Introduction
Although suicidal cognitions are a common feature of major depression, they are also prevalent among people seeking health care (Scott et al.
2012) and observed to some degree in general population samples (Crosby et al.
2011). Interestingly, people differ in the way they experience and react to such cognitions, and whilst for some people suicidal thoughts are fleeting and temporary (Nock et al.
2009), for others they tend to persist and cause distress. Likewise, even in people with severe recurrent depression, the degree of distress that suicidal thoughts and images evoke varies markedly (e.g.; Crane et al.
2014). We have previously suggested that vulnerable individuals’
relationship with and
responses to suicidal cognitions are critical in determining whether suicidal cognitions persist and potentially escalate and that such persistence is most likely when individuals respond to suicidal thoughts with rumination and suppression (Williams et al.
2016). In line with this hypothesis Pettit et al. (
2009) have found that suppression of suicidal thoughts increases their severity over time.
Although it is established that differences in response to suicidal cognitions exist in patient and general population samples, relatively little is known about the factors that determine these. However, models of Obsessive–Compulsive Disorder (OCD); (e.g., Clark
1999), in which distressing and intrusive thoughts are core symptoms, may suggest potential candidate mechanisms. One relevant concept arising from the OCD literature is
Thought–Action Fusion (TAF); (Rachman
1993; Shafran et al.
1996; Craig and Lafreniere
2016). TAF was introduced by Rachman (
1993) to describe a phenomenon in which a person believes that the mere presence of intrusive thoughts can influence events in the real world. Rachman (
1993) describes two components of TAF: First, ‘morality TAF’ describes the tendency to assume that the occurrence of certain intrusions implies immorality of character (e.g., intrusions about killing equal being a bad person). Second, ‘likelihood TAF’ describes the view that the mere presence of thoughts has real life consequences and could, for example, increase the likelihood of catastrophic events (e.g., intrusions about killing somebody make it more likely it will happen). TAF has been proposed to promote engagement in strategies (e.g., suppression, worry, rumination, neutralising behaviours) intended to control intrusions, prompted by the perceived probability of the thought content happening, and/or by the misinterpretation that a person is responsible for harm unless they take action to prevent it (Rachman
1993; Rassin et al.
2000). Importantly, however, since attempts to suppress unwanted thoughts have been found in many cases to lead to a paradoxical increase in their frequency (Abramowitz et al.
2001) and intensity (Wegner
1994), TAF is likely to contribute to the escalation of symptoms over time. Indeed, Rassin et al. (
1999) found that experimentally induced TAF led to an increase in intrusions and perceived discomfort in healthy controls.
Whilst several studies have found a strong relationship between OCD and TAF (e.g., Amir et al.
2001; Shafran et al.
1996), some studies have shown that when depression is controlled for, the association between OCD and TAF is no longer significant (O’Leary et al.
2009; Jonsson et al.
2011). Nevertheless, to our knowledge, only two studies have examined TAF in depressed samples (Abramowitz et al.
2003; Meyer and Brown
2013). Abramowitz et al. (
2003) compared TAF in OCD samples to other anxiety disorder samples, clinically depressed patients and healthy controls. They found that OCD patients had higher likelihood-TAF for others-related events compared to depressed patients, patients with social phobia and healthy controls, but not higher likelihood TAF for self-related events. However, there were no overall differences across groups on Morality TAF. Inferences about group differences across depressed and OCD samples were complicated by the OCD sample scoring higher on the Beck Depression Inventory (BDI) than the depressed sample. However, in this study depression was secondary to the focus on anxiety disorders, and the sample of depressed patients was small (n = 19). Meyer and Brown (
2013), in contrast, examined TAF in a large clinical sample including both depressed and anxious patients. Using the TAF scale developed by Shafran et al. (
1996), they found that both global TAF and TAF-Likelihood were more strongly associated with obsessive–compulsive symptoms than with measures of depression and worry, suggesting that to the extent that TAF is observed in depressive populations it may be present primarily where there is also thought content which has an obsessive quality.
One reason for thinking that TAF may be relevant to suicidal ideation is that it has previously been suggested that suicidal thoughts present as a form of rigid rumination (Kerkhof and van Spijker
2011; see also; Rogers and Joiner
2017), the engagement with which can be characterised as obsessive. Further, the distressing and often graphic content of suicidal urges (e.g., ‘I could hang myself in the back garden’) could be seen to have more in common with obsessions as seen in OCD (e.g., losing control and harming oneself or others), than depressed non-suicidal thoughts (e.g., ‘I’m no good’). However, the strong ruminative nature of non-suicidal depression (e.g., Liu et al.
