Night-time rumination in PTSD: development and validation of a brief measure

ABSTRACT Background: Pre-sleep cognitive activity and arousal have long been implicated in the maintenance of insomnia. However, despite high comorbidity between insomnia and posttraumatic stress disorder (PTSD), pre-sleep thoughts in PTSD and their associations with disturbed sleep, have not yet been investigated. Objective: This study presents the development and preliminary validation of a brief self-report measure of the content of trauma-related pre-sleep thoughts: the Trauma Thoughts before Sleep Inventory (TTSI). Methods: Participants (N = 285) were recruited online into five groups: three groups with clinical symptoms, 1) PTSD; 2) depression without PTSD; 3) insomnia without depression or PTSD; and two healthy control groups 4) nontrauma-exposed controls; 5) trauma-exposed controls. The questionnaire was administered at baseline, and for a subsample (n = 157) again one week later to assess test-retest reliability. At baseline, participants also completed questionnaires of sleep quality, PTSD and depression symptoms, and insomnia-related thoughts. Results: The TTSI had good reliability and validity; it discriminated participants with PTSD from those with depression and insomnia, those with depression from insomnia, and correlated with existing measures of pre-sleep thoughts, self-reported pre-sleep arousal and poor sleep. Conclusions: The results support the utility of the TTSI for measuring thoughts that keep people with PTSD awake, although replication in an independent clinical sample is required.


Introduction
There is good evidence for the role of pre-sleep cognitions in maintaining insomnia (see Harvey, 2002). For example, 'an over-active mind' is one of the most common reasons given by insomniacs to account for poor sleep (Espie, Lindsay, Brooks, Hood, & Turvey, 1989), and excessive pre-sleep cognitive activity has been associated with longer sleep onset latency and shorter sleep duration (Gross & Borkovec, 1982). Further, there is evidence that while trying to sleep, insomniacs are more pre-occupied with thoughts about getting to sleep and the consequences of poor sleep (Wicklow & Espie, 2000), as opposed to the 'nothing in particular' commonly reported by good sleepers (Harvey, 2010). Pre-sleep thoughts of this kind as well as those of 'rehearsal and planning' (e.g. 'what happened today and what I've got on tomorrow') (Espie et al., 2014) predict longer sleep onset latencies (Harvey & Espie, 2004), and are more frequent in insomniacs versus controls (Harvey & Espie, 2004). Accordingly, most models of insomnia propose that cognitive processes such as cognitive arousal, worry and maladaptive beliefs about sleep and its consequences play a role in maintaining insomnia (Espie et al., 1989;Lundh & Broman, 2000).
Insomnia is highly prevalent in PTSD (Ohayon & Shapiro, 2000) and is implicated in PTSD development and maintenance (see Babson & Feldner, 2010). It has been suggested that factors known to perpetuate general insomnia may also contribute to sleep impairment in PTSD (Zayfert & DeViva, 2004). Pre-sleep thoughts may be one candidate factor, but have not yet been directly investigated in PTSD. However, there are some findings that point to a role of fear-related cognitions. For example, studies have found that fear of sleep (Kanady et al., 2018;Pruiksma et al., 2014;Short, Allan, Stentz, Portero, & Schmidt, 2018) linked to nightmares (Davis et al., 2011;Krakow, Tandberg, Scriggins, & Barey, 1995), and fear of loss of vigilance (Pietrzak, Morgan, & Southwick, 2010) are associated with worse sleep and more severe PTSD symptoms in trauma survivors. These findings are consistent with relationships proposed in the maintenance of PTSD symptoms by Ehlers and Clark's (2000) cognitive model of PTSD. This model proposes that trauma-related appraisals related to impending danger (such as 'I will be attacked again') or negative interpretations of symptoms (such as 'If I go to sleep I will not notice intruders' (p. 330) or 'Not sleeping will damage my body') motivate the use of dysfunctional coping strategies such as staying up late because of fear of nightmares, or ruminative thoughts about the trauma, or repetitive thinking about one's emotional state (similar to brooding about feeling sad in depression, Nolen-Hoeksema, 1991), which may then interfere with sleep onset and quality (Ehlers & Clark, 2000). Consistent with this, in non-clinical populations, maladaptive repetitive thoughts have been linked to more severe sleep disturbances (Cox, Ebesutani, & Olatunji, 2016;Nota & Coles, 2015), and shown to mediate the relationship between depressed mood and sleep quality (Slavish & Graham-Engeland, 2015). Studies in clinical populations of individuals with Generalised Anxiety Disorder have also shown that the frequency of evening worry predicts subsequent sleep quality, and vice versa (Thielsch et al., 2015), and rumination following a stressor has been found to predict increased self-reported and objective sleep onset latency (Zoccola, Dickerson, & Lam, 2009). Finally, in trauma survivors, rumination has been linked to worse sleep problems and PTSD symptoms, and is suggested to contribute to impaired sleep (Borders, Rothman, & McAndrew, 2015). It is thus possible that traumarelated pre-sleep thoughts (such as appraisals of being vulnerable while asleep, or repetitive thoughts about how the trauma could have been prevented or about negative feelings) play a role in maintaining sleep problems in PTSD. These trauma-related pre-sleep thoughts could occur alongside pre-sleep thoughts already evidenced in insomnia populations (such as worries about getting enough sleep) (Harvey, 2002;Harvey & Espie, 2004), and both could contribute to poor sleep in PTSD.
The Glasgow Content of Thought Inventory (GCTI; Harvey & Espie, 2004) was developed to assess pre-sleep thought patterns in insomnia populations. Items were generated from pre-sleep thoughts reported by people with insomnia, and so might not assess the full range of pre-sleep thoughts that could interfere with sleep in PTSD. As no measure currently exists to assess presleep thought patterns that are specific to PTSD, it would therefore be of interest to develop such a measure, and to investigate its association with disturbed sleep in PTSD.

