Harnessing machine learning models for non-invasive pre-diabetes screening in children and adolescents

Interpretable machine learning method to predict the risk of pre-diabetes using a national-wide cross-sectional data: evidence from CHNS

OBJECTIVE

The incidence of Type 2 Diabetes Mellitus (T2DM) continues to rise steadily, significantly impacting human health. Early prediction of pre-diabetic risks has emerged as a crucial public health concern in recent years. Machine learning methods have proven effective in enhancing prediction accuracy. However, existing approaches may lack interpretability regarding underlying mechanisms. Therefore, we aim to employ an interpretable machine learning approach utilizing nationwide cross-sectional data to predict pre-diabetic risk and quantify the impact of potential risks.

METHODS

The LASSO regression algorithm was used to conduct feature selection from 30 factors, ultimately identifying nine non-zero coefficient features associated with pre-diabetes, including age, TG, TC, BMI, Apolipoprotein B, TP, leukocyte count, HDL-C, and hypertension. Various machine learning algorithms, including Extreme Gradient Boosting (XGBoost), Random Forest (RF), Support Vector Machine (SVM), Naive Bayes (NB), Artificial Neural Networks (ANNs), Decision Trees (DT), and Logistic Regression (LR), were employed to compare predictive performance. Employing an interpretable machine learning approach, we aimed to enhance the accuracy of pre-diabetes risk prediction and quantify the impact and significance of potential risks on pre-diabetes.

RESULTS

From the China Health and Nutrition Survey (CHNS) data, a cohort of 8,277 individuals was selected, exhibiting a disease prevalence of 7.13%. The XGBoost model demonstrated superior performance with an AUC value of 0.939, surpassing RF, SVM, DT, ANNs, Naive Bayes, and LR models. Additionally, Shapley Additive Explanation (SHAP) analysis indicated that age, BMI, TC, ApoB, TG, hypertension, TP, HDL-C, and WBC may serve as risk factors for pre-diabetes.

CONCLUSION

The constructed model comprises nine easily accessible predictive factors, which prove highly effective in forecasting the risk of pre-diabetes. Concurrently, we have quantified the specific impact of each predictive factor on the risk and ranked them based on their influence. This result may serve as a convenient tool for early identification of individuals at high risk of pre-diabetes, providing effective guidance for preventing the progression of pre-diabetes to T2DM.

Learning from the machine: is diabetes in adults predicted by lifestyle variables? A retrospective predictive modelling study of NHANES 2007-2018

OBJECTIVES

This study aimed to compare the performance of five machine learning algorithms to predict diabetes mellitus based on lifestyle factors (diet and physical activity).

DESIGN

Retrospective cross-sectional predictive modelling study.

SETTING

This study was conducted using publicly available data from the National Health and Nutrition Examination Survey (NHANES), a nationally representative survey designed to assess the health and nutritional status of the US population.

PARTICIPANTS

We analysed data from 29 509 non-pregnant adults who participated in NHANES between 2007 and 2018.

PRIMARY AND SECONDARY OUTCOME MEASURES

The primary outcome was the prediction of type 2 diabetes mellitus (T2DM) by self-reported responses based on machine learning models. The performance of five machine learning algorithms (logistic regression, support vector machine, random forest, XGBoost and CatBoost) was evaluated using accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and the area under the receiver operating characteristic curve (AUC). The secondary outcome measures were feature importance and model performance comparison.

RESULTS

XGBoost exhibited the highest overall predictive performance (AUC 0.8168), followed by random forest and logistic regression (AUCs around 0.79). In terms of accuracy, logistic regression, XGBoost and random forest performed similarly at approximately 85%. While most models demonstrated high specificity (>97%), the SVM stood out for having the highest sensitivity (58.57%), although with a lower accuracy (62.44%). This trade-off underscores the strength of SVM in identifying more true-positive cases, though at the cost of lower overall classification precision. The random forest model, despite having lower sensitivity (7.15%), provided one of the most balanced performances in terms of specificity and interpretability.

CONCLUSION

The results support the use of machine learning models, particularly XGBoost, for early identification of individuals at risk for T2DM. Despite their limited sensitivity, the high specificity and accuracy underscore these models' potential for non-invasive risk assessment. This study is innovative in its integration of machine learning algorithms to predict type 2 diabetes based solely on non-invasive, easily accessible lifestyle and anthropometric variables, demonstrating the potential of data-driven models for early risk assessment without requiring laboratory tests. Despite the lower sensitivity observed in most models, their high specificity makes them valuable for early screening in clinical and public health settings, where they can be complemented with follow-up assessments or ensemble approaches that optimise the balance between sensitivity and specificity for improved risk stratification.

