 20 batches (100 records/batch) with 3 complete iteration cycles. Testing was conducted on Chrome (v120.0; Google) with Intel Core i7 processor and 16 GB RAM. The results demonstrated exceptional computational efficiency: mean prediction latency of 0.0177 (SD 0.0021) seconds, memory utilization of 88.80 (SD 0.01) MB, per-record processing time of 0.18 ms, and high consistency in repeated executions (coefficient of variation<5%).

Risk Stratification and Comparative Performance

Based on ROC analysis, an optimal threshold of 29% was established for risk stratification. The study cohort (n=1927) was dichotomized into high- and low-risk groups, with observed END event rates of 95.24% (420/441) and 2.42% (36/1486), respectively ().

Forward stepwise logistic regression (likelihood ratio method), conducted on the combined dataset of all 1927 patients, identified ENDRAS-derived risk probability as a powerful independent predictor of postthrombolysis END (OR 5080.684, 95% CI 2353.205-10969.440, P<.001). The model demonstrated favorable discriminative capability (AUC 0.988, 95% CI 0.983-0.993), outperforming all individual predictors, including TOAST-LAA classification, moderate-to-severe IAS, NIHSS score, SBP, neutrophil count, and RDW ().

‎

‎

Model Calibration and Clinical Utility

Calibration analysis demonstrated excellent concordance between predicted probabilities and observed END frequencies throughout the risk spectrum. Decision curve analysis revealed that ENDRAS-guided intervention provided substantial net clinical benefit compared to default “prediction all” or “prediction none” strategies across a comprehensive range of threshold probabilities ( and ).

‎

‎

Prospective Validation of ENDRAS

A prospective validation cohort (n=20) was recruited to preliminarily validate the real-world performance of ENDRAS, with baseline characteristics shown in . In this small independent cohort comprising 5 END and 15 non-END patients, ENDRAS demonstrated preliminary discriminative capability with 80.0% sensitivity (4/5 patients with END correctly identified as high-risk) and 86.7% specificity (13/15 patients without END correctly classified as low-risk). The overall predictive accuracy was 85.0% (17/20), with positive and negative predictive values of 66.7% and 92.9%, respectively.

Discussion

Comparison With Prior Work

Machine learning methods leverage computational algorithms to model complex, nonlinear relationships between clinical variables, thereby overcoming the limitations of traditional risk assessment approaches. This has established them as superior predictive tools, demonstrating high accuracy and robust discriminatory performance in outcome prediction while maintaining clinical usability [-]. While prior studies [,] have explored the prediction of END using patient history, laboratory results, and biochemical markers, their clinical applicability remains limited due to insufficient validation. In cases of END following IVT, some predictive models define END as an increase of ≥2 points on the NIHSS score within 24 hours to 7 days [,,]. However, this threshold may lack sensitivity in severe stroke cases (eg, baseline NIHSS ≥20), where a 2-point change in a total score of 42 could underestimate true clinical deterioration []. To address this limitation, recent predictive models have incorporated multifactorial risk assessments. For instance, 1 study [] developed a model integrating 6 key predictors—age, diabetes, atrial fibrillation, antiplatelet therapy, C-reactive protein levels, and baseline NIHSS scores —to improve END risk stratification.

While predictive models for END continue to evolve, significant limitations persist in their clinical application. The model developed by Tian et al [], which incorporates the neutrophil-to-lymphocyte ratio, mean platelet volume, body mass index, and atrial fibrillation, represents an advance in pre-IVT risk stratification. However, its utility is constrained by several factors: (1) limited validation in IVT-treated patients: the model has not been validated specifically in cohorts undergoing IVT, raising questions about its generalizability to this population; (2) methodological limitations: the use of manual grouping rather than randomized allocation may introduce selection bias, reducing the reliability of the predictive outcomes; and (3) incomplete risk factor integration: the model omits established predictors of END, such as baseline NIHSS scores and blood pressure parameters, which have been consistently associated with postthrombolysis neurological decline. To address the limitations in the existing END prediction models, we designed our research approach with several methodological improvements. Based on previous research findings, our study focused on key aspects to enhance predictive accuracy: (1) conducting prospective validation specifically in IVT-treated cohorts; (2) utilizing machine learning approaches to mitigate grouping bias; and (3) implementing comprehensive variable selection that incorporates established predictors such as NIHSS scores and hemodynamic markers.

