Comparison of logistic regression with machine learning methods for the prediction of fetal growth abnormalities: a retrospective cohort study

Multivariable prediction models for fetal macrosomia and large for gestational age: A systematic review

BACKGROUND

The identification of large for gestational age (LGA) and macrosomic fetuses is essential for counselling and managing these pregnancies.

OBJECTIVES

To systematically review the literature for multivariable prediction models for LGA and macrosomia, assessing the performance, quality and applicability of the included model in clinical practice.

SEARCH STRATEGY

MEDLINE, EMBASE and Cochrane Library were searched until June 2022.

SELECTION CRITERIA

We included observational and experimental studies reporting the development and/or validation of any multivariable prediction model for fetal macrosomia and/or LGA. We excluded studies that used a single variable or did not evaluate model performance.

DATA COLLECTION AND ANALYSIS

Data were extracted using the Checklist for critical appraisal and data extraction for systematic reviews of prediction modelling studies checklist. The model performance measures discrimination, calibration and validation were extracted. The quality and completion of reporting within each study was assessed by its adherence to the TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis) checklist. The risk of bias and applicability were measured using PROBAST (Prediction model Risk Of Bias Assessment Tool).

MAIN RESULTS

A total of 8442 citations were identified, with 58 included in the analysis: 32/58 (55.2%) developed, 21/58 (36.2%) developed and internally validated and 2/58 (3.4%) developed and externally validated a model. Only three studies externally validated pre-existing models. Macrosomia and LGA were differentially defined by many studies. In total, 111 multivariable prediction models were developed using 112 different variables. Model discrimination was wide ranging area under the receiver operating characteristics curve (AUROC 0.56-0.96) and few studies reported calibration (11/58, 19.0%). Only 5/58 (8.6%) studies had a low risk of bias.

CONCLUSIONS

There are currently no multivariable prediction models for macrosomia/LGA that are ready for clinical implementation.

Machine learning predictive modeling for condemnation risk assessment in antibiotic-free raised broilers

The condemnation of broiler carcasses in the poultry industry is a major challenge and leads to significant financial losses and food waste. This study addresses the critical issue of condemnation risk assessment in the discarding of antibiotic-free raised broilers using machine learning (ML) predictive modeling. In this study, ML approaches, specifically least absolute shrinkage and selection operator (LASSO), classification tree (CT), and random forests (RF), were used to evaluate and compare their effectiveness in predicting high condemnation rates. The dataset of 23,959 truckloads from 2021 to 2022 contained 14 independent variables covering the rearing, catching, transportation, and slaughtering phases. Condemnation rates between 0.26% and 25.99% were used as the dependent variable for the analysis, with the threshold for a high conviction rate set at 3.0%. As high condemnation rates were in the minority (8.05%), sampling methods such as random over sampling (ROS), random under sampling (RUS), both sampling (BOTH), and random over sampling example (ROSE) were used to account for imbalanced datasets. The results showed that RF with RUS performed better than the other models for balanced datasets. In this study, mean body weight, weight per crate, mortality and culling rates, and lairage time were identified as the 4 most important variables for predicting high condemnation rates. This study provides valuable insights into ML applications for predicting condemnation rates in antibiotic-free raised broilers and provides a framework to improve decision-making processes in establishing farm management practices to minimize economic losses in the poultry industry. The proposed methods are adaptable for different broiler producers, which increases their applicability in the industry.

Development and validation of a prognostic model to predict birth weight: individual participant data meta-analysis

Objective

To predict birth weight at various potential gestational ages of delivery based on data routinely available at the first antenatal visit.

Design

Individual participant data meta-analysis.

Data sources

Individual participant data of four cohorts (237 228 pregnancies) from the International Prediction of Pregnancy Complications (IPPIC) network dataset.

Eligibility criteria for selecting studies

Studies in the IPPIC network were identified by searching major databases for studies reporting risk factors for adverse pregnancy outcomes, such as pre-eclampsia, fetal growth restriction, and stillbirth, from database inception to August 2019. Data of four IPPIC cohorts (237 228 pregnancies) from the US (National Institute of Child Health and Human Development, 2018; 233 483 pregnancies), UK (Allen et al, 2017; 1045 pregnancies), Norway (STORK Groruddalen research programme, 2010; 823 pregnancies), and Australia (Rumbold et al, 2006; 1877 pregnancies) were included in the development of the model.

Results

The IPPIC birth weight model was developed with random intercept regression models with backward elimination for variable selection. Internal-external cross validation was performed to assess the study specific and pooled performance of the model, reported as calibration slope, calibration-in-the-large, and observed versus expected average birth weight ratio. Meta-analysis showed that the apparent performance of the model had good calibration (calibration slope 0.99, 95% confidence interval (CI) 0.88 to 1.10; calibration-in-the-large 44.5 g, -18.4 to 107.3) with an observed versus expected average birth weight ratio of 1.02 (95% CI 0.97 to 1.07). The proportion of variation in birth weight explained by the model (R2) was 46.9% (range 32.7-56.1% in each cohort). On internal-external cross validation, the model showed good calibration and predictive performance when validated in three cohorts with a calibration slope of 0.90 (Allen cohort), 1.04 (STORK Groruddalen cohort), and 1.07 (Rumbold cohort), calibration-in-the-large of -22.3 g (Allen cohort), -33.42 (Rumbold cohort), and 86.4 g (STORK Groruddalen cohort), and observed versus expected ratio of 0.99 (Rumbold cohort), 1.00 (Allen cohort), and 1.03 (STORK Groruddalen cohort); respective pooled estimates were 1.00 (95% CI 0.78 to 1.23; calibration slope), 9.7 g (-154.3 to 173.8; calibration-in-the-large), and 1.00 (0.94 to 1.07; observed v expected ratio). The model predictions were more accurate (smaller mean square error) in the lower end of predicted birth weight, which is important in informing clinical decision making.

Conclusions

The IPPIC birth weight model allowed birth weight predictions for a range of possible gestational ages. The model explained about 50% of individual variation in birth weights, was well calibrated (especially in babies at high risk of fetal growth restriction and its complications), and showed promising performance in four different populations included in the individual participant data meta-analysis. Further research to examine the generalisability of performance in other countries, settings, and subgroups is required.

Trial registration

PROSPERO CRD42019135045.

Repeated Sieving for Prediction Model Building with High-Dimensional Data

Background: The prediction of patients' outcomes is a key component in personalized medicine. Oftentimes, a prediction model is developed using a large number of candidate predictors, called high-dimensional data, including genomic data, lab tests, electronic health records, etc. Variable selection, also called dimension reduction, is a critical step in developing a prediction model using high-dimensional data. Methods: In this paper, we compare the variable selection and prediction performance of popular machine learning (ML) methods with our proposed method. LASSO is a popular ML method that selects variables by imposing an L1-norm penalty to the likelihood. By this approach, LASSO selects features based on the size of regression estimates, rather than their statistical significance. As a result, LASSO can miss significant features while it is known to over-select features. Elastic net (EN), another popular ML method, tends to select even more features than LASSO since it uses a combination of L1- and L2-norm penalties that is less strict than an L1-norm penalty. Insignificant features included in a fitted prediction model act like white noises, so that the fitted model will lose prediction accuracy. Furthermore, for the future use of a fitted prediction model, we have to collect the data of all the features included in the model, which will cost a lot and possibly lower the accuracy of the data if the number of features is too many. Therefore, we propose an ML method, called repeated sieving, extending the standard regression methods with stepwise variable selection. By selecting features based on their statistical significance, it resolves the over-selection issue with high-dimensional data. Results: Through extensive numerical studies and real data examples, our results show that the repeated sieving method selects far fewer features than LASSO and EN, but has higher prediction accuracy than the existing ML methods. Conclusions: We conclude that our repeated sieving method performs well in both variable selection and prediction, and it saves the cost of future investigation on the selected factors.

Development of a regional-based predictive model of incidence of traumatic spinal cord injury using machine learning algorithms

Objective

To develop a predictive model of incidence of traumatic spinal cord injury (TSCI).

Methods

The data for training the model included both the incidence data and the covariates. The incidence data were extracted from systematic reviews and the covariates were extracted from data available in the international road federation database. Then the feature processing measures were taken. First we defined a hyper-parameter, missing-value threshold, in order to eliminate features that exceed this threshold. To tackle the problem of overfitting of model we determined the Pearson correlation of features and excluded those with more than 0.7 correlation. After feature selection three different models including simple linear regression, support vector regression, and multi-layer perceptron were examined to fit the purposes of this study. Finally, we evaluated the model based on three standard metrics: Mean Absolute Error, Root Mean Square Error, and R2.

Results

Our machine-learning based model could predict the incidence rate of TSCI with the mean absolute error of 4.66. Our model found "Vehicles in use, Total vehicles/Km of roads", "Injury accidents/100 Million Veh-Km", "Vehicles in use, Vans, Pick-ups, Lorries, Road Tractors", "Inland surface Passengers Transport (Mio Passenger-Km), Rail", and "% paved" as top predictors of transport-related TSCI (TRTSCI).

Conclusions

Our model is proved to have a high accuracy to predict the incidence rate of TSCI for countries, especially where the main etiology of TSCI is related to road traffic injuries. Using this model, we can help the policymakers for resource allocation and evaluation of preventive measures.

