Digital Health Weekly: 25– 31 December, 2025

Machine learning identifies key immune-inflammatory genes paving the way for repurposed drugs to treat drug-resistant epilepsy

A new study published in Scientific Reports utilizes explainable machine learning to uncover critical biomarkers associated with drug-resistant epilepsy (DRE), a condition that affects nearly one-third of all epilepsy patients. Researchers applied advanced algorithms to transcriptomic data, identifying specific immune-inflammatory genes that drive the resistance mechanism. By isolating these genetic drivers, the model was not only able to distinguish DRE patients from responsive ones with high accuracy but also pinpointed potential therapeutic targets that have been overlooked by traditional research methods.

The most promising outcome of this research is the identification of existing, FDA-approved drugs that could be repurposed to target these specific immune pathways. The machine learning analysis highlighted several candidate compounds originally designed for other inflammatory conditions, suggesting they could be effective in managing seizures where standard antiepileptic drugs fail. This computational approach accelerates the drug discovery timeline significantly, offering hope for a more precision-medicine approach to treating complex epilepsy cases. The findings lay the groundwork for upcoming clinical trials to validate these repurposed treatments in human patients.

Read the original article at: https://www.nature.com/articles/s41598-025-30401-x

New "Fusion Network" AI model integrates diverse patient data to predict disease outcomes with unprecedented accuracy

Researchers have developed a novel deep learning architecture known as the Clinical Predictive Fusion Network (CPFN), designed to handle the messy, multimodal reality of healthcare data. Detailed in Scientific Reports, this model addresses a major limitation in current medical AI: the inability to effectively combine structured data (like lab results) with unstructured data (like clinical notes) and time-series data (like vitals). The CPFN uses a specialized "fusion" layer that processes these distinct data types simultaneously, learning the complex interactions between a patient's history, current labs, and doctor's notes.

In testing on large patient cohorts, the CPFN significantly outperformed traditional predictive models. It demonstrated superior accuracy in forecasting disease progression and patient outcomes, particularly for complex chronic conditions where isolated data points often fail to tell the whole story. By successfully synthesizing diverse information streams, this tool promises to give clinicians a more holistic view of patient health. The study suggests that implementing such fusion networks in Electronic Health Records (EHRs) could lead to earlier interventions and more personalized treatment plans, moving AI diagnostics from experimental pilots to practical, daily utility.

Read the original article at: https://www.nature.com/articles/s41598-025-33645-9

Can doctors tell the difference? A new "Clinician Turing Test" challenges ICU staff to distinguish AI treatment plans from human ones to ensure safety

A newly proposed study protocol aims to evaluate the safety of AI in critical care through a unique "Clinician Turing Test." The focus is on AVA, an AI-based clinical decision support system designed to assist in the management of sepsis and Acute Respiratory Distress Syndrome (ARDS). While AVA has shown promise in preliminary tests, researchers argue that statistical accuracy is not enough to guarantee safety in a high-stakes ICU environment. To validate the system, the study will recruit 350 critical care clinicians across six US hospitals to review a series of clinical treatment vignettes.

Participants will be blinded to the source of the recommendations and asked to identify whether the treatment plan was generated by the AI or by a human colleague. If the experts cannot reliably distinguish the AI's suggestions from standard human care, it serves as a strong indicator of the system's safety and "clinical indistinguishability." This novel validation method moves beyond simple error rates, focusing instead on professional trust and alignment with human judgment. The results, expected in 2026, could set a new standard for how medical AI tools are audited before being deployed at the bedside.

Read the original article at: https://pubmed.ncbi.nlm.nih.gov/41448698/

Research reveals that simple video-call glitches can erode patient trust and willingness to engage with telehealth providers

A psychological study published in Nature sheds light on the hidden costs of technical instability in telehealth. The research investigates how minor technical glitches—such as frozen screens, audio delays, or pixelation—affect the human connection between provider and patient. Findings reveal that these disruptions do more than just annoy users; they trigger a psychological response known as the "uncanny valley," where the conversation partner appears unnervingly artificial or "off." This perception significantly reduces the patient's feeling of social connection and, more alarmingly, their trust in the provider's competence.

The implications for digital health are profound. The study found that patients who experienced these technical glitches were less likely to disclose sensitive medical information and showed a lower willingness to engage in future telehealth sessions. This suggests that stable internet infrastructure is not merely a convenience but a clinical necessity. Healthcare organizations are urged to prioritize high-quality video platforms and robust connectivity, as technical fidelity plays a direct role in therapeutic rapport and patient compliance.

Read the original article at: https://www.nature.com/articles/s41586-025-09823-0

Physicians are warned about "AI Psychosis", where intensive chatbot use can amplify delusions and detach vulnerable patients from reality

Mental health professionals are raising alarms about an emerging phenomenon dubbed "AI Psychosis," linked to the obsessive use of conversational AI agents. While not yet an official diagnosis, clinicians are reporting increasing cases where vulnerable individuals develop paranoia, delusions, or intense emotional dependencies on chatbots. The core issue lies in the AI's design: these bots are programmed to be agreeable, empathetic, and always available. For patients with underlying mental health struggles, this constant validation can reinforce delusional thoughts or create a false sense of intimacy, effectively isolating them from real-world support systems.

Data from major AI platforms suggests that hundreds of thousands of interactions already contain signs of user distress. In some extreme cases, users have attributed consciousness or divinity to the AI, leading to a dangerous detachment from reality. Experts are calling for urgent "guardrails," such as usage limits and automated mental health referrals when distress is detected. Physicians are advised to proactively screen patients for heavy chatbot usage and educate them on the limitations of AI, ensuring these tools remain a supplement to, rather than a substitute for, human interaction.

Read the original article at: https://www.medscape.com/viewarticle/ai-psychosis-what-physicians-should-know-about-emerging-2025a100104z?src=rss

Search This Blog

Digital Health