Principal Results

ENDRAS Development and Validation Performance

We developed ENDRAS incorporating 6 pre-IVT predictors of END: LAA subtype, carotid stenosis ≥50%, NIHSS, SBP, neutrophil count, and RDW. The ENDRAS model demonstrated good discrimination (AUC=0.988, 95% CI 0.983‐0.993 [n=1927, combined development and external validation cohorts]), favorable calibration (calibration curve R=0.998), and improved net benefit on decision curve analysis versus “prediction-all/no” strategies. ENDRAS effectively balances pathophysiological relevance with clinical implementability for postthrombolysis END prediction in AIS.

Methodological Considerations

The high discriminative performance (AUC=0.988) warrants critical evaluation. At least 3 factors potentially contribute to this performance: (1) our feature selection methodology identified variables capturing the complementary pathophysiological mechanisms underlying neurological deterioration, (2) XGBoost’s ensemble architecture detected complex nonlinear relationships between predictors that traditional regression methods cannot capture, and (3) our substantial cohort (n=1927) provided sufficient statistical power to reliably estimate model parameters.

To minimize the possibility of data leakage, all variables included in the ENDRAS model were obtained before thrombolytic therapy, ensuring that no post-treatment information was used during model training or evaluation. Therefore, the model was developed entirely based on prethrombolysis data. As mentioned in the Data Collection and Outcome Assessment section, to ensure temporal precedence and eliminate potential information leakage, all predictor variables were obtained prethrombolysis. Laboratory parameters, including RDW and neutrophil count, were measured within 15‐30 minutes of admission. The electronic medical record system was configured with timestamp restrictions to ensure that only pretreatment data were incorporated into the prediction model.

Despite the strong performance, further validation in larger and more diverse prospective cohorts is still needed to confirm the model’s robustness and generalizability. High performance in our current validation cohorts does not eliminate the need for ongoing evaluation, as the model must demonstrate consistent reliability across varied clinical settings and patient populations before widespread implementation can be recommended.

Despite these strengths, we acknowledge potential limitations in generalizability. Performance metrics may reflect some degree of optimism, and external application across diverse clinical settings may yield more moderate discrimination values. Ongoing prospective validation studies across heterogeneous populations will better establish the model’s transportability and calibration stability.

ENDRAS demonstrated clinically acceptable misclassification rates (false-positive rate=2.42%, false-negative rate=4.76%) with comprehensive validation metrics (true positives=420, false positives=36, false negatives=21, true negatives=1450). The minimal false-negative rate prioritizes the detection of deterioration events, while the moderate false-positive rate represents an acceptable clinical trade-off, particularly in acute stroke management where consequences of unidentified deterioration typically outweigh those of enhanced monitoring.

The computational performance differences between models hold practical significance across multiple dimensions. From a development perspective, XGBoost’s training time is highly efficient (75 ms), which enables rapid model iteration and hyperparameter tuning, significantly accelerating the development cycle through faster experimentation and validation. From an implementation perspective, the final ENDRAS prediction model built with XGBoost occupies only 88 MB of the memory, allowing deployment on resource-constrained platforms such as tablets and emergency department workstations with limited memory capacity. This advantage is especially valuable in community hospitals with older information technology infrastructure. From a maintenance standpoint, XGBoost’s computational efficiency supports continual model retraining with updated data. In our clinical workflow, the model is retrained quarterly, integrating new patient data to preserve predictive performance. The lower computational requirements also lead to significantly reduced cloud computing costs relative to more resource-intensive alternatives. Although these performance differences may appear modest in absolute terms, they profoundly influence real-world implementation, particularly in resource-limited environments where computational efficiency directly determines the accessibility and sustainability of the clinical decision support system.

Pathophysiological Significance of ENDRAS Predictors

The strength of ENDRAS lies in its selection of variables that collectively capture the multifaceted pathophysiological mechanisms underlying END. Each predictor contributes unique insights into END risk stratification.

LAA as a Key Risk Factor for END

This study identified LAA as a significant predictor of END, consistent with prior evidence []. In patients with LAA, progressive ischemia within 24 hours post thrombolysis—driven by slow collateral flow and vulnerable plaque instability—increases END risk. Additionally, moderate-to-severe stenosis (eg, middle cerebral artery or branch atheromatous lesions) correlates with higher NIHSS scores, stroke progression, and worse disability, further elevating END likelihood []. These findings align with previous reports (Nam et al [], Joeng et al []), reinforcing LAA’s role in END pathogenesis.

NIHSS as a Predictor of END Risk

The NIHSS is a validated measure of stroke severity, and our nomogram confirmed that baseline NIHSS ≥4 significantly elevates END risk. This aligns with prior studies [,] showing that moderate-to-severe strokes (NIHSS >4) correlate with larger infarct volumes, higher rates of hemorrhagic transformation or cerebral edema, and consequently, greater END susceptibility. These findings underscore NIHSS’s utility not only for severity stratification but also for early END risk prediction.

