UC Berkeley

Causal inference methodology represents an analytical framework to evaluate and estimate causal effects based on observational data. This framework can be applied to examine and estimate individual contributions of causal factors in epidemiologic studies of aging. These studies are characterized by complex temporal relationships between these different factors, and confounders of these factors and outcomes of interest (i.e. time-dependent confounding). Unbiased estimation of effects under these conditions lies beyond the scope of conventional statistical methods. Causal inference methods allow for examination and unbiased estimation of population-level (i.e., marginal) effects, and can account for time-dependent confounding. Marginal Structural Models (MSMs) can be used to define causal parameters (i.e., marginal effects) of interest. History-adjusted marginal structural models (HAMSMs), a generalization of MSMs, can be used to define and evaluate marginal effects given time-varying covariates.

These models were applied to examine the causal associations between different factors of interest and physical and cognitive functioning outcomes. Estimation procedures were applied (e.g., targeted maximum likelihood estimation, inverse-probability of treatment weights) to address time-dependent confounding in the data and provide unbiased estimates of the effects examined in these models, which would not have been possible with standard statistical methods.

Study 1

Self-reported leisure-time physical activity (LTPA), based on metabolic equivalents, equal to or greater than public-health recommended-levels, and a measure of body composition (lean: fat mass ratio, L/F), estimated from bioelectric impedance using population-specific prediction equations derived from dual x-ray energy absorptiomery, were examined with respect to a measure of physical function that was based on standard self-report questions. In women 55 and older, a one-unit gain in L/F reduced by 65.5 percent (95% CI: 21.8, 87.4) the report of physical limitation at all four surveys of an 8-year study. A similar reduction was not observed in men; however, there was a 3 percent increase in the report of no limitation at any survey. The effect of high levels of LTPA was a reduction in new physical limitation that occurred at the last survey of 36.8 percent (95% CI: 0.0, 0.92) and 52.7 percent (95% CI: 13.5, 91.9) in men and women, respectively. In summary, higher LTPA appeared to reduce the risk of future functional limitation conditional on the level of functioning established by L/F.

Study 2

The association of lifetime household secondhand smoke exposure (SHS) and risk of incident dementia was examined in 970 participants in the Cardiovascular Health Cognition Study who were never-smokers and were free of clinical cardiovascular disease (CVD), dementia, and mild cognitive impairment at baseline. Given prior studies have found that SHS is associated with increased risk of CVD and that CVD is associated with increased risk of dementia, interactions between SHS and measures of clinical and subclinical CVD on dementia risk were examined as well. Moderate (16-25 years) and high (>25 years) SHS exposure levels were not independently associated with dementia risk; however, subjects with >25 years of SHS exposure and >25% carotid artery stenosis had a three-fold increase (Hazard Ratio, 3.00; 95% Confidence Interval: 1.03, 9.72) in hazard of dementia, compared to subjects with no/low (0-15 years) SHS exposure and ≤ 25% carotid artery stenosis. High lifetime SHS exposure may increase the risk of dementia in elderly with undiagnosed CVD.

Study3

Leisure-time physical activity (LTPA), based on metabolic equivalents, was examined with respect to walking speed (WS) that was based on standard protocol. A measure of body composition (Lean:fat mass ratio, L/F) (see Study 1), was included as a surrogate of metabolic function (e.g., glucose tolerance) and as a causal intermediate of LTPA and WS. In sex-specific analyses, the direct effects of LTPA on WS were estimated from four separate surveys of an 8-year study, and pooled. Stratified analyses examined effect estimates in different subgroups (e.g., diabetics vs. non-diabetics). Mean WS increased (2.394 ft/sec vs. 2.238 ft/sec in women; 2.418 ft/sec vs. 2.278 ft/sec in men) with higher LTPA (i.e., greater or equal vs. less than public health recommended levels) and higher L/F (i.e. > median vs. < median). In women, the direct effect of LTPA was an increase in mean WS for < median L/F (2.316 ft/sec vs. 2.238 ft/sec) and > median L/F (2.394 ft/sec vs. 2.316 ft/sec). Similar results were observed for the men. Results of the stratified analysis did not differ from the overall analysis. These results indicate a marginal level of direct protection of LTPA for WS, but underscore the influence of LTPA on metabolic intermediates that affect lower body-function.

The application of MSMs and HAMSMs in these different studies illustrate their use to examine causal factors associated with cognitive and physical decline. Moreover, estimation procedures were employed to provide accurate estimates of effects based on these models of physical activity and other factors associated with function in the elderly.