Breslau N, Roth T, Burduvali E, Kapke A, Schultz L, Roehrs T. Sleep in Lifetime Posttraumatic Stress DisorderA Community-Based Polysomnographic Study. Arch Gen Psychiatry. 2004;61(5):508-516. doi:10.1001/archpsyc.61.5.508
Copyright 2004 American Medical Association. All Rights Reserved.Applicable FARS/DFARS Restrictions Apply to Government Use.2004
Sleep complaints are common in posttraumatic stress disorder (PTSD)
and are included in the DSM criteria. Polysomnographic
studies conducted on small samples of subjects with specific traumas have
yielded conflicting results. We therefore evaluated polysomnographic sleep
disturbances in PTSD.
A representative cohort of young-adult community residents followed-up
for 10 years for exposure to trauma and PTSD was used to select a subset for
sleep studies for 2 consecutive nights and the intermediate day. Subjects
were selected from a large health maintenance organization and are representative
of the geographic area except for the extremes of the socioeconomic status
range. The subset for the sleep study was selected from the 10-year follow-up
of the cohort (n = 913 [91% of the initial sample]). Eligibility criteria
included (1) subjects exposed to trauma during the preceding 5 years; (2)
others who met PTSD criteria; and (3) a randomly preselected subsample. Of
439 eligible subjects, 292 (66.5%) participated, including 71 with lifetime
PTSD. Main outcomes included standard polysomnographic measures of sleep induction,
maintenance, staging, and fragmentation; standard measures of apnea/hypopnea
and periodic leg movement; and results of the multiple sleep latency test.
On standard measures of sleep disturbance, no differences were detected
between subjects with PTSD and control subjects, regardless of history of
trauma or major depression in the controls. Persons with PTSD had higher rates
of brief arousals from rapid eye movement (REM) sleep. Shifts to lighter sleep
and wake were specific to REM and were significantly different between REM
and non-REM sleep (F1,278 = 5.92; P =
We found no objective evidence for clinically relevant sleep disturbances
in PTSD. An increased number of brief arousals from REM sleep was detected
in subjects with PTSD. Sleep complaints in PTSD might represent amplified
perceptions of brief arousals from REM sleep.
The official definition of posttraumatic stress disorder (PTSD) in the DSM-III and subsequent DSM editionsincludes sleep problems as criterion symptoms of reexperiencing (ie, "recurrentdistressing dreams of the event") and increased arousal (ie, "difficulty fallingor staying asleep"). The presence of sleep disturbance (as that of other definingsymptoms) is determined on the basis of subjects' reports and does not requireany objective, independent verification. Although sleep complaints are notnecessary for the diagnosis of PTSD (as diagnostic criteria can be fulfilledin their absence), disturbing dreams and insomnia are reported by most personswith the disorder.1- 3 Thecentrality of traumatic memories in PTSD and their intrusion in sleep haveled to the application of sleep research methods to this disorder, extendingprevious research on affective disorders and other anxiety disorders.
To date, sleep studies on PTSD have focused primarily on combat veteransand, to a lesser extent, on victims of specific traumas who are recruitedin clinical settings. In stark contrast to the consistent findings of highrates of sleep complaints in PTSD, sleep-laboratory studies have yielded conflictingresults. Some studies reported difficulties in initiating and maintainingsleep, reduced sleep efficiency, and increased rates of brief arousals duringsleep.4- 10 Reportsthat sleep difficulties in patients with PTSD were observed only on the firstnight in the laboratory have suggested an increased sensitivity to novel sleepenvironments in PTSD.8 Attenuated first-nighteffects in Vietnam veterans with PTSD were reported as well.11 Otherstudies failed to find objective evidence of sleep disturbances in patientswith PTSD.12- 16
Rapid eye movement (REM) abnormalities are characteristic findings inaffective disorders, which are frequently comorbid with PTSD.17,18 However,indicators of abnormalities in REM sleep (eg, latency to REM and total REMsleep) have not been detected in studies of anxiety disorders such as obsessivecompulsive disorder,19 panic disorder,20 and social phobia.21 Incontrast to other anxiety disorders, the observation that nightmares and recurrentdreams associated with traumatic memories are prototypic symptoms in PTSD1,2,22,23 hassuggested the possibility that REM abnormalities might be observed in PTSD.Studies of REM functions have yielded conflicting results. Some studies reportedshortened latency to REM sleep, increased REM density, and increased percentageof REM sleep.15,24- 26 Otherstudies reported contrasting patterns, characterized by increased REM latencyand reduced REM sleep.4,5 Stillother studies failed to detect differences in REM parameters between personswith PTSD and control subjects.27,28
Recent reports from small clinical samples have called attention tothe possibility of an increased rate of breathing disturbances during sleepin patients with PTSD.29- 31 Thegeneralizability of these results to PTSD in the community has not been examined.
