The effect of clustered versus regular sleep fragmentation on daytime function


Professor Neil J. Douglas, Respiratory Medicine Unit, Department of Medicine, The University of Edinburgh, Royal Infirmary, Lauriston Place, Edinburgh, EH3 9YW, UK. Tel.:+44 131536 3252; fax:+44 131536 3255; e-mail:


Previously, we found that regular sleep fragmentation, similar to that found in patients with sleep apnoea/hypopnoea syndrome (SAHS), impairs daytime function. Apnoeas and hypopnoeas occur in groups in patients with REM or posture related SAHS. Thus, we hypothesised that clustered sleep fragmentation would have a similar impact on daytime function as regular sleep fragmentation. We studied 16 subjects over two pairs of 2 nights and 2 days. The first night of each pair was for acclimatisation. On the second night, subjects either had their sleep fragmented regularly every 90 s, or fragmented every 30 s for 30 min every 90 min, the remaining 60 min being undisturbed. We fragmented sleep with tones to produce a minimum 3 s increase in EEG frequency. During the days following each pair of nights we tested subjects daytime function. Total sleep time (TST) and microarousal frequency were similar on both study nights. We found significantly less stage 2 (55 SD 4, 62±7%; P=0.001) and more slow wave sleep (21 SD 3, 12±6%; P < 0.001) on the clustered night. Mean sleep onset latency was similar on MSLT (clustered 10 SD 5, regular 9±4 min; P=0.7) and MWT (clustered 32 SD 7, regular 30±7 min; P=0.2). There was no difference in subjects mood or cognitive function after either study night. These results suggest that although there is more slow wave sleep (SWS) on the clustered night, similar numbers of sleep fragmenting events produced similar daytime function whether the events were evenly spaced or clustered.


Patients with SAHS suffer from increased daytime sleepiness (Roth et al. 1980) and impaired mood and cognitive function (Greenberg et al. 1987; Cheshire et al. 1992) compared to normal subjects. There is debate as to whether the cause of these deficits in daytime function are due to the nocturnal hypoxaemia or sleep fragmentation suffered by these patients (Roehrs et al. 1989; Bedard et al. 1991). Studies in which sleep has been intermittently disrupted to the extent of producing behavioural awakenings have indicated impaired daytime performance following sleep disruption (Bonnet 1985; Bonnet 1986a; Bonnet 1986b; Bonnet 1987; Bonnet 1989). However, sleep apnoea subjects rarely waken to this level following sleep disruption (Martin et al. 1997a) with brief cortically detected microarousals being much more common. One night of sleep fragmentation as detected by cortical arousal makes normal subjects sleepier during the day (Philip et al. 1994; Roehrs et al. 1994). Neither of these studies found any adverse affect on cognitive function. However, we have previously found that one night of acoustically induced cortical sleep fragmentation causes normal subjects to have increased objective daytime sleepiness and impaired mood and cognitive function (Martin et al. 1996).

Arousals occur in a clustered fashion, particularly in two subsets of patients referred to sleep apnoea clinics; those with REM sleep related apnoea or those with supine/postural related apnoea. REM related respiratory events were originally dismissed as nonpathological (Guilleminault et al. 1976) due to the inherent respiratory instability of REM sleep (Douglas 1994). Lugaresi et al. (1983) described four stages of disease in the progression of snoring to sleep apnoea, the initial phase of which consisted of apnoeas during REM and light sleep. As REM sleep occurs approximately every 90 min during a nights sleep, any sleep fragmentation resulting from REM specific respiratory events therefore occurs in a clustered fashion during sleep. These clusters of sleep fragmentation are interspersed with periods of uninterrupted sleep, in particular SWS. This SWS correlates with objective daytime sleepiness (Philip et al. 1994; Martin et al. 1996) and is thought to be vital for cerebral restitution (Horne 1988) and continued cognitive functioning.

It is uncertain whether these periods of uninterrupted sleep are sufficiently restorative for these patients to overcome the effects of sleep fragmentation on daytime function. We therefore aimed to fragment normal subjects sleep in two different ways so as to produce similar overall microarousal frequencies but different amounts of SWS. In this way we isolated the effects of increased SWS on daytime function while controlling for the effects of sleep fragmentation.