Hybrid statistical and machine-learning approach to hearing-loss identification based on an oversampling technique

BACKGROUND AND OBJECTIVES

Hearing loss is a crucial global health hazard exerting considerable social and physiological effects on spoken language and cognition. Patients affected by this condition may experience social and professional hardships that dominate occupational injuries. Therefore, the identification of the features of recessive hearing loss is important for clinicians to prevent further disease progression. This work aimed to develop a hybrid statistical and machine-learning approach as a decision-support mechanism. We expect the proposed model to help predict hearing-loss disorders and support clinical diagnosis.

METHODS

A three-phase hybrid approach was proposed to implement classification models. A stepwise method and a random forest (RF) technique were utilized as filters during feature selection. Phase I involved reducing the number of input variables and selecting the most influential features. Phase II included the use of an oversampling technique called synthetic minority oversampling technique (SMOTE) to oversample the minority class and balance the sample size between the target and nontarget classes. Phase III focused on the final model selection based on three supervised classification models, namely, the logistic regression, multilayer perceptron, and support vector machine (SVM), for the target identification and prediction of the case of interest (i.e., hearing loss).

RESULTS

The analysis of phase I involved the selection and acquisition of three and seven features through the stepwise technique and RF method, respectively. The SMOTE technique alleviated the imbalanced data issue and improved the predictive capability substantially in phase II and III. Accordingly, in terms of accuracy, precision, recall, and F1 score, our empirical results demonstrated that the proposed hybrid approach involving the SVM method combined with a stepwise technique was competitive against the logistic model featuring all variables. Furthermore, the SVM models that cooperated with the stepwise and RF technique showed superiority to other approaches in terms of the area under the curve (AUC).

CONCLUSION

Compared with multivariate models, the hybrid approach combining the SVM method coupled with a stepwise technique and/or an RF technique is an excellent alternative with a higher efficiency. This approach requires fewer predictors in the model and can be competitive in terms of the accuracy, precision, recall, F1 score, and AUC. This work highlights the potential of hybrid statistical and machine-learning approaches. Our model can be used as a screening tool for upfront forecasting in clinical practice. The proposed hybrid approach also demonstrates a powerful capability to identify vital features and predict hearing loss.

Development and Validation of Machine Learning Models for Identifying Prediabetes and Diabetes in Normoglycemia

BACKGROUND

Prediabetes and diabetes are both abnormal states of glucose metabolism (AGM) that can lead to severe complications. Early detection of AGM is crucial for timely intervention and treatment. However, fasting blood glucose (FBG) as a mass population screening method may fail to identify some individuals who are actually AGM but with normoglycemia. This study aimed to develop and validate machine learning (ML) models to identify AGM among individuals with normoglycemia using routine health check-up indicators.

METHODS

According to the American Diabetes Association (ADA) criteria, participants with normoglycemia (FBG ≤ 5.6 mmol/L) were collected from 2019 to 2023, and then divided into AGM and Normal groups using glycosylated haemoglobin (HbA1c) 5.7% as the threshold. Data from 2019 to 2022 were divided into training and internal validation sets at a 7:3 ratio, while data from 2023 were used as the external validation set. Seven ML algorithms-including logistic regression (LR), random forest (RF), support vector machine (SVM), extreme gradient boosting machine, multilayer perceptron (MLP), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost)-were used to build models for identifying AGM in normoglycemia population. Model performance was evaluated using the area under the receiver operating characteristic curve (auROC) and the precision-recall curve (auPR). The feature contributions to the optimal model was visualised using the SHapley Additive exPlanations (SHAP). Finally, an intuitive and user-friendly interactive interface was developed.

RESULTS

A total of 59,259 participants were finally enroled in this study, and then divided into the training set of 32,810, the internal validation set of 14,060, and the external validation set of 12,389. The Catboost model outperformed the others with auROC of 0.806 and 0.794 for the internal and external validation set, respectively. Age was the most important feature influencing the performance of the CatBoost model, followed by fasting blood glucose, red blood cells, haemoglobin, body mass index, and triglyceride-glucose.

