Comparison of machine learning and conventional logistic regression-based prediction models for gestational diabetes in an ethnically diverse population; the Monash GDM Machine learning model

Early gestational diabetes mellitus risk predictor using neural network with NearMiss

BACKGROUND

Gestational diabetes mellitus (GDM) is globally recognized as a significant pregnancy-related condition, contributing to complex complications for both mothers and infants. Traditional glucose tolerance tests lack the ability to identify the risk of GDM in early pregnancy, hindering effective prevention and timely intervention during the initial stages.

OBJECTIVE

The primary objective of this study is to pinpoint potential risk factors for GDM and develop an early GDM risk prediction model using neural networks to facilitate GDM screening in early pregnancy.

METHODS

Initially, we employed statistical tests and models, including univariate and multivariate logistic regression, to identify 14 potential risk factors. Subsequently, we applied various resampling techniques alongside a multi-layer perceptron (MLP). Finally, we evaluated and compared the classification performances of the constructed models using various metric indicators.

RESULTS

As a result, we identified several factors in early pregnancy significantly associated with GDM (p < 0.05), including BMI, age of menarche, age, higher education, folic acid supplementation, family history of diabetes mellitus, HGB, WBC, PLT, Scr, HBsAg, ALT, ALB, and TBIL. Employing the multivariate logistic model as the baseline achieved an accuracy and AUC of 0.777. In comparison, the MLP-based model using NearMiss exhibited strong predictive performance, achieving scores of 0.943 in AUC and 0.884 in accuracy.

CONCLUSIONS

In this study, we proposed an innovative interpretable early GDM risk prediction model based on MLP. This model is designed to offer assistance in estimating the risk of GDM in early pregnancy, enabling proactive prevention and timely intervention.

GDM-BC: Non-invasive body composition dataset for intelligent prediction of Gestational Diabetes Mellitus

Gestational Diabetes Mellitus (GDM) refers to any degree of impaired glucose tolerance with onset or first recognition during pregnancy. As a high-prevalence disease, GDM damages the health of both pregnant women and fetuses in the short and long term. Accurate and cost-effective recognition of GDM is quite crucial to reduce the risk and economic pressure of this disease. However, existing datasets for the prediction of GDM primarily focus on clinical and biochemical parameters, including a mass of invasive indexes. These variables are hard to obtain and do not always perform well in the prediction of GDM. In this paper, we introduce a large-scale non-invasive body composition dataset, called GDM-BC, for intelligent risk prediction of GDM. Specifically, it contains a cohort of 39,438 pregnant women, of whom 7777 (19.7%) were subsequently diagnosed with GDM. Besides, our dataset includes a large number of body composition indexes that can be acquired non-invasively. In addition, we perform several traditional machine learning and deep learning methods on the GDM-BC dataset, among which the Residual Attention Fully Connected Network (RAFNet) performs the best, achieving an AUC (area under the ROC curve) of 0.920. The results show that our dataset is marvelous and creates a new perspective on the prediction of GDM. Our models may offer an opportunity to establish a cost-effective screening approach for identifying low-risk pregnant women based on body composition data. We believe that our proposed GDM-BC dataset will advance future research on risk prediction for GDM, as well as provide new insights for intelligent prediction of other high-incidence pregnancy-related diseases such as gestational hypertension.

Exploring Primary and Interaction Effects of Minor Physical Anomalies: Development and Validation of Prediction Models Using Explainable Machine Learning Algorithms for Early-Onset Schizophrenia

BACKGROUND AND HYPOTHESIS

Minor physical abnormalities (MPAs) are neurodevelopmental markers that can be traced to prenatal events and may be significant features of early-onset schizophrenia (EOS). Therefore, our study aimed to (1) find the primary and interaction effects of MPAs for EOS and (2) develop and validate the model for EOS based on explainable machine learning algorithms.

STUDY DESIGN

The study included 549 patients with schizophrenia (193 EOS and 356 AOS) and 420 healthy controls (HC) in southern Taiwan. For the feature selection, variable selection using random forests (varSelRF) and recursive feature elimination (RFE) were applied to identify the important variables of MPAs. We used different machine learning algorithms to build the prediction models based on the selected MPAs variables.

