Advancing primary care for childhood pneumonia: a machine learning-based approach to prognosis and case management
Preprint
Abstract
Background Pneumonia is the leading cause of preventable mortality under five years of age. Appropriate case management is as essential as disease prevention interventions, especially in primary care settings. Computer science has been used accurately and widely for pneumonia diagnosis; however, prognosis studies are relatively low. Herein, we developed a machine learning-based clinical decision support system tool for childhood pneumonia to provide prognostic support for case management. Methods We analyzed data from 437 children admitted to our clinic with a pneumonia diagnosis between 2014 and 2020. Pediatricians encoded the raw dataset according to candidate features. Before the experimental study of the machine learning algorithms of Pycaret, SMOTE-Tomek was utilized for managing imbalanced datasets. The feature selection was made by examining the SHAP values of the algorithm with the highest performance and re-modeled with the most important clinical features. We optimized hyperparameters and employed ensemble methods to develop a robust predictive model. Results Optimized models predicted pneumonia prognosis with %77-88 accuracy. It was shown that severity could be determined over %84 by five clinical features: hypoxia, respiratory distress, age, Z score of weight for age, and antibiotic usage before admission. Conclusions In this experimental study, we demonstrated that contemporary data science methods, such as oversampling, feature selection, and machine learning tools, are promising in predicting the critical care need of patients. Even in small-size samples like our study, ML methods can reach current wisdom. Highlights Pneumonia accounts for 14% of mortality in children under 5, with over 740,000 deaths in 2019 alone. WHO and UNICEF’s GAPPD aims to cut mortality rates by focusing on vaccinations, sanitation, breastfeeding, and addressing pediatric HIV. Accurate diagnosis and timely treatment can reduce pneumonia mortality by up to 28%, yet diagnosing can be challenging. Many in underdeveloped regions lack access to essential equipment and trained staff, exacerbating mortality rates. Data science and machine learning offer promising solutions for pneumonia management, especially in LMICs, with a focus on prognostic support.Keywords
This publication has 28 references indexed in Scilit:
- Using big data for risk stratification of childhood pneumonia in low-income and middle-income countries (LMICs): Challenges and opportunitiesEBioMedicine, 2021
- Prediction of Impending Septic Shock in Children With SepsisCritical Care Explorations, 2021
- COVID-19 pandemic and the incidence of community-acquired pneumonia in elderly peopleRespiratory Investigation, 2020
- Data Science for Child HealthThe Journal of Pediatrics, 2019
- Does This Child Have Pneumonia?JAMA, 2017
- Application of a Prognostic Scale to Estimate the Mortality of Children Hospitalized with Community-acquired PneumoniaThe Pediatric Infectious Disease Journal, 2016
- Community-Acquired Pneumonia Requiring Hospitalization among U.S. ChildrenNew England Journal of Medicine, 2015
- The Management of Community-Acquired Pneumonia in Infants and Children Older Than 3 Months of Age: Clinical Practice Guidelines by the Pediatric Infectious Diseases Society and the Infectious Diseases Society of AmericaClinical Infectious Diseases, 2011
- Effect of pneumonia case management on mortality in neonates, infants, and preschool children: a meta-analysis of community-based trialsThe Lancet Infectious Diseases, 2003
- Reduction in total under-five mortality in western Nepal through community-based antimicrobial treatment of pneumoniaThe Lancet, 1991