Determinants of perceived sleep quality in normal sleepers

ABSTRACT

Objective: This study aimed to establish the determinants of perceived sleep quality over a longer period of time, taking into account the separate contributions of actigraphy-based sleep measures and self-reported sleep indices. Methods: Fifty participants (52 ± 6.6 years; 27 females) completed two consecutive weeks of home monitoring, during which they kept a sleep–wake diary while their sleep was monitored using a wrist-worn actigraph. The diary included questions on perceived sleep quality, sleep–wake information, and additional factors such as well-being and stress. The data were analyzed using multilevel models to compare a model that included only actigraphy-based sleep measures (model Acti) to a model that included only self-reported sleep measures to explain perceived sleep quality (model Self). In addition, a model based on the self-reported sleep measures and extended with nonsleep-related factors was analyzed to find the most significant determinants of perceived sleep quality (model Extended). Results: Self-reported sleep measures (model Self) explained 61% of the total variance, while actigraphy-based sleep measures (model Acti) only accounted for 41% of the perceived sleep quality. The main predictors in the self-reported model were number of awakenings during the night, sleep onset latency, and wake time after sleep onset. In the extended model, the number of awakenings during the night and total sleep time of the previous night were the strongest determinants of perceived sleep quality, with 64% of the variance explained. Conclusion: In our cohort, perceived sleep quality was mainly determined by self-reported sleep measures and less by actigraphy-based sleep indices. These data further stress the importance of taking multiple nights into account when trying to understand perceived sleep quality.

A good night’s sleep is important for our overall health. However, it is still not well-defined what a “good night’s sleep” actually entails, in the perception of the sleeper. Currently, there is no broadly accepted definition of perceived sleep quality. Perceived sleep quality can vary from person to person; one subject may link it to not waking up at night, while another may interpret a short sleep latency as good-quality sleep. Moreover, these connotations may even differ for a single person across nights. Besides self-reported sleep measures, perceived sleep quality can be expressed based on more objective measures such as polysomnography or actigraphy-based methods.

Most earlier studies that examined perceived sleep quality are cross-sectional in nature (Åkerstedt, Hume, Minors, & Waterhouse, 1997; Argyropoulos et al., 2003; Keklund & Åkerstedt, 1997; O’Donnell et al., 2009; Riedel & Lichstein, 1998; Rosipal, Lewandowski, & Dorffner, 2013; Westerlund, Lagerros, Kecklund, Axelsson, & Åkerstedt, 2016). The observed associations between objective sleep measures and perceived sleep quality were only low to medium and results were not consistent across studies. Some found that slow-wave sleep was linked with perceived sleep quality (Åkerstedt et al., 1997; Armitage, Trivedi, Hoffmann, & Rush, 1997; Keklund & Åkerstedt, 1997), while others observed a correlation between wake time after sleep onset and perceived sleep quality (Argyropoulos et.al., 2003; Keklund & Åkerstedt, 1997; Rosipal et al., 2013) and others did not (Riedel & Lichstein, 1998; Westerlund et al., 2016). In addition, perceived sleep quality has been associated with nonsleep factors such as employment status, age, and perceived stress (Tworoger, Davis, Vitiello, Lentz, & McTiernan, 2005).

A few studies have examined the relationship between sleep measures and perceived sleep quality in a longitudinal design (Åkerstedt Hume, Minors & Waterhouse, 1994b; Landry, Best, & Liu-Ambrose, 2015; Lemola, Ledermann, & Friedman, 2013). Åkerstedt et al. (1994b) investigated the determinants of perceived sleep quality the next morning using a forced sleep schedule of 8 periods of sleep of 6 hr and 8 naps of 1 hr. They found that perceived sleep quality was related to sleep efficiency but not to sleep stages. Lemola et al. (2013) monitored participants for one week using an actigraph and observed that variability of sleep duration was related to perceived sleep quality and subjective well-being. They took into account variability between nights of the sleep parameters but used the Pittsburgh Sleep Quality Index (PSQI; Buysse, Reynolds, Monk, Berman, & Kupfer, 1989), an overall sleep quality score of the past month, as the perceived sleep quality measure. Most research in perceived sleep quality uses a perceived sleep quality score such as PSQI, where sleep is evaluated passively over time, and as a result, less is known about the individual evening sleep itself.

