Camilo Scherkl, Theresa Dierkes, Michael Metzner, David Czock, Hanna M Seidling, Walter E Haefeli, Andreas D Meid
{"title":"Towards a prescribing monitoring system for medication safety evaluation within electronic health records: a scoping review.","authors":"Camilo Scherkl, Theresa Dierkes, Michael Metzner, David Czock, Hanna M Seidling, Walter E Haefeli, Andreas D Meid","doi":"10.1186/s12911-025-03096-3","DOIUrl":"10.1186/s12911-025-03096-3","url":null,"abstract":"<p><strong>Background: </strong>Medical care can fail for various reasons: diseases can remain undetected and their severity misjudged, therapies can be incorrectly dosed or ineffective, and therapies can trigger new conditions or adverse drug reactions (ADR). To manage the complexity of changing patient circumstances, data-driven techniques play an increasingly important role in monitoring patient safety and treatment success. Therefore, clinical prediction models need to consider longitudinal factors (\"Prescribing Monitoring\") to ensure clinically meaningful results and avoid misclassification in the dynamic health situation of the individual patient.</p><p><strong>Methods: </strong>We have conducted a scoping review (OSF registration: https://doi.org/10.17605/OSF.IO/P93TZ ) on prediction models for ADR to collect potential use cases for Prescribing Monitoring. This review identified 2435 relevant studies in English that were published in MEDLINE or EMBASE. Two reviewers screened the records for inclusion, with a third reviewer making the final decision in the event of discrepancies. In order to derive recommendations on the way towards a Prescribing Monitoring system, the following elements were extracted and interpreted: the prediction models used, selection of candidate predictors, use of longitudinal factors, and model performance.</p><p><strong>Results: </strong>A total of 56 studies were included after the screening process. We identified the main areas of current research in ADR prediction, all covering clinically important outcomes. We identified Prescribing Monitoring use cases based on their potential to (i) make individual predictions considering specific patient characteristics, (ii) make longitudinal predictions in a near time frame, and (iii) make dynamic predictions by updating predictions with previous risk predictions and newly available data. As a further aside, we use hyperkalaemia as an example to discuss the framework for developing Prescribing Monitoring in an electronic health record (EHR).</p><p><strong>Conclusion: </strong>This scoping review provides an overview of the use of time-varying effects and longitudinal variables in current prediction model research. For application to clinical cases, prediction models should be developed, validated and implemented on this basis, so that time-dependent information can enable continuous monitoring of individual patients.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"244"},"PeriodicalIF":3.3,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12224581/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144552313","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Fabio Dennstädt, Maximilian Schmalfuss, Johannes Zink, Janna Hastings, Roberto Gaio, Max Schmerder, Nikola Cihoric, Paul Martin Putora
{"title":"A CDE-based data structure for radiotherapeutic decision-making in breast cancer.","authors":"Fabio Dennstädt, Maximilian Schmalfuss, Johannes Zink, Janna Hastings, Roberto Gaio, Max Schmerder, Nikola Cihoric, Paul Martin Putora","doi":"10.1186/s12911-025-03036-1","DOIUrl":"10.1186/s12911-025-03036-1","url":null,"abstract":"<p><strong>Background: </strong>The growing complexity of oncology and radiation therapy demands structured and precise data management strategies. The National Institutes of Health (NIH) have introduced Common Data Elements (CDEs) as a uniform approach to facilitate consistent data collection. However, there is currently a lack of a comprehensive set of CDEs for describing situations for and within radiation oncology. Specifically for breast cancer, where radiotherapeutic decision-making is complex and based on multiple diverse criteria, there is a clear need for more standardized data. Aim of this study was to create a CDE-based data structure for radiotherapeutic decision-making in breast cancer to promote structured data collection on the level of a local hospital.</p><p><strong>Methods: </strong>Between May 2023 and May 2024, we conducted a case study at the radiation therapy department of a local hospital to develop a CDE-based data structure for radiotherapeutic decision-making in breast cancer. Local Standard Operating Procedures (SOPs) were analyzed to identify relevant decision-making criteria used in clinical practice. Corresponding CDEs were identified, and a structured data framework based on these CDEs was created. The framework was translated into machine-readable JavaScript Object Notation (JSON) format. Six clinical practice guidelines of the American Society for Radiation Oncology (ASTRO) were analyzed as full text to evaluate how many guideline recommendations and corresponding decision-making criteria could be represented using our framework.</p><p><strong>Results: </strong>We identified 31 decision-making criteria from local SOPs, formalized into 46 CDEs. A hierarchical structure within an object-oriented data framework was created and converted into JSON format. 