958
BIOL PSYCHIATRY 1992;32:958-975
Good Sleep, ad Sleep: A Meta-Analysisof PolysomnographicMeasuresin Insomnia, Depression,and Narcolepsy James I, Hudson, Harrison G. Pope, Lynn E. Sullivan, Christine M. Waternaux, Paul E. Keck, and Roger J. Broughton
Primary insomnia, major depression, and narcolepsy are usuallyconsidered to be sep arate disorders, distinguishedby dtrerent polysomnographicprofiles. But do polysomnographic data provide adequate evidence to segregate the three disorders, ot might they displuyfundamentallythe same sleep disturbance, di$ering only in degree? To test the viabilityof these two alternate hypotheses, the authors performed a meta-analysisof controlledpolysomnographicstudies of these disorders. A summarymeasure of degree of sleep disturbance was constructedfrom five variables: wakefulnessa#er sleep onset, percentage of stage I sleep, percentage of stage 3 + 4 sleep, rapid eye movement(REM) latency, and REM density. The results of availablestudiesfor each variable were combined using a weightedaverage of eflect sizes. An overall “sleep disturbance index” was then calculated by combining the estimatesfor the jve above listed variables. On both the individualmeasuresand especiallyon the summaryindex, insomnia, depression, Cmn narcolepsy were arrayed on a simple continuum of progressively more severe sleep disturbanoe-=-congruent with the clinical observation that these disorders display progressivelymore disturbedsleep. Thes, findings suggest that sleep can be disturbed in only a limitednumber of ways: in evaluatingsleep architecture, it may not be possible to elaboratemuch beyond a single aais of good-to-badsleep. Thus, polysomnographicmeasures may not provide adequate evidence to classi&insomnia, depression, and narcolepsy as separate entities.
Introduction Theadventof polysomnographic techniques(or as they shouldmoreaccuratelybe termed, “somnopolygraphic” techniques)has provideda new tool to evaluatedifferencesamong disordersassociatedwith disturbedsleep. Threeof the best studiedof these disordersare primaryinsomnia,majordepression,and narcolepsy. From the Clinical
Neurophysiology Laboratory (JIH, PEK, IS), the Biological PsychiatryLaboratory(JIH, HGP), and the BiostatisticsLaboratory (CMW), Laboratoriesfor PsychiatricResearch, McLeanHospital,Belmont, Massachusetts,and Departmentof Psychiatry,HarvardMedical School; and Division of Neurology. OttawaGeneralHospital and University of Ottawa, Ottawa, Ontario(RIB). Addresscorrespondenceandreprintrequeststo Dr. Hudson,Laboratoriesfor PsychiatricResearch,McLeanHospital, Belmont, Massachusetts02178. ReceivedMarch21,1992; revisedJune 25, 1992.
8 1992 Society of Biological Psychiatry
0006-3223/92/$05.00
Sleepin Insomnia, Depression, andNarcolepsy
BIOL PSYCHIATRY 1992;32:958-975
959
These disorders obviously differ strikingly in their clinical presentation.Primaryinsomnia (also termed psychophysiologic insomnia or idiopathic insomnia)is characterized exclusively by chronic difficulty in initiating or maintainingsleep. Majordepressionis characterizedby depressed mood, anhedonia, and otherpsychologicalandneurovegetative symptoms associated with disturbed sleep. Narcolepsy is characterized by excessive daytime sleepiness and sleep attacks, combined with accessorysymptomssuch as cataplexy, sleep paralysis, and vivid hypnagogic hallucinations. It is also widely maintained that the three disorders show markedly different polysomnographic features. Primary insomnia displays disturbed sleep continuity, butnormaltime from sleep onset to the first rapid eye movement (REM) sleep period (hereafter termed “REM latency”) and REM density (Frankel et al 1976; Reynolds et al 1984). Major depression displays disturbed sleep continuity, decreased stage 3 + 4 sleep, shortened REM latency, and increased REM density (Gillin et al 1984). And narcolepsy displays decreased sleep latency, increased stage-l sleep, shortened REM latency, and increased REM density on all-night studies (Reynolds et al 1983; Broughton et al 1988), combined with shortened sleep latencies and increased number of sleep-onset REM periods during daytime nap studies (multiple sleep latency tests) (Richardson et al 1978). Two hypotheses might be considered to explain the clinical and polysomnographic differences between insomnia, depression, and narcolepsy: (1) they represent fundamentally different disorders characterized by three qualitatively different clinical and polysomnographicprofiles, or (2) they represent fundamentallysimilar disorderswhose clinical and polysomnographic features differ primarily in their degree of severity, rather than qualitatively. Most of the literature favors the former view-that the three disorders are fundamentally different. But certain clinical similarities among the disorders weigh against this view. First,complaintsof insomniaand daytime sleepiness are common not only in insomnia, but also in depression and narcolepsy (Mitchell and Dement 1968;Montplaisir et al 1978; Gillin et al 1979; Reynolds et al 1982; Zorick et al 1986). Second, affective symptoms are present not only in depression, but commonly in insomnia and narcolepsy as well (Broughton et al 1981; Reynolds et al 1983;Tan et al 1984; Vollrath et al 1989). Third, disturbednocturnal sleep, in varying degrees, is a feature of all three disorders (Montplaisir et al 1978; Gillin et al 1979; Reynolds et al 1983, 1984; Broughton et al 1988). Given these clinical similaritiesamongthe three disorders, do the polysomnographicdata nevertheless providesufficientevidenceof differencesamongthemthat they should be classified as separate entities? One way to look at this questionis to examinewhetherthe various polysomnographic abnormalities in these disordersare independent,or whetherthey mightbe accommodated by a simpler model of overall sleep disturbance.To this end, we performed a metaanalysis of polysomnographicmeasuresin insomnia,depression, and narcolepsy.
Methods Selection of Studies To create a database for our meta-analysis, we searched the medical literature for all polysomnographic studies of insomnia, depression, and narcolepsy dating from the introduction of this technique. We used three sources to identify studies: (1) the MEDLINE
960
BIOL PSYCHIATRY 1992;32:958-975
J.I. Hudsonet al
database from January 1983 to May 1992; (2) additional studies cited in the sports ted by &e MEDLINE search; and (3) citations from previous meta-analyses of mnographic studies(Knowles and MacLean 1990; Obermeyer et al 1991). TO be d, studies were required to adopt or closely parallel established diagnostic criteria. Specifically, for insomnia, all studies used the criteria for psychophysiologic insomnia of the Associationof Sleep Disorders Centers ( 1979), except for three studies (Karacan et al 1971; Frankel et al 1976; Seidel et al 1984) that used criteria closely approximating the above. For depression, a major depressive episode was defined by Research Diagnostic Criteria (Spitzer et al 1978) in all studies except one (JI Hudson et al unpublished) that used DSM-III-R criteria (American Psychiatric Association 1987). For our analysis, studies of depression were permitted to include patients with unipolar and bipolar depression, as well as those with and without psychotic features. Other subgroupsof depression, however, such as atypical depression (Quitkin et al 1985) and anergic depression (Thase et al 1989), were excluded, because criteria for these entities are not standardized and often vary significantly from those of major depression. For narcolepsy, all studies used criteriavery similar to those of the Association of Sleep Disorders Centers (1979). In addition, it was required that studies of narcolepsy exclude patients with clinically significant sleep apnea; studies were not required to exclude patients with periodic leg movements during sleep, since such activity is a characteristic feature of narcolepsy (van den Hoed et al 1981; Wittig et al 1983; Baker et al 1987; Bedard et al 1989). Studies meeting the above standards for diagnosis were also required to meet the following criteria: (1) controlled comparison of disorder of interest with either normal control subjects or one of the two other disorders under review; (2) mean age of subject groups over 20 years (because narcolepsy and insomnia have been studied in adults only); (3) age-matched samples, with a difference of at most 10 years in mean age between subject groups; (4) non-selective subject recruitment for each diagnostic group (e.g., studies of patients with depression plus memory loss (Vitiello et al 1990), or insomnia plus decreased sleep efficiency or increased sleep latency (Hauri and Fisher 1986) were excluded; (5) subjects unmedicated at the time of study; (6) all-night polysomnographic studies only (as there are insufficient data to review for other procedures, such as the multiple sleep latency test); (7) use of standard visual scoring procedures; and (8) report of mean and stand deviation (SD) of at least one polysomnographic variable of interest. We also excluded instances in which data from the same subjects were presented in two separate reports. When two or more studies of the same diagnostic group appeared from the same center with similarly aged subjects, we included only the study with the largest sample size, unless we were able to confirm from the authors that no overlap in data had occurred. This selection procedure yielded eight studies of insomnia, 13 studies of depression, and nine studies of narcolepsy. The demographic characteristics, number of nights of polysomnography , as well as mean and SD for polysomnographic variables of interest, for each of these studies are presented in Table 1. The mean ages for patients in the pooled studies were 37.9 years for insomnia, 39.1 years for depression, and 42.5 years for narcolepsy. For all but eight of the studies, subjects were unmedicated for at least 2 weeks prior to evaluation. Five studies useda l-week medication-freeinterval(Richardson et al 1978; Wittig et al 1983; Broughton et al 1988; Lauer et al 1991; Mendlewicz and Kerkhofs 1991); two studies included a small number of patients with a 2-S day medication-free interval (Hishikawa et al 1976; Bedard et al 1989); and one study did not
Sleep in Insomnia,Depression,and Narcolepsy
BIOL PSYCHIATRY 1992;32:958-975
961
specify the medication-free interval (Schneider-Helmert 1987). For studies that presented data from individual nights separately, we calculated the mean and combined SD of all nights for our analysis.