Previous sleep studies in PTSD have been limited by their small samplesizes and unrepresentativeness. Some lacked standardized diagnostic approachesor systematic assessment of all salient variables, including measures of morenuanced sleep fragmentation, adaptation to a novel sleep environment, anddaytime sleepiness. The present study attempts to address some of these limitations.The study was conducted in the context of a large-scale, longitudinal communitystudy. We present data from polysomnographic studies of 2 successive nightsand from a measure of daytime sleepiness in the intervening day, collectedon a subset of this epidemiological sample, including 71 persons with lifetimePTSD (12 with current and 59 with past PTSD).
Our analysis proceeds as follows. First, we compare subjects with lifetimePTSD vs those with no history of PTSD. Second, we compare sleep measures acrossthe following 3 subgroups: PTSD, exposed to trauma/no PTSD, and never exposedto trauma. Third, we compare sleep measures of subjects with PTSD, major depression(MDD), and neither disorder. Our analytic models are designed to detect first-nighteffects overall (main effects) and across subgroups (interactions) and testother interactions of interest. As a preliminary step, we compared subjectswith current vs past PTSD but found no evidence that the groups differed orthat those with past cases display lower sleep quality than those with currentcases. Consequently, we combined current and past cases in the analysis.
The sleep investigation of PTSD was nested in a large-scale longitudinalcommunity study of young adults. The study was described in detail previously.32,33 In brief, a sample of 1200 subjectswas randomly selected in 1989 from all 21- to 30-year-old members of a largehealth maintenance organization in southeast Michigan. The membership of thehealth maintenance organization is representative of the population of thegeographic area as depicted in the 1990 US census data, with the exceptionof the extremes of the socioeconomic range.22 Personalinterviews were conducted in 1989 with 83.9% of the sample (n = 1007). Follow-upinterviews were conducted in 1992, 1994, and 1999-2001. In each wave, morethan 90% follow-up completion was achieved. In the 10-year 1999-2001 follow-up,913 of the initial sample (91.0%) completed interviews. Sleep measures werecollected on a subset of these respondents, whose ages ranged from 31 to 40years at that time.
The following subjects from among those who completed the 10-year follow-upwere eligible: (1) those who were exposed to traumatic events in the intervalfrom 1994 to the last assessment, approximately 5 years later; (2) those notincluded under the first criterion who met PTSD criteria in previous assessments;and (3) those not included under the first or the second criterion from arandomly preselected subsample of the total sample. Subjects who moved awayfrom the metropolitan area were not eligible. As a rule, those with recent(in the past month) substance abuse and use of psychotropic medications arenot eligible for sleep studies. In this general population sample, no suchexclusions were necessary. Respondents were invited to spend 2 consecutivenights and the intervening day at the Sleep Center of Henry Ford Hospital,Detroit, Mich. Of a total of 439 eligible subjects, 292 (66.5%) participated.Complete polysomnographic data are available on 283 respondents.
Each eligible participant was contacted by telephone and recruited forparticipation. Those agreeing to participate were further interviewed to determinetheir usual sleep schedule and use of prescribed, over-the-counter, and recreationaldrugs. On arrival at the sleep laboratory, the study procedures and requirementswere again outlined verbally, and the participant read and signed a writteninformed consent approved by the Henry Ford Hospital institutional reviewboard.
The sleep study consisted of 2 consecutive 8-hour nights of polysomnographyand a multiple sleep latency test (MSLT) conducted during the day betweenthe consecutive nights. Bedtime was established on the basis of the respondent'ssleep schedule, as reported in the telephone interview. Determining the midpointof a participant's usual sleep schedule and adding 4 hours to each side ofthe midpoint set the 8-hour sleep time. Caffeine and nicotine use was excludedin light to moderate users, whereas consumption was reduced and scheduledfor heavy users (ie, daily use of >12 cigarettes and/or >400 mg of caffeine)so as not to interfere with the polysomnographic and MSLT assessments.
