Background

Post-traumatic stress symptoms (PTSS) include re-experiencing the traumatic experience, avoidance of stimuli associated with the traumatic event, negative alterations in cognition and mood, and hyperarousal.1 PTSS is frequently observed in those who have experienced catastrophic illnesses, rape, vehicle accidents or other life-threatening traumatic events.1 In the last 3 years, COVID-19 swept the world, causing more than six million deaths.2 Studies have illustrated that the rapid spread of infection and high mortality rate of COVID-19 were beyond expectations, causing traumatic experiences and PTSS in the survivors and the general public. For example, Kaubisch et al pointed out that the COVID-19 pandemic and associated measures pose a serious threat to mental health, and could be conceptualised as traumatic events.3 Numerous studies have explored the COVID-19-related PTSS. For example, in a review of 19 studies, researchers found that the prevalence of COVID-19-related PTSS ranged from 7% to 53.8%.4 In another review of COVID-19 survivors, researchers found the prevalence of PTSS ranged from 6.5% to 42.8%.5

Several previous studies have tried to ameliorate COVID-19-related PTSS using narrative exposure therapy6 or cognitive behavioural therapy (CBT) based treatments.7 However, these approaches were not consistently effective in reducing PTSS,6 7 and they were highly therapist-dependent. In the current study, we aimed to explore the effectiveness of cognitive bias modification (CBM), an auto-running, computerised programme, which could be implemented with relative ease.8 However, before introducing the intervention effects of CBM on PTSS, it is important to understand the association between cognitive bias and PTSS.

Studies have shown that cognitive bias (eg, attentional bias, interpretation bias, memory bias, etc) was a transdiagnostic risk factor for different mental disorders including anxiety, depression and post-traumatic stress disorder (PTSD).9 10 Cognitive models of PTSD assigned an important role for cognitive biases in the development and maintenance of PTSS.11 Interpretational bias is a type of cognitive bias where ambiguous stimuli are interpreted negatively and as dangerous.12 13 Previous studies have illustrated that interpretation bias was related to PTSS severity,14 and compared with health controls, patients with PTSS generated more trauma-related interpretations for ambiguous sentences.15 Fortunately, negative interpretation bias could be modified. Some studies showed that modification of negative interpretation bias could ammolite PTSS, indicating a causal relationship between negative interpretation bias and PTSS.16 17

Two methods were used in previous studies to achieve the modification of negative interpretational bias. The first was the interpretation modification programme (IMP). The IMP was adapted from the word-sentence association paradigm (WSAP) which was a classical paradigm to assess interpretation bias.18 In WSAP, a benign or a negative interpretation word was showed first, followed by an ambiguous sentence. Participants were asked to judge if the interpretation was related to the sentence. Once participants made their selection, they would move on to the next trial. However, in IMP, participants would receive feedback after selection. Positive IMP training would inhibit negative interpretation bias by presenting positive feedback for endorsing benign interpretations and for rejecting negative interpretations of ambiguous scenarios.18 19 Another way to modify interpretational bias was by use of the Cognitive Bias Modification of Interpretations (CBM-I) training. In CBM-I, participants were presented with a sentence completion task (SCT) and each sentence ended in a to-be-completed word. Participants were asked to complete the word fragment that could give meaning to ambiguous scenarios. Researchers would design the word fragments and guide the participants to produce functional or dysfunctional interpretations. Studies have illustrated the effectiveness of positive CBM-I training.16 17

In previous studies, researchers have modified negative interpretational bias to improve PTSS in clinical samples and analogue PTSS samples, but the findings were not consistent. For example, Woud et al demonstrated that compared with negative training, positive CBM-I training resulted in fewer intrusive memories and lower PTSS severity in analogue PTSS.16 Later, researchers found that positive CBM-I training could reduce intrusion distress and overall PTSS.20 However, these studies were conducted on health participants and the positive training condition was compared with the negative training condition, but not to control conditions (neither negative nor positive, and no explicit guidance was given to the subjects). Several studies have compared the intervention effects of CBM-I and control conditions in PTSD patients, but the results obtained varied widely. For example, de Kleine et al found that the positive training group and the control group did not differ in dysfunctional appraisals and PTSS severity after the intervention, failing to find evidence for CBM-I training.13 Interpretation bias and appraisal bias are related concepts: interpretation bias refers to the tendency to interpret ambiguous information negatively and appraisal bias refers to the tendency to interpret the trauma and its sequelae in a negative manner.13 However, in another study, Woud et al found that compared with the control condition, the positive CBM-I training group showed a greater reduction in appraisal bias and PTSS severity than the control group.21 As comparing positive CBM-I and IMP training to their control trainings is of higher clinical value, more similar research is still needed within the field of PTSS intervention.

