Prediction Models for Intravenous Immunoglobulin Non-Responders of Kawasaki Disease Using Machine Learning.

BACKGROUND AND OBJECTIVE: Intravenous immunoglobulin (IVIG) is a prominent therapeutic agent for Kawasaki disease (KD) that significantly reduces the incidence of coronary artery anomalies. Various methodologies, including machine learning, have been employed to develop IVIG non-responder prediction models; however, their validation and reproducibility remain unverified. This study aimed to develop a predictive scoring system for identifying IVIG nonresponders and rigorously test the accuracy and reliability of this system. METHODS: The study included an exposure group of 228 IVIG non-responders and a control group of 997 IVIG responders. Subsequently, a predictive machine learning model was constructed. The Shizuoka score, including variables such as the "initial treatment date" (cutoff: < 4 days), sodium level (cutoff: < 133 mEq/L), total bilirubin level (cutoff: ≥ 0.5 mg/dL), and neutrophil-to-lymphocyte ratio (cutoff: ≥ 2.6), was established. Patients meeting two or more of these criteria were grouped as high-risk IVIG non-responders. Using the Shizuoka score to stratify IVIG responders, propensity score matching was used to analyze 85 patients each for IVIG and IVIG-added prednisolone treatment in the high-risk group. In the IVIG plus prednisolone group, the IVIG non-responder count significantly decreased (p < 0.001), with an odds ratio of 0.192 (95% confidence interval 0.078-0.441). CONCLUSIONS: Intravenous immunoglobulin non-responders were predicted using machine learning models and validated using propensity score matching. The initiation of initial IVIG-added prednisolone treatment in the high-risk group identified by the Shizuoka score, crafted using machine learning models, appears useful for predicting IVIG non-responders.

Identification of Pediatric Bacterial Gastroenteritis From Blood Counts and Interviews Based on Machine Learning.

INTRODUCTION:  Differentiating between bacterial and viral gastroenteritis is crucial in pediatric enteritis practice. Our objective was to use machine learning (ML) to identify acute gastroenteritis (AG) caused by bacteria based on blood cell counts and interview findings. METHODS:  ML was performed using a decision tree classifier based on data from previously published papers. We included 164 children between one and 108 months diagnosed with gastroenteritis, with 112 having bacterial AG and 52 having viral AG as subjects and controls. Feature selection was performed using least absolute shrinkage and selection operator (LASSO), and the classifier's performance was evaluated by five-fold cross-validation. Additionally, we presented a tree diagram of the decision tree classifier as a flowchart for practical applications. RESULTS:  The area under curve (AUC) was 0.80, indicating a moderate model. Three important features in this model were platelet-lymphocyte ratio, eosinophil count, and leukocyte count. CONCLUSIONS:  In conclusion, this study demonstrates that bacterial AG can be estimated from blood cell counts with moderate accuracy. These findings may be valuable in narrowing down bacterial AG in children with gastrointestinal symptoms.

Laboratory Findings of Benign Convulsions With Mild Gastroenteritis: A Meta-Analysis.

Investigating factors associated with benign convulsions with mild gastroenteritis (CwG) is important for early detection and treatment. In previous studies, uric acid (UA) has been reported to be associated with CwG. However, the association between CwG and abnormal laboratory values remains inconclusive. We performed a meta-analysis of recent reports to determine the association between CwG and laboratory findings, including UA, in patients with acute gastroenteritis without convulsions. We conducted electronic searches of three databases (PubMed, EMBASE, and Cochrane Library) and one scholarly search engine (Google Scholar (Google, Inc., Mountain View, CA, USA)) up to February 2023 for studies on CwG. Eligible studies were observational studies that assessed patients with CwG, reported laboratory data, and stated the presence or absence of convulsions during illness episodes. Patients were children with mild gastroenteritis, with the exposure group developing convulsions and the control group not. The outcome was a comparison of laboratory data between the two groups. The effect size was calculated using the standardized mean difference (SMD), and random-effects models were used for the analysis because of high heterogeneity. In total, 148 articles were included in this study. After the screening, nine studies, including 8,367 patients, were selected for the meta-analysis. The most prevalent laboratory finding was an increased serum UA level, with an SMD of 1.42 (N = 6,411; 95% confidence interval (CI): (1.12, 1.72); Z = 9.242, p< 0.001; I 2 = 81.68%, p= 0.002). The optimal serum UA cutoff value was 7.21 mg/dL, with an area under the receiver operating characteristic (ROC) curve (AUC) of 0.827 (95% CI: (0.807, 0.846)). This meta-analysis suggests that CwG is strongly associated with increased serum UA levels. These results demonstrate that more attention should be paid when interpreting laboratory findings in pediatric patients with acute gastroenteritis.

