BMC Medical Research Methodology最新文献

{"title":"A comparative simulation study of cluster ensemble algorithms integrated with multiple imputation for clustering with missing data.","authors":"Yui Tomo, Funato Sato, Mari Oba","doi":"10.1186/s12874-026-02869-2","DOIUrl":"10.1186/s12874-026-02869-2","url":null,"abstract":"<p><p>Since cluster analysis methods usually cannot be applied directly to data with missing values, various approaches have been investigated to handle the problem. Multiple imputation is one of the standard procedures for addressing the problem of missing data. In cluster analysis, instead of Rubin's rule, cluster ensemble methods have been proposed to be combined with multiple imputation. However, it remains unrevealed which of the cluster ensemble algorithms leads to better performance when integrated with the procedure. Therefore, we conducted numerical comparisons of several algorithms to integrate the results from k-means++ clustering for multiply imputed datasets and also applied the combined approaches to two real datasets. Our results suggest that the non-negative matrix factorization algorithm may be suitable for scenarios with class balance, whereas the greedy and agglomerative cluster algorithms may be suitable for scenarios with class imbalance. Before application to actual datasets, we still recommend performing simulation experiments in scenarios reflecting the characteristics of the datasets and the assumption of missing value mechanisms.</p>","PeriodicalId":9114,"journal":{"name":"BMC Medical Research Methodology","volume":" ","pages":""},"PeriodicalIF":3.7,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13439842/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148142657","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 staged Bayesian framework for longitudinal prediction of adolescent depression using add health data.","authors":"Niloofar Ramezani, Ehiremen Adesua Azugbene, Jeffrey R Wilson","doi":"10.1186/s12874-026-02890-5","DOIUrl":"10.1186/s12874-026-02890-5","url":null,"abstract":"<p><strong>Background: </strong>Accurately predicting adolescent depression from longitudinal behavioral data requires methods that incorporate feedback between prior mental health and substance use while remaining reproducible and interpretable. The objective of this study is predictive rather than explanatory or causal.</p><p><strong>Methods: </strong>Using Waves I-IV of the U.S. National Longitudinal Study of Adolescent to Adult Health (Add Health), we developed a novel staged Bayesian framework for longitudinal prediction: (Stage 1) multivariate logistic models with correlated random intercepts for depression, smoking, and alcohol use; (Stage 2) a composite substance-use indicator jointly modeled with depression; and (Stage 3) a temporal feedback model including lagged depression and lagged substance use. Posterior predictive checks and internal validation were used to assess model adequacy. Discrimination (area under the receiver operating characteristic curve [AUC]) and calibration (Brier score) were summarized across stages. The models are predictive rather than explanatory and are not intended to estimate causal effects.</p><p><strong>Results: </strong>Lagged depression (posterior mean = 0.98; 95% highest posterior density [HPD] interval: 0.45-1.52) and lagged substance use (posterior mean = 0.69; 95% HPD: 0.12-1.25) were associated with increased odds of future depression. Predictive performance improved from Stage 1 to Stage 3: AUC increased from 0.78 to 0.84, and Brier score decreased from 0.185 to 0.152. Posterior predictive checks supported model adequacy.</p><p><strong>Conclusions: </strong>The staged Bayesian pipeline provides a generalizable template for longitudinal prediction with temporal feedback, improving discrimination and calibration while preserving interpretability.</p>","PeriodicalId":9114,"journal":{"name":"BMC Medical Research Methodology","volume":" ","pages":""},"PeriodicalIF":3.7,"publicationDate":"2026-05-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13371721/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148052566","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":"Evaluating Elicit's systematic reviews workflow in an umbrella review on air pollution and acute lower respiratory infections: a methodological study for quality appraisal.","authors":"Cristina Mazzali, Tiziana Pinciroli, Maria Rosa Valetto, Pietro Dri, Antonio Giampiero Russo","doi":"10.1186/s12874-026-02892-3","DOIUrl":"10.1186/s12874-026-02892-3","url":null,"abstract":"<p><strong>Background: </strong>The exponential growth of scientific publications has increased the complexity of evidence synthesis. Systematic reviews remain essential but highly resource-intensive. Large language models (LLMs) offer new opportunities to support or partially automate key steps of this process. This study evaluates the performance of Elicit's Systematic Reviews workflow in comparison to the traditional methodology, using as reference a published umbrella review on the association between air pollution and acute lower respiratory infections (ALRI).</p><p><strong>Methods: </strong>A parallel workflow was developed to reproduce each phase of the traditional review. Considering article retrieval, for the traditional workflow, articles were retrieved through a Boolean search, while for the AI-assisted workflow a natural-language query submitted to Elicit. Screening was conducted in two steps, emulating the traditional PECOS-based criteria. Full-text evaluation and quality appraisal were performed through Elicit's \"data extraction\" functionalities. For quality appraisal the validated AMSTAR-2 EH questionnaire was applied.