gesttools: General Purpose G-Estimation in R

Abstract

In this paper we present gesttools, a series of general purpose, user friendly functions with which to perform g-estimation of structural nested mean models (SNMMs) for time-varying exposures and outcomes in R. The package implements the g-estimation methods found in Vansteelandt and Sjolander (2016) and Dukes and Vansteelandt (2018), and is capable of analysing both end of study and time-varying outcome data that are either binary or continuous, or exposure variables that are either binary, continuous, or categorical. It also allows for the fitting of SNMMs with time-varying causal effects, effect modification by other variables, or both, as well as support for censored data using inverse weighting. We outline the theory underpinning these methods, as well as describing the SNMMs that can be fitted by the software. The package is demonstrated using simulated, and real-world inspired datasets.

Assessing the impact of periodontal therapy on tooth loss: a register-based longitudinal study in Denmark

“…G-estimation of structural nested mean models 13,14,[18][19][20][21] was used to estimate the average treatment effect of having received periodontal therapy in a calendar year on the number of extracted teeth in the subsequent years. Traditional statistical analysis methods are not suitable for such causal inference due to the presence of time-varying exposure and outcome, and time-varying confounding occurring when confounding is affected by earlier exposure or outcome status, or both.…”

“…13,14,19 However, g-estimation of structural nested mean models is one approach to handle this situation and thus to estimate all three desired treatment effects, E1 on O2, E1 on O3, and E2 on O3. The simplified outline of the g-estimation process 13,14,[18][19][20][21] goes as follows: Estimation proceeds by focusing in turn on each outcome time (O2 and O3) and sequentially estimating the treatment effects from previous times (E1 and E2) on that outcome. Importantly, the latest outcome, O3, can be considered as the result of accumulation of the treatment effects from all previous time points (E1 and E2).…”

“…13,14,[18][19][20][21] G-estimation models can also be supplemented with weights from a censoring model to control for potentially differing drop-out probabilities depending on previous exposure, outcome, and confounder values. 13,14,[18][19][20][21] Even though being a sophisticated approach, three common causal inference assumptions need to hold in order to correctly identify the causal effect with g-estimation; conditional exchangeability, that is no unmeasured confounding; positivity, that is above zero probability of receiving exposure; and counterfactual consistency, which dictates that an "individual's counterfactual outcome under a specific set of exposures is equal to their outcome had it been their observed exposure history." 20 Rather than reporting all three effect estimates from the DAG example (E1 on O2, E1 on O3, and E2 on O3) -which with 10 time points could amount to 55 effects (1+2+…+10=55) -one can estimate the effect of exposure on the outcome up to a specified number of time points between exposure and outcome.…”

A Tutorial on Causal Inference in Longitudinal Data With Time-Varying Confounding Using G-Estimation

“…In doing so, we hope to make g-estimation more accessible to psychology researchers who are already familiar with the features and functionalities of the widely used lavaan platform. Readers interested in alternative g-estimation procedures designed to deal with a broad variety of research settings and using other software packages or platforms as further points of investigation are referred to Picciotto and Neophytou (2016), Sterne and Tilling (2002), Tompsett et al (2022), and Wodtke (2018); for closely related approaches, although framed slightly differently, see Ertefaie et al (2021) and Simoneau et al (2018). For example, g-estimation has been extended to noncontinuous outcomes, such as binary outcomes (Dukes & Vansteelandt, 2018) and time-to-event data with censored outcomes (Seaman et al, 2021; Vansteelandt and Sjölander, 2016).…”

A tutorial on causal inference in longitudinal data with time-varying confounding using g-estimation

“…In doing so, we hope to make g-estimation more accessible to psychology researchers who are already familiar with the features and functionalities of the widely-used lavaan platform. Readers interested in alternative g-estimation procedures designed to deal with a broad variety of research settings, as well as using other software packages or platforms, as further points of investigation are referred to Picciotto and Neophytou (2016), Sterne and Tilling (2002), Tompsett et al (2022), andWodtke (2018); as well as Ertefaie et al (2021) andSimoneau et al (2018) for closely related approaches though framed slightly differently. For example, g-estimation has been extended to non-continuous outcomes, such as binary outcomes (Dukes & Vansteelandt, 2018) and time-to-event data with censored outcomes (Seaman et al, 2021;Vansteelandt & Sjölander, 2016).…”