Instrumental variable estimation of the marginal structural Cox model for time-varying treatments

Biometrika ◽  
2021 ◽  
Author(s):  
Y Cui ◽  
H Michael ◽  
F Tanser ◽  
E Tchetgen Tchetgen
Keyword(s):  
Instrumental Variable ◽  
Causal Effect ◽  
Cox Model ◽  
Sufficient Conditions ◽  
Structural Models ◽  
Model Parameters ◽  
Marginal Structural Models ◽  
Time Varying ◽  
Unmeasured Confounding ◽  
Sequential Randomization

Summary Robins (1998) introduced marginal structural models, a general class of counterfactual models for the joint effects of time-varying treatments in complex longitudinal studies subject to time-varying confounding. Robins (1998) established the identification of marginal structural model parameters under a sequential randomization assumption, which rules out unmeasured confounding of treatment assignment over time. The marginal structural Cox model is one of the most popular marginal structural models to evaluate the causal effect of time-varying treatments on a censored failure time outcome. In this paper, we establish sufficient conditions for identification of marginal structural Cox model parameters with the aid of a time-varying instrumental variable, when sequential randomization fails to hold due to unmeasured confounding. Our instrumental variable identification condition rules out any interaction between an unmeasured confounder and the instrumental variable in its additive effects on the treatment process, the longitudinal generalization of the identifying condition of Wang & Tchetgen Tchetgen (2018). We describe a large class of weighted estimating equations that give rise to consistent and asymptotically normal estimators of the marginal structural Cox model, thereby extending the standard inverse probability of treatment weighted estimation of marginal structural models to the instrumental variable setting. Our approach is illustrated via extensive simulation studies and an application to estimate the effect of community antiretroviral therapy coverage on HIV incidence.

scholarly journals A graphical perspective of marginal structural models: An application for the estimation of the effect of physical activity on blood pressure

2016 ◽  
Vol 27 (8) ◽  
pp. 2428-2436
Author(s):  
Denis Talbot ◽  
Amanda M Rossi ◽  
Simon L Bacon ◽  
Juli Atherton ◽  
Geneviève Lefebvre
Keyword(s):  
Physical Activity ◽  
Blood Pressure ◽  
Repeated Measures ◽  
Structural Equation ◽  
Structural Equation Models ◽  
Causal Effect ◽  
Structural Models ◽  
Marginal Structural Models ◽  
Time Varying ◽  
Causal Graph

Estimating causal effects requires important prior subject-matter knowledge and, sometimes, sophisticated statistical tools. The latter is especially true when targeting the causal effect of a time-varying exposure in a longitudinal study. Marginal structural models are a relatively new class of causal models that effectively deal with the estimation of the effects of time-varying exposures. Marginal structural models have traditionally been embedded in the counterfactual framework to causal inference. In this paper, we use the causal graph framework to enhance the implementation of marginal structural models. We illustrate our approach using data from a prospective cohort study, the Honolulu Heart Program. These data consist of 8006 men at baseline. To illustrate our approach, we focused on the estimation of the causal effect of physical activity on blood pressure, which were measured at three time points. First, a causal graph is built to encompass prior knowledge. This graph is then validated and improved utilizing structural equation models. We estimated the aforementioned causal effect using marginal structural models for repeated measures and guided the implementation of the models with the causal graph. By employing the causal graph framework, we also show the validity of fitting conditional marginal structural models for repeated measures in the context implied by our data.

scholarly journals Causal inference in continuous time: an example on prostate cancer therapy

Biostatistics ◽  
2018 ◽  
Vol 21 (1) ◽  
pp. 172-185 ◽  
Author(s):  
Pål Christie Ryalen ◽  
Mats Julius Stensrud ◽  
Sophie Fosså ◽  
Kjetil Røysland
Keyword(s):  
Prostate Cancer ◽  
Continuous Time ◽  
Causal Effect ◽  
Structural Models ◽  
Marginal Structural Models ◽  
Cancer Treatments ◽  
Discrete Parameter ◽  
Inverse Probability Weights ◽  
Using Data ◽  
Prostate Cancer Treatments

Abstract In marginal structural models (MSMs), time is traditionally treated as a discrete parameter. In survival analysis on the other hand, we study processes that develop in continuous time. Therefore, Røysland (2011. A martingale approach to continuous-time marginal structural models. Bernoulli 17, 895–915) developed the continuous-time MSMs, along with continuous-time weights. The continuous-time weights are conceptually similar to the inverse probability weights that are used in discrete time MSMs. Here, we demonstrate that continuous-time MSMs may be used in practice. First, we briefly describe the causal model assumptions using counting process notation, and we suggest how causal effect estimates can be derived by calculating continuous-time weights. Then, we describe how additive hazard models can be used to find such effect estimates. Finally, we apply this strategy to compare medium to long-term differences between the two prostate cancer treatments radical prostatectomy and radiation therapy, using data from the Norwegian Cancer Registry. In contrast to the results of a naive analysis, we find that the marginal cumulative incidence of treatment failure is similar between the strategies, accounting for the competing risk of other death.

scholarly journals The causal effect of dynamic fertility treatment strategies on the probability of pregnancy: a novel application of marginal structural models (MSMs)

2019 ◽  
Vol 112 (3) ◽  
pp. e178
Author(s):  
Soudeh Ansari ◽  
Michael P. LaValley ◽  
Sara Lodi ◽  
Brooke Hayward ◽  
Gilbert L. Mottla ◽  
...  
Keyword(s):  
Causal Effect ◽  
Treatment Strategies ◽  
Structural Models ◽  
Fertility Treatment ◽  
Marginal Structural Models

Estimating the Causal Effect of Some Exposure From Nonrandomized Studies: The Example of Reduced Intensity Conditioning (RIC) in Hematological Diseases.