Blood Pressure Dynamics and END Risk in AIS

While moderately elevated blood pressure may support perfusion in ischemic brain regions, excessive SBP (>185/105 mm Hg)—particularly within the first 24 hours post IVT—is linked to hemorrhagic transformation, cerebral edema, and stroke recurrence, all contributors to END and poor outcomes []. High blood pressure variability and sustained hypertension also correlate with 3-month stroke recurrence, likely via blood-brain barrier disruption (eg, oxidative stress, AQP4 upregulation) [-]. Though admission SBP is a modifiable END risk factor, optimal thresholds and management strategies require further validation [].

Neutrophils as a Predictor of END

Although previous risk models did not assess neutrophil levels, multicenter data identify elevated neutrophil counts as an independent predictor of END []. Higher neutrophil levels correlate with larger infarct volumes, worse outcomes, and greater tissue damage (eg, edema, hemorrhagic transformation) []. Notably, neutrophil depletion in experimental stroke models reduces infarct size and secondary injury, reinforcing the causal role of neutrophils in END pathogenesis. These findings align with our clinical observations, positioning neutrophils as a key modifiable risk factor for END [-].

RDW as a Prognostic Biomarker in AIS With IVT

Baseline RDW independently predicts all-cause mortality in patients with AIS receiving IVT []. Elevated RDW correlates with moderate-to-severe stroke severity, poor functional outcomes (modified Rankin scale 3‐6), and lower Barthel index scores (<85 at 3 mo) []. Notably, patients with futile recanalization exhibit significantly higher RDW than those with successful reperfusion. Mechanistically, pro-inflammatory cytokines (tumor necrosis factor α, interleukin-1 β, interleukin-6) suppress erythropoiesis, increasing RDW, while simultaneously promoting endothelial activation, blood-brain barrier disruption, and neutrophil infiltration, exacerbating ischemic injury [,]. Our findings further support this inflammatory cascade, demonstrating significantly higher neutrophil counts in patients with END, suggesting a potential interplay between RDW, inflammation, and END risk.

Clinical Implementation Framework

The robust predictive performance of ENDRAS, coupled with the clear pathophysiological relevance of its component predictors, establishes a strong foundation for clinical implementation. ENDRAS provides real-time risk stratification for END in patients with prethrombolysis, addressing the current reliance on subjective assessment. We propose a risk-stratified management framework:

• Patients with high risk (ENDRAS ≥0.29)
  1. Hourly neurological monitoring for 24 h post thrombolysis
  2. Immediate advanced imaging (CT perfusion or magnetic resonance imaging)
  3. Blood pressure control (140‐160 mm Hg systolic)
  4. Priority assessment for neurovascular intervention
  5. Prophylactic neuroprotective measures
  6. Consider empirical antiedema therapy for large infarcts
• Patients with low risk (ENDRAS <0.29)
  1. Neurological assessments every 2 hours for 24 hours
  2. Standard follow-up imaging at 24 hours
  3. Standard blood pressure management
  4. Standard glycemic and temperature monitoring
  5. Early mobilization when appropriate
  6. Expedited rehabilitation planning

This framework complements standard guidelines, adding personalized risk assessment while preserving clinical judgment when patient-specific factors warrant deviation. Although our model development established 29% as the optimal probability threshold for creating a comprehensive optimal model (sensitivity: 95.24%, specificity: 97.58%, positive predictive value: 92.11%, negative predictive value: 98.57%), we recognize that different clinical environments may require different thresholds depending on their specific priorities.

Institutions with emergency departments, intensive care units, or those treating high-risk patients may opt for a lower threshold of 20% (high-sensitivity conservative model) to enhance “rule-out” capability and reduce missed END events, given its strong negative predictive value (98.61%) in high-stakes situations. For standard clinical pathways, multidisciplinary team decisions, and clinical trials, a threshold of 25% (high-performance balanced model) provides balanced and reliable performance across all metrics, delivering trustworthy positive and negative predictions.

In contrast, specialist clinics emphasizing resource efficiency and confirmatory testing may choose a threshold of 35% (precision diagnostic model) to limit false positives while preserving diagnostic accuracy. Resource-limited settings focused on cost containment and avoidance of overtreatment might implement a 40% threshold (high-specificity precision model), leveraging its “rule-in” capability to initiate treatment protocols based on highly specific positive results.

