BMC Medical Research Methodology最新文献

Introduction and objectives: Over recent decades, the exponential growth of data, especially in healthcare, has necessitated advanced analytical methods. Conventional machine learning algorithms often assume independence among data points, limiting their effectiveness with longitudinal and hierarchical data. This study introduces a novel algorithm called GMEXGBoost, a methodological extension of generalized mixed-effects models that leverages the boosting framework of XGBoost for estimating fixed effects while simultaneously accounting for random effects. The innovation lies in GMEXGBoost's ability to explicitly incorporate data correlations while retaining the predictive power of boosted trees.

Methods: The GMEXGBoost model was evaluated through extensive simulations and a real-world cohort study, benchmarking against GLMM, GLMMTree, GMERF, and XGBoost. Also, its performance was assessed using predictive mean absolute deviation (PMAD), predictive misclassification rate (PMCR), sensitivity, specificity, accuracy, and AUC. Simulation analyses were conducted using multiple synthetic datasets, each comprising training and testing groups with varying effect structures, including random intercepts and slopes. All computations were performed in RStudio(version 2023.06.0).

Results: Our results indicate that while XGBoost achieved the lowest average errors across most scenarios, GMEXGBoost consistently demonstrated superior stability and accuracy when random-effect variance was large or correlations were strong. Also, in real data, GMEXGBoost outperformed other models in terms of the performance metrics.

Conclusion: The GMEXGBoost algorithm, by combining the estimates of the GLMM and XGBoost models, leverages the capabilities of both and delivers improved performance in complex problems. Although it is not universally superior, but demonstrates clear advantages in the analysis of hierarchical and longitudinal datasets with strong correlations. These properties make it a valuable tool for decision-making in healthcare and other domains that involve complex, structured data.

Background: Popularized in the 1980s, N-of-1 trials have emerged as a useful study design to assess the effects of interventions in single individuals. This study design consists of observing outcomes over time for the same individual under periods of exposure to an intervention and a comparator. Despite the simple idea, N-of-1 trials can require strong assumptions in the analysis phase to identify and estimate causal effects. As an illustrative example, we present an N-of-1 trial aiming at assessing the effect of alcohol abstinence on mood.

Methods: The N-of-1 trial participant decided to join a month-long nationwide alcohol abstinence campaign and was interested in the effects of alcohol abstinence on his mood. Every eight hours, the participant collected data about his own mood levels, number of alcohol units consumed, and social interactions, before, during, and after the alcohol abstinence period. Mood levels were measured using a 5-point Likert scale ranging from -2 to 2. To analyze the N-of-1 trial data, we relied on an explicit causal framework and made precise assumptions about the data generating process. We used a g-computation algorithm to estimate, for each time point, the individual-specific difference between the expected mood outcomes under the "always abstain from alcohol" intervention and "always drinking as usual" comparator.

Results: Overall, 264 time points were recorded, 171 under no intervention, and 93 during the intervention (alcohol abstinence) period. After adjusting for the other time-varying causes of mood, no statistically significant effect of alcohol units on mood level was found for measurements at the same time point; however, the number of alcohol units reported had a statistically significant negative effect on mood levels at the subsequent time point. The mean of the individual-specific average treatment effects across the entire study period was 0.05 (95%CI: -0.06, 0.15).

Conclusions: N-of-1 trials can be truly individual-centric studies, tailored to the needs and preferences of the participants. Analyzing data from N-of-1 trials can be complex, and the use of a causal framework can help inform the analyses.

Background: Cluster randomized trials (CRTs) require balanced baseline covariates to yield unbiased estimates of treatment effects. Existing approaches such as constrained randomization can improve balance but may compromise allocation randomness. We introduce Cluster Minimal Sufficient Balance (CMSB), a cluster randomization method designed to enhance covariate balance while preserving allocation randomness and computational efficiency.

Methods: CMSB integrates dynamic imbalance monitoring with conditional biased randomization into a single procedure. The method accommodates both continuous and categorical covariates and was evaluated through simulation studies comparing its performance with constrained randomization, simple randomization, block randomization, stratified randomization, and minimization across varying numbers of clusters and covariate dimensions. An empirical application further assessed its practical utility.

Results: In high-dimensional settings with 10 covariates, CMSB achieved a 45% greater improvement in covariate balance compared to constrained randomization, while maintaining near-optimal allocation randomness (correct guess probability: 49.79% versus 61.03% for minimization). CMSB reduced mean allocation time from over 100 s under constrained randomization to 0.116 s per allocation, even when simulating up to 2,000 randomization schemes. In the empirical application, CMSB demonstrated a 76% improvement in covariate balance relative to simple randomization.

Conclusions: CMSB effectively balances the competing demands of covariate balance, allocation randomness, and computational efficiency in cluster randomized trials. Its ability to handle high-dimensional covariates makes it particularly suitable for large-scale trials.

Background: As the number of systematic reviews (SRs) and meta-analyses (MAs) evaluating drug-eluting stents (DES) for coronary artery disease (CAD) continues to grow, the need for standardized primary outcomes has become increasingly important. This study aimed to examine the specification and selection of primary outcomes in published SRs/MAs on DES for CAD and to identify factors associated with their reporting.