CONCLUSION

A well-performed ML model to identify AGM in the normoglycemia population was built, offering significant potential for early intervention and treatment of AGM that would otherwise have been missed.

Multiomics and eXplainable artificial intelligence for decision support in insulin resistance early diagnosis: A pediatric population-based longitudinal study

Pediatric obesity can drastically heighten the risk of cardiometabolic alterations later in life, with insulin resistance standing as the cornerstone linking adiposity to the increased cardiovascular risk. Puberty has been pointed out as a critical stage after which obesity-associated insulin resistance is more difficult to revert. Timely prediction of insulin resistance in pediatric obesity is therefore vital for mitigating the risk of its associated comorbidities. The construction of effective and robust predictive systems for a complex health outcome like insulin resistance during the early stages of life demands the adoption of longitudinal designs for more causal inferences, and the integration of factors of varying nature involved in its onset. In this work, we propose an eXplainable Artificial Intelligence-based decision support pipeline for early diagnosis of insulin resistance in a longitudinal cohort of 90 children. For that, we leverage multi-omics (genomics and epigenomics) and clinical data from the pre-pubertal stage. Different data layers combinations, pre-processing techniques (missing values, feature selection, class imbalance, etc.), algorithms, training procedures were considered following good practices for Machine Learning. SHapley Additive exPlanations were provided for specialists to understand both the decision-making mechanisms of the system and the impact of the features on each automatic decision, an essential issue in high-risk areas such as this one where system decisions may affect people's lives. The system showed a relevant predictive ability (AUC and G-mean of 0.92). A deep exploration, both at the global and the local level, revealed promising biomarkers of insulin resistance in our population, highlighting classical markers, such as Body Mass Index z-score or leptin/adiponectin ratio, and novel ones such as methylation patterns of relevant genes, such as HDAC4, PTPRN2, MATN2, RASGRF1 and EBF1. Our findings highlight the importance of integrating multi-omics data and following eXplainable Artificial Intelligence trends when building decision support systems.

Machine Learning Prediction of Prediabetes in a Young Male Chinese Cohort with 5.8-Year Follow-Up

The identification of risk factors for future prediabetes in young men remains largely unexamined. This study enrolled 6247 young ethnic Chinese men with normal fasting plasma glucose at the baseline (FPGbase), and used machine learning (Mach-L) methods to predict prediabetes after 5.8 years. The study seeks to achieve the following: 1. Evaluate whether Mach-L outperformed traditional multiple linear regression (MLR). 2. Identify the most important risk factors. The baseline data included demographic, biochemistry, and lifestyle information. Two models were built, where Model 1 included all variables and Model 2 excluded FPGbase, since it had the most profound effect on prediction. Random forest, stochastic gradient boosting, eXtreme gradient boosting, and elastic net were used, and the model performance was compared using different error metrics. All the Mach-L errors were smaller than those for MLR, thus Mach-L provided the most accurate results. In descending order of importance, the key factors for Model 1 were FPGbase, body fat (BF), creatinine (Cr), thyroid stimulating hormone (TSH), WBC, and age, while those for Model 2 were BF, white blood cell, age, TSH, TG, and LDL-C. We concluded that FPGbase was the most important factor to predict future prediabetes. However, after removing FPGbase, WBC, TSH, BF, HDL-C, and age were the key factors after 5.8 years.

The Efficacy of Machine Learning Models for Predicting the Prognosis of Heart Failure: A Systematic Review and Meta-Analysis

INTRODUCTION

Heart failure (HF) is a major global public health concern. The application of machine learning (ML) to identify individuals at high risk and enable early intervention is a promising approach for improving HF prognosis. We aim to systematically evaluate the performance and value of ML models for predicting HF prognosis.

METHODS

PubMed, Web of Science, Scopus, and Embase online databases were searched up to April 30, 2023, to identify studies on the use of ML models to predict HF prognosis. HF prognosis primarily encompasses readmission and mortality. The meta-analysis was conducted by MedCalc software. Subgroup analyses include grouping based on types of ML models, time intervals, sample sizes, the number of predictive variables, validation methods, whether to conduct hyperparameter optimization and calibration, data set partitioning methods.