We recommend that institutions select the threshold most appropriate for their clinical context, monitoring resources, and risk management strategy. The detailed performance metrics and clinical interpretations for each threshold are provided in Table S3 in .

Clinical Impact Demonstration

An example of a high-risk case is as follows: a 72-year-old male with left MCA occlusion (NIHSS 18, ASPECTS 8) received an intravenous recombinant tissue plasminogen activator at 2.5 h post onset, with an ENDRAS score of 0.82. Hourly assessments detected subtle deterioration (+2 NIHSS points) at hour 4, triggering immediate advanced imaging that revealed salvageable penumbra. Emergent thrombectomy resulted in favorable outcomes (modified Rankin scale: 2). Without ENDRAS-guided monitoring, this intervention window might have been missed.

An example of a low-risk case is as follows: A 65-year-old female with small cortical infarct (NIHSS 4) received an intravenous recombinant tissue plasminogen activator at 3 h post onset, with an ENDRAS score of 0.28. Standard monitoring enabled early mobilization at 24 h and discharge planning by day 2, optimizing resource utilization without compromising safety.

These cases demonstrate how ENDRAS transforms from a predictive tool into a practical clinical decision support system guiding resource allocation, monitoring intensity, and intervention decisions. Prospective validation of these management protocols represents an important direction for future research.

Limitations

Despite ENDRAS demonstrating robust predictive performance (AUC=0.988, 95% CI 0.983‐0.993), several limitations warrant consideration. While the model’s error profile is clinically acceptable (false-positive rate=2.42%, false-negative rate=4.76%), these misclassifications could still impact treatment decisions. The time span of the validation cohort (September 2023 to April 2024, totaling 8 mo) is relatively short, which does not allow for a thorough assessment of the model’s stability across different time periods. All validation data originate from hospitals within a single geographical area, limiting the evaluation of the model’s generalizability across different regional populations. We acknowledge a significant lack of statistical power in the prospective validation portion (n=20). Given an expected incidence rate of END events of 8%‐28%, this sample size is only projected to yield 1‐6 events, far below the recommended standard for prediction model validation.

Furthermore, a critical limitation is ENDRAS’s development and validation exclusively within Chinese stroke populations. Established ethnic variations in stroke pathophysiology—including higher intracranial atherosclerosis prevalence in East Asian populations versus predominant extracranial atherosclerosis in Western cohorts—may affect model performance across different demographic contexts.

We recognize that dependence on CTA for IAS assessment poses an implementation challenge for our prediction model in resource-limited settings. To address this limitation, we have conducted a sensitivity analysis using a modified model without IAS, which demonstrated reduced but still clinically meaningful performance (AUC=0.857, 95% CI 0.837‐0.877). The identification of IAS as a strong predictor is consistent with pathophysiological mechanisms and previous studies linking it to stroke progression.

We acknowledge that excluding patients receiving bridging therapy (IV thrombolysis followed by mechanical thrombectomy) introduces selection bias and restricts applicability to the broader stroke population. This exclusion was necessary due to substantially different neurological courses, monitoring protocols, and clinical trajectories between patients undergoing emergent endovascular intervention and those receiving IV thrombolysis alone.

To address these limitations, we have initiated several complementary research directions. First, recognizing the selection bias introduced by excluding bridging therapy patients, we are developing a dedicated prediction model for this population (ENDRAS-MT). Preliminary analysis (n=216) indicates distinct predictive features in this group, with procedural factors (time to groin puncture, number of passes) and angiographic findings (collateral status, thrombus location) emerging as important predictors. This complementary model will extend risk stratification to the broader population of patients with AIS eligible for reperfusion therapies.

Second, we acknowledge CTA as the current reference standard for intracranial stenosis evaluation, offering higher sensitivity than magnetic resonance angiography or transcranial Doppler. We plan to validate alternative IAS assessment methods (transcranial Doppler, magnetic resonance angiography) against CTA in future work to develop conversion algorithms across modalities, potentially enabling resource-adapted model implementation.

Third, to enhance generalizability and validation rigor, we are planning a larger multicenter prospective validation study (target n>200) across diverse Chinese regions with extended follow-ups (12+ mo), while establishing international collaborations to assess applicability in non-Asian populations. This expanded validation will provide more robust evidence for ENDRAS’s clinical utility across heterogeneous populations.

Finally, future development will focus on optimizing computational efficiency for real-time implementation, refining risk stratification thresholds, and integrating additional relevant biomarkers and imaging parameters to improve predictive accuracy. These comprehensive steps are essential for ENDRAS’s successful clinical translation and widespread adoption in acute stroke management. Information on the proportion of missing data and interrater reliability for imaging and classification variables was not available in this study. These aspects should be addressed in future research to improve data quality and reproducibility.

