Chronotype and emotion processing: a pilot study testing timing of online cognitive bias modification training

A pilot study to understand whether the effectiveness of CBM for facial emotion processing is improved when delivered at a time-of-day aligned to chronotype. Healthy, young adults (aged 18–25) with either an early or late chronotype completed CBM training online in either the morning (06:00 hours to 10:00 hours) or evening (18:00 hours to 22:00 hours) session. We predicted that the effectiveness of CBM for shifting interpretation of emotional facial expressions (learning rate and post-training balance point) would be optimised when delivered in a ‘synchronous’ condition (evening session for late chronotypes and morning session for early chronotypes) compared with an ‘asynchronous’ condition (morning session for late chronotypes and evening session for early chronotypes), and this effect would be greater in the late chronotypes compared with early chronotypes (group×condition interaction). We also predicted that online CBM would have a greater effect on improving positive and negative effects in late chronotypes compared with early chronotypes.

The study design can be viewed in figure 1. Young adults were pre-screened using the online recruitment platform Prolific (https://www.prolific.co↗) and Qualtrics (https://www.qualtrics.com/uk/↗). Prolific hosts a large pool of diverse participants to recruit from and was used to avoid recruiting a student-only sample and improve generalisability. Adults with a late or early chronotype were then recruited to the main study through Prolific which was conducted remotely and online using Gorilla (https://gorilla.sc/↗). Half of each chronotype group was allocated to CBM training in a morning (06:00 hours to 10:00 hours) or evening (18:00 hours to 22:00 hours) session, that is, in a synchronous (morning for early chronotypes, evening for late chronotypes) or asynchronous (morning for late chronotypes, evening for early chronotypes) condition (2×2 design). An effort was made to stratify the conditions by sex at birth. Participants completed the study between February and April 2023.

The study was conducted according to the revised Declaration of Helsinki (2011) and the 1996 International Council for Harmonisation (ICH) Guidelines for Good Clinical Practice E6 (R2). All participants provided informed consent online prior to taking part in the study and were compensated for their time. The study protocol was pre-registered on the Open Science Framework (OSF) https://doi.org/10.17605/OSF.IO/F95A2↗.

1242 young adults were assessed for eligibility during the pre-screening phase. Inclusion criteria included a late or early chronotype (as determined by a score of 4–11 or 18–26 on the reduced Morningness-Eveningness Questionnaire (rMEQ)15), aged between 18 and 25, residing in the UK, English language fluency and normal or corrected-to-normal vision. Exclusion criteria included a current diagnosis of depression or any sleep disorder (determined via self-report), use of antidepressant medication, daily nicotine (smoking/vaping) or recreational drug use, drinking more than 35 alcoholic units per week and drinking more than 8 cups of caffeinated beverages a day. 208 eligible adults (109 late chronotypes and 99 early chronotypes) were recruited and successfully completed the main study.

At the start of the main study session, the following measures were collected; sleep quality (Pittsburgh Sleep Quality Index (PSQI)16), number of hours slept the night before, positive and negative affect (Positive And Negative Affect Schedule (PANAS)17) and consumption of alcohol, caffeine, nicotine and recreational drugs within 24 hours. A repeated measure of positive and negative affect (PANAS) was collected after CBM training. All questionnaires are described in full in online supplemental material 1.

All participants completed an active version of CBM training for facial emotion processing as described previously.11 18 Briefly, the task involved viewing 15 images of facial expressions morphing from happy to sad. In the baseline block, participants judged the emotion (ie, happy or sad, two-option forced choice) and a ‘balance point’ is calculated (interpretation bias towards ambiguous facial expressions). Next, participants underwent three blocks of active CBM training where participants were given feedback that the two faces adjacent to their balance point (training threshold) should be classed as ‘happy’ instead of ‘sad’, positively shifting responses to the faces (Supp figure 1). The baseline block was then repeated to assess the post-training balance point. Further task details are available in supplementary materials.

