BMC Medical Informatics and Decision Making最新文献

{"title":"Secondary use of health records for prediction, detection, and treatment planning in the clinical decision support system: a systematic review.","authors":"Dipendra Pant, Øystein Nytrø, Bennett L Leventhal, Carolyn Clausen, Kaban Koochakpour, Line Stien, Odd Sverre Westbye, Roman Koposov, Thomas Brox Røst, Thomas Frodl, Norbert Skokauskas","doi":"10.1186/s12911-025-03021-8","DOIUrl":"10.1186/s12911-025-03021-8","url":null,"abstract":"<p><strong>Background: </strong>This study aims to understand how secondary use of health records can be done for prediction, detection, treatment recommendations, and related tasks in clinical decision support systems.</p><p><strong>Methods: </strong>Articles mentioning the secondary use of EHRs for clinical utility, specifically in prediction, detection, treatment recommendations, and related tasks in decision support were reviewed. We extracted study details, methods, tools, technologies, utility, and performance.</p><p><strong>Results: </strong>We found that secondary uses of EHRs are primarily retrospective, mostly conducted using records from hospital EHRs, EHR data networks, and warehouses. EHRs vary in type and quality, making it critical to ensure their completeness and quality for clinical utility. Widely used methods include machine learning, statistics, simulation, and analytics. Secondary use of health records can be applied in any area of medicine. The selection of data, cohorts, tools, technology, and methods depends on the specific clinical utility.</p><p><strong>Conclusion: </strong>The process for secondary use of health records should include three key steps: 1. Validation of the quality of EHRs, 2. Use of methods, tools, and technologies with proactive training, and 3. Multidimensional assessment of the results and their usefulness.</p><p><strong>Trial registration: </strong>PROSPERO registration number CRD42023409582.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"190"},"PeriodicalIF":3.3,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12083156/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144085927","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":"Federated SPARQL query performance evaluation for exploring disease model mouse: combining gene expression, orthology, and disease knowledge graphs.","authors":"Tatsuya Kushida, Tarcisio Mendes de Farias, Ana C Sima, Christophe Dessimoz, Hirokazu Chiba, Frederic B Bastian, Hiroshi Masuya","doi":"10.1186/s12911-025-03013-8","DOIUrl":"10.1186/s12911-025-03013-8","url":null,"abstract":"<p><strong>Background: </strong>The RIKEN BRC develops and maintains the RIKEN BioResource MetaDatabase to help users explore appropriate target bioresources for their experiments and prepare precise and high-quality data infrastructures. The Swiss Institute of Bioinformatics develops two databases across multi-species for the study of gene expression and orthology: Bgee and Orthologous MAtrix (OMA, an orthology database).</p><p><strong>Methods: </strong>This study combines the RIKEN BioResource data with Resource Description Framework (RDF) datasets from Bgee, a gene expression database, the OMA, the DisGeNET, a human gene-disease association, Mouse Genome Informatics (MGI), UniProt, and four disease ontologies in the RIKEN BioResource MetaDatabase. Our aim is to evaluate the distributed SPARQL query performance when exploring which model organisms are most appropriate for specific medical science research applications across the aforementioned interoperable datasets. More precisely in our biomedical use cases, we investigate disease-related genes, as well as anatomical parts where these genes are expressed and subsequently identify appropriate bioresource candidates available for specific disease research applications.</p><p><strong>Results: </strong>We illustrate the above through two use cases targeting either Alzheimer's disease or melanoma. We identified 14 Alzheimer's disease-related genes that were expressed in the prefrontal cortex (e.g., APP and APOE) and 55 RIKEN bioresources, which were genetically modified mice related to these genes, predicted to be relevant to Alzheimer's disease research. Furthermore, executing a transitive search for the Uberon terms by using the Property Paths function, we identified 14 melanoma-related genes (e.g., HRAS and PTEN), and 12 anatomical parts in which these genes were expressed, such as the \"skin of limb\" as an example. Finally, we compared the performance of the federated SPARQL query via the remote Bgee SPARQL endpoint with the performance of a centralized SPARQL query using the Bgee dataset as part of the RIKEN BioResource MetaDatabase.</p><p><strong>Conclusions: </strong>As a result, we confirmed that the performance of the federated approach degraded. We concluded that we reduced the degradation of the query performance of the federated approach from the BioResource MetaDatabase to the SIB by refining the transferred data through a subquery and enhancing the server specifications thereby optimizing the triple store query evaluation.