Clinical risk prediction of progression to severe dengue illness during the febrile phase in primary healthcare settings

Abstract

To date, no specific antiviral treatment is available for dengue. The recognition of clinical progression to severe disease during the early phases of the illness and the decisions made at triage are key for the clinical management and often determine the outcome. Previous studies have been developed to explore the relationship between risk factors and disease severity and have identified the risk predictors of severe dengue illness. However, the results of these studies have been inconsistent due to studies with small sample sizes and the significant variation in predictors on different days of illness. In addition, previous risk prediction models for early progression to severe disease were based on hospitalised patient data, thus missing to investigate early predictors potentially capable to inform triage and decision making in outpatient settings, where infections are first seen. This thesis aims to fill this knowledge gap and investigates how early risk predictors, collected within the first four days since symptom onset, can be used in conjunction with clinical prediction models to estimate individualised risks of progression to severe disease in outpatient settings. This thesis provides evidence that platelet count, liver function test, and serum albumin can be used as risk predictors of progression to severe disease, backed both by a systematic review and meta-analysis of the existing literature, and by the analysis of clinical data from two large cohort studies conducted in Thailand and Vietnam. To investigate the development of the first signs of severe diseases and compensate for the infrequency of dengue shock syndrome, we investigate two clinical endpoints (i.e. dengue shock syndrome and a combined endpoint of dengue shock syndrome and/or moderate plasma leakage) using a variety of statistical models, including logistic regression and machine learning techniques, such as extreme gradient boosted tree and artificial neural network. The risk prediction models were developed following best practices, were optimised through internal and external validation, and using independent datasets from Thailand and Vietnam, two countries in dengue-endemic areas. We also characterised the temporal dynamics of clinical and laboratory parameters throughout disease progression and investigated the use and predictive power of sequential measurements for prediction, using recurrent neural networks. The use of statistical predictive models to inform triage and clinical management of dengue in the outpatient setting is promising and has the potential to support decision making in the early phases of disease in clinical practice. While further and extensive validation across healthcare contexts and populations is needed, this work lays the foundations for integrating evidence-based and data-driven methods into a decision support system in clinical practice, which in turn can contribute to informing decision making and optimising healthcare allocation and the optimal use of resources.