94 recommendations with mentioning of decision-making criteria in 216 cases were identified across the six ASTRO guidelines. In 151 cases (70.0%) the mentioned criterion could be presented with the data framework.</p><p><strong>Conclusions: </strong>The CDE-based data structure provides a standardized, machine-readable framework for documenting and exchanging radiotherapeutic decision-making data in breast cancer patients. While further refinement is needed for broader interoperability, this approach facilitates structured data collection, enhances IT integration and supports standardized communication across different stakeholders.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"220"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12211407/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144539064","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Data-driven decision making in patient management: a systematic review.","authors":"Guoliang Lyu","doi":"10.1186/s12911-025-03072-x","DOIUrl":"10.1186/s12911-025-03072-x","url":null,"abstract":"<p><strong>Introduction: </strong>Data-Driven Decision Making (DDDM) plays a pivotal role in healthcare, specifically patient management. This review aims to provide a comprehensive understanding of the technologies used in DDDM and provide a framework of how DDDM is involved in patient management.</p><p><strong>Methodology: </strong>This study follows the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) framework, studies from Web of Science, Pubmed, and Embase are screened for consideration. The inclusion criteria are outlined to identify studies on patient management utilizing DDDM.</p><p><strong>Result: </strong>The studies included in the review explore DDDM in patient management from data-driven approaches to decision making methods. In the former, artificial intelligence, together with other methods, is the dominant method utilized. As a comparison, the decision support system, Markov decision process, and shared decision making are exploited in the latter. Disease diagnosis and treatment was the most common area of patient management application along with precision medicine, patient care, nursing, and other related fields of patient management. A framework of how DDDM is involved in patient management was identified.</p><p><strong>Conclusion: </strong>While challenges such as data quality and interpretability exist, advantages of DDDM lie in unprecedented personalization, streamlined decision-making, and the potential for a future where technology complements healthcare expertise for more effective and patient-centered care. DDDM is not only a useful option for patient management but also to many other aspects of healthcare and the systems around healthcare.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"239"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12219683/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144539013","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Priority setting of teleoncology services, and their barriers, and facilitators from oncologists and patients perspectives: a mixed-methods study.","authors":"Sareh Keshvardoost, Kambiz Bahaadinbeigy, Soodabeh Shahidsales, Farhad Fatehi, Sharareh Rostam Niakan Kalhori","doi":"10.1186/s12911-025-03042-3","DOIUrl":"10.1186/s12911-025-03042-3","url":null,"abstract":"<p><strong>Background: </strong>Cancer is the second leading cause of death worldwide. Unlike the potential of telemedicine in the follow-up of chronic patients, many patients are still being examined in-person traditionally. This study aimed to prioritize teleoncology services and identify implementation barriers, and facilitators from oncologists' and patients' perspectives.</p><p><strong>Method: </strong>The research employed a mixed-methods approach, combining open-ended questions and interviews with both patients and oncologists. A total of 63 patients and 13 oncologists participated in the study. The data analysis utilized chi-square tests and descriptive statistics. Additionally, a qualitative content analysis was performed using MAXQDA 20 to analyze the data obtained from the open-ended questions and interviews.</p><p><strong>Result: </strong>According to oncologists, the prioritization of teleoncology was described as follows based on several factors such as screening protocols, prevalence rates and complexity of diagnosis: (1) breast/gastrointestinal, (2) respiratory/cervical/prostate/skin, (3) neurology/head and neck, (4) gynecological and genitourinary, (5) endocrine and neuroendocrine, musculoskeletal system, and (6) eye. Ten out of 13 (77%) oncologists identified remote consultation between physicians as their highest priority, while 11 of them (85%) considered follow-up as the second most important. Moreover, 12 oncologists (92%) ranked remote diagnosis as their third priority. The majority of patients (66.7%) expressed a preference for in-person visits. Three key factors influencing the choice of in-person visits for patients were: feeling inadequate care during telemedicine visits, lack of digital knowledge, and being in the active treatment period. For patients, education level and preference of treatment method (in-person versus telemedicine) showed statistically significant association (Value < 0.05).</p><p><strong>Conclusion: </strong>Telemedicine was most favorable for teleconsultation between treating physicians and consultants. Lack of insurance cover, limited internet coverage, lack of local guidance, and insufficient continuous education were expressed as the main challenges for teleoncology. Addressing these challenges and prioritizing the needs of specialists are key factors in the promotion of telemedicine.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"224"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12211455/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144539049","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Shuifa Sun, Qin Hu, Fengjiao Xu, Feng Hu, Yirong Wu, Ben Wang