Aims
This study describes the development and initial validation of a new measure designed to assess traumarelated pre-sleep thought patterns that may interfere with sleep in people with PTSD; the Trauma Thoughts before Sleep Inventory (TTSI).
The specific aims of the present study were to 1) develop a measure of trauma-related pre-sleep thoughts, 2) to assess the psychometric properties of the new measure, and 3) to investigate whether trauma-related presleep thoughts discriminate people with PTSD from those with insomnia or depression, and from traumatised and non-traumatised controls. Due to the high comorbidity between PTSD and depression (Kessler, Chiu, Demler, & Walters, 2005), and the high prevalence of insomnia in depression (Riemann, Berger, & Voderholzer, 2001) individuals with depression were included as a comparison group to determine whether the trauma-related pre-sleep thought patterns were specific to PTSD.

Methods
An online study evaluated the reliability of a new questionnaire of trauma-related pre-sleep thoughts and examined its associations with established measures of sleep disturbance, arousal, and PTSD symptoms and its specificity to PTSD compared to depression and insomnia.

Participants
A sample of 285 participants aged 18-65 (M = 30.32 years, SD = 11.37) took part in the study between 2013 and 2016 (see Table 1 for demographics). This included 159 women (55.8%). Inclusion criteria for all groups were that they were aged 18-65, could read and write in English, reported no present or past diagnosis of psychosis or bipolar disorder, reported no present substance or alcohol dependence, and were not currently receiving psychological therapy (as this may have affected the test-retest reliability analyses).
Participants were recruited by adverts and sent a screening questionnaire including measures of trauma exposure, PTSD symptoms, depression, insomnia and general questions about their mental health. On the basis of the self-reported symptoms in standardised questionnaires, participants were divided into one of five groups. Three groups reported clinically significant symptoms, using empirically established cut-offs: (1) PTSD: These participants reported at least one traumatic event and a PTSD symptom severity of 18 or above (Ehring, Kleim, Clark, Foa, & Ehlers, 2007)  Two control groups: (1) Controls: These participants reported no trauma and no depression, insomnia, PTSD, Age ( and did not report any other mental health diagnoses (n = 80).
(2) Trauma controls: These participants reported at least one trauma, but did not report depression, insomnia, PTSD, and did not report any other mental health diagnoses (n = 77).
Of 420 advert respondents, 354 people (84.29%) were suitable for one of the groups, and 285 (80.51%) completed the first online TTSI questionnaire. One week later, 157 of these participants (55.10%) completed the questionnaire again to assess test-retest reliability (Table 1).