2017) indicates that TAF might also be relevant here. Thus, there appears to be merit in exploring the potential commonalities between OCD intrusions and suicidal cognitions, and to compare this with non-suicidal depression. It is also possible that one possible reason why existing studies of TAF in depression have not yielded strong findings, is that the standard measure to assess TAF, TAF-R (Shafran et al.
1996), was developed to assess TAF in OCD, and so may be less sensitive to identify TAF in other domains, such as in depression and suicidality. Thus, it is possible that TAF may also be observed in people with depression, particularly where there are thoughts with an obsessive quality, if items were appropriately worded to elicit it.
If TAF is present in people suffering from suicidal depression what impact would this have for symptom maintenance and exacerbation? In OCD, it is suggested that TAF increases the distress associated with intrusive thoughts and compounds attempts to neutralise them (Rassin et al.
1999; Rachman
1997). However, there is no suggestion that TAF increases the actual likelihood of enactment of such thoughts. Likewise, it might be suggested that TAF related to suicidal thoughts would similarly increase distress and cognitive preoccupation, but would not increase risk of enactment. However, suicidal cognition is characterised not only by explicit thoughts and images related to suicidal acts, but also to broader cognitions including those relating to perceived burdensomeness (e.g., ‘my death is worth more than my life to others’; Van Orden et al.
2010) and entrapment (‘nobody can help me out of this mess’; Williams
2014). Although entirely hypothetical, it is possible to speculate that if TAF also operated in relationship to these thoughts, the responsibility they evoke (e.g., to unburden the family) might serve as a volitional moderator (O’Connor
2011), reinforcing and promoting suicidal distress, and facilitating the progress from thoughts to acts. To the best of our knowledge, however, no studies have systematically examined TAF in suicidal depression, nor compared TAF in this population to individuals with a history of non-suicidal depression and healthy controls.
If TAF is present in people with recurrent suicidal depression, a second issue concerns its mood dependence. Cognitive science accounts of recurrent depression (i.e., the Differential Activation framework) posit that in people with a history of recurrent depression, depressive and suicidal thoughts can be reactivated by subtle and transient changes in mood (Williams et al.
2016; Scherrer et al.
2014; Brockmeyer et al.
2012). Indeed, mood-dependent changes in dysfunctional thinking have been shown to predict depressive relapse in previously depressed patients (Segal et al.
2006). Such reactivation involves not just negative
content but also maladaptive
cognitive processes. For example, our previous work has shown that people with a history of suicidal depression can be distinguished from those with a history of non-suicidal depression in the extent to which increases in negative mood impair interpersonal problem solving, a cognitive deficit characteristic of people at times of suicidal crisis (Williams et al.
2005). Likewise, people with a history of suicidality report more hopeless thoughts when asked to imagine being in a slightly sad mood, and this measure of hopelessness reactivity predicts actual cognitive deficits in positive future thinking after an experimental mood challenge (e.g.; Williams et al.
2008). Some forms of mood-dependent dysfunctional thinking (e.g., entrapment) have also been shown to increase the risk of suicidal behaviour longitudinally (O’Connor et al.
2013). If TAF is also observed in people with a history of recurrent suicidal depression, a first step in examining whether this bias predicts subsequent suicidal relapse is to investigate whether such potential endorsement of thoughts increases in the context of transient low mood. Identifying factors uniquely affected by changes in mood is helpful in understanding what serves to maintain and progress suicidal crises, and in formulating potential clinical targets. Indeed, although as stated above there is relatively little research exploring TAF in depression, Shafran et al. (
1996) suggest a reciprocal relationship between mood disorders and TAF, in which depression might increase the occurrence and believability of thoughts, and in which TAF, in turn, might exacerbate low self-esteem, depression and anxiety.
Thus, there appears to be a strong rationale for exploring the potential presence of TAF in individuals with a history of suicidal depression, including the extent to which TAF might be mood-state dependent, and elevated at times of low mood or crisis. However, since the original TAF scale was developed for OCD, for the present study we considered that a revision of the items would be required in order to make the scale more amenable to depression and suicidality.
Assessment of Thought–Action Fusion
The first version of the Thought Action Fusion Scale (Shafran et al.
