Insomnia disorder is characterised by persistent difficulties with the initiation and/or maintenance of sleep, leading to daytime impairment [1
]. Epidemiological studies have shown that approximately one in ten people meet the criteria for insomnia disorder [3
]. Insomnia can have severe consequences for many aspects of life, including work performance, social functioning, and health [5
A diagnosis of insomnia disorder is typically based on self-reported symptoms alone. However, numerous studies have demonstrated a mismatch between subjective reports and objective estimates of sleep in people with insomnia [11
]. Specifically, many individuals with a diagnosis of insomnia disorder do not demonstrate sleep abnormalities according to objective assessments such as Polysomnography (PSG). Where objective measures do corroborate abnormalities in sleep, it is often not to the extent that the subjective report suggests [12
]. There is debate in the field as to whether the subjective/objective mismatch represents a distinct subtype of insomnia (“sleep state misperception”, “subjective insomnia”, “paradoxical insomnia”), or is a more general feature of the disorder [11
]. Whilst there is considerable variation in the accuracy of sleep perceptions within insomnia samples [14
], on average, individuals with insomnia have a tendency to underestimate actual sleep [15
]. This is in contrast to good sleepers, who tend to overestimate sleep [20
] or estimate objective sleep parameters accurately [18
]. These findings have implications for the assessment and diagnosis of insomnia and highlight uncertainties that exist around whether insomnia is best captured by subjective or objective methods. They also expose the potential importance of perceptions of sleep for our understanding of insomnia.
The discrepancy between subjective reports and objective estimates has been demonstrated in a variety of indices of sleep, including total sleep time (TST) [12
], sleep onset latency (SOL) [13
], wake-time after sleep onset (WASO) [24
], and sleep efficiency (SE) [26
]. Discrepancy occurs in primary insomnia and also in patient groups where insomnia is comorbid with other health or psychiatric conditions [31
]. There is evidence that discrepancy increases with advancing age and may play a role in the higher rates of self-reported insomnia in later life [29
The tendency for people with insomnia to underestimate objective sleep has been conceptualised in several ways; (i) as an exaggeration of sleep difficulties, perhaps due to more general psychological characteristics and personality traits [19
]; (ii) as a meaningful phenomenon which reflects a localised sleep disturbance with candidate physiological markers [20
]; and (iii) as a cognitive distortion which contributes to the maintenance and escalation of insomnia [37
]. According to Harvey’s Cognitive Model (2002), those who underestimate their sleep may be more at risk of developing objective sleep deficits due to increased preoccupation with sleep and increased sleep related anxiety and arousal, which is antithetical to optimal sleep onset and maintenance [37
]. In line with this model, studies which have sought to correct sleep discrepancy have demonstrated changes in insomnia related anxiety and distress and concomitant reductions in self-reported insomnia symptoms [38
]. There is emerging evidence that Cognitive Behavioural Therapy for Insomnia (CBT-I) improves the accuracy of sleep perceptions [30
], a finding which highlighted the possibility that correction of discrepancy could account for some of the efficacy of treatments for insomnia.
A variety of psychological and physiological mechanisms have been proposed to underlie sleep discrepancy (for a review see Harvey and Tang [11
]). These include cognitive arousal [41
], physiological arousal [41
], cortical arousal [20
], selective attention [45
], memory bias [11
], and sleep fragmentation [46
]. A recent study by Takano and colleagues (2016) [48
] in a community sample comprising individuals with and without insomnia symptoms, reported that higher levels of pre-sleep cognitive arousal were associated with underestimations of TST and overestimations of SOL. This work is consistent with correlational and experimental studies that have reported associations between pre-sleep cognitive activity and sleep discrepancy in insomnia samples [19
The studies conducted thus far have assessed sleep discrepancy over a single night or averaged data from multiple nights. Therefore, little is known about intra-individual variability in sleep discrepancy and whether it is affected by natural, day-to-day variations in psychological factors. There is some evidence for high-levels of night-to-night variability in sleep discrepancy in older adults, where the discrepancy between self-report and actigraphy based estimates of sleep onset latency was found to be 150% more variable within the same individual across nights, compared to between individuals [29
]. In addition, there is emerging evidence for intra-individual variability in some of those factors which are proposed to underlie discrepancy, such as arousal [49
] and sleep fragmentation [51
]. The importance of examining intra-individual variability in sleep/wake patterns is increasingly being recognized [52
]. Although daily and average values of sleep parameters tend to be highly correlated, information may be concealed when only average or single values are considered. It is possible that daily values and averaged values have overlapping but distinctive aetiology. For example, daily values may be more highly associated with state like psychophysiological variables than average values.