Initial development of the trauma thoughts before sleep inventory
The item-pool for the Trauma Thoughts before Sleep Inventory (TTSI) was developed from the literature, and through discussions with PTSD patients and clinical psychologists experienced in the treatment of PTSD. Respondents rated how often each thought had kept them awake (when they were trying to fall asleep, or when they had woken in the night) over the previous week, on a scale from 0 (Not at all) to 3 (5 or more times a week/nearly every day). The initial pool of items was refined through informal interviews and feedback after completion of the scale by a small sample of PTSD patients, healthy trauma survivors, and healthy controls, to check for understanding and repetition. The final item pool comprised 9 items representing empirically and theoretically supported possible trauma-related pre-sleep thoughts (e.g. 'being vulnerable if I sleep'). The total score is the sum of the items (range: 0-27); higher scores indicated greater frequency of pre-sleep thoughts (see Figure 3 for scale).

Validation measures
To examine the validity of the TTSI, participants also completed the following measures: 2.3.1. Pre-sleep cognitive activity and arousal Glasgow Content of Thoughts Inventory (GCTI; Harvey & Espie, 2004); The GCTI is a 25-item measure of pre-sleep thoughts that have been shown to be linked to insomnia. Respondents rate how often each thought kept them awake over the previous 7 days on a 4-point scale from never (= 1) to always (= 4) (range: 25-100). The scale has good psychometric properties (Cronbach's α = .87; Harvey & Espie, 2004). To test whether TTSI items constitute a separate factor from general insomnia-related thoughts, 13 non-trauma related items were adapted from the GCTI (see Table 3) and rated on the same rating scale as the TTSI.

Sleep symptoms
Pittsburgh Sleep Quality Index (PSQI; Buysse, Reynolds, Monk, Berman, & Kupfer, 1989) The PSQI is a widely used 19-item scale, which assesses seven components of sleep quality and disturbances over the previous month. Items and sub-scales are scored from 0 (no difficulty) to 3 (severe difficulty) and summed to produce a PSQI global score (range: 0-21; higher scores indicate more severe global sleep disturbances). A global score of greater than 5 can be used to indicate clinically significant sleep disturbances (Buysse et al., 1989). To assess time taken to fall asleep, PSQI Component 2 was used (Sleep Onset Latency (SOL); range: 0-3). The PSQI Addendum for PTSD (PSQIA) (Germain, Hall, Krakow, Shear, & Buysse, 2005) was also included as a measure of the severity of PTSD related sleep disturbances over the past month (range: 0-21).

Trauma exposure and PTSD symptoms
Life Events Checklist (LEC). The LEC (Gray, Litz, Hsu, & Lombardo, 2016) is a self-report questionnaire containing a list of stressful and traumatic (according to DSM-5 criteria) life events. Respondents indicate whether they have experienced each event in their lifetime. Participants also answered questions about the worst event they had experienced, to determine if it met DSM-5 criteria for a traumatic event (Criterion A; American Psychiatric Association, 2013).
Posttraumatic Diagnostic Scale (PDS) (Foa et al., 1997). The PDS is a validated self-report measure assessing the severity of DSM specified PTSD symptoms over the previous week, rated on a scale from 0 (not at all) to 3 (5 or more times a week/nearly every day). A cut-off of 18 has been found to show the greatest diagnostic accuracy in predicting a PTSD diagnosis according to the Structured Clinical Interview for DSM-IV (Ehring et al., 2007). In the present study, the Clinician Administered PTSD Scale for DSM-5 (CAPS; Weathers et al., 2013) was administered by one of the authors (EW or JS) at Week 1 for a sub-set of trauma controls (n = 36) and participants with PTSD (n = 40). Cohen's kappa showed 'very good' agreement (94.29%) between the CAPS and the PDS cut off, κ = .87, p < .001.

