Wearable Device and Deep Learning Predict Dengue Deterioration
Key Insights
A new study uses a wearable device to continuously monitor patients with dengue (search) fever, collecting photoplethysmography (PPG) signals.
Deep learning models analyze PPG data to predict the onset of clinical shock and assess illness severity in real-time.
The models demonstrated the ability to forecast clinical shock two hours in advance, potentially enabling earlier intervention.
A recent study published in Nature Digital Medicine details the development and evaluation of deep learning models that utilize continuous photoplethysmography (PPG) signals, captured via a non-invasive wearable device, to predict clinical outcomes in patients hospitalized with dengue (search) fever. The research, conducted at the Hospital for Tropical Diseases in Ho Chi Minh City, Vietnam, highlights the potential of wearable technology combined with artificial intelligence to improve real-time monitoring and prediction of disease progression in dengue patients.
Study Design and Methodology
The prospective observational study enrolled adult and pediatric patients (age 8 or older) admitted to the hospital with a clinical diagnosis of dengue (search) and hospitalization of fewer than 28 hours at the time of enrollment. Participants wore a wrist wearable device that continuously captured PPG signals using red and infra-red wavelengths at 100 Hz for up to 24 hours. Clinical data, including demographics, symptoms, vital signs, and administered treatments, were also recorded. The primary outcome was the prediction of clinical shock (dengue shock syndrome (search) or recurrent shock) two hours in advance, using a 10-minute segment of PPG recording. Secondary outcomes included patient illness severity during the initial 24-hour period of hospitalization, assessed using the NEWS2 score and the dengue-specific modified Sequential Organ Failure Assessment score (mSOFA).
The study enrolled 153 patients, with data partitioned into training (n = 132) and independent hold-out test sets (n = 21). Deep learning models, including Spatio-temporal Fusion Transformer (SFT), convolutional neural networks (CNN), and CNN-LSTM, were employed to analyze the PPG signals and predict outcomes. Multi-modal approaches incorporating demographic and clinical information were also developed and evaluated.
Key Findings
The deep learning models demonstrated the ability to predict the onset of clinical shock two hours in advance. The models also showed promise in assessing patient illness severity during the initial 24 hours of hospitalization. The study suggests that continuous PPG monitoring, combined with deep learning algorithms, can provide valuable insights into disease progression and enable timely clinical intervention.
Clinical Implications
Dengue (search) fever is a significant public health concern, particularly in tropical and subtropical regions. Early prediction of severe dengue and timely intervention are crucial for improving patient outcomes. This study introduces a low-cost, non-invasive method for continuous clinical monitoring of dengue patients, potentially enabling healthcare providers to identify patients at high risk of deterioration and provide appropriate treatment. The use of wearable sensors and AI could be particularly beneficial in resource-limited settings where continuous monitoring is challenging.
Future Directions
The researchers emphasize the need for further validation of the models in larger, more diverse populations. Future studies could also explore the integration of additional data sources, such as laboratory results and electronic health records, to further improve the accuracy and reliability of the predictions. The development of user-friendly interfaces and decision support tools could facilitate the implementation of this technology in clinical practice.