Objective: Utilizing lung cancer risk prediction models at the screening stage can enhance the accuracy of identifying high-risk individuals eligible for lung cancer screening. However, there is a relative lack of research on such prediction models in China, particularly regarding machine learning algorithms.
Methods: A stratified random sampling method was employed to randomly divide the dataset into a training set (70%) and a validation set (30%). Key variables were screened using LASSO regression. Then logistic regression and XGBoost algorithm were utilized to construct a lung cancer risk prediction model in the training set and validate it in the validation set, respectively.
Results: A lung cancer risk prediction model was constructed using 11,708 participants enrolled in a prospective cohort, the Guangzhou Lung-Care Project Program. In the constructed lung cancer risk prediction models, the AUC of the logistic regression model in the validation set was 0.647 (95% CI: 0.574–0.720); in contrast, the AUC of the XGBoost model based on the machine learning algorithm in the validation set was 0.658 (95% CI: 0.589–0.727), demonstrating slightly better discriminative ability compared to the logistic regression model. In addition, this study found the important effect of childhood exposure to cooking fuels on the risk of lung cancer, which has been rarely considered in previous research.
Conclusion: The lung cancer risk prediction model constructed based on the XGBoost algorithm is better than the logistic regression algorithm in terms of prediction accuracy and robustness, aiding in the risk assessment of individuals undergoing screening.
Objective: Academic adjustment is essential for medical students facing rigorous academic demands. While individual and instructional factors have been well-studied, the role of family functioning remains underexplored. This study examined the association between family functioning and academic adjustment, explored the mediating role of coping styles, and compared these pathways between at-risk students (those who had failed at least one final examination) and controls.
Methods: A cross-sectional study was conducted using validated scales for assessment of academic adjustment, family functioning, and coping styles. Structural equation modeling and bootstrap analyses were used to test mediation effects.
Results: 1022 Chinese medical students (293 at-risk, 729 controls) were included. Family functioning was significantly and positively associated with academic adjustment (p < 0.001). Positive coping mediated this relationship in both groups (at-risk: β = 0.13, 95% confidence interval [CI] [0.068, 0.218]; controls: β = 0.21, 95% CI [0.182, 0.243]), while negative coping mediated the effect only in at-risk students (β = 0.12, 95% CI [0.090, 0.167]). At-risk students showed significantly lower academic adjustment (t = –6.56, p < 0.001, Cohen's d = –0.45) and relied on distinct coping mechanisms compared to controls.
Conclusions: This study reveals distinct mediating pathways of coping styles between at-risk and other students, deepening our understanding of family influences and providing practical guidance for targeted interventions in medical education.
Objective: The overwhelming majority of prediction models have not been applied. An evidence-based review is needed to show that the new research is justified. This study aimed to develop an assessment tool for researchers and peer reviewers to conduct a rapid and comprehensive evaluation on the necessity and feasibility of planning clinical prediction model before its startup.
Methods: The framework for developing quality assessment tools was followed to develop the necessity and Feasibility Assessment Tool of CLInical Prediction models for individual prognosis (FATCLIP). Firstly, the scope, framework, and item pool of the FATCLIP was identified by a steering group comprising 15 experts through a web-based meeting. Then, an iterative Delphi process was conducted to refine the FATCLIP, in which the Delphi group enrolled 34 experts from multidiscipline, including epidemiologists, statisticians, clinicians, evidence-based medicine specialists, health care administrators and academic journal editors.
Results: Through twice steering group meetings and 2 rounds of the Delphi process, the framework of FATCLIP was determined based on expert consensus, including 6 domains and 31 signaling questions. The six domains were as follows: prediction outcome, review of existing models, candidate predictors, data, development and validation, and application and extension. At the same time, the usage manual of FATCLIP was also presented.
Conclusions: The FATCILP aims to assist researchers and peer reviewers to detect potential challenges during the development and application of the clinical prediction model for individual prognosis before its start-up, so that the research of clinical prediction models could be efficient and avoid research waste.
Objective: This study aimed to comprehensively evaluate the efficacy of music intervention as a non-pharmacological approach for improving physiological and psychological outcomes in patients with hypertension through a systematic review and meta-analysis of randomized controlled trials (RCTs).
Methods: We systematically searched PubMed, Embase, The Cochrane Library, Web of Science Core Collection, Wanfang Data, and CNKI for RCTs investigating the effects of music therapy on blood pressure, heart rate (HR), anxiety, and depression in hypertensive adults. Data were pooled using random-effects models, and weighted mean differences (WMDs) with 95% confidence intervals (CIs) were calculated. The robustness of findings was assessed via sensitivity analysis, and publication bias was evaluated using Egger's and Begg's tests.