</p><p><strong>Results: </strong>The traditional Boolean search identified 324 unique articles. When compared with the 500 records retrieved by Elicit, an overlap of 8% was observed, which prevented a direct, recall-oriented comparison of search performance. To enable a controlled comparison of downstream steps, we applied Elicit's screening and data-extraction functions to the 324 records identified through the Boolean search, using empirically defined screening-score threshold in Elicit to select studies for further evaluation. In the screening on title and abstract, 33 articles were identified through the traditional workflow and 70 through Elicit, 30 articles overlapping (recall 90.9%, precision 42.9%). The full-text screening selected 15 articles with the traditional methodology and 24 with Elicit, all 15 articles from the traditional methodology being included in Elicit selection (recall 100%, precision 62.5%). In the quality assessment, Elicit showed 24.4% disagreement on general items and 30.4% on additional items of the AMSTAR-2 EH. Errors clustered around multi-component questions, items requiring expert interpretation, and information located in supplementary materials.</p><p><strong>Conclusions: </strong>Elicit can support several phases of systematic reviews and reduce manual workload, but it cannot independently reproduce the methodological rigor required for high-quality evidence synthesis. At present, LLM-based tools are best positioned as complementary systems within human-supervised workflows.</p>","PeriodicalId":9114,"journal":{"name":"BMC Medical Research Methodology","volume":" ","pages":""},"PeriodicalIF":3.7,"publicationDate":"2026-05-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13412304/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148101076","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":"Clustering methods for categorical time series and sequences : a scoping review.","authors":"Ottavio Khalifa, Alan Balendran, Viet-Thi Tran, François Petit","doi":"10.1186/s12874-026-02857-6","DOIUrl":"10.1186/s12874-026-02857-6","url":null,"abstract":"<p><strong>Objective: </strong>To provide an overview of clustering methods for categorical time series (CTS), a data structure common in epidemiology, sociology, biology, and marketing, and to support method selection according to data characteristics.</p><p><strong>Materials and methods: </strong>We searched PubMed (via MEDLINE), Web of Science, and Google Scholar up to November 2024 for articles proposing and evaluating CTS clustering techniques. Methods were classified into three families-distance-based, feature-based, and model-based-and assessed for their ability to address challenges such as variable sequence length, multivariate data, continuous time, missing data, covariates, and large data volumes.</p><p><strong>Results: </strong>Of 14,607 records retrieved, 124 articles describing 129 methods were included. Distance-based approaches, especially those using Optimal Matching, were most common, with 56 methods. We found 28 model-based methods, which covered a broader range of complex data structures such as multivariate data, continuous time and time-invariant covariates. We recorded 45 feature-based approaches, which were on average more scalable but less flexible. Fewer than half of the methods provided public implementations. A searchable Web application ( https://cts-clustering-scoping-review-7sxqj3sameqvmwkvnzfynz.streamlit.app/ ) was developed to support method selection.</p><p><strong>Discussion: </strong>CTS clustering methods are highly heterogeneous in assumptions, capabilities, and scalability. Distance-based approaches dominate, but model-based methods offer richer modeling potential, while feature-based ones emphasize performance at the cost of flexibility.</p><p><strong>Conclusion: </strong>This review highlights methodological diversity and gaps in CTS clustering. The proposed typology and Web application aim to help researchers choose appropriate methods to choose appropriate methods for their data.</p>","PeriodicalId":9114,"journal":{"name":"BMC Medical Research Methodology","volume":"26 1","pages":""},"PeriodicalIF":3.7,"publicationDate":"2026-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13217966/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148052551","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":"Translating treatment effects between correlated endpoints.","authors":"Nusrat Rabbee","doi":"10.1186/s12874-026-02852-x","DOIUrl":"10.1186/s12874-026-02852-x","url":null,"abstract":"<p><p>Clinical researchers frequently need to answer: 'If a treatment changes endpoint 1 by X%, what change in endpoint 2 should we expect?' Current meta-analytic approaches either produce biased estimates (Daniels-Hughes) or provide correlation measures without explicit translational coefficients. We demonstrate how multivariate random-effects meta-analysis can derive the Expected Translational Association (η), which translates treatment effects across correlated endpoints, simultaneously fitting all correlated endpoints through their shared variance-covariance structure, yielding unbiased estimates and