Screening and allocation
Participants were recruited via different online adverts in the community in Oxford and at the University of Oxford. Adverts invited people with either (1) difficulties falling or staying asleep, (2) low mood, (3) experience of a trauma, or (4) healthy controls to contact the research team. Advert respondents completed online screening questionnaires, which consisted of the LEC (Gray et al., 2016) to identify trauma exposure, the PDS (Foa et al., 1997) to assess PTSD symptoms (for those who reported a traumatic event), and questions about their current and past mental health. Participants responding to adverts for low mood or sleeping problems (in the past two weeks) also completed the PHQ-9 (Kroenke et al., 2001), and the ISI (Bastien et al., 2001) as screening measures for depression and insomnia, respectively. Suitable participants were allocated to a group and emailed the online baseline questionnaire pack.

Baseline questionnaire
Participants completed the questionnaires in a fixed order. First, participants completed the LEC again and identified the worst event they had experienced from those reported, then answered questions to determine if the event met criteria for a traumatic event (Criterion A) according to DSM 5 criteria. Participants were then asked to answer the PTSD symptom measure (PDS) with the 'traumatic event' they had identified in mind. They then completed the PSQI (and addendum), GCTI, PSAS and the TTSI. Finally they completed the BDI and GAD-7 to assess mood and anxiety. Non-trauma-exposed participants were asked to answer the PDS, TTSI and PSQIA with the 'worst event' they had identified in mind, and the wording was amended to 'the event' rather than 'the trauma/traumatic event/experience'.

One-week questionnaire
After completing the baseline questionnaire pack (Week 1), the TTSI was emailed again one week later (Week 2), to examine test-retest reliability.

Analysis
Analyses were conducted using SPSS version 22. Data was checked for skewness and kurtosis, and homogeneity of variance, and parametric and nonparametric tests were used where appropriate.

Reliability
Internal consistency was evaluated with Cronbach's alpha (α > 0.8 indicates good internal consistency; Nunnally & Bernstein, 1994), calculated for the entire sample and the PTSD group. For participants who completed the TTSI a second time, the mean sum score at Week 1 and Week 2 was correlated (intraclass correlation) to examine test-retest reliability.

Validity
Factorial validity of the new measure was evaluated with Exploratory Factor Analysis (Principal Axis Factoring), to explore whether pre-sleep thoughts related to insomnia and trauma were represented by two separate factors in the PTSD sample only. The Kaiser-Meyer-Olkin criterion (0.84) and the Bartlett test (χ 2 (231) = 822.11, p < .001) indicated that data were highly suitable for factor analysis. Factors were rotated using oblique rotation (Oblimin method). Factor extraction was supported by parallel analysis (Horn, 1965), which suggests extracting factors whose eigenvalues are bigger than 95% of random eigenvalues, taking into account the sample size and number of items. Item retention was based on factor loadings larger than 0.4 and differences between primary and secondary loadings not smaller than 0.2, to determine the most interpretable factor solution (Matsunaga, 2015).
To examine convergent validity, the new measure was correlated with existing self-report measures of pre-sleep cognitive activity in insomnia (GCTI); sleep quality (PSQI); sleep onset latency (SOL; PSQI Component 2), and pre-sleep arousal (PSAS). Presleep cognitive activity (GCTI) in patients with insomnia (Harvey & Espie, 2004) has been previously correlated with increased pre-sleep mental arousal (PSAS-cog) and increased self-reported sleep onset latency (Harvey & Espie, 2004), therefore these arousal measures were also used to evaluate the construct validity of the TTSI. In trauma survivors only (n = 152), the TTSI was correlated with PTSD symptom severity excluding the sleep item (PDSwos).
To investigate criterion validity, the ability of the TTSI to discriminate between experimental groups was examined using multinomial regressions. For comparison, the ability of the GCTI to discriminate between groups was also examined.

Logistic regression
The contribution of trauma-and insomnia-related presleep thoughts to poor sleep in trauma survivors was examined using logistic regression. Sum scores from the TTSI and GCTI were entered as continuous predictors of sleeper status according to the PSQI, where a global score greater than 5 was taken to indicate poor sleep.