Identifying First-Trimester Risk Factors for SGA-LGA Using Weighted Inheritance Voting Ensemble Learning

The classification of fetuses as Small for Gestational Age (SGA) and Large for Gestational Age (LGA) is a critical aspect of neonatal health assessment. SGA and LGA, terms used to describe fetal weights that fall below or above the expected weights for Appropriate for Gestational Age (AGA) fetuses, indicate intrauterine growth restriction and excessive fetal growth, respectively. Early prediction and assessment of latent risk factors associated with these classifications can facilitate timely medical interventions, thereby optimizing the health outcomes for both the infant and the mother. This study aims to leverage first-trimester data to achieve these objectives. This study analyzed data from 7943 pregnant women, including 424 SGA, 928 LGA, and 6591 AGA cases, collected from 2015 to 2021 at the Third Affiliated Hospital of Sun Yat-sen University in Guangzhou, China. We propose a novel algorithm, named the Weighted Inheritance Voting Ensemble Learning Algorithm (WIVELA), to predict the classification of fetuses into SGA, LGA, and AGA categories based on biochemical parameters, maternal factors, and morbidity during pregnancy. Additionally, we proposed algorithms for relevance determination based on the classifier to ascertain the importance of features associated with SGA and LGA. The proposed classification solution demonstrated a notable average accuracy rate of 92.12% on 10-fold cross-validation over 100 loops, outperforming five state-of-the-art machine learning algorithms. Furthermore, we identified significant latent maternal risk factors directly associated with SGA and LGA conditions, such as weight change during the first trimester, prepregnancy weight, height, age, and obstetric factors like fetal growth restriction and birthing LGA baby. This study also underscored the importance of biomarker features at the end of the first trimester, including HDL, TG, OGTT-1h, OGTT-0h, OGTT-2h, TC, FPG, and LDL, which reflect the status of SGA or LGA fetuses. This study presents innovative solutions for classifying and identifying relevant attributes, offering valuable tools for medical teams in the clinical monitoring of fetuses predisposed to SGA and LGA conditions during the initial stage of pregnancy. These proposed solutions facilitate early intervention in nutritional care and prenatal healthcare, thereby contributing to enhanced strategies for managing the health and well-being of both the fetus and the expectant mother.

Development and evaluation of machine learning models for predicting large-for-gestational-age newborns in women exposed to radiation prior to pregnancy

INTRODUCTION

The correlation between radiation exposure before pregnancy and abnormal birth weight has been previously proven. However, for large-for-gestational-age (LGA) babies in women exposed to radiation before becoming pregnant, there is no prediction model yet.

MATERIAL AND METHODS

The data were collected from the National Free Preconception Health Examination Project in China. A sum of 455 neonates (42 SGA births and 423 non-LGA births) were included. A training set (n = 319) and a test set (n = 136) were created from the dataset at random. To develop prediction models for LGA neonates, conventional logistic regression (LR) method and six machine learning methods were used in this study. Recursive feature elimination approach was performed by choosing 10 features which made a big contribution to the prediction models. And the Shapley Additive Explanation model was applied to interpret the most important characteristics that affected forecast outputs.

RESULTS

The random forest (RF) model had the highest average area under the receiver-operating-characteristic curve (AUC) for predicting LGA in the test set (0.843, 95% confidence interval [CI]: 0.714-0.974). Except for the logistic regression model (AUC: 0.603, 95%CI: 0.440-0.767), other models' AUCs displayed well. Thereinto, the RF algorithm's final prediction model using 10 characteristics achieved an average AUC of 0.821 (95% CI: 0.693-0.949).

CONCLUSION

The prediction model based on machine learning might be a promising tool for the prenatal prediction of LGA births in women with radiation exposure before pregnancy.

Predicting Chronic Pain and Treatment Outcomes Using Machine Learning Models Based on High-dimensional Clinical Data From a Large Retrospective Cohort

PURPOSE

To identify factors and indicators that affect chronic pain and pain relief, and to develop predictive models using machine learning.

METHODS

We analyzed the data of 67,028 outpatient cases and 11,310 valid samples with pain from a large retrospective cohort. We used decision tree, random forest, AdaBoost, neural network, and logistic regression to discover significant indicators and to predict pain and treatment relief.

FINDINGS

The random forest model had the highest accuracy, F1 value, precision, and recall rates for predicting pain relief. The main factors affecting pain and treatment relief included body mass index, blood pressure, age, body temperature, heart rate, pulse, and neutrophil/lymphocyte × platelet ratio. The logistic regression model had high sensitivity and specificity for predicting pain occurrence.

IMPLICATIONS

Machine learning models can be used to analyze the risk factors and predictors of chronic pain and pain relief, and to provide personalized and evidence-based pain management.

Machine learning models on a web application to predict short-term postoperative outcomes following anterior cervical discectomy and fusion

BACKGROUND

The frequency of anterior cervical discectomy and fusion (ACDF) has increased up to 400% since 2011, underscoring the need to preoperatively anticipate adverse postoperative outcomes given the procedure's expanding use. Our study aims to accomplish two goals: firstly, to develop a suite of explainable machine learning (ML) models capable of predicting adverse postoperative outcomes following ACDF surgery, and secondly, to embed these models in a user-friendly web application, demonstrating their potential utility.

METHODS

We utilized data from the National Surgical Quality Improvement Program database to identify patients who underwent ACDF surgery. The outcomes of interest were four short-term postoperative adverse outcomes: prolonged length of stay (LOS), non-home discharges, 30-day readmissions, and major complications. We utilized five ML algorithms - TabPFN, TabNET, XGBoost, LightGBM, and Random Forest - coupled with the Optuna optimization library for hyperparameter tuning. To bolster the interpretability of our models, we employed SHapley Additive exPlanations (SHAP) for evaluating predictor variables' relative importance and used partial dependence plots to illustrate the impact of individual variables on the predictions generated by our top-performing models. We visualized model performance using receiver operating characteristic (ROC) curves and precision-recall curves (PRC). Quantitative metrics calculated were the area under the ROC curve (AUROC), balanced accuracy, weighted area under the PRC (AUPRC), weighted precision, and weighted recall. Models with the highest AUROC values were selected for inclusion in a web application.

RESULTS

The analysis included 57,760 patients for prolonged LOS [11.1% with prolonged LOS], 57,780 for non-home discharges [3.3% non-home discharges], 57,790 for 30-day readmissions [2.9% readmitted], and 57,800 for major complications [1.4% with major complications]. The top-performing models, which were the ones built with the Random Forest algorithm, yielded mean AUROCs of 0.776, 0.846, 0.775, and 0.747 for predicting prolonged LOS, non-home discharges, readmissions, and complications, respectively.

CONCLUSIONS

Our study employs advanced ML methodologies to enhance the prediction of adverse postoperative outcomes following ACDF. We designed an accessible web application to integrate these models into clinical practice. Our findings affirm that ML tools serve as vital supplements in risk stratification, facilitating the prediction of diverse outcomes and enhancing patient counseling for ACDF.

A deep learning framework for noninvasive fetal ECG signal extraction

Introduction: The availability of proactive techniques for health monitoring is essential to reducing fetal mortality and avoiding complications in fetal wellbeing. In harsh circumstances such as pandemics, earthquakes, and low-resource settings, the incompetence of many healthcare systems worldwide in providing essential services, especially for pregnant women, is critical. Being able to continuously monitor the fetus in hospitals and homes in a direct and fast manner is very important in such conditions. Methods: Monitoring the health of the baby can potentially be accomplished through the computation of vital bio-signal measures using a clear fetal electrocardiogram (ECG) signal. The aim of this study is to develop a framework to detect and identify the R-peaks of the fetal ECG directly from a 12 channel abdominal composite signal. Thus, signals were recorded noninvasively from 70 pregnant (healthy and with health conditions) women with no records of fetal abnormalities. The proposed model employs a recurrent neural network architecture to robustly detect the fetal ECG R-peaks. Results: To test the proposed framework, we performed both subject-dependent (5-fold cross-validation) and independent (leave-one-subject-out) tests. The proposed framework achieved average accuracy values of 94.2% and 88.8%, respectively. More specifically, the leave-one-subject-out test accuracy was 86.7% during the challenging period of vernix caseosa layer formation. Furthermore, we computed the fetal heart rate from the detected R-peaks, and the demonstrated results highlight the robustness of the proposed framework. Discussion: This work has the potential to cater to the critical industry of maternal and fetal healthcare as well as advance related applications.

Application of machine learning algorithms to predict dead on arrival of broiler chickens raised without antibiotic program

Understanding the factors of dead-on-arrival (DOA) incidents during pre-slaughter handling is crucial for informed decision-making, improving broiler welfare, and optimizing farm profitability. In this study, 3 different machine learning (ML) algorithms - least absolute shrinkage and selection operator (LASSO), classification tree (CT), and random forest (RF) - were used together with 4 sampling techniques to optimize imbalanced data. The dataset comes from 22,115 broiler truckloads from a large producer in Thailand (2021-2022) and includes 14 independent variables covering the rearing, catching, and transportation stages. The study focuses on DOA% in the range of 0.10 to 1.20%, with a threshold for high DOA% above 0.3%, and records DOA% per truckload during pre-slaughter ante-mortem inspection. With a high DOA rate of 25.2%, the imbalanced dataset prompts the implementation of 4 methods to tune the imbalance parameters: random over sampling (ROS), random under sampling (RUS), both sampling (BOTH), and synthetic sampling or random over sampling example (ROSE). The aim is to improve the performance of the prediction model in classifying and predicting high DOA%. The comparative analysis of the different error metrics shows that RF outperforms the other models in a balanced dataset. In particular, RUS shows a significant improvement in prediction performance across all models compared to the original unbalanced dataset. The identification of the 4 most important variables for predicting high DOA percentages - mortality and culling rate, rearing stocking density, season, and mean body weight - emphasizes their importance for broiler production. This study provides valuable insights into the prediction of DOA status using an ML approach and contributes to the development of more effective strategies to mitigate high DOA percentages in commercial broiler production.

An artificial intelligence approach to predict infants' health status at birth

BACKGROUND

Machine learning could be used for prognosis/diagnosis of maternal and neonates' diseases by analyzing the data sets and profiles obtained from a pregnant mother.

PURPOSE

We aimed to develop a prediction model based on machine learning algorithms to determine important maternal characteristics and neonates' anthropometric profiles as the predictors of neonates' health status.

METHODS

This study was conducted among 1280 pregnant women referred to healthcare centers to receive antenatal care. We evaluated several machine learning methods, including support vector machine (SVM), Ensemble, K-Nearest Neighbor (KNN), Naïve Bayes (NB), and Decision tree classifiers, to predict newborn health state.