Subjects were recruited from the local student population using advertisements, which did not refer to sleep. Responders with potential sleep disorders were excluded using our in house sleep/wake questionnaire, which is used to assess patients prior to attendance at the Sleep Laboratory. Ethical permission was obtained from the Lothian Research and The University of Edinburgh Ethics Committees. We studied 16 subjects (9 men, 7 women) with a mean age 23±3 SD years. They were all nonobese (body mass index, 21±3 kg/m2) and had Epworth Sleepiness Scores (Johns 1991) in the normal range (mean 5; range 0–9).


We used a two limb randomised crossover design in this study, similar to previous sleep fragmentation studies (Martin et al. 1996, 1997b). The two limbs were separated by a week, with subjects spending a total of two pairs of 2 nights and 2 days in the sleep laboratory. Prior to the first night in the laboratory, subjects received instructions on each test of mood and cognitive function that we used in this study and underwent one practice session. The first night of each pair of nights was for acclimatisation to the laboratory to avoid any first night effect. On the acclimatisation night subjects had electrodes applied to the scalp and were allowed to sleep undisturbed. On the second night of each pair, subjects were randomly assigned to have either of two fragmentation paradigms:

Regular: sleep fragmentation every 90 s from the onset of stage 2 sleep

Clustered: sleep fragmentation every 30 s, for 30 min every 90 min

On the clustered limb, fragmentation started 1 h after the onset of stage 2 sleep. These two paradigms were designed to present the subjects with equal frequencies of sleep fragmentation but to allow for periods of uninterrupted sleep on the clustered study night.

During fragmentation we varied the duration and volume of tones of 1000 Hz to try to produce a standard microarousal response, i.e. a return to alpha or theta rhythm for longer than 3 s but, where possible, not longer than 15 s on the EEG channels. As reported previously (Martin et al. 1996, 1997b), we did not use any EMG criteria (ASDA 1992) as we had found from pilot studies that an increase in EMG tone was difficult to induce without causing an awakening. If we achieved an arousal or awakening we began the next intertone interval from the re-appearance of stage 2 sleep, defined as the first occurrence of a well defined K-complex or sleep spindle, or the re-appearance of REM sleep. If we did not achieve an arousal response on the first tone, we allowed a 10-s lapse before repeating with a louder and/or longer tone. Lights out on all nights was standardised to 23.00 h and the study time finished at 06.45 h on all nights.

On study nights, sleep was recorded by our standard techniques (Douglas et al. 1992) and manually scored according to standard criteria (Rechtschaffen and Kales 1968) from EEG, EOG and submental electro-myography (EMG). We scored arousals as a return to alpha or theta on the EEG channels for a minimum of 3 s regardless of sleep stage. The arousal frequency consisted of the number of microarousals plus the number of Rechtschaffen and Kales awakenings (Rechtschaffen and Kales 1968), per hour of TST. Arousal frequencies were subdivided into arousals of duration 3–5 s, 5–10 s, 10–15 s, and arousals of more than 15 s that could not be scored as R & K awakenings. Subjects spent the day following each pair of nights undergoing testing of daytime sleepiness, cognitive function and mood.

Daytime assessment

We measured objective daytime sleepiness using the Multiple Sleep Latency Test (MSLT) (Carskadon et al. 1986) with naps at 10.00, 12.00, 14.00, and 16.00 h, and the Maintenance of Wakefulness Test (Poceta et al. 1992) with naps at 10.45, 12.45, 14.45, and 16.45 h. We stopped all naps after one epoch of Rechtschaffen and Kales stage 1 sleep thus preventing subjects obtaining any recuperative sleep, which may have affected their subsequent daytime function. We assessed subjective daytime sleepiness using the Stanford Sleepiness Scale (SSS), (Hoddes et al. 1973) and mood using the UWIST (University of Wales Institute of Science and Technology) mood adjective checklist (Matthews et al. 1990) at 07.00 h and prior to each nap on the MSLT. This checklist scores mood dimensions of Energetic Arousal (EA), Hedonic Tone (HT) and Tense Arousal (TA), and has been described in detail in previous sleep fragmentation studies (Martin et al. 1996, 1997b).