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 Suppl 1","pages":"189"},"PeriodicalIF":3.3,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12082848/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144085930","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":"Deep learning health space model for ordered responses.","authors":"Chanhee Lee, Taesung Park","doi":"10.1186/s12911-025-03026-3","DOIUrl":"10.1186/s12911-025-03026-3","url":null,"abstract":"<p><strong>Background: </strong>As personalized medicine becomes more prevalent, the objective measurement and visualization of an individual's health status are becoming increasingly crucial. However, as the dimensions of data collected from each individual increase, this task becomes more challenging. The Health Space (HS) model provides a statistical framework for visualizing an individual's health status on biologically meaningful axes. In our previous study, we developed HS models using statistical models such as logistic regression model (LRM) and the proportional odds model (POM). However, these statistical HS models are limited in their ability to accommodate complex non-linear biological relationships.</p><p><strong>Methods: </strong>In order to model complex non-linear biological relationship, we developed deep learning HS models. Specifically, we formulated five distinct deep learning HS models: four standard binary deep neural networks (DNNs) for binary outcomes and one deep ordinal neural network (DONN) that accounts for the ordinality of the dependent variable. We trained these models using 32,140 samples from the Korea National Health and Nutrition Examination Survey (KNHANES) and validated them with data from the Ewha-Boramae cohort (862 samples) and the Korea Association Resource (KARE) project (3,199 samples).</p><p><strong>Results: </strong>The proposed deep learning HS models were compared with the existing statistical HS model based on the POM. Deep learning HS model using DONN demonstrated the best performance in discriminating health status in both the training and external datasets.</p><p><strong>Conclusion: </strong>We developed deep learning HS models to capture complex non-linear biological relationships in HS and compared their performance with our previously best-performing statistical HS model. The deep learning HS models show promise as effective tools for objectively and meaningfully visualizing an individual's health status.</p><p><strong>Clinical trial number: </strong>Not applicable.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"191"},"PeriodicalIF":3.3,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12084989/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144085985","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":"A meta-analysis of the diagnostic test accuracy of artificial intelligence predicting emergency department dispositions.","authors":"Kuang-Ming Kuo, Chao Sheng Chang","doi":"10.1186/s12911-025-03010-x","DOIUrl":"10.1186/s12911-025-03010-x","url":null,"abstract":"<p><strong>Background: </strong>The rapid advancement of Artificial Intelligence (AI) has led to its widespread application across various domains, showing encouraging outcomes. Many studies have utilized AI to forecast emergency department (ED) disposition, aiming to forecast patient outcomes earlier and to allocate resources better; however, a dearth of comprehensive review literature exists to assess the objective performance standards of these predictive models using quantitative evaluations. This study aims to conduct a meta-analysis to assess the diagnostic accuracy of AI in predicting ED disposition, encompassing admission, critical care, and mortality.</p><p><strong>Methods: </strong>Multiple databases, including Scopus, Springer, ScienceDirect, PubMed, Wiley, Sage, and Google Scholar, were searched until December 31, 2023, to gather relevant literature. Risk of bias was assessed using the Prediction Model Risk of Bias Assessment Tool. Pooled estimates of sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC) were calculated to evaluate AI's predictive performance. Sub-group analyses were performed to explore covariates affecting AI predictive model performance.</p><p><strong>Results: </strong>The study included 88 articles possessed with 117 AI models, among which 39, 45, and 33 models predicted admission, critical care, and mortality, respectively. The reported statistics for sensitivity, specificity, and AUROC represent pooled summary measures derived from the component studies included in this meta-analysis. AI's summary sensitivity, specificity, and AUROC for predicting admission were 0.81 (95% Confidence Interval [CI] 0.74-0.86), 0.87 (95% CI 0.81-0.91), and 0.87 (95% CI 0.84-0.93), respectively. For critical care, the values were 0.86 (95% CI 0.79-0.91), 0.89 (95% CI 0.83-0.93), and 0.93 (95% CI 0.89-0.95), respectively, and for mortality, they were 0.85 (95% CI 0.80-0.89), 0.94 (95% CI 0.90-0.96), and 0.93 (95% CI 0.89-0.96), respectively. Emergent sample characteristics and AI techniques showed evidence of significant covariates influencing the heterogeneity of AI predictive models for ED disposition.</p><p><strong>Conclusions: </strong>The meta-analysis indicates promising performance of AI in predicting ED disposition, with certain potential for improvement, especially in sensitivity. Future research could explore advanced AI techniques such as ensemble learning and cross-validation with hyper-parameter tuning to enhance predictive model efficacy.