STUDY RESULTS

The results showed that the mouth anomalies are significant MPAs variables and have interaction effects with craniofacial MPAs variables for EOS. The prediction models using the selected MPAs variables performed better in discriminating EOS vs HC compared to AOS vs HC. The AUC values for distinguishing EOS vs HC were 0.85-0.93, AOS vs HC were 0.80-0.87, and EOS vs AOS were 0.67-0.77 in validation sets.

CONCLUSIONS

This risk prediction model provides a clinical decision support system for detecting patients at high risk of developing EOS and enables early intervention in clinical practice.

Analyzing electronic medical records to extract prepregnancy morbidities and pregnancy complications: Toward a learning health system

Introduction

Preexisting and pregnancy-related medical conditions frequently co-occur, leading to multimorbidity (≥2 morbidities) in pregnant women, and much of this information is in semi-structured format in electronic medical records (EMRs). The aim was to advance the learning health system as a platform for automating information extraction from EMRs and to uncover the prevalence of common morbidities during pregnancy and their association with pregnancy-related complications.

Methods

This study included 48 502 pregnant women attending Monash Health maternity hospitals from 2016 to 2021. Natural language processing (NLP) was used to extract morbidities from semi-structured text in EMRs. Chi-squared tests were used to assess the association between morbidities of gestational diabetes mellitus (GDM) and other pregnancy complications. The k-means clustering algorithm identified clusters of comorbid conditions associated with GDM.

Results

The most common comorbidities during pregnancy were vitamin deficiency (14 019; 28.9%), overweight (13 918; 28.7%), obesity (11 026; 22.7%), anemia and other blood-related disorders (4821; 9.9%), mental health disorders (4314; 9.8%), asthma (4126; 8.5%), thyroid diseases (3576; 7.4%), endometrial disease (1927; 3.9%), cardiovascular disease (1525; 3.1%), and polycystic ovary syndrome (PCOS) (1464; 3.0%). While 22.5% of women had no medical conditions, 77.5% had one or more. Multimorbidity was associated with conditions including overweight, obesity, vitamin deficiency, thyroid disease, substance use, PCOS, GDM, and endometrial diseases. On cluster analysis, aged 35 years or older, overweight, vitamin deficiency, obesity, thyroid disease, asthma, uterine disease, other blood disorders, mental disorders, and PCOS were associated with GDM.

Conclusions

More than three-quarters of pregnant women in the Australian urban setting experienced one or more morbidities during pregnancy, which can be associated with adverse pregnancy outcomes. This project contributes to developing a learning health system infrastructure to deliver high-value maternal health care while reducing costs.

Machine learning based model for the early detection of Gestational Diabetes Mellitus

BACKGROUND

Gestational Diabetes Mellitus (GDM) is one of the most common medical complications during pregnancy. In the Gulf region, the prevalence of GDM is higher than in other parts of the world. Thus, there is a need for the early detection of GDM to avoid critical health conditions in newborns and post-pregnancy complexities of mothers.

METHODS

In this article, we propose a machine learning (ML)-based techniques for early detection of GDM. For this purpose, we considered clinical measurements taken during the first trimester to predict the onset of GDM in the second trimester.

RESULTS

The proposed ensemble-based model achieved high accuracy in predicting the onset of GDM with around 89% accuracy using only the first trimester data. We confirmed biomarkers, i.e., a history of high glucose level/diabetes, insulin and cholesterol, which align with the previous studies. Moreover, we proposed potential novel biomarkers such as HbA1C %, Glucose, MCH, NT pro-BNP, HOMA-IR- (22.5 Scale), HOMA-IR- (405 Scale), Magnesium, Uric Acid. C-Peptide, Triglyceride, Urea, Chloride, Fibrinogen, MCHC, ALT, family history of Diabetes, Vit B12, TSH, Potassium, Alk Phos, FT4, Homocysteine Plasma LC-MSMS, Monocyte Auto.

CONCLUSION

We believe our findings will complement the current clinical practice of GDM diagnosis at an early stage of pregnancy, leading toward minimizing its burden on the healthcare system.Source code is available in GitHub at: https://github.com/H-Zaky/GD.git.

