Research And Application Of Extreme Weight Solution In The Inverse Probability Weight Of Propensity Score

Objective: This paper intends to compare inverse probability weight(IPW)and its alternative methods trimming and overlap weights(OW),optimal covariate balancing propensity score(o CBPS),entropy balancing(EB)and stable weights that balance covariates(SBW)under finite overlap and propensity score model misspecification.In addition,these methods were used to explore the relationship between serum vitamin D deficiency,dietary vitamin D intakes and sleep duration in adults.Methods: The simulation set three sample sizes,the proportion of two groups receiving treatment,homogeneous and heterogeneous treatment effects,overlap degree of three groups of propensity score,the correct and wrong assignment of propensity score model.Based on the indicators of bias,root mean square error(RMSE),standard deviation(SD),the performance of IPW and alternative methods,such as trimming,OW,o CBPS,EB,and SBW,are compared with limited overlap and misspecified propensity score model.Examples are based on the US National Health and Nutrition Examination Survey(NHANES)database,Methods such as IPW and trimming,OW,o CBPS,EB and SBW are used to explore the causal relationship between serum total 25(OH)D,serum 25(OH)D2,serum 25(OH)D3,dietary vitamin D intake and sleep duration of adults.Results: Simulation results show that,in all simulation scenarios,IPW is sensitive to poor overlap and incorrect specification of the PS model.However,alternative methods Trimming,OW,o CBPS,EB and SBW maintain the ability to estimate causal effect with relatively small deviation and small standard deviation,showing stronger stability and higher efficiency.Among them,OW method provides the best causal effect estimation.A total of 4390 subjects were included in the case study.The results of 25(OH)D deficiency were basically consistent and statistically significant.Compared with 25(OH)D deficiency,the sleep time decreased by about 0.2h.The results of 25(OH)D3 deficiency were similar to those of 25(OH)D deficiency,and 25(OH)D3 deficiency was significantly correlated with short sleep duration.In the absence of 25(OH)D2,the results of each method were not statistically significant.Sleep duration decreased by about 0.12 h when dietary vitamin D intake was insufficient compared with adequate dietary vitamin D intake,and unweighted results were not statistically significant.Conclusions: Both simulation and case results show that OW method provides the best estimation of causal effect and can be used as an optimal alternative to IPW method.Therefore,it is suggested that when applying IPW method,OW method can be preferred to avoid extreme weight.This study found that a lack of serum vitamin D and its metabolites and a lack of dietary vitamin D intake reduced sleep duration in adults.