The need to consider the impact of previous stressors on current stress parameter measurements

Dear Editor, in their very interesting article, Lucas et al. (2007) use logistic and fractional polynomial regression techniques and show in a large group (n = 302) of healthy 41- to 46-year-old men and women that higher stress levels were associated with higher serum neopterin concentration and both elevated and depressed salivary IgA levels. The authors conclude from their findings that stress may have bi-directional effects on immune mechanisms depending on the individual's amount of allostatic load (McEwen 1998): individuals with low level of allostatic load respond to stress with an up-regulation of immune parameters whereas others with high level of allostatic load respond with a down-regulation indicating a decompensation of the system. This does not seem to be a reasonable interpretation as the individuals in the Lucas et al. (2007) study were healthy and middle-aged with little reason to assume that they were chronically stressed according to the concept of allostatic load (McEwen 1998). Moreover, if different levels of allostatic load had been the reason for the bi-directional findings on stress-associated IgA concentrations neopterin should have been affected in the same way, which was not the case.

Rather, the findings by Lucas et al. (2007) may reflect some kind of inconsistency which calls attention to giving much more consideration to the complex nature of stress research data than has already been done. The common way to deal with circadian rhythms and other sources of variation in stress parameter levels (e.g. activity level, food ingestion) is to strictly control the timing of specimen collection, i.e. all individuals under study provide their samples at the same time of day (Tremblay et al. 1995). However, this static approach and its associated statistical methods are not adequate means to avoid biased results in stress research as they cannot sufficiently describe the dynamic characteristics of the individual–environment interaction.

In our “integrative single-case studies” on patients with systemic lupus erythematosus (SLE) (Schubert et al. 1999, 2002, 2003) extensive time series were gathered in 12 h intervals over approximately two months (at least 50 consecutive measurements) in order to carefully analyze the impact of everyday incidents on the dynamic course of stress- and SLE disease-related biochemical parameters by applying time series analysis according to Box and Jenkins (1976) (i.e. ARIMA modelling, cross-correlational analysis). In these studies, modelling time series revealed strong regularities (e.g. circadian rhythm) in urinary cortisol and neopterin concentrations as well as more or less clear deviations from these regularities. Moreover, the deviations were found to be bi-directional, i.e. the deviation went either up or down, thus accentuating or antagonizing the time series regular behaviour.

In a search for any explanations for these deviations, data from careful qualitative analysis and coding of everyday stressors that occurred during the study period were then applied. Adjusted cross-correlational analysis showed that some alterations in urinary cortisol and neopterin levels occurred with a time delay of 12–36 h (cortisol) and 36–60 h (neopterin) following emotionally meaningful everyday stressors and that these alterations were cyclic in nature: cortisol levels first increased in response to stressors after 12–24 h and then decreased after a total of 36 h, whereas neopterin levels first decreased after 36 h and then increased after a total of 60 h. Such cyclic stress parameter patterns following everyday incidents may be indicators of delayed, adaptive immunoneuroendocrine circuits activated to counter psychosocially triggered perturbations (Schubert et al. 2002, 2003).

Taking these first findings on the dynamic characteristics of stress research data into consideration, an alternative explanation for the bi-directional findings on salivary IgA levels in the Lucas et al. (2007) study is possible: First, an individual's salivary IgA concentration on the study day could have been caused not only by stress experienced on that day but also by stressors preceding the current measurement by one or even more days. This might count particularly for those individuals who perceive themselves as having experienced more stress in the recent past. Second, if the deviation from the regular IgA behaviour (e.g. circadian rhythm) on the study day had been caused by preceding stressors, this deviation could have been either up or down (bi-directional) depending on the phase of the cyclic pattern in response to the preceding stressors. Thus, not controlling for the possible impact of recent stressors that may have occurred up to several days before the current salivary IgA bears the risk of producing biased findings when aggregated across individuals and correlated with ratings of perceived stress.

It is theoretically well accepted in modern stress research (Lazarus 1991) but hardly ever taken empirically into account that individuals entering a study protocol have a history of emotionally negative (and positive) experiences that can influence a current stress measurement in a rather intense and complex way. We suggest that neglecting these potential influences in conventional group studies can cause inconsistencies within as well as between studies and that careful analysis of an individual's psychosocial reality together with sophisticated time series analysis can help to minimize this problem.

Matching the study design to the research question

Dr Schubert raises several interesting questions in his comments on our study (Lucas et al. 2007; Schubert 2008), particularly in relation to study design. He highlights the complexities of studying biological effects of a highly dynamic exposure, where the biological response may be conditional on previous experiences and exposures.

Different study designs will have different strengths and weaknesses with regard to evaluation of these issues. Our study (Lucas et al. 2007) used a cross-sectional design to examine the association between perceived stress and two measures of immune function in a large population-based sample of healthy middle-aged men and women. “Healthy” in this study meant the absence of overt physical disease. No assumptions were made about the underlying stress status of participants – within this population representative of the socioeconomic spectrum it would seem likely that the full range of underlying allostatic load status could be present. Indeed, although Schubert (2008) notes that if underlying allostatic load variation was the reason for the finding of a bi-directional effect on immune function one would expect both IgA and neopterin to be affected similarly, we did demonstrate a (non-significant) bi-directional effect on neopterin.

Schubert's studies on stress in an individual with SLE using a longitudinal design with intense exposure and outcome assessment (Schubert et al. 1999, 2003) have provided considerable data on the dynamic association between stress and various biological measures. While this setting provides data to examine questions relating to individual diurnal rhythms in stress parameters and associations between stressor exposure and biological measures, the results are not generalizable at the population level (diseased or healthy) due to the very small sample size (n = 1 in each study).

A strength of our study is that it is a large population-based sample, representative of the underlying population from which it is drawn. In addition, our study sample consists of healthy middle-aged people (a somewhat neglected research group), rather than those affected by an autoimmune disease, which itself may influence the association between stress and the immune response.

It is important to match the study design to the research question. Both the longitudinal intense study of few individuals and the population-based study of a larger, representative, sample contribute useful information. Notably, neither of these study designs is ideal for looking at long-term previous stressors. This would require a large population-based longitudinal study beginning during the in utero period to take account of maternal stress and its effects on the stress/immune axis of the developing fetus.

We recognise and acknowledged the limitations of our study design (Lucas et al. 2007). In this study setting controlled timing of specimen collection was not feasible and we used statistical methods to adjust for the time of specimen collection. The study questionnaire did not aim to measure the stress immediately before sample collection, but aimed to provide a global assessment of perceived stress in the time proximal to the biological measurement. The measurement of the effect of immediately antecedent stress is perhaps better suited to laboratory stress challenges.

Schubert (2008) suggests that study validity requires a “careful analysis of an individual's psychosocial reality together with sophisticated time series analysis.” We agree that a carefully designed large population-based prospective study, considering not only the psychosocial reality, but a wide range of other factors (some of which we did examine), including lifestyle factors such as smoking, drug and alcohol intake and physical activity, with frequent measurement of both stressor exposure and perception and a range of biological parameters would provide valuable data on the association between stress perception and immune function. One realistic alternative is to use a variety of study designs and statistical tools appropriate to the (narrowly defined) research question, and integrate the ensuing research findings.

R. M. Lucas

National Centre for Epidemiology and Population Health, ANU College of Medicine and Health Sciences, The Australian National University, Canberra, ACT, 0200, Australia, 61 02 6125 3448, 61 02 6125 5614, [email protected]

A.-L. Ponsonby

Murdoch Childrens Research Institute, Melbourne, Victoria, 3052, Australia