Neural competition between prefrontal and auditory cortex constrains novel sound strategy learning.

Acute myocardial infarction (AMI), the most severe clinical manifestation of coronary artery disease, remains one of the leading causes of mortality worldwide and impose a substantial burden on healthcare systems [,]. Epidemiological data reveal a concerning upward trend in AMI mortality in China, with both urban and rural areas exceeding 60 deaths per 100,000 population in 2020 []. Similarly, in the United States, approximately 750,000 individuals experience either the first or recurrent AMI annually []. Notably, patients with diabetes mellitus (DM) represent a distinct subgroup within the myocardial infarction population, accounting for 20%-30% of all AMI cases [,]. Previous studies indicated that due to multiple underlying mechanisms, including more complex coronary artery lesions, endothelial dysfunction, prothrombotic state, and greater comorbidity burden, patients with diabetes experience significantly worse short- and long-term outcomes compared with those without diabetes []. Specifically, the in-hospital mortality rate following AMI is 1.5 to 2 times higher in patients with diabetes than non-diabetic individuals [,]. Although advances in revascularization techniques have substantially improved outcomes in AMI patients overall, those with DM have disproportionately high risks, with the reinfarction rate exceeding 40% and long-term mortality remaining high [,]. Consequently, the clinical management of AMI in diabetic patients remains particularly challenging, underscoring the critical need for accurate risk stratification and prognostic evaluation to identify high-risk inpatients and guide timely, tailored treatment strategies. Current guidelines predominantly recommend traditional risk prediction models, exemplified by the GRACE score, as standardized tools for assessing in-hospital mortality risk in patients with ACS []. However, the application of these models in contemporary clinical practice presents several notable limitations. First, most are based on conventional logistic regression methods, which may oversimplify the complex, often non-linear relationships between predictors and outcomes. Second, these models were developed in earlier periods. Nevertheless, patient characteristics and treatment strategies have changed dramatically in decades []. More importantly, as models designed for the broader ACS population, they often fail to adequately account for the unique clinical characteristics of patients with DM or incorporate metabolically relevant prognostic variables, thereby limiting their predictive performance in this high-risk subpopulation. Recent years have witnessed rapid growth in the application of machine learning (ML) techniques for cardiovascular risk prediction. These approaches have been demonstrated to possess superior performance to conventional models across multiple clinical scenarios, owing to their capacity for processing high-dimensional data and capturing complex nonlinear relationships [,]. Concurrently, emerging composite metabolic indicators, such as the triglyceride-glucose index (TyG) and the stress hyperglycemia ratio (SHR), have provided new perspectives for risk prediction beyond conventional metabolic markers. These indicators have been shown to be closely associated with outcomes in patients with AMI, with particularly strong predictive values observed in those with diabetes. For example, compared with the admission glucose levels, SHR has exhibited superior accuracy in predicting in-hospital mortality [–]. Building upon these foundations, the purpose of the present study is to develop specialized risk prediction tools for the diabetic AMI subpopulation by applying ML algorithms, leveraging representative multicenter real-world data from China, and incorporating diabetes-specific metabolic indicators. This modelling strategy may improve risk stratification while aligning with current guideline recommendations that emphasize high-risk subpopulations and individualized assessment. []. To accommodate different clinical application scenarios, this study plans to develop two distinct models: CAMI-DM 1.0, a simplified model intended for rapid initial screening and bedside evaluation, and CAMI-DM 2.0, a more sophisticated model designed to provide higher predictive accuracy for comprehensive risk assessment supported by electronic health systems. This study utilized data from the China Acute Myocardial Infarction (CAMI) registry, a prospective, nationwide, multicenter observational study conducted across tertiary hospitals in all provinces and municipalities of mainland China (excluding Hong Kong and Macau). The registry included 108 participating hospitals, comprising 31 provincial-level, 45 municipal-level, and 32 county-level institutions, reflecting the diversity and hierarchical nature of the healthcare system in China. The study design has been described in detail in previous publications []. This project was approved by the Institutional Review Board of the ce