A Neonatal Sepsis Prediction Algorithm Using Electronic Medical Record Data
![Thumbnail](/xmlui/bitstream/handle/123456789/2064/A%20Neonatal%20Sepsis%20Prediction%20Algorithm%20Using%20Electronic%20Medical%20Record%20Data.pdf.jpg?sequence=4&isAllowed=y)
View/ Open
Date
2022Author
Ezeobi, Dennis Peace
Wasswa, William
Musimenta, Angella
Kyoyagala, Stella
Metadata
Show full item recordAbstract
Background: Neonatal sepsis is a significant cause of neonatal death and has 23 been a major challenge worldwide. The difficulty in early diagnosis of neonatal sepsis leads to delay in treatment. The early diagnosis of neonatal sepsis has been predicted to improve neonatal outcomes. The use of machine learning techniques with the relevant screening parameters provides new ways of understanding neonatal sepsis and having possible solutions to tackle the challenges it presents. This work proposes an algorithm for predicting neonatal sepsis using electronic medical record (EMR) data from Mbarara Regional Referral Hospital (MRRH) that can improve the early recognition and treatment of sepsis in neonates.
Methods: A retrospective analysis was performed on datasets composed of de-identified electronic medical records collected between 2015 to 2019. The dataset contains records of 482 neonates hospitalized in Mbarara Regional Referral Hospital, Uganda. The proposed algorithm implements Support Vector Machine (SVM), Logistic regression (LR), K-nearest neighbor (KNN), Naïve Bayes (NB), and Decision tree (DT) algorithms, which were trained, tested, and compared based on the acquired data. The performance of the proposed algorithm was evaluated by comparing it with the physician’s diagnosis. The experiment used a Stratified K-fold cross-validation technique to evaluate the performance of the models. Statistical significance of the experimental results was carried out using the Wilcoxon Signed-Rank Test.
Results: The results of this study show that the proposed algorithm (with the lowest Sensitivity of 0.95, lowest Specificity of 0.95) outperformed the physician diagnosis (Sensitivity = 0.89, Specificity = 44 0.11). SVM model with radial basis function, polynomial kernels, and DT model (with the highest AUROC values of 0.98) performed better than the other 45 models in predicting neonatal sepsis as their results were statistically significant.
Conclusions: The study provides evidence that the combination of maternal risk factors, neonatal clinical signs, and laboratory tests effectively diagnose neonatal sepsis. Based on the study result, the proposed algorithm can help identify neonatal sepsis cases as it exceeded clinicians' sensitivity and specificity. A prospective study is warranted to test the algorithm's clinical utility, which could provide a decision support aid to clinicians.
Collections
- Research Articles [37]