Opening the black box: interpretable machine learning for predictor finding of metabolic syndrome

Predicting metabolic syndrome: Machine learning techniques for improved preventive medicine

Objectives: Metabolic syndrome (MetS) has a significant impact on health. MetS is the umbrella term for a group of interdependent metabolic threats that contribute to the emergence of diseases that can lead to death. This study was designed to better predict the risks associated with MetS to enable medical personnel to make more optimal preventive medical decisions. Study design: Data from a large hospital survey database was used to train data mining classification techniques to predict patient-level risk subsequent to extensive data engineering that included aggregating predictors from multiple visits. Methods: A prospective group of seemingly healthy volunteers from the database was studied based on data obtained during their regular annual health checkups. Results: After aggregating the variables over time, the findings indicated that the predictive power of our model outperformed methods presented in other studies (AUC = 0.947). Specific lifestyle factors were identified as contributing to MetS. Conclusion: Involvement to avoid recurring diseases can significantly decrease medical problems and treatment expenses. The findings emphasize the importance of using predictive tools in healthcare and preventive medicine. The results can be used for future prevention strategies that encourage lifestyle changes and implement directed medical treatment protocols to decrease the burden of illness.

Detecting Alzheimer's Disease Stages and Frontotemporal Dementia in Time Courses of Resting-State fMRI Data Using a Machine Learning Approach

Early, accurate diagnosis of neurodegenerative dementia subtypes such as Alzheimer's disease (AD) and frontotemporal dementia (FTD) is crucial for the effectiveness of their treatments. However, distinguishing these conditions becomes challenging when symptoms overlap or the conditions present atypically. Resting-state fMRI (rs-fMRI) studies have demonstrated condition-specific alterations in AD, FTD, and mild cognitive impairment (MCI) compared to healthy controls (HC). Here, we used machine learning to build a diagnostic classification model based on these alterations. We curated all rs-fMRIs and their corresponding clinical information from the ADNI and FTLDNI databases. Imaging data underwent preprocessing, time course extraction, and feature extraction in preparation for the analyses. The imaging features data and clinical variables were fed into gradient-boosted decision trees with fivefold nested cross-validation to build models that classified four groups: AD, FTD, HC, and MCI. The mean and 95% confidence intervals for model performance metrics were calculated using the unseen test sets in the cross-validation rounds. The model built using only imaging features achieved 74.4% mean balanced accuracy, 0.94 mean macro-averaged AUC, and 0.73 mean macro-averaged F1 score. It accurately classified FTD (F1 = 0.99), HC (F1 = 0.99), and MCI (F1 = 0.86) fMRIs but mostly misclassified AD scans as MCI (F1 = 0.08). Adding clinical variables to model inputs raised balanced accuracy to 91.1%, macro-averaged AUC to 0.99, macro-averaged F1 score to 0.92, and improved AD classification accuracy (F1 = 0.74). In conclusion, a multimodal model based on rs-fMRI and clinical data accurately differentiates AD-MCI vs. FTD vs. HC.

Towards the prediction of drug solubility in binary solvent mixtures at various temperatures using machine learning

Drug solubility is an important parameter in the drug development process, yet it is often tedious and challenging to measure, especially for expensive drugs or those available in small quantities. To alleviate these challenges, machine learning (ML) has been applied to predict drug solubility as an alternative approach. However, the majority of existing ML research has focused on the predictions of aqueous solubility and/or solubility at specific temperatures, which restricts the model applicability in pharmaceutical development. To bridge this gap, we compiled a dataset of 27,000 solubility datapoints, including solubility of small molecules measured in a range of binary solvent mixtures under various temperatures. Next, a panel of ML models were trained on this dataset with their hyperparameters tuned using Bayesian optimization. The resulting top-performing models, both gradient boosted decision trees (light gradient boosting machine and extreme gradient boosting), achieved mean absolute errors (MAE) of 0.33 for LogS (S in g/100 g) on the holdout set. These models were further validated through a prospective study, wherein the solubility of four drug molecules were predicted by the models and then validated with in-house solubility experiments. This prospective study demonstrated that the models accurately predicted the solubility of solutes in specific binary solvent mixtures under different temperatures, especially for drugs whose features closely align within the solutes in the dataset (MAE < 0.5 for LogS). To support future research and facilitate advancements in the field, we have made the dataset and code openly available. Scientific contribution Our research advances the state-of-the-art in predicting solubility for small molecules by leveraging ML and a uniquely comprehensive dataset. Unlike existing ML studies that predominantly focus on solubility in aqueous solvents at fixed temperatures, our work enables prediction of drug solubility in a variety of binary solvent mixtures over a broad temperature range, providing practical insights on the modeling of solubility for realistic pharmaceutical applications. These advancements along with the open access dataset and code support significant steps in the drug development process including new molecule discovery, drug analysis and formulation.

