Nonlinear associations between insomnia symptoms and circadian preferences in the general population: symptom-specific and lifespan differences in men and women

OBJECTIVES
This study investigated (non)linear associations between different eveningness characteristics (bedtime, wake time, morning affect, and peak performance time) and insomnia symptoms (difficulties initiating sleep, difficulties maintaining sleep, and nonrestorative sleep) in a large general population sample.


METHODS
The data came from digital surveys about insomnia (Minimal Insomnia Scale) and circadian preferences (Children's Chronotype Questionnaire/Composite Scale of Morningness) completed by the Dutch general population (37,389 participants aged 4-91years, 42.4% men) in the Lifelines cohort substudy Comorbid Conditions of ADHD.


RESULTS
Using generalized additive modeling, we found that different characteristics of eveningness related to insomnia either exponentially (later wake time/peak performance time, worse morning affect) or quadratically (early and late bedtime/midpoint of sleep). While difficulties initiating sleep and nonrestorative sleep were strongly associated with all eveningness characteristics, difficulties maintaining sleep related only to earlier bedtimes. These relationships were similar for men and women but varied partly in shapes and strengths across the lifespan. Additional analyses showed that bedtime and wake time were associated with insomnia symptoms only when their combination would result in an unusually long or short preferred time in bed.


CONCLUSION
The association between eveningness and insomnia symptoms highly depends on whether eveningness is reflected by daytime performance or sleep-wake time. The pattern and strength of these associations also vary depending on age and insomnia symptom, but less so on sex. Future sleep-related research and policies relying on circadian preferences should account for the nonlinearity, dimension/symptom-related specificity and age-related differences in the association between eveningness and insomnia symptoms.

The values in the cells show means and standard deviations in parentheses for continuous variables, or a number of observations and the proportion in the whole (sub)sample in parentheses for categorical variables.Abbreviations: N/A = not applicable * Several exact values have been concealed (replaced with "-") due to privacy considerations.The standard Lifelines export rule requires a group size of n≥10 to prevent traceability to participants.Overall, we fit a set of models for each combination of the insomnia symptoms (outcome) and eveningness dimensions (predictor) by adding linear and nonlinear terms.Each following model was compared to a previous model with the best fit.If the model fit was not significantly better after adding a new term, this term was removed; for example, if the separate smooths for men and women did not improve the model fit, the smooth was not separated by sex.The order of the comparisons was as follows:      The values in the cells show the effective degrees of freedom (edf), which indicate the degree of nonlinearity, and the corresponding p-value as a superscript, which indicates statistical significance.In the case when the effect is not statistically significant (p>0.05), the exact p-value is indicated, and when significant, the asterisks correspond to p<0.05 (*), p<0.01 (**), or p<0.001 (***).If the nonlinear effect was estimated separately for men and women, the value on the left corresponds to men, and on the right, to women, separated with a slash.