</p><p><strong>Trial registration: </strong>This systematic review was not registered with PROSPERO or any other similar registry because the review was completed prior to the opportunity for registration, and PROSPERO currently does not accept registrations for reviews that are already completed. We are committed to transparency and have adhered to best practices in systematic review methodology throughout this study.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"187"},"PeriodicalIF":3.3,"publicationDate":"2025-05-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12082892/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144076114","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":"Developing a multiomics data-based mathematical model to predict colorectal cancer recurrence and metastasis.","authors":"Bing Li, Ming Xiao, Rong Zeng, Le Zhang","doi":"10.1186/s12911-025-03012-9","DOIUrl":"10.1186/s12911-025-03012-9","url":null,"abstract":"<p><strong>Background: </strong>Colorectal cancer is the fourth most deadly cancer, with a high mortality rate and a high probability of recurrence and metastasis. Since continuous examinations and disease monitoring for patients after surgery are currently difficult to perform, it is necessary for us to develop a predictive model for colorectal cancer metastasis and recurrence to improve the survival rate of patients.</p><p><strong>Results: </strong>Previous studies mostly used only clinical or radiological data, which are not sufficient to explain the in-depth mechanism of colorectal cancer recurrence and metastasis. Therefore, this study proposes such a multiomics data-based predictive model for the recurrence and metastasis of colorectal cancer. LR, SVM, Naïve-bayes and ensemble learning models are used to build this predictive model.</p><p><strong>Conclusions: </strong>The experimental results indicate that our proposed multiomics data-based ensemble learning model effectively predicts the recurrence and metastasis of colorectal cancer.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 Suppl 2","pages":"188"},"PeriodicalIF":3.3,"publicationDate":"2025-05-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12082861/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144076122","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":"A knowledge-based clinical decision support system for personalized health examination items in China: design and evaluation.","authors":"Dan Wu, Jiye An, Shan Nan, Yutong She, Huilong Duan, Ning Deng","doi":"10.1186/s12911-025-03019-2","DOIUrl":"https://doi.org/10.1186/s12911-025-03019-2","url":null,"abstract":"<p><strong>Background: </strong>Health examination identifies risk factors and diseases at an early stage through a series of health examination items. In China, however, the incidence of consulting services for health examination items is low and the current health examination item package is insufficiently personalized. Therefore, we created and evaluated a clinical decision support system (CDSS) for personalized health examination items.</p><p><strong>Methods: </strong>An ontology with the data properties as the core design was created to guide the knowledge expression. A knowledge graph composed of ontology-guided property graphs was developed to provide rich and clear decision-making knowledge. The system, including the web for primary care clinicians and the app for participants, was constructed to directly assist primary care clinicians through personalized and interpretable health examination item recommendations. The enter rate and mapping rate were created to evaluate the system's capability to process input health feature data. The two-step expert evaluation was designed to assess whether recommendations with several health examination items were appropriate for participants. The system recommendations and existing packages were compared to the expert's gold standard.</p><p><strong>Results: </strong>There were 15 classes, 2-level class hierarchies, 3 types of object properties, and 16 types of data properties in the health examination item recommendation ontology. Several different data properties could express a piece of complex decision-making knowledge and reduce the number of classes. There were 584 classes, 781 object properties, and 1094 data properties in the knowledge graph. Retrospective data from 70 participants, with a total of 472 health features, were selected for system evaluation. The ontology can cover 96.2% of the health features. 56.4% health features entered into the system had corresponding health examination items. The precision and recall of the system were 96.3% and 84.8%, and the packages were 72.5% and 69.1%.</p><p><strong>Conclusions: </strong>The performance of this system was close to experts and outperformed the current impersonalized health examination item packages. This system could improve the personalization of health examination items and the health examination consultation services, and promote participants' engagement in the health examination.