Prediction model of pressure injury occurrence in diabetic patients during ICU hospitalization--XGBoost machine learning model can be interpreted based on SHAP

BACKGROUND

The occurrence of pressure injury in patients with diabetes during ICU hospitalization can result in severe complications, including infections and non-healing wounds.

AIMS

The aim of this study was to predict the occurrence of pressure injury in ICU patients with diabetes using machine learning models.

STUDY DESIGN

In this study, LASSO regression was used for feature screening, XGBoost was employed for machine learning model construction, ROC curve analysis, calibration curve analysis, clinical decision curve analysis, sensitivity, specificity, accuracy, and F1 score were used for evaluating the model's performance.

RESULTS

Out of the 503 ICU patients with diabetes included in the study, pressure injury developed in 170 cases, resulting in an incidence rate of 33.8 %. The XGBoost model had a higher AUC for predicting pressure injury in patients with diabetes during ICU hospitalization (train: 0.896, 95 %CI: 0.863 to 0.929; test: 0.835, 95 % CI: 0.761-0.908). The importance of SHAP variables in the model from high to low was: 'Days in ICU', 'Mechanical Ventilation', 'Neutrophil Count', 'Consciousness', 'Glucose', and 'Warming Blanket'.

CONCLUSION

The XGBoost machine learning model we constructed has shown high performance in predicting the occurrence of pressure injury in ICU patients with diabetes. Additionally, the SHAP method enables the interpretation of the results provided by the machine learning model.

RELEVANCE TO CLINICAL PRACTICE

Improve the ability to predict the early occurrence of pressure injury in diabetic patients in the ICU. This will enable clinicians to intervene early and reduce the occurrence of complications.

Construction of a Diagnostic Model for Small Cell Lung Cancer Combining Metabolomics and Integrated Machine Learning

BACKGROUND

To date, no study has systematically explored the potential role of serum metabolites and lipids in the diagnosis of small cell lung cancer (SCLC). Therefore, we aimed to conduct a case-cohort study that included 191 cases of SCLC, 91 patients with lung adenocarcinoma, 82 patients with squamous cell carcinoma, and 97 healthy controls.

METHODS

Metabolomics and lipidomics were applied to analyze different metabolites and lipids in the serum of these patients. The SCLC diagnosis model (d-model) was constructed using an integrated machine learning technology and a training cohort (n = 323) and was validated in a testing cohort (n=138).

RESULTS

Eight metabolites, including 1-mristoyl-sn-glycero-3-phosphocholine, 16b-hydroxyestradiol, 3-phosphoserine, cholesteryl sulfate, D-lyxose, dioctyl phthalate, DL-lactate and Leu-Phe, were successfully selected to distinguish SCLC from controls. The d-model was constructed based on these 8 metabolites and showed improved diagnostic performance for SCLC, with the area under curve (AUC) of 0.933 in the training cohort and 0.922 in the testing cohort. Importantly, the d-model still had an excellent diagnostic performance after adjusting the stage and related clinical variables and, combined with the progastrin-releasing peptide (ProGRP), showed the best diagnostic performance with 0.975 of AUC for limited-stage patients.

CONCLUSION

This study is the first to analyze the difference between metabolomics and lipidomics and to construct a d-model to detect SCLC using integrated machine learning. This study may be of great significance for the screening and early diagnosis of SCLC patients.

