Self-reported sleep quality is more closely associated with mental and physical health than chronotype and sleep duration in young adults: A multi-instrument analysis

Sleep and circadian rhythms are considered to be important determinants of men tal and physical health. Epidemiological studies have established the contribution of self-reported sleep duration, sleep quality and chronotype to health outcomes. Mental health and sleep problems are more common in women and men are more likely to be evening types. Few studies have compared the relative strength of these contributions and few studies have assessed these contributions separately in men and women. Furthermore,


| INTRODUC TI ON
Sleep problems are common in people with mental health disorders (Freeman et al., 2017). While disrupted sleep was previously thought to be a consequence of mental health disorders, sleep problems are now being increasingly recognised as an important element of the complex and multi-factorial causation of the symptoms and functional disability associated with psychiatric disorders (Harvey, Murray, Chandler, & Soehner, 2011).
Sleep quality is often assessed by the total score of the PSQI, which is one of the most widely used questionnaires. The PSQI consists of 24 items and 7 domains and the single sleep quality question of the PSQI has also been used to assess subjective sleep quality.
The KSD measures sleep quality of the previous night by one question. Another measure of sleep quality is the Insomnia Severity Index (ISI) primarily used in insomnia research but now also used outside insomnia research (Bastien, Vallieres, & Morin, 2001;Lazar et al., 2013).
In recent years there has been increasing interest in the contribution of circadian disturbances and individual differences in circadian rhythmicity to mental and physical health. Circadian processes may contribute to health through their role in sleep regulation or through the separate wide range of biological and behavioural processes that are influenced by circadian rhythmicity (Logan & McClung, 2019).
The effect of chronotype on mental health and well-being has been suggested to be due to the impact of an evening-type preference on sleep quality and sleep duration. Evening-type adults commonly complain of decreased subjective sleep quality, insufficient sleep, excessive day time sleepiness and trouble initiating sleep (Kivela, Papadopoulos, & Antypa, 2018). Evening chronotype, low sleep quality and excessive daytime sleepiness have been shown to independently predict common mental disorders (Rose et al., 2015).
Eveningness has been associated with increased social jet leg (Roenneberg, Pilz, Zerbini, & Winnebeck, 2019). Social jetlag refers to a misalignment between biological time and sleep timing imposed by social schedules such as school and work times. Social jetlag may contribute to emerging mental health difficulties especially in adolescents and young adults (Doi, Ishihara, & Uchiyama, 2015). However, the association of social jetlag with mental health in young people is equivocal according to a recent systematic review (Henderson, Brady, & Robertson, 2019).
Despite the apparently overlapping effects of sleep and chronotype the independent contribution of sleep duration, sleep quality, chronotype and social jet lag on physical and mental health has rarely been assessed in surveys in which all three aspects were covered simultaneously and with multiple validated instruments.
Furthermore in most studies reporting on associations between sleep and health outcomes the results are controlled for sex effects. Few studies have investigated sex specific associations between chronotype, sleep duration and sleep quality and mental health. Women have a greater risk of developing mental health conditions while a greater proportion of evening-types are men (Antypa, Vogelzangs, Meesters, Schoevers, & Penninx, 2016).
Although there are other factors associated with mental health to consider, from a sleep perspective, this presents a discrepancy in the link between mental health and the evening chronotype.
Sex differences in the association between sleep and health deserve further exploration as women sleep longer than men and epidemiological studies suggest that poor sleep and sleep-related problems are more strongly associated with poor health outcomes in women than in men (Lauderdale et al., 2006;Suarez, 2008).
Women who report 'unhealthy' sleep have greater psychological distress and risk of cardiovascular disease, type 2 diabetes, depression and mood disorders (Suarez, 2008). Additionally, daytime sleepiness and poor sleep quality is higher among women, and female predominance in the rate of depression was observed in subjects with delayed sleep-wake schedule (Fabbian et al., 2016).
These combined issues are addressed in the study reported below.

| Aims and Significance
The major aim of this study was to establish whether chronotype, sleep quality and sleep duration, assessed with a number of frequently used instruments, are independent predictors of physical and mental health (e.g. affect, wellbeing), lifestyle (e.g. diet), and if so, which is the strongest predictor. The second aim was to investigate the independent associations of chronotype, sleep quality and sleep duration with stable trait-like psychological characteristics. A third aim of this study was to assess how sex modulates the associations between chronotype, sleep quality, sleep duration and physical and mental health and psychological characteristics.

| Participants
The study was reviewed and approved by the University of Surrey Ethics Committee and conducted in accordance with the principles of the Declaration of Helsinki. Participants were recruited using posters, advertisements in local newspapers, by radio, and through Web sites. Following a 10 min long telephone interview to check for basic eligibility criteria 677 participants attended the face-to-face screening session at Surrey Clinical Research Centre (SCRC) and 675 participants completed the session, which lasted for approximately two hours and included 15 questionnaires and two paper based tests. Approximately 99% of the participants completed all questionnaires. In total 671 participants were included in the current analyses. For some questionnaires with derived outcome measures (e.g. global sleep quality measured by PSQI) the number of observations included in the analyses was lower because some questions were not answered. Participants were healthy, non-smokers, aged between 20 and 35 years. We only studied young adults to minimise potential confounding effects of age. The participants did not do any shift-work during year preceding the study and had a BMI between 18 and 30. There was no exclusion criteria related to ethnicity or country of origin.
The data were collected as part of the screening for a sleep-circadian experiment and participants were reimbursed for reasonable out of pocket expenses, e.g. bus, train fares, etc and paid 10 pounds for their time. For additional details, please see the supplementary material and our previous report on this data set .

| Health and sleep measures
Chronotype, habitual sleep-wake timing and quality of sleep, as well as psychological and self-reported health characteristics were assessed by multiple questionnaires and scales (see Supplementary material for more details).
Demographic data, such as age, sex, BMI, income, ethnicity, marital status, etc., were assessed by a medical questionnaire (MQ) and the BSS.
Diurnal preference (chronotype) measures were taken from the total score over all the 19 items of the MEQ (Horne & Ostberg, 1976), the midpoint of sleep during free days (MSF) and the self-assessment questions referring to being early or late type using the 7-point Likert scale (MCTQ Iam ) from the MCTQ questionnaire.
Sleep-wake timing and duration were assessed using multiple questionnaires that differed from each other with respect to the time period to which they refer (MQ -in general, PSQI -last month, BSS -last week, and MCTQ -usual workdays and free days). PSQI and BSS were used to provide a direct measure of self-reported sleep duration and the MCTQ was used for the assessment of both  (Roenneberg et al., 2004).
Quality of sleep was assessed by the ISI total score across all the seven items (Bastien et al., 2001), the KSD single sleep quality question measured on 9-point Likert scale ('How would you rate your quality of sleep?' -referring to the last night (Akerstedt et al., 1994), the PSQI global score across all the 24 items and the single sleep quality question within the PSQI measured on a 4-point Likert scale ('During the past month, how would you rate your sleep quality overall?') (Buysse et al., 1989).

Social jetlag was computed as the absolute difference between
the uncorrected midpoint of sleep during free-days and workdays.
For further details on the sleep-timing outcome measures please refer to the supplementary material.
Health profile was assessed using Version 2 of the 36-Item Short-Form Health Survey (SF-36v2) which provides a separate measure for physical and mental health (Ware & Sherbourne, 1992). The SF-36 has been previously used to investigate sleep and health associations in healthy young adults (Gulec et al., 2013). The General Health Questionnaire (GHQ) provides a composite total score as a reliable measure for general psychiatric health (Goldberg, 1978). In this way mental health was directly measured by two different questionnaires the SF36 and the GHQ, but labelled as mental health and general psychiatric health, respectively. BMI was also included as a proxy measure of physical health.
Eating behaviour was assessed using the English version of the Dutch Eating Behaviour Questionnaire (DEBQ) (Van Strien, Frijters,

Individual traits of general motivation systems were assessed by
the Behavioural Inhibition System-Behavioural Activation System (BIS-BAS) scales (Carver & White, 1994).
General affective orientation reported for the last couple of weeks was measured by the Positive Affect and Negative Affect Schedule (PANAS) (Watson, Clark, & Tellegen, 1988).

| Statistical analysis
Statistical analyses were conducted using SPSS (version 25) and SAS (version 9.4).
In a first step for assessment of sex differences amongst demographic, sleep and health measures simple t-tests were used (Table 1). For measures that were grouped categorical data, a chisquare test was used.
In a second step regression models were applied for the assessment of the independent associations of chronotype, sleep quality and sleep duration on health and psychological wellbeing measures.
Analyses were performed with and without control for sex, age, ethnicity, work status, gross income, marital status and alcohol consumption (Table 2). Multicollinearity was assessed by examining tolerance and the Variance Inflation Factor (VIF) ( Table S1). For more details on the multivariate regression models please refer to the supplementary material.
The independent effect of chronotype, sleep duration and sleep quality on physical and mental health and psychological wellbeing was estimated by Local effect size calculations of Cohen's f 2 . This quantifies the proportion of variance explained by adding a sleep or chronotype predictor to the model with confounders alone (Selya, Rose, Dierker, Hedeker, & Mermelstein, 2012).
In order to further explore sex differences in the association between chronotype, sleep quality, sleep duration and the main three health outcome measure we computed Pearson correlations separately for men and women and compared the correlation coefficients using Fisher's transformation. Additionally, the local effect size for the associations of chronotype, sleep duration and the four different sleep quality measures with general psychiatric, mental and physical health for each sex group was calculated. Finally, we also assessed the independent contribution of social jetlag to measures of health and psychological well-being.
In order to address multiplicity we estimated false discovery rate analysis using the Benjamini-Hochberg correction (Benjamini & Hochberg, 1995). The False discovery rate was set at 0.05 and significance was established if the q values were smaller than 0.05.
For the regression analyses we did not apply FDR correction. This is because the p values associated with the individual predictors cannot be considered independent within the same multivariate model.

| Sex differences across demographic, sleep and health measures
Demographic data did not show any significant differences between men and women, except for alcohol intake which indicated that men drink significantly more (~1.4 unit more) per week than women (Table 1).

| Sleep measures
Women had significantly earlier bedtime and sleep onset time during both workdays and free-days, as measured by the MCTQ. Women also reported a significantly earlier midpoint of sleep during workdays (MSW) than men, but this effect was present only as a trend during free-days. Women spent more time in bed throughout the week and mainly during free-days, and they also reported an overall worse sleep quality as measured by KSD single sleep quality question compared to men. Nonetheless, these sex effects on time in bed during free-days and sleep quality lost significance after statistical correction for multiplicity (Table 1).

| Health/Psychological/Personality measures
Women had significantly poorer mental health and lower BMI scores. Women were significantly more likely to eat in response to emotional challenges and required significantly greater effort to refrain from eating. Women were significantly more Conscientious, Agreeable and inclined towards Neuroticism. Men were significantly more driven in anticipation of rewards and to avoid punishment.

| The effect of chronotype, sleep quality and sleep duration associations on health and psychological measures
In the second step, we analysed the associations between chronotype, sleep quality and sleep duration with health and psychological measures. This was repeated with covariates of sex, age, ethnicity, work status, gross income, marital status and alcohol consumption included in the model (Table 2).
Chronotype and sleep quality were found to predict inde-  According to Cohen's (1988) (Figure 1 and Table S2a). Chronotype is most strongly associated with general health whereas sleep duration associates most strongly with physical health (Figure 1 and Table 2).
To clarify the extent to which the results might be dependent on the individual questionnaires we ran the same analyses using

