Concordance of Chronotype Categorisations Based on Dim Light Melatonin Onset, the Morningness-Eveningness Questionnaire, and the Munich Chronotype Questionnaire

The body clock of each human individual is uniquely timed. Individual differences in body clock timing impact biological rhythms, resulting in a wide range of preferred and actual daily sleep and activity patterns. Chronotype is a construct that reflects individual differences underlying circadian timing [1,2]. Earlier chronotypes get up, achieve peak performance, and go to bed earlier than later chronotypes [3]. Mounting evidence suggests chronotype plays an important role in physical health (e.g., diabetes, metabolic disorders), mental health (e.g., depression, anxiety), and shift-work tolerance [4,5].

Chronobiology researchers have applied varied approaches to the operationalisation and analysis of chronotype for more than one hundred years. Chronotype operationalised as circadian phase or timing can be based on the timing of objective markers in biological variables that exhibit circadian rhythms (e.g., melatonin concentration, core body temperature). Dim light melatonin onset (DLMO) is the most common and reliable circadian phase marker and can be detected in periodic blood, saliva and urine samples [6]. However, collection of biological samples is invasive, laborious, and costly, limiting chronotype determination from objective measures to laboratory experiments with small samples [3]. Self-report questionnaires are simple to administer, and several have been developed to determine circadian preference or typical timing. Chronotype operationalised as preferred timing of sleep and activity is usually determined from the Morningness-Eveningness Questionnaire (MEQ) [7]. Chronotype operationalised as the phase of entrainment between the sleep-wake cycle and the 24-h clock can be based on sleep markers sourced from the Munich Chronotype Questionnaire (MCTQ) [8], which incorporates questions about typical times for sleep and activity.

Inconsistency in the approach to the operationalisation of chronotype has stimulated discussion in the literature [1,9]. Whilst chronotype based on objectively-measured biological rhythms is considered appropriate for examination of the impact of circadian timing on outcome variables (e.g., cognitive performance) [10], chronotype based on daily preference is appropriate for research on psychological traits, and chronotype based on the phase of entrainment is appropriate for research on circadian traits that are influenced by the entraining environment [9]. Several studies have assessed relationships between biological rhythms and MEQ/MCTQ measures, e.g., [11,12,13]. DLMO consistently correlates with both MEQ score and MSF_sc (corrected midpoint of sleep on free days derived from the MCTQ), however the strength of correlations varies between populations (e.g., DLMO-MEQ score: r = −0.40, DLMO-MSF_sc: r = 0.54 [14]; DLMO-MEQ score: r = −0.70, DLMO-MSF_sc: r = 0.68 [2]). Although measures of daily preferences and phase of entrainment may correlate, they are not comparable or interchangeable [9]. It follows that any research applying circadian typology should clearly justify the selection of the chronotype definition and measure. Furthermore, to simplify analyses and interpretations, distributions of continuous measures are often stratified into chronotype (e.g., early, intermediate, and late) categories. However, the approach to determining the cut-offs that separate categories is also inconsistent; most researchers apply percentile splits, e.g., [15,16], however, others use self-decided or recommended absolute cut-offs, e.g., [17,18].

Although DLMO, MEQ score and MSF_sc correlate, chronotype categorisations based on them do not necessarily concord. An individual may be categorised as early-type based on one measure, but as intermediate or late-type based on another measure. Low concordance between categorisations based on different chronotype measures and arbitrary cut-off scores limit meaningful comparisons of results between studies [1]. Some researchers have sought to mitigate this issue by categorising using a combination of subjective and objective measures, e.g., [19]. A weighted combination of multiple continuous circadian measures may provide more reliable chronotype categorisations, and several approaches have been suggested [20,21].

Surprisingly, there do not appear to be any studies that have systematically quantified concordance between chronotype classifications based on independent, subjective and objective circadian measures [22]. This study will close this gap by quantifying the concordance of chronotype categorisations based on an objective measure (DLMO) with categorisations based on two subjective measures (MEQ and MCTQ), for a relatively large sample of participants. As MSF_sc (from MCTQ) is a measure of behaviour influenced by circadian timing, we predict greater concordance between DLMO and MCTQ chronotype categorisations than between DLMO and MEQ chronotype categorisations.

