To assess, to control, to exclude: Effects of biobehavioral factors on circulating inflammatory markers

Abstract

Behavioral scientists have increasingly included inflammatory biology as mechanisms in their investigation of psychosocial dynamics on the pathobiology of disease. However, a lack of standardization of inclusion and exclusion criteria and assessment of relevant control variables impacts the interpretation of these studies. The present paper reviews and discusses human biobehavioral factors that can affect the measurement of circulating markers of inflammation. Keywords relevant to inflammatory biology and biobehavioral factors were searched through PubMed. Age, sex, and hormonal status, socioeconomic status, ethnicity and race, body mass index, exercise, diet, caffeine, smoking, alcohol, sleep disruption, antidepressants, aspirin, and medications for cardiovascular disease are all reviewed. A tiered set of recommendations as to whether each variable should be assessed, controlled for, or used as an exclusion criteria is provided. These recommendations provide a framework for observational and intervention studies investigating linkages between psychosocial and behavioral factors and inflammation.

Introduction

Inflammation, as indexed by increases of circulating levels of C-reactive protein (CRP) and proinflammatory cytokines such as interleukin-6 (IL-6), is thought to influence the onset and course of a wide spectrum of diseases including cardiovascular disease, arthritis, type 2 diabetes, and certain cancers (Ferrucci et al., 1999). Further data show that elevated levels of IL-6 prospectively predict future disability, declines of health status, and mortality risk, particularly in older adults (Reuben et al., 2002, Volpato et al., 2001). Given substantial evidence that psychosocial factors such as depression, psychological stress, and social isolation increase cardiovascular morbidity and mortality, and hasten the onset and negatively impact the course of other inflammatory disorders, behavioral scientists have increasingly evaluated circulating levels of inflammatory markers in an effort to understand the signaling pathways by which these psychosocial dynamics impact the pathobiology of disease.

Part of the difficulty in evaluating changes in circulating levels of inflammatory markers in relation to psychosocial factors is that human populations display wide variability in a vast array of variables that are known to affect the production and expression of proinflammatory markers. Hence, reliable conclusions about the influence of biobehavioral processes on circulating markers of inflammation can be jeopardized when comparing different populations and/or different clinical settings without a common methodological framework.

However, standardization of inclusion and exclusion criteria and assessment of relevant control variables has provided a platform for the study of immunity in older adults, as well as prior psychoneuroimmunology studies of innate and cellular immunity. For example, the SENIEUR protocol identified a set of demographic and clinical criteria that guided the selection of relatively healthy participants for the study of the effects of aging on immunity in such a way that exogenous and endogenous influences on immunity were reduced to a minimum. Ultimately this standardized protocol influenced a generation of research that dissected the contribution of disease, as opposed to the impact of normal aging processes, on the immune system (Castle et al., 2001, Ligthart, 2001), which arguably could be generalized to other populations including those with medical comorbidities. Likewise Kiecolt-Glaser and Glaser (1988) identified a number of key behavioral factors that exert an influence on the measurement of cellular immunity, and discussed issues related to assessment in evaluation of psychosocial influences on measures of lymphocyte proliferation, natural killer cell activity, and antibodies to latent viruses. To our knowledge, no prior review has discussed the methodological issues relevant for the study of behavioral processes and inflammation in humans.

The goal of the present paper is to review and discuss human biobehavioral factors that can affect the measurement of circulating markers of inflammation. We chose to focus on select circulating inflammatory markers (e.g., interleukins (ILs) and tumor necrosis factor-α (TNF-α), their soluble receptors, and CRP) because of their broad use (and assay procedures) and well-established connections with clinical health outcomes afflicting a large proportion of studied populations. This review is primarily based on studies of non-medical adult populations, rather than those who have a particular medical disorder which might require assessment of certain behavioral factors at a much more detailed level. In the case of medications, antidepressants, statins, and antihypertensives are quite common, and empirical evidence for their effects on inflammatory markers typically comes from individuals with depression or existing cardiovascular disease. In an effort to guide biobehavioral research, key behavioral factors are identified for assessment consideration, with a tiered set of recommendations as to whether the variable should be assessed, controlled for, or used as an exclusion criteria, recognizing that clinical research effectively balances these methodological issues with cost and effort of assessment. Suggestions for assessment of these variables are also provided.

We are sensitive to the appropriate selection of control variables in biobehavioral research, due to the potential for “overfitting” regression models. Variables should be chosen for a priori reasons, and this paper may help investigators do just that. However, controlling for too many variables may lead to sample-specific results, limiting the replicability. Although it is beyond the scope of this manuscript, guidelines exist for choosing control variables in relationship to sample size, and some journals even require these guidelines to be followed (Babyak, 2004).

Although assessing, controlling, and excluding variables is discussed, investigators may examine some variables as important components in a theoretical model. This may be either as a moderating variable that changes the relationship between the variables of interest (e.g., smoking magnifies the relationship between depression and inflammation), or as a mediating variable that explains the relationship between the variables of interest (e.g., the relationship between depression and inflammation might be partly explained by sleep disturbance (Irwin et al., 2006)).

We note that if investigators wish to examine some of these variables of interest as mediators, a key issue is the timing of assessment. Beyond the statistical criteria for mediation (Kraemer et al., 2001, MacKinnon et al., 2007), an important criterion is temporal precedence; that is, the mediating variables must be measured at some point after the independent variable and before the dependent variable. For instance, to establish that smoking explains/mediates the relationship between depression and inflammation, the smoking measures must be assessed sometime after depression is measured and before inflammatory markers are measured. However, it is beyond the scope of this paper to determine how to place these variables in investigators’ theoretical and design frameworks.