Section A1. The rationale behind the choice of the particular dimensional structure of the circadian preference questionnaires
In our study, we measured circadian preferences on the spectrum of morningness-eveningness (M/E).We used the M/E scale of the parent-report Children's Chronotype Questionnaire (CCTQ) in children and adolescents, and the self-report Composite Scale of Morningness (CSM) in adults.[1,2] Although the instruments measuring M/E in our studies were developed as unidimensional, the accumulating evidence from the last decades of research supports their multidimensionality.Herewith, we provide the detailed literature overview and our rationale behind choosing a specific multidimensional structure of the CCTQ's M/E scale and CSM.
The CSM was developed by uniting items from two other M/E scales, Horne and Ostberg's Morningness-Eveningness Questionnaire (MEQ) and Torsvall and Akerstedt's 7-item Scale.[3,4] Although the items of the CSM were chosen based on factor analyses of the aforementioned questionnaires, the dimensional structure of the CSM itself was not investigated in the original study.[2] Multiple later studies showed that questions in the CSM can be divided into distinct dimensions, but the number and content of these dimensions varied between studies.[5][6][7][8][9][10][11][12][13][14][15] Despite these inconsistencies, some of the resulting dimensions generally reflected similar constructs.Two-factor solutions yielded dimensions related to morning affect/alertness (items 4, 5, 12 and sometimes 3,6,11) and activity planning/general typology (items 1-2, 7-10, 13 and sometimes 3,6,11).[5][6][7] Three-factor solutions separated some items from activity planning/general typology to a separate dimension, either bedtime (items 2, 7, 9, 13 and sometimes 1, 8) or wake time (items 1, 6, 10, 11 and sometimes 3, 8).[8][9][10][11][12][13] Furthermore, another study found a four-factor structure that includes all the aforementioned dimensions of morning affect, activity planning, wake time and bedtime.[15] Morning affect was the only dimension that was relatively stable across cultures/countries and age groups.[8,11] The other two dimensions usually comprised a combination of items that related to bedtime/wake time and activity planning/general typology.General typology items 9 and 13 assessed overall M/E and inconsistently loaded on either the dimension of bedtime or wake time, or even comprised their own dimension, depending on the study sample.[8] Activity planning usually included items 6 and 8 with questions about the time of best cognitive and physical performance.Although the question about the time of best cognitive performance seemed to clearly belong to its own dimension, a similar item about physical performance sometimes loaded on the factors of wake time and morning affect.[7][8][9][10][11][12][13] After a careful review, we concluded that activity planning was not defined as its own dimension only when factor solutions could not separate the dimensions of bedtime and wake time.Inconsistencies in the dimensional structure may have resulted from differences in study samples and methodology; for example, use of principal component versus factor analysis.[16] Furthermore, the irregular loading patterns partially resulted from questions that were formulated assuming one process behind M/E, hence not phrased in a way to reflect a specific dimension.This non-specificity of the items can explain the absence of strong dimensionality in the CSM.The study that allowed for such coexistence of a global construct with meaningful specificities was confirmed that the general typology items (9 and 13), indeed, reflect the global M/E, which can be specified into the factors of morning affect and wake time, whereas bedtime-related items (2 and 7) reflect a specific factor separate from the global construct.[9] Therefore, research has provided enough evidence for the multidimensionality of CSM, but the exact structure and interpretation of the dimensions is still in dispute.
The dimensional structure of CCTQ was never reported in the original publication, and to our knowledge, in any of the subsequent studies that used CCTQ.
However, CCTQ was developed by adapting two self-report M/E questionnaires to parent-report.[1] Questions measuring sleep-wake time were derived from the Munich Chronotype Questionnaire (MCTQ) and the M/E scale from the Morningness-Eveningness Scale for Children (MESC).[17,18] Although the authors of MESC did not mention the exact source of the survey's items, it appears that these items comprise questions from the CSM adapted for better understanding by children.In our study, we used the M/E scale of the CCTQ whose questions largely overlap with the CSM.The only major deviation is that MESC and CCTQ lack three questions asked in CSM (items 5, 10 and 13).
Another difference is that item 13 of the CSM, although present in the CCTQ, is not included in the M/E scale sumscore of the CCTQ.Therefore, the M/E scale of the CCTQ appears to be a reduced parent-report version of the CSM.
Although the exact dimensions in both MEQ and CSM remain undetermined, these dimensions are important for the interpretation of total M/E scores and their relationship to other health outcomes.
Therefore, our study defined the M/E dimensions in CSM and CCTQ based on the consistencies among the factor structures from earlier studies on CSM so that these dimensions aligned as much as possible with the general established operationalizations.We used own judgement in interpreting the meaning of the items' questions to classify the items which had inconsistent factor loadings in prior research.
Although the response scale for the same questions in CCTQ sometimes deviates from that in the CSM and some questions from CSM are absent from CCTQ, we think the dimension scores of both questionnaires should mostly reflect the same constructs and, thus, will be comparable in our sample.
Even though three-factor models varied in which dimensions they include, the item content in these dimensions was very similar to the content of the dimensions in the four-factor solutionmorning affect, bedtime, wake time, activity planning/general typology.[8][9][10][11][12][13]15] Our confirmatory factor analyses resulted in good fit of the four-factor structure of the CSM as shown by the fit indices (see the detailed output in Table S8 and Figure S4).Notably, the dimensions wake time, morning affect and peak performance time were not sufficiently distinct from as indicated by their high correlation.
Although the factors of the CCTQ's M-E scale were less differentiated than in the CSM, we decided to use them to allow comparison across the lifespan.
We decided that four factors propose a refined dimensional structure of M/E and would be the most informative for our research purposes.The first dimension of "bedtime" (items 20, 23, 25 from CCTQ and 2, 7 from CSM) reflected the preferred time of going to bed.The second dimension of "wake time" (items 19, 24 from CCTQ and 1, 10, 11 from CSM) reflected the preferred time of rising in the morning.The third dimension of "morning affect" (items 17, 18, 26 from CCTQ and 3-5, 12 from CSM) reflected affective state and alertness after sleep.The fourth dimension of "peak performance time" (items 21, 22 from CCTQ and 6, 8 from CSM) reflected the best time of cognitive/physical functioning during the day.For the best specificity of these dimensions, we did not use the broad general typology items (9, 13 from CSM and 27 from CCTQ).The exact numbers and questions of the items that belong to the specific dimensions of the CCTQ and CSM are summarized in the Table S7.