Curiously, only a limited number of studies examined the association between perceived sleep quality and self-reported sleep measures, obtained for instance from a diary. Åkerstedt, Hume, Minors & Waterhouse, (1994a) determined the subjective meaning of good sleep and proposed a sleep quality index consisting of sleep quality, calmness of sleep, ease of falling asleep, and ability to sleep throughout the allotted time. In another study, they predicted the sleep quality index with subjective sleep variables and other factors, such as health and stress measures (Åkerstedt et.al., 2012). The most profound factor in this latter study was stress before bedtime, followed by late awakening, short prior sleep, high quality of prior sleep and good health the prior day.

It is clear that many factors play a role when judging sleep quality; however, their relative importance needs to be determined. In addition, longitudinal studies on sleep quality are scarce and outcomes differ between reports. The aim of the current study was to further establish the determinants of perceived sleep quality scored on a daily basis, assessing the contribution of actigraphy-based sleep measures as well as self-reported sleep indices.

Methods

Participants

A recruitment agency located in Amsterdam, the Netherlands, was responsible for finding participants, from an available panel of 95,000 persons. Participants had to be between 40 and 65 years old, assuming that people around this age have a relatively stable daily routine and sleep rhythm. Participants were excluded when (a) diagnosed with any neurological, cardiovascular, psychiatric, pulmonary, or endocrinological disorder, (b) diagnosed with a sleep disorder, (c) found to be using sleep, antidepressant, or cardiovascular medication, (d) found to be using drugs or excessive alcohol (> 3 units per day), (e) pregnant, or (6) working shifts or crossing more than two time zones in the last two months. Initially, 58 participants started the study but 7 dropped out. The reasons for withdrawing were (a) study protocol deemed too intensive (2), (b) dysfunctionality of devices (2), (c) discomfort with wearing devices during sleep (2), and (d) personal reasons (1). Data of one participant was excluded, because it was unreliable, as the morning diary was continuously filled in too late, resulting in a total sample of 50 participants. All participants received a reimbursement of 50 euros.

Study design and procedure

This study was conducted from February 2015 to May 2015. All procedures and measures were approved by the Internal Committee of Biomedical Experiments of Philips Research. No medical approval was required. The study was part of a larger project examining the sleep quality in middle-aged persons. The study design consisted of two weeks of home monitoring, during which participants wore a wrist actigraph and filled in a sleep and wake diary in the evening as well as in the morning (Figure 1). Participants were invited for an intake that was held two days prior to the start of the study. During the intake, written informed consent was obtained and the study design and use of the equipment were explained. Additionally, demographic information (age, gender, number of working days), general health, and the Pittsburgh Sleep Quality Index (PSQI; Buysse et al., 1989) were assessed. Participants filled in the sleep and wake diary through a laptop that was provided to them during the intake. On the laptop, a link was installed to access the software package Tempest (Batalas & Markopoulos, 2012a, 2012b), which enables the user to fill in the sleep–wake diary online as well as offline. When information was entered without an Internet connection present, data was stored locally until an Internet connection was established and data was sent to the research server.

Figure 1. Overview of the procedure of the study. Indicated by the arrows are the predictors that are utilized in the analysis to predict perceived sleep quality the next morning. Sleep variables of the preceding sleep are only used in the analysis of model Extended.

Measures

Perceived sleep quality

As the primary outcome, perceived sleep quality was assessed using a self-rating visual analog scale, ranging from 0 (bad night’s sleep) to 100 (good night’s sleep), in response to the following question: “How well did you sleep last night?” Higher scores indicated better sleep quality, referred to as “perceived sleep quality score” (PSQ).

