Hospital management plays a pivotal role in ensuring the efficient delivery of medical services, especially in the face of challenges posed by pandemics such as COVID-19. This paper explores the application of machine learning techniques in addressing the challenge of hospitalization during pandemics. Leveraging a comprehensive dataset sourced from the Mexican government, various supervised learning algorithms including Random Forest, Gradient Boosting, Support Vector Machine, K-Nearest Neighbors, and Multilayer Perceptron are trained and evaluated to discern factors contributing to hospitalizations. Feature importance analysis and dimensionality reduction techniques are employed to enhance models predictive performance. The best model was Gradient Boosting algorithm with an accuracy of 85.63% and AUC score of 0.8696. The interpretability plots showed that pneumonia had a positive impact on the hospitalization prediction of the model. Our analysis indicates that women aged over 45 with pneumonia and concurrent COVID-19 exhibit the highest likelihood of hospitalization. This study underscores the potential of interpretable machine learning in aiding hospital managers to optimize resource allocation, hospitalization cases, and make data-driven decisions during pandemics.
{"title":"An interpretable machine learning model to predict hospitalizations","authors":"Hagar Elbatanouny , Hissam Tawfik , Tarek Khater , Anatoliy Gorbenko","doi":"10.1016/j.ceh.2025.03.004","DOIUrl":"10.1016/j.ceh.2025.03.004","url":null,"abstract":"<div><div>Hospital management plays a pivotal role in ensuring the efficient delivery of medical services, especially in the face of challenges posed by pandemics such as COVID-19. This paper explores the application of machine learning techniques in addressing the challenge of hospitalization during pandemics. Leveraging a comprehensive dataset sourced from the Mexican government, various supervised learning algorithms including Random Forest, Gradient Boosting, Support Vector Machine, K-Nearest Neighbors, and Multilayer Perceptron are trained and evaluated to discern factors contributing to hospitalizations. Feature importance analysis and dimensionality reduction techniques are employed to enhance models predictive performance. The best model was Gradient Boosting algorithm with an accuracy of 85.63% and AUC score of 0.8696. The interpretability plots showed that pneumonia had a positive impact on the hospitalization prediction of the model. Our analysis indicates that women aged over 45 with pneumonia and concurrent COVID-19 exhibit the highest likelihood of hospitalization. This study underscores the potential of interpretable machine learning in aiding hospital managers to optimize resource allocation, hospitalization cases, and make data-driven decisions during pandemics.</div></div>","PeriodicalId":100268,"journal":{"name":"Clinical eHealth","volume":"8 ","pages":"Pages 53-65"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143799502","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-11-02DOI: 10.1016/j.ceh.2025.10.001
I. Biliunaite , M.A. Adriaanse , A.P.Y. Hoogendoorn , A. Montvila , M.R. Crone , L.C. van Gestel
Healthcare professionals (HCPs) commonly see the potential of health apps for their patients, but in practice do not actively recommend them during consultation. As quality concerns have been identified as a key barrier, a health and wellness app assessment framework and related quality label was previously developed. Yet, even when health apps are of high quality, recommendation behavior may not necessarily follow due to other factors that are yet to be identified and targeted. The main aim of this study was to explore a wide range of HCP behavioral determinants and identify the key determinants of HCP app recommendation behavior. We used the TDF-checklist, which is based on the Theoretical Domains Framework (TDF), an evidence-based framework for the systematic assessment of behavioral determinants of HCP behavior, and adapted it to the study context. 290 Catalan HCPs filled in the survey. For all determinants, room for improvement (deviation from the maximum), relevance (correlation with anticipated behavior), and the potential for change (based on combining room for improvement and relevance) were assessed. A large majority of HCPs indicated they would recommend high-quality apps to their patients. Overall, HCPs were motivated, but more room for improvement was found for capability and opportunity-related domains. Anticipated recommendation behavior correlated strongest with motivational factors like beliefs about consequences and beliefs about capabilities. The potential for change was highest for nature of the behaviors (habit), beliefs about capabilities and knowledge. When implementing the label, efforts should focus on promoting habit formation for recommending high-quality apps, boosting confidence of HCPs, and providing further knowledge regarding health apps.