At 07.00 h, prior to the first daytime nap, subjects underwent a battery of performance tests. They were selected to test a broad range of functions: WAIS-R subtests-digit symbol substitution and block design (Lezak 1983), testing general cognition; Trailmaking A & B (Lezak 1983), testing mental flexibility and attention; steer clear, testing vigilance (Findley et al. 1989); and rapid visual information processing (Petrie and Deary 1989) and paced auditory serial addition test (PASAT) at 4 and 2 s (Lezak 1983), testing sustained attention and information processing. The durations of these tests were: Digit symbol 90 s, block design 15 min, Trailmaking A & B approximately 2 min, steer clear 30 min, rapid visual information processing 30 min, PASAT 4 and 2 s approximately 10 min. In addition, we have previously found that some of these tests relate significantly to nocturnal hypoxaemia and sleep fragmentation in SAHS patients (Cheshire et al. 1992) and that they are sensitive to improvements in performance in SAHS patients after CPAP therapy (Engleman et al. 1994). Furthermore, we have used these tests in previous modelling studies on the effects of sleep fragmentation on daytime function (Martin et al. 1996, 1997b).

Statistical analysis

We analysed our data using a mixed two-way analysis of variance (SPSS-PC+) for repeated measures with order of conditions as a between subjects effect. We examined all outcome measures for non-normal distribution and found ceiling effects on the PASAT 4 s-test and the 10.45, 14.45 and 16.45 h naps on the MWT. These variables were subsequently analysed using the Wilcoxon matched pairs test. There was an order effect on the PASAT 2 s-test. This was subsequently analysed, as suggested by Hills and Armitage (1979), using an unpaired t-test on first limb data only. The a priori end points were mean MSLT and mean MWT. However, a post hoc analysis of the MSLT and MWT results at individual time points was carried out by paired t-tests.


Sleep architecture

There was no difference in TST (P=0.4), or the percentage of wakefulness (P=0.9) between the clustered and regular fragmentation study nights (Table 1). There was significantly more stage 2 (P=0.001) and less SWS (P < 0.001) on the regular study night (Table 1).

Table 1.  Comparison of sleep stages and microarousals on clustered and regular fragmentation study nights. All values are mean±SD. Values for stage 1 to REM are expressed a percentage of TST Thumbnail image of

Sleep fragmentation

Subjects received a similar number of tones on both study nights (Table 2) (P=0.9). There was no significant difference in the number of tones inducing arousals (P=0.9) or R & K awakenings (P=0.5) between study nights. The overall arousal frequency was similar on both study nights (P=0.6) (Table 2). There were significantly more arousals of 5–10 s duration (P=0.03) and a trend towards significantly less arousals of 3–5 s duration (P=0.09) on the regular study night. There were no significant differences in arousal frequencies of other durations between study nights (Table 2).

Table 2.  Sleep fragmentation data for both study nights including microarousal and R & K frequencies per hour slept for varying durations of microarousals, and number of tones subjects received Thumbnail image of

Daytime sleepiness

There was no difference between clustered or regular fragmentation in their effects on mean sleep onset latency on the MSLT (clustered 10 SD 5, regular 9±4 min; P=0.7) (Fig. 1) or on the MWT (clustered 32±7, regular 30±7; P=0.2) (Fig. 2). There were no significant differences between study nights on individual naps on the MSLT (all P > 0.2). On the MWT there were no significant differences between study nights on the 10.45, 14.45 or 16.45 h individual naps (all P > 0.2). Subjects had a trend towards a shorter sleep latency after the regular night on the 12.45 h nap on the MWT (34 SD 10, 29±9; P=0.06). There was no difference in subjective sleepiness on the SSS at any time of day after clustered or regular fragmentation (all P > 0.3).

Figure 1.

Paired point data of mean sleep onset latency on the Multiple Sleep Latency Test after both study nights. Single points are mean±SD.

Figure 2.

Paired point data of mean sleep onset latency on the MWT after both study nights. Single points are mean±SD.


Tense arousal at 16.00 h was significantly increased after the clustered night (clustered 13 SD 4, regular 12±3; P=0.03). There were no other significant differences between clustered and regular fragmentation in their effect on mood dimensions at any time of day. EA, all P > 0.1; TA, all P > 0.2 (except at 16.00 h); HT, all P > 0.4.

Cognitive function

There was no significant difference between either fragmentation paradigm in their effects on cognitive function in any of the tests that we used in this study (all P > 0.2) (Table 3).

