As artificial intelligence (AI) becomes increasingly integrated into mainstream healthcare applications, there has been a surge in the number of AI-centered diagnostic test accuracy studies. In September 2025, the international journal Nature Medicine published the Reporting Guideline for Diagnostic Accuracy Studies Using AI (STARD-AI). The guideline represents a comprehensive reporting framework specifically updated for AI-based studies within the Standards for Reporting Diagnostic Accuracy Studies (STARD). It consists of a main checklist (40 major items with 48 sub-items) and an abstract checklist (11 items). This article provides an interpretation of STARD-AI's development process, main content, scope of application, and specific item details, aiming to assist researchers, clinicians, and editors in thoroughly understanding and correctly applying STARD-AI. The goal is to enhance the quality and transparency of reporting in AI-centered diagnostic accuracy studies and to promote their standardized and ethical integration into the medical field.
With the aging population and the widespread implementation of lung cancer screening programs, an increasing number of elderly patients are undergoing curative lung resection. Due to diminished physiological reserve and complex comorbidities, this demographic exhibits a significantly higher incidence of severe perioperative complications (defined as Clavien-Dindo grade≥Ⅲ), which adversely affects both perioperative safety and long-term prognosis. In recent years, the research paradigm has shifted from univariate analysis toward multidimensional risk integration and the development of predictive models. This article systematically reviews the key risk factors for severe perioperative complications in elderly lung cancer patients, encompassing biological aging processes, frailty and sarcopenia, cardiopulmonary and renal functional reserves, inflammatory-immune and coagulation status, and perioperative interventions. Furthermore, it traces the evolution of risk assessment tools from traditional regression models to machine learning models that integrate multimodal data. The review also discusses common challenges in this field, including the standardization of outcome definitions, external validation, calibration assessment, and clinical translation. Future efforts should prioritize the deep integration of predictive tools with clinical decision support systems to establish a closed-loop care pathway from risk identification to stratified intervention, thereby effectively reducing complication rates and enhancing surgical outcomes for elderly lung cancer patients.