{"title":"Medical named entity recognition based on domain knowledge and position encoding.","authors":"Shuifa Sun, Qin Hu, Fengjiao Xu, Feng Hu, Yirong Wu, Ben Wang","doi":"10.1186/s12911-025-03037-0","DOIUrl":"10.1186/s12911-025-03037-0","url":null,"abstract":"<p><p>A model for recognizing named entities in Chinese electronic medical records is proposed, focusing on accurate boundary detection, by leveraging medical domain knowledge and positional encoding. Medical domain-specific terms are integrated into a BERT module by a lexical adapter firstly. After pre-training, the model captures the dynamic character feature representation containing lexical information and boundary information. In the feature encoding module, Star-Transformer and BiLSTM are employed to extract local features and long-distance features respectively in order to generate the sequence's feature representation. Additionally, considering the influence of the relative position information between characters in the text on recognition results, Rotary Position Embedding (RoPE) is incorporated to improve Star-Transformer to enhance the ability of extracting semantic features. Experimental results on the CCKS2020 dataset show an improvement in the F1-score, reaching 85.78%. Compared to the baseline model, the F1-score increases by 2.96%. For the self-build breast cancer ultrasound report dataset, improvement is also observed, which proves the effectiveness and applicability of the model in medical field.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"235"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12220262/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144539036","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Francesco Branda, Pierangelo Veltri, Francesco Chiodo, Massimo Ciccozzi, Fabio Scarpa, Pietro Hiram Guzzi
{"title":"Computational modeling of infectious diseases: insights from network-based simulations on measles.","authors":"Francesco Branda, Pierangelo Veltri, Francesco Chiodo, Massimo Ciccozzi, Fabio Scarpa, Pietro Hiram Guzzi","doi":"10.1186/s12911-025-03063-y","DOIUrl":"10.1186/s12911-025-03063-y","url":null,"abstract":"<p><strong>Background: </strong>Computational modelling of disease spread is crucial for understanding the dynamics of infectious outbreaks and assessing the effectiveness of control measures. In particular, network-based models for disease spreading offer detailed, granular insights into heterogeneous interactions and enable dynamic simulation of intervention strategies. Therefore, they offer valuable insights into the factors influencing disease spread, enabling public health authorities to develop effective containment strategies. Vaccination is among the most impactful interventions in controlling disease spread and has proven essential in preventing the spread of infectious diseases such as measles. However, recent trends indicate a concerning decline in the fraction of vaccinated individuals in various populations, increasing the risk of outbreaks.</p><p><strong>Methods: </strong>In this study, we utilize computational simulations on graph-based models to analyze how vaccination affects the spread of infectious diseases. By representing populations as networks in which individuals (nodes) are connected by potential spread pathways (edges), we simulate different vaccination coverage scenarios and assess their impact on disease spread. Our simulations incorporate high and low vaccination coverage to reflect real-world trends and explore various conditions under which disease spread can be effectively blocked.</p><p><strong>Results: </strong>The results demonstrate that adequate vaccination coverage is critical for halting outbreaks, with a marked reduction in disease spread observed as the fraction of vaccinated individuals increases. Conversely, insufficient vaccination rates lead to widespread outbreaks, underscoring the importance of maintaining high vaccination levels to achieve herd immunity and prevent resurgence. These findings highlight the vital role of vaccination as a preventative tool and emphasize the potential risks posed by declining vaccination rates.</p><p><strong>Conclusion: </strong>This study provides a deeper understanding of how vaccination strategies can mitigate the spread of infectious diseases and serves as a reminder of the importance of maintaining robust immunization programs to protect public health.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"238"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12219886/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144538996","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Sanghoon Han, Yu-Rim Lee, Ji-Ho Lee, JinHee Jeon, Choongki Min, Kyungnam Kim, Donghoon Kim, Myung Pyo Kim, Young Mi Park, Uiri An, Kyoung Min Moon