Model 1 :Figure A1 .
Figure A1.Perspective (3D) plots of the differences in the severity of total and individual insomnia symptoms depending on age and A) eveningness, or B-E) its separate dimensions.Insomnia severity is lower than average in the green area and higher in the orange area.The missing areas show the combinations of variables that are not represented in the data and, therefore, might not yield good predictions in the model (at 10% extrapolation).
Figure A2.Visualization of the follow-up analyses: interaction between bedtime and wake time in their effect on insomnia symptoms.Insomnia severity is lower than average in the green area and higher in the orange area.The missing areas show the combinations of variables that are not represented in the data and, therefore, might not yield good predictions in the model (at 10% extrapolation).
Figure A3.Visualization of the follow-up analyses: Estimated severity of total and individual insomnia symptoms (colored lines) depending on midpoint of sleep (higher values corresponds to later midpoint) and preferred time in bed (higher values correspond to longer time in bed).Shaded areas over the colored lines indicate 95% confidence intervals.

Figure A4 .
Figure A4.Differences in the associations between circadian preferences and insomnia symptoms in men and women aged 50-85 with different employment status.The colored lines show the estimated severity of total or individual insomnia symptoms depending on A) bedtime, B) wake time, C) morning affect, and D) peak performance time, from 0 (extremely early preferences) to 4 (extremely late preferences).Shaded areas over the coloured lines indicate 95% confidence intervals.

Figure A5 .
Figure A5.Dimensional relations of the Composite Scale of Morningness (left figure) and the morningness-eveningness scale of the Children's Chronotype Questionnaire (right figure) The circles represent each of the four dimensions of the CSM (on the left) or the M/E scale of CCTQ (on the right) and the squares correspond to the items belonging to them.Green lines and corresponding values indicate the correlations between the attached components.Abbreviations: bd_ = bedtime; wk_ = wake time; mr_ = morning affect; pk_ = peak performance time

Table A2 .
Demographic and sleep-related characteristics of the sample per age-sex subgroup

Table A3 .
Comparison of the model fit based on maximum likelihood The values in the table are taken from the final models that we chose based on the model fit test based on maximum likelihood comparison (compareML function the package itsadug).Generally, a lower value indicates a better fit.Maximum likelihoods can only be compared within certain families of tests (for nested models, such as comparing models nonrestorative sleep ~ age + sex vs. nonrestorative sleep ~ age + sex + bedtime, but not nonrestorative sleep ~ sex + age vs. difficulties initiating sleep ~ age + sex).

Table A4 .
Follow-up analyses: generalized additive models for the effects of bedtime, wake time, and the interaction between bedtime and wake time on insomnia and its individual symptoms

Table A5 .
Results of the follow-up analyses: generalized additive models for the interaction effect of midpoint of sleep and time in bed on insomnia symptoms

Table A6 .
Results of the follow-up analyses: generalized additive models for the associations between circadian preferences and insomnia symptoms in the sample stratified by employment status

Table A7 .
The dimensional structure of the Composite Scale of Morningness (CSM) and the morningness-eveningness scale of the Children's Chronotype assume that your child has to be at peak performance for a test that will be mentally exhausting for 2 hours.Considering your child's "feeling best" rhythm and that you are entirely free to plan your child's day, which ONE of the three time intervals would you choose for the test?8 You wish to be at your peak performance for a test which you know is going to be mentally exhausting and lasting for two hours.You are entirely free to plan your day, and considering only your own "feeling best" rhythm, which ONE of the four testing times would you choose?22 Let's assume that you have decided to enroll your child in an athletic activity (e.g., swimming).The only class available meets twice a week at 7 to 8 am.How do you think he/she will perform?6 You have decided to engage in some physical exercise.A friend suggests that you do this one hour twice a week and the best time for him is 7:00-8:00 a.m.Bearing in mind nothing else but your own "feeling best" rhythm, how do you think you would perform?

Table A8 .
Results of the confirmatory factor analyses testing the four-factor model