RESULTS

A total of 31 studies were included. The most common ML models were random forest, boosting, support vector machine, neural network. The area under the receiver operating characteristic curve (AUC) for predicting HF readmission was 0.675 (95% CI: 0.651-0.699, p < 0.001), and the AUC for predicting HF mortality was 0.790 (95% CI: 0.765-0.816, p < 0.001). Subgroup analyses revealed that models with the prediction time interval of 1 year, sample sizes ≥10,000, the number of predictive variables ≥100, external validation, hyperparameter tuning, calibration adjustment, and data set partitioning using 10-fold cross-validation exhibited favorable performance within their respective subgroups.

CONCLUSION

The performance of ML models in predicting HF readmission is relatively poor, while its performance in predicting HF mortality is moderate. The quality of the relevant studies is generally low, it is essential to enhance the predictive capabilities of ML models through targeted improvements in practical applications.

Artificial intelligence in paediatric endocrinology: conflict or cooperation

Artificial intelligence (AI) in medicine is transforming healthcare by automating system tasks, assisting in diagnostics, predicting patient outcomes and personalising patient care, founded on the ability to analyse vast datasets. In paediatric endocrinology, AI has been developed for diabetes, for insulin dose adjustment, detection of hypoglycaemia and retinopathy screening; bone age assessment and thyroid nodule screening; the identification of growth disorders; the diagnosis of precocious puberty; and the use of facial recognition algorithms in conditions such as Cushing syndrome, acromegaly, congenital adrenal hyperplasia and Turner syndrome. AI can also predict those most at risk from childhood obesity by stratifying future interventions to modify lifestyle. AI will facilitate personalised healthcare by integrating data from 'omics' analysis, lifestyle tracking, medical history, laboratory and imaging, therapy response and treatment adherence from multiple sources. As data acquisition and processing becomes fundamental, data privacy and protecting children's health data is crucial. Minimising algorithmic bias generated by AI analysis for rare conditions seen in paediatric endocrinology is an important determinant of AI validity in clinical practice. AI cannot create the patient-doctor relationship or assess the wider holistic determinants of care. Children have individual needs and vulnerabilities and are considered in the context of family relationships and dynamics. Importantly, whilst AI provides value through augmenting efficiency and accuracy, it must not be used to replace clinical skills.

Approaching Personalized Medicine: The Use of Machine Learning to Determine Predictors of Mortality in a Population with SARS-CoV-2 Infection

The COVID-19 pandemic demonstrated the need to develop strategies to control a new viral infection. However, the different characteristics of the health system and population of each country and hospital would require the implementation of self-systems adapted to their characteristics. The objective of this work was to determine predictors that should identify the most severe patients with COVID-19 infection. Given the poor situation of the hospitals in the first wave, the analysis of the data from that period with an accurate and fast technique can be an important contribution. In this regard, machine learning is able to objectively analyze data in hourly sets and is used in many fields. This study included 291 patients admitted to a hospital in Spain during the first three months of the pandemic. After screening seventy-one features with machine learning methods, the variables with the greatest influence on predicting mortality in this population were lymphocyte count, urea, FiO2, potassium, and serum pH. The XGB method achieved the highest accuracy, with a precision of >95%. Our study shows that the machine learning-based system can identify patterns and, thus, create a tool to help hospitals classify patients according to their severity of illness in order to optimize admission.

Artificial intelligence for automated detection of diabetic foot ulcers: A real-world proof-of-concept clinical evaluation

OBJECTIVE

Conduct a multicenter proof-of-concept clinical evaluation to assess the accuracy of an artificial intelligence system on a smartphone for automated detection of diabetic foot ulcers.

METHODS

The evaluation was undertaken with patients with diabetes (n = 81) from September 2020 to January 2021. A total of 203 foot photographs were collected using a smartphone, analysed using the artificial intelligence system, and compared against expert clinician judgement, with 162 images showing at least one ulcer, and 41 showing no ulcer. Sensitivity and specificity of the system against clinician decisions was determined and inter- and intra-rater reliability analysed.

RESULTS

Predictions/decisions made by the system showed excellent sensitivity (0.9157) and high specificity (0.8857). Merging of intersecting predictions improved specificity to 0.9243. High levels of inter- and intra-rater reliability for clinician agreement on the ability of the artificial intelligence system to detect diabetic foot ulcers was also demonstrated (Kα > 0.8000 for all studies, between and within raters).

CONCLUSIONS

We demonstrate highly accurate automated diabetic foot ulcer detection using an artificial intelligence system with a low-end smartphone. This is the first key stage in the creation of a fully automated diabetic foot ulcer detection and monitoring system, with these findings underpinning medical device development.