Constructionof the Sleep Disturbance Index TO determine whether the
three disorders were associated with differing degrees of disturbed sleep, we attempted to devise a summary measure of degree of sleep disturbance that would be composed of several variables, derived from visual scoring of records, commonly reported in the polysomnographic literature on insomnia, depression, and narcolepsy. Five variables seemed to best summarizeoverall quality of sleep: wakefulness aftersleeponset(WAS@(minutes of awake time after sleep onset); percentage of stageI sleep(expressed as percentage of the total sleep time); percentage of stage 3 + 4 deep (as percentage of the total sleep time); REM latency(time between onset of sleep and onset of the first REM sleep period); and REMdensity(amount of eye-movement activity per min of REM sleep). The first three of these variables comprehensively and nonredundantly measure the presence of wakefulness and the depth of nonREM sleep: specifically, increased wakefulness after sleep onset, increased stage 1 sleep (“light” sleep), and decreased stage 3 + 4 sleep (“deep sleep”) each represent poorer quality of sleep. Amounts of the other two sleep stages, stage 2 and stage REM, were not included, because changes in these variables have not been associated with poor quality of sleep. Stages 3 and 4 were necessarily considered together because many of the available studies did not report data on these stages separately. The other two variables, REM latency and REM density, were chosen because shortened REM latency and increased REM density are reported in several disorders to be associated with poor sleep, particularly depression, and because they are nonredundant with the three measures of sleep stages. Thus, it seemed likely that they would add information on overall quality of sleep beyond that supplied by sleep stage measures. Furthermore, this choice of variables for our summary index has construct validity, in that all of these measures (with the exception of REM density, which will be discussed later) worsen with age (Feinberg 1974;Ehlers and Kupfer 1989; Knowles and MacLean 1990; Lauer et al 1991), as does subjective sleep quality. We combined the resultsof polysomnographicstudiesfor each variableusing a weighted average of effect sizes, and then we derived an overall “sleep disturbance index” (SDI) by combining the estimates for effect sizes for the five above variables listed. Details of the statistical method for this analysis are provided in the Appendix. Although these five variables seemed to provide the simplest a priorisummary model of sleep quality, we considered several alternate choices of variables. We will return to a consideration of the merits of these alternate models, and their effect on our calculations, in the discussion.
Results The weighted averages of effect sizes, 6*‘s, calculated for each of the five polysomnographic variables contributing to the SD1 in each of the three disorders, are presented in Figure 1. For all variables considered, the h*‘s for insomnia were found to be either statistically similar to or higher than normal individuals; those for depression were either similar to or higher than insomnia; and those for narcolepsy were either similar to or
Karacanet al (1971) Frankel et al (1976) Coates et al (1982) Mendelson et al (1984) Seidel et al (1984) Mendelson et al (1986) SchneiderHelmert (1987) Reynolds et al (1984) Gillin et al (1979)’ Quitkin et al (1985) Reynolds et al (1985) Duhe et al (1986) de Maertelaeret al (1987) Mendelson et al (1987) Latteret al (1991)’
Study 10 10 18 lgg 12 12 10 10 38 38 10 10 16 16 10 2ok 74’ 41 19 21 2s 2s 52 23 19 9 8 8 74 51
Insomnia Depression Depression Normal Depressiot# Normal Depression Normal Depression Normal Depression Normal Depression Nomal Depression Normal
N
Insomnia Normal Insomnia Normal Insomnia Normal Insomnia Normal Insomnia Normal Insomnia Normal Insomnia Normal
Diagnosis
43.8 39.9 43.0 44.0 40.4 35.1 69.9 69.0 40.2 36.9 41.4 35.0 43.9 43.9 39.0 38.3
37.5’ 37.5’ 44.5 45.1 41 .r 41.7’ 35.3 35.4 29 29 39.5 40.1 46.1 46.1
Mean age 2-3d.e
Id Id 3-tjd 3-sd 2-3 2-3 2d 2d Id Id 3d 3d Id Id Id Id
4d 4d 3d 36 2d 2d Id Id 2d 2d Id Id
2-3d.e
(51.5) (25.8)
(53.7) (32.8)
(78.0) (31.0) (37.3) (25.8) (67.6) (26.1)
77.4 33.7
116.8 47.4
91.6 41.0 26.3 21.3 64.2 26.7
(22. I) (20. I) (49.4) (42.8)
23.6 23.5 122.8 63. I (37.5)
(21.8) (17.3) (24.0) (19.0)
30.1 19.9 47.9 25.9
49.4
(34.2)
49.5
WA!@’ (min)
It).0 IS.5 16.1 9.7 4.7 3.2 21.4 19.0 5.8 8.5 9.4 7.7
9.1
10.1 6.6 4.5 2.9 9.0 7.1 9.2 6.2 8.1 6.3
(8.7) (4.1) (10.7) (8.2) (4.1) (2.4) (10.9) (6.0) (5.7) (11.1) (4.9) (4.8)
(4.5)
(5.8) (2.7) (3.0) (2.4) (5.0) (3.0) (7.1) (3.9) (4.4) (2.5)
Percentage stage 1
4.7 6.1 2.1 2.5 10.5 18.1 10.3 14.0 7.5 11.8 11.0 11.2
2.8
9.1 10.4 17.6 26.2
2.0 3.1 17.3 15.9
7.4 12.3 2.5
(4.3) (5.3) (4.1) (3.9) (9.3) (7.4) (7.9) (9.4) (5.9) (5.8) (8.9) (8.9)
(4.6)
(1.9) (1.8) (7.8) (7.3)
(3.2) (3.0) (7.3) (4.2)
(10.8) (8.7) (3.9)
Percentage stage 3 + 4
47.8 78.1 53.9 92.8 33.3 57.6 55.8 90.1 56.9 113.0 86.5 63.3 54.3 70.9
82.9
77.4 90.7
86.7 84.5
81.6 86.0 68.6
(29.5) (26.4) (40.1) (33.6) (36.5) (16.9) (36.5) (33.5) (32.8) (S2.2) (63.6) (28.3) (32.1) (23.9)
(28.0)
(27.0) (45.9)
(40.0) (19.9)
(37.0) (29.0) (16.6)
REM latency (min)
Mean (SD) values for polysomnographic variables”
Measures in Studies Used for
Nights of sleep
Table I. Clinical Characteristics and Polysomnographic
(0.29) (0.76) (0.59) (1.47) (1.23) (0.67) (0.48)
(0.83) (0.25) (0.51) (0.72) (12.3) (9.7)
2.01 1.73 4.53 3.45 1.57 1.35
1.44 0.47 1.38 1.61 41.7 27.0
(0.41) (0.38)
I.44 1.30
1.07
(0.43) (0.33)
(0.39)r
1.39 1.38
1.s3
REM density (units/min)
k
z
E CD
k E
4
2
2-4 2-4 2 2 1 1 2-3 2-3 2 2 1 1 2 2 3d 3d 1 1 2 2