{"title":"Deep learning-based cough classification using application-recorded sounds: a transfer learning approach with VGGish.","authors":"Sanghoon Han, Yu-Rim Lee, Ji-Ho Lee, JinHee Jeon, Choongki Min, Kyungnam Kim, Donghoon Kim, Myung Pyo Kim, Young Mi Park, Uiri An, Kyoung Min Moon","doi":"10.1186/s12911-025-03065-w","DOIUrl":"10.1186/s12911-025-03065-w","url":null,"abstract":"<p><strong>Background: </strong>Coughing sounds contain various bio-metric information with regards to respiratory diseases that can help in the assessment of respiratory diseases. While clinicians find coughs insightful, non-experts struggle to identify abnormalities in cough sounds. Furthermore, respiratory diseases has characterized by widespread health complications and elevated mortality rates, the development of early diagnostic systems is imperative for ensuring timely intervention and improving outcomes for both clinicians and patients. Accordingly, we propose a deep learning-based model for early diagnosis. To enhance the reliability of the training data, we utilized annotations provided by multiple medical specialists. Additionally, we examined how clinical expertise and diagnostic input influence the model's generalization performance.</p><p><strong>Methods: </strong>This study introduces a deep learning framework utilizing VGGish as a transfer learning model, enhanced with additional detection and classification networks. The detection model identifies cough events within recorded audio, and then the classification model determines whether a detected cough is normal or abnormal. Both models were trained on raw cough sound data collected via smartphones and labeled by medical experts through a rigorous inspection process.</p><p><strong>Results: </strong>Experimental evaluations demonstrated that the cough detection model achieved an average accuracy of 0.9883, while the cough classification model attained accuracies of 0.8417, 0.8629, and 0.8662 among dataset1, 2, and 3. To enhance interpretability, we applied Grad-CAM to visualize the features that influenced the model's decision-making. Model performance was further evaluated using the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC).</p><p><strong>Conclusions: </strong>Our proposed cough classification model has the potential to assist individuals with limited access to healthcare as well as medical professionals with limited experience in diagnosing cough-related conditions. By leveraging deep learning and smartphone-recorded cough sounds, this approach aims to enhance early detection and management of respiratory diseases.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"228"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12218819/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144539015","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Nonlinear association between visceral fat metabolism score and heart failure: insights from LightGBM modeling and SHAP-Driven feature interpretation in NHANES.","authors":"Ningyi Cheng, Yukun Chen, Lei Jin, Liangwan Chen","doi":"10.1186/s12911-025-03076-7","DOIUrl":"10.1186/s12911-025-03076-7","url":null,"abstract":"<p><strong>Objective: </strong>Using 2005-2018 NHANES data, this study examined the association between the visceral fat metabolism score (METS-VF) and heart failure (HF) prevalence in U.S. adults, leveraging machine learning (LightGBM/XGBoost) and SHAP for classfication performance evaluation and feature interpretation.</p><p><strong>Methods: </strong>After excluding missing data, 30,704 participants were analyzed via survey-weighted statistics, restricted cubic splines (RCS), stratified analyses, and multivariate logistic regression. Ensemble models were compared for HF classification, with SHAP quantifying feature importance.</p><p><strong>Results: </strong>HF patients exhibited higher METS-VF (7.35 ± 0.53 vs. 6.79 ± 0.72, P < 0.001) and worse cardiometabolic profiles. Multivariate adjustment revealed a 2.249-fold increased HF prevalence per 1-unit METS-VF increase (95% CI: 1.503-3.366, P < 0.001), with a nonlinear threshold effect (inflection point = 7.151; OR = 3.321, 95% CI: 3.464-8.494 for METS-VF ≥ 7.151). Obesity (BMI ≥ 30 kg/m²) amplified the association (OR = 5.857). LightGBM outperformed logistic regression in classification (AUC = 0.964 vs. 0.907), with SHAP identifying METS-VF as the top contributor (importance weight = 18.6%), surpassing hypertension (10.8%) and coronary artery disease (11.7%). Correlations validated METS-VF as a composite index of visceral adiposity and metabolic dysfunction (waist circumference r = 0.43, high-density lipoprotein cholesterol r = - 0.38, all P < 0.001).</p><p><strong>Conclusion: </strong>METS-VF is independently and nonlinearly associated with HF prevalence, particularly in obese individuals. Machine learning enhances predictive accuracy by capturing complex interactions, while SHAP-based interpretability establishes METS-VF as a key biomarker integrating metabolic-adipose abnormalities, offering a novel target for personalized HF prevention.