{"title":"Recommending high-quality health apps: Identifying key behavioral determinants of healthcare professional behavior","authors":"I. Biliunaite , M.A. Adriaanse , A.P.Y. Hoogendoorn , A. Montvila , M.R. Crone , L.C. van Gestel","doi":"10.1016/j.ceh.2025.10.001","DOIUrl":"10.1016/j.ceh.2025.10.001","url":null,"abstract":"<div><div>Healthcare professionals (HCPs) commonly see the potential of health apps for their patients, but in practice do not actively recommend them during consultation. As quality concerns have been identified as a key barrier, a health and wellness app assessment framework and related quality label was previously developed. Yet, even when health apps are of high quality, recommendation behavior may not necessarily follow due to other factors that are yet to be identified and targeted. The main aim of this study was to explore a wide range of HCP behavioral determinants and identify the key determinants of HCP app recommendation behavior. We used the TDF-checklist, which is based on the Theoretical Domains Framework (TDF), an evidence-based framework for the systematic assessment of behavioral determinants of HCP behavior, and adapted it to the study context. 290 Catalan HCPs filled in the survey. For all determinants, room for improvement (deviation from the maximum), relevance (correlation with anticipated behavior), and the potential for change (based on combining room for improvement and relevance) were assessed. A large majority of HCPs indicated they would recommend high-quality apps to their patients. Overall, HCPs were motivated, but more room for improvement was found for capability and opportunity-related domains. Anticipated recommendation behavior correlated strongest with motivational factors like beliefs about consequences and beliefs about capabilities. The potential for change was highest for nature of the behaviors (habit), beliefs about capabilities and knowledge. When implementing the label, efforts should focus on promoting habit formation for recommending high-quality apps, boosting confidence of HCPs, and providing further knowledge regarding health apps.</div></div>","PeriodicalId":100268,"journal":{"name":"Clinical eHealth","volume":"8 ","pages":"Pages 218-229"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145519367","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-04-28DOI: 10.1016/j.ceh.2025.04.004
Khadijeh Moulaei , Abbas Sheikhtaheri
Introduction
Smart diabetic shoes can be essential in preventing and monitoring foot ulcers. We developed a smart diabetic shoe to monitor pressure, temperature, and humidity and send the data to patients’ phones via Bluetooth for foot care. This study aimed to evaluate the usability of this smart diabetic shoe.
Methods
Seven patients were interviewed using a semi-structured interview. They were asked to use the shoes and application in different positions and then express their opinions.
Results
We identified a total number of 35 unique usability problems and recommendations. Hardware and software were responsible for 8 and 27 of them, respectively. The majority of the issues concerned the application. The most common software-related complaints raised by the participants were warning presentation, application appearance, and customization. Participants highlighted foot comfort as the most important concern among hardware-related issues.
Conclusion
By addressing various hardware and software issues—such as foot comfort, shoe design and layout, system performance, data collection, remote monitoring, and communication with healthcare providers—we can enhance the usability and overall experience of wearable devices for users.
{"title":"Usability evaluation of wearable technology: A pilot study on a smart diabetic shoe for foot care","authors":"Khadijeh Moulaei , Abbas Sheikhtaheri","doi":"10.1016/j.ceh.2025.04.004","DOIUrl":"10.1016/j.ceh.2025.04.004","url":null,"abstract":"<div><h3>Introduction</h3><div>Smart diabetic shoes can be essential in preventing and monitoring foot ulcers. We developed a smart diabetic shoe to monitor pressure, temperature, and humidity and send the data to patients’ phones via Bluetooth for foot care. This study aimed to evaluate the usability of this smart diabetic shoe.</div></div><div><h3>Methods</h3><div>Seven patients were interviewed using a semi-structured interview. They were asked to use the shoes and application in different positions and then express their opinions.</div></div><div><h3>Results</h3><div>We identified a total number of 35 unique usability problems and recommendations. Hardware and software were responsible for 8 and 27 of them, respectively. The majority of the issues concerned the application. The most common software-related complaints raised by the participants were warning presentation, application appearance, and customization. Participants highlighted foot comfort as the most important concern among hardware-related issues.</div></div><div><h3>Conclusion</h3><div>By addressing various hardware and software issues—such as foot comfort, shoe design and layout, system performance, data collection, remote monitoring, and communication with healthcare providers—we can enhance the usability and overall experience of wearable devices for users.</div></div>","PeriodicalId":100268,"journal":{"name":"Clinical eHealth","volume":"8 ","pages":"Pages 94-102"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143885978","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-08-07DOI: 10.1016/j.ceh.2025.08.001
Rajib Kumar Halder, Marzana Akter Lima, Mohammed Nasir Uddin, Md.Aminul Islam, Adri Saha
Breast cancer is a leading cause of morbidity and mortality among women worldwide, arising from malignant cell transformations in breast tissue. Early detection is paramount as it significantly improves survival rates and reduces the complexity and cost of treatment. Machine learning has revolutionized this field, providing more precise, efficient, and personalized diagnostic methods. Our research aims to develop a robust predictive model for breast cancer classification through rigorous preprocessing, diverse feature selection techniques, and advanced ensemble learning strategies. A central component of our methodology is the employment of a Stacking Classifier integrated with multiple base classifiers, optimized using RandomizedSearchCV to fine-tune hyperparameters. This process enhances the model’s accuracy, reliability, and generalizability. Significantly, our feature selection process involves three methodologies: filter, wrapper, and embedded methods. By applying these techniques, we identify the most critical features that are consistently selected across all methods. These features are then used to train the model, ensuring that our approach focuses on the most relevant data points for breast cancer classification. Utilizing the Wisconsin Breast Cancer Dataset from the UCI repository, which comprises 569 patient records, our model demonstrates exceptional performance. It achieves a perfect accuracy of 100% and an AUC-ROC of 1.00, indicating flawless sensitivity and specificity. The proposed framework was evaluated using two distinct datasets: the Wisconsin Prognostic Breast Cancer (WPBC) dataset and the Wisconsin Original Breast Cancer (WOBC) dataset. This model stands out for its potential to significantly enhance early detection and treatment strategies, marking a significant advance in applying machine learning to improve healthcare outcomes. Additionally, we have developed a user-friendly web app for breast cancer detection using our predictive model.