In this study, we utilised multilevel modelling to investigate whether self-reported arousal (cognitive, physiological), sleep effort, sleep fragmentation, and mood upon awakening predicted sleep discrepancy across seven nights, in a group of poor sleepers reporting insomnia symptoms. We chose to investigate multiple constructs in order to tease out the most important contributors. Multilevel modelling allows for the analysis of within-person changes in variables across nights whilst accounting for the influence of between-subject variations in the relationships of interest. By using this approach, we sought to examine intra-individual variability in sleep discrepancy and assess the relationship between sleep discrepancy and psychophysiological variables over multiple nights, without the requirement to aggregate data [53
]. We chose to focus on cognitive arousal and sleep fragmentation because these constructs have the strongest evidence as predictors of subjective/objective sleep discrepancy [11
]. We included two measures of cognitive arousal; one that assesses general cognitive arousal and another that assesses the content and frequency of thoughts during the pre-sleep period, due to preliminary evidence that certain aspects of cognitive arousal may be more closely associated with sleep disturbance [54
]. We included a measure of self-reported physiological arousal because an experimental study has shown that increases in physiological arousal lead to increases in TST and SOL discrepancy [41
]. The inclusion of a measure of sleep effort was based on research demonstrating that sleep effort is strongly associated with subjective reports of sleep disturbance but not objective sleep parameters (PSG) [55
]. Mood upon awakening was assessed due to evidence that low mood and general feeling state, at the time of reporting of subjective sleep, may mediate underestimations of objective sleep parameters [35
]. It was hypothesised that higher levels of arousal (cognitive and physiological), sleep effort, sleep fragmentation, and worse mood on awakening would be associated with the overestimation of TST. It was also hypothesised that higher levels of arousal (cognitive and physiological), sleep effort, and worse mood on awakening would be associated with the overestimation of SOL. We focused on discrepancies in TST and SOL because the subjective/objective discrepancy has been demonstrated more robustly in these indices.
A mismatch between subjective and objective estimates of sleep parameters is commonly observed in people with insomnia, however little is known about the mechanisms underlying this phenomenon. This study sought to determine predictors of subjective/objective sleep discrepancy in individuals with insomnia symptoms. Using actigraphy and sleep diaries, we conducted repeated longitudinal assessments of sleep discrepancy, pre-sleep, and next-day psychophysiological factors, across seven days and nights. Our results highlight roles for arousal, sleep effort, mood upon awakening, and sleep fragmentation.
Examination of subjective and objective sleep over multiple nights enabled us to identify high levels of intra-individual variability for MI and SOLd in this sample. Overall, 54.1% of the variation in MI was due to differences between days within the same participant. Both overestimation and underestimation of TST was evident. Of the 41 poor-sleepers who took part in the study, 30 (73%) displayed a mixture of over and underestimation of TST. These results do not support the proposal that underestimation of TST is a consistent and trait-like feature of people with insomnia. There was also considerable intra-individual variability in SOLd, where 82.9% of the variation was due to differences between days within the same participant. Both overestimation and underestimation of SOL was evident, however the frequency of underestimation of SOL was relatively small, with just 14.7% of subjective SOL values representing an underestimation.
4.1. Predictors of Subjective/Objective Sleep Discrepancy
Univariate analyses revealed that cognitive arousal (general cognitive arousal and specific pre-sleep cognitive activity measured using the GCTI), sleep effort, sleep fragmentation, and mood upon awakening were all significant predictors of MI. Multivariable analysis revealed that pre-sleep cognitive activity and mood upon awakening provided statistically significant, independent contributions to MI. With regards to SOLd, the univariate analyses identified sleep effort as the only significant predictor.
These findings suggest that cognitive arousal is associated with subjective/objective discrepancy in TST. These data corroborate evidence from an experimental study in which provoking an increase in cognitive arousal led to increases in TST sleep discrepancy [41
]. They also support the work of Takano and colleagues [48
], who found that cognitive arousal was uniquely associated with TST discrepancy in a community sample. Mechanisms through which cognitive arousal contributes to sleep discrepancy have been proposed, however the evidence is limited. One suggestion is that cognitive arousal distorts the perception of time because a unit of time is perceived as longer when more information is processed (through greater levels of mentation under high arousal conditions) [41
]. Another proposal is that cognitive arousal maintains an enhanced level of sensory and memory processing during sleep onset, which obscures the distinction between sleep and wakefulness [20
]. In line with this, an association between high-frequency Electroencephalogram (EEG) activity during non-rapid eye movement (NREM) sleep and subjective/objective sleep discrepancy has been observed [20
]. High frequency EEG activity is thought to be a marker of sensory processing and memory formation. Further research implementing the fine grained measurement of sleep using techniques such as high density EEG is required to advance our understanding of the possible neurophysiological processes underlying associations between cognitive arousal and sleep discrepancy.
In this study, cognitive arousal was assessed using two measures; the GCTI which evaluates the content and frequency of pre-sleep cognitions and a visual analogue scale rating the extent to which participants experienced thoughts running through their minds (general cognitive arousal) during the pre-sleep period. In the univariate analyses, both measures of cognitive arousal were significant predictors of MI, however in the multivariable analysis, only the GCTI was a significant predictor. Clearly there is substantial overlap between these two measures, as indicated by the moderate strength correlation between responses (r (266) = 0.63, p < 0.001). Shared variance may explain why only the GCTI was a significant predictor of MI in the multivariable analysis, however there is little change in the predictive value of general cognitive arousal when scores on the GCTI were omitted from the multivariable analysis. The GCTI probes a wide variety of intrusive thoughts which are known to be commonly experienced by individuals with insomnia in the pre-sleep period and contains items such as “How frustrated/upset I am feeling” and “How nervous/anxious I am feeling”, which may capture the emotional and physiological sequelae of intrusive thoughts in a way that a single question about a racing mind does not. Although the GCTI is predominantly a measure of cognitive arousal, it appears to tap into hyperarousal more broadly and this may be the reason that it is the strongest predictor of MI in this study.
The findings from this study suggest that self-reported physiological arousal is not related to subjective/objective discrepancy in TST or SOL. This is contrary to reports from a previous study in which manipulations of physiological arousal using caffeine have led to changes in sleep discrepancy [41
] and a study in which physiological arousal was shown to predict discrepancy in SOL [48
]. We implemented a single item, self-report assessment of physiological arousal, which demonstrated sufficient sensitivity to detect associations between physiological arousal and various subjective and objective sleep parameters in a previous study of chronic pain patients [68
]. However, self-report may be less sensitive in the domain of physiological arousal. Future work should administer validated subjective and objective measures of physiological arousal to fully assess its contribution to sleep discrepancy.
Mood upon awakening was a significant predictor of MI in both the univariate and multivariable analyses. Previous work has revealed associations between sleep discrepancy and depressive symptoms assessed at baseline [75
]. The findings from this study extend that work by showing that daily fluctuations in morning mood are associated with subjective/objective discrepancy in TST. The relationship between sleep discrepancy and mood upon awakening may be explained by mood congruent memory bias, in which an individual recalls or selectively processes information that is consistent with their current mood. This is a phenomenon which has been documented in individuals with clinical depression [77
] and during depressed mood induction in non-clinical populations [78
]. Specifically, when an individual is making a judgement about how well they slept the previous night, current feeling state may distort memory such that low mood or dysphoria at the time of reporting leads to negatively biased judgements of sleep quantity and/or quality. It has long been established that memory is a reconstructive process affected by bias and error [79
]. Moreover, mood congruent memory biases have been demonstrated in a variety of contexts, including symptom reporting [81
Another possible explanation for the association between sleep discrepancy and mood upon awakening is that greater sleep discrepancy leads to worse mood upon awakening or that these variables are related by means of a third factor, such as sleep quality. An association between sleep quality and next-day affect is well established [83
] and a number of studies have suggested a link between poor sleep quality and increased subjective/objective sleep discrepancy [28
]. Due to the nature of the study design, causal inferences with regards to the relationships uncovered cannot be made. Experimental investigations are required to determine the direction of the effect and potential mediators of the relationship. For example, future studies could use a mood induction paradigm to examine the impact of mood upon awakening on subsequent subjective reports of sleep quantity and quality.
Consistent with our hypothesis, sleep effort was a significant predictor of both MI and SOLd
. These findings lend support to the attention-intention-effort model of insomnia [85
] which proposes that explicit intention to sleep inhibits normal de-arousal and subsequently hinders sleep. Sleep effort appears to play a particularly important role in SOL discrepancy, where it was the only significant predictor. In line with our findings, a previous study reported that reductions in sleep effort mediated the improved accuracy of sleep perceptions following paradoxical intention [86
]. One possibility is that sleep effort maintains and exacerbates sleep difficulties through distorting perceptions of sleep.
Finally, this study revealed a significant association between sleep fragmentation (assessed by actigraphy) and MI in the univariate analysis, whereby higher levels of sleep fragmentation were associated with underestimation of TST. These findings concur with reports from an experimental study, in which inducing brief awakenings led to overestimates of sleep onset latency in normal sleepers [46
]. More frequent awakenings may lead to shallower forms of sleep and greater levels of cortical activity, resulting in difficulties distinguishing wake from sleep [87
This study has several limitations. First, we examined associations between variables and therefore no causal inferences can be made. Second, potentially overlapping constructs were assessed, as indicated by moderate/strong correlations between many of the predictor variables. This complicates the interpretation of the results from the multivariable analysis. Third, the use of actigraphy enabled the assessment of sleep across multiple days in the home environment which increases the ecological validity of findings, however actigraphy is known to overestimate sleep time in individuals with insomnia [88
]. This has implications for the reliability of the sleep discrepancy outcome variables. Fourth, our sample consisted predominantly of females, which limits the generalisability of the results. Fifth, we did not assess sleep microstructure and therefore the contribution of EEG parameters that are proposed to play a role in sleep discrepancy were not examined. Sleep discrepancy has been associated with heightened brain activity during PSG defined sleep [44
]. Brief arousals from Rapid Eye Movement (REM) sleep and time spent in REM sleep have also been shown to correlate with the degree of discrepancy [89
]. It will be important for future studies to include measures such as high density EEG, to understand how the relationships uncovered in the current study are expressed across different levels of explanation (i.e., underlying physiological processes). Finally, we did not conduct rigorous screening for sleep, physical health, or psychiatric comorbidities and we did not assess whether participants were taking substances that might induce sleeplessness (e.g., medications, caffeine, alcohol, illicit drugs). Therefore, it is possible that these factors influenced our findings.