25.8 24.5 42.2 42.8 28.2 28.2 44.3 43.6 44.6 44.1 44.7 49.3 39.4 39.4 47.5’ 47.5’ 45.3 45.6 43.6 44.5
30 19 25 25 20” IO 20 10 8 8
Depression Normal Depression Narcolepsy Narcolepsy Normal Narcolepsy Normal Narcolepsy Normal N yolepsy Normal Narcolepsy Normal Narcolepsy Normal Narcolepsy Normal Narcolepsy Normal 20 II 11 I7 17 IO 10 I29 6
549
2 2 2 2 Id Id Id ld
24.4 24.4 30.8 29.3 35.4 32.5 42.5 41.4
8 8 60 20 139 47 67 66
Depression Normal Depression Normal Depression Normal Depression Normal (22.7) (8.2) (30.4) (46.6)
(42.9) (31.3) (38.2) (29.6) (34.9) (41.6)
(21 .O) (36.9) (6.5) (3.9)
56.7 39.1 97.7 62.3 55.2 52.3
56.0 36.2 y.8 5.2
(33.9) (20.8)
23.3 16.0
22.5 9.6 31 .s 60.1
(14.1) (7.1)
12.0 7.5
(5.2) (2.4) (4.8) (8.9) (7.2) (6.1) (7.6)” (3.3)”
(7.1) (4.9) (6.3) (5.8)
(7.1) (7.1) (3.3) (3.4)
18.1 14.3 IS.9 13.6
12.3 8.9 21.1 10.8
(3.0) (2.6) (8.8) (4.8) (4.6) (4.5) (8.7) (6.4)
7.3 3.7 7.9 16.2 25.2 13.5 17.9 8.0
4.5 4.0 10.2 10.0 6.7 8.4 9.6 6.8
20.0 14.8 11.2 9.3
9.6 7.2
15.0 13.0 2.2 3.0 13.6 13.7 10.2 14.2
12.5 20.0 68.5 83.6 8.3 7.9 14.7 15.0
(8.0) (5.1) (3.3) (7.5)
(6.5) (6.8)
(13.2) (8.3) (4.7) (7.5) (7.5) (4.7) (7.2)p (7.2)p
(8.0) (10.1) (23.8)” (lS.3)“’ (8.1) (8.8) (10.6) (8.2)
17.2 61.7 48.2 85.5
15.9 86.3
41.1 81.1 68.4 79.6
58.7 85.9 47.4 25.6
71 .O 74.5 1.90 1.30 61.2 71.3 69.1 80.6
(33.0) (17.8) (28.9) (37.2)
(29.6) (33.7)
(37.5) (42.6) (61 .O) (20.9)
(25.6) (21.9) (212) (31.2)
(50.6) (22.5) (0.50) (0.30) (30.5) (21.6) (48.8) (30.3) (0.43) (0.20) (0.46) (0.57)
(0.42) (0.43) (1.69) (1.65)
(0.55) (0.51)
5.05 (I.60)’ 4.03 (2.25)
1.25 0.92 1.3s 1.61
1.29 1.07 4.40 2.80
1.s5 1.30
OWhenvalues for more than one night of sleep were reported in an individual study, the mean and SD for all were calculated and used for analysis. 6Wakefulnessafter sleep onset. ‘Midpoint of age range (mean age not reported). dPolysomnographyperformed after 1-2 nights of t 1 vdws werenot used, since the variable number of subsequentnightsprecludedaccuratecombiningof data. ‘Value reJrottedfor this group reportedby Feinberget al was not included in analysis, becauseof presumed overlap with Gillin et al (1979). *JIepressiongroup was not included in analysis, becauseof presumedoverlap with ReynoJdset 3). ‘Combined 56 patientsin initial study and 18 patientsin replicationstudy.‘Data for 56 of 74 patientsreportedin tabular form in Feinberget al (1982). ‘Patientswith e ~sion inchtded, but patientswith atypical depression excluded.‘Data for WAS0 from Latter et al (1988). “Data from Giles et al (1990). which reporteda larger sample(177 and 54 healtby individuals)from this centerREM latency only. “Patientswith major depressionor bipolar disorder, depressed.by DSM-III-R criteria; studiedaccording bed in Hudsonet al (1988); data reportedin part in Hudsonet al (1992). “Narcolepticpatientswith only sleep attacksnot included. becausethey would not meet trite in this analysis(see text). PMean values reported and total sleep time, qJncJ&s patients wih and for one night for this variable in Montplaisir (1976); since SD was not reported,it was estimatedfrom the SD of without periodic leg movements(see text). ‘Includes unpublisheddata from this study (Browman CP; prsonal communication, 1991). Data calculatedfrom figure 28 of Uris study,
Kupfer et al (1989) Wailer et al (1989) Buysse et al (1990) Mendlewicz and Kerhofs (1991) Hudson et al (unpublished)” Reynolds et al (1983) Hishikawa et al (1976) Montplaisir ct al (1978) Rtchardsonet al (1978) Wittig et al (1983) Browman et al (1986) Geisler et al (1987) Broughton et al (1983) Bedardet al (1989)
L z !Y
964
J.I. Hudson et al
BIOLPSYCHIATRY 1992;32:95&975
A. WAKEFULNESSAFTERSLEEPONSET
Insomnb
B. PERCEN
Narcolepsy
E OF STAGE1 SLEEP
Insomnia Figure
Depressbn
Depressian
Natcolepsy
1, s*‘s (mean I SEM) for diagnostic groups for the following polysomnographicmeasures: (A) wakefulness after sleep onset; (B) percentage of stage 1 sleep; (C) percentage of stage 3 + 4 sleep; (D) REM latency; (E) REM density. Significanceof differences: (A) insomnia versus normal. individuals,p < 0.001; depression versus normal individuals,p < 0.001; narcolepsy versus normal individuals, p = 0.001; (B) insomnia versus normal individuals, p = 0.003; insomnia versus narcolepsy, p = 0.032; depression versus narcolepsy, p = 0.003; narcolepsy versus normal individuals, p < 0.001; (C) insomnia versus normal individuals, p = 0.002; depression versus norm,il individuals, p = 0.019; (D) insomnia versus depression, p = 0.002; insomnia versus narcolepsy,p -C 0.001; depression versus normal individuals, p < 0.001; narcolepsy versus normal individuals, p < O.OOl; (E) insomnia versus depression, p = 0.001; insomnia versus narcolepsy, p = 0.029; depression versus normal individuals,p < 0.001; narcolepsy versus normal individuals, P = 0.016. All other differences nonsignificant (p > 0.05).
Sleepin Insomnia, Depression, and Narcolepsy
BIOL PSYCHIATRY 1992;32:958-975
965
C. PERCENTAGEOF STAGE3+4 SLEEP 0.8
r
0.6
0.4
0.2
0.0
Insomnia
Depressbn
Narcolepsy
Figure 1. Continued
D. REM LATENCY I.U
1.2 ’.
G +I x
o*8
0.4 . 0.0
Insomnia
Depressbn
Narcolepsy
higher than depression. Combining these variables to derive the SD1yielded an upward slope (Figure 2), in which each disorder showed an SD1 that was higher (to a degree either exceeding or approaching statistical significance) than its predecessor.
Discussion We constructed a “sleep disturbance index” based on five commonly measured polysomnographic variables, to provide a summary description of quality of sleep. Calculating the SD1for three widely studied disorders-insomnia, depression, and narcolepsy-using a statistical meta-analytic approach, we found these disorders to be arrayed along a continuum of progressively more disturbed sleep.