{"title":"Integrated feature selection-based stacking ensemble model using optimized hyperparameters to predict breast cancer with smart web application","authors":"Rajib Kumar Halder, Marzana Akter Lima, Mohammed Nasir Uddin, Md.Aminul Islam, Adri Saha","doi":"10.1016/j.ceh.2025.08.001","DOIUrl":"10.1016/j.ceh.2025.08.001","url":null,"abstract":"<div><div>Breast cancer is a leading cause of morbidity and mortality among women worldwide, arising from malignant cell transformations in breast tissue. Early detection is paramount as it significantly improves survival rates and reduces the complexity and cost of treatment. Machine learning has revolutionized this field, providing more precise, efficient, and personalized diagnostic methods. Our research aims to develop a robust predictive model for breast cancer classification through rigorous preprocessing, diverse feature selection techniques, and advanced ensemble learning strategies. A central component of our methodology is the employment of a Stacking Classifier integrated with multiple base classifiers, optimized using RandomizedSearchCV to fine-tune hyperparameters. This process enhances the model’s accuracy, reliability, and generalizability. Significantly, our feature selection process involves three methodologies: filter, wrapper, and embedded methods. By applying these techniques, we identify the most critical features that are consistently selected across all methods. These features are then used to train the model, ensuring that our approach focuses on the most relevant data points for breast cancer classification. Utilizing the Wisconsin Breast Cancer Dataset from the UCI repository, which comprises 569 patient records, our model demonstrates exceptional performance. It achieves a perfect accuracy of 100% and an AUC-ROC of 1.00, indicating flawless sensitivity and specificity. The proposed framework was evaluated using two distinct datasets: the Wisconsin Prognostic Breast Cancer (WPBC) dataset and the Wisconsin Original Breast Cancer (WOBC) dataset. This model stands out for its potential to significantly enhance early detection and treatment strategies, marking a significant advance in applying machine learning to improve healthcare outcomes. Additionally, we have developed a user-friendly web app for breast cancer detection using our predictive model.</div></div>","PeriodicalId":100268,"journal":{"name":"Clinical eHealth","volume":"8 ","pages":"Pages 146-161"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144828913","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Cyberchondria is defined as an excessive or repeated online health-related information-seeking behavior exacerbated by information overload and quarantine, resulting in amplified health anxiety. A total of 406 Lebanese participants, residing in Lebanon, participated in this cross-sectional study conducted between February and March 2022 to investigate the association between social media use and cyberchondria severity. Participants filled an online questionnaire assessing the severity of cyberchondria (via short Cyberchondria Severity Scale (CSS-12)), fear of COVID-19 (via the Fear of COVID-19 Scale (FCV–19S)), and social media use. The majority of recruited participants were females (76.6 %) with an average age of 30.87 ± 12.68 years. The average time spent on social media per day was 4.19 ± 2.86 h, and the mean scores per item were 2.27 ± 0.73 and 2 ± 0.71 of CSS-12 and Fear of COVID-19, respectively. Social media use for health-related information and considering health-related information from social media, google, and medical websites reliable, were found to be significantly associated with cyberchondria. The developed multiple linear regression model justified 23.3 % of the variation of cyberchondria severity score. Besides, social media use for health-related information (p-value < 0.001), Fear of COVID-19 (p-value < 0.001), and age (p-value = 0.046) were significantly associated with cyberchondria severity. This implies the importance of social media implementation in the health care field in the forms of e-medicine and telehealth.