Comparison of Logistic Regression and Machine Learning Approaches in Predicting Depressive Symptoms: A National-Based Study

OBJECTIVE

Machine learning (ML) has been reported to have better predictive capability than traditional statistical techniques. The aim of this study was to assess the efficacy of ML algorithms and logistic regression (LR) for predicting depressive symptoms during the COVID-19 pandemic.

METHODS

Analyses were carried out in a national cross-sectional study involving 21,916 participants. The ML algorithms in this study included random forest (RF), support vector machine (SVM), neural network (NN), and gradient boosting machine (GBM) methods. The performance indices were sensitivity, specificity, accuracy, precision, F1-score, and area under the receiver operating characteristic curve (AUC).

RESULTS

LR and NN had the best performance in terms of AUCs. The risk of overfitting was found to be negligible for most ML models except for RF, and GBM obtained the highest sensitivity, specificity, accuracy, precision, and F1-score. Therefore, LR, NN, and GBM models ranked among the best models.

CONCLUSION

Compared with ML models, LR model performed comparably to ML models in predicting depressive symptoms and identifying potential risk factors while also exhibiting a lower risk of overfitting.

Predicting low density lipoprotein cholesterol target attainment using machine learning in patients with coronary artery disease receiving moderate-dose statin therapy

Low-density lipoprotein cholesterol (LDL-C) is an important factor in the development of cardiovascular disease, making its management a key aspect of cardiovascular health. While high-dose statin therapy is often recommended for LDL-C reduction, careful consideration is needed due to patient-specific factors and potential side effects. This study aimed to develop a machine learning (ML) model to estimate the likelihood of achieving target LDL-C levels in patients hospitalized for coronary artery disease and treated with moderate-dose statins. The predictive performance of three ML models, including Extreme Gradient Boosting (XGBoost), Random Forest, and Logistic Regression, was evaluated using electronic medical records from the Asan Medical Center in Seoul across six performance metrics. Additionally, all three models achieved an average AUROC of 0.695 despite reducing features by over 43%. SHAP analysis was conducted to identify key features influencing model predictions, aiming insights into patient characteristics associated with achieving LDL-C targets. This study suggests that ML-based approaches may help identify patients likely to benefit from moderate-dose statins, potentially supporting personalized treatment strategies and clinical decision-making for LDL-C management.

First-Trimester Prediction Models Based on Maternal Characteristics for Adverse Pregnancy Outcomes: A Systematic Review and Meta-Analysis

BACKGROUND

Early risk stratification can facilitate timely interventions for adverse pregnancy outcomes, including preeclampsia (PE), small-for-gestational-age neonates (SGA), spontaneous preterm birth (sPTB) and gestational diabetes mellitus (GDM).

OBJECTIVES

To perform a systematic review and meta-analysis of first-trimester prediction models for adverse pregnancy outcomes.

SEARCH STRATEGY

The PubMed database was searched until 6 June 2024.

SELECTION CRITERIA

First-trimester prediction models based on maternal characteristics were included. Articles reporting on prediction models that comprised biochemical or ultrasound markers were excluded.

DATA COLLECTION AND ANALYSIS

Two authors identified articles, extracted data and assessed risk of bias and applicability using PROBAST.

MAIN RESULTS

A total of 77 articles were included, comprising 30 developed models for PE, 15 for SGA, 11 for sPTB and 35 for GDM. Discriminatory performance in terms of median area under the curve (AUC) of these models was 0.75 [IQR 0.69-0.78] for PE models, 0.62 [0.60-0.71] for SGA models of nulliparous women, 0.74 [0.72-0.74] for SGA models of multiparous women, 0.65 [0.61-0.67] for sPTB models of nulliparous women, 0.71 [0.68-0.74] for sPTB models of multiparous women and 0.71 [0.67-0.76] for GDM models. Internal validation was performed in 40/91 (43.9%) of the models. Model calibration was reported in 21/91 (23.1%) models. External validation was performed a total of 96 times in 45/91 (49.5%) of the models. High risk of bias was observed in 94.5% of the developed models and in 58.3% of the external validations.

CONCLUSIONS

Multiple first-trimester prediction models are available, but almost all suffer from high risk of bias, and internal and external validations were often not performed. Hence, methodological quality improvement and assessment of the clinical utility are needed.