nominal confidence coverage. For convenience, we refer to this stacked implementation as SLIM (Stacked LInear Mixed Effects Model). Extensive simulations demonstrated that SLIM maintains near-zero bias and nominal coverage across scenarios, while the Daniels-Hughes approach exhibits substantial bias due to measurement-error-in-covariates, with bias persisting as sample size increases. Using the abdominal aortic aneurysm (AAA) dataset, we demonstrate how this approach quantifies translational associations between growth-rate and rupture-risk endpoints and enables quantitative biomarker-outcome \"what if\" analyses, while showing that Bayesian meta-regression approaches can underestimate risk at clinically critical ranges. The framework generalizes naturally to multivariate systems, offering a rigorous and transparent foundation for evidence synthesis in translational science.</p>","PeriodicalId":9114,"journal":{"name":"BMC Medical Research Methodology","volume":"26 1","pages":""},"PeriodicalIF":3.7,"publicationDate":"2026-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13220494/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148052585","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":"Quality and performance of machine learning versus logistic regression for predicting IVIG resistance in Kawasaki disease: a PROBAST+AI systematic comparison.","authors":"Jiaying Zhang, Difan Wang, Jinfeng Dong, Ying He, Ying Liu, Tingjiao You, Jing Li, Lizhi Li, Xiaodan Wu, Qiuyu Tang, Shurong Ma, Panpan Liu, Haitao Lv, Hongbiao Huang","doi":"10.1186/s12874-026-02893-2","DOIUrl":"10.1186/s12874-026-02893-2","url":null,"abstract":"<p><strong>Background: </strong>This study aimed to systematically compare the predictive performance and methodological quality of logistic regression (LR) and machine learning (ML) models for intravenous immunoglobulin (IVIG) resistance in Kawasaki disease (KD) using the PROBAST + AI framework.</p><p><strong>Methods: </strong>We searched PubMed, Embase, and Web of Science to identify studies on prediction models for IVIG resistance in KD published between January 1, 2006, and July 31, 2025. We assessed methodological rigour, risk of bias, and applicability using PROBAST + AI. A meta-analysis was performed using random-effects models with logit-transformed area under the receiver operating characteristic curve (AUC) values. Subgroup, sensitivity, and publication bias analyses were additionally conducted.</p><p><strong>Results: </strong>We identified 52 eligible studies (40 LR and 12 ML). In external validation, pooled AUCs were similar between ML and LR models (0.76 [95% CI 0.64-0.86] vs. 0.75 [95% CI 0.68-0.81]). In internal validation, ML showed a slightly higher pooled AUC than LR (0.86 [95% CI 0.78-0.92] vs. 0.76 [95% CI 0.72-0.79]), although no statistically significant differences were observed. All studies were judged to be at high risk of bias, mainly due to retrospective single-centre designs, inadequate handling of missing data and continuous predictors, and poor reporting of calibration and clinical utility. No study reported sample size calculations.</p><p><strong>Conclusions: </strong>Given the limited external validation and substantial heterogeneity across studies, ML does not consistently outperform LR in predicting IVIG resistance in KD. Future studies should prioritise rigorous external validation and adherence to TRIPOD + AI and PROBAST + AI.</p>","PeriodicalId":9114,"journal":{"name":"BMC Medical Research Methodology","volume":" ","pages":""},"PeriodicalIF":3.7,"publicationDate":"2026-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13393907/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148041023","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":"Information borrowing in phase II randomized dose-ranging clinical trials in oncology.","authors":"Guillaume Mulier, Vincent Lévy, Lucie Biard","doi":"10.1186/s12874-026-02883-4","DOIUrl":"10.1186/s12874-026-02883-4","url":null,"abstract":"&lt;p&gt;&lt;strong&gt;Introduction: &lt;/strong&gt;Over the past decades, the advent of new therapeutics such as immunotherapies and targeted therapies has challenged conventional clinical trial designs, such as single-arm studies. Selecting a single dose in phase I trials with short follow-up, typically based solely on toxicity endpoints, has been shown to lead to suboptimal dosing decisions. Consequently, dose optimization is now increasingly encouraged in oncology. This study was motivated by the case of ibrutinib in chronic lymphocytic leukemia, for which the initially approved dose of 420 mg/day, determined using conventional phase I designs based on the maximum tolerated dose, was later shown to achieve comparable response rates at lower doses. This example highlights the potential value of dose-ranging phase II studies in oncology. Assuming that borrowing information across doses can improve statistical power, our objective was to compare several strategies for information borrowing in phase II randomized trials involving multiple doses of the same drug.