1996) was explicitly developed for use with OCD patients, and consisted of 34 items, which, following factor analyses and revisions, led to the 19 item Thought–action fusion scale-revised (TAF-R). This scale consisted of three subscales all denoting negative events: TAF-moral (12 items), TAF-likelihood-self (i.e., TAF likelihood for self-related events; 3 items) and TAF-likelihood-other (i.e., TAF likelihood for self-related events: 4 items). Overall, the TAF literature shows that the three factors of TAF-R have good reliability, as reflected in Cronbach’s alphas ranging from 0.75 to 0.89 (Rassin et al.
2001).
Whilst it seems clear that patients suffering from anxiety disorders report higher TAF than healthy controls on the probability dimension (e.g., Rassin et al.
2001), comparisons with other clinical groups are sparse. Two revisions of the original scale have since been made; one including TAF for thoughts about
positive events happening to self and others, across domains of both gain and harm avoidance (Craig and Lafreniere
2016), and one including items about likelihood of preventing harm by means of positive thoughts (Amir et al.
2001). To our knowledge, no TAF scale has been developed to specifically assess TAF for depression or suicidality.
Shafran et al. (
1996) have noted that responses on the TAF tended to be idiosyncratic, and that the scale would be most usefully considered as a clinical tool to identify idiosyncratic beliefs, rather than generating sum-scores. Moreover, they emphasised that the construct of TAF was in its early stages and that associated metacognitions, specificity to OCD, and clinical implications were still unclear. Furthermore, with the notable exception of Craig and Lafreniere’s (
2016) examination on TAF for positive events, the TAF literature has focussed solely on negative events (e.g., being in a car accident, either self or other; Shafran et al.
1996). Another important dimension which has not been assessed in existing versions of the TAF-R but which might influence the extent to which TAF is observed is the controllability of the event to which a thought relates. There is some evidence that thinking about an event increases the likelihood of engaging in that behaviour at least for desirable actions (e.g., imagining to vote and subsequently doing so; Libby et al.
2007), and thus there is to some degree a reality to the belief that thinking about a potentially self-initiated action may increase its probability. In contrast, there is no evidence that thinking about uncontrollable actions has any bearing on one’s behaviour. As such it is possible that pathological TAF may be most easily observed when it relates to the likelihood of other-initiated actions (e.g., friend committing a crime) or self-relevant uncontrollable actions (e.g., being called to Jury duty) for which there is no logical mechanism through which thought might lead to action. Controllability may be a particularly relevant aspect of TAF in the context of both suicidal and non-suicidal depression given the emphasis on locus of control in depressive and suicidal symptomatology. Thus, both the self-other and controllable-uncontrollable dimensions appear important in understanding the phenomenon in the context of recurrent suicidal and non-suicidal depression. Finally, in terms of convergent validity, despite the theoretical emphasis on TAF eliciting control strategies, little is known about relationship of TAF to relevant cognitive strategies other than obsessionality, such as suppression, rumination as well as a divergent construct, trait mindfulness (i.e., the capacity to pay attention to momentary experience with acceptance and non-judgement, Baer et al.
2006,
2008).
The objectives of the present study were therefore to examine TAF in suicidal and non-suicidal depression as compared to healthy controls, and also to examine the effect of a mood induction within the two depressed groups. To address the above issues concerning the measurement of TAF we (a) developed and validated a TAF scale adapted to include items whose content was relevant to suicidal depression (including items with both positive and negative content, including suicide related content), and which included items covering both self-other and controllable-uncontrollable dimensions using exploratory and confirmatory factor analysis, (b) compared the derived TAF scale in individuals with a history of suicidal depression and non-suicidal depressed compared to non-clinical controls, and (c) examined whether TAF increased in the depressed groups when these groups were asked to respond from the perspective of a time of crisis, as compared to their baseline responses. We hypothesised (1) that total TAF score would correlate positively with suppression and rumination and correlate negatively with facets of trait mindfulness across the sample as a whole, (2) that total TAF would be related linearly to lifetime depression severity (i.e., increased in depressed controls compared to healthy controls and highest in individuals with a history of suicidal depression), (3) that if a factor emerged that reflected item content related to suicidality that this factor would be higher in individuals with a history of suicidal depression than in non-suicidal depressed and healthy controls and compared to other factors for this group due to its salience, and (4) that TAF would increase for clinical groups when reported in the context of low mood as compared to baseline.
Results
Participant Characteristics
The sample consisted of N = 361 individuals, consisting of 130 HC, 134 D-NS, and 97 D-S individuals (the random subsamples consisted of N = 181 and N = 180, respectively). Baseline characteristics of the full sample and subgroups are shown in Table S1. Gender, education, ethnicity, and employment differed significantly across subsamples, with significantly higher number of men in the depressed non-suicidal group (p < 0.05).