{"title":"Association between social media use and cyberchondria during the COVID-19 pandemic: a cross-sectional study","authors":"Nadia Koleilat , Abir Ghosson , Adel Ghandour , Fatima Soufan , Hussein Kaddoura , Mohammad Jounblat , Saria Abdallah , Issam Shaarani","doi":"10.1016/j.ceh.2025.10.004","DOIUrl":"10.1016/j.ceh.2025.10.004","url":null,"abstract":"<div><div>Cyberchondria is defined as an excessive or repeated online health-related information-seeking behavior exacerbated by information overload and quarantine, resulting in amplified health anxiety. A total of 406 Lebanese participants, residing in Lebanon, participated in this cross-sectional study conducted between February and March 2022 to investigate the association between social media use and cyberchondria severity. Participants filled an online questionnaire assessing the severity of cyberchondria (via short Cyberchondria Severity Scale (CSS-12)), fear of COVID-19 (via the Fear of COVID-19 Scale (FCV–19S)), and social media use. The majority of recruited participants were females (76.6 %) with an average age of 30.87 ± 12.68 years. The average time spent on social media per day was 4.19 ± 2.86 h, and the mean scores per item were 2.27 ± 0.73 and 2 ± 0.71 of CSS-12 and Fear of COVID-19, respectively. Social media use for health-related information and considering health-related information from social media, google, and medical websites reliable, were found to be significantly associated with cyberchondria. The developed multiple linear regression model justified 23.3 % of the variation of cyberchondria severity score. Besides, social media use for health-related information (p-value < 0.001), Fear of COVID-19 (p-value < 0.001), and age (p-value = 0.046) were significantly associated with cyberchondria severity. This implies the importance of social media implementation in the health care field in the forms of e-medicine and telehealth.</div></div>","PeriodicalId":100268,"journal":{"name":"Clinical eHealth","volume":"8 ","pages":"Pages 230-239"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145519366","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2024-12-18DOI: 10.1016/j.ceh.2024.12.002
Da-Wei Yang , Hui-Fen Weng , Jing Li , Min-Jie Ju , Hao Wang , Yi-Chen Jia , Xiao-Dan Wang , Jia Fan , Zuo-qin Yan , Jian Zhou , Cui-Cui Chen , Yin-Zhou Feng , Xiao-Yan Chen , Dong-Ni Hou , Xing-Wei Lu , Wei Yang , Yin Wu , Zheng-Guo Chen , Tao Bai , Xiao-Han Hu , Yuan-Lin Song
Objectives
To explore the clinical prognosis and the risk factors of death from COVID-19 patients complicated with one of the three major comorbidities (coronary heart disease, diabetes, or hypertension) based on real-world data.
Methods
This single-centre retrospective real-world study investigated all in-hospital patients who were transferred to the Coronavirus Special Ward of the Elderly Center of Zhongshan Hospital from March to June 2022 with a positive COVID-19 virus nucleic acid test and with at least one of the three comorbidities (coronary heart disease, diabetes or hypertension). Clinical data and laboratory test results of eligible patients were collected. A multivariate logistic regression analysis was performed to explore the risk associated with the prognosis.
Results
For the 1,281 PCR-positive patients at the admission included in the analysis, the mean age was 70.5 ± 13.7 years, and 658 (51.4 %) were males. There were 1,092 (85.2 %) patients with hypertension, 477(37.2 %) patients with diabetes, and 124 (9.7 %) patients with coronary heart disease. The length of hospital stay (LOS) was 9.2 ± 5.1 days. Among all admitted patients,1112 (91.5 %) were fully recovered, 77 (6.9 %) were improved, and 29 (2.6 %) died. Over the hospitalization, 172 (13.4 %) PCR-positive patients experienced rebound COVID following initial recovery with a negative PCR test. A multivariate logistic regression analysis showed that vaccination had no protective effects in this study population; Paxlovid was associated with a lower risk of death(OR = 0.98, 95 % CI: 0.95–1.00). Whereas the presence of solid malignancies and nerve system disease were significantly associated with increased risk of death (OR = 1.04, 95 % CI:1.02–1.05; OR = 1.10, 95 % CI:1.05–1.14; OR = 1.08, 95 % CI:1.03–1.13; respectively).
Conclusion
The vast majority of the hospitalized COVID patients were fully recovered. Paxlovid was associated with a lower risk of death. In contrast, the presence of solid malignancies and nerve system disease and some treatments were all significantly associated with an increased risk of death.
