How to use marginal structural models in randomized trials to estimate the natural direct and indirect effects of therapies mediated by causal intermediates

Oba, Koji; Sato, Tosiya; Ogihara, Toshio; Saruta, Takao; Nakao, Kazuwa

doi:10.1177/1740774511402526

Cited by 6 publications

(8 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Mediation analyses seek a more in-depth understanding by decomposing the intention-to-treat effect into a direct and indirect effect, mediated by given intermediaries [1]. For instance, Oba et al [2] contrasted two treatments that equally suppressed the incidence of cardiovascular events but had a small but significantly different effect on systolic blood pressure. A mediation analysis provided a more in-depth understanding by clarifying what the (relative) treatment effect would be if an effect on systolic blood pressure could be avoided.…”

Section: Introductionmentioning

confidence: 99%

Boosting the precision of mediation analyses of randomised experiments through covariate adjustment

Vandenberghe

Vansteelandt

Loeys

2017

Statistics in Medicine

View full text Add to dashboard Cite

Analyses of randomised experiments frequently include attempts to decompose the intention-to-treat effect into a direct and indirect effect, mediated by given intermediaries, with the aim to shed light onto the treatment mechanism. Methods from causal mediation analysis have facilitated this by allowing for arbitrary models for the outcome and the mediator. They thereby generalise the traditional approach to direct and indirect effects, which is essentially limited to linear models. The default maximum likelihood methods make use of a model for the conditional distribution of the mediator, given treatment and baseline covariates, but are prone to bias when that model is misspecified. In randomised experiments, specification of such model can be easily avoided, but at the expense of a sometimes major efficiency loss when those baseline covariates are predictive of the mediator. In this article, we develop a compromise approach: it makes use of a model for the mediator to optimally extract information from the baseline covariate data but is insulated from the impact of misspecification of that model; it achieves this by exploiting the known randomisation probabilities. Simulation studies and the analysis of a randomised study show major efficiency gains and confirm our theoretical findings that the default methods from causal mediation analysis are sometimes, although not always, reasonably robust to model misspecification. Copyright © 2017 John Wiley & Sons, Ltd.

show abstract

Section: Introductionmentioning

confidence: 99%

Boosting the precision of mediation analyses of randomised experiments through covariate adjustment

Vandenberghe

Vansteelandt

Loeys

2017

Statistics in Medicine

View full text Add to dashboard Cite

show abstract

“…In addition to the sequential g‐estimation method of SNMMs, the inverse‐probability weighted estimation of MSMs and g‐computation with Monte Carlo simulations after fitting the parametric models of each variable's conditional distribution using maximum likelihood methods (parametric g‐formula) may also be readily applicable methods for estimating controlled direct effects using standard software. Yet SNMMs , MSMs , and g‐formula differ in causal parameters that analysts target for and in their modeling assumptions.…”

Section: Discussionmentioning

confidence: 99%

“…Although these methods have been originally intended to cope with longitudinal settings where treatment changes during the follow‐up period and intermediate variables also change over time, most studies demonstrated these methods on point‐treatment or point‐intermediate situations or both . In the presence of repeated measures data, treatment itself is likely to be confounded by covariates that are affected by prior treatment, which results in a time‐dependent confounding problem .…”

Section: Introductionmentioning

confidence: 99%

“…G‐computation can be directly applied to repeated measures data especially by parametric modeling of covariate densities followed by Monte Carlo simulation . Direct‐effect MSMs have also been illustrated to define controlled (and ‘natural’ ) direct effects in repeated measures data that can be estimated through inverse‐probability weighted model fitting (and additional assumptions and steps) .…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Estimation of controlled direct effects in time‐varying treatments using structural nested mean models: application to a primary prevention trial for coronary events with pravastatin

Shinozaki

Matsuyama

Ohashi

2014

Statistics in Medicine

View full text Add to dashboard Cite

For the estimation of controlled direct effects (i.e., direct effects controlling intermediates that are set at a fixed level for all members of the population) without bias, two fundamental assumptions must hold: the absence of unmeasured confounding factors for treatment and outcome and for intermediate variables and outcome. Even if these assumptions hold, one would nonetheless fail to estimate direct effects using standard methods, for example, stratification or regression modeling, when the treatment influences confounding factors. For such situations, the sequential g-estimation method for structural nested mean models has been developed for estimating controlled direct effects in point-treatment situations. In this study, we demonstrate that this method can be applied to longitudinal data with time-varying treatments and repeatedly measured intermediate variables. We sequentially estimate the parameters in two structural nested mean models: one for a repeatedly measured intermediate and the other one for direct effects of a time-varying treatment. The method was applied to data from a large primary prevention trial for coronary events, in which pravastatin was used to lower the cholesterol levels in patients with moderate hypercholesterolemia.