Development and validation of a prediction model for coronary heart disease risk in depressed patients aged 20 years and older using machine learning algorithms

Background

Depression is being increasingly acknowledged as an important risk factor contributing to coronary heart disease (CHD). Currently, there is no predictive model specifically designed to evaluate the risk of coronary heart disease among individuals with depression. We aim to develop a machine learning (ML) model that will analyze risk factors and forecast the probability of coronary heart disease in individuals suffering from depression.

Methods

This research employed data from the National Health and Nutrition Examination Survey (NHANES) from 2007-2018, which included 2,085 individuals who had previously been diagnosed with depression. The population was randomly divided into a training set and a validation set, with an 8:2 ratio. Univariate and multivariate logistic regression analyses were employed to identify independent risk factors for coronary heart disease in individuals with depression. Eight machine learning algorithms were applied to the training set to construct the model, including logistic regression (LR), random forest (RF), gradient boosting machine (GBM), support vector machine (SVM), extreme gradient boosting (XGBoost), classification and regression tree (CART), k-nearest neighbors (KNN), and neural network (NNET). The validation set are used to evaluate the various performances of eight machine learning models. Several evaluation metrics were employed to assess and compare the performance of eight different machine learning models, aiming to identify the most effective algorithm for predicting coronary heart disease risk in individuals with depression. The evaluation metrics applied in this study included the area under the receiver operating characteristic (ROC) curve, calibration curve, Brier scores, decision curve analysis (DCA), and the precision-recall (PR) curve. And internally validated by the bootstrap method.

Results

Univariate and multivariate logistic regression analyses identified age, chest pain status, history of myocardial infarction, serum triglyceride levels, and education level as independent predictors of coronary heart disease risk. Eight machine learning algorithms are applied to construct the models, among which the Random Forest model has the best performance, with an (Area Under Curve) AUC of 0.987 for the random forest model in the training set, and an AUC of 0.848 for the PR curve. In the validation set, the random forest model achieves an AUC of 0.996, and an AUC of 0.960 for the PR curve, which demonstrates an excellent discriminative ability. Calibration curves indicated high congruence between observed and predicted odds, with minimal Brier scores of 0.026 and 0.021 for the training, respectively, reinforcing the model's ability to discriminate. Set and validation set, respectively, reinforcing the model's predictive accuracy. DCA curves confirmed net benefits of the random forest model across. Furthermore, the AUC of the random forest model was 0.928 after internal validation by bootstrap method, indicating that its discriminative ability is good, and the model is useful for clinical assessment of the risk of coronary heart disease in depressed people.

Conclusion

The random forest algorithm exhibited the best predictive performance, potentially aiding clinicians in assessing the risk probabilities of coronary heart disease within this population.

Leveraging Shapley Additive Explanations for Feature Selection in Ensemble Models for Diabetes Prediction

Diabetes, a significant global health crisis, is primarily driven in India by unhealthy diets and sedentary lifestyles, with rapid urbanization amplifying these effects through convenience-oriented living and limited physical activity opportunities, underscoring the need for advanced preventative strategies and technology for effective management. This study integrates Shapley Additive explanations (SHAPs) into ensemble machine learning models to improve the accuracy and efficiency of diabetes predictions. By identifying the most influential features using SHAP, this study examined their role in maintaining high predictive performance while minimizing computational demands. The impact of feature selection on model accuracy was assessed across ten models using three feature sets: all features, the top three influential features, and all except these top three. Models focusing on the top three features achieved superior performance, with the ensemble model attaining a better performance in most of the metrics, outperforming comparable approaches. Notably, excluding these features led to a significant decline in performance, reinforcing their critical influence. These findings validate the effectiveness of targeted feature selection for efficient and robust clinical applications.

Dietary inflammatory index as a predictor of prediabetes in women with previous gestational diabetes mellitus

INTRODUCTION

Gestational diabetes mellitus (GDM) is associated with an increased risk of developing type 2 diabetes mellitus (T2DM). The inflammatory potential of diet is crucial in GDM development. This study compares dietary inflammatory indices (DII) in females with and without a history of GDM and constructs a predictive model for prediabetes risk.

METHODS

Cross-sectional data from NHANES cycles (2011-2014) were analyzed using the DII. Independent t tests, chi-square test, and Mann-Whitney U test examined DII scores in relation to GDM history. Multivariate logistic regression assessed DII's association with prediabetes in females with GDM history. Restricted cubic spline (RCS) and LASSO regression modeled non-linear relationships and predicted prediabetes risk.

