Towards improved risk prediction of clinical deterioration in acutely ill children in LMICs

Under-five mortality rates have declined globally, yet disparities persist, especially in Sub-Saharan Africa and Southern Asia. Key factors in low- and middle-income countries (LMICs) include malnutrition, acute illnesses, and limited healthcare access. Timely identification of children most at risk for clinical deterioration or mortality remains an important challenge. This thesis explores potential predictive tools, including dermatological assessment, risk prediction models, and fecal volatile organic compound (VOC) analysis. Skin changes were common but not specific to malnutrition. Risk prediction models, used across different hospital settings, including emergency departments, pediatric wards, and pediatric intensive care units were assessed, revealing limitations for use in LMICs. The thesis underscores the need for a multi-model validation approach, stakeholder engagement, and qualitative studies to overcome implementation barriers. Machine learning models can fairly accurately discriminate between mortality and survival in children with an acute illness or complicated severe malnutrition. Larger VOC studies with additional metabolomics could further our understanding of the intestinal dysfunction seen in these children and potentially identify biomarkers. Finally, precision nutrition highlights how personalized treatments based on individual needs and microbial data could improve outcomes in malnourished children. Conclusively, no single prediction model can be recommended for risk prediction in the LMIC setting; specific skin changes were not seen, and VOC analysis demonstrated moderate accuracy in predicting mortality and highlighted potential biomarkers that warrant further research.