Factorial Validity
Principal axis factor analysis was conducted on TTSI items together with the 13 insomnia-related items adapted from the GCTI (Table 3), using the PTSD sample (n = 57). Parallel analysis suggested extracting two factors, which explained 54.45% of the overall variance. Factor 1 (Eigenvalue = 9.78, Percentage of explained variance 44.45%) could be interpreted as insomnia-related thoughts (GCTIadapted items) and Factor 2 (Eigenvalue = 2.20, Percentage of explained variance 10.00%) as trauma-related thoughts (TTSI items) (see Table  3). The two factors correlated highly, r = .54, p < .001. Item 3 ('noises I hear') did not substantially load on any of the two factors. No substantial cross-loadings were observed. Significance level of group difference indicated by: * p < .05; ** p < .01, *** p < .001.

Group comparison: glasgow content of thought inventory
For comparison, groups were compared on the GCTI (see Figure 2). Greater scores on the GCTI were associated with a higher likelihood of belonging to the PTSD group compared to all the other groups (non-exposed controls, b (SE) = 0.13 (0.02), exp. b = 0.85, p < .001; trauma-exposed controls, b (SE) = 0.12 (0.02), exp. b = 0.85, p < .001; insomnia,   p > .05, and between the trauma-exposed and nonexposed control group, b (SE) = 0.01 (0.02), exp. b = 1.00, p > .05. Effect sizes of the significant group differences were small to medium.

Predicting poor sleeper status in trauma survivors
A logistic regression was performed to identify the effects of insomnia (GCTI) and trauma-related thoughts (TTSI) on poor sleep in trauma survivors  (n = 77). The model was statistically significant χ 2 (2) = 91.05, p < 0.001, and correctly classified 82.1% of cases, Nagelkerke R 2 = .61. Both greater traumarelated and insomnia-related thoughts were associated with increasing likelihood of poor sleep (Table 4).

Discussion
This study presented the development and initial validation of a measure designed to assess traumarelated thoughts that may interfere with sleep (e.g. 'Being vulnerable if I sleep', or 'dwelling on how my life has been changed by the event'). It was also explored whether these thoughts discriminated people with PTSD from those with insomnia and depression, to determine the TSSI's utility as a tool to identify trauma-and PTSD-related pre-sleep thought content, which might contribute to poor sleep in PTSD. The TSSI had good psychometric properties, including good test-retest reliability and internal consistency, and good validity, correlating highly with the GCTI, an existing validated measure of pre-sleep cognitive activity in insomnia (Harvey & Espie, 2004).
The TTSI showed good convergent validity. Participants with a greater frequency of traumarelated thoughts before sleep reported higher presleep arousal on the PSAS (both somatic and cognitive) and more severe sleep difficulties (longer sleep onset, and worse sleep quality) on the PSQI. This is consistent with previous findings that presleep cognitive activity is associated with insomnia (Harvey, 2002(Harvey, , 2003(Harvey, , 2010Harvey & Espie, 2004), and interferes with sleep by prolonging sleep onset latency (Gross & Borkovec, 1982). The results of this study could similarly suggest that in PTSD there are trauma-related thought patterns associated with elevated pre-sleep arousal and worse sleep (perhaps by prolonging sleep onset). However, the present results do not indicate a causal relationship.
The TTSI also showed good criterion validity. The PTSD group reported more frequent traumarelated thoughts before sleep than any other group. The TTSI discriminated people with PTSD from other clinical groups associated with poor sleep (depression and insomnia). It also discriminated depression from insomnia, whereas the GCTI did not. This could be explained by a larger number of trauma survivors in the depression versus insomnia group, who may be more likely to ruminate on the trauma before sleep, but do not differ on PTSD-related sleep complaints (PSQIA) and PTSD symptoms compared to the insomnia group. It further suggests that, in addition to PTSD, the TTSI could also be relevant to understanding the symptom experience in other trauma and stressor-related disorders, and therefore may have wider clinical applications; for example with individuals with depression following a traumatic stressor.
More frequent insomnia-related thoughts before sleep (GCTI) were also more likely to be associated with PTSD than the other groups, although to a lesser extent than trauma-related thoughts. People with depression and insomnia reported more frequent insomnia-related thoughts than both control groups, and did not differ from each other. This suggests that poor sleep in the presence or absence of a co-morbid mental health problem is associated with more frequent pre-sleep thoughts about sleep and its consequences. Trauma-exposed and non-exposed controls did not differ in the frequency of either insomnia or trauma-related thoughts, suggesting that traumaexposure alone is not related to increased pre-sleep thoughts of any kind.
Factor analysis showed that insomnia and traumarelated thoughts were reflected by separate factors. This suggests that in PTSD, although correlated, trauma-related pre-sleep thoughts are distinct from insomnia-related pre-sleep thoughts. However, this interpretation is limited by the small sample size.
In trauma-survivors the TTSI showed good sensitivity and specificity to detecting poor sleeper status. In trauma survivors both insomnia-related and trauma-related pre-sleep thoughts increased the likelihood of being a poor sleeper.
In sum, the pattern of group differences suggests similarities and differences between pre-sleep thought content in PTSD and insomnia, i.e. thoughts about sleep and its consequences are common to both groups, and those with PTSD reported additional thoughts about the trauma, as did those with depression (but to a lesser extent than those with PTSD). Both types of content are related to poor sleep in trauma survivors, and both are reported more frequently in PTSD than other clinical groups with poor sleep. These findings are consistent with evidence that similar mechanisms may underlie insomnia with or without a mental health problem (Kohn & Espie, 2005), and with suggestions that insomnia alone and in PTSD may have both similarities and differences (Ulmer, Edinger, & Calhoun, 2011). The TTSI may provide a brief, clinically useful tool to assess the type and frequency of trauma-related pre-sleep thoughts that might be keeping people with PTSD awake.