{"title":"Clinical prognosis and risk factors of death for COVID-19 patients complicated with coronary heart disease/diabetes/hypertension-a retrospective, real-world study","authors":"Da-Wei Yang , Hui-Fen Weng , Jing Li , Min-Jie Ju , Hao Wang , Yi-Chen Jia , Xiao-Dan Wang , Jia Fan , Zuo-qin Yan , Jian Zhou , Cui-Cui Chen , Yin-Zhou Feng , Xiao-Yan Chen , Dong-Ni Hou , Xing-Wei Lu , Wei Yang , Yin Wu , Zheng-Guo Chen , Tao Bai , Xiao-Han Hu , Yuan-Lin Song","doi":"10.1016/j.ceh.2024.12.002","DOIUrl":"10.1016/j.ceh.2024.12.002","url":null,"abstract":"<div><h3>Objectives</h3><div>To explore the clinical prognosis and the risk factors of death from COVID-19 patients complicated with one of the three major comorbidities (coronary heart disease, diabetes, or hypertension) based on real-world data.</div></div><div><h3>Methods</h3><div>This single-centre retrospective real-world study investigated all in-hospital patients who were transferred to the Coronavirus Special Ward of the Elderly Center of Zhongshan Hospital from March to June 2022 with a positive COVID-19 virus nucleic acid test and with at least one of the three comorbidities (coronary heart disease, diabetes or hypertension). Clinical data and laboratory test results of eligible patients were collected. A multivariate logistic regression analysis was performed to explore the risk associated with the prognosis.</div></div><div><h3>Results</h3><div>For the 1,281 PCR-positive patients at the admission included in the analysis, the mean age was 70.5 ± 13.7 years, and 658 (51.4 %) were males. There were 1,092 (85.2 %) patients with hypertension, 477(37.2 %) patients with diabetes, and 124 (9.7 %) patients with coronary heart disease. The length of hospital stay (LOS) was 9.2 ± 5.1 days. Among all admitted patients,1112 (91.5 %) were fully recovered, 77 (6.9 %) were improved, and 29 (2.6 %) died. Over the hospitalization, 172 (13.4 %) PCR-positive patients experienced rebound COVID following initial recovery with a negative PCR test. A multivariate logistic regression analysis showed that vaccination had no protective effects in this study population; Paxlovid was associated with a lower risk of death(OR = 0.98, 95 % CI: 0.95–1.00). Whereas the presence of solid malignancies and nerve system disease were significantly associated with increased risk of death (OR = 1.04, 95 % CI:1.02–1.05; OR = 1.10, 95 % CI:1.05–1.14; OR = 1.08, 95 % CI:1.03–1.13; respectively).</div></div><div><h3>Conclusion</h3><div>The vast majority of the hospitalized COVID patients were fully recovered. Paxlovid was associated with a lower risk of death. In contrast, the presence of solid malignancies and nerve system disease and some treatments were all significantly associated with an increased risk of death.</div></div>","PeriodicalId":100268,"journal":{"name":"Clinical eHealth","volume":"8 ","pages":"Pages 26-31"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143563476","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-11-19DOI: 10.1016/j.ceh.2025.11.003
Putta Durga , Ali B.M. Ali , Deepthi Godavarthi , Sachi Nandan Mohanty , Shoira Formanova , M. Ijaz Khan
Parkinson’s disease (PD), a condition of the brain, causes accidental or intractable tendencies including shaking, stiffness, and issues with balance and coordination. In most cases, symptoms start mildly and get worse with time. Patients may have problems speaking and walking as the illness worsens. Additionally, they may exhibit altered mental and behavioral patterns, sleep disorders, sadness, memory difficulty, and exhaustion. In general, it is difficult to forecast sickness. Additionally, more than 25 % of PD diagnoses are incorrect because of the significant similarity between PD symptoms and other neurological symptoms. This prompted us to conduct a comparative literature review of how cutting-edge Machine Learning (ML) implementations are used in these methodologies with their corresponding datasets, including Logistic Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), K-nearest neighbors (KNN), Random Forest (RF), and Naïve Bayes (NB) classifiers. To increase accuracy, we have used multi-ensemble methods like the XGBoost Classifier and Ensemble (Majority Voting: RF & LSTM) are also used. Our results are contrasted with those from each study. The Static Spiral Test, which is used to identify tremors, performed significantly better in all experiments after applying XGBoost. As a result, it can be deduced that the multi-modal technique is efficient when used in conjunction with the ensemble method Xgboost classification (Extreme gradient boosting) and Ensemble (Majority Voting: RF & LSTM) that it offers a high accuracy of (95 %, and 96 %) in comparison to other classifier models. The approaches’ performance was assessed using a trustworthy dataset from the UCI ML repository.