RESULTS

The minimum redundancy-maximum relevance (MRMR) algorithm revealed that variables, including head circumference of neonates, pregnancy intention, and drug consumption history during pregnancy, were top-scored features for classifying normal and unhealthy infants. Among the different classification methods, the SVM classifier had the best performance. The average values of accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC) in the test group were 75%, 75%, 76%, 76%, and 65%, respectively, for SVM model.

CONCLUSION

Machine learning methods can efficiently forecast the neonate's health status among pregnant women. This study proposed a new approach toward the integration of maternal data and neonate profiles to facilitate the prediction of neonates' health status.

Predicting risk of preterm birth in singleton pregnancies using machine learning algorithms

We aimed to develop, train, and validate machine learning models for predicting preterm birth (<37 weeks' gestation) in singleton pregnancies at different gestational intervals. Models were developed based on complete data from 22,603 singleton pregnancies from a prospective population-based cohort study that was conducted in 51 midwifery clinics and hospitals in Wenzhou City of China between 2014 and 2016. We applied Catboost, Random Forest, Stacked Model, Deep Neural Networks (DNN), and Support Vector Machine (SVM) algorithms, as well as logistic regression, to conduct feature selection and predictive modeling. Feature selection was implemented based on permutation-based feature importance lists derived from the machine learning models including all features, using a balanced training data set. To develop prediction models, the top 10%, 25%, and 50% most important predictive features were selected. Prediction models were developed with the training data set with 5-fold cross-validation for internal validation. Model performance was assessed using area under the receiver operating curve (AUC) values. The CatBoost-based prediction model after 26 weeks' gestation performed best with an AUC value of 0.70 (0.67, 0.73), accuracy of 0.81, sensitivity of 0.47, and specificity of 0.83. Number of antenatal care visits before 24 weeks' gestation, aspartate aminotransferase level at registration, symphysis fundal height, maternal weight, abdominal circumference, and blood pressure emerged as strong predictors after 26 completed weeks. The application of machine learning on pregnancy surveillance data is a promising approach to predict preterm birth and we identified several modifiable antenatal predictors.

Lifestyle and clinical factors as predictive indicators of cardiometabolic multimorbidity in Chinese adults: Baseline findings of the Beijing Health Management Cohort (BHMC) study

BACKGROUND

Cardiometabolic multimorbidity (CMM) is increasing globally as a result of lifestyle changes and the aging population. Even though previous studies have examined risk factors associated with CMM, there is a shortage of prediction models that can accurately identify high-risk individuals for early prevention.

METHODS

In the baseline survey of the Beijing Health Management Cohort, a total of 77,752 adults aged 18 years or older were recruited from 2020 to 2021. Data on lifestyle factors, clinical profiles, and diagnoses of diabetes, coronary heart disease, and stroke were collected. Logistic regression models were used to identify risk factors for CMM. Nomograms were developed to estimate an individual's probability of CMM based on the identified risk factors. The performance of the model was evaluated using the area under the receiver operating characteristic curve (AUC).

RESULTS

In men, the top three risk factors for CMM were hypertension (OR: 3.52, 95 % CI: 2.97-4.18), eating very fast (3.43, 2.27-5.16), and dyslipidemia (2.59, 2.20-3.06). In women, hypertension showed the strongest association with CMM (3.62, 2.90-4.52), followed by night sleep duration less than 5 h per day (2.41, 1.67-3.50) and dyslipidemia (1.91, 1.58-2.32). The ORs for holding passive and depressed psychological traits were 1.49 (95%CI: 1.08-2.06) in men and 1.58 (1.03-2.43) in women. Prediction models incorporating these factors demonstrated good discrimination in the test set, with AUC 0.84 (0.83-0.86) for men and 0.90 (0.89-0.91) for women. The sex-specific nomograms were established based on selected predictors.

CONCLUSIONS

Modifiable lifestyle factors, metabolic health and psychological trait are associated with the risk of CMM. The developed prediction models and nomograms could facilitate early identification of individuals at high-risk of CMM.

Machine Learning Prediction Algorithm for In-Hospital Mortality following Body Contouring

BACKGROUND

Body contouring is a common procedure, but it is worth attention because of concern for a variety of complications, and even the potential for death. As a result, the purpose of this study was to determine the key predictors following body contouring and create models for the risk of mortality using diverse machine learning (ML) models.

METHODS

The National Inpatient Sample database from 2015 to 2017 was queried to identify patients undergoing body contouring. Candidate predictors, such as demographics, comorbidities, personal history, postoperative complications, and operative features, were included. The outcome was in-hospital mortality. Models were compared by area under the curve, accuracy, sensitivity, specificity, positive and negative predictive values, and decision curve analysis.

RESULTS

Overall, 8214 patients undergoing body contouring were identified, among whom 141 (1.72%) died in the hospital. Variable importance plot demonstrated that sepsis was the variable with greatest importance across all ML algorithms, followed by Elixhauser Comorbidity Index, cardiac arrest, and so forth. The naive Bayes model had a higher predictive performance (area under the curve, 0.898; 95% CI, 0.884 to 0.911) among these eight ML models. Similarly, in the decision curve analysis, the naive Bayes model also demonstrated a higher net benefit (ie, the correct classification of in-hospital deaths considering a tradeoff between false-negatives and false-positives) compared with the other seven models across a range of threshold probability values.

CONCLUSION

The ML models, as indicated by this study, can be used to predict in-hospital death for patients at risk who undergo body contouring.

Node embedding-based graph autoencoder outlier detection for adverse pregnancy outcomes

Adverse pregnancy outcomes, such as low birth weight (LBW) and preterm birth (PTB), can have serious consequences for both the mother and infant. Early prediction of such outcomes is important for their prevention. Previous studies using traditional machine learning (ML) models for predicting PTB and LBW have encountered two important limitations: extreme class imbalance in medical datasets and the inability to account for complex relational structures between entities. To address these limitations, we propose a node embedding-based graph outlier detection algorithm to predict adverse pregnancy outcomes. We developed a knowledge graph using a well-curated representative dataset of the Emirati population and two node embedding algorithms. The graph autoencoder (GAE) was trained by applying a combination of original risk factors and node embedding features. Samples that were difficult to reconstruct at the output of GAE were identified as outliers considered representing PTB and LBW samples. Our experiments using LBW, PTB, and very PTB datasets demonstrated that incorporating node embedding considerably improved performance, achieving a 12% higher AUC-ROC compared to traditional GAE. Our study demonstrates the effectiveness of node embedding and graph outlier detection in improving the prediction performance of adverse pregnancy outcomes in well-curated population datasets.

Combined application of inflammation-related biomarkers to predict postoperative complications of rectal cancer patients: a retrospective study by machine learning analysis

BACKGROUND

Postoperative complications in patients of rectal cancer pose challenges to postoperative recovery. Accurately predicting these complications is crucial for developing effective treatment plans for patients.

METHODS

In this retrospective study, 493 patients with rectal cancer who underwent radical resection between January 2020 and December 2021 were examined. We evaluated logistic regression, support vector machines, regression trees, and random forests to predict the incidence of postoperative complications in patients and evaluate the performance of the model. The results will be analyzed to make recommendations for reducing complications.

RESULTS

Among the four machine learning models, random forest demonstrated the highest results. The performance of this model was showed with an AUC of 0.880 (95% CI 0.807-0.949), an accuracy of 88.0% (95% CI 0.815-0.929), a sensitivity of 96.6%, and a specificity of 45.8%. Notably, factors such as inflammation related prognostic index, prognostic nutritional index, tumor location, and T stage were found to significantly increase the probability of postoperative complications.

CONCLUSION

Our study provided evidence that machine learning models can effectively evaluate early postoperative complications of the patients after surgery.

Similarity network fusion to identify phenotypes of small-for-gestational-age fetuses

Fetal growth restriction (FGR) affects 5-10% of pregnancies, is the largest contributor to fetal death, and can have long-term consequences for the child. Implementation of a standard clinical classification system is hampered by the multiphenotypic spectrum of small fetuses with substantial differences in perinatal risks. Machine learning and multiomics data can potentially revolutionize clinical decision-making in FGR by identifying new phenotypes. Herein, we describe a cluster analysis of FGR based on an unbiased machine-learning method. Our results confirm the existence of two subtypes of human FGR with distinct molecular and clinical features based on multiomic analysis. In addition, we demonstrated that clusters generated by machine learning significantly outperform single data subtype analysis and biologically support the current clinical classification in predicting adverse maternal and neonatal outcomes. Our approach can aid in the refinement of clinical classification systems for FGR supported by molecular and clinical signatures.

Development of machine learning models for predicting outcome in patients with distal medium vessel occlusions: a retrospective study

Background

While numerous prognostic factors have been reported for large vessel occlusion (LVO)-acute ischemic stroke (AIS) patients, the same cannot be said for distal medium vessel occlusions (DMVOs). We used machine learning (ML) algorithms to develop a model predicting the short-term outcome of AIS patients with DMVOs using demographic, clinical, and laboratory variables and baseline computed tomography (CT) perfusion (CTP) postprocessing quantitative parameters.

Methods

In this retrospective cohort study, consecutive patients with AIS admitted to two comprehensive stroke centers between January 1, 2017, and September 1, 2022, were screened. Demographic, clinical, and radiological data were extracted from electronic medical records. The clinical outcome was divided into two categories, with a cut-off defined by the median National Institutes of Health Stroke Scale (NIHSS) shift score. Data preprocessing involved addressing missing values through imputation, scaling with a robust scaler, normalization using min-max normalization, and encoding of categorical variables. The data were split into training and test sets (70:30), and recursive feature elimination (RFE) was employed for feature selection. For ML analyses, XGBoost, LightGBM, CatBoost, multi-layer perceptron, random forest, and logistic regression algorithms were utilized. Performance evaluation involved the receiver operating characteristic (ROC) curve, precision-recall curve (PRC), the area under these curves, accuracy, precision, recall, and Matthews correlation coefficient (MCC). The relative weights of predictor variables were examined using Shapley additive explanations (SHAP).