For this sample of young, healthy adults, the continuous measures of DLMO, MEQ score and MSF_sc were normally distributed and correlated. Circadian questionnaires such as the MEQ and MCTQ are simple and suited to the field and large-scale studies, but their measures are not interchangeable with each other, or with DLMO. DLMO was more strongly correlated with MSF_sc than MEQ score, although neither correlation is strong. MSF_sc, a subjective measure of sleep timing, appears more closely related to endogenous circadian timing than the subjective preference for sleep timing measured by MEQ [2]. The correlation between DLMO and MEQ score (r = −0.25) is weaker than the same correlation reported for other studies (r = −0.35 to −0.76) [23]. The correlation between DLMO and MSF_sc (r = 0.32) may be weaker than the same correlation reported by others (r = 0.54 [14], r = 0.68 [2]) because our participants were mostly international travellers or students. Application of the MCTQ to individuals with irregular work arrangements may be problematic, as its calculations assume structured work schedules [24]. Furthermore, chronotype can only be determined from the MCTQ when an alarm clock is not used on free days [25], and therefore MEQ is sometimes used in preference to MCTQ simply to maximise available data, e.g., [26]. In the present study, MSF_sc could not be calculated for 14 participants, reducing our sample to 57. A recently developed short-form of MCTQ (µMCTQ) allows MSF_sc to be calculated for all respondents by capturing wake times only on free days on which an alarm clock is not used [27], and a similar modification to the standard MCTQ may enhance its utility.

For chronotype categorisations based on DLMO, there were no differences in mean MEQ scores between each category, and only a difference in mean MSF_sc between the early and late categories. This suggests that for this sample, MEQ scores and MSF_sc of participants in the early, intermediate, and late chronotype categories based on DLMO were indistinguishable. Chronotype categorisations based on splits of both the MEQ score distribution and the MSF_sc distribution had low concordance with chronotype categorisations based on a split of the DLMO distribution. Compared with categorisations based on DLMO, concordance for categorisations based on both MEQ score and MSF_sc was 37% (i.e., 63% error rate). However, categorisations based on the MEQ score had a greater proportion of 2-step errors (21%) than categorisations based on MSF_sc (14%). The low concordance may be a consequence of applying percentile splits to normally distributed continuous variables to create a categorical variable. This approach allows for the use of ANOVA statistical methods that simplify analyses and interpretation of differences between groups. A shortcoming of this methodology is that participants with measures near, but below or above, the percentile cut-offs can be assigned to different categories, although their measures are actually closer to each other than other members of their assigned category. Secondly, percentile splits of each study sample will yield different thresholds, limiting the generalisability of the results and comparisons with other studies. Although samples are sometimes dichotomised by a median split, e.g., [26], trichotomisation by tertile split, as in the present study, should provide superior between-group comparisons. The risk of drawing incorrect conclusions arising from categorisations based on continuous measures can usually be mitigated or overcome by regression and correlational analysis [28].

An important limitation of this study is that extreme chronotypes were excluded due to the selection criteria for the main shift work-study. However, our findings are generalisable to populations of younger, healthy adults, as epidemiological studies report that extreme chronotypes are uncommon [29]. Mean DLMO of our early and late DLMO chronotype categories (20:25 h and 23:03 h respectively) are comparable with criteria applied in other studies (e.g., early: <21:30 h, late >22.30 h [20]), indicating our sample included a broad range of chronotypes. We demonstrated greater concordance between categorisations based on DLMO and categorisations based on each of the subjective measures for the early and late categories than for the intermediate category, suggesting concordance may be greater with more extreme chronotypes. A second limitation to our study is that to illustrate the underlying measurement issue, we divided our sample into chronotype categories using tertile splits. Although this method is commonly used by chronotype researchers, categorisations based on other arbitrary or recommended percentiles or absolute cut-offs would likely yield different levels of categorisation concordance.

Valid measurement is a key element of scientific research. Our results suggest that laboratory experimental research on the differential effects of body clock timing on outcome variables should use objective biological measures where possible. DLMO, MEQ and MCTQ measure related, but different constructs that are not interchangeable. Although these measures were correlated for this sample, the correlations were not strong. Consequently, chronotype categorisations based on percentile splits of these distributions showed low concordance. To enable valid comparisons with previous results and reduce the likelihood of misleading conclusions, chronotype researchers should select the most appropriate measures and statistical techniques for their construct of interest and research question.