Researchers have demonstrated that modification of interpretations could also reduce PTSS-related emotions like hostility, anger, fear and anxiety. For example, Dillon et al found in a pilot study that intervention of interpretation bias could reduce hostile interpretation bias and anger in veterans.22 van Teffelen et al found that compared with the control condition, CBM-I training could better improve behavioural aggression in highly hostile samples.23 Fear and anxiety might be the most important emotions in the aetiology and maintenance of PTSS,24 but they were poorly studied. In the single study that included anxiety, Lichtenthal et al found that the ‘attention and interpretation modification for fear of breast cancer recurrence’ intervention could improve negative interpretation bias and health worry in survivors of breast cancer.25 However, due to the excessive failure rate (only 26% of participants who screened into the study agreed to participate), the level of evidence of the study was affected. Taking together, the effectiveness of IMP and CBM-I on reducing negative affect remains to be studied in the field of PTSS treatment.

Primary evidence has shown that interpretation biases and PTSS could be reduced even after a single session of CBM-I training. For example, Woud et al found that explicit appraisal bias and overall PTSS were significantly improved in the positive CBM-I training condition compared with the negative training condition.20 Moreover, studies of anxious samples also found significant reductions in interpretation bias with only one training session.26 27 The single-session training can be a useful way to assess a new intervention before testing efficacy in a multisession format.28 29 As IMP training is a novel intervention in PTSS reduction, before conducting studies with multiple training sessions, the effects of single IMP training need to be explored.

Objective

In most previous studies on modifying interpretation bias, positive results were found by comparing positive and negative training conditions, not neutral ones. Most previous studies used CBM-I training while few randomised controlled trials focused on IMP training. To fill the gap, the present study designed IMP training to improve negative interpretation bias and PTSS. We would further explore whether interpretation bias modification could alleviate fear and how the change of fear would promote PTSS improvement.

MethodsParticipants

This study was conducted at the Behaviour Measurement Laboratory in the Faculty of Psychology, Naval Medical University in April 2023. The recruitment invitations indicated that the current study wished to recruit subjects who were infected with COVID-19 in December 2022 and January 2023, when the epidemic prevention and control in China began to enter the normalised stage.30 This period was chosen because we witnessed and experienced the rapidity of the spread and severe impact of COVID-19 during this time, and more potentially eligible subjects were available. Participants were recruited through the mainstream social media WeChat. They received links from WeChat and were asked to finish online questionnaires (demographics, whether they were infected with COVID-19 between December 2022 and January 2023, pretests). If subjects met the inclusion criteria, they would participate in the training in 2 weeks.

A total of 59 eligible subjects (34 male and 25 female) with an average age of 22.41±2.88 years participated in the present study. The inclusion criteria were: (1) Infected with COVID-19 between December 2022 and January 2023; (2) The PTSD Checklist for Diagnostic and Statistical Manual of mental disorders, 5th edition (DSM-5; PCL-5) total score ≥33; (3) 18 years of age or older; (4) Comfortable using the computer. Subjects who refused to participate or suffered from any type of psychotic disorder were excluded from the study.

Sample size calculation

The required sample size was determined using G*Power software prior to sample inclusion.31 Based on an estimated average small-to-medium effect size of IMP on PTSS (d=0.25),16 a total of 54 participants (ie, 27 participants in each arm) were needed to detect a significant effect (a=0.05, b=0.95, analysis of variance (ANOVA): repeated measures, within-between interaction). Considering the potential sample loss, an additional 10% sample was included.

Intervention

IMP was used to modify negative interpretation bias.18 Due to word limitations, the description of the intervention is detailed in the online supplemental materials.

OutcomesPrimary outcomeInterpretation bias

The SCT adapted from Naim et al’s study was used to assess participants’ interpretation bias.32 SCT was chosen because it was a holistic and ecologically valid approach.33 The SCT consisted of sentences describing ambiguous scenarios with the last word missing. Adding the final word could determine explicitly the valence of the sentence. In the current study, a total of 12 sentences were presented. They featured a range of ambiguous situations that had the potential to be interpreted as threatening to physical well-being. Subjects were asked to imagine themselves in the situation and generate as many endings as possible for each sentence (eg, ‘You feel different recently, because you ____.’). To reduce the demand characteristic, we did not ask participants to code the valence of their sentences, instead, the sentence completions were coded by QL and YS as positive, negative or other, thus providing relatively objective perspectives. Negative interpretation bias scores were calculated by summarising the number of negative responses divided by the total number of responses provided by each participant and positive interpretation bias scores were calculated by summarising the number of positive responses divided by the total number of responses. Rater reliability was high (r=0.89~0.98, p<0.05).