The sleep and wake diary

The sleep and wake diary included a morning section and an evening section. The morning section consisted of the PSQ, the Consensus Sleep Diary (CSD; Carney et al., 2012), and questions about sleep rhythm and specific sleep issues, such as problems falling or staying asleep, early morning awakenings, and feeling refreshed in the morning. The evening section contained questions about daily habits, stress, and well-being during the day.

The Consensus Sleep Diary was developed by sleep experts and potential users in order to meet the need of a standardized sleep diary facilitating comparison across studies (Carney et al., 2012). It contains questions such as, “When did you go to bed?” (bedtime), “How long did it take you to fall asleep?” (sleep onset latency [SOL]), “How many times did you wake up during the night?” (WASO), and, “Did you sleep well?” In our study, these are referred to as the self-reported sleep measures. The morning section of the sleep and wake diary had to be filled in within 1 hr after waking up.

The evening section contained questions regarding daily habits and mood, such as stress level during the day (stress_day), stress level before bedtime (stress_bedtime), and well-being during the day (WB_emotion/WB_energy). Stress level was assessed on a scale from 1 to 5; with 5 being very stressed. Well-being during the day was determined by a part of the Short Form Health Survey (SF-36; Jenkinson, Coulter, & Wright, 1993). The subscores energy/fatigue (WB_energy) and emotional well-being (WB_emotion) were assessed, with higher scores indicating better well-being during the day (0–100). In addition, the evening diary inquired whether participants took a nap during the day (yes/no). Participants were asked to fill in the evening section of the sleep and wake diary 1 hr prior to going to bed.

Actigraphy-based sleep measures

During all nights, participants wore the Actiwatch Spectrum (AW) on the dominant wrist (Philips Respironics, Inc., Murrysville, USA). It was set to a standard sampling rate of 120 per hr. Actigraphy-based sleep parameters were determined using the Actiware software (version 5.57.0006). Participants were instructed to push a marker button when they were going to sleep and after their final awaking, such that a sleep interval was annotated based on these inputs. The following parameters were derived from this interval: bedtime (00:00:00), total sleep time (TST), wake time after sleep onset (WASO), number of awakenings during the night, sleep onset latency (SOL), and final wake-up time (00:00:00), all reported in minutes unless stated otherwise. Number of awakenings was calculated as “the total number of continuous blocks of epochs where each epoch was scored as wake for the given sleep interval.” These sleep outcomes are referred to below as the actigraphy-based sleep measures.

Statistical analysis

We performed a multilevel analysis, as this enables us to determine the sleep quality and its determinants on a night-to-night basis. Multilevel analyses were performed on the response variable “perceived sleep quality score” (PSQ). The analyses were done on complete-record data, since less than 12% of the data was missing.

First, it was examined how much actigraphy-based and self-reported sleep measures explain perceived sleep quality. Two multilevel models with random intercepts for participants and independent intrasubject variability were built—model Acti and model Self. The variables included were TST, SOL, bedtime, WASO, number of awakenings, and final wake-up time, corrected for age and gender, including all two-way interactions. The selection of the significant variables and their interactions was done through stepwise backward elimination (keeping the models hierarchical at all times). Pearson’s correlation coefficients between the observed and (the best linear unbiased) predictions of PSQ were calculated to quantify model prediction.

As a second step, the model that performed best in the first analysis was extended by including explanatory variables other than sleep characteristics of the present night (model Extended), that is, the self-reported sleep variables of the preceding night (the night prior to the previous night; named: _pre variables; Figure 1), weekday or weekend day, employment status (employed or not), daytime naps (naps), stress levels during the day (stress_day), and at bedtime (stress_bedtime), WB_energy and WB_emotion.

The variables involving a length of time (i.e., TST, SOL, WASO) were converted into hours, and bedtime was centered around midnight, that is, negative values indicated times before midnight. For interpretation and direct understanding of the size effects respectively, we consider both the variables in their original scale and their corresponding standardized versions. For the latter, all the continuous explanatory variables were standardized to zero mean and unit standard deviation (with respect to the means and standard deviations provided in Table 1). Reported relationships were significant at a two-tailed p value of ≤ .05. All analyses were conducted using the MIXED procedure in the SAS software (SAS 9.4 for Windows).

Table 1. Descriptive demographics with sleep and nonsleep characteristics of the study sample

Characteristics
Age (years)	51.8 ± 6.6
Female gender, N, (%)	26, 54%
PSQI	4.16 ± 2.51
PSQ (vas)	71.35 ± 23.22
Weekend nights, (%)	28%
Employed, (%)	76%
Stress_day	1.54 ± 0.76
Stress_bedtime	1.23 ± 0.55
WB_energy	76.48 ± 16.27
WB_emotion	86.19 ± 12.21
	Actigraphy-based sleep measures	Self-reported sleep measures	r
TST (min)	454.31 ± 87.76	436.95 ± 87.93	0.66*
SOL (min)	10.24 ± 17.09	17.21 ± 26.45	0.24*
Bedtime (time)	24:02 ± 1:19	24:04 ± 1:15	0.95*
WASO (min)	28.76 ± 14.02	14.95 ± 24.91	0.22*
Number of awakenings (#)	29.67 ± 10.55	1.77 ± 1.46	0.16*
Final wake-up time (time)	7:46 ± 1:24	7:37 ± 1:28	0.93*

Note. All reported in mean ± standard deviation, unless stated otherwise. PSQ = perceived sleep quality score, WB = well-being during the day, TST = total sleep time, SOL = sleep onset latency, WASO = wake time after sleep onset, r = Pearson correlations between the actigraphy-based sleep measures and self-reported sleep measures, * = significant at the < .01 level (two-tailed).

Results

The characteristics of the study sample are shown in Table 1. Approximately half of the participants (54%) were female and the age range of the sample was between 41 and 65 years. The PSQI global score ranged from 1 to 13, with 9 persons above the cutoff score of 5, meaning that our sample generally did not experience sleep complaints.

In Table 2 we show the outcome of model Acti. The strongest predictors were WASO, number of awakenings, and total sleep time. For the latter two, an interaction term adjusted the coefficients for the female group. The positive effect of total sleep time is stronger for women (7.29) than for men (3.26), while the effect of the number of awakenings has different directions, being positive for males and negative for females (for each awakening during sleep the PSQ increases 0.59 points for men, while it decreases 0.14 points for women). Moreover, we observed a significant interaction effect between sleep onset latency and time in bed: the negative effect of sleep onset latency is made positive by late bedtime (i.e., later than 00:46).

Table 2. Model acti. Coefficients for the fixed effects with corresponding estimation error and p value

	Model Acti
Effect	β	SE	p	b	SE	p
Intercept	69.53	2.34	< .001	48.05	8.52	< .001
Gender (effect of females)	5.23	3.23	0.11	−3.68	10.86	0.73
WASO	−9.89	1.25	< .001	−43.32	5.36	< .001
Number of awakenings	6.26	1.76	< .001	0.59	0.17	< .001
TST	4.27	1.65	0.01	3.26	1.26	0.01
Bedtime	2.61	1.09	0.02	1.03	0.90	0.25
SOL	−1.02	0.99	0.30	−4.23	3.59	0.24
Number of awakenings *Gender	−7.69	2.28	< .001	−0.73	0.22	< .001
TST*Gender	5.29	2.11	0.01	4.03	1.61	0.01
SOL*Bedtime	2.07	0.80	0.01	5.50	2.13	0.01

Note. Model Acti = only the actigraphy-based sleep measures of the present night were included in this model, β = standardized coefficient, b = unstandardized coefficient, SE = standard error, p = p value, WASO = wake time after sleep onset, TST = total sleep time, SOL = sleep onset latency. The * denotes interaction between two variables. Variance of the random intercept:${\hat{\tau }}^2=97.42$ and residual variance ${\hat{\sigma }}^2=323.95$.