Table 3.  Data from all cognitive function tests used in this study Thumbnail image of


This study shows that there is no difference between the relative impacts of clustered or regular sleep fragmentation on daytime sleepiness, mood, or cognitive function, even though there was significantly more SWS and less stage 2 sleep on the clustered sleep fragmentation night. The clustered sleep fragmentation night was designed to allow subjects to obtain some uninterrupted sleep and this is reflected in the altered sleep architecture on the regular fragmentation night.

In this study, we were successful in fragmenting our subjects sleep to induce similar numbers of arousals on both study nights. It could be argued that we may have overestimated arousal frequencies during REM sleep owing to our lack of EMG criteria for scoring arousals. However, we made every effort in this study to distinguish between alpha intrusion as a normal part of REM sleep and arousals as a consequence of tones. It may be expected that allowing subjects to have blocks of uninterrupted sleep may lower their acclimatisation thresholds to tones. However, this did not appear to be the case as there was no difference between study nights in the number of tones that were presented to subjects, with 75% of tones causing arousals, and less than 10% of tones causing full awakenings on both study nights. These figures include repeat tones if the initial tone was not successful in inducing the standard response.

There was no difference between study nights in their relative impacts on daytime sleepiness on MSLT or MWT apart from a trend towards subjects being significantly sleepier at 12.45 h on the MWT. This lack of difference in the effects of two types of fragmentation on daytime sleepiness may be due to the performance of MSLT and MWT on the same day, suggesting an interaction between these tests. In defence of this protocol, subjects had a minimum of 25 min between each MSLT and MWT nap where they sat quietly in a bright sitting room. This is in accordance with published guidelines on performance of the MSLT (Carskaddon et al. 1986). It may be argued that an increase in the number of subjects may reveal further significant differences between the two sleep fragmentation paradigms in their relative effects on daytime sleepiness. We therefore performed a power calculation using 4 min as our clinically relevant difference. This was based on results from our previous visible sleep fragmentation study where daytime sleepiness was increased by 4 min after regular sleep fragmentation (Martin et al. 1996). Our calculation indicates that this study had adequate power, 90% power at alpha equals 0.05.

In a similar fashion, there were no differences in mood or cognitive function between either study condition. We found no differences in daytime function between study nights despite finding less stage 2 and more SWS on the clustered fragmentation night. These results suggest that the increased SWS found on the clustered study night is not sufficient to lead to improved scores on the daytime sleepiness, mood and cognitive function tests that we used in this study. This data is in agreement with Bonnet (1986b) who compared induced sleep disruption with induced sleep disruption + elimination of SWS in normal subjects. Performance, mood and daytime sleepiness scores were not affected by the elimination of SWS. The results from Bonnet (1986b) in conjunction with those from the current study suggest that deficits in daytime function caused by sleep fragmentation are not dependent on SWS, and that it is the overall frequency of sleep fragmentation rather than its spacing that determines its effect on daytime function. This is in contrast to the sleep continuity theory as suggested by Downey and Bonnet (1987). This suggests that there may be a minimum period of uninterrupted sleep required for sleep to be restorative of daytime function. These authors (Downey and Bonnet 1987) suggest that this period may be as short as 10 min. The hour long periods of unfragmented sleep on the clustered limb in this study did not, however, lead to improved daytime function. Previous studies using intermittent behavioural awakenings have shown greater performance deficits than more frequent awakenings (Bonnet 1986a), and also no difference in daytime performance if sleep was disturbed 20 min every 30 min, 40 min every 60 min or 80 min every 120 min (Bonnet 1989).

The study could be criticised for not having an undisturbed sleep limb. This was not included in the study design because both we and colleagues have demonstrated that regular fragmentation has marked effects on daytime sleepiness (Philip et al. 1994; Roehrs et al. 1994; Martin et al. 1996) and daytime function (Martin et al. 1996). Indeed, the sleep fragmentation obtained in the regular protocol in the current study was, if anything, greater than that previously shown to cause sleepiness, with tones being generated every 90 s as opposed to every 120 s and a resulting microarousal frequency of 42 SD 5/h as opposed to 34 SD 5/h (Martin et al. 1996). Thus, because severe sleep fragmentation was achieved in the current study, as in the previous studies (Martin et al. 1996; Martin et al. 1997b) in which increased daytime sleepiness was shown to result from sleep fragmentation by the same workers in the same laboratory studying similar subjects, we feel that an undisturbed sleep limb was not required. Having an undisturbed limb would have decreased patient acceptability (requiring a sixth night and 3 day commitment to the laboratory) and decreased the statistical power to detect any difference between the clustered and regular sleep fragmentation limbs, this being the principle question addressed by the current study.

