COVID-19 Modulates the Effect of Trait Anxiety on Adverse Delivery Outcomes: An Exploratory XAI-Guided Study

Previous reports suggest a link between gestational COVID-19 and adverse gestational outcomes. However, many factors other than COVID-19 impact pregnancy. Our objective was to use an explainable artificial intelligence (XAI) screening strategy to identify maternal exposures associated with both COVID-19 infection and delivery complications, and to test whether COVID-19 acts as an effect modifier of those exposures. In a cross-sectional cohort of 120 pregnant people (42 COVID, 78 CONTROL), we measured fifteen socio-demographic, obstetric and psychological variables, including State-Trait Anxiety Inventory scores. XAI Logistic models with nested cross-validation first predicted COVID-19 status, and then delivery complications, to select features. Variables highlighted by both screens were evaluated by moderation models. XAI showed that trait anxiety was associated with both pregnancy complications and COVID-19. This variable was set as the exposure in a moderation analysis where COVID-19 was a moderator and gestation complications were the outcome. COVID-19 showed no main effect on delivery complications but significantly amplified the effects of trait anxiety. These hypothesis-generating findings illustrate how XAI can uncover patterns that traditional methods might overlook. Future studies intend to include replication in larger cohorts.既往研究表明，妊娠期新冠病毒感染（gestational COVID-19）与不良妊娠结局存在关联。然而，除新冠病毒感染外，尚有诸多因素可对妊娠造成影响。本研究旨在采用可解释人工智能（explainable artificial intelligence, XAI）筛查策略，识别同时与新冠病毒感染及分娩并发症相关的孕产妇暴露因素，并检验新冠病毒感染是否可作为上述暴露因素的效应修饰因子。本研究纳入120名孕产妇组成的横断面队列（其中新冠感染组42例、对照组78例），测定了15项社会人口学、产科及心理学变量，包括状态-特质焦虑量表（State-Trait Anxiety Inventory）评分。采用嵌套交叉验证的可解释人工智能逻辑回归模型，首先预测新冠病毒感染状态，随后预测分娩并发症，以此筛选特征变量。经两轮筛查均被筛选出的变量，通过调节效应模型进行评估。可解释人工智能分析结果显示，特质焦虑与妊娠并发症及新冠病毒感染均存在关联。在以新冠病毒感染为调节因子、妊娠并发症为结局变量的调节效应分析中，该变量被设定为暴露因素。新冠病毒感染对分娩并发症无显著主效应，但可显著强化特质焦虑的效应。这些具有假说生成价值的研究结果，阐明了可解释人工智能如何挖掘传统方法可能遗漏的潜在关联模式。未来研究计划纳入更大规模的队列以验证本研究结果。