The polysomnography included the standard central (C3-A2) and occipital(Oz-A2) electroencephalograms, bilateral horizontal electro-oculograms, asubmental electromyogram, and an electrocardiogram recorded with a V5 lead.34 In addition, airflow was monitoredwith oral and nasal thermistors, and leg movements were monitored with electrodesplaced over the left tibialis muscles.35,36 Therecordings, collected at a speed of 10 mm/s, were traced onto paper with polygraphs(Model 7D; Grass-Telefactor, West Warwick, RI) or digitized and stored (Heritage;Grass-Telefactor) electronically on equipment located in a separate monitoringroom. Respiration and tibialis electromyogram recordings were evaluated andtabulated as to frequency of respiratory and leg movement events using thestandard scoring criteria.35,36 Briefly,apneas, defined as 10-second or longer cessations of airflow, and hypopneas,defined as 10-second or longer 50% reductions of airflow, were summed andexpressed as number of events per hour. In scoring periodic leg movements,only those tibialis electromyogram flexions of 0.5 second and greater associatedwith electroencephalographic arousals were tabulated. All sleep recordingswere scored in 30-second epochs, according to the standards of Rechtschaffenand Kales34 for sleep stages. Scorers maintaineda 90% scoring reliability and were unaware of study night or respondent'sgroup membership.
The primary polysomnographic measures used to compare groups were latencyto persistent sleep (latency from lights out to the first 10 minutes of continuoussleep); minutes of wake during sleep (epochs scored wake after sleep onsetand before the final awakening divided by 2); sleep efficiency ([Total SleepTime/Time in Bed] × 100); number of awakenings (2 consecutive epochsscored wake); percentages of stages 1, 2, 3 and 4 (combined), and REM sleep;latency to REM sleep (nonwake minutes from lights out to first epoch scoredREM; and number of entries to stage 1 sleep and wake (1 epoch or more) fromnon-REM and REM sleep.
Measurement of daytime sleepiness by means of the MSLT was performedaccording to the standard protocol37 on theday between the 2 nocturnal recordings, at 2-hour intervals after arising,typically at 10 AM, 12 PM, 2 PM,and 4 PM. Participants laid down in a bed in quiet and darkrooms with the instruction to go to sleep. They remained in bed for 20 minutesafter 3 consecutive epochs of stage 1 sleep, an epoch of another sleep stage,or 20 minutes of continuous wake. Sleep latency was scored as the minutesto the first epoch of sleep. The mean latency of the 4 tests was used as themeasure of daytime sleepiness.
The National Institute of Mental Health–Diagnostic Interview Schedule(NIMH-DIS)38 for DSM-III-R was used to diagnose psychiatric disorders. The baseline interviewin 1989 inquired about lifetime history of disorders, and each follow-up assessmentinquired about disorders occurring during the interval period since the previousassessment. The diagnosis of PTSD in DSM-III-R requiresexposure to a qualifying traumatic event and the presence of PTSD criterionsymptoms that are linked to the traumatic event. Two earlier studies reportedhigh concordance between the diagnosis of lifetime PTSD by lay interviewersusing structured interviews based on the DIS and independent clinical reinterviews.38,39 The latter used the Clinician-AdministeredPTSD Scale (CAPS)40,41 and reportedsensitivity of 76% and specificity of 97%.42
Statistical analyses were performed on log-transformed data. In thefew instances when the log transformation did not correct for the distribution'sskewness, nonparametric tests were used (eg, Wilcoxon rank sum test). To facilitatecomparisons with previous work, the tables display raw means and standarddeviations (SDs). Three series of analyses were conducted. The first seriescompared the PTSD group with no-PTSD controls. The second series consistedof comparisons across 3 groups, in which the no-PTSD group was divided intosubjects exposed to trauma with no PTSD (exposed/no-PTSD group) and thosewho had never been exposed to trauma (nonexposed group). This series provided2 control groups with which the PTSD group was compared. The third seriesconsisted of comparisons across the following 3 groups, defined by historyof PTSD and MDD: (1) neither disorder; (2) MDD only; and (3) PTSD with andwithout history of MDD. A separate analysis revealed no differences betweenPTSD alone vs PTSD comorbid with MDD.
We used multiple regression analysis applying generalized estimatingequations (GEE),43- 45 totest and estimate associations between group membership (eg, PTSD vs no-PTSD)and each of the sleep laboratory measures across the 2 nights. The GEE approachpermits simultaneous modeling of the relationship between group classificationand sleep measures from each of the 2 nights. The GEE takes into account correlationswithin subjects across the 2 nights. The addition of interaction terms allowedus to examine, in addition to main effects, whether differences across groupsvaried between the 2 nights and by covariates of interest (eg, sex). The modelfor testing interactions (using sex as an example) is illustrated in the followingequation:
Y = α + β1(Group) + β2(Sex) + β3(Group × Sex) + β4(Time),
where sleep measures at 2 times (nights 1 and 2) are the outcomes (Y).
Although there were main effects of first vs second night, no significantinteractions between night (first vs second) and group membership were detectedin any model. Therefore, post hoc tests of differences by group membershipacross nights were not performed. No significant group × sex interactionwas detected. The raw means and standard deviations appearing in the tablesare averages of the 2 nights.
The statistical power available to test differences in this analysisis as follows. At α = .05 and power of 80%, a comparison between thePTSD and no-PTSD groups could detect an effect size as small as 0.386, aneffect size falling between small (0.2) and medium (0.5) for a t test, as defined by Cohen.46 The effectsize in the 3 group models that can be detected between any 2 groups (eg,PTSD vs no exposure) was slightly larger but less than medium.
Characteristics of the subsample on which sleep-laboratory measuresare available and the total sample of respondents who participated in the10-year follow-up are depicted in Table1. The subsample reflects closely the total sample composition withrespect to sex, race, education, age, and history of alcohol or other drugabuse or dependence. No significant differences were observed on these variableswhen groups were defined by history of exposure and diagnostic categories.The prevalence of sleep complaints in persons with PTSD, measured by itemsin the NIMH-DIS PTSD module, was 87% (n = 62). Of all PTSD cases, 42% (n =30) were attributable to assaultive violence (physical assault in 19 and rapein 11), 17% (n = 12) were attributable to sudden unexpected death of a lovedone, 14% (n = 10) to accidents, 13% (n = 9) to witnessing violence, and 14%(n = 10) to miscellaneous traumatic events.
The GEE analysis of sleep-laboratory measures across the 2 nights revealedno significant differences in measures of sleep induction and sleep maintenancebetween the PTSD vs no-PTSD groups (Table2). Significant differences were observed in the rate per hour ofshifts to lighter sleep and wake (ie, stage 1 and stage 1 plus entry to wake)from REM. The specificity of these findings to REM sleep was evaluated bytesting interactions between group membership (PTSD vs no-PTSD) and sleeptype (REM vs non-REM), using analysis of variance with repeated measures.For entry to stage 1, F1,278 = 7.16 (P =.008), and for entry to stage 1 plus entry to wake, F1,278 = 5.92(P = .02). These results indicate that the findingson these arousal variables were significantly different for REM vs non-REMsleep. The mean rate per hour of entry from REM to wake also was higher inthe PTSD than in the no-PTSD group, but the difference did not reach statisticalsignificance (P = .07). A related finding was thesignificantly lower percentage of REM sleep in the PTSD vs the no-PTSD group(P = .046).
Other findings of interest are the differences between first and secondnight in percentage of REM sleep, latency to REM, percentage of stage 2 sleep,and entry to stage 1 from non-REM sleep (Table 2). (The statistics for group comparisons on these variablesare adjusted for first-night effects.) No interactions between group membership(PTSD vs no-PTSD) and assessment night were detected on any variable, indicatingthat subjects with PTSD were not different from controls in their adaptationto the first night in the laboratory.
Subjects with current PTSD were compared with those with past PTSD onthe array of sleep variables covered in the study (Table 3). No significant differences were detected on any variable.Furthermore, the means of the 2 groups are close and, to the extent that theydiffer, past PTSD was generally associated with worse sleep.
We next evaluated sleep disturbance in PTSD by dividing the no-PTSDgroup into the following 2 subsets: exposed subjects with no PTSD and subjectswith no history of exposure (Table 4).The GEE analysis on the 2 nights of data revealed main effects of group membership(ie, PTSD, exposed only, and nonexposed) on the number of shifts per hourto lighter sleep (ie, stage 1, wake, and stage 1 plus wake) from REM sleep.Subjects with a history of exposure to traumatic events who did not have PTSDhad significantly lower rates of arousals from REM sleep than those with PTSD,whereas the nonexposed group was similar to the PTSD group. No differenceswere detected on any of the variables that signify sleep disturbance.
As in the 2-group analysis, presented in Table 2, we found evidence of significant main effects of firstvs second night but no significant interactions between night and group membership.