Application of Machine Learning Models for Early Detection and Accurate Classification of Type 2 Diabetes

Early detection of diabetes is essential to prevent serious complications in patients. The purpose of this work is to detect and classify type 2 diabetes in patients using machine learning (ML) models, and to select the most optimal model to predict the risk of diabetes. In this paper, five ML models, including K-nearest neighbor (K-NN), Bernoulli Naïve Bayes (BNB), decision tree (DT), logistic regression (LR), and support vector machine (SVM), are investigated to predict diabetic patients. A Kaggle-hosted Pima Indian dataset containing 768 patients with and without diabetes was used, including variables such as number of pregnancies the patient has had, blood glucose concentration, diastolic blood pressure, skinfold thickness, body insulin levels, body mass index (BMI), genetic background, diabetes in the family tree, age, and outcome (with/without diabetes). The results show that the K-NN and BNB models outperform the other models. The K-NN model obtained the best accuracy in detecting diabetes, with 79.6% accuracy, while the BNB model obtained 77.2% accuracy in detecting diabetes. Finally, it can be stated that the use of ML models for the early detection of diabetes is very promising.

A machine learning-based diagnosis modelling of type 2 diabetes mellitus with environmental metal exposure

BACKGROUND AND OBJECTIVE

Increasing and compelling evidence has been proved that urinary and dietary metal exposure are underappreciated but potentially modifiable biomarkers for type 2 diabetes mellitus (T2DM). The aims of this study were (1) to identify the key potential biomarkers which contributed to T2DM with effective and parsimonious features and (2) to assess the utility of baseline variables and metal exposure in the diagnosis of T2DM.

METHODS

Based on the National Health and Nutrition Examination Survey (NHANES), we selected 9822 screening records with 82 significant variables covering demographics, lifestyle, anthropometric measures, diet and metal exposure for this study. Combining extreme gradient boosting (XGBoost), random forest and light gradient boosting machine (lightGBM), a soft voting ensemble model was proposed to measure the importance of 82 features. With this soft voting ensemble model and variance inflation factor (VIF), strong multicollinear features with low importance scores were further removed from candidate biomarkers. Then, a soft voting ensemble classifier was adopted to demonstrate the efficiency of the proposed feature selection method.

RESULTS

With the novel feature selection method, 12 baseline variables and 3 metal variables were selected to detect patients at risk for T2DM in our study. For metal variables, the dietary copper (Cu), urinary cadmium (Cd) and urinary mercury (Hg) metals were selected as the most remarkable metal exposure and the corresponding P-values were all less than 0.05. In a classification model of T2DM with 12 baseline biomarkers, the addition of 3 metal exposure improved the classification accuracy of T2DM from a traditional area under the curve (AUC) 0.792 of the receiver operating characteristic (ROC) to an AUC 0.847.

CONCLUSIONS

This was the first demonstration of T2DM classification with machine learning under urinary and dietary metal exposure. Improved prediction precision illustrated the effectiveness of the proposed machine learning-based diagnosis model facilitated lifestyle/dietary intervention for T2DM prevention.

Risk Prediction for the Development of Hyperuricemia: Model Development Using an Occupational Health Examination Dataset

OBJECTIVE

Hyperuricemia has become the second most common metabolic disease in China after diabetes, and the disease burden is not optimistic.

METHODS

We used the method of retrospective cohort studies, a baseline survey completed from January to September 2017, and a follow-up survey completed from March to September 2019. A group of 2992 steelworkers was used as the study population. Three models of Logistic regression, CNN, and XG Boost were established to predict HUA incidence in steelworkers, respectively. The predictive effects of the three models were evaluated in terms of discrimination, calibration, and clinical applicability.

RESULTS

The training set results show that the accuracy of the Logistic regression, CNN, and XG Boost models was 84.4, 86.8, and 86.6, sensitivity was 68.4, 72.3, and 81.5, specificity was 82.0, 85.7, and 86.8, the area under the ROC curve was 0.734, 0.724, and 0.806, and Brier score was 0.121, 0.194, and 0.095, respectively. The XG Boost model effect evaluation index was better than the other two models, and similar results were obtained in the validation set. In terms of clinical applicability, the XG Boost model had higher clinical applicability than the Logistic regression and CNN models.

CONCLUSION

The prediction effect of the XG Boost model was better than the CNN and Logistic regression models and was suitable for the prediction of HUA onset risk in steelworkers.