966
J.I. Hudsonet al
BIOL PSYCHIATRY 1992;32:958-975
E. REM DENSITY 0.8
I
0.6
Insomnia
Depression
I
Figure1. Continued
Naroolepsy
Before considering further the implications of these findings, several criticisms of our method should be considered. These criticisms fall into two groups: choice of inclusion criteria for studies, and choice of variables for the SDI. First, with regard to inclusion criteria for studies, it might be argued that greater precision could have been obtained by narrowing the analysis to a specific age group throughout, since age has such a strong effect on polyf;omnographicfindings (Knowles and MacLean 1990). For example, one might restrict analysis to studies with a mean age of 35-45 years, as this would include most of the studies of insomnia (seven of eight), depression (eight of 13), and narcolepsy (eight of nine). We decided against this approach, and included all studies, because the mean ages of the groups were similar, even though
Figure2. Sleep disturbanceindex ( f: SEM) for diagnostic groups. Significance of differences: insomnia versus normal individuals, p < 0.001; insomnia versus depression, P = 0.062; insomnia versus narcolepsy, p < 0.008; depression versus normal individuals, p < 0.001; depression versus narcolepsy, p = 0.076; narcolepsy versus normal individuals, p < 0.001.
Insomnia
Depression
Narcolepsy
Sleep in Insomnia,Depression,and Narcolepsy
BIOLPSYCHIATRY 1992;32:958-975
969
nonREM sleep, and REM sleep. If so, the SD1 would reflect several unrelated aspects of sleep disturbance, rather than overall sleep disturbance. Although this possibility cannot be excluded, three lines of evidence suggest that the SD1does reflectoverallsleep quality. First, all of the component measures in the SD1worsen with increased arousal. Increased arousal is associated with a shift towards more wakefulness, increased stage-l sleep, and decreased stage 3 + 4 sleep. Because REM latency represents the length of the first NREM period, factors that decrease NREM sleep (particularly slow-wave sleep), such as increased arousal, may thereby shorten REM latency (Feinberg et al 1988; Ehlers and Kupfer 1989). Finally, increased arousal is associated with increased REM density (Feinberg et al 1987, 1988).
Second, the SD1 worsens with age, as sleep becomes lighter. All of the measures in the SDI, with the exception of REM density, worsen with age (Feinberg 1974; El&s and Kupfer 1989; Knowles and MacLean 1990; Lauer et al 1991). However, even the changes in REM density with age may be consistent with the effects of the age-related lightening of sleep. Ehlers and Kupfer ( 1989)found that middle-aged individuals displayed increased REM density compared with younger individuals, but elderly individuals had REM density values similar to young individuals. But elderly individuals also displayed increased time awake in the last one-third of the night, a time dominated by REM sleep with high REM density in young and middle-aged individuals. Thus, they have proposed that the failure to find a progressively increased REM density with age may be a result of gradual replacement of high-density REM sleep with awakening due to increased levels of arousal. Third, depressed individuals show a greater change with age than do normal individuals in most the component measuresof the SDI, accordingto many (Gillin et al 1981; Knowles and MacLean 1990; Lauer et al 1991), but not all (Obermeyer et al 1991) studies. In other words, the sleep disturbance in depression may be analogous to “accelerated aging” (Gillin et al 1981). Hence, for at least one of the disorders under study, the data are consistent with the possibility that a unitary process, similar to that involved in normal aging, disrupts most of the measures in the SDI. Thus, although it is possible that the differentcomponentmeasuresof the SD1 assess independent systems, the above three observationssuggest thatthese measures are not independent and that the SD1may assess a general tendency toward more disturbed sleep. In summary, therefore, the findings generated using the SD1 are consistent with a “unitary” hypothesis of sleep disturbance in insomnia, depression, and narcolepsy: that these three disorders are positioned along a single axis of progressively more disturbed sleep. Admittedly, this analysis does not disprove the currently more popular “pluralistic” hypothesis:that the sleep disturbancesin insomnia,depression,and narcolepsyare fundamentallydifferentand are associatedwith pathognomonic polysomnographic features. However, attempts to segregate the disorders on the basis of polysomnographic features requiremorecomplicated assumptions. For example, one must assumethatREMlatency andREMdensityaremeaningful to distinguish narcolepsy and depression frominsOda, but not depression from narcolepsy, and that sleep efficiency and sleep latency are meaningful for distinguishingnarcolepsyfrominsomniaanddepression,butnot insomnia fromdepression. The unitary hypothesis, by comparison, explains the polysomnographic findingsin these disordersmoreparsimoniously. Stateddifferently, it appearsthat sleep can be affected in only a limitednumberof ways-and that present dataare insufficientto permitelaborationmuchbeyonda single axis of god sleep to bad sleep. If so, polysornnographic measures cannot be used as
970
J.I. Hudsonet al
“biOL PSYCHIATRY 1992;32:958-975
evidence for classifying insomnia, depression, and narcolepsy as separate entities. Indeed, the possibility remains open that the three disorders are closely related, as perhaps implied by their frequently shared phenomenologic fentures and treatment response (Tan et al 1984; Jacobs et al 1988; Hudson and Pope 1990).
Appendix For each polysomnographic variable for each study, we computed the unbiased estimate for the effect size using the method proposed by Hedges and Olkin (1985, p 81)
>
(X, - X,)/SD,,
Cl1
where N = 11~+ n, is the sum of the sample sizes for control and experimental groups, respectively; X, is the mean value of the normal control group; X, is the mean value of the experimental group, and SD, is the pooled standard deviation for the normal control group and the experimental group. For the two studies (Reynolds et al 1983, 1984) that did not have normal control groups, we used data from an age-matched normal control group studied at the same center as these two studies, but reported separately (Thase et al 1989). Note that for polysomnographic variables in which a larger value represented better quality sleep (sleep efficiency, stage 3 + 4 sleep, REM latency), we multiplied the effect size by - 1 so that a larger positive value of ES always represented poorer qualit! of sleep. Before pooling the estimates of the effect size for a series of k studies, it is important to determine whcthcr the studies cstimute a common effect size 6. The assumption that all of the studies have the same constant but unknown effect size was tested for each polysomnographic variable using the x2 statistic for homogeneity with k-l degrees of freedom, k
Qw=
Wi
(ESi - S)‘,
M
where 8 is the weighted estimator under the assumption of a common effect size
133
and the weights wi are the reciprocal of the estimated variance of the effect size ESi for the ith study (Hedges and Olkin 1986, p 86). When the hypothesis of homogeneity is rejected, the variability among studies is greater than expected under the simple model of random deviations from a common effect size, and the pooled estimated in [3] is not appropriate. We used random effects models to account for the unexplained variability in the observed effect sizes (National Research
Sleep in Insomnia, Depression, and Narcolepsy
BIOL PSYCHIATRY 1992;32:958-975
971
Council 1992). The average true effect is estimated using a weighted average of the effect sizes
8* =
(i
WTEsJ/(g wf),
HI
where the revised weights w* are the reciprocal of the variance of the effect sizes under the random effects model and are given by WT
=
(wj-’
+
ep-’ ,
PI
and T*is the method-of-mome$ estimate of the variance component (DerSimonian and Laird 1986). The variance of 6* is estimated by v&b” =
w1
To derive the overall sleep disturbance index (SDI), we computed the $*‘s for five polysomnographic variables: wake after sleep onset (6&,), percentage of stage 1 sleep (&), percentage of stage 3 + 4 sleep (@s4), REM latency (&), and REM density (&id). We assigned each variable equal weight, as we lacked justification Q priori for differential weighting. Thus, the SD1 for each diagnostic group represented the mean of the $*‘s of the five variables:
and the standard error of SD1was estimated using the square root of the variance of SDI. Using the formula for computing the variance of a sum of correlated components, the estimate of the variance of SD1 is equal to
viirSD1 = (1125) 5
5
i=l
i=l
j-l
(V&
8:)“’
(Vb
ST)“’
rijg
PI
j
BIOL PSYCHIATRY 1992;32:958-975
Good Sleep, ad Sleep: A Meta-Analysisof PolysomnographicMeasuresin Insomnia, Depression,and Narcolepsy James I, Hudson, Harrison G. Pope, Lynn E. Sullivan, Christine M. Waternaux, Paul E. Keck, and Roger J. Broughton
Primary insomnia, major depression, and narcolepsy are usuallyconsidered to be sep arate disorders, distinguishedby dtrerent polysomnographicprofiles. But do polysomnographic data provide adequate evidence to segregate the three disorders, ot might they displuyfundamentallythe same sleep disturbance, di$ering only in degree? To test the viabilityof these two alternate hypotheses, the authors performed a meta-analysisof controlledpolysomnographicstudies of these disorders. A summarymeasure of degree of sleep disturbance was constructedfrom five variables: wakefulnessa#er sleep onset, percentage of stage I sleep, percentage of stage 3 + 4 sleep, rapid eye movement(REM) latency, and REM density. The results of availablestudiesfor each variable were combined using a weightedaverage of eflect sizes. An overall “sleep disturbance index” was then calculated by combining the estimatesfor the jve above listed variables. On both the individualmeasuresand especiallyon the summaryindex, insomnia, depression, Cmn narcolepsy were arrayed on a simple continuum of progressively more severe sleep disturbanoe-=-congruent with the clinical observation that these disorders display progressivelymore disturbedsleep. Thes, findings suggest that sleep can be disturbed in only a limitednumber of ways: in evaluatingsleep architecture, it may not be possible to elaboratemuch beyond a single aais of good-to-badsleep. Thus, polysomnographicmeasures may not provide adequate evidence to classi&insomnia, depression, and narcolepsy as separate entities.
Introduction Theadventof polysomnographic techniques(or as they shouldmoreaccuratelybe termed, “somnopolygraphic” techniques)has provideda new tool to evaluatedifferencesamong disordersassociatedwith disturbedsleep. Threeof the best studiedof these disordersare primaryinsomnia,majordepression,and narcolepsy. From the Clinical
Neurophysiology Laboratory (JIH, PEK, IS), the Biological PsychiatryLaboratory(JIH, HGP), and the BiostatisticsLaboratory (CMW), Laboratoriesfor PsychiatricResearch, McLeanHospital,Belmont, Massachusetts,and Departmentof Psychiatry,HarvardMedical School; and Division of Neurology. OttawaGeneralHospital and University of Ottawa, Ottawa, Ontario(RIB). Addresscorrespondenceandreprintrequeststo Dr. Hudson,Laboratoriesfor PsychiatricResearch,McLeanHospital, Belmont, Massachusetts02178. ReceivedMarch21,1992; revisedJune 25, 1992.
8 1992 Society of Biological Psychiatry
0006-3223/92/$05.00
Sleepin Insomnia, Depression, andNarcolepsy
BIOL PSYCHIATRY 1992;32:958-975
959
These disorders obviously differ strikingly in their clinical presentation.Primaryinsomnia (also termed psychophysiologic insomnia or idiopathic insomnia)is characterized exclusively by chronic difficulty in initiating or maintainingsleep. Majordepressionis characterizedby depressed mood, anhedonia, and otherpsychologicalandneurovegetative symptoms associated with disturbed sleep. Narcolepsy is characterized by excessive daytime sleepiness and sleep attacks, combined with accessorysymptomssuch as cataplexy, sleep paralysis, and vivid hypnagogic hallucinations. It is also widely maintained that the three disorders show markedly different polysomnographic features. Primary insomnia displays disturbed sleep continuity, butnormaltime from sleep onset to the first rapid eye movement (REM) sleep period (hereafter termed “REM latency”) and REM density (Frankel et al 1976; Reynolds et al 1984). Major depression displays disturbed sleep continuity, decreased stage 3 + 4 sleep, shortened REM latency, and increased REM density (Gillin et al 1984). And narcolepsy displays decreased sleep latency, increased stage-l sleep, shortened REM latency, and increased REM density on all-night studies (Reynolds et al 1983; Broughton et al 1988), combined with shortened sleep latencies and increased number of sleep-onset REM periods during daytime nap studies (multiple sleep latency tests) (Richardson et al 1978). Two hypotheses might be considered to explain the clinical and polysomnographic differences between insomnia, depression, and narcolepsy: (1) they represent fundamentally different disorders characterized by three qualitatively different clinical and polysomnographicprofiles, or (2) they represent fundamentallysimilar disorderswhose clinical and polysomnographic features differ primarily in their degree of severity, rather than qualitatively. Most of the literature favors the former view-that the three disorders are fundamentally different. But certain clinical similarities among the disorders weigh against this view. First,complaintsof insomniaand daytime sleepiness are common not only in insomnia, but also in depression and narcolepsy (Mitchell and Dement 1968;Montplaisir et al 1978; Gillin et al 1979; Reynolds et al 1982; Zorick et al 1986). Second, affective symptoms are present not only in depression, but commonly in insomnia and narcolepsy as well (Broughton et al 1981; Reynolds et al 1983;Tan et al 1984; Vollrath et al 1989). Third, disturbednocturnal sleep, in varying degrees, is a feature of all three disorders (Montplaisir et al 1978; Gillin et al 1979; Reynolds et al 1983, 1984; Broughton et al 1988). Given these clinical similaritiesamongthe three disorders, do the polysomnographicdata nevertheless providesufficientevidenceof differencesamongthemthat they should be classified as separate entities? One way to look at this questionis to examinewhetherthe various polysomnographic abnormalities in these disordersare independent,or whetherthey mightbe accommodated by a simpler model of overall sleep disturbance.To this end, we performed a metaanalysis of polysomnographicmeasuresin insomnia,depression, and narcolepsy.
Methods Selection of Studies To create a database for our meta-analysis, we searched the medical literature for all polysomnographic studies of insomnia, depression, and narcolepsy dating from the introduction of this technique. We used three sources to identify studies: (1) the MEDLINE
960
BIOL PSYCHIATRY 1992;32:958-975
J.I. Hudsonet al
database from January 1983 to May 1992; (2) additional studies cited in the sports ted by &e MEDLINE search; and (3) citations from previous meta-analyses of mnographic studies(Knowles and MacLean 1990; Obermeyer et al 1991). TO be d, studies were required to adopt or closely parallel established diagnostic criteria. Specifically, for insomnia, all studies used the criteria for psychophysiologic insomnia of the Associationof Sleep Disorders Centers ( 1979), except for three studies (Karacan et al 1971; Frankel et al 1976; Seidel et al 1984) that used criteria closely approximating the above. For depression, a major depressive episode was defined by Research Diagnostic Criteria (Spitzer et al 1978) in all studies except one (JI Hudson et al unpublished) that used DSM-III-R criteria (American Psychiatric Association 1987). For our analysis, studies of depression were permitted to include patients with unipolar and bipolar depression, as well as those with and without psychotic features. Other subgroupsof depression, however, such as atypical depression (Quitkin et al 1985) and anergic depression (Thase et al 1989), were excluded, because criteria for these entities are not standardized and often vary significantly from those of major depression. For narcolepsy, all studies used criteriavery similar to those of the Association of Sleep Disorders Centers (1979). In addition, it was required that studies of narcolepsy exclude patients with clinically significant sleep apnea; studies were not required to exclude patients with periodic leg movements during sleep, since such activity is a characteristic feature of narcolepsy (van den Hoed et al 1981; Wittig et al 1983; Baker et al 1987; Bedard et al 1989). Studies meeting the above standards for diagnosis were also required to meet the following criteria: (1) controlled comparison of disorder of interest with either normal control subjects or one of the two other disorders under review; (2) mean age of subject groups over 20 years (because narcolepsy and insomnia have been studied in adults only); (3) age-matched samples, with a difference of at most 10 years in mean age between subject groups; (4) non-selective subject recruitment for each diagnostic group (e.g., studies of patients with depression plus memory loss (Vitiello et al 1990), or insomnia plus decreased sleep efficiency or increased sleep latency (Hauri and Fisher 1986) were excluded; (5) subjects unmedicated at the time of study; (6) all-night polysomnographic studies only (as there are insufficient data to review for other procedures, such as the multiple sleep latency test); (7) use of standard visual scoring procedures; and (8) report of mean and stand deviation (SD) of at least one polysomnographic variable of interest. We also excluded instances in which data from the same subjects were presented in two separate reports. When two or more studies of the same diagnostic group appeared from the same center with similarly aged subjects, we included only the study with the largest sample size, unless we were able to confirm from the authors that no overlap in data had occurred. This selection procedure yielded eight studies of insomnia, 13 studies of depression, and nine studies of narcolepsy. The demographic characteristics, number of nights of polysomnography , as well as mean and SD for polysomnographic variables of interest, for each of these studies are presented in Table 1. The mean ages for patients in the pooled studies were 37.9 years for insomnia, 39.1 years for depression, and 42.5 years for narcolepsy. For all but eight of the studies, subjects were unmedicated for at least 2 weeks prior to evaluation. Five studies useda l-week medication-freeinterval(Richardson et al 1978; Wittig et al 1983; Broughton et al 1988; Lauer et al 1991; Mendlewicz and Kerkhofs 1991); two studies included a small number of patients with a 2-S day medication-free interval (Hishikawa et al 1976; Bedard et al 1989); and one study did not
Sleep in Insomnia,Depression,and Narcolepsy
BIOL PSYCHIATRY 1992;32:958-975
961
specify the medication-free interval (Schneider-Helmert 1987). For studies that presented data from individual nights separately, we calculated the mean and combined SD of all nights for our analysis.
Constructionof the Sleep Disturbance Index TO determine whether the
three disorders were associated with differing degrees of disturbed sleep, we attempted to devise a summary measure of degree of sleep disturbance that would be composed of several variables, derived from visual scoring of records, commonly reported in the polysomnographic literature on insomnia, depression, and narcolepsy. Five variables seemed to best summarizeoverall quality of sleep: wakefulness aftersleeponset(WAS@(minutes of awake time after sleep onset); percentage of stageI sleep(expressed as percentage of the total sleep time); percentage of stage 3 + 4 deep (as percentage of the total sleep time); REM latency(time between onset of sleep and onset of the first REM sleep period); and REMdensity(amount of eye-movement activity per min of REM sleep). The first three of these variables comprehensively and nonredundantly measure the presence of wakefulness and the depth of nonREM sleep: specifically, increased wakefulness after sleep onset, increased stage 1 sleep (“light” sleep), and decreased stage 3 + 4 sleep (“deep sleep”) each represent poorer quality of sleep. Amounts of the other two sleep stages, stage 2 and stage REM, were not included, because changes in these variables have not been associated with poor quality of sleep. Stages 3 and 4 were necessarily considered together because many of the available studies did not report data on these stages separately. The other two variables, REM latency and REM density, were chosen because shortened REM latency and increased REM density are reported in several disorders to be associated with poor sleep, particularly depression, and because they are nonredundant with the three measures of sleep stages. Thus, it seemed likely that they would add information on overall quality of sleep beyond that supplied by sleep stage measures. Furthermore, this choice of variables for our summary index has construct validity, in that all of these measures (with the exception of REM density, which will be discussed later) worsen with age (Feinberg 1974;Ehlers and Kupfer 1989; Knowles and MacLean 1990; Lauer et al 1991), as does subjective sleep quality. We combined the resultsof polysomnographicstudiesfor each variableusing a weighted average of effect sizes, and then we derived an overall “sleep disturbance index” (SDI) by combining the estimates for effect sizes for the five above variables listed. Details of the statistical method for this analysis are provided in the Appendix. Although these five variables seemed to provide the simplest a priorisummary model of sleep quality, we considered several alternate choices of variables. We will return to a consideration of the merits of these alternate models, and their effect on our calculations, in the discussion.
Results The weighted averages of effect sizes, 6*‘s, calculated for each of the five polysomnographic variables contributing to the SD1 in each of the three disorders, are presented in Figure 1. For all variables considered, the h*‘s for insomnia were found to be either statistically similar to or higher than normal individuals; those for depression were either similar to or higher than insomnia; and those for narcolepsy were either similar to or
Karacanet al (1971) Frankel et al (1976) Coates et al (1982) Mendelson et al (1984) Seidel et al (1984) Mendelson et al (1986) SchneiderHelmert (1987) Reynolds et al (1984) Gillin et al (1979)’ Quitkin et al (1985) Reynolds et al (1985) Duhe et al (1986) de Maertelaeret al (1987) Mendelson et al (1987) Latteret al (1991)’
Study 10 10 18 lgg 12 12 10 10 38 38 10 10 16 16 10 2ok 74’ 41 19 21 2s 2s 52 23 19 9 8 8 74 51
Insomnia Depression Depression Normal Depressiot# Normal Depression Normal Depression Normal Depression Normal Depression Nomal Depression Normal
N
Insomnia Normal Insomnia Normal Insomnia Normal Insomnia Normal Insomnia Normal Insomnia Normal Insomnia Normal
Diagnosis
43.8 39.9 43.0 44.0 40.4 35.1 69.9 69.0 40.2 36.9 41.4 35.0 43.9 43.9 39.0 38.3
37.5’ 37.5’ 44.5 45.1 41 .r 41.7’ 35.3 35.4 29 29 39.5 40.1 46.1 46.1
Mean age 2-3d.e
Id Id 3-tjd 3-sd 2-3 2-3 2d 2d Id Id 3d 3d Id Id Id Id
4d 4d 3d 36 2d 2d Id Id 2d 2d Id Id
2-3d.e
(51.5) (25.8)
(53.7) (32.8)
(78.0) (31.0) (37.3) (25.8) (67.6) (26.1)
77.4 33.7
116.8 47.4
91.6 41.0 26.3 21.3 64.2 26.7
(22. I) (20. I) (49.4) (42.8)
23.6 23.5 122.8 63. I (37.5)
(21.8) (17.3) (24.0) (19.0)
30.1 19.9 47.9 25.9
49.4
(34.2)
49.5
WA!@’ (min)
It).0 IS.5 16.1 9.7 4.7 3.2 21.4 19.0 5.8 8.5 9.4 7.7
9.1
10.1 6.6 4.5 2.9 9.0 7.1 9.2 6.2 8.1 6.3
(8.7) (4.1) (10.7) (8.2) (4.1) (2.4) (10.9) (6.0) (5.7) (11.1) (4.9) (4.8)
(4.5)
(5.8) (2.7) (3.0) (2.4) (5.0) (3.0) (7.1) (3.9) (4.4) (2.5)
Percentage stage 1
4.7 6.1 2.1 2.5 10.5 18.1 10.3 14.0 7.5 11.8 11.0 11.2
2.8
9.1 10.4 17.6 26.2
2.0 3.1 17.3 15.9
7.4 12.3 2.5
(4.3) (5.3) (4.1) (3.9) (9.3) (7.4) (7.9) (9.4) (5.9) (5.8) (8.9) (8.9)
(4.6)
(1.9) (1.8) (7.8) (7.3)
(3.2) (3.0) (7.3) (4.2)
(10.8) (8.7) (3.9)
Percentage stage 3 + 4
47.8 78.1 53.9 92.8 33.3 57.6 55.8 90.1 56.9 113.0 86.5 63.3 54.3 70.9
82.9
77.4 90.7
86.7 84.5
81.6 86.0 68.6
(29.5) (26.4) (40.1) (33.6) (36.5) (16.9) (36.5) (33.5) (32.8) (S2.2) (63.6) (28.3) (32.1) (23.9)
(28.0)
(27.0) (45.9)
(40.0) (19.9)
(37.0) (29.0) (16.6)
REM latency (min)
Mean (SD) values for polysomnographic variables”
Measures in Studies Used for
Nights of sleep
Table I. Clinical Characteristics and Polysomnographic
(0.29) (0.76) (0.59) (1.47) (1.23) (0.67) (0.48)
(0.83) (0.25) (0.51) (0.72) (12.3) (9.7)
2.01 1.73 4.53 3.45 1.57 1.35
1.44 0.47 1.38 1.61 41.7 27.0
(0.41) (0.38)
I.44 1.30
1.07
(0.43) (0.33)
(0.39)r
1.39 1.38
1.s3
REM density (units/min)
k
z
E CD
k E
4
2
2-4 2-4 2 2 1 1 2-3 2-3 2 2 1 1 2 2 3d 3d 1 1 2 2
25.8 24.5 42.2 42.8 28.2 28.2 44.3 43.6 44.6 44.1 44.7 49.3 39.4 39.4 47.5’ 47.5’ 45.3 45.6 43.6 44.5
30 19 25 25 20” IO 20 10 8 8
Depression Normal Depression Narcolepsy Narcolepsy Normal Narcolepsy Normal Narcolepsy Normal N yolepsy Normal Narcolepsy Normal Narcolepsy Normal Narcolepsy Normal Narcolepsy Normal 20 II 11 I7 17 IO 10 I29 6
549
2 2 2 2 Id Id Id ld
24.4 24.4 30.8 29.3 35.4 32.5 42.5 41.4
8 8 60 20 139 47 67 66
Depression Normal Depression Normal Depression Normal Depression Normal (22.7) (8.2) (30.4) (46.6)
(42.9) (31.3) (38.2) (29.6) (34.9) (41.6)
(21 .O) (36.9) (6.5) (3.9)
56.7 39.1 97.7 62.3 55.2 52.3
56.0 36.2 y.8 5.2
(33.9) (20.8)
23.3 16.0
22.5 9.6 31 .s 60.1
(14.1) (7.1)
12.0 7.5
(5.2) (2.4) (4.8) (8.9) (7.2) (6.1) (7.6)” (3.3)”
(7.1) (4.9) (6.3) (5.8)
(7.1) (7.1) (3.3) (3.4)
18.1 14.3 IS.9 13.6
12.3 8.9 21.1 10.8
(3.0) (2.6) (8.8) (4.8) (4.6) (4.5) (8.7) (6.4)
7.3 3.7 7.9 16.2 25.2 13.5 17.9 8.0
4.5 4.0 10.2 10.0 6.7 8.4 9.6 6.8
20.0 14.8 11.2 9.3
9.6 7.2
15.0 13.0 2.2 3.0 13.6 13.7 10.2 14.2
12.5 20.0 68.5 83.6 8.3 7.9 14.7 15.0
(8.0) (5.1) (3.3) (7.5)
(6.5) (6.8)
(13.2) (8.3) (4.7) (7.5) (7.5) (4.7) (7.2)p (7.2)p
(8.0) (10.1) (23.8)” (lS.3)“’ (8.1) (8.8) (10.6) (8.2)
17.2 61.7 48.2 85.5
15.9 86.3
41.1 81.1 68.4 79.6
58.7 85.9 47.4 25.6
71 .O 74.5 1.90 1.30 61.2 71.3 69.1 80.6
(33.0) (17.8) (28.9) (37.2)
(29.6) (33.7)
(37.5) (42.6) (61 .O) (20.9)
(25.6) (21.9) (212) (31.2)
(50.6) (22.5) (0.50) (0.30) (30.5) (21.6) (48.8) (30.3) (0.43) (0.20) (0.46) (0.57)
(0.42) (0.43) (1.69) (1.65)
(0.55) (0.51)
5.05 (I.60)’ 4.03 (2.25)
1.25 0.92 1.3s 1.61
1.29 1.07 4.40 2.80
1.s5 1.30
OWhenvalues for more than one night of sleep were reported in an individual study, the mean and SD for all were calculated and used for analysis. 6Wakefulnessafter sleep onset. ‘Midpoint of age range (mean age not reported). dPolysomnographyperformed after 1-2 nights of t 1 vdws werenot used, since the variable number of subsequentnightsprecludedaccuratecombiningof data. ‘Value reJrottedfor this group reportedby Feinberget al was not included in analysis, becauseof presumed overlap with Gillin et al (1979). *JIepressiongroup was not included in analysis, becauseof presumedoverlap with ReynoJdset 3). ‘Combined 56 patientsin initial study and 18 patientsin replicationstudy.‘Data for 56 of 74 patientsreportedin tabular form in Feinberget al (1982). ‘Patientswith e ~sion inchtded, but patientswith atypical depression excluded.‘Data for WAS0 from Latter et al (1988). “Data from Giles et al (1990). which reporteda larger sample(177 and 54 healtby individuals)from this centerREM latency only. “Patientswith major depressionor bipolar disorder, depressed.by DSM-III-R criteria; studiedaccording bed in Hudsonet al (1988); data reportedin part in Hudsonet al (1992). “Narcolepticpatientswith only sleep attacksnot included. becausethey would not meet trite in this analysis(see text). PMean values reported and total sleep time, qJncJ&s patients wih and for one night for this variable in Montplaisir (1976); since SD was not reported,it was estimatedfrom the SD of without periodic leg movements(see text). ‘Includes unpublisheddata from this study (Browman CP; prsonal communication, 1991). Data calculatedfrom figure 28 of Uris study,
Kupfer et al (1989) Wailer et al (1989) Buysse et al (1990) Mendlewicz and Kerhofs (1991) Hudson et al (unpublished)” Reynolds et al (1983) Hishikawa et al (1976) Montplaisir ct al (1978) Rtchardsonet al (1978) Wittig et al (1983) Browman et al (1986) Geisler et al (1987) Broughton et al (1983) Bedardet al (1989)
L z !Y
964
J.I. Hudson et al
BIOLPSYCHIATRY 1992;32:95&975
A. WAKEFULNESSAFTERSLEEPONSET
Insomnb
B. PERCEN
Narcolepsy
E OF STAGE1 SLEEP
Insomnia Figure
Depressbn
Depressian
Natcolepsy
1, s*‘s (mean I SEM) for diagnostic groups for the following polysomnographicmeasures: (A) wakefulness after sleep onset; (B) percentage of stage 1 sleep; (C) percentage of stage 3 + 4 sleep; (D) REM latency; (E) REM density. Significanceof differences: (A) insomnia versus normal. individuals,p < 0.001; depression versus normal individuals,p < 0.001; narcolepsy versus normal individuals, p = 0.001; (B) insomnia versus normal individuals, p = 0.003; insomnia versus narcolepsy, p = 0.032; depression versus narcolepsy, p = 0.003; narcolepsy versus normal individuals, p < 0.001; (C) insomnia versus normal individuals, p = 0.002; depression versus norm,il individuals, p = 0.019; (D) insomnia versus depression, p = 0.002; insomnia versus narcolepsy,p -C 0.001; depression versus normal individuals, p < 0.001; narcolepsy versus normal individuals, p < O.OOl; (E) insomnia versus depression, p = 0.001; insomnia versus narcolepsy, p = 0.029; depression versus normal individuals,p < 0.001; narcolepsy versus normal individuals, P = 0.016. All other differences nonsignificant (p > 0.05).
Sleepin Insomnia, Depression, and Narcolepsy
BIOL PSYCHIATRY 1992;32:958-975
965
C. PERCENTAGEOF STAGE3+4 SLEEP 0.8
r
0.6
0.4
0.2
0.0
Insomnia
Depressbn
Narcolepsy
Figure 1. Continued
D. REM LATENCY I.U
1.2 ’.
G +I x
o*8
0.4 . 0.0
Insomnia
Depressbn
Narcolepsy
higher than depression. Combining these variables to derive the SD1yielded an upward slope (Figure 2), in which each disorder showed an SD1 that was higher (to a degree either exceeding or approaching statistical significance) than its predecessor.
Discussion We constructed a “sleep disturbance index” based on five commonly measured polysomnographic variables, to provide a summary description of quality of sleep. Calculating the SD1for three widely studied disorders-insomnia, depression, and narcolepsy-using a statistical meta-analytic approach, we found these disorders to be arrayed along a continuum of progressively more disturbed sleep.