Model Self was mostly determined by number of awakenings, sleep onset latency, and WASO (Table 3). In this case, only the coefficient of WASO was different for the female and male groups, with the negative effect of WASO on PSQ being greater for females: for each extra hour of WASO, women report on average a PSQ score 14.52 points lower than males. Furthermore, we found significant interactions between the number of awakenings and age, sleep onset latency, and WASO, and between sleep onset latency and final wake-up time. Due to the first interaction term, the negative effect of the number of awakenings is softened by increased age though never becoming positive (the coefficient for number of awakenings is in fact −17.51 + 0.25*Age). Furthermore, the latter two interactions mean that, although a higher sleep onset latency lowers the total PSQ score, the negative effect of sleep onset latency on PSQ is reduced along with a later wake-up time and a higher WASO. In particular, the coefficient for SOL becomes −40.88 + 9.38*WASO +3.21*final wake-up time, which can even become positive for certain values of the variables involved. The total variance explained by the self-reported model was 61% and for the actigraphy-based model 41%. When replacing the best linear unbiased predictions (including random effects) with exclusively fixed effects predictions, only 18% of the total variance of perceived sleep quality was explained by the actigraphy sleep measures, and for the self-reported model this resulted in 49%.

Table 3. Model self. Coefficients for the fixed effects with corresponding estimation error and p value

	Model Self
Effect	β	SE	p	b	SE	p
Intercept	70.05	1.72	< .001	74.74	12.42	< .001
Gender (effect of females)	3.39	2.38	0.16	7.01	2.51	0.007
Age	0.91	1.19	0.45	−0.30	0.22	0.18
WASO	−4.58	1.03	< .001	−13.72	2.81	< .001
Number of awakenings	−6.95	0.80	0.001	−17.51	4.47	0.001
TST	3.00	1.17	0.01	2.05	0.80	0.01
Bedtime	−3.17	1.15	0.01	−2.54	0.92	0.006
Final wake-up time	3.78	1.31	0.004	1.63	0.93	0.08
SOL	−6.08	0.80	< .001	−40.88	9.59	< .001
WASO*Gender	−6.03	1.47	< .001	−14.52	3.54	< .001
Number of awakenings * Age	2.38	0.83	0.004	0.25	0.086	0.004
SOL*WASO	1.72	0.52	0.001	9.38	2.87	0.001
SOL*Final wake-up time	2.10	0.80	0.009	3.21	1.22	0.009

Note. Model Self = only the self-reported sleep measures of the present night were included in this model, β = standardized coefficient, b = unstandardized coefficient, SE = standard error, p = p value, WASO = wake time after sleep onset, TST = total sleep time, SOL = sleep onset latency. The * denotes interaction between two variables. Variance of the random intercept was estimated to be ${\hat{\tau }}^2=49.54$ and the residual variance was ${\hat{\sigma }}^2=214.28$.

Subsequently, model Self was extended by including nonsleep characteristics (model Extended). In Table 4 the outcome of the extended model is shown. The strongest effect was represented by number of awakenings during the night, followed by total sleep time of the preceding night, sleep onset latency, and WASO. For each awakening during the night, the PSQ score was reduced with 18.15 points in this model. Moreover, due to significant interaction terms, the negative effect of both number of awakenings and total sleep time of the preceding night alleviates with increased age (the negative effect of bedtime_pre can even be inverted by old age). In addition, both WASO and bedtime of the preceding night show a difference among genders. For women, the effect of WASO is more negative than for men, and the effect of bedtime of the preceding night is positive while it is negative for men. The extended model explained approximately 64% of the total variance.

Table 4. Model extended. Coefficients for the fixed effects with corresponding estimation error and p-value

	Model Extended (self-reported)
Effect	β	SE	p	b	SE	p
Intercept	64.13	10.01	< .001	132.48	34.68	< .001
Gender (effect of female)	4.21	2.47	0.01	7.49	2.63	0.006
Age	0.12	0.19	0.52	−1.53	0.65	0.02
Number of awakenings	−24.77	6.70	< .001	−18.15	4.91	< .001
TST_pre	−12.89	6.11	0.04	−9.0	4.27	0.04
SOL	−5.66	0.91	< .001	−12.09	1.94	< .001
WASO	−4.37	1.16	< .001	−10.72	2.85	< .001
Final wake-up time	3.83	1.43	0.008	2.73	1.02	0.008
Bedtime	−3.81	1.27	0.003	−3.17	1.06	0.003
TST	3.26	1.39	0.02	2.25	0.96	0.02
Bedtime_pre	−1.75	1.34	0.19	−1.52	1.17	0.19
Number of awakenings*Age	0.38	0.13	0.004	0.28	0.09	0.004
TST_pre*Age	0.23	0.11	0.05	0.16	0.08	0.05
WASO*Gender	−5.40	1.55	< .001	−13.25	3.80	< .001
Bedtime_pre*Gender	3.60	1.67	0.03	3.14	1.45	0.03

Note. Model Extended includes self-reported sleep variables of the present night as well as from the preceding night. _pre = variable of preceding night: the night prior to the present night, β = standardized coefficient, b = unstandardized coefficient, SE = standard error, p = p value, TST = total sleep time, SOL = sleep onset latency, WASO = wake time after sleep onset. The * denotes interaction effects between variables. Variance of the random intercept was estimated to be ${\hat{\tau }}^2=50.83$ and the residual variance was ${\hat{\sigma }}^2=196.73$.

Discussion

This study investigated the perceived sleep quality in a longitudinal design, using a daily measure of perceived sleep quality. We performed a modeling approach to determine the predictive value of actigraphy-based versus self-reported sleep measures. It was found that self-reported sleep measures were stronger predictors of perceived sleep quality than actigraphy-based sleep indices. The self-reported number of awakenings and total sleep time of the preceding night were the strongest predictors of perceived sleep quality the next morning. This means that the present night as well as the preceding night are indicative of perceived sleep quality the next morning.

Wake time during the night was a strong indicator in our models, which is in line with previous research (Argyropoulos et al., 2003; Keklund & Åkerstedt, 1997; Rosipal et al., 2013). However, in our results, this association was more evident in females, based in the significant interaction between wake time after sleep onset and gender in the self-reported model. The actigraphy-based model showed a positive effect for number of awakenings during the night. This is a counterintuitive result, as one would expect that the number of awakenings negatively affects the perceived sleep quality. In fact, this positive effect was found only for males, since the interaction term of the number of awakenings with gender would make the influence of number of awakenings for females negative. This counterintuitive relation may indicate that males notice awakenings during the night less, and would therefore not necessarily evaluate the perceived sleep quality as negative or possibly even as positive. The difference between genders was also observed in the overall effect of total sleep time (model Acti). The effect for sleep duration was stronger for females than for males; the longer their sleep duration, the better their rating of perceived sleep quality.

The sleep variables that were significant in the self-reported model remained when additional nonsleep factors and sleep measures of the preceding night were added in an extended model. Among the additional variables, total sleep time of the preceding night and the time the participants went to bed the night before (only for females) were indicators of the perceived sleep quality the following morning. Lemola et al. (2013), found an association between the variability of total sleep time (coefficient of variation based on 7 nights, resulting in a cross-sectional design) and sleep quality. This finding was confirmed in our study where total sleep time of the preceding nights had an effect on the sleep quality rating the following night. In our data, employment status, day of the week, daytime naps, emotional and energy well-being scores of the day before, and stress levels were not significant explanatory variables for the perceived sleep quality. This is different from previous studies, where associations were found between perceived sleep quality and stress (Kashani, Eliasson, & Vernalis, 2012; Winbush, Gross, & Kreitzer, 2007), and well-being (Ford & Cooper-Patrick, 2001; Mayers, Grabau, Campbell, & Baldwin, 2009). This could be due to differences in assessing perceived sleep quality: most studies used an overall sleep quality score over the previous month in a cross-sectional design, while we incorporated a direct sleep quality score that was rated every morning. In addition, the mean and standard deviation of stress level during the day of all participants was low, which could have yielded a variance too low to find possible associations between stress and perceived sleep quality.

The subject-specific intercept helped to improve the fit of the actigraphy-based model substantially, increasing the explained variation from 0.18 to 0.41. In fact, this additional random term captures the subjective baselines, thus a prediction of 41% would be achievable only when some baseline PSQ information of a participant is present. To calculate the random intercept, multiple nights need to be taken into account. The prediction of this perceived sleep quality will be more accurate as it is more tailored to the person. In the self-reported model, we observed that the same inclusion of a random intercept had a smaller effect (49–61%). In this case, some of the subjectivity was already included in the explanatory variables.

This shows that psychological constructs, such as perceived sleep quality, are difficult to determine for a whole group. People have their own individual interpretation strategies regarding the course of the night and when answering sleep-related questions. Personality aspects such as general positive or negative affect may be represented in multiple domains (i.e., the perceived sleep quality score and individual sleep characteristics), implying an association that might be caused partly by method bias (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). These could be reasons why the relationship between the actigraphy-based sleep characteristics and perceived sleep quality was only modest.

It is also well known that subjective variables correlate best with other self-reported measures. Therefore, the difference found between the actigraphy-based sleep model and the self-reported sleep model could be inherent in the assessment method they represent. Earlier research has found that self-reported total sleep time was consistently higher compared to actigraphy-based total sleep time (Lauderdale, Knutson, Yan, Liu, & Rathouz, 2008). Moreover, the actigraph algorithm operationalizing the number of awakenings may influence the differences between objective and self-reported number of awakenings. For example, recording continuous wake time is difficult for an actigraph, as it is challenging for these devices to differentiate between lying still but awake and being asleep. This could mean that the actigraph records several awakenings during an actual single period of wake time. Nevertheless, the above results suggest that the perceived sleep experience is an important factor in determining overall sleep quality and therefore it may affect other parts of people’s life as well. For instance, functioning during the day might be influenced by the feeling of being well rested after a night sleep (Jean-Louis, Kripke, & Ancoli-Israel, 2000; Lemola et al., 2013; Taylor, Lichstein, & Durrence, 2003).

With regard to the current study, some strengths and limitations are important to note. We were able to examine the sleep quality determinants in a real-life setting for 14 days, enabling analysis of perceived sleep quality in a between- and within-subject design. However, since there are currently no validated questionnaires that evaluate well-being and stress on a daily basis, questions from other questionnaires were used. The included age range of 40 to 65 years limits the generalizability of our findings. Future studies should assess the perceived sleep quality in daily life in other age groups, such as students, given that these persons typically have different sleep–wake rhythms. In addition, future studies that focus on categorization of individuals based on several sleep characteristics may improve the prediction of perceived sleep quality (Krystal & Edinger, 2008).

In conclusion, we were able to examine the determinants of perceived sleep quality, by using a longitudinal approach over multiple nights and a daily measure of perceived sleep quality in a middle-aged sample. Perceived sleep quality was best described by self-reported sleep measures that include the number of awakenings during the night and the total sleep time of the preceding night. The use of the umbrella term sleep quality as a reference to either objective or subjective sleep quality is not appropriate and a distinction should be made between objective sleep quality and perceived sleep quality. To improve the accuracy of perceived sleep quality predictions, multiple nights should be taken into account and a subject-specific intercept should be included.

Acknowledgments

This research is performed within the framework of the Data Science Flagship. We would like to thank Charlotte Lunsingh Scheurleer, Leonie van den Heuvel, and Mustafa Radha for helping with the data collection.

Funding

This work was supported by POINT ONE, a government agency providing general funding for the project. In addition, Royal Philips, a commercial sponsor provided general funding for the project.

Additional information

Funding

This work was supported by POINT ONE, a government agency providing general funding for the project. In addition, Royal Philips, a commercial sponsor provided general funding for the project.