Open Collections

Kafi, Mohammad Abdullah Heel

University of British Columbia

In epidemiological research, the association between treatment and outcome may vary across values of a third variable, termed an “effect modifier”. Studying effect modifiers can help to identify subgroups for which treatment is beneficial or harmful. Standard approaches to assessing the impact of an effect modifier include subgroup analysis or adding an interaction term between treatment and the effect modifier in the regression analysis. Various propensity score (PS) based methods, also have been proposed, but it is unclear whether the PS should be derived based on the full dataset or within subgroups, or how these PS’s should then be used (matching, weights, stratification, covariate adjustment). We explored the performance of various PS-based approaches using simulation to compare how well the approach works under various scenarios. We performed comprehensive Monte Carlo simulation studies considering various scenarios, plasmode simulation, and empirical data analysis to evaluate the performance of all possible combinations of effect modifier approaches within all propensity score methods for various scenarios. We found that, for various simulation scenarios, there were comparable differences among various effect modifier approaches. For example, under simulation scenarios such as the model with additivity, mild non-linearity, moderate non-additivity, the subgroup-specific PS approach produced higher bias than the overall PS approach. PS matching approaches performed better than PS weighting methods in the model’s scenarios with mild non-linearity, mild non-additivity. In Monte Carlo simulation with interaction, the useful interaction approach had more downward bias, lower mean squared error than the overall PS, and subgroup-specific approach, in various scenarios such as additive, mild-linearity, moderate non-linearity. Our findings demonstrate that, under different scenarios, different effect modifiers approaches can be performed better than others. Therefore, this study will help the researcher to identify the appropriate approaches to deal with effect modifier with the appropriate PS methods under a given practical setting.

Text

2021-05-04

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/78089

Population and Public Health

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/