{"title":"An ensemble model for detection of Parkinson’s disease by comparing numerous machine learning models and XGBoost based on vocal features","authors":"Putta Durga , Ali B.M. Ali , Deepthi Godavarthi , Sachi Nandan Mohanty , Shoira Formanova , M. Ijaz Khan","doi":"10.1016/j.ceh.2025.11.003","DOIUrl":"10.1016/j.ceh.2025.11.003","url":null,"abstract":"<div><div>Parkinson’s disease (PD), a condition of the brain, causes accidental or intractable tendencies including shaking, stiffness, and issues with balance and coordination. In most cases, symptoms start mildly and get worse with time. Patients may have problems speaking and walking as the illness worsens. Additionally, they may exhibit altered mental and behavioral patterns, sleep disorders, sadness, memory difficulty, and exhaustion. In general, it is difficult to forecast sickness. Additionally, more than 25 % of PD diagnoses are incorrect because of the significant similarity between PD symptoms and other neurological symptoms. This prompted us to conduct a comparative literature review of how cutting-edge Machine Learning (ML) implementations are used in these methodologies with their corresponding datasets, including Logistic Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), K-nearest neighbors (KNN), Random Forest (RF), and Naïve Bayes (NB) classifiers. To increase accuracy, we have used multi-ensemble methods like the XGBoost Classifier and Ensemble (Majority Voting: RF & LSTM) are also used. Our results are contrasted with those from each study. The Static Spiral Test, which is used to identify tremors, performed significantly better in all experiments after applying XGBoost. As a result, it can be deduced that the multi-modal technique is efficient when used in conjunction with the ensemble method Xgboost classification (Extreme gradient boosting) and Ensemble (Majority Voting: RF & LSTM) that it offers a high accuracy of (95 %, and 96 %) in comparison to other classifier models. The approaches’ performance was assessed using a trustworthy dataset from the UCI ML repository.</div></div>","PeriodicalId":100268,"journal":{"name":"Clinical eHealth","volume":"8 ","pages":"Pages 273-287"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145617646","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-11-02DOI: 10.1016/j.ceh.2025.10.003
Abir Chowdhury , Md.Mahbubur Rahman Druvo , Md.Shariful Islam , Khandaker Mohammad Mohi Uddin , Md Ashraf Uddin
Millions of people worldwide suffer greatly from osteoporosis, a chronic bone disease marked by decreased bone mass and structural degradation. Timely intervention and therapy of osteoporosis depend heavily on accurate early osteoporosis prediction. In the proposed method, use a chronic dataset of patient characteristics and risk variables to present a machine learning framework for osteoporosis prediction. Class imbalance is handled by the pipeline by utilizing synthetic minority over-sampling technique (SMOTE) and other data preprocessing techniques including scaling and normalization. Then, the data was split in an 80:20 ratio and seven features were selected by mutual information. Using an ensemble learning technique and also adjusted the hyperparameters of several classification algorithms such as random forest, k-nearest neighbors, support vector machine, XGBoost and logistic regression. XGBoost, the top-performing algorithm has an AUC score of 81.08%, showing excellent classification performance. Furthermore, the interpretability of the model was improved through the utilization of shapley additive explanations (SHAP) and local interpretable model-agnostic explanations (LIME) by the XGBoost. This facilitated a more profound comprehension of the fundamental elements propelling the prediction. At last, develops a web interface where patients can know about their own condition by it. According to the work, the suggested framework is a useful tool for osteoporosis early prognosis which could help medical practitioners make treatment decisions.
