Selection bias and information bias are common and difficult problems in clinical research. Selection bias means that patients with different characteristics have different opportunities to enter the study (for example, patients with severe illness are more likely to participate in a therapeutic study), resulting in study results that are not representative of the target population. Information bias is mostly related to inaccurate measurement or information collection (for example, patients with different characteristics have bias in recalling past medical history). Both of these biases may make the results deviate from the real situation and reduce the credibility of the conclusions. How to evaluate and control the two kinds of bias through effective analysis and supplement sensitivity analysis is a difficult point. The use of inverse probability of treatment weighting (IPTW) based on propensity score (PS) can be used for sensitivity analysis in this case. For example, in a study, patients with ideal efficacy have a higher proportion of shedding during long-term follow-up, which will cause the actual study subjects to deviate from the target population. At the same time, this deviation itself is related to prognosis, and the selection bias generated at this time may lead to wrong conclusions. Since the actual outcome of shedding patients cannot be obtained, the magnitude and direction of bias cannot be judged by simple multivariate analysis, and PS-based IPTW can be used with the following steps:
