The present analyses used data from the Raine Study, a multigenerational longitudinal epidemiological study established in 1989 (for more study details visit; rainestudy.org.au). Data from Generation 2 (Gen2) participants¹ who completed the sleep study, actigraphy, and cognitive testing at the Gen2-22-year follow-up were used (n = 902).

Individuals with either a diagnosis of insomnia (DSM-5) or who met the insomnia criterion on the insomnia symptom questionnaire (ISQ; n = 156) were separated into NSDI (n = 63) or SSDI (n = 93) based on the Research Diagnostic Criteria (RDC) for PSG (28). For NSDI the RDC requires that PSG shows scored total sleep time (TST) ≥6.5 h, sleep efficiency (SE; TST/Time in bed × 100) ≥85%. For SSDI, the RDC requires short-sleep duration, PSG-based TST <6.5 h.

While participants were categorised by sleep study, they were retained if their self-report sleep diary data validated short or normal sleep duration: normal-sleep duration (n = 53) or short-sleep duration (n = 71). This was to ascertain if this single night of sleep under PSG was typical of an individual's sleep.

A healthy sleeper sample was constructed from those participants from the Raine Study who were without insomnia or other sleep disorders, had SE ≥85%, and TST ≥6.5 h on PSG (n = 324). Participants were retained if they reported normal sleep duration (TST>= 6.5 h) on their sleep diary (n = 246).

Participants who did not meet the insomnia selection criteria, healthy sleeper criteria, had a history of neurological, neurodevelopmental, significant psychiatric history, shift-work, or other sleep disorder/s (e.g., sleep apnoea with apnoea hypopnoea index≥15 and/or restless legs syndrome) were excluded (n = 532) from analyses.

A flow chart of study group selection is shown in Figure 1.

Participants were administered full overnight Level 1 polysomnography (PSG) [Compumedics E-Series (Compumedics, Melbourne Australia)] and scored using Compumedics PSG 3 software at the Centre for Sleep Science, University of Western Australia. Equipment placement, sleep staging and event scoring was completed by experienced sleep technologists according to American Academy of Sleep Medicine criteria (29).

This computerised battery (30) provided tasks of attention (Card Identification), executive functioning (Set Shifting), learning (Continuous Paired Associates), psychomotor function (Detection Test), and working memory (One-back Task). The tasks show good correlations to traditional neuropsychological assessments (30) even when measured in populations that exhibit subtle cognitive changes (31). The Cogstate records both accuracy (whether the trial was answered correctly) and speed (time to make a correct response) on a trial-by-trial basis, allowing calculation of accuracy, mean response time, and variability in response time (inconsistency) using intra-individual standard deviations, for each task assessed.

To date, a few small sample studies examining cognition in obstructive sleep apnoea have included cognitive inconsistency amongst their measures. All have found greater IIV (i.e., more inconsistency or less cognitive stability) in those with obstructive sleep apnoea compared to healthy controls (32). IIV has not been reported in insomnia.

A 16-item questionnaire (33) that provides a self-reported assessment of prospective and retrospective memory errors. The scale demonstrates good construct validity and internal reliability (α = 0.80–89) (33). Scores were generated for prospective and retrospective subscales.

This 12-item questionnaire (34) provides a measure of everyday mistakes made when not paying sufficient attention to a task. The values are summed to give an overall score. ARCES scores are highly correlated with other scales assessing errors of sustained attention (35).

A 13-item self-report tool (36) designed and used to identify insomnia symptoms and provide a case definition of insomnia. The items are consistent with the RDC (28) for insomnia and compared with interview methods of diagnosing insomnia the ISQ has good internal reliability (α = 0.89), moderate sensitivity (50–67%) and good specificity (91%) (36).

This questionnaire (37) has 10-items designed to assess the impact of daytime sleepiness on daytime activity. The FOSQ-10 has good internal consistency (α = 0.87) and demonstrates changes over time with successful treatment of sleep disorders (38). Scores are reported for general productivity, vigilance, social outcomes, activity level, and sexual desire.

An 18-item questionnaire (39) that is designed to assess sleep habits, disturbances, and daytime impairments. This scale shows good internal consistency (α = 0.73) and shows a strong correlation with other scales assessing daytime function (40).

This self-administered 8-item scale (41) assesses sleepiness in several every-day situations. The questionnaire has good internal consistency (α = 0.73–0.90). Excessive sleepiness is a score of 10 or more.

This well-established 21-item questionnaire (42) assesses symptoms of depression, anxiety, and stress. It has demonstrated good internal reliability (α = 0.82–0.97) (42) and is an appropriate mood measure for sleep disordered populations, as it does not contain sleep-related items (43). Scores were generated for depression, anxiety, and stress subscales.

Neuropsychological testing and questionnaires were administered to all participants the night of or the morning after their sleep study at the Centre for Sleep Science. There was no adaptation night for this sleep study, hence we examined if this night was representative of an individual's habitual sleep against a weeklong sleep diary.

Descriptive statistics for the SSDI, NSDI, and healthy-sleeper groups are presented in Table 1.

Residuals of the dependent variables were checked for and approximated normality as assessed through visual inspection of normality plots, p-p plots, and metrics of skew and kurtosis. Due to the differences in sample size, homogeneity of variance was assessed and was not violated. Independence of observations was met as all participants were counted in only one group (healthy, long-sleep duration, or short-sleep).

MANOVAs were conducted to compare the effect of group (normal-sleep duration insomnia, short-sleep type insomnia, and healthy sleepers) on lab-assessed sleep differences (PSG—-percentage N1, N2, N3, and REM), lab-based (Cogstate) and self-report cognition (retrospective and prospective memory subscales of the PRMQ and ARCES total score), self-reported mood (DASS-21- stress, anxiety and depression subscale scores), and self-reported daytime function (FOSQ—-general productivity, vigilance, social outcomes, activity level and sexual desire subscale scores). Lab-based (Cogstate) domains of cognition examined were attention, executive functioning, working memory, learning, and psychomotor function. For all domains, tests of accuracy, speed, and inconsistency of performance were assessed for group differences. Where tests of sphericity were violated an adjusted F value is reported. An alpha level of 0.05 was used, except where omnibus group interaction or main effect differences were found, then post-hoc Bonferroni-corrected tests were conducted (0.05 × number of tests).

Means for sleep study data are presented in Table 1.

Interactions were significant for group by sleep stage, Roy's Largest Root, F_{(3, 427)} = 2267.69, p < 0.001. Indicating that time spent in sleep stages varied by group.

Post-hoc comparisons indicated that participants with SSDI spent a larger percentage of TST in N1 compared to the healthy sleepers [p < 0.001; Cohen's d = 0.39 (95% CI, LL = 0.12, UL = 0.65)] and to those with NSDI [p < 0.001; Cohen's d = 0.67 (95% CI, LL = 0.31, UL = 1.02)]. Those with SSDI also spent a higher percentage of TST in N3 than those with healthy sleep [p < 0.001; Cohen's d = 0.63 (95% CI, LL = 0.36, UL = 0.89)] or with NSDI [p < 0.001; Cohen's d = 0.45 (95% CI, LL = 0.09, UL = 0.81)]. Finally, those with SSDI had lower %REM than healthy sleep [p < 0.001; Cohen's d = 0.72 (95% CI, LL = 0.46, UL = 0.99)] and NSDI [p < 0.001; Cohen's d = 0.88 (95% CI, LL = 0.52, UL = 1.23)].

In summary, participants with SSDI demonstrated a greater percentage of their total sleep time in NREM sleep and less time in REM sleep than the healthy sleepers or those with NSDI.

Interactions were not significant for accuracy or reaction time, however, a significant group by cognitive test interaction was found for inconsistency, Roy's Largest Root, F_{(8, 1212)} = 3.15, p = 0.002, revealing a different pattern of cognitive performance for each group.

For attention (inconsistency) the healthy sleeper group showed more consistent response times than the NSDI [p < 0.001; Cohen's d = 0.91 (95% CI, LL = 0.61, UL = 1.20)] and the SSDI (p = 0.006; Cohen's d = 0.44 (95% CI, LL = 0.18, UL = 0.71) groups. For working memory (inconsistency), the healthy sleepers had more consistent response times than the NSDI [p = 0.005; Cohen's d = 0.31 (95% CI, LL = 0.01, UL = 0.61)] but not the SSDI group. For executive function, shifting (inconsistency) healthy sleepers were more consistent in their response times than the SSDI group [p < 0.001; Cohen's d = 0.55 (95% CI, LL = 0.29, UL = 0.82)], but not the NSDI. Means for all lab-assessed cognitive tests are presented in Table 2.

There were no group differences on psychomotor function or learning, nor any interactions with group.

Taken together, these results indicate that the healthy participant group experienced better lab-assessed cognition overall and that both insomnia groups demonstrated greater inconsistency; the NSDI group demonstrated poorer working memory than the healthy sleepers; and, the SSDI group demonstrated poorer executive function, shifting, than the healthy sleepers.

Interactions were significant for group by self-report cognitive test, Roy's largest root, F_{(3, 423)} = 11.29, p < 0.001.

Post-hoc comparisons indicated that all mean scores for all self-reported cognition assessments for the healthy sleeper group were significantly better than the NSDI (p < 0.001) and SSDI groups (p < 0.001), however the insomnia groups did not differ. Means are presented in Table 3.

These results indicate that people from both the SSDI and NSDI groups reported poorer attention, retrospective, and prospective memory than individuals with healthy sleep, and that the insomnia groups did not differ from one another.

Interactions were significant for group by mood [Roy's Largest Root: F_{(3, 402)} = 5.868, p < 0.001] and daytime function by group [Roy's Largest Root: F_{(5, 391)} = 26.596, p < 0.001].

Post-hoc comparisons indicated that mean scores for all assessments of self-reported mood (p < 0.001) and functional sleep outcomes (p < 0.001, with the exception of the FOSQ vigilance subscale, p = 0.003) were significantly higher in both insomnia groups than in healthy sleepers, though the insomnia groups did not differ. Means are presented in Table 4.

These results suggest that both the NSDI and SSDI groups report poorer mood across higher depression, stress, and anxiety, and report that their poor sleep impacts on their ability to function day-to-day, with regards to productivity, vigilance, social outcomes, activity levels, and sexual desire, than healthy sleepers.

In line with the literature (1, 6, 13), those with both SSDI and NSDI self-report problems with cognition (attention and memory), daytime function, mood, and sleep quality. As self-reported sleep quality, daytime function, and mood are core components of an insomnia diagnosis (2), this result is not surprising. Further, these findings suggest that SSDI and NSDI do not differ in terms of self-report measures of cognition, however they do differ on objective assessments.

Both phenotypes of insomnia exhibit inconsistency in attention. Whilst Wardle-Pinkston et al. (6) also reported attention problems in insomnia in their meta-analysis, Fulda and Schulz (10) and Fortier Brochu et al. (8) did not, but all three studies reported accuracy. This sample, using a younger sample than previous studies, found no evidence of problems with attention accuracy, but did find greater inconsistency in the speed of responding to attention trials in insomnia. Likewise, whilst past studies, also with older participants, have reported deficits in working memory and executive function accuracy (6, 8, 10), we found more subtle effects in inconsistency of responding in these domains. These differences were varied across the insomnia phenotypes: those with NSDI were less consistent in working memory despite relatively healthy sleep, while those with SSDI were less consistent in executive functioning, and had different sleep architecture (more N1 and N3, and less REM, as a percentage of total sleep time) than healthy sleepers and those with NSDI.

