Sensitivity analysis
We performed a sensitivity analysis to evaluate the possible selective
biases made by PSM and the multivariate non-conditional logistics
regression was employed. The main pregnancy outcomes mentioned before
were analyzed respectively using logistics-regression to evaluate the
efficacy of aspirin. The independent variables including aspirin-intake,
age, BMI, BMI increase, gravidity, Chorionic types, and fertilization
methods.
As shown in supplementary tables, the logistic regression models were
made for each outcome. The protective role of LDA on PE and PTB is
undoubted, for the identical conclusions between the two methods.
However, the effect of LDA on SGA was uncertain as to the disparate
outcomes before (AOR: 0.76, 95%CI: 0.45-0.91) and after PSM (OR:0.74,
95%CI: 0.55-1.00). In the whole-data regression model, no matter which
variable was removed, the results still kept the same as before, and so
did in the regression model after PSM. That’s to say, the discrepancy
was not caused by single variables. When PSM was performed, 42 and 420
subjects were excluded in the LDA and control group. Among them, the
percentage of SGA was a little more in the control group (23.8%vs. 27.4%, p =0.717), and this may make some minor
difference in the conclusion. (Table S1-S5)
According to sensitivity analysis, we can conclude no obvious biases
were made by PSM, and we can confirm LDA as a protector to PE and PTB.
For SGA, we still considered it could also provide some benefits against
SGA, and this point needs further studies to confirm.