Conclusions

ENDRAS represents a promising approach in postthrombolysis stroke care by providing clinicians with an objective tool for END risk assessment. The model shows encouraging discriminative performance in combined development and external validation cohorts, suggesting potential clinical utility in identifying high-risk patients who may benefit from intensified monitoring and early intervention, though further validation in larger prospective studies is needed.

Future research should focus on prospective implementation studies across varied health care settings, refinement of risk thresholds to optimize clinical decision-making, and investigation of whether ENDRAS-guided management actually improves patient outcomes compared to standard care. As we continue to validate and refine this model, ENDRAS offers a promising framework for personalized postthrombolysis monitoring that balances resource utilization with patient safety.

SEOUL, South Korea (AP) — South Korea will require advertisers to label their ads made with artificial intelligence technologies from next year as it seeks to curb a surge of deceptive promotions featuring fabricated experts or deep-faked celebrities endorsing food or pharmaceutical products on social media.

Following a policy meeting chaired by Prime Minister Kim Min-seok on Wednesday, officials said they will ramp up screening and removal of problematic AI-generated ads and impose punitive fines, citing growing risks to consumers — especially older people who struggle to tell whether content is AI-made.

Lee Dong-hoon, director of economic and financial policy at the Office for Government Policy Coordination, said in a briefing that such ads are “disrupting the market order,” and that “swift action is now essential.”

“Anyone who creates, edits, and posts AI-generated photos or videos will be required to label them as AI-made, and the users of the platform will be prohibited from removing or tampering with those labels,” he said.

AI-generated ads using digitally fabricated experts or deepfake videos and audios of celebrities, promoting everything from weight-loss pills and cosmetics to illegal gambling sites, have become staples across the South Korean spaces of YouTube, Facebook and other social media platforms.

The government will seek to revise the telecommunications act and other related laws so the AI-labeling requirement, along with strengthened monitoring and punitive measures, can take effect in early 2026. Companies operating the platforms will also be responsible for ensuring that advertisers comply with the labeling rules, Lee said.

AI fuels surge in false ads

Officials say it’s becoming increasingly difficult to monitor and detect the growing number of false ads fueled by AI. South Korea’s Food and Drug Safety Ministry identified more than 96,700 illegal online ads of food and pharmaceutical products in 2024 and 68,950 through September this year, up from around 59,000 in 2023.

The problem is also spreading into areas such as private education, cosmetics and illegal gambling services, leaving the Korea Consumer Agency and other watchdogs struggling to keep pace, the Government Policy Coordination Office said.

Beyond deceptive ads and misinformation, South Korea is also grappling with sexual abuse enabled by AI and other digital technologies. A Seoul court last month sentenced a 33-year-old man to life in prison for running an online blackmail ring that sexually exploited or abused more than 200 victims, including many minors who were threatened with deepfakes and other manipulated sexual images and videos.

Officials plan to raise fines and also introduce punitive penalties next year to discourage the creation of false AI-generated ads, saying those who knowingly distribute false or fabricated information online or through other telecommunications networks could be held liable for damages up to five times the losses incurred.

Officials will also strengthen monitoring and faster takedown procedures, including enabling reviews within 24 hours and introducing an emergency process to block harmful ads even before deliberation is complete. They also plan to bolster the monitoring capabilities of the Food and Drug Safety Ministry and the Korea Consumer Agency — using AI, of course.

Despite risks, South Korea’s love for AI grows

Prime Minister Kim, Seoul’s No. 2 official behind President Lee Jae Myung, said during the policy meeting that it’s crucial to “minimize the side effects of new technologies” as the country embraces the “AI era.”

The plans to label AI-generated ads were announced as Lee, in a separate meeting with business leaders, reiterated his government’s ambitions for AI, pledging national efforts to strengthen South Korea’s capabilities in advanced computer chips that power the global AI race.

Government plans include more research and development spending on AI-specific chips and other advanced semiconductor products as well as expanding the country’s chip manufacturing hubs beyond metropolitan areas near the capital city of Seoul to the southern regions. South Korean chipmakers, including Samsung Electronics and SK Hynix, combined for more than 65% of the global memory chip market last year.

The science and telecommunications ministry also said Wednesday it will require the country’s wireless carriers to transition to 5G standalone networks, which are seen as optimal for advanced AI applications because of their higher bandwidth and lower latency, as a condition for renewing their 3G and LTE licenses.