Analyse de la relation entre la fréquence cardiaque et la température corporelle chez les enfants en état critique par apprentissage automatique

Abstract

The main objective of this project is to explore the relationship between heart rate (HR) and body temperature (BT) specifically in critical ill children aged 0 to 18 years admitted to the Pediatric Intensive Care Unit (PICU). To achieve this goal, we employed Machine learning (ML) algorithms to investigate the complex relationship between these vital signs, going beyond traditional linear regression methods for a more in-depth understanding of the association. The results reveal a consistent trend of decreasing HR with increasing patient age, confirming the observed inverse correlation. Initially, we employed conventional linear regression models, and their performance was notably low. Specifically, we observed an R-squared (R2) value ranging from 0.2967 to 0.3220, along with a Mean squared error (MSE) ranging from 610.0736 to 620.5992. Furthermore, a detailed analysis identifies Gradient Boosting Machines (GBM) implemented with Quantile regression (QR), as the best model performance compared to others. Utilizing total quantile loss as an evaluation metric for QR models, the top-performing model demonstrated the lowest total quantile loss of 6.5069 ± 5.2507e-05. This model distinguished by its non-linear kernel, effectively capture the intricate non-linear relationships between HR, BT, and age in critically ill young patients at different quantiles. Moreover, we’ve designed a simple user interface specifically for generating HR predictions based on three essential parameters : current HR, current BT, and patient’s age. The interface provides the predicted HR at various percentiles along with the current HR position, allowing caregivers to determine whether the HR falls within the normal range (between the 5th and 95th percentiles) or not. In summary, this research contributes to enhancing our understanding of HR trends related to BT and age in critically ill children, challenging established assumptions about the linear relationship between HR and BT. The selected ML model, QR performed with GBM, demonstrate their effectiveness in capturing the non-linear dynamics of these physiological parameters, emphasizing the importance of reconsidering traditional assumptions in clinical contexts. Additionally, we aspire that the user interface developed for HR prediction can aid clinicians in making clinical decisions.

Date	28 Jun 2024
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Rita Noumeir (Supervisor) & Philippe Jouvet (Co-supervisor)