NPJ Digital Medicine最新文献

Faced with challenging cases, doctors are increasingly seeking diagnostic advice from large language models (LLMs). This study aims to compare the ability of LLMs and human physicians to diagnose challenging cases. An offline dataset of 67 challenging cases with primary gastrointestinal symptoms was used to solicit possible diagnoses from seven LLMs and 22 gastroenterologists. The diagnoses by Claude 3.5 Sonnet covered the highest proportion (95% confidence interval [CI]) of instructive diagnoses (76.1%, [70.6%–80.9%]), significantly surpassing all the gastroenterologists (p < 0.05 for all). Claude 3.5 Sonnet achieved a significantly higher coverage rate (95% CI) than that of the gastroenterologists using search engines or other traditional resource (76.1% [70.6%–80.9%] vs. 45.5% [40.7%-50.4%], p < 0.001). The study highlights that advanced LLMs may assist gastroenterologists with instructive, time-saving, and cost-effective diagnostic scopes in challenging cases.

Coordinated access to multi-domain health data can facilitate the development and implementation of artificial intelligence-augmented clinical decision support (AI-CDS). However, scalable institutional frameworks supporting these activities are lacking. We present the PULSE framework, aimed to establish an integrative and ethically governed ecosystem for the patient-guided, patient-contextualized use of multi-domain health data for AI-augmented care. We describe deliverables related to stakeholder engagement and infrastructure development to support routine engagement of patients for consent-guided data abstraction, pre-processing, and cloud migration to support AI-CDS model development and surveillance. Central focus is placed on the routine collection of social determinants of health and patient self-reported health status to contextualize and evaluate models for fair and equitable use. Inaugural feasibility is reported for over 30,000 consecutively engaged patients. The described framework, conceptually developed to support a multi-site cardiovascular institute, is translatable to other disease domains, offering a validated architecture for use by large-scale tertiary care institutions.

Digital therapeutics (DTx) are software-driven solutions for prevention, treatment, and management of medical conditions. Despite a pro-DTx momentum in China, global DTx trial assessments overlooked the country. We identified 756 DTx trials in China and analyzed their characteristics and quality parameters. Over 70% were funded by governments, hospitals, and universities, with tertiary hospitals in eastern China leading most trials. 44.8% used automated DTx, with 39.2% DTx-guided. Most trials focused on management (52.5%) and treatment (38.1%), with few on prevention (9.4%). Mental, behavioral, or neurodevelopmental disorders represented the leading condition category of focus. Recent declines in median sample size, median duration, and mean number of sites were noted. Only 18% of trials were at low overall risk of bias. While recognizing the rapid development of DTx trials in China, we call for better trial design and methodological rigor, prioritization of preventive and primary care, wider condition category scope, and higher inclusivity.

This study demonstrates the potential of multimodal large language models in calculating safety indicators and predicting contraindications for laser vision correction. ChatGPT-4 effectively analyzed ocular data, calculated key indicators, generated calculator codes, and outperformed traditional machine learning models and indicators in handling unstructured data and corneal topography. Its modality-independent system enabled efficient and accurate data analysis. Despite longer processing times, ChatGPT-4’s performance highlights its potential as a decision-support tool, offering advancements in improving safety.

Body composition prediction from 3D optical imagery has previously been studied with linear algorithms. In this study, we present a novel application of deep 3D convolutional graph networks and nonlinear Gaussian process regression for human body shape parameterization and body composition estimation. We trained and tested linear and nonlinear models with ablation studies on a novel ensemble body shape dataset containing 4286 scans. Nonlinear GPR produced up to a 20% reduction in prediction error and up to a 30% increase in precision over linear regression for both sexes in 10 tested body composition variables. Deep shape features produced 6–8% reduction in prediction error over linear PCA features for males only, and a 4–14% reduction in precision error for both sexes. All coefficients of determination (R²) for all predicted variables were above 0.86 and achieved lower estimation RMSEs than all previous work on 10 metrics of body composition.

Endovascular treatment (EVT) for vertebrobasilar artery occlusion (VBAO) with atrial fibrillation presents complex clinical challenges. This comprehensive multicenter study of 525 patients across 15 Chinese provinces investigated nuanced predictors beyond conventional metrics. While 45.1% achieved favorable outcomes at 90 days, our advanced machine learning approach unveiled subtle interaction effects among clinical variables not captured by traditional statistical methods. The predictive model distinguished high-risk subgroups by integrating multiple parameters, demonstrating superior prognostic precision compared to standard NIHSS-based assessments. Novel findings include nonlinear relationships between dyslipidemia, stroke severity, and functional recovery. The developed predictive algorithm (AUC 0.719 internally, 0.684 externally) offers a more sophisticated risk stratification tool, potentially guiding personalized treatment strategies in high-complexity VBAO patients with atrial fibrillation.

This study investigates the application of prompt engineering to optimize prompt-driven ChatGPT for generating electronic medical records (EMRs) during lung nodule screening. We assessed the performance of ChatGPT in generating EMRs from patient–provider verbal consultations and integrated this approach into practical tools, such as WeChat mini-programs, accessible to patients before hospital visits. The findings highlight ChatGPT’s potential to enhance workflow efficiency and improve diagnostic processes in clinical settings.

Real-time monitoring of hydration biomarkers in tandem with biophysical markers can offer valuable physiological insights about heat stress and related thermoregulatory response. These metrics have been challenging to achieve with wearable sensors. Here we present a closed-loop electrochemical/biophysical wearable sensing device and algorithms that directly measure whole-body sweat loss, sweating rate, sodium concentration, and sodium loss with electrode arrays embedded in a microfluidic channel. The device contains two temperature sensors for skin temperature and thermal flux recordings, and an accelerometer for real-time monitoring of activity level. An onboard haptic module enables vibratory feedback cues to the wearer once critical sweat loss thresholds are reached. Data is stored onboard in memory and autonomously transmitted via Bluetooth to a smartphone and cloud portal. Field studies conducted in physically demanding activities demonstrate the key capabilities of this platform to inform hydration interventions in highly challenging real-world settings.

Large language models (LLMs) have the potential to enhance evidence synthesis efficiency and accuracy. This study assessed LLM-only and LLM-assisted methods in data extraction and risk of bias assessment for 107 trials on complementary medicine. Moonshot-v1-128k and Claude-3.5-sonnet achieved high accuracy (≥95%), with LLM-assisted methods performing better (≥97%). LLM-assisted methods significantly reduced processing time (14.7 and 5.9 min vs. 86.9 and 10.4 min for conventional methods). These findings highlight LLMs’ potential when integrated with human expertise.