Last Month In Healthcare AI (January 2025)

Contents: ?? 1) Prediction & Diagnosis | ??2) Treatment & Care | ?? 3) Generative AI ??4) Adoption & Governance

?? 1) Prediction & Diagnosis

??The NHS will trial Aire-DM, an AI tool predicting type 2 diabetes risk up to 13 years early by analyzing subtle ECGchanges. Developed at Imperial College London, it achieves 70% accuracy and could improve with added data. Starting in 2025, the tool aims to enable targeted prevention and stop diabetes before it develops.

??A new AI tool predicts atrial fibrillation (AF) risk before symptoms appear, using GP records to analyze factors like age, sex, and medical history. Developed by the University of Leeds, it’s being trialed to reduce stroke risk, with high-risk individuals using handheld ECGs for early detection.

??A U.S. study found that AI in mammography screening may boost breast cancer detection rates by 21%. Data from over 747,000 women showed AI-enhanced scans helped radiologists spot anomalies more effectively. While 22% of the increase was linked to selection bias from higher-risk patients enrolling, the rest was attributed to AI.

??An AI-powered laser analysis method detects stage 1a breast cancer with 98% accuracy. This non-invasive technique combines Raman spectroscopy with machine learning to analyze chemical changes in blood plasma, achieving over 90% accuracy in distinguishing cancer subtypes.

??FaceAge uses AI to analyze selfies, estimating biological age and predicting health outcomes, including lifespan. In a study by Mass General Brigham, it outperformed doctors in predicting cancer survival rates by detecting facial aging signs linked to health. While promising, the tool needs peer review, and questions remain about its reliability against cosmetic interventions.

??AI aging clocks predict health and lifespan using blood metabolites, measuring biological age (“MileAge”) and its gap with chronological age. Tested on 225,000 UK Biobank participants, they link accelerated aging to frailty and higher mortality, enabling early interventions to slow aging.

?? 2) Treatment & Care

??A machine learning model combining clinical and genomic data improved predictions for HR-positive, HER2-negative metastatic breast cancer treated with CDK4/6 inhibitors. It identified four risk groups with distinct survival outcomes, aiding personalized treatment. Key factors included TP53 loss, a mutation tied to cancer progression, and clinical markers like liver metastasis.

??The NHS is leveraging AI to identify patients at risk of frequent emergency visits, providing targeted care to reduce A&Epressure. By predicting high-risk cases, the initiative focuses on early intervention and personalized support, improving outcomes while easing the strain on emergency services.

??AI models can accurately identify suicidality in hospital patients by analyzing psychiatric admission records for signs of suicidal thoughts or intent. Using advanced language processing, the models extract key information from free-text health records, offering early warning and monitoring tools for psychiatric emergencies.

??Carta Healthcare acquired Realyze Intelligence to enhance its AI platform for matching patients to clinical trials. Realyze’s clinician-trained AI streamlines trial enrollment by analyzing electronic health records, increasing efficiency and outcomes. At UPMC, the software matched seven times more patients and doubled trial enrollment.

?? 3) Generative AI

??Claimable uses generative AI to fight claim denials, crafting appeal letters that integrate policy details, clinical research, and patient data. Initially focused on autoimmune disorders, the platform charges $50 per appeal and overturns 85% of denials, offering hope to patients battling unjust rejections.

??A study shows GPT-4 can assist with physical exams by providing tailored guidance based on patient symptoms, scoring over 80% for accuracy, comprehensiveness, and readability. The AI performed well for cases like leg pain but occasionally lacked diagnostic specificity, highlighting the importance of physician oversight.

??Suki has launched new features for its AI assistant, integrating Google Cloud’s Vertex AI platform to enhance clinical workflows. The assistant now includes patient summaries for concise overviews of medical history and a Q&A function to quickly retrieve information like drug interactions or past test dates.

?? 4) Adoption & Governance

??The FDA's new guidelines streamline approvals for AI-enabled medical devices, allowing developers to implement updates without extra submissions if a Predetermined Change Control Plan (PCCP) is included. The plan ensures safety and effectiveness for nearly 1,000 approved devices while balancing innovation and regulatory efficiency.

??The MHRA's AI Airlock pilot is testing five AI-powered devices to improve evidence collection and safely fast-track innovation into the NHS. Tools include those for COPD risk prediction, radiology reporting, AI monitoring, cancer care personalization, and clinician support.

??ECRI's 2024 report highlights AI risks as the top healthcare hazard, warning of biases, model drift, and insufficient governance, especially in smaller facilities. Cybersecurity threats from third-party vendors and inadequate support for homecare devices also rank high, exposing vulnerabilities in healthcare systems.

??A Health Foundation survey found 75% of the UK public supports sharing health data to develop AI in the NHS, especially for eye health (59%), medication (58%), and chronic illness data (57%). However, fewer were open to sharing data from smartphones (47%) or on sexual health (44%). Support varied across socioeconomic groups, with lower-income groups being less willing.

??A global study highlights biases in healthcare AI and calls for diverse, representative datasets to improve equity. Recommendations include identifying biases in datasets, testing AI for fairness, and ensuring data suitability for development. NHS England supports these measures to promote equitable and responsible AI use in healthcare.