RESULTS

971 female participants were included. Those with GDM history had lower DII scores (1.62 (0.58, 2.93) vs. 2.05 (0.91, 2.93)). Higher DII scores in females with GDM were linked to prediabetes, remaining significant after adjusting for confounders. RCS analysis found no non-linear correlation (non-linear p = 0.617). The prediabetes model for GDM history had strong predictive performance (AUC = 88.6%, 95% CI: 79.9-97.4%).

CONCLUSION

Females with GDM history show lower DII levels, potentially reflecting improved diet and health awareness. Higher DII scores correlate with increased prediabetes risk in this group, emphasizing diet's role in diabetes risk. Further studies are needed to confirm these findings.

Prediction models for cognitive frailty in community-dwelling older adults: A scoping review

OBJECTIVES

This review investigates the current status of cognitive frailty risk prediction models for community-dwelling older adults, aiming to explore the shortcomings and provide insights for model optimisation.

METHODS

We adhered to the PRISMA guidelines for scoping review and followed the Joanna Briggs Institute Manual for Evidence Synthesis.

RESULTS

This article includes a total of 10 studies, revealing a prevalence of cognitive frailty ranging from 4.8 % to 39.6 %. The methods used for model construction included both logistic regression and machine learning. The predictors varied across the models, with age, education level, gender, and physical activity level being the most frequently cited factors.

CONCLUSIONS

While most models showed good applicability, all models displayed a high risk of bias. Future endeavors should concentrate on leveraging existing tools to ensure standardization in development and conducting rigorous evaluations of prediction models for cognitive frailty in community-dwelling older adults.

Prediction of neurologic outcome after out-of-hospital cardiac arrest: An interpretable approach with machine learning

Out-of-hospital cardiac arrest (OHCA) is a critical condition with low survival rates. In patients with a return of spontaneous circulation, brain injury is a leading cause of death. In this study, we propose an interpretable machine learning approach for predicting neurologic outcome after OHCA, using information available at the time of hospital admission.

METHODS

The study population were 55 615 OHCA cases registered in the Swedish Cardiopulmonary Resuscitation Registry between 2010 and 2020. The dataset was split to training and validation sets (for model development) and test set (for evaluation of the final model). We used an XGBoost algorithm with stratified, repeated 10-fold cross-validation along with Optuna framework for hyperparameters tuning. The final model was trained on 10 features selected based on the importance scores and evaluated on the test set in terms of discrimination, calibration and bias-variance tradeoff. We used SHapley Additive exPlanations to address the 'black-box' model and align with eXplainable artificial intelligence.

RESULTS

The final model achieved: area under the receiver operating characteristic value 0.964 (95% confidence interval (CI) [0.960-0.968]), sensitivity 0.606 (95% CI [0.573-0.634]), specificity 0.975 (95% CI [0.972-0.978]), positive predictive value (PPV) 0.664 (95% CI [0.625-0.696]), negative predictive value (NPV) 0.969 (95% CI [0.966-0.972]), macro F1 0.803 (95% CI [0.788-0.816]), and showed a very good calibration. SHAP features with the highest impact on the model's output were:'ROSC on arrival to hospital', 'Initial rhythm asystole' and 'Conscious on arrival to hospital'.

CONCLUSIONS

The XGBoost machine learning model with 10 features available at the time of hospital admission showed good performance for predicting neurologic outcome after OHCA, with no apparent signs of overfitting.

The construction of machine learning-based predictive models for high-quality embryo formation in poor ovarian response patients with progestin-primed ovarian stimulation

OBJECTIVE

To explore the optimal models for predicting the formation of high-quality embryos in Poor Ovarian Response (POR) Patients with Progestin-Primed Ovarian Stimulation (PPOS) using machine learning algorithms.