Identifying Group A Streptococcal Pharyngitis in Children Through Clinical Variables Using Machine Learning.

Background Group A Streptococcus (GAS) is the most common bacterial cause of pharyngitis in children. GAS pharyngitis requires antimicrobial agents, and rapid antigen detection tests (RADTs) are currently considered useful for diagnosis. However, the decision to perform the test is based on the pediatrician's examination findings, but the indicators are not clear. Therefore, we used machine learning (ML) to create a model to identify GAS pharyngitis from clinical findings and to explore important features. Methods ML with Python programming language was used for this study. Data from the included study involved 676 children aged 3 to 15 years diagnosed with pharyngitis, with positive results on the RADT serving as exposures, and negative results serving as controls. The ML performances served as the outcome. We utilized six types of ML classifiers, namely, logistic regression, support vector machine, k-nearest neighbor algorithm, random forest, an ensemble of them, Voting Classifier, and the eXtreme Gradient Boosting (XGBoost) algorithm. Additionally, we used SHapley Additive exPlanations (SHAP) values to identify important features. Results Moderately performing models were generated for all six ML classifiers. XGBoost produced the best model, with an area under the receiver operating characteristics curve of 0.75 ± 0.01. The order of important features in the model was palatal petechiae, followed by scarlatiniform rash, tender cervical lymph nodes, and age. Conclusion Through this study, we have demonstrated that ML models can predict childhood GAS pharyngitis with moderate accuracy using only commonly recorded clinical variables in children diagnosed with pharyngitis. We have also identified four important clinical variables. These findings may serve as a reference for considering indicators under the current guidelines recommended for selective RADTs.

Hyponatremia and Recurrent Febrile Seizures During Febrile Episodes: A Meta-Analysis.

Several studies have investigated the potential effects of hyponatremia on recurrent febrile seizures (RFS) during febrile illness. Because findings were inconsistent across studies, we aimed to evaluate the serum sodium levels in febrile seizures (FS) of children with or without RFS during the same episode. We conducted electronic searches in three databases (PubMed, EMBASE, Cochrane Library) and one scholarly search engine (Google Scholar) up to June 2021 for studies on FS. Screening was done based on the titles and abstracts of primary studies. Then, eligibility was reviewed based on the abstracts. Finally, in order to match the inclusion and exclusion criteria, full-text articles were evaluated by two authors and inconsistencies were discussed. Data extraction was carried out by two independent authors. The extracted variables were author's name, article title, journal name, year of publication, study location, study design, sample size, and mean and standard deviation of blood Na concentration in FS. We performed a risk of bias assessment of included studies using the Newcastle-Ottawa Scale (NOS). The effect size was calculated using the standardized mean difference (SMD), and random-effects models were used for the analysis. A total of 12 articles were included with a single outlier. This analysis suggested that serum sodium level was lower in patients with RFS during the same febrile episode than in those with single FS, with SMD of -0.70, (n=1784; 95% CI: -1.03, -0.36; Z=-4.10, p<0.01; I 2 86.67%, p<0.01). In the sensitivity analysis, no significant change was observed in pooled SMD. The optimal cutoff value of serum sodium level was 134.72 mmol/L with an area under the receiver operating characteristic curve of 0.81 (95% CI: 0.61, 1.00), with sensitivity of 80.0% and specificity of 70.0%. This result indicated a significant association between hyponatremia and RFS during the same febrile episode. Decreased serum sodium levels may be involved in seizure recurrence and may play a role in FS pathogenesis.

Childhood-Onset Multifocal Motor Neuropathy With Immunoglobulin M Antibodies to Gangliosides GM1 and GM2: A Case Report and Review of the Literature.

Multifocal motor neuropathy is a rare immune-mediated neuropathy characterized by progressive asymmetric weakness and atrophy without sensory abnormalities. Although disease onset is usually in adulthood, a few childhood-onset cases have been reported. Here, we report the case of an 8-year-old boy with multifocal motor neuropathy who presented with a slowly progressive left and distal upper limb weakness without sensory loss. The initial high-dose intravenous immunoglobulin treatment significantly improved left upper limb muscle weakness. Continued monthly intravenous immunoglobulin treatment gradually improved muscle strength for several months initially. While the muscle strength decreased slightly after 8 months of therapy, it was better than that before intravenous immunoglobulin treatment. One year and eight months after the initiation of treatment, serum testing for IgM antibodies to gangliosides, GM1 and GM2, was negative. This is the first pediatric report of the serum IgM autoantibodies positive to GM1 and GM2. The clinical course is similar to that of partial intravenous immunoglobulin responders among patients with adulthood-onset multifocal motor neuropathy. Since the symptoms plateaued after the initial intravenous immunoglobulin therapy, prognosis appears to be determined by the patient's initial response to intravenous immunoglobulin treatment.