Previous studies of CBM versus sham in healthy individuals with low mood show a large effect of CBM training on the post-training balance point.10 Previous studies of synchrony effects on executive functions are large.13 Therefore, we cautiously powered the current study for the main effect of synchrony (2×2 analysis of variance (ANOVA)) on post-training balance point assuming a smaller effect size (Cohen’s f=0.2). A total sample of 200 (50 per group) gave us 80% power to detect this effect at an alpha rate of 0.05. We recruited an extra eight participants (total n=208) in error before the study was closed.

Data wrangling was done in R and analysed using IBM SPSS Statistics (V.29). Participants were excluded from analyses if: they failed to complete all CBM blocks, had an extreme baseline balance point (<4 or >12) or >2× the IQR away from the median, or completed the study >2 hours outside of their time slot. Participants were asked not to consume alcohol or recreational drugs within 24 hours, nicotine within 12 hours or caffeine within 1 hour of completing the study and were excluded from analysis if they reported they had (self-reported at the beginning of the study session).

An updated version of an ALCOVE computational model of associative learning19 was fitted to trial-by-trial responses during CBM training and was used to compute the primary outcome measure; effective learning rate (ε_effthr). Model details are described in the Supplementary materials. Briefly, the effective learning rate is a measure of how quickly the participant incorporates task feedback to update their responses to new faces. People generalise what they learn about each face stimulus according to its relationship to neighbouring stimuli on the facial expression morph continuum. This is accounted for in the effective learning rate which adjusts for individual training threshold. Higher effective learning rates indicate the participants respond more quickly and effectively to task feedback and therefore shift their classifications of ambiguous faces accordingly. Other parameters included: (1) inverse temperature (θ) (response reliability), (2) emotion bias (gH – gS) (positive values represent excess sad judgements on overt facial expressions) and (3) pre-training indifference point (a model-based measure analogous to balance point).

The primary outcome measure was effective learning rate (ε_effthr) which was transformed to a normal distribution using an inverse rank transformation. The secondary outcome measure was the post-training balance point, calculated as: (the number of ‘happy’ responses in the post-training baseline block/45 trials)×15 facial stimuli. A 2×2 ANOVA tested for a main effect of group (late chronotype, early chronotype), condition (synchronous, asynchronous) and a group×condition interaction on the primary (learning rate) and secondary (post-training balance point) outcome measures. Exploratory analysis of covariance were also performed adjusting for pre-training outcome measures, negative affect symptoms (PANAS score) and sleep quality (PSQI score) by including these variables as covariates in the model. Pre-training outcome measures (eg, pre-training balance point) were included as covariates to adjust for the participant’s baseline (eg, this can be interpreted as the effect of group and condition on change in balance point). Sleep quality and negative affect were included as covariates due to the effects of poor sleep quality and higher negative effects on attention, cognition and the perception of negative facial expressions.20 Since age was within a narrow range10 14 18–23 and sex was balanced between conditions, these variables were not added as covariates. All analyses were conducted on eligible participants (n=173), however for completeness, analyses were also conducted on the full sample (including those who had used nicotine, alcohol, caffeine and recreational drugs close to the study, n=208). Exploratory analyses included post-training positive and negative affect (subscales of PANAS score) as the outcome variables. Further details in supplementary materials.

All participants completed all CBM blocks. Two participants were excluded as they had balance points lower than 4. Three participants completed the session outside of their allotted time slot (due to experimenter error) but none >2 hours as per the quality control criteria. 34 participants were excluded for reporting using stimulants (alcohol, nicotine, recreational drugs, caffeine) before the study session leaving 173 eligible participants (47 late chronotype synchronous, 45 late chronotype asynchronous, 40 early chronotype synchronous, 41 early chronotype asynchronous).

Baseline characteristics are presented in table 1. Across the sample (n=173), 114 reported their ethnicity as ‘White’, 20 as ‘Black’, 25 as ‘Asian’, 10 as ‘Mixed’ and 4 as ‘Other’. 77 were in full or part-time employment, 58 were studying full or part-time, 18 were unemployed and 20 were unknown. The average age was 22.5 years and 45% were men. Sleep quality (M=5.0, SD=2.3) was poor as some studies suggest that a PSQI score >5 represents sleep disturbance.21 Late chronotypes were significantly younger, had poorer sleep quality, slept fewer hours the night before the study session and had lower positive effect compared with early chronotypes. A larger proportion of late chronotypes were also men. The proportion of participants who used nicotine, alcohol, caffeine and recreational drugs did not differ between groups. The mean times that participants completed the study session within groups and conditions were at: 19:29 late chronotype synchronous, 09:08 late chronotype asynchronous, 08:35 early chronotype synchronous and 19:44 early chronotype asynchronous.

Participant effective learning rate and generalisation were consistent with a prior report8 (table 1). As expected, the pre-training indifference point was strongly correlated to the pre-training balance point (r(172=0.87, p<0.001).

Overall, there was strong evidence that CBM produced positive changes in participant balance point rising from 6.75 (SD=1.24) at pre-training to 8.29 (SD=1.47) post-training (mean difference=1.52, SD=0.86, paired t-test t(172)=−22.1, p<0.001) (figure 2). This means that participants changed their responses to ambiguous facial expressions in line with the feedback they received. This positive change is in line with previous studies.11 However, there was an overall dip in balance point after training which could represent participants tiring of doing all blocks in a single session. Indeed, previous studies have asked participants to do five training blocks spread out over a week.11

There was no evidence for a main effect of the chronotype group, synchrony condition or interaction effect on the primary outcome measure; effective learning rate ε_effthr. However, when sleep quality and pre-training negative affect were added as covariates to the model, there was weak evidence for a group×condition interaction effect (F(1,166)=2.92, p=0.090, η_p²=0.017) where pairwise comparisons provided weak evidence that late chronotypes had a higher ε_effthr in the synchronous (M=0.012, SE=0.048) compared with asynchronous (M=−0.13, SE=0.051) condition (p=0.050) (figure 3C). This effect disappeared when participants who had recently taken stimulants were included in analyses.

There was no evidence for a main effect of the chronotype group on the secondary outcome measure; the post-training balance point. However, there was weak evidence for a main effect of synchrony (F(1,169)=3.40, p=0.067, η_p²=0.020) where the post-training balance point was higher in the synchronous (M=8.50, SE=0.16) compared with the asynchronous (M=8.08, SE=0.16) condition. This effect was strengthened when sleep quality, pre-training balance point and negative affect were added as covariates (F(1,166)=4.28, p=0.040, η_p²=0.025, synchronous condition M=8.43, SE=0.097, asynchronous condition M=8.15, SE=0.098) (figure 3A,B). There was moderate evidence for a group×condition interaction effect (F(1,166)=4.34, p=0.039, η_p²=0.025) where pairwise comparisons (with Bonferroni correction) showed that late chronotypes had a higher post-training balance point in the synchronous (M=8.58, SE=0.13) compared with asynchronous (M=8.00, SE=0.14) condition (p=0.003) (figure 3D).

First, the rMEQ is an indirect, behavioural measure of the circadian phase. Future studies would benefit from measuring other indices, for example, dim light melatonin onset. It would be interesting to investigate which aspects determining chronotype (eg, genetics, light exposure) are most important for synchrony effects. Second, the timings of CBM training sessions (morning/evening) were based around working hours and were generous (4-hour slots) meaning participants may have completed the session at different phases of their circadian physiology. Future studies would benefit from fixing the session to circadian time (eg, hours after waking). Third, all measures were self-reported and there were likely other factors, for example, light, activity, breakfast or naps affecting cognitive performance that were not accounted for. Future studies would benefit from controlled laboratory conditions (eg, to control stimulant use) to validate and extend this work.

Initial evidence is that an online psychological intervention can be optimised (learning is improved) when delivered at a time-of-day, that is, synchronised to chronotype, and this might be most important for late chronotypes that are vulnerable to poor mental health. This could be considered during intervention delivery. Chronotherapy has shown that medication delivery timing for physical and mental health conditions can improve treatment effects29 however there are concerns about whether treatments can be delivered at certain times and if circadian rhythms can be measured accurately enough. Interventions that can be delivered online could avoid some of these practical concerns. Future studies in clinical populations, for example, depression, would be important for understanding the effect on mood.