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Methods: &lt;/strong&gt;The backbone design considered was the Bayesian Optimal Phase II (BOP2) design, adapted to a multi-arm setting and allowing for co-primary binary endpoints as well as interim analyses. This design relies on a multinomial conjugate model to describe the endpoints within a Bayesian framework, with decision rules for early stopping due to futility and/or toxicity based on posterior probabilities. We adapted and compared several information-borrowing approaches to estimate efficacy and toxicity: (i) power prior, (ii) Bayesian hierarchical modeling, (iii) Bayesian calibrated hierarchical modeling, and (iv) Bayesian logistic regression. These approaches were applied alongside BOP2 decision rules. In addition, a Simon's two-stage design with added toxicity monitoring was used as a comparator. A simulation study was conducted to evaluate the operating characteristics of the designs, using a hypothetical randomized dose-ranging trial with efficacy and toxicity as co-primary endpoints against reference values.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Results: &lt;/strong&gt;Our results indicate that the power prior, without dynamic adaptation of the borrowing strength, is unsuitable in this context as it substantially increases the false positive rate. Bayesian hierarchical modeling shrinks estimates toward a common mean, reducing variance but also leading to inflated false positive rates. In contrast, Bayesian logistic regression provides a more balanced trade-off, achieving moderate gains in power while increasing the false positive rates.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Conclusion: &lt;/strong&gt;In multi-arm trials, the BOP2 design without information borrowing offers stricter control of the false positive rate than borrowing-based approaches when only toxic or futile doses are considered. Nevertheless, Bayesian logistic regression modeling of the dose-toxicity and dose-efficacy relationships, combined with BOP2 decision rules, could be","PeriodicalId":9114,"journal":{"name":"BMC Medical Research Methodology","volume":" ","pages":""},"PeriodicalIF":3.7,"publicationDate":"2026-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13397689/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148041053","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":"Simulating crisis triage: a methodological framework for evaluating ventilator allocation under crisis standards of care.","authors":"B Corbett Walsh, Jianan Zhu, Yang Feng, Rebecca A Betensky, Deepak Pradhan","doi":"10.1186/s12874-026-02878-1","DOIUrl":"10.1186/s12874-026-02878-1","url":null,"abstract":"<p><strong>Background: </strong>Crisis Standards of Care (CSC) may require rationing of life-sustaining resources, such as mechanical ventilation, during public health emergencies. Simulation modeling offers a scalable, transparent method to evaluate triage frameworks before implementation. While ventilator triage frameworks vary in their use of exclusion criteria, comorbidity adjustments, and reassessment frequency, few have been rigorously compared using real-world data under realistic surge conditions. Robust platforms that simulate both front-end (initiation) and back-end (reassessment or reallocation) triage are critical for evaluating clinical, operational, and ethical performance.</p><p><strong>Methods: </strong>We developed a computational simulation platform using retrospective real-world data from intubated adults during the Spring 2020 COVID-19 surge across a large New York City health system. The simulated surge cohort included all patients mechanically ventilated between March 1 and June 30, 2020. A crisis cohort was defined as those patients receiving ventilation once 95% of the health system's pre-pandemic ventilator supply was in use. Eight CSC strategies were evaluated, including policies from New York, Pennsylvania, Maryland, Canada, two academic frameworks, a lottery-based system, and first-come first-served. Strategies varied in their use of exclusion criteria, comorbidity modifiers, reassessment intervals, and prioritization for special populations. Daily ICU census and ventilator availability were used to simulate resource strain and drive triage decision-making. Patients simulated for ventilator rationing were simulated to expire. Manual abstraction of comorbidities and structured rules for imputing missing SOFA subscores were applied uniformly.</p><p><strong>Results: </strong>The platform simulated 10,000 iterations per strategy and included 2,365 intubated patients. Final analyses will be published separately.</p><p><strong>Conclusion: </strong>This platform provides a scalable, reproducible framework for evaluating ventilator triage strategies under pandemic-like conditions. By integrating both initial triage and serial reassessment (front- and back-end) logic, operational constraints, and clinical trajectories, it enables detailed comparisons of survival, resource utilization, and prognostic accuracy. The simulation also supports ethical evaluation by testing the practical impact of exclusion and comorbidity-based criteria. Such models can assist governments, health systems, and public health agencies in designing triage protocols that are evidence-informed, ethically defensible, and operationally feasible. This work demonstrates how computational modeling can strengthen health system preparedness and support public trust.