METHODS

A retrospective analysis was conducted on the clinical data of 4,216 POR cycles who underwent in vitro fertilization (IVF) / intracytoplasmic sperm injection (ICSI) at Sichuan Jinxin Xinan Women and Children's Hospital from January 2015 to December 2021. Based on the presence of high-quality cleavage embryos 72 h post-fertilization, the samples were divided into the high-quality cleavage embryo group (N = 1950) and the non-high-quality cleavage embryo group (N = 2266). Additionally, based on whether high-quality blastocysts were observed following full blastocyst culture, the samples were categorized into the high-quality blastocyst group (N = 124) and the non-high-quality blastocyst group (N = 1800). The factors influencing the formation of high-quality embryos were analyzed using logistic regression. The predictive models based on machine learning methods were constructed and evaluated accordingly.

RESULTS

Differential analysis revealed that there are statistically significant differences in 14 factors between high-quality and non-high-quality cleavage embryos. Logistic regression analysis identified 14 factors as influential in forming high-quality cleavage embryos. In models excluding three variables (retrieved oocytes, MII oocytes, and 2PN fertilized oocytes), the XGBoost model performed slightly better (AUC = 0.672, 95% CI = 0.636-0.708). Conversely, in models including these three variables, the Random Forest model exhibited the best performance (AUC = 0.788, 95% CI = 0.759-0.818). In the analysis of high-quality blastocysts, significant differences were found in 17 factors. Logistic regression analysis indicated that 13 factors influence the formation of high-quality blastocysts. Including these variables in the predictive model, the XGBoost model showed the highest performance (AUC = 0.813, 95% CI = 0.741-0.884).

CONCLUSION

We developed a predictive model for the formation of high-quality embryos using machine learning methods for patients with POR undergoing treatment with the PPOS protocol. This model can help infertility patients better understand the likelihood of forming high-quality embryos following treatment and help clinicians better understand and predict treatment outcomes, thus facilitating more targeted and effective interventions.

The role of machine learning algorithms in detection of gestational diabetes; a narrative review of current evidence

Gestational Diabetes Mellitus (GDM) poses significant health risks to mothers and infants. Early prediction and effective management are crucial to improving outcomes. Machine learning techniques have emerged as powerful tools for GDM prediction. This review compiles and analyses the available studies to highlight key findings and trends in the application of machine learning for GDM prediction. A comprehensive search of relevant studies published between 2000 and September 2023 was conducted. Fourteen studies were selected based on their focus on machine learning for GDM prediction. These studies were subjected to rigorous analysis to identify common themes and trends. The review revealed several key themes. Models capable of predicting GDM risk during the early stages of pregnancy were identified from the studies reviewed. Several studies underscored the necessity of tailoring predictive models to specific populations and demographic groups. These findings highlighted the limitations of uniform guidelines for diverse populations. Moreover, studies emphasised the value of integrating clinical data into GDM prediction models. This integration improved the treatment and care delivery for individuals diagnosed with GDM. While different machine learning models showed promise, selecting and weighing variables remains complex. The reviewed studies offer valuable insights into the complexities and potential solutions in GDM prediction using machine learning. The pursuit of accurate, early prediction models, the consideration of diverse populations, clinical data, and emerging data sources underscore the commitment of researchers to improve healthcare outcomes for pregnant individuals at risk of GDM.

Knowledge framework of intravenous immunoglobulin resistance in the field of Kawasaki disease: A bibliometric analysis (1997-2023)

BACKGROUND

Kawasaki disease (KD) is an autoimmune disease with cardiovascular disease as its main complication, mainly affecting children under 5 years old. KD treatment has made tremendous progress in recent years, but intravenous immunoglobulin (IVIG) resistance remains a major dilemma. Bibliometric analysis had not been used previously to summarize and analyze publications related to IVIG resistance in KD. This study aimed to provide an overview of the knowledge framework and research hotspots in this field through bibliometrics, and provide references for future basic and clinical research.

METHODS

Through bibliometric analysis of relevant literature published on the Web of Science Core Collection (WoSCC) database between 1997 and 2023, we investigated the cooccurrence and collaboration relationships among countries, institutions, journals, and authors and summarized key research topics and hotspots.

RESULTS

Following screening, a total of 364 publications were downloaded, comprising 328 articles and 36 reviews. The number of articles on IVIG resistance increased year on year and the top three most productive countries were China, Japan, and the United States. Frontiers in Pediatrics had the most published articles, and the Journal of Pediatrics had the most citations. IVIG resistance had been studied by 1889 authors, of whom Kuo Ho Chang had published the most papers.

