Introduction Marginal structural models (MSMs) were developed to address time-varying confounding in nonrandomized exposure effect studies. It is unclear how estimates from MSMs to conventional models differ in real settings and how the MSMs are implemented in the literature.
Methods We systematically reviewed the literature of MSMs since 2000 retrieving papers from both PubMed and ISI Web of Knowledge databases.
Results Data to compare MSMs and conventional models were obtained from 65 papers reporting 164 exposure-outcome associations. In 18 (11.0%), the two techniques resulted in estimates with opposite interpretations, and in 58 (39.7%) estimates differed by at least 20%. The SEs of the MSM associations were in median 19.4% greater than the respective conventional SEs (IQR: 2.4% to 47.5%) in the 156 available associations. In 88 papers, MSMs were used to analyse real data; only 53 (60.2%) of these papers reported that stabilised inverse-probability weights (IPWs) were used, and only 28 (31.8%) reported that they verified that the mean of the stabilised IPWs was close to one.
Conclusions We found important differences between MSMs and conventional models in real studies. Reporting of MSMs can be improved.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.