Results

Sixty-nine patients were included and divided into two groups: 35 patients with favorable outcomes and 34 patients with unfavorable outcomes. Utilizing ten selected features, the XGBoost algorithm achieved the best performance in predicting unfavorable outcomes, with an area under the ROC curve (AUROC) of 0.894 and an area under the PRC curve (AUPRC) of 0.756. The SHAP analysis ranked the following features in order of importance for the XGBoost model: mismatch volume, time-to-maximum of the tissue residue function (Tmax) >6 s, diffusion-weighted imaging (DWI) volume, neutrophil-to-platelet ratio (NPR), mean corpuscular volume (MCV), Tmax >10 s, hemoglobin, potassium, hypoperfusion index (HI), and Tmax >8 s.

Conclusions

Our ML models, trained on baseline quantitative laboratory and CT parameters, accurately predicted the short-term outcome in patients with DMVOs. These findings may aid clinicians in predicting prognosis and may be helpful for future research.

Building a predictive model of low birth weight in low- and middle-income countries: a prospective cohort study

BACKGROUND

Low birth weight (LBW, < 2500 g) infants are at significant risk for death and disability. Improving outcomes for LBW infants requires access to advanced neonatal care, which is a limited resource in low- and middle-income countries (LMICs). Predictive modeling might be useful in LMICs to identify mothers at high-risk of delivering a LBW infant to facilitate referral to centers capable of treating these infants.

METHODS

We developed predictive models for LBW using the NICHD Global Network for Women's and Children's Health Research Maternal and Newborn Health Registry. This registry enrolled pregnant women from research sites in the Democratic Republic of the Congo, Zambia, Kenya, Guatemala, India (2 sites: Belagavi, Nagpur), Pakistan, and Bangladesh between January 2017 - December 2020. We tested five predictive models: decision tree, random forest, logistic regression, K-nearest neighbor and support vector machine.

RESULTS

We report a rate of LBW of 13.8% among the eight Global Network sites from 2017-2020, with a range of 3.8% (Kenya) and approximately 20% (in each Asian site). Of the five models tested, the logistic regression model performed best with an area under the curve of 0.72, an accuracy of 61% and a recall of 72%. All of the top performing models identified clinical site, maternal weight, hypertensive disorders, severe antepartum hemorrhage and antenatal care as key variables in predicting LBW.

CONCLUSIONS

Predictive modeling can identify women at high risk for delivering a LBW infant with good sensitivity using clinical variables available prior to delivery in LMICs. Such modeling is the first step in the development of a clinical decision support tool to assist providers in decision-making regarding referral of these women prior to delivery. Consistent referral of women at high-risk for delivering a LBW infant could have extensive public health consequences in LMICs by directing limited resources for advanced neonatal care to the infants at highest risk.

Development and validation of an interpretable machine learning model-Predicting mild cognitive impairment in a high-risk stroke population

Background

Mild cognitive impairment (MCI) is considered a preclinical stage of Alzheimer's disease (AD). People with MCI have a higher risk of developing dementia than healthy people. As one of the risk factors for MCI, stroke has been actively treated and intervened. Therefore, selecting the high-risk population of stroke as the research object and discovering the risk factors of MCI as early as possible can prevent the occurrence of MCI more effectively.

Methods

The Boruta algorithm was used to screen variables, and eight machine learning models were established and evaluated. The best performing models were used to assess variable importance and build an online risk calculator. Shapley additive explanation is used to explain the model.

Results

A total of 199 patients were included in the study, 99 of whom were male. Transient ischemic attack (TIA), homocysteine, education, hematocrit (HCT), diabetes, hemoglobin, red blood cells (RBC), hypertension, prothrombin time (PT) were selected by Boruta algorithm. Logistic regression (AUC = 0.8595) was the best model for predicting MCI in high-risk groups of stroke, followed by elastic network (ENET) (AUC = 0.8312), multilayer perceptron (MLP) (AUC = 0.7908), extreme gradient boosting (XGBoost) (AUC = 0.7691), and support vector machine (SVM) (AUC = 0.7527), random forest (RF) (AUC = 0.7451), K-nearest neighbors (KNN) (AUC = 0.7380), decision tree (DT) (AUC = 0.6972). The importance of variables suggests that TIA, diabetes, education, and hypertension are the top four variables of importance.

Conclusion

Transient ischemic attack (TIA), diabetes, education, and hypertension are the most important risk factors for MCI in high-risk groups of stroke, and early intervention should be performed to reduce the occurrence of MCI.

Development and validation of machine learning-based clinical decision support tool for identifying malnutrition in NICU patients

Hospitalized newborns have an increased risk of malnutrition and, especially preterm infants, often experience malnutrition-related extrauterine growth restriction (EUGR). The aim of this study was to predict the discharge weight and the presence of weight gain at discharge with machine learning (ML) algorithms. The demographic and clinical parameters were used to develop the models using fivefold cross-validation in the software-R with a neonatal nutritional screening tool (NNST). A total of 512 NICU patients were prospectively included in the study. Length of hospital stay (LOS), parenteral nutrition treatment (PN), postnatal age (PNA), surgery, and sodium were the most important variables in predicting the presence of weight gain at discharge with a random forest classification (AUROC:0.847). The AUROC of NNST-Plus, which was improved by adding LOS, PN, PNA, surgery, and sodium to NNST, increased by 16.5%. In addition, weight at admission, LOS, gestation-adjusted age at admission (> 40 weeks), sex, gestational age, birth weight, PNA, SGA, complications of labor and delivery, multiple birth, serum creatinine, and PN treatment were the most important variables in predicting discharge weight with an elastic net regression (R² = 0.748). This is the first study on the early prediction of EUGR with promising clinical performance based on ML algorithms. It is estimated that the incidence of EUGR can be improved with the implementation of this ML-based web tool ( http://www.softmed.hacettepe.edu.tr/NEO-DEER/ ) in clinical practice.

Prediction of elevated groundwater fluoride across India using multi-model approach: insights on the influence of geologic and environmental factors

Elevated fluoride in groundwater is a severe problem in India due to its extensive occurrence and detrimental health impacts on the large population that thrives on groundwater. Although fluoride is primarily a geogenic pollutant, existing model-based studies lack the amalgamation of the influence of geologic factors, specifically tectonics, for identifying groundwater fluoride distribution. This drawback encourages the present study to investigate the association of the tectonic framework with fluoride in a multi-model approach. We have applied three machine learning models (random forest, boosted regression tree, and logistic regression) to predict elevated groundwater fluoride based on fluoride measurements across India. The random forest model outperformed other models with an accuracy of 93%. Tectonics was found to be one of the most important predictors alongside "depth to water table." Two major areas of high risk identified were the northwest parts and the south-southeast cratonic peninsular region. The random forest model also performed significantly well over the validation dataset. We estimate that nearly 257 million people are exposed to elevated fluoride risk in India. We endeavor that the findings of our study would be an effective tool for identifying the areas at risk of elevated fluoride and also assist in undertaking effective groundwater management strategies.

A Machine Learning-Based Online Prediction Tool for Predicting Short-Term Postoperative Outcomes Following Spinal Tumor Resections

Background: Preoperative prediction of short-term postoperative outcomes in spinal tumor patients can lead to more precise patient care plans that reduce the likelihood of negative outcomes. With this study, we aimed to develop machine learning algorithms for predicting short-term postoperative outcomes and implement these models in an open-source web application. Methods: Patients who underwent surgical resection of spinal tumors were identified using the American College of Surgeons, National Surgical Quality Improvement Program. Three outcomes were predicted: prolonged length of stay (LOS), nonhome discharges, and major complications. Four machine learning algorithms were developed and integrated into an open access web application to predict these outcomes. Results: A total of 3073 patients that underwent spinal tumor resection were included in the analysis. The most accurately predicted outcomes in terms of the area under the receiver operating characteristic curve (AUROC) was the prolonged LOS with a mean AUROC of 0.745 The most accurately predicting algorithm in terms of AUROC was random forest, with a mean AUROC of 0.743. An open access web application was developed for getting predictions for individual patients based on their characteristics and this web application can be accessed here: huggingface.co/spaces/MSHS-Neurosurgery-Research/NSQIP-ST. Conclusion: Machine learning approaches carry significant potential for the purpose of predicting postoperative outcomes following spinal tumor resections. Development of predictive models as clinically useful decision-making tools may considerably enhance risk assessment and prognosis as the amount of data in spinal tumor surgery continues to rise.

Application of Natural Language Processing (NLP) in Detecting and Preventing Suicide Ideation: A Systematic Review

(1) Introduction: Around a million people are reported to die by suicide every year, and due to the stigma associated with the nature of the death, this figure is usually assumed to be an underestimate. Machine learning and artificial intelligence such as natural language processing has the potential to become a major technique for the detection, diagnosis, and treatment of people. (2) Methods: PubMed, EMBASE, MEDLINE, PsycInfo, and Global Health databases were searched for studies that reported use of NLP for suicide ideation or self-harm. (3) Result: The preliminary search of 5 databases generated 387 results. Removal of duplicates resulted in 158 potentially suitable studies. Twenty papers were finally included in this review. (4) Discussion: Studies show that combining structured and unstructured data in NLP data modelling yielded more accurate results than utilizing either alone. Additionally, to reduce suicides, people with mental problems must be continuously and passively monitored. (5) Conclusions: The use of AI&ML opens new avenues for considerably guiding risk prediction and advancing suicide prevention frameworks. The review's analysis of the included research revealed that the use of NLP may result in low-cost and effective alternatives to existing resource-intensive methods of suicide prevention.

Towards an explainable clinical decision support system for large-for-gestational-age births

A myriad of maternal and neonatal complications can result from delivery of a large-for-gestational-age (LGA) infant. LGA birth rates have increased in many countries since the late 20th century, partially due to a rise in maternal body mass index, which is associated with LGA risk. The objective of the current study was to develop LGA prediction models for women with overweight and obesity for the purpose of clinical decision support in a clinical setting. Maternal characteristics, serum biomarkers and fetal anatomy scan measurements for 465 pregnant women with overweight and obesity before and at approximately 21 weeks gestation were obtained from the PEARS (Pregnancy Exercise and Nutrition with smart phone application support) study data. Random forest, support vector machine, adaptive boosting and extreme gradient boosting algorithms were applied with synthetic minority over-sampling technique to develop probabilistic prediction models. Two models were developed for use in different settings: a clinical setting for white women (AUC-ROC of 0.75); and a clinical setting for women of all ethnicity and regions (AUC-ROC of 0.57). Maternal age, mid upper arm circumference, white cell count at the first antenatal visit, fetal biometry and gestational age at fetal anatomy scan were found to be important predictors of LGA. Pobal HP deprivation index and fetal biometry centiles, which are population-specific, are also important. Moreover, we explained our models with Local Interpretable Model-agnostic Explanations (LIME) to improve explainability, which was proven effective by case studies. Our explainable models can effectively predict the probability of an LGA birth for women with overweight and obesity, and are anticipated to be useful to support clinical decision-making and for the development of early pregnancy intervention strategies to reduce pregnancy complications related to LGA.

Analytical Comparison of Risk Prediction Models for the Onset of Macrosomia Based on Three Statistical Methods

Background and Purpose. Fetal overgrowth can pose a serious threat to the safety of a mother and child. Early identification of high-risk pregnant women and timely pregnancy intervention and guidance are of great value in preventing the development of giant babies and improving adverse maternal and infant outcomes. The current clinical methods for predicting macrosomia mainly rely on obstetric examination and imaging, but their accuracy is controversial. And there is no accepted method for accurately predicting macrosomia. We investigated the risk factors influencing the occurrence of macrosomia and established a prediction model for the occurrence of macrosomia to provide a reference basis for interventions to prevent macrosomia. Method. A retrospective selection of 93 women who were hospitalized in our hospital from March 2019 to May 2022 with a singleton pregnancy and delivered at term with macrosomia were the study group. And 356 women who delivered a normal size baby during the same period were the control group. The variables that were associated with the onset of macrosomia were screened from maternal medical records. Logistic regression models, random forest, and CART decision tree models were developed using the screened variables as input variables and whether they were macrosomia as outcome variables, respectively. The performance of the three models was evaluated by accuracy, precision, recall, F1 score, and receiver operating characteristic curve (ROC). Result. The risk prediction models for the onset of macrosomia, logistic regression model, random forest model, and decision tree, were successfully developed, with accuracies of 0.904, 1.000, and 0.901 in the training set and 0.926, 0.582, and 0.852 in the validation set, respectively. The AUC in the training set were 0.898, 1.000, and 0.789, and in the validation set were 0.906, 0.913, and 0.731, respectively. In general, the logistic regression model has the highest diagnostic efficiency, followed by the random forest model. Conclusion. Logistic regression models have high application value in the assessment of predicting the risk of macrosomia, and it is suggested that the advantages of logistic regression models and random forest models should be combined in future studies and applications to make them work better in the prediction of the risk of macrosomia.

Predictive models for small-for-gestational-age births in women exposed to pesticides before pregnancy based on multiple machine learning algorithms

Background

The association between prenatal pesticide exposures and a higher incidence of small-for-gestational-age (SGA) births has been reported. No prediction model has been developed for SGA neonates in pregnant women exposed to pesticides prior to pregnancy.

Methods

A retrospective cohort study was conducted using information from the National Free Preconception Health Examination Project between 2010 and 2012. A development set (n = 606) and a validation set (n = 151) of the dataset were split at random. Traditional logistic regression (LR) method and six machine learning classifiers were used to develop prediction models for SGA neonates. The Shapley Additive Explanation (SHAP) model was applied to determine the most influential variables that contributed to the outcome of the prediction.

Results

757 neonates in total were analyzed. SGA occurred in 12.9% (n = 98) of cases overall. With an area under the receiver-operating-characteristic curve (AUC) of 0.855 [95% confidence interval (CI): 0.752–0.959], the model based on category boosting (CatBoost) algorithm obtained the best performance in the validation set. With the exception of the LR model (AUC: 0.691, 95% CI: 0.554–0.828), all models had good AUCs. Using recursive feature elimination (RFE) approach to perform the feature selection, we included 15 variables in the final model based on CatBoost classifier, achieving the AUC of 0.811 (95% CI: 0.675–0.947).

Conclusions

Machine learning algorithms can develop satisfactory tools for SGA prediction in mothers exposed to pesticides prior to pregnancy, which might become a tool to predict SGA neonates in the high-risk population.

Infant birth weight estimation and low birth weight classification in United Arab Emirates using machine learning algorithms

Accurate prediction of a newborn’s birth weight (BW) is a crucial determinant to evaluate the newborn’s health and safety. Infants with low BW (LBW) are at a higher risk of serious short- and long-term health outcomes. Over the past decade, machine learning (ML) techniques have shown a successful breakthrough in the field of medical diagnostics. Various automated systems have been proposed that use maternal features for LBW prediction. However, each proposed system uses different maternal features for LBW classification and estimation. Therefore, this paper provides a detailed setup for BW estimation and LBW classification. Multiple subsets of features were combined to perform predictions with and without feature selection techniques. Furthermore, the synthetic minority oversampling technique was employed to oversample the minority class. The performance of 30 ML algorithms was evaluated for both infant BW estimation and LBW classification. Experiments were performed on a self-created dataset with 88 features. The dataset was obtained from 821 women from three hospitals in the United Arab Emirates. Different performance metrics, such as mean absolute error and mean absolute percent error, were used for BW estimation. Accuracy, precision, recall, F-scores, and confusion matrices were used for LBW classification. Extensive experiments performed using five-folds cross validation show that the best weight estimation was obtained using Random Forest algorithm with mean absolute error of 294.53 g while the best classification performance was obtained using Logistic Regression with SMOTE oversampling techniques that achieved accuracy, precision, recall and F1 score of 90.24%, 87.6%, 90.2% and 0.89, respectively. The results also suggest that features such as diabetes, hypertension, and gestational age, play a vital role in LBW classification.

Machine learning approach for the prediction of 30-day mortality in patients with sepsis-associated encephalopathy

Objective

Our study aimed to identify predictors as well as develop machine learning (ML) models to predict the risk of 30-day mortality in patients with sepsis-associated encephalopathy (SAE).

Materials and methods

ML models were developed and validated based on a public database named Medical Information Mart for Intensive Care (MIMIC)-IV. Models were compared by the area under the curve (AUC), accuracy, sensitivity, specificity, positive and negative predictive values, and Hosmer–Lemeshow good of fit test.

Results

Of 6994 patients in MIMIC-IV included in the final cohort, a total of 1232 (17.62%) patients died following SAE. Recursive feature elimination (RFE) selected 15 variables, including acute physiology score III (APSIII), Glasgow coma score (GCS), sepsis related organ failure assessment (SOFA), Charlson comorbidity index (CCI), red blood cell volume distribution width (RDW), blood urea nitrogen (BUN), age, respiratory rate, PaO₂, temperature, lactate, creatinine (CRE), malignant cancer, metastatic solid tumor, and platelet (PLT). The validation cohort demonstrated all ML approaches had higher discriminative ability compared with the bagged trees (BT) model, although the difference was not statistically significant. Furthermore, in terms of the calibration performance, the artificial neural network (NNET), logistic regression (LR), and adapting boosting (Ada) models had a good calibration—namely, a high accuracy of prediction, with P-values of 0.831, 0.119, and 0.129, respectively.

Conclusions

The ML models, as demonstrated by our study, can be used to evaluate the prognosis of SAE patients in the intensive care unit (ICU). Online calculator could facilitate the sharing of predictive models.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12874-022-01664-z.

Risk scores for predicting small for gestational age infants in Japan: The TMM birthree cohort study

This study aimed to construct a prediction model for small-for-gestational-age (SGA) infants in Japan by creating a risk score during pregnancy. A total of 17,073 subjects were included in the Tohoku Medical Megabank Project Birth and Three-Generation Cohort Study, a prospective cohort study. A multiple logistic regression model was used to construct risk scores during early and mid-gestational periods (11–17 and 18–21 weeks of gestation, respectively). The risk score during early gestation comprised the maternal age, height, body mass index (BMI) during early gestation, parity, assisted reproductive technology (ART) with frozen-thawed embryo transfer (FET), smoking status, blood pressure (BP) during early gestation, and maternal birth weight. The risk score during mid-gestation also consisted of the maternal age, height, BMI during mid-gestation, weight gain, parity, ART with FET, smoking status, BP level during mid-gestation, maternal birth weight, and estimated fetal weight during mid-gestation. The C-statistics of the risk scores during early- and mid-gestation were 0.658 (95% confidence interval [CI]: 0.642–0.675) and 0.725 (95% CI: 0.710–0.740), respectively. In conclusion, the predictive ability of the risk scores during mid-gestation for SGA infants was acceptable and better than that of the risk score during early gestation.

Development and Evaluation of a Machine Learning Prediction Model for Small-for-Gestational-Age Births in Women Exposed to Radiation before Pregnancy

Exposure to radiation has been associated with increased risk of delivering small-for-gestational-age (SGA) newborns. There are no tools to predict SGA newborns in pregnant women exposed to radiation before pregnancy. Here, we aimed to develop an array of machine learning (ML) models to predict SGA newborns in women exposed to radiation before pregnancy. Patients’ data was obtained from the National Free Preconception Health Examination Project from 2010 to 2012. The data were randomly divided into a training dataset (n = 364) and a testing dataset (n = 91). Eight various ML models were compared for solving the binary classification of SGA prediction, followed by a post hoc explainability based on the SHAP model to identify and interpret the most important features that contribute to the prediction outcome. A total of 455 newborns were included, with the occurrence of 60 SGA births (13.2%). Overall, the model obtained by extreme gradient boosting (XGBoost) achieved the highest area under the receiver-operating-characteristic curve (AUC) in the testing set (0.844, 95% confidence interval (CI): 0.713–0.974). All models showed satisfied AUCs, except for the logistic regression model (AUC: 0.561, 95% CI: 0.355–0.768). After feature selection by recursive feature elimination (RFE), 15 features were included in the final prediction model using the XGBoost algorithm, with an AUC of 0.821 (95% CI: 0.650–0.993). ML algorithms can generate robust models to predict SGA newborns in pregnant women exposed to radiation before pregnancy, which may thus be used as a prediction tool for SGA newborns in high-risk pregnant women.