Other potential criticisms of the study include the daytime tests used. We believe these were appropriate as we had previously shown they were sensitive to sleep fragmentation (Martin et al. 1996, Martin et al. 1997b). Similarly, we do not believe that we obtained a negative result due to problems with scoring of the data, as scoring of both sleep and daytime function tests is regularly validated in our laboratory and the scorer in this study has previously published on this area (Martin et al. 1997a). A potentially valid criticism is that this study was not adequately powered to detect the small differences in daytime function (see above). We accept this, but it was adequately powered to detect a 4-min difference in MWT [in comparison to a 10-min difference induced by regular sleep fragmentation (Martin et al. 1996)] and it is arguable whether a lesser difference would be clinically relevant.

In this study daytime sleepiness after regular sleep fragmentation was slightly less severe than that after regular sleep fragmentation in our previous study (Martin et al. 1996), although both studies were conducted under similar laboratory conditions using similar daytime protocols. While we made every effort to use a similar group of subjects to our previous sleep fragmentation study, it may have been that the group of subjects in this study were less sleepy at baseline, although we did not make objective measures at this time. However, it is the within subject, not between subject, comparison that is important and that shows similar effects of clustered and regular fragmentation.

These study results may have relevance for the excessively sleepy patient who has intermittent sleep fragmentation. This may be due to sleep apnoea, which only occurs during REM or light sleep (Lugaresi et al. 1983) or while the patient is lying supine, or perhaps due to periodic limb movements, which normally occur in a clustered fashion during light NREM sleep (Montplaisir et al. 1994). These patients may have periods of uninterrupted sleep, which would consist largely of SWS, during the night. Recently Kass et al. (1996) described a group of these patients. They had mild SAHS (AHI < 15) with a high AHI specific to REM sleep and moderately severe daytime sleepiness. These patients are similar to our experimental design in that, as they had over half their respiratory events during REM sleep, they may have had periods of uninterrupted sleep. Although we should be careful about extrapolating our results from one night of sleep fragmentation to those patients with SAHS whose sleep fragmentation is chronic, evidence from Kribbs et al. (1993) suggests it may be reasonable to do so. They found that after one night off CPAP in patients with SAHS daytime sleepiness returned to pretreatment levels. We did not however, limit our sleep fragmentation to REM sleep only in the clustered limb and therefore a further study investigating the effects on daytime function of sleep fragmentation during REM sleep only, would clarify whether these patients are at risk of suffering from actual deficits in daytime function.

Furthermore, snoring and increased upper airway resistance may also lead to fragmentation of sleep and subsequent daytime sleepiness (Guilleminault et al. 1991, 1993). Patients who have respiratory events in a clustered fashion during REM sleep may snore or have increased upper airway resistance during NREM sleep. Series et al. (1994) have suggested that sleep fragmentation may contribute to increased upper airway collapsibility during sleep. They found that sleep fragmentation increases upper airway collapsibility during recovery sleep compared to sleep deprivation. This data suggests that respiratory events during REM sleep alone may lead to increased upper airway collapsibility during subsequent sleep stages. Although this suggests that patients with sleep fragmentation found in clusters may have increased upper airway collapsibility leading to increased arousal frequencies during periods of uninterrupted sleep, further studies are required to characterise whether there are actual deficits in daytime function in this group of patients.

In conclusion, although we have delivered sleep fragmentation to produce similar arousal frequencies on two study limbs but different amounts of SWS, this had no differential effect on daytime function. This suggests that it may be the overall number of arousals from sleep, rather than their placement, which determines subsequent daytime function.


Pirkko E. Brander was supported by funding from the Finnish Anti–Tuberculosis Association Foundation. Supported by grant no. K/MRS/50/C2151 from Chief Scientist, Scottish Office Home and Health Department.

Accepted in revised form 10 June 1999; received 29 September 1998