{"title":"An application to detect osteoporosis using ensemble Machine learning with hyperparameter tuning and model interpretability","authors":"Abir Chowdhury , Md.Mahbubur Rahman Druvo , Md.Shariful Islam , Khandaker Mohammad Mohi Uddin , Md Ashraf Uddin","doi":"10.1016/j.ceh.2025.10.003","DOIUrl":"10.1016/j.ceh.2025.10.003","url":null,"abstract":"<div><div>Millions of people worldwide suffer greatly from osteoporosis, a chronic bone disease marked by decreased bone mass and structural degradation. Timely intervention and therapy of osteoporosis depend heavily on accurate early osteoporosis prediction. In the proposed method, use a chronic dataset of patient characteristics and risk variables to present a machine learning framework for osteoporosis prediction. Class imbalance is handled by the pipeline by utilizing synthetic minority over-sampling technique (SMOTE) and other data preprocessing techniques including scaling and normalization. Then, the data was split in an 80:20 ratio and seven features were selected by mutual information. Using an ensemble learning technique and also adjusted the hyperparameters of several classification algorithms such as random forest, k-nearest neighbors, support vector machine, XGBoost and logistic regression. XGBoost, the top-performing algorithm has an AUC score of 81.08%, showing excellent classification performance. Furthermore, the interpretability of the model was improved through the utilization of shapley additive explanations (SHAP) and local interpretable model-agnostic explanations (LIME) by the XGBoost. This facilitated a more profound comprehension of the fundamental elements propelling the prediction. At last, develops a web interface where patients can know about their own condition by it. According to the work, the suggested framework is a useful tool for osteoporosis early prognosis which could help medical practitioners make treatment decisions.</div></div>","PeriodicalId":100268,"journal":{"name":"Clinical eHealth","volume":"8 ","pages":"Pages 177-200"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145465132","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-01-17DOI: 10.1016/j.ceh.2025.01.003
Luca Cossu , Francesco Prendin , Giacomo Cappon , David Herzig , Lia Bally , Andrea Facchinetti
Background
Post-bariatric hypoglycemia (PBH) is a severe and often overlooked complication of bariatric surgery (BS), characterized by dangerously low blood glucose levels after meals, particularly those high in carbohydrates. Unlike in Type 1 and Type 2 diabetes (T1D, T2D), where decision support systems (DSS) and continuous glucose monitoring (CGM) tools aid blood glucose management, no dedicated DSS exists for PBH. This leaves individuals vulnerable to recurrent, unpredictable hypoglycemia, posing significant health risks. To address this gap, we propose Glu4, an open-source software package designed to predict and notify users of impending PBH events using CGM data.
Methods
Glu4 employs a two-step approach to predict PBH. A run-to-run algorithm forecasts future glucose levels using past CGM data, identifying potential hypoglycemic events 30 min in advance. An intelligent alarm system alerts users when glucose levels are predicted to drop below a critical threshold, prompting preventive action. A pilot study involving three PBH patients collected real-time glucose data to validate the system’s predictive performance.
Results
The pilot study demonstrated that Glu4 reliably predicted impending hypoglycemia in all participants, providing timely alerts 30 min before glucose drops. The system showed a high specificity, with no false alarms being triggered during the monitoring period. The proactive notifications enabled participants to manage their glucose levels more effectively by taking preventive actions such as consuming rescue carbohydrates before the onset of severe hypoglycemia.
Conclusions
Glu4 represents a promising tool for managing PBH, leveraging CGM data to deliver accurate, timely alerts that enable proactive intervention. By improving safety and quality of life for individuals with PBH, Glu4 addresses a critical unmet need. Future work will focus on enhancing system capabilities and conducting larger-scale studies to validate its effectiveness and refine its usability for clinical adoption.
{"title":"Glu4: An open-source package for real-time forecasting and alerting post-bariatric hypoglycemia based on continuous glucose monitoring","authors":"Luca Cossu , Francesco Prendin , Giacomo Cappon , David Herzig , Lia Bally , Andrea Facchinetti","doi":"10.1016/j.ceh.2025.01.003","DOIUrl":"10.1016/j.ceh.2025.01.003","url":null,"abstract":"<div><h3>Background</h3><div>Post-bariatric hypoglycemia (PBH) is a severe and often overlooked complication of bariatric surgery (BS), characterized by dangerously low blood glucose levels after meals, particularly those high in carbohydrates. Unlike in Type 1 and Type 2 diabetes (T1D, T2D), where decision support systems (DSS) and continuous glucose monitoring (CGM) tools aid blood glucose management, no dedicated DSS exists for PBH. This leaves individuals vulnerable to recurrent, unpredictable hypoglycemia, posing significant health risks. To address this gap, we propose Glu4, an open-source software package designed to predict and notify users of impending PBH events using CGM data.</div></div><div><h3>Methods</h3><div>Glu4 employs a two-step approach to predict<!--> <!-->PBH. A run-to-run algorithm forecasts future glucose levels using past CGM data, identifying potential hypoglycemic events 30 min in advance. An intelligent alarm system alerts users when glucose levels are predicted to drop below a critical threshold, prompting preventive action. A pilot study involving three PBH patients collected real-time glucose data to validate the system’s predictive performance.