Limitations
The present study had a number of limitations. Firstly, measures of arousal and sleep were selfreported rather than objective. Further investigation is required to determine whether TTSI scores are also related to objective measures of poor sleep and physiological arousal. Secondly, insomnia and depression groupings were based on self-reports and the results may be different for individuals with a clinical diagnosis, particularly those seeking treatment, who may be more severe. Finally the analyses were limited by the smaller clinical group sample sizes. There is a need for further evaluation of the scale using multi-group confirmatory factor analyses to test factorial invariance across clinical groups and across clinical and nonclinical groups. Future research could investigate the psychometric properties further in clinical, ideally treatment-seeking, populations, with larger samples.

Future directions
A number of experts have recommended including measures of pre-sleep thoughts, behaviours and arousal with sleep outcome measures when investigating the experience of insomnia (Kohn & Espie, 2005) and the effects of treatment on sleep disturbances (Harvey, 2002). Therefore future studies investigating the effect of insomnia or PTSD treatments on sleep problems in PTSD could aim to include measures of insomnia-related (e.g. GCTI; see Espie et al. (2014) for a 9-item version) and trauma-related (e.g. TTSI) pre-sleep thoughts. This would help determine whether insomnia and PTSD treatments, both of which improve sleep in PTSD (e.g. Belleville, Guay, & Marchand, 2011;Lommen et al., 2015;Talbot et al., 2014;Ulmer et al., 2011), also produce change in presleep thoughts. A possibility is that trauma-focused therapy for PTSD may improve trauma-but not insomnia-related pre-sleep thoughts, which could theoretically contribute to residual sleep problems after otherwise effective PTSD treatment (Gutner, Casement, Gilbert, & Resick, 2013;Zayfert & DeViva, 2004).

Conclusion
In conclusion, a brief measure to assess trauma-related pre-sleep thoughts was developed and validated. The TTSI had good psychometric properties, and discriminated people with PTSD from insomnia and depression, and was sensitive to detecting poor sleep in trauma survivors. The utility of the TTSI for measuring pre-sleep thought frequency and content in PTSD (and possibly in other stressor or traumarelated disorders such as depression) is supported, although replication of the psychometric properties in a clinical sample is required. Future studies could examine the contribution of pre-sleep thoughts to sleep problems in PTSD, and persistence after treatment.