In Table 5 we display resultsin which the no-PTSD control group was classified according to the presenceor absence of history of MDD. The PTSD group includes persons with and withoutlifetime MDD. A separate series of analyses showed that PTSD comorbid withMDD (n = 49) did not differ from PTSD with no history of MDD (n = 22) on anyof the sleep variables. The purpose in this set of analyses was to comparethe PTSD group with a healthy control group that had neither PTSD nor MDD.Results for rates per hour of entry to stage 1 sleep and the combined ratesof entry to stage 1 plus entry to wake from REM sleep are significant, indicatingmore arousals to lighter sleep and wake specific to REM sleep in PTSD, comparedwith controls with neither disorder. No differences were detected on variablesthat signify clinical sleep disturbance.
We used a cutoff of 10 events/h to measure sleep breathing disturbances(apnea-hypopnea). Findings were 14% in the first night and 7% in the secondnight (P = .20). The difference between the PTSDand no-PTSD groups was not significant (P = .20).The only significant finding was the higher rate in men than women (odds ratio,6.8; 95% confidence interval, 2.2-20.7). Using a cutoff of 10 events/h, analysisof periodic leg movement revealed no significant differences between the PTSDvs no-PTSD groups (P = .43). There were no significantnight or sex effects.
The GEE analyses of MSLT scores used the same models applied to thepolysomnographic data. Differences across groups were not significant in anyof the analyses. The 2-group comparison (PTSD vs no-PTSD) showed a higheraverage MSLT score for the PTSD (mean, 10.4; SD, 5.0) than for the no-PTSD(mean, 9.7; SD, 4.7) (P = .30) group. There wereno significant differences between the PTSD, exposed/no-PTSD, and not-exposedgroups. The 3-group comparison by psychiatric history (PTSD, MDD, and neitherdisorder) showed a lower MSLT average in persons with MDD (mean, 8.9; SD,4.8) than in persons in the other 2 groups (P = .24).No difference was found between the PTSD and no-PTSD groups on sleep-onset–REMperiods; 15.5% and 16.1%, respectively, registered 1 or more sleep-onset–REMperiod (χ2 = 0.02; P = .90).
This study failed to find objective evidence of sleep disturbance inpersons with lifetime PTSD. On measures of sleep induction and sleep maintenance,no differences were detected between PTSD subjects and controls, regardlessof history of exposure to trauma or history of MDD in the controls. No evidencewas found of a first-night effect specific to PTSD, although the prototypicindicators of a first-night effect47- 49 wereobserved in the sample as a whole. Ross et al50 alsoreported a first-night effect that was not specific to PTSD. Furthermore,there was no evidence to support a deficit in daytime alertness in PTSD. Incontrast, we found significant associations of PTSD with shifts from REM sleepto lighter stages and to wake. The specificity of these findings to REM sleepwas tested and confirmed.
Exposed subjects with no PTSD displayed a lower number of arousals fromREM sleep than did PTSD and nonexposed subjects. Two potential explanationsshould be considered. First, a lower rate of arousal from REM might reflecta coping mechanism in trauma victims who did not have PTSD. Previous studiesreported elevated awaking thresholds in REM30 andnon-REM29 sleep in chronic PTSD and interpretedthe findings as active coping. Our findings do not replicate that pattern,in that the lower arousal rates were specific to REM sleep and were observedonly in trauma victims who did not have PTSD.
A second potential explanation is that history of exposure to traumasegregated the respondents into a susceptible group (ie, those with PTSD)vs a nonsusceptible group (ie, those who were exposed to trauma but did nothave PTSD). The nonexposed group (which showed more arousals from REM sleepthan the exposed/no-PTSD group) might have included respondents with a preexistingsusceptibility to PTSD correlated with REM arousals, who would have PTSD ifexposed. The presence of susceptible persons with elevated REM-related arousalswould push upward the average of the nonexposed group as a whole. A similarpattern has been observed for level of catecholamines.51 Considerationshould be given to the use of subjects who were exposed to traumatic eventsbut did not have PTSD as a control group for evaluating sleep correlates ofPTSD.
The findings from this study are consistent with recent reports fromclinical samples that have concluded that objective measurements of sleepby standard polysomnographic methods do not demonstrate clinically relevantsleep disturbances in PTSD.52 The interpretationof the discrepancy between subjective complaints of disturbed sleep, includinginsomnia, nightmares, and recurrent distressing dreams about the trauma, andobjective evidence of normal or adequate sleep is unclear.