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"183"},"PeriodicalIF":3.3,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12079905/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144076113","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":"Application of artificial intelligence medical imaging aided diagnosis system in the diagnosis of pulmonary nodules.","authors":"Ya Yang, Pan Wang, Chengzhou Yu, Jing Zhu, Jinping Sheng","doi":"10.1186/s12911-025-03009-4","DOIUrl":"https://doi.org/10.1186/s12911-025-03009-4","url":null,"abstract":"<p><p>The application of artificial intelligence (AI) technology has realized the transformation of people's production and lifestyle, and also promoted the rapid development of the medical field. At present, the application of intelligence in the medical field is increasing. Using its advanced methods and technologies of AI, this paper aims to realize the integration of medical imaging-aided diagnosis system and AI, which is helpful to analyze and solve the loopholes and errors of traditional artificial diagnosis in the diagnosis of pulmonary nodules. Drawing on the principles and rules of image segmentation methods, the construction and optimization of a medical image-aided diagnosis system is carried out to realize the precision of the diagnosis system in the diagnosis of pulmonary nodules. In the diagnosis of pulmonary nodules carried out by traditional artificial and medical imaging-assisted diagnosis systems, 231 nodules with pathology or no change in follow-up for more than two years were also tested in 200 cases. The results showed that the AI software detected a total of 881 true nodules with a sensitivity of 99.10% (881/889). The radiologists detected 385 true nodules with a sensitivity of 43.31% (385/889). The sensitivity of AI software in detecting non-calcified nodules was significantly higher than that of radiologists (99.01% vs 43.30%, P < 0.001), and the difference was statistically significant.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"185"},"PeriodicalIF":3.3,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12080151/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144076116","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":"Classification of lung cancer severity using gene expression data based on deep learning.","authors":"Ali Bou Nassif, Nour Ayman Abujabal, Aya Alchikh Omar","doi":"10.1186/s12911-025-03011-w","DOIUrl":"https://doi.org/10.1186/s12911-025-03011-w","url":null,"abstract":"<p><p>Lung cancer is one of the most prevalent diseases affecting people and is a main factor in the rising death rate. Recently, Machine Learning (ML) and Deep Learning (DL) techniques have been utilized to detect and classify various types of cancer, including lung cancer. In this research, a DL model, specifically a Convolutional Neural Network (CNN), is proposed to classify lung cancer stages for two types of lung cancer (LUAD and LUSC) using a gene dataset. Evaluating and validating the performance of the proposed model required addressing some common challenges in gene datasets, such as class imbalance and overfitting, due to the low number of samples and the high number of features. These issues were mitigated by deeply analyzing the gene dataset and lung cancer stages from a medical perspective, along with extensive research and experiments. As a result, the optimized CNN model using F-test feature selection method, achieved high classification accuracies of approximately 93.94% for LUAD and 88.42% for LUSC.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"184"},"PeriodicalIF":3.3,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12080119/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144076119","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 application of an early prediction model for risk of bloodstream infection based on real-world study.","authors":"Xiefei Hu, Shenshen Zhi, Yang Li, Yuming Cheng, Haiping Fan, Haorong Li, Zihao Meng, Jiaxin Xie, Shu Tang, Wei Li","doi":"10.1186/s12911-025-03020-9","DOIUrl":"https://doi.org/10.1186/s12911-025-03020-9","url":null,"abstract":"&lt;p&gt;&lt;strong&gt;Background: &lt;/strong&gt;Bloodstream Infection (BSI) is a severe systemic infectious disease that can lead to sepsis and Multiple Organ Dysfunction Syndrome (MODS), resulting in high mortality rates and posing a major public health burden globally. Early identification of BSI is crucial for effective intervention, reducing mortality, and improving patient outcomes. However, existing diagnostic methods are flawed by low specificity, long detection times and high demands on testing platforms. The development of artificial intelligence provides a new approach for early disease identification. This study aims to explore the optimal combination of routine laboratory data and clinical monitoring indicators, and to utilize machine learning algorithms to construct an early, rapid, and universally applicable BSI risk prediction model, to assist in the early diagnosis of BSI in clinical practice.