966
J.I. Hudsonet al
BIOL PSYCHIATRY 1992;32:958-975
E. REM DENSITY 0.8
I
0.6
Insomnia
Depression
I
Figure1. Continued
Naroolepsy
Before considering further the implications of these findings, several criticisms of our method should be considered. These criticisms fall into two groups: choice of inclusion criteria for studies, and choice of variables for the SDI. First, with regard to inclusion criteria for studies, it might be argued that greater precision could have been obtained by narrowing the analysis to a specific age group throughout, since age has such a strong effect on polyf;omnographicfindings (Knowles and MacLean 1990). For example, one might restrict analysis to studies with a mean age of 35-45 years, as this would include most of the studies of insomnia (seven of eight), depression (eight of 13), and narcolepsy (eight of nine). We decided against this approach, and included all studies, because the mean ages of the groups were similar, even though
Figure2. Sleep disturbanceindex ( f: SEM) for diagnostic groups. Significance of differences: insomnia versus normal individuals, p < 0.001; insomnia versus depression, P = 0.062; insomnia versus narcolepsy, p < 0.008; depression versus normal individuals, p < 0.001; depression versus narcolepsy, p = 0.076; narcolepsy versus normal individuals, p < 0.001.
Insomnia
Depression
Narcolepsy
Sleep in Insomnia,Depression,and Narcolepsy
BIOLPSYCHIATRY 1992;32:958-975
969
nonREM sleep, and REM sleep. If so, the SD1 would reflect several unrelated aspects of sleep disturbance, rather than overall sleep disturbance. Although this possibility cannot be excluded, three lines of evidence suggest that the SD1does reflectoverallsleep quality. First, all of the component measures in the SD1worsen with increased arousal. Increased arousal is associated with a shift towards more wakefulness, increased stage-l sleep, and decreased stage 3 + 4 sleep. Because REM latency represents the length of the first NREM period, factors that decrease NREM sleep (particularly slow-wave sleep), such as increased arousal, may thereby shorten REM latency (Feinberg et al 1988; Ehlers and Kupfer 1989). Finally, increased arousal is associated with increased REM density (Feinberg et al 1987, 1988).
Second, the SD1 worsens with age, as sleep becomes lighter. All of the measures in the SDI, with the exception of REM density, worsen with age (Feinberg 1974; El&s and Kupfer 1989; Knowles and MacLean 1990; Lauer et al 1991). However, even the changes in REM density with age may be consistent with the effects of the age-related lightening of sleep. Ehlers and Kupfer ( 1989)found that middle-aged individuals displayed increased REM density compared with younger individuals, but elderly individuals had REM density values similar to young individuals. But elderly individuals also displayed increased time awake in the last one-third of the night, a time dominated by REM sleep with high REM density in young and middle-aged individuals. Thus, they have proposed that the failure to find a progressively increased REM density with age may be a result of gradual replacement of high-density REM sleep with awakening due to increased levels of arousal. Third, depressed individuals show a greater change with age than do normal individuals in most the component measuresof the SDI, accordingto many (Gillin et al 1981; Knowles and MacLean 1990; Lauer et al 1991), but not all (Obermeyer et al 1991) studies. In other words, the sleep disturbance in depression may be analogous to “accelerated aging” (Gillin et al 1981). Hence, for at least one of the disorders under study, the data are consistent with the possibility that a unitary process, similar to that involved in normal aging, disrupts most of the measures in the SDI. Thus, although it is possible that the differentcomponentmeasuresof the SD1 assess independent systems, the above three observationssuggest thatthese measures are not independent and that the SD1may assess a general tendency toward more disturbed sleep. In summary, therefore, the findings generated using the SD1 are consistent with a “unitary” hypothesis of sleep disturbance in insomnia, depression, and narcolepsy: that these three disorders are positioned along a single axis of progressively more disturbed sleep. Admittedly, this analysis does not disprove the currently more popular “pluralistic” hypothesis:that the sleep disturbancesin insomnia,depression,and narcolepsyare fundamentallydifferentand are associatedwith pathognomonic polysomnographic features. However, attempts to segregate the disorders on the basis of polysomnographic features requiremorecomplicated assumptions. For example, one must assumethatREMlatency andREMdensityaremeaningful to distinguish narcolepsy and depression frominsOda, but not depression from narcolepsy, and that sleep efficiency and sleep latency are meaningful for distinguishingnarcolepsyfrominsomniaanddepression,butnot insomnia fromdepression. The unitary hypothesis, by comparison, explains the polysomnographic findingsin these disordersmoreparsimoniously. Stateddifferently, it appearsthat sleep can be affected in only a limitednumberof ways-and that present dataare insufficientto permitelaborationmuchbeyonda single axis of god sleep to bad sleep. If so, polysornnographic measures cannot be used as
970
J.I. Hudsonet al
“biOL PSYCHIATRY 1992;32:958-975
evidence for classifying insomnia, depression, and narcolepsy as separate entities. Indeed, the possibility remains open that the three disorders are closely related, as perhaps implied by their frequently shared phenomenologic fentures and treatment response (Tan et al 1984; Jacobs et al 1988; Hudson and Pope 1990).
Appendix For each polysomnographic variable for each study, we computed the unbiased estimate for the effect size using the method proposed by Hedges and Olkin (1985, p 81)
>
(X, - X,)/SD,,
Cl1
where N = 11~+ n, is the sum of the sample sizes for control and experimental groups, respectively; X, is the mean value of the normal control group; X, is the mean value of the experimental group, and SD, is the pooled standard deviation for the normal control group and the experimental group. For the two studies (Reynolds et al 1983, 1984) that did not have normal control groups, we used data from an age-matched normal control group studied at the same center as these two studies, but reported separately (Thase et al 1989). Note that for polysomnographic variables in which a larger value represented better quality sleep (sleep efficiency, stage 3 + 4 sleep, REM latency), we multiplied the effect size by - 1 so that a larger positive value of ES always represented poorer qualit! of sleep. Before pooling the estimates of the effect size for a series of k studies, it is important to determine whcthcr the studies cstimute a common effect size 6. The assumption that all of the studies have the same constant but unknown effect size was tested for each polysomnographic variable using the x2 statistic for homogeneity with k-l degrees of freedom, k
Qw=
Wi
(ESi - S)‘,
M
where 8 is the weighted estimator under the assumption of a common effect size
133
and the weights wi are the reciprocal of the estimated variance of the effect size ESi for the ith study (Hedges and Olkin 1986, p 86). When the hypothesis of homogeneity is rejected, the variability among studies is greater than expected under the simple model of random deviations from a common effect size, and the pooled estimated in [3] is not appropriate. We used random effects models to account for the unexplained variability in the observed effect sizes (National Research
Sleep in Insomnia, Depression, and Narcolepsy
BIOL PSYCHIATRY 1992;32:958-975
971
Council 1992). The average true effect is estimated using a weighted average of the effect sizes
8* =
(i
WTEsJ/(g wf),
HI
where the revised weights w* are the reciprocal of the variance of the effect sizes under the random effects model and are given by WT
=
(wj-’
+
ep-’ ,
PI
and T*is the method-of-mome$ estimate of the variance component (DerSimonian and Laird 1986). The variance of 6* is estimated by v&b” =
w1
To derive the overall sleep disturbance index (SDI), we computed the $*‘s for five polysomnographic variables: wake after sleep onset (6&,), percentage of stage 1 sleep (&), percentage of stage 3 + 4 sleep (@s4), REM latency (&), and REM density (&id). We assigned each variable equal weight, as we lacked justification Q priori for differential weighting. Thus, the SD1 for each diagnostic group represented the mean of the $*‘s of the five variables:
and the standard error of SD1was estimated using the square root of the variance of SDI. Using the formula for computing the variance of a sum of correlated components, the estimate of the variance of SD1 is equal to
viirSD1 = (1125) 5
5
i=l
i=l
j-l
(V&
8:)“’
(Vb
ST)“’
rijg
PI
j