A high amount of N3 sleep, seen here in those with SSDI, has been noted to indicate rebound sleep. This is considered ‘recovery sleep’ as shown after sleep deprivation or chronic sleep restriction (44). This supports the diagnosis of SSDI, provided by PSG, diary, and self-report symptoms in this study, and supports the idea that the cognitive problems shown in SSDI are the result of chronic sleep loss. Conversely, those with NSDI showed subtle objective and self-reported cognitive deficits, despite no evidence of lab-based sleep loss.

That separable cognitive profiles and sleep profiles were demonstrated for the different phenotypes suggests future directions for providing a differential diagnosis. Currently, as in this paper, overnight sleep study or actigraphy are used to assess the mismatch between self-report and lab-assessed sleep and distinguish short- from normal normal-sleep duration insomnia (28). However, sleep studies, actigraphy, and full neuropsychological assessment can be expensive and time consuming, and are activities requiring a high degree of specialised training. The present paper suggests that cognitive tasks such as those used here, may provide further information to profile those with insomnia. When computerised, these tasks are relatively quick (7-min in healthy participants), and easy to administer. However, these results require replication by other groups and the ability of these differences to discriminate groups requires validation.

Working memory and executive functions are separable but related components of cognition. Working memory is a limited capacity cognitive system that can hold information ready for processing for a limited time (45). The executive function factor assessed here, shifting, is related to working memory, as it assists with shifting attentional control quickly (46). Other aspects of executive function [updating, generativity, fluid problem solving, and inhibition (46, 47)] were not assessed in the present paper, and as such we have an incomplete picture of executive function in these subtypes of insomnia. A complete examination of executive functions in insomnia phenotypes will provide greater understanding of how executive functions are impacted and deepen our understanding of different cognitive profiles in insomnia.

There were no psychomotor or learning problems for either phenotype uncovered in these analyses. These findings are in line with past studies of psychomotor function (6, 8, 10) and in contrast to past explorations of learning in insomnia (6). Learning, in the current sample, may not have been problematic as Generation-2 from the Raine Study were young (Age, M = 22 years, at the time of assessment). Young age is protective for cognition. Wardle-Pinkston et al. (6) investigated age as a moderator in their analyses showing that older age was associated with larger effect sizes in the differences between healthy sleepers and individuals with insomnia. To explore the impact of age, future research could investigate phenotypic differences in the progression of cognitive deficits in longitudinal datasets, or compare cognitive function in older and younger samples. The Raine Study will make an excellent space to explore this concept as data continue to be collected on the same individuals, their parents (Generation 1), their grandparents, (Generation 0) and now the children of Generation 2 (i.e., Generation 3).

Where there were cognitive differences, for both phenotypes of insomnia these were in maintaining consistent response times throughout a testing trial, whilst accuracy and speed were not impacted (6, 8, 10). This finding, taken together with the small effects evidenced in meta-analyses and inconsistent findings across the field, identifies a need for sensitive measures of cognition that capture moment-to-moment performance stability, such as intra-individual variability (IIV) (48). Previous studies have not examined inconsistency, reporting only measures of accuracy and speed, meaning it is possible these consistency differences were present in earlier samples. The literature on IIV indicates it is a sensitive measure of early cognitive change and is predictive of later cognitive dysfunction and decline (49). Future follow-up assessments of the Raine Study participants will be able to track those showing early cognitive instability to see if clearer cognitive problems, in accuracy, speed, and/or a wider set of cognitive domains, develop.

Further, the present sample were relatively healthy, young individuals involved in research from before birth to their mid-twenties, possibly leading to selection bias. As comorbidity, including overweight and psychological diagnoses, can independently impact sleep, sleep disorders, and cognition (50), future studies may wish to investigate the interaction of comorbidity and/or other demographic features with cognition in those with insomnia.

