Elsevier

Psychoneuroendocrinology

Volume 74, December 2016, Pages 380-386
Psychoneuroendocrinology

Assessment of the cortisol awakening response: Real-time analysis and curvilinear effects of sample timing inaccuracy

https://doi.org/10.1016/j.psyneuen.2016.09.026Get rights and content

Highlights

  • All cortisol data objectively verified for awakening and saliva sampling times.

  • Cortisol growth curve plot against real-time not affected by protocol deviance.

  • In healthy young adults the mean CAR was a 100% increase from awakening cortisol.

  • Curvilinear delay effect on CAR size if protocol times wrongly assumed accurate.

  • Electronic-monitoring vital for CAR measurement and meaningful interpretation.

Abstract

The cortisol awakening response (CAR) is typically measured in the domestic setting. Moderate sample timing inaccuracy has been shown to result in erroneous CAR estimates and such inaccuracy has been shown partially to explain inconsistency in the CAR literature. The need for more reliable measurement of the CAR has recently been highlighted in expert consensus guidelines where it was pointed out that less than 6% of published studies provided electronic-monitoring of saliva sampling time in the post-awakening period.

Analyses of a merged data-set of published studies from our laboratory are presented. To qualify for selection, both time of awakening and collection of the first sample must have been verified by electronic-monitoring and sampling commenced within 15 min of awakening. Participants (n = 128) were young (median age of 20 years) and healthy. Cortisol values were determined in the 45 min post-awakening period on 215 sampling days. On 127 days, delay between verified awakening and collection of the first sample was less than 3 min (‘no delay’ group); on 45 days there was a delay of 4–6 min (‘short delay’ group); on 43 days the delay was 7–15 min (‘moderate delay’ group).

Cortisol values for verified sampling times accurately mapped on to the typical post-awakening cortisol growth curve, regardless of whether sampling deviated from desired protocol timings. This provides support for incorporating rather than excluding delayed data (up to 15 min) in CAR analyses. For this population the fitted cortisol growth curve equation predicted a mean cortisol awakening level of 6 nmols/l (±1 for 95% CI) and a mean CAR rise of 6 nmols/l (±2 for 95% CI). We also modelled the relationship between real delay and CAR magnitude, when the CAR is calculated erroneously by incorrectly assuming adherence to protocol time. Findings supported a curvilinear hypothesis in relation to effects of sample delay on the CAR. Short delays of 4–6 min between awakening and commencement of saliva sampling resulted in an overestimated CAR. Moderate delays of 7–15 min were associated with an underestimated CAR. Findings emphasize the need to employ electronic-monitoring of sampling accuracy when measuring the CAR in the domestic setting.

Introduction

Typically, awakening triggers a marked rise in cortisol secretion, normally peaking within 45 min post-awakening at around 30 min, followed by a normal diurnal decline. This rise has by convention been termed the ‘Cortisol Awakening Response’ (CAR). Since its discovery, researchers have become interested in exploring this phenomenon in relation to possible trait and state correlates, especially in the domains of cognition, affect, health and well-being (Ennis et al., 2016, Evans et al., 2012, Evans et al., 2007, Juster et al., 2011, Lovell et al., 2011, Stalder et al., 2010a, Stalder et al., 2010b, Stalder et al., 2009, Steptoe et al., 2007, Steptoe et al., 2008). Assessment of the CAR is typically in the domestic setting, with self-collection of saliva samples on awakening and at fixed intervals up to 45 min post-awakening. Given the brief time-window of the post-awakening cortisol rise, accurate sampling times relative to awakening are imperative for assessment of the CAR (Smyth et al., 2013a), an issue that has been highlighted in a recent expert consensus guidelines paper (Stalder et al., 2016). Awakening and sampling times are typically based on participants’ self-reports and only a small proportion (5.7%) of published studies conducted in the domestic setting (between 2013 and 2014) provided electronic-monitoring of sampling time relative to awakening. This is alarming given that delays between awakening and collection of saliva samples in the post-awakening period result in erroneous CAR measures (Broderick et al., 2004, DeSantis et al., 2010, Dockray et al., 2008, Golden et al., 2014, Griefahn and Robens, 2011, Kudielka et al., 2003, Kudielka et al., 2007, Kupper et al., 2005, Okun et al., 2010, Smyth et al., 2013a). There is also growing evidence that incorporation of uncorroborated data in computing CAR measures potentially influences findings and may partially explain some noted inconsistencies in the CAR literature (Smyth et al., 2015b).