The brief arousals from REM sleep associated with lifetime PTSD suggestthe possibility that complaints of disturbed sleep in PTSD might have somebasis in objectively measured facts. Although the subtle excess of arousalsfrom REM sleep does not constitute evidence of clinical sleep disturbance,these arousals may signify phenomena that are misperceived as marked disturbances.Whether or not the REM phenomena are an aspect of the syndrome of PTSD orinstead represent preexisting characteristics associated with vulnerabilityfor PTSD cannot be answered in this study. Our finding that trauma-exposedsubjects who did not have PTSD had a lower rate of arousals from REM sleepis consistent with the second interpretation.
The results of this study should be interpreted in light of the followingconsiderations. First, the 66.5% participation in this community-based study,which required a 32-hour stay in the sleep laboratory, is high, consideringthe respondents' burden, but it limits the generalizability of the findings.However, the exceptionally high 10-year follow-up rate across multiple assessmentsallows us to evaluate the representativeness of the subset that participatedin the sleep study by comparing it with the total sample from which it wasdrawn. As depicted in Table 1,no differences between the subset in the sleep study and the total samplewere detected on sociodemographic characteristics and history of substanceuse disorders. Second, although previous sleep studies have used patientswith current PTSD, we studied lifetime cases. The typically low current prevalenceof PTSD in the community is reflected in our sample, in which less than onefifth of lifetime cases were current. A comprehensive comparison of currentvs past cases across the entire array of sleep variables used in the studyfailed to detect any significant differences. Furthermore, an examinationof the means reveals a pattern that is inconsistent with poorer sleep in currentvs past cases. Thus, the failure to find sleep disturbances in PTSD cannotbe attributed to the large proportion of past cases. The number of currentPTSD cases is similar to that of previous studies of clinical samples of activePTSD, in which small sizes of samples (n<20) were typical.
Third, the model in which we separated the no-PTSD controls into thosewith history of MDD and those with no PTSD or MDD combined in a single categoryPTSD cases with and without history of MDD. Separate analyses failed to findsignificant differences between PTSD cases with MDD vs PTSD cases withoutcomorbid MDD. Similar findings were reported by Hurwitz et al.14 Fourth,we used the NIMH-DIS across all assessments. The DIS has been found to behighly specific and more conservative than a clinically administered instrument.The possibility of misclassification of PTSD cases as noncases (false-negativeclassification) cannot be ruled out. However, its implications in this studywere thoroughly evaluated in the comparison of the PTSD, exposed/no-PTSD,and nonexposed groups presented in Table4. Because PTSD, by definition, depends on exposure to DSM-qualifying traumatic events, there can be no false-negative casesamong the nonexposed group. With respect to the failure to find evidence ofsleep disturbances associated with PTSD, diagnostic uncertainty is an unlikelyexplanation.
Finally, we tested more than 20 individual comparisons in several series,leaving ample opportunity for chance findings. In giving weight to the observedsignificant differences, we rely on the pattern of the results, specifically,the clustering of the significant results in 1 specific domain, ie, briefarousals from REM sleep, which were observed in each series of analyses.
Important strengths of the study deserve mention. To our knowldege,this is the first study to evaluate objective indicators of sleep in PTSDin a large community sample. Previous studies, by and large, have used smallclinical samples or samples of volunteers, which are biased in terms of severityof psychopathology, other clinical features, and social factors that influenceself-selection to treatment and clinical research. The use of 2 nights inthe sleep laboratory and standard measurement of daytime sleepiness in thislarge sample are noteworthy. Our statistical analysis that used data on the2 nights and tested interactions among variables maximized statistical powerto detect group differences. An important strength is the longitudinal natureof the epidemiological study, with multiple assessments of the occurrenceof traumatic events. As a result, the classification of the subset with nohistory of exposure to trauma in lifetime is exceptionally strict, especiallyin view of the common occurrence of exposure to traumatic events in recentpopulation studies and the high proportions of persons with history of multipletraumas.53
Corresponding author and reprints: Naomi Breslau, PhD, Departmentof Epidemiology, College of Human Medicine, Michigan State University, B645West Fee Hall, East Lansing, MI 48824 (e-mail: email@example.com).
Submitted for publication September 4, 2003; final revision receivedOctober 29, 2003; accepted November 18, 2003.
This study was supported by grant NIH-48802 from the National Instituteof Mental Health, Rockville, Md.