Results: Twenty-one RCTs involving 1436 participants were included. Meta-analysis revealed that music intervention significantly reduced systolic blood pressure (SBP) (WMD = −8.26 mmHg, 95% CI: −10.56 to −5.96), diastolic blood pressure (DBP) (WMD = −5.91 mmHg, 95% CI: −8.03 to −3.79), HR (WMD = −4.17, 95% CI: −7.22 to −1.12), anxiety levels (measured by Self-Rating Anxiety Scale, SAS) (WMD = −5.22, 95% CI: −7.03 to −3.40), and depression levels (measured by Self-Rating Depression Scale, SDS) (WMD = −7.12, 95% CI: −10.27 to −3.98). Sensitivity analyses confirmed the stability of these findings, and statistical tests showed no significant publication bias for primary outcomes.
Conclusion: Music therapy is an effective complementary intervention for reducing blood pressure, HR, anxiety, and depression in hypertensive patients. Personalized music selections and longer intervention sessions may enhance efficacy. Future research should focus on standardizing intervention protocols, clarifying underlying mechanisms, and exploring long-term efficacy.
Traditional Chinese medicine (TCM) faces persistent gaps between evidence generation and clinical use. Building on Qian Xuesen's theory of open complex giant systems (OCGSs) and its qualitative-to-quantitative metasynthesis, we advance a complex-systems evidence framework tailored to TCM's hallmarks: a holistic perspective and pattern-based diagnosis and therapy. The framework integrates systems science, artificial intelligence, and allied disciplines to coordinate qualitative and quantitative approaches and to align macro-level effectiveness evaluation with micro-level mechanistic inquiry. It organizes multi-source evidence into a four-phase loop—production, differentiation, application, and validation: (1) standardizes evidence production; (2) conducts integrated evaluation along the disease–pattern–formula axis; (3) supports individualized effectiveness evaluation and decision-making; and (4) uses real-world feedback to verify and refine evidence. By linking clinical phenotypes, pathways, and outcomes into an evidence chain, the framework is intended to improve clinical decision quality and accelerate translational research. Beyond TCM, it offers a generalizable model for complex interventions acting on complex human systems, positioning TCM research for international scientific dialogue and modernization.
Aim: In traditional Chinese medicine, different dosage forms of orally administered Chinese herbal medicine (CHM) may introduce bias in estimating treatment and adverse effects. This meta-epidemiological study aimed to evaluate whether the use of different orally administered CHM dosage forms is associated with overestimation or underestimation of treatment and adverse effects in randomized controlled trials (RCTs).
Methods: Seven electronic databases were searched to identify potentially eligible meta-analyses (MAs) of RCTs evaluating CHM interventions. A two-step meta-epidemiological analysis was performed, using ratios of odds ratios for binary outcomes and differences in standardized mean differences for continuous outcomes. These metrics assessed whether different orally administered CHM dosage forms—including CHM decoctions, Chinese patent medicines (CPMs), and CHM granules influenced the magnitude of reported treatment effects or adverse effects.
Results: Eighty-two MAs comprising 1263 RCTs were analyzed. Overall, there was no consistent evidence that any oral dosage form systematically overestimated or underestimated treatment effects or adverse effects. Sensitivity analyses confirmed these findings, with the exception that CHM decoctions showed slightly larger binary treatment effects compared to CPMs after adjusting for incomplete outcome data. However, when adjusted for all confounders, CPMs yielded significantly greater continuous treatment effects than CHM decoctions. Additionally, CHM granules were associated with larger continuous treatment effects than CHM decoctions after adjusting for RCT funding. Subgroup analyses indicated that RCTs on digestive diseases tended to report larger effect estimates when using CHM decoctions, whereas RCTs on endocrine, nutritional, and metabolic diseases tended to report larger effect estimates when using CPMs.
Conclusions: This meta-epidemiological study suggests that while oral dosage forms of CHM are associated with minimal differences in reported treatment and adverse effect estimates, specific dosage forms may offer advantages in certain contexts. Subgroup analyses indicate that digestive disease trials tend to report larger estimates with CHM decoctions, and endocrine/metabolic disease trials with CPMs. When adjusting for confounders, CPMs yield greater continuous treatment effects compared with CHM decoctions, while CHM granules are associated with larger estimates than CHM decoctions after adjusting for RCT funding. Further research is needed to confirm their clinical relevance and guide formulation choices in CHM practice.
Background: Traditional Chinese medicine (TCM) is widely used in managing lumbar disc herniation (LDH), but heterogeneous outcome reporting in its trials hinders evidence synthesis. This study intends to develop a core outcome set (COS) for TCM-LDH to standardize reporting and improve research quality.