</p>","PeriodicalId":9114,"journal":{"name":"BMC Medical Research Methodology","volume":" ","pages":""},"PeriodicalIF":3.7,"publicationDate":"2026-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13267345/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148026676","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 comparative analysis of parametric, semiparametric and machine learning models for survival prediction.","authors":"Salomi Arul, Sujatha V","doi":"10.1186/s12874-026-02863-8","DOIUrl":"10.1186/s12874-026-02863-8","url":null,"abstract":"<p><strong>Background: </strong>Survival analysis is crucial in medical research. This study provides a comprehensive comparison of parametric, semi-parametric, and machine learning models for predicting disease-free survival (DFS) in colorectal cancer.</p><p><strong>Methods: </strong>Using data from 2,088 patients in a Phase III trial (Project Data Sphere), we evaluated parametric models (Lognormal, Generalized Gamma), semi-parametric models (CoxPH and stratified Cox model) and machine learning models (Random Survival Forests, XGBoost). Performance was assessed using Area under the curve (AUC) and Brier score.</p><p><strong>Results: </strong>After 10-fold cross-validation, the Random Survival Forest (RSF) model demonstrated superior discriminative performance (1-year AUC: 0.713, 5-year AUC: 0.799) and the best overall prediction accuracy (Brier score at 3 years: 0.148). Weighted Brier scores confirmed its robustness across clinical scenarios (equal weights: 0.144; FN weight=3: 0.179; FP weight=2: 0.205). Parametric and semi-parametric models showed stable intermediate performance (CV AUC: 0.68-0.70), while XGBoost exhibited severe overfitting with near-random CV AUC (0.49-0.57).</p><p><strong>Conclusions: </strong>Ensemble methods, particularly Random Survival Forest, provide superior predictive accuracy for Disease-free survival compared to traditional models. These advanced techniques offer significant potential for improving risk stratification and personalizing cancer care, though their integration requires addressing interpretability challenges.</p>","PeriodicalId":9114,"journal":{"name":"BMC Medical Research Methodology","volume":" ","pages":""},"PeriodicalIF":3.7,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13377819/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148013423","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":"Clustering longitudinal data: comparison of model-based and distance-based approaches using simulated and real-world data in psychiatric research.","authors":"Laura Julià, Ana M Sánchez-Torres, Jessica Merchan, Silvia Amoretti, Josep Lluis Carrasco, Sergi Mas","doi":"10.1186/s12874-026-02882-5","DOIUrl":"10.1186/s12874-026-02882-5","url":null,"abstract":"<p><strong>Background: </strong>Clustering longitudinal symptom trajectories is increasingly used to characterise clinical heterogeneity in psychiatric disorders. However, the relative performance of distance-based and model-based approaches remains insufficiently studied in psychiatric settings, where limited follow-up waves, measurement variability, and heterogeneous patient responses may affect clustering reliability. We aimed to systematically compare distance-based and model-based (Latent Class Mixed Models, LCMMs) methods, evaluating their performance in recovering the number of groups, classification accuracy, and clinical interpretability.</p><p><strong>Methods: </strong>We analysed longitudinal symptom data from 237 first-episode psychosis patients (PEPs cohort). We also performed simulations (500 replicates per scenario) with known group structures (3 or 4), varying group-size distributions (balanced/unbalanced) and levels of group separation (well-separated, partially overlapping, strongly overlapping). Recovery of the true or optimal number of groups was assessed using multiple validity indices. Classification accuracy was measured by weighted Cohen's [Formula: see text]. Clinical interpretability was evaluated in terms of within-group homogeneity, stability across replications, and minimum group size.</p><p><strong>Results: </strong>In the PEPs cohort, distance-based methods consistently identified two stable and clinically interpretable trajectory groups, whereas LCMMs frequently suggested more complex structures, often including very small classes. Concordance between methodological families was low, indicating that patient stratification was strongly method-dependent: distance-based methods grouped patients primarily by overall symptom severity, whereas LCMMs distinguished groups according to trajectory slope. In simulations, both approaches performed well when groups were clearly separated, but performance declined with increasing overlap and group-size imbalance. Distance-based methods generally produced more stable and homogeneous partitions, whereas LCMMs showed greater variability across replications.</p><p><strong>Conclusions: </strong>In psychiatric longitudinal studies with few assessment waves and moderate sample sizes, distance-based clustering methods may provide a robust strategy for trajectory-based partitioning. Model-based approaches such as LCMMs may offer advantages when complex nonlinear trajectories are expected, but their stability may be limited under noisy or weakly separated conditions. Combining multiple evaluation criteria and considering clinical interpretability are essential for reliable subgroup identification.</p>","PeriodicalId":9114,"journal":{"name":"BMC Medical Research Methodology","volume":" ","pages":""},"PeriodicalIF":3.7,"publicationDate":"2026-05-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13374183/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148004497","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}