When time derived from electronic monitoring does not match time defined by protocol, data are typically determined “inaccurately timed” and then excluded from CAR calculations (Ramachandran et al., 2016, Smyth et al., 2015b). Management of such inaccurate data in this way is costly, and could be avoided if it were possible to incorporate cortisol values outside of the fixed sample protocol times. Maximal use of hard-won data is not an insignificant issue and therefore it is vital to examine whether data obtained from such inaccurately timed samples are fully useable when analyzed in real-time. Such an approach was recommended in the expert consensus guidelines (Stalder et al., 2016). However, no study has conducted an analysis of post-awakening cortisol data in electronically verified real-time. Real-time analysis would also allow the cortisol growth curve to be plotted effectively using something closer to near continuous sampling. Little is known about cortisol levels between the common fixed sampling points of most study protocols. Areas under the curve are typically presented as illustrations, with lines joining adjacent time points, with an unspoken assumption of interpolated linearity. However, where an assumption of linearity has been investigated with repeated time-verified sampling in the first 15 min interval following awakening (Smyth et al., 2013a), linearity was not supported. Rather there was a brief latency period immediately after awakening and ending sometime between 5 and 10-min later when cortisol rise is clearly evident with growth curve modelling estimating the point of up-swing in cortisol starting at 8-min.

If delayed cortisol values of known timing, plotted in real-time, fit the normal cortisol growth curve, the corollary has to be that they would perforce give rise to distorted CAR values if plotted in protocol time, significantly reinforcing concerns about imprecise measurement. Existing electronic monitoring studies tentatively suggest a curvilinear relationship between delay and CAR magnitude. Short delays between awakening and collection of the first sample (Md = 7-min with a modal 28% of sample delayed by only 5-min) have yielded erroneously larger CARs if the first sample is wrongly assumed to be undelayed (Smyth et al., 2013a). Intensive 5 min sampling suggest this is a consequence of the ‘latency’ period which as we have just noted typically ends at or just after this time point relative to real awakening time (Smyth et al., 2013a, Smyth et al., 2015a). Thus the erroneous awakening value from which CAR rise is calculated is in reality delayed sufficiently to be just still in the latency period and likely therefore to be similar to what the hypothetical real awakening time sample would have been. However the short delay when carried forward to later samples will be associated with a higher average value in terms of their positioning under the real-time cortisol growth curve. The net effect is to yield erroneously higher CAR rise measures if protocol time is wrongly assumed. By contrast, a longer established evidence base has consistently found that longer delays are associated with smaller CARs (Dockray et al., 2008, Griefahn and Robens, 2011, Kupper et al., 2005, Okun et al., 2010), since the erroneously assumed base starting value, beyond the end of the brief latency period, will now most likely and rapidly become much higher than it should be and the potential for further CAR rise thus constrained.

In the present study, we present analyses of what we believe is the largest merged data-set yet assembled, where timing of awakening and collection of the first sample were verified using electronic-monitoring. Data were derived from studies of healthy young adults in our laboratory using the saliva collection protocol of an awakening sample and three further samples at 15 min intervals over a 45 min period. Findings from each contributing data-set, in respect of diverse aims, have already been published (Ramachandran et al., 2016, Smyth et al., 2013a, Smyth et al., 2015a), but their merger permits exploitation of cortisol data of known collection timing previously excluded as not sufficiently protocol accurate. Inclusion of such data permits analysis, for the first time, of cortisol data in verified real-time, to clarify whether the typical growth curve of post-awakening salivary cortisol is evident, regardless of sampling accuracy. Data will also provide the most accurate parameters yet published of the typical CAR period growth curve, using real-time data, not just at fixed 15 min time intervals. Such data, allows us to compute predicted values for post-awakening cortisol and CAR measures as reference values for a young healthy sample.