CONCLUSION

Research in the field was focused on risk factors, therapy (atorvastatin, tumor necrosis factor-alpha inhibitors), pathogenesis (gene expression), and similar diseases (multisystem inflammatory syndrome in children, MIS-C). "Treatment," "risk factor," and "prediction" were important keywords, providing a valuable reference for scholars studying this field. We suggest that, in the future, more active international collaborations are carried out to study the pathogenesis of IVIG insensitivity, using high-throughput sequencing technology. We also recommend that machine learning techniques are applied to explore the predictive variables of IVIG resistance.

Development and validation of machine learning-based models for predicting healthcare-associated bacterial/fungal infections among COVID-19 inpatients: a retrospective cohort study

BACKGROUND

COVID-19 and bacterial/fungal coinfections have posed significant challenges to human health. However, there is a lack of good tools for predicting coinfection risk to aid clinical work.

OBJECTIVE

We aimed to investigate the risk factors for bacterial/fungal coinfection among COVID-19 patients and to develop machine learning models to estimate the risk of coinfection.

METHODS

In this retrospective cohort study, we enrolled adult inpatients confirmed with COVID-19 in a tertiary hospital between January 1 and July 31, 2023, in China and collected baseline information at admission. All the data were randomly divided into a training set and a testing set at a ratio of 7:3. We developed the generalized linear and random forest models for coinfections in the training set and assessed the performance of the models in the testing set. Decision curve analysis was performed to evaluate the clinical applicability.

RESULTS

A total of 1244 patients were included in the training cohort with 62 healthcare-associated bacterial/fungal infections, while 534 were included in the testing cohort with 22 infections. We found that patients with comorbidities (diabetes, neurological disease) were at greater risk for coinfections than were those without comorbidities (OR = 2.78, 95%CI = 1.61-4.86; OR = 1.93, 95%CI = 1.11-3.35). An indwelling central venous catheter or urinary catheter was also associated with an increased risk (OR = 2.53, 95%CI = 1.39-4.64; OR = 2.28, 95%CI = 1.24-4.27) of coinfections. Patients with PCT > 0.5 ng/ml were 2.03 times (95%CI = 1.41-3.82) more likely to be infected. Interestingly, the risk of coinfection was also greater in patients with an IL-6 concentration < 10 pg/ml (OR = 1.69, 95%CI = 0.97-2.94). Patients with low baseline creatinine levels had a decreased risk of bacterial/fungal coinfections(OR = 0.40, 95%CI = 0.22-0.71). The generalized linear and random forest models demonstrated favorable receiver operating characteristic curves (ROC = 0.87, 95%CI = 0.80-0.94; ROC = 0.88, 95%CI = 0.82-0.93) with high accuracy, sensitivity and specificity of 0.86vs0.75, 0.82vs0.86, 0.87vs0.74, respectively. The corresponding calibration evaluation P statistics were 0.883 and 0.769.

CONCLUSIONS

Our machine learning models achieved strong predictive ability and may be effective clinical decision-support tools for identifying COVID-19 patients at risk for bacterial/fungal coinfection and guiding antibiotic administration. The levels of cytokines, such as IL-6, may affect the status of bacterial/fungal coinfection.

Routine screening for gestational diabetes: a review

PURPOSE OF REVIEW

Rates of gestational diabetes mellitus (GDM) throughout the world continue to increase associated with the increasing rates of obesity. Given this epidemiologic burden, the importance of proper screening, diagnosis, and management cannot be understated. This review focuses on the current screening guidelines utilized throughout the world and new data recently published regarding the most optimal screening techniques and future directions for research.

RECENT FINDINGS

Despite unanimous opinion that GDM warrants screening, the optimal screening regimen remains controversial. Notably, in the United States per the consensus recommendation by the American College of Obstetrics and Gynecology and the Society for Maternal-Fetal Medicine, a 2-step screening approach is often used. Recently, there have been multiple studies published that have compared the 1-step and 2-step screening process with respect to GDM incidence and perinatal outcomes. These new findings are summarized below.

SUMMARY

Utilization of the 1-step screening as opposed to the 2-step screening results in an increased diagnosis of GDM without significant population level benefit in outcomes. However, these studies remain underpowered to allow for meaningful comparison of outcomes in those diagnosed with GDM.