Secondary outcomePTSS

PTSS was evaluated using PCL-5, a 20-item self-report measure corresponding to PTSS clusters defined by DSM-5.34 These clusters include re-experiencing (Criterion B), avoidance (Criterion C) and negative alterations in cognition or mood and arousal (Criterion D). Participants were asked to rate the severity of each item on a five-point Likert Scale from 0 (not at all) to 4 (extremely). The total score ranged from 0 to 80. Individuals with a total score of 33 and above were more likely to be diagnosed with PTSD, as previous studies have indicated that the cut-off of 33 was optimally efficient for diagnosing PTSD.35 36 Thus, the score of 33 on the PCL-5 Scale was used as the cut-off for PCL-5. In the current study, subjects were required to rate PCL-5 according to their response to COVID-19 in the past month. PCL-5 showed good internal consistency in the present study (pretest: McDonald’s ω=0.92; post-test: McDonald’s ω=0.88).

Fear of COVID-19

The Fear of COVID-19 Scale (FCV-19S) has been widely applied to assess fear of COVID-1937 and was translated into Chinese (Chinese version of FCV-19S, FCV-19S-C).38 FCV-19S-C is a 5-point Likert Scale containing seven items assessing fear of COVID-19 with good validity.38 FCV-19S-C showed good internal consistency in the present study (pretest: McDonald’s ω=0.88; post-test: McDonald’s ω=0.91).

Anxiety and depression

Self-rating Anxiety Scale (SAS) and Self-rating Depression Scale (SDS) were used to assess anxiety and depression changes. SAS and SDS are 5-point Likert Scales and each contains 20 items to assess anxiety and depression in the last 2 weeks. Internal consistency (McDonald’s ω) of SAS and SDS ranged from 0.86 to 0.91 in the present study, illustrating good reliability.

Procedure

A detailed description of the procedure is presented in the online supplemental material. Figure 1 depicts the passage of participants.

Figure 1

Flow chart depicting the passage of participants. IMP, interpretation modification programme; ICC, interpretation control condition.

Data analysis

Data analysis was conducted according to the intention-to-treat principle. One participant did not take the pretraining SAS and SDS tests, and his data were replaced with mean values. Data were analysed using SPSS V.19. McDonald’s ω (omega) was calculated using the SPSS OMEGA macro written by Hayes.39 Preliminary analyses included descriptives, reliability analysis and test of normality distribution (Shapiro-Wilk test, skewness/kurtosis analysis and inspection of histograms). SAS scores were log-transformed because of a violation of the normal distribution. Repeated measures of ANOVA were used to explore the time (pretraining, post-training) × condition (IMP, interpretation control condition (ICC)) interaction effect on primary and secondary outcome measurements, and the simple effect analysis with Bonferroni correction for multiple comparisons was applied whenever the interaction effect was significant. Paired t-tests (two-tailed) were used if the main effect of time was significant. Pretraining and post-training variables were compared between the two groups using independent t-tests (two-tailed). We calculated the change of outcome variables by subtracting post-treatment scores from pretreatment scores and correlation analysis was applied to explore the relationships between variables. SPSS PROCESS macro V.3.4 bootstrapping procedures with 5000 resamples were performed to estimate the possible effects of change of fear and change of interpretation bias on PTSS reduction.40

FindingsPrimary outcome: interpretation bias

Table 1 presents the means and standard errors of outcome variables. To test the training effect of IMP on interpretation bias, the mean values of SCT were entered into a 2×2 repeated measurement ANOVA with time as a within-subject factor and group as a between-subject factor. Results illustrated a significant time × group interaction effect, F(1,57)=4.38, p=0.041, ηp2=0.07. A simple effect analysis with Bonferroni correction for multiple comparisons indicated that both groups showed significant change during training, and the improvement of negative interpretation bias was more obvious in the IMP group (F(1,57)=35.03, p<0.001, ηp2=0.38 for IMP, F(1,57)=7.49, p=0.008, ηp2=0.12 for ICC). The main effects of time were also significant, F(1,57)=36.74, p<0.001, ηp2=0.39; the main effect of the group was not significant. For positive interpretation bias, ANOVA demonstrated the main effect of time F(1,57)=18.64, p<0.001, ηp2=0.25. There was no significant main effect of group nor interaction effect.

Table 1

Comparison of pretraining and post-training between the training and control groups

Secondary outcomesPost-traumatic stress symptoms

For PTSS, a 2×2 ANOVA revealed that the interaction effect of time × group was not significant, F(1,57)=0.95, p=0.95; the main effect of time was significant, F(1,57)=22.54, p<0.01, ηp2=0.28, and the main effect of group was not significant, F(1,57)=0.62, p=0.43. Before training and after training, there were no differences in PCL-5 scores between IMP and ICC (see table 1), but both groups showed significant alleviation of PTSS after training (t(30)=0.348, p=0.002 for IMP, t(27)=0.33, p=0.003 for ICC).

