ARDS预后预测的评估框架:数据集、模型和特征

An evaluation framework for ARDS prognostic prediction:datasets,models,and features

急性呼吸窘迫综合征(acute respiratory distress syndrome,ARDS)预后旨在根据患者的身体状况预测后期出现某种风险的概率,有效的预后方案能极大降低死亡率,优化资源分配。近年来,研究人员通过增强模型计算能力的方式提升了预后的时效性和准确率。然而,ARDS预后研究中数据格式不统一和评估基线不一致等问题突出,限制了ARDS预后研究的深入发展。为解决上述问题,本文提出了ARDS预后预测的评估框架。一方面,提出了针对多源数据的治理方案,解决了ARDS数据格式不统一的问题;另一方面,该框架形成了由13种机器学习模型和20种特征集合组成的评估体系,解决了评估基线不一致的问题。实验结果表明,该框架有效解决了上述问题并评估了数据集、模型和特征的影响程度。

Acute respiratory distress syndrome(ARDS)prognosis aims to predict the probability of developing a certain risk at a later stage based on current physical condition of a patient.An effective prognostic strategy can significantly reduce mortality and optimize resource allocation.In recent years,researchers have improved the timeliness and accuracy of prognosis by enhancing the computational power of models.However,the problems of non-uniform data formats and inconsistent comparison baselines in ARDS prognostic studies are still severe,which limit the in-depth development of ARDS prognosis studies.To address above problems,this paper proposes an evaluation framework for ARDS prognostic prediction.On the one hand,it proposes a governance strategy for multi-source data to solve the problem of non-uniform ARDS data format;on the other hand,the framework forms an evaluation system consisting of 13 machine learning models and 20 feature sets to solve the problem of inconsistent comparison baselines.Experimental results show that the framework effectively addresses above problems and evaluates the impact of datasets,models,and features.