This merged database has also enabled investigation of the impact of curvilinear effects of sample timing accuracy on composite measures of the CAR commonly used in the literature: the simple cortisol rise from awakening to 30 min and average rise across the whole 45 min period. The curvilinear hypothesis is currently based on plausible inference from a limited number of small studies; it has not been tested in a single large data-set. Using data in this pooled data-set, we modelled the impact and effect size of short and moderate sampling delay on measures of the CAR, with the expectation of finding a significant curvilinear effect. Specifically, if adherence to fixed protocol times is wrongly assumed for delayed data, the following curvilinear relationship should pertain: CAR magnitude will increase as sampling delay increases from a minimal range of 0–3-min to short delays of between 4 and 7-min, followed by a rapid decline in CAR magnitude as delay increases further (from 8 to 15 min).

Section snippets

Database

Data originating from four previously published studies by our group were merged into a single database. Data were drawn from two studies presented in a single paper by Smyth et al. (2013a), from Smyth et al. (2015a) and Ramachandran et al. (2016). In total, data derived from 128 healthy participants, recruited from the academic community at the University of Westminster (median age 20 years and inter-quartile range 18–24 years). The study samples were predominantly female (N = 102, 79%),

Modelling the growth curve of cortisol in verified real time

The first analysis addressed the question: do verified known sampling times, regardless of accuracy in relation to desired protocol sampling times, result in the typical post-awakening cortisol growth curve when plotted against verified real-time from awakening? The obtained growth curve of cortisol over 60 min post-awakening revealed the usual significant linear (F = 267.857; df = 1, 730.831; p < 0.001) and quadratic components (F = 46.813; df = 1, 653.823; p < 0.001), underpinned by the typical steep rise

Discussion

For the first time, using a merged data-set of electronically-monitored awakening and collection of the awakening sample, we have shown that a typical post-awakening growth curve is evident when cortisol values are plotted against verified real-sampling times, regardless of protocol sampling time accuracy. This means that data, which are delayed by less than 15 min between awakening and collection of the first sample, do not need to be excluded from analyses if real-time and not fixed

Conclusion

The novel analyses reported here make an important additional contribution to the evidence reviewed in the recent Experts’ Guidelines on assessment of the CAR. Real-time analyses of all samples with less than 15 min delay between verified awakening and initiation of saliva sampling are presented. The expected cortisol curve is evident regardless of accuracy to the desired sample protocol timings, meaning that protocol inaccurate data need not be excluded from analyses. It needs emphasis that we

Acknowledgements

Studies providing data for the analyses presented here were made possible by funding from The Bial Foundation (grants 96/10 and 72/12) and the British Academy (grant SG132690).

References (30)

Cited by (18)

  • Evaluation and update of the expert consensus guidelines for the assessment of the cortisol awakening response (CAR)

    2022, Psychoneuroendocrinology
    Citation Excerpt :

    Beyond novel objective verification approaches, recent findings also inform about the best strategy for dealing with data objectively verified as inaccurately sampled (e.g., due to a delayed initiation of sampling). Smyth et al. (2016) revealed that when using objectively verified sampling times, actual cortisol data accurately mapped onto a typical post-awakening cortisol growth curve, even if respective sampling times deviated from scheduled protocol times. These data are thus in favor of the incorporation of verified inaccurate data (up to a delay of 15 min) into a CAR estimation model, rather than excluding these data.

  • The cortisol awakening response predicts a same-day index of executive function in healthy young adults

    2020, International Journal of Psychophysiology
    Citation Excerpt :

    Sampling delay on each day of CAR assessment was computed using Actiwatch and MEMS cap recordings. Data were treated as accurate where the difference between electronically measured awakening time and timing of the first sample was ≤8 min, as delays greater than this have been shown to result in under-estimated CARs (Smyth et al., 2016). Where delay was >8 but ≤15 min, these data were included in the analyses, and analysed in comparison to accurate data, as per the recommendations of Smyth et al. (2016).

View all citing articles on Scopus
View full text