Analysis Of The Causal Effect Of COVID-19 Pandemic On Maternal Thyroid Function Based On Additive Model

The COVID-19 epidemic outbreak in early 2020 has deeply affected everyone.This article focuses on the causal effect of the development of the epidemic on the level of thyroid function in pregnant women.The epidemic,as an external stimulus,is likely to increase psychological stress on pregnant women,thereby affecting their thyroid function,which may cause adverse consequences for pregnant women and their fetuses.Considering the characteristics of thyroid function test data during pregnancy provided by Shanghai First Maternity and Infant Hospital,this paper uses a regression method based on additive model as the baseline method,and chooses three causal effect estimation methods suitable for continuous treatment and nonlinear effect curves: an inverse probability weighting method based on the generalized propensity score,a doubly robust weighted regression method and a non-parametric doubly robust method.These methods also estimate the curve by using additive models.Among the last three methods,this paper uses a Super-Learner-based non-parametric method to estimate weights,and proposes a truncated version of the weights.All these methods are applied to estimate the effect curve of the number of new reported infected cases per day on the FT4 level of pregnant women’s thyroid function the next day.The results show that as the number of new reported cases increases daily,the FT4 level of pregnant women will decrease slightly.The curve estimated by the regression method is linear,and the curve estimated by the non-parametric doubly robust method with truncated weights is very close to a straight line.Furthermore,assuming that the effect curve is linear,the regression method and the non-parametric doubly robust method calculate the effect coefficients of-0.0144 and-0.0123,respectively,which have significant statistical significance at the significance level of 0.05.However,the coefficient itself is quite small,so it can be considered that the causal effect of the development of the COVID-19 epidemic on the FT4 level of pregnant women exists but is not that strong.In addition,this paper constructs simulation datasets with different characteristics to compare the performance of these seven estimation methods.The two doubly robust methods do show doubly robustness.The non-parametric method often performs better,in comparison with the weighted regression method.Regardless of whether the weights are truncated or not,the imbalance of the covariates’ distribution can be well adjusted.The use of truncated weights usually results in a smaller RMSE of the estimate.