ClarusC64/clinical-coagulation-instability-v0.1

该数据集用于评估模型是否能够从简短的临床代理轨迹中检测凝血系统的不稳定性。每个数据行代表在三个时间点观察到的简化止血状态，任务是判断凝血系统是否保持稳定或正在向凝血或出血不稳定性发展。数据集的核心思想是凝血不稳定性很少由单一实验室值引起，而是多个相互作用组件同时漂移的结果。数据集包括血小板轨迹、INR轨迹、纤维蛋白原轨迹、D-二聚体轨迹、肝脏合成代理、纤维蛋白溶解代理、输血反应和干预延迟等变量，以及一些干扰变量。预测目标是判断凝血不稳定性是否正在形成（label=1）或保持稳定（label=0）。

This dataset evaluates whether models can detect instability in the coagulation system from short clinical proxy trajectories. Each row represents a simplified hemostatic state observed across three time points. The task is to determine whether the coagulation system remains stable or is moving toward clotting or bleeding instability. The core idea is that coagulation instability rarely arises from a single laboratory value, but rather from the simultaneous drift of several interacting components. The dataset includes proxies for platelet trajectory, INR trajectory, fibrinogen trajectory, D-dimer trajectory, liver synthesis proxy, fibrinolysis proxy, transfusion response, and intervention delay, along with some decoy variables. The prediction target is to determine whether coagulation instability is emerging (label=1) or the state remains stable (label=0).