Factor Analysis
The number of factors obtained by principal axis factor analysis was cross-validated using the Kaiser criteria (3 factors > eigenvalue of 1), a scree plot (3 factors above break), and parallel analysis (4 factors > 95th percentile of simulated eigenvalues; see Fig. S1). Although parallel analysis diverted from a 3-factor solution, this solution was taken forward since the difference between observed and simulated eigenvalues for the fourth factor was negligible. EFA of this 3-factor solution with maximum likelihood estimation and promax rotation revealed a relatively clear cut factor structure (see Table
1): Factor 1 (Uncontrollable TAF) combined 24 items describing uncontrollable events related to either self and/or other (e.g., being in a plane crash, winning the lottery, others’ self-harm), Factor 2 (Suicidal TAF) combined three items that were related to own self-harm or suicide (i.e., not others’ self-harm) and were hence controllable (logically, if not phenomenologically), and Factor 3 (Controllable TAF) combined three items with self-related positive content that were also controllable (e.g., myself having fun on a holiday). The percentage of variance explained by these factors was 44.1, 7.5, and 5.8 for factors 1, 2, and 3, respectively.
The number of items was reduced from 30 to 20 items based on low factor loadings (5 items below 0.5) as well as keeping with item mirroring in the self-other category (5 corresponding items; see Table
1). All item reductions were from the uncontrollable TAF subscale.
Next, CFA was conducted on the reduced 20-item version of the TAF-SR. More specifically, a model with three correlated latent factors, where all 14 items describing uncontrollable events loaded on one factor, the three items related to self-harm loaded on another factor, and the three items describing controllable positive events loaded on a third factor. Results indicated good model fit; χ2(167) = 250.52, CFI = 0.95, TLI = 0.95, RMSEA = 0.053 (90% CI 0.039–0.066) of the three-factor solution. We compared the three-factor model to a single factor model where all items loaded on one overall TAF factor. However, fit for this model was not adequate; χ2(170) = 429.18, CFI = 0.85, TLI = 0.83, RMSEA = 0.092 (90% CI 0.081–0.103).
Convergent and Discriminant Validity
Cronbach’s α and Omega Total showed good to excellent internal validity for total TAF (α = 0.94, ω = 0.94), uncontrollable TAF (α = 0.96, ω = 0.96), and self-suicidal TAF (α = 0.86, ω = 0.87). Internal consistency was only moderate for positive controllable TAF (α = 0.68, ω = 0.69).
Associations between TAF and other scales are presented in a correlation matrix in Table S3. Total TAF score, uncontrollable TAF, and self-suicidal TAF correlated moderately with RRS. Looking at RRS subscales (Table S4) revealed somewhat stronger correlations between TAF scales and the Reflection and Depression subscales as compared to the Brooding subscale (z = 2.84, p < 0.01). Correlations to other scales (BRFL, WBSI, SMQ, FFMQ) was low. However, subscale analyses for FFMQ and WBSI yielded somewhat different findings. There was a significant moderate correlation between the Describing subscale of the FFMQ and TAF total (r = .4) and TAF self-suicidal (r = .32) (Table S4). The correlation matrix (Table S4) indicates that TAF was moderately correlated with WBSI: Unwanted Intrusive Thoughts (TAF total: r = .38, TAF Uncontrollable: r = .4) but not with the other WBSI subscales Thought Suppression (r = − .05) and Self-Distraction (r = − .08). Moreover, WBSI-Unwanted Intrusive Thoughts was moderately correlated with TAF self-suicidal (r = .31).
TAF Across Groups
Results from multiple regression analyses in conjunction with permutation tests revealed significant differences in TAF between groups (see Table S5). The D-NS group had significantly higher total TAF, uncontrollable TAF, self-suicidal TAF, and lower positive controllable TAF when compared with healthy controls. Contrary to our hypothesis, the D-S group did not show differences to HCs on overall TAF, uncontrollable TAF, and positive controllable TAF, but did show significantly higher self-suicidal TAF.
Differences Between Normal and Crisis State
Finally, we tested whether the two clinical groups differed in how their TAF scores changed from normal to crisis states using permutation-based linear regression of outcome crisis state TAF with predictors being dummy-coded group, normal state TAF, and their interaction. D-NS individuals increased significantly more than D-S individuals from normal to crisis states on total and uncontrollable TAF scores. In contrast, the D-S group increased significantly more for suicidal TAF and decreased significantly more for the TAF positive controllable (Table S6).