| Sex differences in the effect of chronotype, sleep quality and sleep duration on health measures
In the following step we aimed to clarify the extent to which the predictive power of chronotype, sleep quality, sleep duration on health may be different for men and women. We first ran a model in which we used all the four sleep quality measures as presented in Figure 1, but this time separately for men and women. Sleep quality was the strongest predictor of health across men and women, with significant effects for each sex group, however, effect size was considerably larger for women than for men, across all sleep quality measures ( Figure S1 and Tables S4a-d). Mental health was also independently predicted by chronotype and significantly so for both men and women but with a stronger effect size in women.
However, the effect of chronotype on general psychiatric health was significant in men only with a stronger effect as compared to women Comparing the correlations coefficients between chronotype, sleep quality, sleep duration and the three main health outcome measures showed that the relationship between sleep quality and mental health was significantly stronger in women than in men (Table S5).

| The independent effect of social jet lag on measures of health and psychological characteristics
In a final step we conducted a non-exhaustive analysis of the effect of  (Table S6)

| D ISCUSS I ON
Our results based on numerous validated questionnaires and measures of chronotype, sleep quality and sleep duration show unequivocally that self-reported sleep quality is the strongest predictor of mental and physical health and more so in women than in men.
Our results for sleep quality and mental health are in general in accordance with the depression and anxiety literature (Baglioni et al., 2011;Franzen & Buysse, 2008;Tafoya et al., 2019). However, in contrast to previous studies, our study did not show many effects of sleep duration on mental health (Table 2 and Tables S2a-c) and the local effect size of sleep duration was much smaller compared to the effect size of sleep quality and chronotype. Previous studies have reported associations between sleep duration and mental health (Baum et al., 2014;Roberts & Duong, 2017;Zhai, Zhang, & Zhang, 2015). It may be that in previous studies the association between sleep duration and mental health was mediated by effects on sleep quality and partly by chronotype.
This study found numerous significant associations between sleep quality and proxies of mental health. Positive affect and negative affect were shown to be associated with sleep quality in accordance with a previous study (Hoag, Tennen, Stevens, Coman, & Wu, 2016). Markarian, Pickett, Deveson, & Kanona, (2013) suggested that BIS sensitivity may be related to poor sleep quality, which is in support of the findings of the current study. Markarian and colleagues also reported that high BIS sensitivity is related to neuroticism which is the personality factor that had the strongest association with sleep quality in our study. Weaker but significant associations were found between Conscientiousness, Extraversion, Agreeableness and sleep quality. These associations between personality and sleep quality have been previously reported in the literature (Stephan, Sutin, Bayard, Krizan, & Terracciano, 2018). All of these personality factors have been shown to affect subjective well-being, which is a significant topic in mental health research.
Sleep quality has recently been identified as the mediator of this relationship (Lai, 2018).
Previous studies have demonstrated associations between both diurnal preference and chronotype with mental and physical health conditions (Knutson & von Schantz, 2018;Merikanto et al., 2015;Yu et al., 2015). Diurnal preference as measured by the MEQ correlates with chronotype as assessed by reported sleep timing. Diurnal preference correlates sleep-wake timing, the phase of the melatonin rhythm and with circadian period (Hasan et al., 2012;Lazar et al., 2013 Cohen's (1988) guidelines f 2 ≥ 0.02, f 2 ≥ 0.15, and f 2 ≥ 0.35 represent small, medium, and large effect sizes, respectively that participants were not sleeping and waking at preferred times.
Not sleeping at preferred timings can impact sleep quality, particularly for evening-type individuals (Gangwar et al., 2018) and this may explain the association between diurnal preference and not chronotype, with mental health. We did not find a major contribution of social jetlag to any of our outcome measures and this is in accordance with a recent systematic review (Henderson et al., 2019).
Morning preference was associated with stronger presence of favourable (Conscientiousness, Extraversion, Agreeableness) and weaker presence of unfavourable personality traits for mental health (neuroticism), which is in accordance with previous reports (Randler, Baumann, & Horzum, 2014).
In our study physical health was only associated with sleep quality. This is in disagreement with previous studies which found that physical health, which included increased risk of cardiovascular disease (Grandner, 2017;Merikanto et al., 2013), diabetes and metabolic syndrome (Yu et al., 2015), as well as higher caloric intake, BMI and rates of obesity (Arora & Taheri, 2015;Knutson, 2012) (Druiven et al., 2019) reports that a later chronotype does not predict a persistent course of depressive or anxiety disorders.
Furthermore the association between chronotype and health factors could be mediated by the impact of a later chronotype on other sleep factors