Hyperarousal is a core pathophysiological feature of insomnia, in general, and is explained in two different models: a psychological model and a physiological model. The psychological model posits that worry and rumination about life stress, and about sleep itself, disrupt sleep, whereas, the physiological model posits that hyperarousal is due to a higher level of neuroendocrine and metabolic functioning, which disrupts sleep.

While it is possible that cognition and sleep, in both insomnia phenotypes, are impacted by psychological processes, as certainly belief impacts biological health in other areas, including stress (51) and treatment uptake (52, 53), such a model, with one road to pathology, does not explain the different cognitive and sleep profiles of these disorders, as evidenced here. There is some discussion in the literature of differing types of hyperarousal across the two phenotypes. Short sleep is associated with physiological hyperarousal while NSDI is associated with cortical hyperarousal (17). While physiological hyperarousal is a heightened stress state due to negative thoughts, cortical hyperarousal is increased activation of the reticular formation causing an increase in wakefulness. The result from the present paper provide further evidence of two different disorders that may have different underlying causes of hyperarousal.

Disrupted sleep appears to be a core feature of short-sleep duration insomnia, with less total time asleep, a higher % of NREM, and lower percent of REM sleep, when compared to those with healthy sleep and NSDI. This pattern of sleep architecture is similar to that seen in attention deficit hyperactivity disorder (ADHD) (54), another disorder of hyperarousal. This suggests potential for some shared pathophysiology, for example, problems in the dorsolateral pons, an area of the brain implicated in modulating arousal and sleep states (55, 56). This biological basis for poor sleep may then impact mood and thinking, and then move into a more cyclic relationship between these features.

Further, short sleep duration is associated with chronic insomnia (17). As chronicity in many other disorders is associated with poorer health and cognitive outcomes (24), it is surprising that the results of the present paper do not indicate more severe cognitive problems for those with SSDI. However, the sample was young and relatively healthy, and as cognitive problems were only witnessed in inconsistency (an indicator of early cognitive change), it is possible these individuals have not had sufficient exposure to insomnia, and/or that young age may provide a “buffer” for cognitive problems. For example, N3 declines with age, whereas in the present sample, there was greater quantity of N3 sleep, perhaps reflecting a homeostatic way of compensating for sleep loss that may not be present in an older sample. Examination of the impact of chronicity among these two phenotypes is an important future direction.

By contrast, poor lab-assessed sleep is not a core feature of NSDI, despite less consistent response times to cognitive tasks in comparison to healthy sleepers. This suggests that something else is affecting cognition and mood than the sleep disruption seen in SSDI.

The DMN is a network of interacting brain regions that is active when a person is at rest (18) and is thought to be responsible for “off-line” cognitive functions. This network has been implicated in the paradoxical experience of feeling awake while lab-assessed assessments detect sleep, and has been purported to be responsible for cognitive problems in other neurological disorders, including schizophrenia (57), where there are also working memory deficits and problems with self-referential thoughts (58). As such, NSDI appears not to be a sleep disorder per se, rather a disorder of neural networks.

Different pathophysiology, as suggested by these results, directly impacts the most appropriate therapy and suggests a need for early differential diagnosis.

Cognitive behavioural therapy for insomnia (CBTi) is considered a first-line treatment for insomnia with results superior to benzodiazepines (59). Among other aspects, CBTi includes core components to address unhelpful health beliefs, low mood, and unconsolidated sleep. These factors are important modifiable precipitating and maintaining features of insomnia. While CBTi demonstrates good results, not all patients receive benefits from this treatment (59). It is plausible that those who do not benefit are those for whom the underlying cause has not been addressed. For example, if short-sleep duration is due to biologically-based hyperarousal then CBTi may have limited ability to address such predisposing factors.

Further work to understand these potentially different disorders and their pathophysiology is crucial to inform therapy and provide more individualised treatments.