Evidence synthesis is the process of systematically gathering, analyzing, and integrating available research evidence. The quality of evidence synthesis depends on the quality of the original studies included. Validity assessment, also known as risk of bias assessment, is an essential method for assessing the quality of these original studies. Currently, there are numerous validity assessment tools available, but some of them lack a rigorous development process and evaluation. The application of inappropriate validity assessment tools to assessing the quality of the original studies during the evidence synthesis process may compromise the accuracy of study conclusions and mislead the clinical practice. To address this dilemma, the LATITUDES Network, a one-stop resource website for validity assessment tools, was established in September 2023, led by academics at the University of Bristol, U.K. This Network is dedicated to collecting, sorting and promoting validity assessment tools to improve the accuracy of original study validity assessments and increase the robustness and reliability of the results of evidence synthesis. This study introduces the background of the establishment of the LATITUDES Network, the included validity assessment tools, and the training resources for the use of validity assessment tools, in order to provide a reference for domestic scholars to learn more about the LATITUDES Network, to better use the appropriate validity assessment tools to conduct study quality assessments, and to provide references for the development of validity assessment tools.
In recent years, the rapid advancement of artificial intelligence has profoundly affected societal structures and a broad range of sectors, including healthcare, and is reshaping modes of production and everyday life. Across the evolution of AI technologies, high-quality datasets provide the essential foundation for model training and iterative improvement. Nevertheless, the construction of datasets for AI training remains challenged by ambiguous objective definition and insufficient core technical support, which together hinder the development of high-quality AI datasets. To address these challenges, the China Academy of Information and Communications Technology and Tsinghua University, together with other institutions, jointly released the Guidelines for the Construction of High-Quality Datasets for Artificial Intelligence (hereinafter, the Guidelines), proposing an engineering-oriented, full-lifecycle framework that encompasses data collection, governance, annotation, quality inspection, and operational management. Centered on the Guidelines, this article reviews and interprets their release context, key definitions, and recommended implementation pathways, with the aim of providing a structured reference for researchers seeking to systematically advance high-quality AI dataset development.
Objective To systematically investigate the implementation and reporting quality of statistical analysis methods in observational studies for the clinical evaluation of heart failure treatment and management devices, and to provide references for the standardized design and reporting of statistical analyses in future studies within this field. Methods A comprehensive search was conducted in the PubMed database for observational studies published between October 2014 and September 2024 that aimed to evaluate the effectiveness and/or safety of heart failure treatment devices with a control group. Two researchers independently screened the literature and extracted data. The basic characteristics of the included studies and the implementation and reporting features of their statistical analysis methods were analyzed. Results A total of 65 studies were included, comprising 63 (96.92%) cohort studies and 2 (3.08%) case-control studies. Among these, only 39 (60.00%) studies performed multivariable analyses. The median number of confounders included was 9 (IQR 5 to 16), and only 22 (56.41%) studies reported specific methods for identifying confounders. None of the studies considered procedure-related confounders such as operator experience or institutional procedure volume. The most frequently used multivariable method was Cox regression (20, 51.28%), followed by propensity score methods (13, 33.33%). Only 15 (23.08%) studies conducted subgroup analyses and 11 (16.92%) performed sensitivity analyses. Compared with studies published in non-Q1 journals according to the journal citation reports (JCR), studies published in Q1 journals had larger sample sizes and higher proportions of using multivariable analysis. Conclusion Observational studies on the clinical evaluation of heart failure treatment devices exhibit notable deficiencies in the implementation of statistical analysis methods, including inadequate identification and control of confounding factors and low proportions of subgroup and sensitivity analyses. Addressing these methodological limitations in future research will be essential for generating robust, high-quality evidence to inform clinical decision-making.