Updates: Top Trending News in Data and Technology
In the ongoing battle against COVID-19, artificial intelligence (AI) systems have shown impressive capabilities in predicting ICU admission for infected patients. These systems, relying on machine learning models, have demonstrated high performance in forecasting hospitalization, ICU admission, and death by analysing clinical and administrative health data[1].
A 2025 study, for instance, revealed that machine learning-based predictive models showed strong accuracy in predicting ICU admission among COVID-19 patients, utilising large administrative databases regularly updated and compliant with data protection regulations. The study also identified potential protective medication effects, such as ACE inhibitors, ARBs, and metformin, which the AI took into account during risk stratification[1].
AI's role extends beyond prediction, as it also enhances the prognostic evaluation of COVID-19 pneumonia severity. By quantitatively assessing lung injury on CT scans, AI provides an efficient, reproducible, and faster analysis compared to radiologists alone. Imaging biomarkers combined with inflammatory markers serve as independent predictors of disease severity, aiding clinical risk stratification and informing ICU needs[2].
Despite these promising predictive capabilities, a systematic review up to mid-2024 revealed that most AI models (74%) remain in early development and lack clinical integration, with only a small fraction (2%) operationalized in real ICU settings[4]. This highlights the need for a paradigm shift towards prospective testing and implementation of AI tools in clinical workflows.
Noteworthy, AI models developed at large medical centres, such as Mount Sinai, exhibit strong reliability in predicting early hospital admissions, which can be extended to ICU admission, offering actionable insights hours earlier than traditional methods and potentially improving patient flow and outcomes[3].
In summary, AI systems are highly effective in retrospective and predictive modelling of ICU admission for COVID-19 patients, with advanced models incorporating clinical, demographic, and imaging data. However, their clinical operationalization remains limited, and further prospective validation and integration into healthcare practice are required to realise their full potential[1][2][3][4].
Meanwhile, advancements in AI continue in various fields. Honda partners with the Ohio Department of Transportation to monitor road conditions with connected car technology. Researchers at the University of Exeter develop an AI system to predict a patient's chances of developing dementia within the next 2 years with 92 percent accuracy. Researchers at Durham University use a supercomputer to model impact scenarios explaining Uranus's unusual planetary tilt. Lastly, researchers at Microsoft create an AI system that can identify common bugs in Python code, and those at Massachusetts General Hospital create a predictive model to detect signs of lung cancer in asymptomatic patients from a blood sample.