Fear

A 2×2 repeated measurement ANOVA revealed that the interaction effect of time × group on fear of COVID-19 was significant, F(1,57)=4.14, p=0.047, ηp2=0.07. A simple effect analysis with Bonferroni correction for multiple comparisons illustrated that only the IMP group showed a significant change in fear of COVID-19 during training, F(1,57)=16.34, p<0.001, ηp2=0.22, and fear in the ICC group did not show significant improvement over time, F(1,57)=1.07, p=0.305. The main effect of time was also significant, F(1,57)=12.50, p=0.001, ηp2=0.18. The main effect of the group was not significant.

Depressive and anxious symptoms

For both SDS and SAS scores, the only significant result was the main effect of time, F(1,57)=14.19, p<0.001, ηp2=0.20 for SDS scores, and F(1,57)=69.70, p<0.001, ηp2=0.55 for SAS scores. The main effects of group and group × time interaction effect were not significant.

Correlation and moderation effect

To explore the correlation among changes in variables, post-training data were subtracted from pretraining data to illustrate the changes in each variable. Correlation analyses revealed that the change in interpretation bias was correlated with improvement of fear and PTSS. Table 2 presents the results. We further explored the potential medication and moderation effect of training conditions, fear reduction and interpretation bias reduction on PTSS improvement. There was a significant interaction of training condition × reduction in fear of COVID-19 on the improvement of PTSS (β=0.96, t=2.01, p=0.049, 95% CI 0.005 to 1.91, R2=0.05). Figure 2 illustrates the moderation effect: only in IMP, the reduction of fear of COVID-19 could predict improvement of PTSS, and the prediction effect was not significant for the ICC condition.

Table 2

Correlations among changes in variables

Figure 2

Reduction in post-traumatic stress symptoms (PTSS) predicted by training condition by reduction in fear of COVID-19 interactions. DSM-5, Diagnostic and Statistical Manual of Mental Disorders, 5th edition; IMP, interpretation modification programme; ICC, interpretation control condition; PCL-5, PTSD Checklist for DSM-5; PTSD, post-traumatic stress disorder.

Discussion

The results from this study suggested that a single session of IMP, designed to modify health-related and COVID-19-related interpretation bias was helpful in reducing fear of COVID-19 and negative interpretation bias. Results illustrated a significant time-by-group effect on negative interpretation reduction. Compared with the ICC group, the improvement of negative interpretation bias was more obvious in the IMP group. However, results demonstrated that participants' PTSS improvements were similar across training conditions (IMP and ICC). This result replicated several previous studies. For example, de Kleine found that the active CBM-I training group and control group illustrated similar PTSS levels from pretraining to post-training.13 Of note, in studies showing the effectiveness of interpretation bias modification, positive training conditions were typically compared with negative training conditions, that is, training participants to interpret ambiguous scenarios in a negative way.16 17 20 In another study showing the effectiveness of positive interpretation bias modification in reducing PTSS, the peripheral vision task (PVT) was selected as the sham training condition.21 However, PVT was not a common control condition. The current study’s control and intervention training conditions were more similar, providing more direct evidence for the effectiveness of IMP and higher clinical value.

Fear was considered as an important factor in the aetiology and maintenance of PTSS.41 Mild threatening and ambiguous stimulus, although not fearful in itself, could be interpreted as threatening, eliciting fear and maintaining PTSS.11 42 Primary evidence has illustrated that modification of attention and interpretation bias could reduce fear of breast cancer recurrence.25 This was in line with the current findings and we extended the previous study by linking fear with PTSS. Our results illustrated that although IMP training was not directly related to PTSS change, training conditions could moderate the relationship between fear reduction and PTSS improvement. In the previous study, Kanady et al found that the interaction of treatment conditions (CBT vs waitlist) by reduction in PTSS could predict change in fear of sleep.43 This result was similar to our study, but in the present study reduction in PTSS was treated as the dependent variable. To our knowledge, this was the first study to link interpretation bias modification and fear to PTSS improvement. By evaluating fear, this study might provide insight into the potential mechanism of IMP in the improvement of PTSS.

There were several limitations in the current study. First, follow-up data were not collected and the long-term effect of a single session of IMP was not clear. Second, Criterion A of the PTSD diagnosis was not evaluated, and the participants could not be clinically diagnosed with PTSD. Our results may be strengthened by researching clinical samples. Third, there was only one session of IMP training in our study, and additional training sessions might result in more significant improvement. Future studies could further evaluate the effectiveness of multiple sessions of IMP on negative interpretation, PTSS and fear reduction.