Methods: Candidate outcomes were identified via a systematic review of TCM-related randomized controlled trials (RCTs) for LDH, with studies retrieved from multiple databases between January 1 2019 and December 31 2023 and supplemented by clinical trial registry searches. Semistructured interviews with LDH patients and clinician questionnaires were conducted to refine candidate outcomes. Two Delphi rounds were carried out among clinicians, pharmaceutical researchers, journal editors, methodologists, and patients, followed by an online–offline consensus meeting to finalize the COS.
Results: A candidate outcome pool was established via a systematic review (413 RCTs, 51 registered studies), 30 LDH patient interviews, and 73 clinician surveys. After integration, deduplication, and steering committee refinement, two rounds of Delphi surveys were conducted. Following a consensus meeting attended by 24 multidisciplinary experts, 7 core outcomes were finalized for LDH: lumbar dysfunction, pain/discomfort, recurrence rate, straight leg raise angle, adverse reactions/adverse events, TCM syndromes, and sciatica frequency.
Conclusion: The developed COS for TCM-related LDH clinical trials provides standardized recommendations for outcome selection and reporting, which can enhance the consistency of research evidence, facilitate meta-analysis, and ultimately advance the quality of TCM-based interventions for LDH.
Background: Traditional Chinese medicine (TCM) with knowledge-intensive framework poses unique challenges to performance for large language models (LLMs). Although TCM-specific benchmarks and models have been developed, the performance of lightweight LLMs remains insufficiently investigated. This study presents a systematic evaluation and comparison of large-scale and lightweight LLMs to assess their capabilities and deployment trade-offs.
Methods: We developed TCM-related question-answering, a dataset comprising 801 questions derived from TCM textbooks. Eleven LLMs were evaluated under zero-shot and few-shot prompting conditions in both English and Chinese. Performance was primarily measured by accuracy.
Results: Large-scale LLMs achieved high accuracy on single-choice (69.01%–90.92%) and true/false questions (52.34%–59.38%) but performed poorly on multiple-choice questions, with a maximum accuracy of only 8.40%. Lightweight LLMs (2.10%–49.48%) generally lagged behind larger LLMs (6.30%–95.07%). However, Qwen3-1.7B (5.92%–54.20%) stood out and even surpassed the domain-specialized TCMChat-7B (2.10%–36.98%). Few-shot prompting enhanced performance in 8/11 (72.7%) of the models, Chinese prompts yielded better results than English in 9/11 (81.8%) of the models. Symptomatic diagnosis emerged as the most challenging reasoning category across all models (16.75%–48.07%).
Conclusion: This study demonstrates that although large-scale LLMs exhibit strong knowledge recall in TCM, their suboptimal performance on multiple-choice questions and substantial computational costs may limit their practical applicability in clinical settings. The robust performance of Qwen3-1.7B indicates that effective model optimization and domain-specific training may offer greater advantages than simply increasing model size. While the current evaluation is based on examination-style tasks and does not involve real-world clinical decision-making, our findings provide insights to support the deployment of optimized models in resource-constrained healthcare environments.
Accurate subtyping of lung cancer is crucial for developing personalized treatment plans and improving patient outcomes. This study established machine learning models for lung cancer subtyping by integrating multidimensional hematological indicators, offering advantages such as non-invasiveness, repeatability, and the capability for dynamic disease monitoring. The study utilized data from 771 lung cancer patients for model construction and validation, and an additional 510 independent cases were collected to assess the clinical applicability of the models. Ten supervised learning algorithms were screened, ultimately identifying the XGBoost model as optimal for differentiating small cell lung cancer from non-small cell lung cancer, and the Random Forest model as optimal for distinguishing lung squamous cell carcinoma from lung adenocarcinoma. In the independent clinical validation cohorts from two centers, these two models achieved accuracies of 95% and 91%, respectively, demonstrating good clinical applicability and serving as a valuable complement to pathological biopsy.
Aim: In the post-operative period of cardiac surgeries, the majority of patients are admitted to the intensive care unit (ICU) for ongoing management, where delirium frequently occurs as a complication. However, the association between various analgesic regimens and the onset of postoperative delirium in ICU patients following cardiac surgery remains poorly understood. The present study employs a target trial emulation (TTE) framework to examine the effects of early postoperative pain management on the incidence of delirium in patients immediately admitted to the ICU after cardiac surgery.
Patients and Methods: Study participants were selected from the MIMIC-IV version 3.1 database, comprising 5356 adult patients who were admitted for the first time and underwent cardiac surgery, followed by immediate transfer to the intensive care unit (ICU). The TTE framework was applied, utilizing the clone-censor-weighting (CCW, refer to the Statistical Analysis section) method for data analysis.