</div></div><div><h3>Results</h3><div>The pilot study demonstrated that Glu4 reliably predicted impending hypoglycemia in all participants, providing timely alerts 30 min before glucose drops. The system showed a high specificity, with no false alarms being triggered during the monitoring period. The proactive notifications enabled participants to manage their glucose levels more effectively by taking preventive actions such as consuming rescue carbohydrates before the onset of severe hypoglycemia.</div></div><div><h3>Conclusions</h3><div>Glu4 represents a promising tool for managing PBH, leveraging CGM data to deliver accurate, timely alerts that enable proactive intervention. By improving safety and quality of life for individuals with PBH, Glu4 addresses a critical unmet need. Future work will focus on enhancing system capabilities and conducting larger-scale studies to validate its effectiveness and refine its usability for clinical adoption.</div></div>","PeriodicalId":100268,"journal":{"name":"Clinical eHealth","volume":"8 ","pages":"Pages 1-6"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143169850","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-11-26DOI: 10.1016/j.ceh.2025.11.001
Anika Tahsin Meem, Sifat Momen
Gastric and intestinal cancers are among the deadliest gastrointestinal diseases, necessitating precise organ segmentation for effective detection and treatment planning. Conventional deep-learning models, such as CNN-based U-Net architectures, struggle with long-range dependencies and complex anatomical variations. This study introduces OVT-Net (Optimized Vision Transformer Network), an innovative deep-learning framework integrating Swin Transformer blocks, EfficientNetB7, Adaptive Contextual Attention (ACA) module, Atrous Spatial Pyramid Fusion (ASPF), and Squeeze-and-Excite (SE) Networks. Unlike traditional architectures, OVT-Net employs a hybrid dual-encoder structure, combining EfficientNetB7 for low-level feature extraction and Swin Transformers for global context modeling, addressing intricate anatomical complexities and imaging variabilities of the gastrointestinal tract. The model is trained on 38,496 MRI/CT scans paired with RLE-encoded masks that contain structural and labeling inconsistencies. These inconsistencies are resolved through a comprehensive preprocessing pipeline incorporating path generation, label restructuring, and augmentation to improve generalizability. Experimental results demonstrate superior performance, with a Dice score of 0.9350, an IoU score of 0.9218, a BCE loss of 0.0716, and robust surface distance metrics (HD95 and ASSD), outperforming conventional segmentation methods. To enhance clinical applicability, Explainable AI (XAI) techniques, including Grad-CAM and Grad-CAM++, provide interpretability by highlighting critical regions, improving model transparency in decision-making. Furthermore, OVT-Net is deployed in a Django-based web application, facilitating real-time segmentation and classification with an average accuracy of 97.5 %. This research presents OVT-Net as a transformative AI-driven segmentation model, bridging advanced vision transformers with XAI for enhanced medical diagnostics. Its integration into real-world clinical settings paves the way for improved cancer detection and early intervention.
{"title":"OVT-Net: Semantic segmentation of gastrointestinal cancer using an optimized vision transformer model with explainable AI","authors":"Anika Tahsin Meem, Sifat Momen","doi":"10.1016/j.ceh.2025.11.001","DOIUrl":"10.1016/j.ceh.2025.11.001","url":null,"abstract":"<div><div>Gastric and intestinal cancers are among the deadliest gastrointestinal diseases, necessitating precise organ segmentation for effective detection and treatment planning. Conventional deep-learning models, such as CNN-based U-Net architectures, struggle with long-range dependencies and complex anatomical variations. This study introduces OVT-Net (Optimized Vision Transformer Network), an innovative deep-learning framework integrating Swin Transformer blocks, EfficientNetB7, Adaptive Contextual Attention (ACA) module, Atrous Spatial Pyramid Fusion (ASPF), and Squeeze-and-Excite (SE) Networks. Unlike traditional architectures, OVT-Net employs a hybrid dual-encoder structure, combining EfficientNetB7 for low-level feature extraction and Swin Transformers for global context modeling, addressing intricate anatomical complexities and imaging variabilities of the gastrointestinal tract. The model is trained on 38,496 MRI/CT scans paired with RLE-encoded masks that contain structural and labeling inconsistencies. These inconsistencies are resolved through a comprehensive preprocessing pipeline incorporating path generation, label restructuring, and augmentation to improve generalizability. Experimental results demonstrate superior performance, with a Dice score of 0.9350, an IoU score of 0.9218, a BCE loss of 0.0716, and robust surface distance metrics (HD95 and ASSD), outperforming conventional segmentation methods. To enhance clinical applicability, Explainable AI (XAI) techniques, including Grad-CAM and Grad-CAM++, provide interpretability by highlighting critical regions, improving model transparency in decision-making. Furthermore, OVT-Net is deployed in a Django-based web application, facilitating real-time segmentation and classification with an average accuracy of 97.5 %. This research presents OVT-Net as a transformative AI-driven segmentation model, bridging advanced vision transformers with XAI for enhanced medical diagnostics. Its integration into real-world clinical settings paves the way for improved cancer detection and early intervention.</div></div>","PeriodicalId":100268,"journal":{"name":"Clinical eHealth","volume":"8 ","pages":"Pages 288-313"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145736816","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号