Vitamin D Deficiency, Excessive Gestational Weight Gain, and Oxidative Stress Predict Small for Gestational Age Newborns Using an Artificial Neural Network Model

(1) Background: Size at birth is an important early determinant of health later in life. The prevalence of small for gestational age (SGA) newborns is high worldwide and may be associated with maternal nutritional and metabolic factors. Thus, estimation of fetal growth is warranted. (2) Methods: In this work, we developed an artificial neural network (ANN) model based on first-trimester maternal body fat composition, biochemical and oxidative stress biomarkers, and gestational weight gain (GWG) to predict an SGA newborn in pregnancies with or without obesity. A sensibility analysis to classify maternal features was conducted, and a simulator based on the ANN algorithm was constructed to predict the SGA outcome. Several predictions were performed by varying the most critical maternal features attained by the model to obtain different scenarios leading to SGA. (3) Results: The ANN model showed good performance between the actual and simulated data (R2 = 0.938) and an AUROC of 0.8 on an independent dataset. The top-five maternal predictors in the first trimester were protein and lipid oxidation biomarkers (carbonylated proteins and malondialdehyde), GWG, vitamin D, and total antioxidant capacity. Finally, excessive GWG and redox imbalance predicted SGA newborns in the implemented simulator. Significantly, vitamin D deficiency also predicted simulated SGA independently of GWG or redox status. (4) Conclusions: The study provided a computational model for the early prediction of SGA, in addition to a promising simulator that facilitates hypothesis-driven constructions, to be further validated as an application.

Contributions of artificial intelligence reported in Obstetrics and Gynecology journals: a systematic review

Background

The applications of artificial intelligence (AI) processes have grown significantly in all medical disciplines during the last decades. Two main types of AI have been applied in medicine: symbolic AI (eg, knowledge base and ontologies) and nonsymbolic AI (eg, machine learning and artificial neural networks). Consequently, AI has also been applied across most obstetrics and gynecology (OB/GYN) domains, including general obstetrics, gynecology surgery, fetal ultrasound, and assisted reproductive medicine, among others.

Objective

The aim of this study was to provide a systematic review to establish the actual contributions of AI reported in OB/GYN discipline journals.

Methods

The PubMed database was searched for citations indexed with “artificial intelligence” and at least one of the following medical subject heading (MeSH) terms between January 1, 2000, and April 30, 2020: “obstetrics”; “gynecology”; “reproductive techniques, assisted”; or “pregnancy.” All publications in OB/GYN core disciplines journals were considered. The selection of journals was based on disciplines defined in Web of Science. The publications were excluded if no AI process was used in the study. Review, editorial, and commentary articles were also excluded. The study analysis comprised (1) classification of publications into OB/GYN domains, (2) description of AI methods, (3) description of AI algorithms, (4) description of data sets, (5) description of AI contributions, and (6) description of the validation of the AI process.

Results

The PubMed search retrieved 579 citations and 66 publications met the selection criteria. All OB/GYN subdomains were covered: obstetrics (41%, 27/66), gynecology (3%, 2/66), assisted reproductive medicine (33%, 22/66), early pregnancy (2%, 1/66), and fetal medicine (21%, 14/66). Both machine learning methods (39/66) and knowledge base methods (25/66) were represented. Machine learning used imaging, numerical, and clinical data sets. Knowledge base methods used mostly omics data sets. The actual contributions of AI were method/algorithm development (53%, 35/66), hypothesis generation (42%, 28/66), or software development (3%, 2/66). Validation was performed on one data set (86%, 57/66) and no external validation was reported. We observed a general rising trend in publications related to AI in OB/GYN over the last two decades. Most of these publications (82%, 54/66) remain out of the scope of the usual OB/GYN journals.

Conclusions

In OB/GYN discipline journals, mostly preliminary work (eg, proof-of-concept algorithm or method) in AI applied to this discipline is reported and clinical validation remains an unmet prerequisite. Improvement driven by new AI research guidelines is expected. However, these guidelines are covering only a part of AI approaches (nonsymbolic) reported in this review; hence, updates need to be considered.

Decreased psoas muscle area is a prognosticator for 90-day and 1-year survival in patients undergoing surgical treatment for spinal metastasis

BACKGROUND AND AIMS

Survival estimation for patients with spinal metastasis is crucial to treatment decisions. Psoas muscle area (PMA), a surrogate for total muscle mass, has been proposed as a useful survival prognosticator. However, few studies have validated the predictive value of decreased PMA in an Asian cohort or its predictive value after controlling for existing preoperative scoring systems (PSSs). In this study, we aim to answer: (1) Is PMA associated with survival in Han Chinese patients with spinal metastasis? (2) Is PMA a good prognosticator according to concordance index (c-index) and decision curve analysis (DCA) after controlling for six existing and commonly used PSSs?

METHODS

This study included 180 adult (≥18 years old) Taiwanese patients with a mean age of 58.3 years (range: 22-85) undergoing surgical treatment for spinal metastasis. A patient's PMA was classified into decreased, medium, and large if it fell into the lower (0-33%), middle (33-67%), and upper (67-100%) 1/3 in the study cohort, respectively. We used logistic and cox proportional-hazard regressions to assess whether PMA was associated with 90-day, 1-year, and overall survival. The model performance before and after addition of PMA to six commonly used PSSs, including Tomita score, original Tokuhashi score, revised Tokuhashi score, modified Bauer score, New England Spinal Metastasis Score, and Skeletal Oncology Research Group machine learning algorithms (SORG-MLAs), was compared by c-index and DCA to determine if PMA was a useful survival prognosticator.

RESULTS

Patients with a larger PMA is associated with better 90-day, but not 1-year, survival. The model performance of 90-day survival prediction improved after PMA was incorporated into all PSSs except SORG-MLAs. PMA barely improved the discriminatory ability (c-index, 0.74; 95% confidence interval [CI], 0.67-0.82 vs. c-index, 0.74; 95% CI, 0.66-0.81) and provided little gain of clinical net benefit on DCA for SORG-MLAs' 90-day survival prediction.

CONCLUSIONS

PMA is a prognosticator for 90-day survival and improves the discriminatory ability of earlier-proposed PSSs in our Asian cohort. However, incorporating PMA into more modern PSSs such as SORG-MLAs did not significantly improve its prediction performance.

Are Machine Learning Algorithms More Accurate in Predicting Vegetable and Fruit Consumption Than Traditional Statistical Models? An Exploratory Analysis

Machine learning (ML) algorithms may help better understand the complex interactions among factors that influence dietary choices and behaviors. The aim of this study was to explore whether ML algorithms are more accurate than traditional statistical models in predicting vegetable and fruit (VF) consumption. A large array of features (2,452 features from 525 variables) encompassing individual and environmental information related to dietary habits and food choices in a sample of 1,147 French-speaking adult men and women was used for the purpose of this study. Adequate VF consumption, which was defined as 5 servings/d or more, was measured by averaging data from three web-based 24 h recalls and used as the outcome to predict. Nine classification ML algorithms were compared to two traditional statistical predictive models, logistic regression and penalized regression (Lasso). The performance of the predictive ML algorithms was tested after the implementation of adjustments, including normalizing the data, as well as in a series of sensitivity analyses such as using VF consumption obtained from a web-based food frequency questionnaire (wFFQ) and applying a feature selection algorithm in an attempt to reduce overfitting. Logistic regression and Lasso predicted adequate VF consumption with an accuracy of 0.64 (95% confidence interval [CI]: 0.58–0.70) and 0.64 (95%CI: 0.60–0.68) respectively. Among the ML algorithms tested, the most accurate algorithms to predict adequate VF consumption were the support vector machine (SVM) with either a radial basis kernel or a sigmoid kernel, both with an accuracy of 0.65 (95%CI: 0.59–0.71). The least accurate ML algorithm was the SVM with a linear kernel with an accuracy of 0.55 (95%CI: 0.49–0.61). Using dietary intake data from the wFFQ and applying a feature selection algorithm had little to no impact on the performance of the algorithms. In summary, ML algorithms and traditional statistical models predicted adequate VF consumption with similar accuracies among adults. These results suggest that additional research is needed to explore further the true potential of ML in predicting dietary behaviours that are determined by complex interactions among several individual, social and environmental factors.

Mapping spatial distribution and geographic shifts of East African highland banana (Musa spp.) in Uganda

East African highland banana (Musa acuminata genome group AAA-EA; hereafter referred to as banana) is critical for Uganda's food supply, hence our aim to map current distribution and to understand changes in banana production areas over the past five decades. We collected banana presence/absence data through an online survey based on high-resolution satellite images and coupled this data with independent covariates as inputs for ensemble machine learning prediction of current banana distribution. We assessed geographic shifts of production areas using spatially explicit differences between the 1958 and 2016 banana distribution maps. The biophysical factors associated with banana spatial distribution and geographic shift were determined using a logistic regression model and classification and regression tree, respectively. Ensemble models were superior (AUC = 0.895; 0.907) compared to their constituent algorithms trained with 12 and 17 covariates, respectively: random forests (AUC = 0.883; 0.901), gradient boosting machines (AUC = 0.878; 0.903), and neural networks (AUC = 0.870; 0.890). The logistic regression model (AUC = 0.879) performance was similar to that for the ensemble model and its constituent algorithms. In 2016, banana cultivation was concentrated in the western (44%) and central (36%) regions, while only a small proportion was in the eastern (18%) and northern (2%) regions. About 60% of increased cultivation since 1958 was in the western region; 50% of decreased cultivation in the eastern region; and 44% of continued cultivation in the central region. Soil organic carbon, soil pH, annual precipitation, slope gradient, bulk density and blue reflectance were associated with increased banana cultivation while precipitation seasonality and mean annual temperature were associated with decreased banana cultivation over the past 50 years. The maps of spatial distribution and geographic shift of banana can support targeting of context-specific intensification options and policy advocacy to avert agriculture driven environmental degradation.