Results: In this study of 5356 adult patients who underwent cardiac surgery and were immediately transferred (within 24 h post-surgery) to the intensive care unit postoperatively, in comparison with the morphine group, patients receiving fentanyl exhibited a significantly elevated risk of delirium within seven days postoperatively (hazard ratio [HR] = 1.69; 95% confidence interval [CI]: 1.41–1.95) and demonstrated an earlier onset of delirium.
Conclusion: The findings of this study, conducted within a TTE framework, indicate that the utilization of fentanyl for pain management during the initial 24 h following cardiac surgery is associated with a higher incidence and earlier onset of postoperative delirium compared to morphine.
Background: Although basic research and observational clinical studies have shown an association between vascular calcification (VC) and heart failure (HF), the low level of evidence cannot directly indicate a causal relationship, and no Mendelian randomization (MR) study has been conducted to explore the relationship between VC and HF.
Methods: This study used bidirectional, multivariable, and mediation MR to comprehensively analyze the associations between different VC subtypes and HF as well as its subtypes from multiple perspectives. Accessible genome-wide association study (GWAS) data of VC were included, covering coronary artery calcification (CAC), abdominal aortic calcification (AAC), and calcific aortic valve stenosis (CAVS). Accessible public GWAS data of HF were also included, covering overall heart failure (HFall), ischemic heart failure (IHF), non-ischemic heart failure (ni-HF) recently published in Nature Genetics, and GWAS data of ni-HF with reduced ejection fraction (ni-HFrEF) and ni-HF with preserved ejection fraction (ni-HFpEF) which were classified based on ejection fraction.
Results: After sensitivity analysis and multiple correction, the following findings were obtained: (1) VC significantly increases the risk of HF. Specifically, CAC, AAC, and CAVS all significantly increase the risks of HFall and IHF; only CAVS is associated with an increased risk of ni-HF, while CAC and AAC have no impact on ni-HF; VC has no impact on ni-HFrEF or ni-HFpEF. (2) HF also promotes the progression of VC, indicating a bidirectional causal relationship between the two. Specifically, HFall and IHF significantly increase the risks of CAC and AAC; IHF increases the risk of CAVS; no reverse causal relationship is found between ni-HF (including its subtypes) and VC. (3) After two-step MR and multivariable MR correction, atrial fibrillation (AF) is found to partially mediate the causal effect of CAVS on HFall, with a mediation proportion of 21.65%.
Conclusion: This study reveals a bidirectional causal relationship between VC and HF through two-sample bidirectional MR, suggesting that early detection and management of VC are conducive to the prevention and treatment of HF, especially in high-risk populations such as those with ischemic heart disease. Controlling HF also helps delay the progression of VC and improve vascular status, which is the cornerstone of the prognosis of various cardiovascular diseases. Mediation analysis identifies AF as an important mediating factor, suggesting that screening and intervention of AF may be the key link to block the “VC → AF → HF” pathway, providing new genetic evidence for the prevention and treatment of HF.
Background: Evidence regarding the impact of age at hypertension onset on brain atrophy, independent of hypertension duration, remains limited. This study investigated whether the association between hypertension and brain volume differs by age at hypertension diagnosis.
Methods: In a multicenter, community-based cohort study (initiated in 2006), 948 participants were included: 117 early-onset hypertensives (diagnosed ≤40 years), 354 late-onset hypertensives (diagnosed >40 years), and 477 non-hypertensive controls, matched through propensity score matching. Neuroimaging data have been collected since 2020 to assess brain volume. The associations between early-onset and late-onset hypertension with brain volume were evaluated using the voxel-wise and generalized linear models.
Results: Compared to controls, early-onset hypertension was associated with lower volumes of cerebral parenchyma (β = −0.302; 95% confidence interval [CI], −0.541 to −0.063) and gray matter (β = −0.338; 95% CI, −0.590 to −0.087). The frontal, occipital, and temporal lobes showed the most prominent volume loss. Compared to participants with late-onset hypertension, early-onset hypertensives exhibited more pronounced brain volume reductions, especially in the frontal lobe. In contrast, compared with the matched non-hypertensive controls, the pattern of brain volume reduction in the late-onset hypertension group was similar to that in the early-onset hypertension group at the voxel level but did not reach statistical significance after full adjustment for covariates.
Conclusions: There is a significant correlation between early-onset hypertension and brain atrophy. It is crucial to manage blood pressure at a young age. For patients diagnosed with hypertension before the age of 40, it is recommended to strengthen brain health monitoring while undergoing antihypertensive treatment.