Comparison of the Inverse Probability of Treatment Weighted (IPTW) Estimator With a Naïve Estimator in the Analysis of Longitudinal Data With Time-Dependent Confounding: A Simulation Study

Romain Neugebauer 01/2004;
Source: OAI

ABSTRACT A simulation study was conducted to compare estimates from a naïve estimator, using standard conditional regression, and an IPTW (Inverse Probability of Treatment Weighted) estimator, to true causal parameters for a given MSM (Marginal Structural Model). The study was extracted from a larger epidemiological study (Longitudinal Study of Effects of Physical Activity and Body Composition on Functional Limitation in the Elderly, by Tager et. al [accepted, Epidemiology, September 2003]), which examined the causal effects of physical activity and body composition on functional limitation. The simulation emulated the larger study in terms of the exposure and outcome variables of interest-- physical activity (LTPA), body composition (LNFAT), and physical limitation (PF), but used one time-dependent confounder (HEALTH) to illustrate the effects of estimating causal effects in the presence of time-dependent confounding. In addition to being a time-dependent confounder (i.e. predictor of exposure and outcome over time), HEALTH was also affected by past treatment. Under these conditions, naïve estimates are known to give biased estimates of the causal effects of interest (Robins, 2000). The true causal parameters for LNFAT (-0.61) and LTPA (-0.70) were obtained by assessing the log-odds of functional limitation for a 1-unit increase in LNFAT and participation in vigorous exercise in an ideal experiment in which the counterfactual outcomes were known for every possible combination of LNFAT and LTPA for each subject. Under conditions of moderate confounding, the IPTW estimates for LNFAT and LTPA were -0.62 and -0.94, respectively, versus the naïve estimates of -0.78 and -0.80. For increased levels of confounding of the LNFAT and LTPA variables, the IPTW estimates were -0.60 and -1.28, respectively, and the naïve estimates were -0.85 and -0.87. The bias of the IPTW estimates, particularly under increased levels of confounding, was explored and linked to violation of particular assumptions regarding the IPTW estimation of causal parameters for the MSM.

1 Bookmark
  • Source
    Article: Name
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We used agent-based simulation to examine the problem of time-varying confounding when estimating the effect of an adverse event on hospital length of stay. Conventional analytic methods were compared with inverse probability weighting (IPW). A cohort of hospitalized patients, at risk for experiencing an adverse event, was simulated. Synthetic individuals were assigned a severity of illness score on admission. The score varied during hospitalization according to an autoregressive equation. A linear relationship between severity of illness and the logarithm of the discharge rate was assumed. Depending on the model conditions, adverse event status was influenced by prior severity of illness and, in turn, influenced subsequent severity. Conditions were varied to represent different levels of confounding and categories of effect. The simulation output was analyzed by Cox proportional hazards regression and by a weighted regression analysis, using the method of IPW. The magnitude of bias was calculated for each method of analysis. Estimates of the population causal hazard ratio based on IPW were consistently unbiased across a range of conditions. In contrast, hazard ratio estimates generated by Cox proportional hazards regression demonstrated substantial bias when severity of illness was both a time-varying confounder and intermediate variable. The direction and magnitude of bias depended on how severity of illness was incorporated into the Cox regression model. In this simulation study, IPW exhibited less bias than conventional regression methods when used to analyze the impact of adverse event status on hospital length of stay.
    Medical Care 11/2007; 45(10 Supl 2):S108-15. · 3.23 Impact Factor