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"223"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12211878/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144539048","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Development and validation of machine learning classifiers for predicting treatment-needed retinopathy of prematurity.","authors":"Nasser Shoeibi, Majid Abrishami, Seyedeh Maryam Hosseini, Mohammad-Reza Ansari-Astaneh, Razieh Farrahi, Bahareh Gharib, Fatemeh Neghabi, Mojtaba Abrishami, Mehdi Sakhaee, Mehrdad Motamed Shariati","doi":"10.1186/s12911-025-03057-w","DOIUrl":"10.1186/s12911-025-03057-w","url":null,"abstract":"<p><strong>Background: </strong>This study aims to design and evaluate various supervised machine-learning models for identifying premature infants who require treatment based on demographic data and clinical findings from screening examinations.</p><p><strong>Methods: </strong>We conducted a retrospective review of medical records for infants screened for retinopathy of prematurity (ROP) at our clinic over the past decade. We extracted demographic and clinical data, including eleven features: sex, maternal education, paternal education, birth weight, gestational age, ROP stage, zone of retinal involvement, age at examination, weight at examination, and CPR. We developed and assessed several classifiers: logistic regression (LR), decision tree (DT), support vector machine (SVM), naïve Bayes (NB), K-nearest neighbors (KNN), XGBoost, artificial neural networks (ANN), and random forest (RF). The target variable was defined as whether the neonate received any treatment during the follow-up period.</p><p><strong>Results: </strong>Our analysis included data from 9,692 infants. Among the machine learning models evaluated, the XGBoost and ANN models achieved the highest accuracy at 96%. In terms of sensitivity (recall), the NB model exhibited the lowest false negative rate, indicating the highest sensitivity (0.99). In the context of premature neonates, accurately diagnosing those who require treatment is crucial. Therefore, from a clinical perspective, prioritizing a model with the lowest false negative rate may be more beneficial than selecting one based solely on the highest accuracy.</p><p><strong>Conclusion: </strong>While AI can enhance decision-making processes by providing real-time risk assessments, these tools must be used to augment-not replace-clinical judgment. Clinicians must remain involved in interpreting model outputs and making final treatment decisions based on a holistic understanding of each patient's unique circumstances.</p><p><strong>Clinical trial number: </strong>Not applicable.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"221"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12211727/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144539016","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jean Digitale, Deborah Franzon, Jin Ge, Charles McCulloch, Mark J Pletcher, Efstathios D Gennatas
{"title":"Expert-augmented machine learning for predicting extubation readiness in the pediatric intensive care unit.","authors":"Jean Digitale, Deborah Franzon, Jin Ge, Charles McCulloch, Mark J Pletcher, Efstathios D Gennatas","doi":"10.1186/s12911-025-03070-z","DOIUrl":"10.1186/s12911-025-03070-z","url":null,"abstract":"<p><strong>Background: </strong>Determining extubation readiness in pediatric intensive care units (PICU) is challenging. We used expert-augmented machine learning (EAML), a method that combines machine learning with human expert knowledge, to predict successful extubation.</p><p><strong>Methods: </strong>We extracted electronic health record data from patients in two PICUs. Data from patients in one unit was split into 80% training and 20% test, while patients in the other served as an external test set. EAML begins by training RuleFit, which converts gradient-boosted trees into decision rules. Then, expert clinicians were asked to assess the relative probability of successful extubation of the subgroup defined by each rule compared with the entire sample. The rules were ranked in order of increasing chance of successful extubation according to (1) the RuleFit model and (2) clinician assessment, and differences between the two rankings were calculated. The initial RuleFit model was then regularized based on these differences, producing the EAML model.</p><p><strong>Results: </strong>The RuleFit model selected 46 rules; we surveyed 25 clinician experts to provide feedback on them. All clinicians worked in a PICU setting and were from multiple disciplines; over half (56%) had > 5 years of PICU experience. As expected, the added regularization slightly lowered performance of EAML compared with RuleFit in the internal test set, although the difference was not statistically significant (RuleFit AUC = 0.817 vs. best-performing EAML model AUC = 0.814, difference = 0.003, 95% CI of difference = -0.009, 0.003). EAML had superior performance in the external test set (RuleFit AUC = 0.791 vs. best-performing EAML model AUC = 0.799, difference = 0.007, 95% CI of difference = 0.002, 0.013).</p><p><strong>Conclusions: </strong>When creating a model to predict successful extubation in PICU patients, incorporating expert knowledge directly into the model construction process via EAML produced a model more generalizable to an external test set.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"232"},"PeriodicalIF":3.3,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12220236/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144539020","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}