Global Digi-Health Monitor: Calendar Week 16, 2025
AMIE: Google’s Conversational Diagnostic AI Outperforms Doctors
Google researchers have developed AMIE (Articulate Medical Intelligence Explorer), a large language model optimized for diagnostic medical conversations that outperformed primary care physicians in a randomized study. The AI demonstrated superior diagnostic accuracy and was rated higher on most communication metrics by both specialist evaluators and patient-actors.
AI+EQ: The Human Element in Healthcare Technology
Owen Tripp, CEO of Included Health, argues that while AI is helping doctors become more present with patients by handling administrative tasks, the healthcare experience extends far beyond the exam room. True transformation requires integrating emotional intelligence (EQ) with artificial intelligence throughout the entire healthcare ecosystem to create connected, personalized experiences that see patients as whole people.
AI Models Show Healthcare Bias Based on Patient Demographics
A study published in Nature Medicine reveals that healthcare AI models can recommend different treatments for identical clinical scenarios based solely on socioeconomic and demographic factors, mirroring real-world healthcare inequities. Researchers found high-income patients were more likely to receive advanced diagnostic tests like CT scans, while low-income patients were often advised to undergo no further testing.
CRISPR Screen Identifies Commander Complex as Key Player in Parkinson’s Disease
Northwestern Medicine scientists have discovered a previously unknown role of the Commander complex—a group of 16 proteins that deliver specific proteins to lysosomes—in Parkinson’s disease risk using a genome-wide CRISPR interference screen. The research, published in Science Advances, found that loss-of-function variants in Commander genes appear more frequently in Parkinson’s patients than in healthy individuals across two independent cohorts (UK Biobank and AMP-PD), potentially explaining why some people with pathogenic GBA1 variants develop the disease while others don’t. This breakthrough provides new therapeutic targets, as the Commander complex maintains lysosomal function—a critical cellular recycling system that malfunctions in several neurodegenerative conditions—suggesting that drugs enhancing Commander protein function could improve lysosomal activity and potentially complement existing treatments aimed at increasing lysosomal GCase activity.
Electronic Health Records Enable Large-Scale Analysis of COVID-19’s Cardiovascular Impact in Children
This groundbreaking study leveraged electronic health records (EHRs) from 19 U.S. children’s hospitals to analyze cardiovascular outcomes in 297,920 SARS-CoV-2-positive children and 915,402 controls, demonstrating how standardized EHR data can power sophisticated epidemiological research at scale. The researchers utilized the PCORnet Common Data Model to harmonize patient data across institutions, applied propensity score stratification to balance hundreds of covariates between comparison groups, and extracted diagnostic codes (ICD-10-CM, ICD-9-CM, SNOMED) to identify cardiovascular outcomes occurring 28-179 days post-infection. This methodology revealed significantly increased risks for multiple cardiovascular conditions including hypertension, arrhythmias, myocarditis, and heart failure (relative risks 1.26-2.92), with findings persisting across demographic subgroups and virus variants. While the study showcases EHRs’ tremendous value for population-level research, it also highlights inherent limitations including potential misclassification bias from undiagnosed cases, variations in documentation practices across institutions, and challenges in capturing healthcare encounters outside hospital systems—illustrating both the power and constraints of using EHR data for longitudinal health outcomes research.
