Living forwards or understanding backwards? A comparison of Inverse Probability of Treatment Weighting and G-estimation methods for targeting hypothetical full adherence estimands in longitudinal cohort studies

TL;DR

Two established longitudinal causal approaches designed to handle time-varying confounding under the ``no unmeasured confounding''(NUC) assumption are considered: G-estimation and inverse probability of treatment weighting (IPTW) and inverse probability of treatment weighting (IPTW).

Abstract

Medication adherence is essential to ensure treatment effectiveness, but too often in routine care non-adherence compromises the desired outcome. We explore longitudinal causal modelling using observational data to estimate the time-varying effects of continuous drug adherence measures on health outcomes over a sustained period. The goal of such analyses is to quantify the potential impact of interventions to improve adherence on long-term health. We consider two established longitudinal causal approaches designed to handle time-varying confounding under the ``no unmeasured confounding'' (NUC) assumption: G-estimation and inverse probability of treatment weighting (IPTW). In randomized controlled trial, NUC-based methods have been applied to address non-adherence as an intercurrent event, and instrumental variable (IV) extensions of G-estimation have also been introduced for settings where the NUC assumption may fail. We adapt these methods to observational data settings and illustrate their use for assessing how adherence over time impacts health outcomes. We align the causal parameters across methods and show they can target the same causal estimand: the average effect among treated individuals of full adherence versus zero adherence. We set out the identification conditions for IPTW and G-estimation under NUC, and for an IV-based extension that has specific utility when the NUC assumption is implausible. We assess the statistical properties, strengths and weaknesses of each approach through Monte Carlo simulations designed to reflect longitudinal studies with a continuous exposure. We demonstrate these methods by quantifying the effect of full statin adherence on LDL cholesterol control in 13,000 UK Biobank participants with linked primary care data.

Figures (5)

• Figure 1: Illustration of the assumed longitudinal causal structure with three time points. Arrows are colour-coded to reflect the components of the total causal effect of adherence at each time point: red for $A_1$, blue for $A_2$ and green for $A_3$. For visual clarity, we omit some arrows (such as the direct effect of $A_1$ on $\bm{L}_{3}$), but the framework allows for general relationships where variables may depend on the entire past history.
• Figure 2: Conceptual illustration of (a) IPTW, (b) SNMM G-estimation, and (c) IV-based G-estimation for the effect of adherence history $(A_1,A_2,A_3)$ on the selected final outcome $Y_3$. For visual clarity, we omit some arrows (such as the direct effect of $A_1$ on $\bm{L}_{3}$), but the framework allows for general relationships where variables may depend on the entire past history.
• Figure 3: Distributions of (a) LDL-c follow-up times and (b) MPR from baseline to the first LDL-c follow-up.
• Figure 4: Estimation results of IPTW, G-estimation and IV.
• Figure 5: Conceptual illustration of conditioning on statin initiation. $S$ indicates statin initiation. Solid black arrows represent the assumed causal structure. Conditioning on $S = 1$ (restricting to treated individuals) can induce additional associations between the instrument $G$, unobserved confounder $U$ and outcome $Y_k$ (shown as red dashed arrows), potentially violating the IV independence and exclusion assumptions within the selected sample.