Blood ◽  
2009 ◽  
Vol 114 (22) ◽  
pp. 3365-3365
Author(s):  
Matthieu Resche-Rigon ◽  
Marie Robin ◽  
Regis Peffault de Latour ◽  
Sylvie Chevret ◽  
Gerard P Socie
Keyword(s):  
Quantitative Methods ◽  
Causal Effect ◽  
Autologous Transplantation ◽  
Structural Models ◽  
Marginal Structural Models ◽  
Reduced Intensity Conditioning ◽  
Hematological Diseases ◽  
Causal Pathway ◽  
The Impact ◽  
Reduced Intensity

Abstract Abstract 3365 Poster Board III-253 Introduction: Although allogeneic SCT with RIC has now gained wide acceptance, its eventual benefit again non-transplant approach is largely unknown (outside the setting of large randomized trials). When evaluating the impact on survival of reduced intensity conditioning in malignant hematological diseases, standard estimations based on Cox regression from observational databases could be biased because they ignore covariates that confound treatment decision. In this setting, we applied and compared two different statistical methods that were developed to control for confounding in estimating exposure (or treatment) effect from epidemiological studies. Patients and Methods: The statistical challenge was that allograft tended to be given when a patient was in advanced phase of his/her hematological malignancy, so that treatment was confounded by performance indicators, which in turn lie on the causal pathway between treatment and outcome. Thus, comparison of outcome first used propensity score (PS) analyses that attempt to create a comparison group of non-treated patients that closely resembles the group of treated patients by matching for the likelihood that a given patient has received the treatment. Then, we used marginal structural models (MSMs) that consist in creating, by using inverse probability of treatment weights, a pseudo-population in which the probability of treatment does no longer depend on covariates, and the effect of treatment on outcome is the same as in the original population. Result: Reduced intensity conditioning allograft was performed in 82 patients with chemotherapy-sensitive patients relapsing after autologous transplantation. Patients with myeloma (MM, 23 pts), follicular lymphoma (FL, 28 pts) or Hodgkin disease (HD, 31 pts), were compared to 276 patients who relapsed after autologous transplantation but did not underwent allogeneic stem cell transplantation (142 MM, 115 FL and 19 HD). From original datasets, 21 (91%) matched pairs could be constituted from MM patients, as compared to 19 (68%) of the FL patients, down to 15 (48%) of the HD patients. Based on these PS-matched samples, a significant benefit of reduced intensity conditioning as compared with non allografted patients was observed in MM, with estimated hazard ratio (HR) of death at 0.34 (95% confidence interval, CI: 0.14-0.88), as well as in FL (HR= 0.78, 95%CI: 0.27;2.30) and in HD (HR= 0.24; 95%CI: 0.09-0.62). MSM-based analyses that applied to the reweighted populations confirmed these trends towards survival benefits in FL, though partially erased in MM and HD. Conclusions: We reported the application of marginal structural models, a new class of causal models to estimate the effect of nonrandomized treatments as an alternative to PS based approaches in small samples. We expect that an increasing number of physicians involved in clinical cohorts become familiar with these novel and appealing quantitative methods when assessing innovative treatment effects. Disclosures: No relevant conflicts of interest to declare.

scholarly journals Causal inference under over-simplified longitudinal causal models

2021 ◽  
Vol 0 (0) ◽  
Author(s):  
Lola Étiévant ◽  
Vivian Viallon
Keyword(s):  
Causal Inference ◽  
Causal Effect ◽  
Sufficient Conditions ◽  
Repeated Measurements ◽  
Causal Models ◽  
Causal Effects ◽  
Sensitivity Analyses ◽  
Time Varying ◽  
Weighted Averages

Abstract Many causal models of interest in epidemiology involve longitudinal exposures, confounders and mediators. However, repeated measurements are not always available or used in practice, leading analysts to overlook the time-varying nature of exposures and work under over-simplified causal models. Our objective is to assess whether – and how – causal effects identified under such misspecified causal models relates to true causal effects of interest. We derive sufficient conditions ensuring that the quantities estimated in practice under over-simplified causal models can be expressed as weighted averages of longitudinal causal effects of interest. Unsurprisingly, these sufficient conditions are very restrictive, and our results state that the quantities estimated in practice should be interpreted with caution in general, as they usually do not relate to any longitudinal causal effect of interest. Our simulations further illustrate that the bias between the quantities estimated in practice and the weighted averages of longitudinal causal effects of interest can be substantial. Overall, our results confirm the need for repeated measurements to conduct proper analyses and/or the development of sensitivity analyses when they are not available.

Sign in / Sign up

Export Citation Format

Share Document