Using Machine Learning to Predict Complications in Pregnancy: A Systematic Review

Introduction: Artificial intelligence is widely used in medical field, and machine learning has been increasingly used in health care, prediction, and diagnosis and as a method of determining priority. Machine learning methods have been features of several tools in the fields of obstetrics and childcare. This present review aims to summarize the machine learning techniques to predict perinatal complications.

Objective: To identify the applicability and performance of machine learning methods used to identify pregnancy complications.

Methods: A total of 98 articles were obtained with the keywords “machine learning,” “deep learning,” “artificial intelligence,” and accordingly as they related to perinatal complications (“complications in pregnancy,” “pregnancy complications”) from three scientific databases: PubMed, Scopus, and Web of Science. These were managed on the Mendeley platform and classified using the PRISMA method.

Results: A total of 31 articles were selected after elimination according to inclusion and exclusion criteria. The features used to predict perinatal complications were primarily electronic medical records (48%), medical images (29%), and biological markers (19%), while 4% were based on other types of features, such as sensors and fetal heart rate. The main perinatal complications considered in the application of machine learning thus far are pre-eclampsia and prematurity. In the 31 studies, a total of sixteen complications were predicted. The main precision metric used is the AUC. The machine learning methods with the best results were the prediction of prematurity from medical images using the support vector machine technique, with an accuracy of 95.7%, and the prediction of neonatal mortality with the XGBoost technique, with 99.7% accuracy.

Conclusion: It is important to continue promoting this area of research and promote solutions with multicenter clinical applicability through machine learning to reduce perinatal complications. This systematic review contributes significantly to the specialized literature on artificial intelligence and women’s health.

Interpretable machine learning to identify important predictors of birth weight: A prospective cohort study

BACKGROUND

Predicting birth weight and identifying its risk factors are clinically important. This study aims to use interpretable machine learning to predict birth weight and identity important predictors.

METHODS

This prospective cohort study was conducted in Tongzhou Maternal and Child Health Care Hospital of Beijing, China, recruiting pregnant women between June 2018 and February 2019. We used 24 features to predict infant birth weight, including gestational age, mother's age, parity, history of macrosomia delivery, pre-pregnancy body mass index (BMI), height, father's BMI, lifestyle (diet, physical activity, smoking), and biomarker (fasting glucose and lipids) features. Study outcome was birth weight of infant. We used 8 supervised learning models including 4 individual [linear regression, ridge regression, lasso regression, support vector machines regression (SVR)], and 4 ensemble estimators (random forest, AdaBoost, gradient boosted trees, and voting ensemble for regression) to predict birth weight. Model accuracy was measured by root mean squared error (RMSE) of 10-fold cross validation on the training set and RMSE of prediction on the test set. We used permutation importance algorithm to understand the prediction from the models and what affected them.

RESULT

This study included 4,754 mother-child dyads. RMSEs were lower in voting ensemble for regression, linear regression, and SVR than random forest, AdaBoost, and gradient boosted tree. The 5 most important predictors for infant birth weight were gestational age, fetal sex, preterm birth, mother's height, and pre-pregnancy BMI. After adding ultrasound-measured indicators of fetal growth into predictors, mother's height and pre-pregnancy BMI remained the most important predictors in predicting the outcome.

CONCLUSION

Mother's height and pre-pregnancy BMI were identified as important predictors for infant birth weight. Interpretable machine learning is a promising tool in the prediction of birth weight.

Gradient boosted trees with individual explanations: An alternative to logistic regression for viability prediction in the first trimester of pregnancy

BACKGROUND

Clinical models to predict first trimester viability are traditionally based on multivariable logistic regression (LR) which is not directly interpretable for non-statistical experts like physicians. Furthermore, LR requires complete datasets and pre-established variables specifications. In this study, we leveraged the internal non-linearity, feature selection and missing values handling mechanisms of machine learning algorithms, along with a post-hoc interpretability strategy, as potential advantages over LR for clinical modeling.

METHODS

The dataset included 1154 patients with 2377 individual scans and was obtained from a prospective observational cohort study conducted at a hospital in London, UK, from March 2014 to May 2019. The data were split into a training (70%) and a test set (30%). Parsimonious and complete multivariable models were developed from two algorithms to predict first trimester viability at 11-14 weeks gestational age (GA): LR and light gradient boosted machine (LGBM). Missing values were handled by multiple imputation where appropriate. The SHapley Additive exPlanations (SHAP) framework was applied to derive individual explanations of the models.

RESULTS

The parsimonious LGBM model had similar discriminative and calibration performance as the parsimonious LR (AUC 0.885 vs 0.860; calibration slope: 1.19 vs 1.18). The complete models did not outperform the parsimonious models. LGBM was robust to the presence of missing values and did not require multiple imputation unlike LR. Decision path plots and feature importance analysis revealed different algorithm behaviors despite similar predictive performance. The main driving variable from the LR model was the pre-specified interaction between fetal heart presence and mean sac diameter. The crown-rump length variable and a proxy variable reflecting the difference in GA between expected and observed GA were the two most important variables of LGBM. Finally, while variable interactions must be specified upfront with LR, several interactions were ranked by the SHAP framework among the most important features learned automatically by the LGBM algorithm.

CONCLUSIONS

Gradient boosted algorithms performed similarly to carefully crafted LR models in terms of discrimination and calibration for first trimester viability prediction. By handling multi-collinearity, missing values, feature selection and variable interactions internally, the gradient boosted trees algorithm, combined with SHAP, offers a serious alternative to traditional LR models.

Development and validation of a new predictive model for macrosomia at late-term pregnancy: A prospective study

OBJECTIVE

Fetal macrosomia is defined as a birth weight more than 4,000 g and is associated with maternal and fetal complications. This early metabolic disease may influence the entire life of the infant. Currently, macrosomia is predicted by using the estimated fetal weight (EFW). However, the EFW is inaccurate when the gestational week is gradually increasing. To assess precisely the risk of macrosomia, we developed a new predictive model to estimate the risk of macrosomia.

METHODS

We continuously collected data on 655 subjects who attended regular antenatal visits and delivered at the Second Hospital of Hebei Medical University (Shijiazhuang, China) from November 2020 to September 2021. A total of 17 maternal features and 2 fetal ultrasonographic features were included at late-term pregnancy. The 655 subjects were divided into a model training set and an internal validation set. Then, 450 pregnant women were recruited from Handan Central Hospital (Handan, China) from November 2021 to March 2022 as the external validation set. The least absolute shrinkage and selection operator method was used to select the most appropriate predictive features and optimize them via 10-fold cross-validation. The multivariate logistical regressions were used to build the predictive model. Receiver operating characteristic (ROC) curves, C-indices, and calibration plots were obtained to assess model discrimination and accuracy. The model's clinical utility was evaluated via decision curve analysis (DCA).

RESULTS

Four predictors were finally included to develop this new model: prepregnancy obesity (prepregnancy body mass index ≥ 30 kg/m²), hypertriglyceridemia, gestational diabetes mellitus, and fetal abdominal circumference. This model afforded moderate predictive power [area under the ROC curve 0.788 (95% confidence interval [CI] 0.736, 0.840) for the training set, 0.819 (95% CI 0.744,0.894) for the internal validation set, and 0.773 (95% CI 0.713,0.833) for the external validation set]. On DCA, the model evidenced a good fit with, and positive net benefits for, both the internal and external validation sets.

CONCLUSIONS

We developed a predictive model for macrosomia and performed external validation in other regions to further prove the discrimination and accuracy of this predictive model. This novel model will aid clinicians in easily identifying those at high risk of macrosomia and assist obstetricians to plan accordingly.

Validating machine learning models for the prediction of labour induction intervention using routine data: a registry-based retrospective cohort study at a tertiary hospital in northern Tanzania

OBJECTIVES

We aimed at identifying the important variables for labour induction intervention and assessing the predictive performance of machine learning algorithms.

SETTING

We analysed the birth registry data from a referral hospital in northern Tanzania. Since July 2000, every birth at this facility has been recorded in a specific database.

PARTICIPANTS

21 578 deliveries between 2000 and 2015 were included. Deliveries that lacked information regarding the labour induction status were excluded.

PRIMARY OUTCOME

Deliveries involving labour induction intervention.

RESULTS

Parity, maternal age, body mass index, gestational age and birth weight were all found to be important predictors of labour induction. Boosting method demonstrated the best discriminative performance (area under curve, AUC=0.75: 95% CI (0.73 to 0.76)) while logistic regression presented the least (AUC=0.71: 95% CI (0.70 to 0.73)). Random forest and boosting algorithms showed the highest net-benefits as per the decision curve analysis.

CONCLUSION

All of the machine learning algorithms performed well in predicting the likelihood of labour induction intervention. Further optimisation of these classifiers through hyperparameter tuning may result in an improved performance. Extensive research into the performance of other classifier algorithms is warranted.

C1431T Variant of PPARγ Is Associated with Preeclampsia in Pregnant Women

Peroxisome proliferator-activated receptor γ (PPARγ) is essential for placental development, whose SNPs have shown increased susceptibility to pregnancy-related diseases, such as preeclampsia. Our aim was to investigate the association between preeclampsia and three PPARγ SNPs (Pro12Ala, C1431T, and C681G), which together with nine clinical factors were used to build a pragmatic model for preeclampsia prediction. Data were collected from 1648 women from the EDEN cohort, of which 35 women had preeclamptic pregnancies, and the remaining 1613 women had normal pregnancies. Univariate analysis comparing preeclamptic patients to the control resulted in the SNP C1431T being the only factor significantly associated with preeclampsia (p < 0.05), with a confidence interval of 95% and odds ratio ranging from 4.90 to 8.75. On the other hand, three methods of multivariate feature selection highlighted seven features that could be potential predictors of preeclampsia: maternal C1431T and C681G variants, obesity, body mass index, number of pregnancies, primiparity, cigarette use, and education. These seven features were further used as input into eight different machine-learning algorithms to create predictive models, whose performances were evaluated based on metrics of accuracy and the area under the receiver operating characteristic curve (AUC). The boost tree-based model performed the best, with respective accuracy and AUC values of 0.971 ± 0.002 and 0.991 ± 0.001 in the training set and 0.951 and 0.701 in the testing set. A flowchart based on the boost tree model was constructed to depict the procedure for preeclampsia prediction. This final decision tree showed that the C1431T variant of PPARγ is significantly associated with susceptibility to preeclampsia. We believe that this final decision tree could be applied in the clinical prediction of preeclampsia in the very early stages of pregnancy.

