Clinical Scorecard: Utilizing AI for Early Identification of Infectious Diseases in Dutch Primary Care with BERT and ERNIE
At a Glance
Category
Detail
Condition
Emerging infectious diseases with early, non-specific clinical presentations
Key Mechanisms
Unsupervised natural language processing using ERNIE framework leveraging BERT-based semantic representations to detect atypical disease clusters from unstructured primary care text data
Target Population
Patients presenting to Dutch general practitioners with early or mild symptoms of infectious diseases
Care Setting
Primary care/general practice in the Netherlands
Key Highlights
ERNIE autonomously detects emerging infectious disease clusters without reliance on diagnostic codes or laboratory confirmation.
The framework identified early COVID-19-like clusters before confirmed cases and detected RSV and simulated West Nile Virus patterns.
Dutch primary care free-text data offers a rich, dense source for early surveillance beyond structured coding systems.
Guideline-Based Recommendations
Diagnosis
Incorporate AI-driven unsupervised NLP methods like ERNIE to analyze unstructured primary care text for early detection of emerging infections.
Do not rely solely on predefined diagnostic codes or laboratory results for surveillance of novel infectious diseases.
Management
Use early identification of atypical symptom clusters to enable timely public health interventions and resource allocation.
Integrate AI surveillance outputs with existing public health monitoring systems for enhanced outbreak response.
Monitoring & Follow-up
Continuously monitor primary care consultation text data using ERNIE to detect emerging disease signals in near real-time.
Validate AI-detected clusters with external datasets to ensure stability and accuracy.
Risks
Traditional surveillance systems may miss early, non-specific disease signals leading to delayed outbreak detection.
Dependence on structured codes alone limits detection of novel or atypical infectious disease presentations.
Patient & Prescribing Data
Patients attending Dutch general practices presenting with early or undifferentiated symptoms potentially indicative of emerging infections.
Early AI-driven detection allows clinicians and public health officials to anticipate outbreaks and optimize treatment and containment strategies before laboratory confirmation.
Clinical Best Practices
Leverage unstructured clinical text from primary care electronic health records for surveillance of emerging infectious diseases.
Apply unsupervised AI models that do not require predefined diagnostic labels to capture novel disease patterns.
Combine anomaly detection, clustering, and interpretability techniques to enhance understanding and usability of AI surveillance outputs.
by Maarten Homburg, Gijs Danoe, Marjolein Y. Berger, Tim olde Hartman, Jean Muris, Andreas Voss, Axel Hamprecht, Maarten F. Brilman, Lilian L. Peters, Matthijs S. Berends