Development and validation of a lifestyle risk index to screen for metabolic syndrome and its components in two multi-ethnic cohorts

BACKGROUND

Metabolic syndrome (MetS) is a precursor to cardiovascular diseases and type 2 diabetes. Existing MetS prediction models relied heavily on biochemical measures and those based on non-invasive predictors such as lifestyle behaviours were limited. We aim to (1) develop a weighted lifestyle risk index for MetS and (2) externally validate this index using two Asian-based cohorts in Singapore.

METHODS

Using data from the Multi-Ethnic Cohort (MEC) 1 (n = 2873, 41% male), multiple logistic regression was used to identify predictors associated with MetS. A weighted lifestyle risk index was generated using coefficients of the selected predictors in the development cohort (MEC1). Subsequently, the performance of the lifestyle risk index in predicting the occurrence of MetS within 10 years was assessed by discrimination and calibration in an external validation cohort (MEC2) (n = 6070, 43% male).

RESULTS

A lifestyle risk index for MetS with nine predictors was developed (age, sex, ethnicity, having a family history of diabetes, BMI, diet, physical activity, smoking status, and screen time). This index demonstrated acceptable discrimination in the development cohort [AUC (95% CI) = 0.74 (0.71, 0.76)] and the validation cohort [AUC (95% CI) = 0.79 (0.77, 0.81)].

CONCLUSION

This lifestyle risk index exhibits potential for risk stratification in population-based screening programmes. Future research could apply a similar methodology to develop disease-specific lifestyle risk indices using nationwide registry-based data.

Survival Prediction Model for Patients with Hepatocellular Carcinoma and Extrahepatic Metastasis Based on XGBoost Algorithm

Purpose

Accurate estimation of survival is of utmost importance in patients with hepatocellular carcinoma (HCC) and extrahepatic metastasis. This study aimed to develop a survival prediction model using real-world data.

Patients and Methods

A total of 993 patients with treatment-naïve HCC and extrahepatic metastasis were included from 13 Korean hospitals between 2013 and 2018. Patients were randomly divided into a training set (70.0%) and a test set (30.0%). The eXtreme Gradient Boosting (XGBoost) algorithm was applied to predict survival at 3, 6, and 12 months.

Results

The mean age of the patients was 60.8 ± 12.3 years, and 85.4% were male. During the study period, 96.1% died, and median survival duration was 4.0 months. In multivariate analysis, Child-Pugh class, number and size of tumors, presence of vascular or bile duct invasion, lung or bone metastasis, serum AFP, and primary anti-HCC treatment were associated with survival. We constructed a model for survival prediction based on the relevant variables, which is available online (https://metastatic-hcc.onrender.com/form). Our model demonstrated high performance, with areas under the receiver operating characteristic curves of 0.778, 0.794, and 0.784 at 3, 6, and 12 months, respectively. Feature importance analysis indicated that the primary anti-HCC treatment had the highest importance.

Conclusion

We developed a model to predict the survival of patients with HCC and extrahepatic metastasis, which demonstrated good discriminative ability. Our model would be helpful for personalized treatment and for improving the prognosis.

Machine Learning Approach for Metabolic Syndrome Diagnosis Using Explainable Data-Augmentation-Based Classification

Metabolic syndrome (MetS) is a cluster of risk factors including hypertension, hyperglycemia, dyslipidemia, and abdominal obesity. Metabolism-related risk factors include diabetes and heart disease. MetS is also linked to numerous cancers and chronic kidney disease. All of these variables raise medical costs. Developing a prediction model that can quickly identify persons at high risk of MetS and offer them a treatment plan is crucial. Early prediction of metabolic syndrome will highly impact the quality of life of patients as it gives them a chance for making a change to the bad habit and preventing a serious illness in the future. In this paper, we aimed to assess the performance of various algorithms of machine learning in order to decrease the cost of predictive diagnoses of metabolic syndrome. We employed ten machine learning algorithms along with different metaheuristics for feature selection. Moreover, we examined the effects of data augmentation in the prediction accuracy. The statistics show that the augmentation of data after applying feature selection on the data highly improves the performance of the classifiers.