Machine learning algorithms for prediction of health-related quality-of-life after surgery for mild degenerative cervical myelopathy

BACKGROUND

Degenerative cervical myelopathy (DCM) is the most common cause of spinal cord dysfunction worldwide. Current guidelines recommend management based on the severity of myelopathy, measured by the modified Japanese Orthopedic Association (mJOA) score. Patients with moderate to severe myelopathy, defined by an mJOA below 15, are recommended to undergo surgery. However, the management for mild myelopathy (mJOA between 15 and 17) is controversial since the response to surgery is more heterogeneous.

PURPOSE

To develop machine learning algorithms predicting phenotypes of mild myelopathy patients that would benefit most from surgery.

STUDY DESIGN

Retrospective subgroup analysis of prospectively collected data.

PATIENT SAMPLES

Data were obtained from 193 mild DCM patients who underwent surgical decompression and were enrolled in the multicenter AOSpine CSM clinical trials.

OUTCOME MEASURES

The mJOA score, an assessment of functional status, was used to isolate patients with mild DCM. The primary outcome measures were change from baseline for the Short Form-36 (SF-36) mental component summary (MCS) and physical component summary (PCS) at 1-year postsurgery. These changes were dichotomized according to whether they exceeded the minimal clinically important difference.

METHODS

The data were split into training (75%) and testing (25%) sets. Model predictors included baseline demographic variables and clinical presentation. Seven machine learning algorithms and a logistic regression model were trained and optimized using the training set, and their performances were evaluated using the testing set. For each outcome (improvement in MCS or PCS), the machine learning algorithm with the greatest area under the curve (AUC) on the training set was selected for further analysis.

RESULTS

The generalized boosted model (GBM) and earth models performed well in the prediction of significant improvement in MCS and PCS respectively, with AUCs of 0.72 to 0.78 on the training set. This performance was replicated on the testing set, in which the GBM and earth models showed AUCs of 0.77 and 0.78, respectively, as well as fair to good calibration across the predicted range of probabilities. Female patients with a low initial MCS were less likely to experience significant improvement in MCS than males. The presence of certain signs and symptoms (eg, lower limb spasticity, clumsy hands) were also predictive of worse outcome.

CONCLUSIONS

Machine learning models showed good predictive power and provided information about the phenotypes of mild DCM patients most likely to benefit from surgical intervention. Overall, machine learning may be a useful tool for management of mild DCM, though external validation and prospective analysis should be performed to better solidify its role.

Artificial intelligence in nutrition research: perspectives on current and future applications

Artificial intelligence (AI) is a rapidly evolving area that offers unparalleled opportunities of progress and applications in many healthcare fields. In this review, we provide an overview of the main and latest applications of AI in nutrition research and identify gaps to address to potentialize this emerging field. AI algorithms may help better understand and predict the complex and non-linear interactions between nutrition-related data and health outcomes, particularly when large amounts of data need to be structured and integrated, such as in metabolomics. AI-based approaches, including image recognition, may also improve dietary assessment by maximizing efficiency and addressing systematic and random errors associated with self-reported measurements of dietary intakes. Finally, AI applications can extract, structure and analyze large amounts of data from social media platforms to better understand dietary behaviours and perceptions among the population. In summary, AI-based approaches will likely improve and advance nutrition research as well as help explore new applications. However, further research is needed to identify areas where AI does deliver added value compared with traditional approaches, and other areas where AI is simply not likely to advance the field. Novelty: Artificial intelligence offers unparalleled opportunities of progress and applications in nutrition. There remain gaps to address to potentialize this emerging field.

Relative Performance of Machine Learning and Linear Regression in Predicting Quality of Life and Academic Performance of School Children in Norway: Data Analysis of a Quasi-Experimental Study

BACKGROUND

Machine learning techniques are increasingly being applied in health research. It is not clear how useful these approaches are for modeling continuous outcomes. Child quality of life is associated with parental socioeconomic status and physical activity and may be associated with aerobic fitness and strength. It is unclear whether diet or academic performance is associated with quality of life.

OBJECTIVE

The purpose of this study was to compare the predictive performance of machine learning techniques with that of linear regression in examining the extent to which continuous outcomes (physical activity, aerobic fitness, muscular strength, diet, and parental education) are predictive of academic performance and quality of life and whether academic performance and quality of life are associated.

METHODS

We modeled data from children attending 9 schools in a quasi-experimental study. We split data randomly into training and validation sets. Curvilinear, nonlinear, and heteroscedastic variables were simulated to examine the performance of machine learning techniques compared to that of linear models, with and without imputation.

RESULTS

We included data for 1711 children. Regression models explained 24% of academic performance variance in the real complete-case validation set, and up to 15% in quality of life. While machine learning techniques explained high proportions of variance in training sets, in validation, machine learning techniques explained approximately 0% of academic performance and 3% to 8% of quality of life. With imputation, machine learning techniques improved to 15% for academic performance. Machine learning outperformed regression for simulated nonlinear and heteroscedastic variables. The best predictors of academic performance in adjusted models were the child's mother having a master-level education (P<.001; β=1.98, 95% CI 0.25 to 3.71), increased television and computer use (P=.03; β=1.19, 95% CI 0.25 to 3.71), and dichotomized self-reported exercise (P=.001; β=2.47, 95% CI 1.08 to 3.87). For quality of life, self-reported exercise (P<.001; β=1.09, 95% CI 0.53 to 1.66) and increased television and computer use (P=.002; β=-0.95, 95% CI -1.55 to -0.36) were the best predictors. Adjusted academic performance was associated with quality of life (P=.02; β=0.12, 95% CI 0.02 to 0.22).

CONCLUSIONS

Linear regression was less prone to overfitting and outperformed commonly used machine learning techniques. Imputation improved the performance of machine learning, but not sufficiently to outperform regression. Machine learning techniques outperformed linear regression for modeling nonlinear and heteroscedastic relationships and may be of use in such cases. Regression with splines performed almost as well in nonlinear modeling. Lifestyle variables, including physical exercise, television and computer use, and parental education are predictive of academic performance or quality of life. Academic performance is associated with quality of life after adjusting for lifestyle variables and may offer another promising intervention target to improve quality of life in children.

Effect of a Real-Time Risk Score on 30-day Readmission Reduction in Singapore

OBJECTIVE

To develop a risk score for the real-time prediction of readmissions for patients using patient specific information captured in electronic medical records (EMR) in Singapore to enable the prospective identification of high-risk patients for enrolment in timely interventions.

METHODS

Machine-learning models were built to estimate the probability of a patient being readmitted within 30 days of discharge. EMR of 25,472 patients discharged from the medicine department at Ng Teng Fong General Hospital between January 2016 and December 2016 were extracted retrospectively for training and internal validation of the models. We developed and implemented a real-time 30-day readmission risk score generation in the EMR system, which enabled the flagging of high-risk patients to care providers in the hospital. Based on the daily high-risk patient list, the various interfaces and flow sheets in the EMR were configured according to the information needs of the various stakeholders such as the inpatient medical, nursing, case management, emergency department, and postdischarge care teams.

RESULTS

Overall, the machine-learning models achieved good performance with area under the receiver operating characteristic ranging from 0.77 to 0.81. The models were used to proactively identify and attend to patients who are at risk of readmission before an actual readmission occurs. This approach successfully reduced the 30-day readmission rate for patients admitted to the medicine department from 11.7% in 2017 to 10.1% in 2019 (p < 0.01) after risk adjustment.

CONCLUSION

Machine-learning models can be deployed in the EMR system to provide real-time forecasts for a more comprehensive outlook in the aspects of decision-making and care provision.

Prediction of Worse Functional Status After Surgery for Degenerative Cervical Myelopathy: A Machine Learning Approach

BACKGROUND

Surgical decompression for degenerative cervical myelopathy (DCM) is one of the mainstays of treatment, with generally positive outcomes. However, some patients who undergo surgery for DCM continue to show functional decline.

OBJECTIVE

To use machine learning (ML) algorithms to determine predictors of worsening functional status after surgical intervention for DCM.

METHODS

This is a retrospective analysis of prospectively collected data. A total of 757 patients enrolled in 2 prospective AO Spine clinical studies, who underwent surgical decompression for DCM, were analyzed. The modified Japanese Orthopedic Association (mJOA) score, a marker of functional status, was obtained before and 1 yr postsurgery. The primary outcome measure was the dichotomized change in mJOA at 1 yr according to whether it was negative (worse functional status) or non-negative. After applying an 80:20 training-testing split of the dataset, we trained, optimized, and tested multiple ML algorithms to evaluate algorithm performance and determine predictors of worse mJOA at 1 yr.

RESULTS

The highest-performing ML algorithm was a polynomial support vector machine. This model showed good calibration and discrimination on the testing data, with an area under the receiver operating characteristic curve of 0.834 (accuracy: 74.3%, sensitivity: 88.2%, specificity: 72.4%). Important predictors of functional decline at 1 yr included initial mJOA, male gender, duration of myelopathy, and the presence of comorbidities.

CONCLUSION

The reasons for worse mJOA are frequently multifactorial (eg, adjacent segment degeneration, tandem lumbar stenosis, ongoing neuroinflammatory processes in the cord). This study successfully used ML to predict worse functional status after surgery for DCM and to determine associated predictors.