show abstract

“…We found that LDL levels lower than target values could reduce both cardiovascular events (HR = 0.72, 95% CI = 0.48-1.06, P = 0.10) and diabetes-related events (HR = 0.82, 95% CI = 0.58-1.17, P = 0.29) independently of atorvastatin, which might show the presence of an LDL-loweringmediated effect of atorvastatin. Formal statistical assessment as to atorvastatin's cholesterol-lowering-mediated effects is the subject of future work, which might be carried out using the marginal structural models 33,34 or the structural nested models. 35,36 Several studies have reported the possibility that statins might increase diabetic risk by increasing blood glucose levels.…”

mentioning

confidence: 99%

Effective prevention of cardiovascular disease and diabetes‐related events with atorvastatin in Japanese elderly patients with type 2 diabetes mellitus: Adjusting for treatment changes using a marginal structural proportional hazards model and a rank‐preserving structural failure time model

Shinozaki

Matsuyama²,

Iimuro³

et al. 2012

Geriatrics Gerontology Int

View full text Add to dashboard Cite

Aim:To assess the preventive effect of atorvastatin on cardiovascular disease and on diabetes-related events in elderly type 2 diabetic patients enrolled in the Japanese Elderly Diabetes Intervention Trial (J-EDIT).Methods: Data were obtained from 1173 patients aged 65-84 years who were enrolled in the J-EDIT. Patients were followed prospectively for 6 years to determine the effects of atorvastatin on serum cholesterol levels, and cardiovascular and diabetes-related events. Because the study protocol allowed atorvastatin to be prescribed according to the clinical needs of each patient, we regarded the J-EDIT data as if they came from a cohort study. We adjusted for clinical characteristics during the study as time-dependent confounders using two methods, inverse-probability-of-treatment (IPT) weighting and g-estimation method. Results: The total follow-up period was 5310.8 person-years (5.7 years of median follow up), during which 202 patients received atorvastatin treatment. Atorvastatin was associated with moderate reductions in cholesterol levels: 24.2 mg/dL for total cholesterol, 22.9 mg/dL for low-density lipoprotein (LDL) cholesterol and 24.3 mg/dL for non-highdensity lipoprotein cholesterol at the first post-treatment year. As a result, the proportion of patients who achieved targeted levels of LDL cholesterol clearly increased after atorvastatin treatment. Eight patients in 476.6 person-years among atorvastatin-treated and 113 untreated patients in 4721.4 person-years had cardiovascular events (the composite endpoint of fatal/non-fatal myocardial infarction, angina pectoris, coronary intervention, and fatal/non-fatal cerebrovascular disease); hazard ratio (HR) = 0.48, 95% confidence interval (CI) = 0.19-1.16, P = 0.10, and HR = 0.32, 95% CI = 0.05-1.87, P = 0.21 from IPT weighting and g-estimation method, respectively. Furthermore, seven in 475.0 person-years among atorvastatin-treated and 149 untreated patients in 4682.4 person-years had diabetes-related events (the composite end-point of sudden death, renal failure death, death as a result of hyperglycemia or hypoglycemia, diabetic gangrene and congestive heart failure in addition to cardiovascular event); HR = 0.30, 95% CI = 0.12-0.77, P = 0.01, and HR = 0.40, 95% CI = 0.09-0.89, P = 0.03 from IPT weighting and g-estimation method, respectively. When cardiovascular events were further differentiated into coronary vascular and cerebrovascular events, atorvastatin especially decreased the cerebrovascular risk. Conclusion:The use of atorvastatin to lower cholesterol levels in elderly Japanese patients with type 2 diabetes mellitus appears to reduce the risk of cardiovascular and diabetes-related events. Geriatr Gerontol Int 2012; 12 (Suppl. 1): 88-102.

show abstract

How to use marginal structural models in randomized trials to estimate the natural direct and indirect effects of therapies mediated by causal intermediates

Abstract: Background Although intention-to-treat analysis is a standard approach, additional

Cited by 6 publications

References 33 publications

Boosting the precision of mediation analyses of randomised experiments through covariate adjustment

Boosting the precision of mediation analyses of randomised experiments through covariate adjustment

Estimation of controlled direct effects in time‐varying treatments using structural nested mean models: application to a primary prevention trial for coronary events with pravastatin

Effective prevention of cardiovascular disease and diabetes‐related events with atorvastatin in Japanese elderly patients with type 2 diabetes mellitus: Adjusting for treatment changes using a marginal structural proportional hazards model and a rank‐preserving structural failure time model

Contact Info

Product

Resources

About