| Gender differences in the associations between sleep and health factors
Sleep quality associations were stronger in women while chronotype associations with general psychiatric health were only significant for men and chronotype associations with mental health were significant predominantly in women. There was also a significantly stronger correlation between chronotype and mental health in women compared to men. These results are in line with the literature. Eveningness was more significantly associated with impulsivity and anger, depression, anxiety disorders and low mood in women (Fabbian et al., 2016).
Our finding that sleep quality had a significantly stronger association with mental health in women than in men is also in accordance with the literature. Suarez (Suarez, 2008) reported that poor sleep quality and prolonged sleep latency incurred a greater psychological and physiological toll on women relative to men.
In our study, not all measures of sleep quality showed these relationships equally clearly. The ISI and the PSQI global score have shown a considerably stronger association with mental health in women as compared to men. These sex differences were much smaller when sleep quality was measured by the PSQI individual question and the KSD.

| Limitations
Our analyses were based on questionnaire data. Subjective methods have a sensitivity between 73% and 97.7%, while their specificity ranges from 50%-96% (Ibanez, Silva, & Cauli, 2018). As sleep quality is a subjective measure, it is justified to assess it using a questionnaire (Fatima, Doi, & Mamun, 2016). Furthermore, Bei et al. (Bei, Wiley, Allen, & Trinder, 2015) found that subjective sleep quality mediated the relationship between objective sleep and negative mood, consequently highlighting the importance of subjective sleep perception in the development of sleep related mood problems. This suggests that conclusions derived from sleep quality measures are valid.
Nevertheless, the self-report of sleep quality may be influenced by the participants' mood and in that sense may be confounded (Krystal & Edinger, 2008).
The restricted age range of the participants in this sample is both a limitation and a strength; a strength because it helps to eliminate many confounding factors that are associated with age; a limitation because it prevents extrapolating our conclusions to the older population.
Another limitation is that the study population included more males than females. This was due to a differential response during the laboratory study recruitment phase, which generally attracts more men. However, the observed gender differences warrant further research into this area with larger samples sizes of equal proportions of men and women.
Finally the "work-start time" may interact with chronotype in modulating sleep duration and quality and therefore could have been included as a predictor in the multivariate models. In the current analyses, however, we were interested in the predictive value of direct measures of self-reported sleep quality, duration and chronotype and it was not our aim to analyse how other factors such as work-start time, may interact with these direct measures.

| CON CLUS I ON AND OUTLOOK
This study found that sleep quality is the strongest independent predictor of mental and physical health outcomes in young healthy adults, with significant effect for both sex groups but stronger association seen in women than in men. Therefore, the results of this study imply that sleep quality should be used in the assessment and treatment of mental health conditions in clinical settings, particularly for women. Further research is needed to understand the relationship between subjective sleep quality and objective sleep measures (e.g. Della Monica, Johnsen, Atzori, Groeger, & Dijk, 2018) and identify other determinants of subjective sleep quality.

Data collection was supported by BBSRC grant BB/F022883
and Air Force Office of Scientific Research grant FA9550-08-1-

(DJD et al.) DJD is supported by the UK Dementia Research
Institute. ASL is supported by a seed award in science from the Wellcome trust (207799/Z/17/Z). We thank Drs Nayantara Santhi, June Lo, Ana Slak and Sibah Hasan for their help with data acquisition.

AUTH O R CO NTR I B UTI O N S
JAG and D-JD designed research; D-JD directed the research; ASL performed research; KM and ASL analyzed data; and KM, ASL, JAG, and D-JD wrote the paper.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available on request from the authors.