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Methods: &lt;/strong&gt;Clinical data of 2582 suspected BSI patients admitted to the Chongqing University Central Hospital, from January 1, 2021 to December 31, 2023 were collected for this study. The data were divided into a modeling dataset and an external validation dataset based on chronological order, while the modeling dataset was further divided into a training set and an internal validation set. The occurrence rate of BSI, distribution of pathogens, and microbial primary reporting time were analyzed within the training set. During the feature selection stage, univariate regression and ML algorithms were applied. First, Univariate logistic regression was used to screen for predictive factors of BSI. Then, the Boruta algorithm, Lasso regression, and Recursive Feature Elimination with Cross-validation (RFE-CV) were employed to determine the optimal combination of predictors for predicting BSI. Based on the optimal combination, six machine learning algorithms were used to construct an early BSI risk prediction model. The best model was selected by models' performance, and the Shapley Additive Explanations (SHAP) method was used to explain the model. The external validation set was used to evaluate the predictive performance and generalizability of the selected model, and the research findings were ultimately applied in clinical practice.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Results: &lt;/strong&gt;The incidence of BSI among inpatients at the Chongqing University Central Hospital was 12.91%. Following further feature selection, a set of 5 variables was determined, including white blood cell count, standard bicarbonate, base excess of extracellular fluid, interleukin-6, and body temperature. BSI early risk prediction models were constructed using six machine learning algorithms, with the XGBoost model demonstrating the best performance, achieving an AUC value of 0.782 in the internal validation set and an AUC value of 0.776 in the external validation set. This model is made publicly available as an online webpage tool for clinical use.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Conclusions: &lt;/strong&gt;This","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"186"},"PeriodicalIF":3.3,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12079808/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144076120","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":"Optimizing unsupervised feature engineering and classification pipelines for differentiated thyroid cancer recurrence prediction.","authors":"Emmanuel Onah, Uche Jude Eze, Abdullahi Salahudeen Abdulraheem, Ugochukwu Gabriel Ezigbo, Kosisochi Chinwendu Amorha, Fidele Ntie-Kang","doi":"10.1186/s12911-025-03018-3","DOIUrl":"10.1186/s12911-025-03018-3","url":null,"abstract":"<p><strong>Background: </strong>Differentiated thyroid cancer (DTC) is a common endocrine malignancy with rising incidence and frequent recurrence, despite a generally favorable prognosis. Accurate recurrence prediction is critical for guiding post-treatment strategies. This study aimed to enhance predictive performance by refining feature engineering and evaluating a diverse ensemble of machine learning models using the UCI DTC dataset.</p><p><strong>Methods: </strong>Unsupervised data engineering-specifically dimensionality reduction and clustering-was used to improve feature quality. Principal Component Analysis (PCA) and Truncated Singular Value Decomposition (t-SVD) were selected based on superior clustering metrics: adjusted Rand Index (ARI > 0.55) and V-measure (> 0.45). These were integrated into classification pipelines using Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), Feedforward Neural Network (FNN), and Gradient Boosting (GB). Model performance was evaluated through bootstrapping on an independent test set, stratified 10-fold cross-validation (CV), and subgroup analyses. Metrics included balanced accuracy, F1 score, AUC, sensitivity, specificity, and precision, each reported with 95% confidence intervals (CIs). SHAP analysis supported model interpretability.</p><p><strong>Results: </strong>The PCA-based LR pipeline achieved the best test set performance: balanced accuracy of 0.95 (95% CI: 0.90-0.99), AUC of 0.99 (95% CI: 0.97-1.00), and sensitivity of 0.94 (95% CI: 0.84-1.00). In stratified CV, it maintained strong results (balanced accuracy: 0.86; AUC: 0.97; sensitivity: 0.80), with consistent performance across clinically relevant subgroups. The t-SVD-based LR pipeline showed comparable performance on both test and CV sets. SVM and FNN pipelines also performed robustly (test AUCs > 0.99; CV AUCs > 0.96). RF and KNN had high specificity but slightly lower sensitivity (test: ~0.87; CV: 0.77-0.80). GB pipelines showed the lowest overall performance (test balanced accuracy: 0.86-0.88; CV: 0.85-0.88).</p><p><strong>Conclusions: </strong>Dimensionality reduction via PCA and t-SVD significantly improved model performance, particularly for LR, SVM, FNN, RF and KNN classifiers. The PCA-based LR pipeline showed the best generalizability, supporting its potential integration into clinical decision-support tools for personalized DTC management.</p><p><strong>Clinical trial number: </strong>Not applicable.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"182"},"PeriodicalIF":3.3,"publicationDate":"2025-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12070754/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143966186","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}