Methods: We conducted a cross-sectional analysis of SRs/MAs on DES for CAD. They were retrieved from English-language databases (PubMed, Embase, Cochrane Library) and Chinese-language databases (CNKI, Wanfang, VIP) from the beginning of their establishment to May 3, 2025. We assessed whether primary outcomes were explicitly specified and employed multivariable logistic regression to identify factors associated with such specification. For frequently reported outcomes, we compared how often they were designated as primary outcomes. In SRs/MAs that clearly stated primary outcomes, we categorized them using the Core Outcome Measures in Effectiveness Trials (COMET) framework and calculated the Corrected Covered Area (CCA) to assess outcome overlap.

Results: A total of 271 SRs/MAs from 29 countries or regions were included. Only 188 (188/271, 69.37%) specified primary outcomes. Specification was significantly associated with prospective protocol registration (adjusted odds ratio = 2.69, 95% confidence interval: 1.11-6.54, P = 0.03). We identified 18 distinct primary outcomes among these reviews, with a low overlap (Corrected Covered Area = 0.0544). The most frequently reported outcomes were specified as primary outcomes in just 19.56% (53/271) of studies, while the most designated primary outcomes were not reported in 43.54% (118/271) of reviews. According to the COMET framework, coverage of life impact, resource use, and other non-physiological domains was limited.

Conclusion: The inconsistent specification and limited standardization of primary outcomes in SRs/MAs of DES for CAD may undermine evidence quality. The pathway from "unspecified primary outcomes" to "lack of standardization" to an "incomplete core outcome set" suggests a need for urgent methodological improvement in outcome reporting.

Background: Research findings must be representative by creating a sample of individuals, ensuring the results can be generalized and applicable to a larger population, which has historically been guided by a power analysis. However, the varied research design methods require a unique approach to sampling and a formula for recruitment and size. Therefore, the purpose of this study was to analyze historical data from published manuscripts in the Journal of Athletic Training (JAT) relative to study design and sample sizes. A secondary purpose was to further explore metrics for survey-based research.

Methods: This descriptive analysis explored 1267 publications in each issue of the JAT from January 2012 (Volume 47) to December 2022 (Volume 57). We extracted publications from the JAT website. Every article was entered into a spreadsheet (year of publication, publication title) and data specific to the study design and sample size were used for analysis. For studies that were coded as survey-based research, access, response, and completion rates were completed, and topic area and use of a power analysis were extracted. Data were analyzed using measures of central tendency (mean, median, range).

Results: Of the 1267 published studies, the most frequent design was cross-sectional (394, 31.1%). In total, 1080 publications (85.2%) were not survey-based, with a median sample size of 34 participants, while 187 publications (14.8%) were survey-based, with a median sample size of 429. Among those surveys, most were cross-sectional (n = 151/187, 80.8%), with 80.7% (n = 151/187) reporting the number initially recruited and 50.8% (n = 95/187) reporting the number of surveys started. The survey publications reported recruiting an average of 4453 potential participants (median = 2500; min = 101, max = 48752), with 985 participants starting the study (median = 816, min = 57, max = 7067), and a final sample size of 819 (median = 429; min = 17, max = 13002). The grand mean access rate was 22.1%, the grand mean response rate was 18.4%, and the grand mean completion rate was 83.1%.

Conclusion: Researchers and reviewers can use these trends to guide authorship and review processes for athletic training research. However, sampling strategies should be consistent with the research question, which may lead to deviations from these reported trends.

Background: Combining multiple biomarkers into a single diagnostic score can improve disease classification. However, traditional methods such as logistic regression and linear discriminant analysis depend on restrictive distributional assumptions, which can limit their effectiveness when dealing with complex or heterogeneous datasets. To address these limitations, more practical methodological alternatives are required.

Methods: This study investigates the utility of two sufficient dimension reduction (SDR) methods, Minimum Average Variance Estimation (MAVE) and the Outer Product of Gradients (OPG), for constructing composite biomarker scores in binary diagnostic classification. A comprehensive simulation study was conducted under four data-generating scenarios, varying in sample sizes, mean shifts, variance heterogeneity, normal and non-normal distributional forms. SDR based scores were constructed using likelihood ratio statistics under both the central subspace and central mean subspace frameworks. Classification performance was evaluated using the area under the receiver operating characteristic curve (AUC). Traditional methods, logistic regression and linear discriminant analysis, were included as benchmarks. To demonstrate practical utility, the methods were applied to the Breast Cancer Wisconsin (Diagnostic) dataset.

Results: In simulations, SDR methods outperformed traditional approaches consistently in settings with variance heterogeneity and mixture structures, yielding higher AUC values. The performance of SDR methods was less robust where the data had strong skewness, though it remained comparable to that of traditional methods. In the real dataset, SDR methods achieved similar discriminative accuracy to traditional methods while offering more compact and interpretable summaries of biomarker contributions.

Conclusions: SDR methods combined with likelihood-based scoring show potential for a relatively robust and interpretable framework in the context of this paper. They are particularly advantageous in settings with complex diagnostic structures, while maintaining competitiveness in well-structured data. These findings support the use of SDR methods as practical tools for combining biomarkers in precision medicine.