Study objectives and hypotheses
This study aims to develop and evaluate the effects of a novel online training paradigm that is grounded in the principles of EC in reducing loneliness in adults with a randomised controlled trial (RCT) design. The details of our study design, procedures and conditions are visualised in Fig. 1. The study schedule and the measurements are detailed in Table 1 according to the Standard Protocol Items: Recommendations for Intervention Trials (SPIRIT) guidelines. The SPIRIT checklist is reported as Supplementary Material (Table S1) [32]. We hypothesise that:
-
(i)
The EC intervention will reduce self-reported and implicit loneliness after intervention and at follow-up after 3 months.
-
(ii)
The EC intervention will bring positive changes to other outcomes (DMN connectivity, emotional processing and social motivation towards social stimuli, anxiety and depression levels, social network, perceived social support, emotional state and reactivity) after intervention and at follow-up after 3 months.
-
(iii)
DMN connectivity, emotional processing and social motivation towards social stimuli will mediate the effects of EC in reducing loneliness.
-
(iv)
Age, gender and baseline level of depression will moderate the effects of EC intervention.
Flow diagram of the 2-arm randomised controlled trial protocol. EC = evaluative conditioning
Ethical approval and informed consent
This study is approved by the Human Research Ethics Committee of The Education University of Hong Kong (Reference: E2022-2023-0082) and is registered with the ClinicalTrials.gov (NCT06521099) on July 2024. Written consent will be obtained from all participants. Participants can withdraw from the study any time without negative consequences. Data collected will be securely stored, password protected and anonymised. If a serious adverse event occurs, it will be reported to the principal investigator immediately.
Eligibility and participant recruitment
The inclusion criteria are (i) aged ≥ 18 years, (ii) with no histories of learning impairment, major psychiatric disorders or neurological disorders other than depression, (iiii) with at least primary school education, and (iv) having a score of ≥ 6 on the 3-item UCLA Loneliness Scale [33]. Those who are on medication or treatments that would affect the individual’s brain, cognitive and affective functions within 2 weeks prior to the beginning of the study will be excluded. People who are severely depressed (i.e., scores on depression symptoms ≥ 15 in the Hospital Anxiety and Depression Scale) will also be excluded [34]. For the MRI assessments, participants will first undergo safety screening. Only participants without MRI contraindications, not claustrophobic, not pregnant (for females), and those who provide consent to participate in the MRI scanning, will be invited. All participants’ demographics, including age, gender, education levels, and living arrangements will be recorded. Recruitment is anticipated to be completed by June 2026.
Sample size
Power analysis using pwr in R is performed to estimate the required sample size for this study. With medium effect sizes (d ~ = 0.5), α = 0.05 two-tailed, and statistical power of 0.8, it is estimated that 68 participants are required for each group. Accounting for a 30% of non-response rate and dropout, we target to recruit 180 people at baseline to participate in the study.
Study design and procedure
We will apply a 2 × 3 repeated-measures RCT design. People who are interested in this study will first be invited to complete a screening questionnaire online on eligibility and consent will then be collected. Non-responders will be further contacted twice. Behavioural and MRI assessments will be conducted at baseline, immediately after, and 3 months after the intervention in our eligible participants. At baseline prior to the intervention sessions, participants will first complete tests and questionnaires, and they will also receive resting-state functional MRI (fMRI) scanning in a 3 Tesla MRI scanner. Participants will then be randomly allocated to either the EC group or the control group with a computer algorithm administered by an independent research personnel, stratified by gender and baseline loneliness scores to ensure balanced representation. Participants will be blinded to the group allocation. They will receive six 30-minute training sessions online over the course of three weeks. Immediately after (i.e., within 3 days), and also 3 months after the intervention, participants will be invited to complete again the same behavioural and MRI assessments. Due to minimal risk, emergency unblinding will not be implemented. Data monitoring committee will not be established, and no formal interim analyses are planned. Behavioural data will be self-report or collected by trained research personnel, and MRI data will be collected by trained scanner operator. Participants who complete the study and all assessments will receive a maximum of HK$500 as a token of appreciation for their participation.
Intervention paradigms
Participants in the EC or control group will receive the intervention online, which includes six 30-minute training sessions scheduled over the course of three weeks. Participants allocated to the EC group will view the socio-affective stimuli (CS) paired with positive emotional stimuli (US1) for 120 trials in random order in each session (Fig. 2). Participants allocated to the control group will view the CS paired with neutral emotional stimuli (US2) for 120 trials in random order. The CS, US1 and US2 were pictures selected from The Nencki Affective Picture System, according to their valence reported in a previous study and the involvement of humans [35]. These stimuli, together with other stimuli used in the Emotional Processing task, were all validated in a separate sample. To ensure high compliance, we will follow up on the participants if a null response is recorded in any of the scheduled sessions.
(A) Example trials in the third and fourth blocks of the Implicit Association Test – Loneliness. Participants are instructed to sort items to the left or right with the key ‘E’ or ‘I’ on the keyboard respectively. (B) Schematic diagram of trials in a training session of the evaluative conditioning (EC) and control groups. Both trainings consist of 6 sessions across 3 weeks. In each training session, participants allocated to the EC group will undergo a CS-US conditioning phase and view their CS paired with positive emotional stimuli (US1) for 120 trials in random order. Participants allocated to the control group will view their CS paired with neutral emotional stimuli (US2). (C) The mean ratings of the valence and social motivation of the picture stimuli from our pilot study. Error bars = 95% CI
Primary outcomes
Self-reported loneliness. The 20-item UCLA Loneliness Scale version 3 will be used. It is reliable and has been widely used to measure subjective perception of loneliness instead of objective social isolation in local samples [36]. It is also sensitive to changes over time [37]. The scale includes 20 items, 11 of which are positively worded and the other 9 are negatively worded (which will be reversely coded). Participants will be asked to rate on a 4-point Likert scale (never to often) to indicate how frequently they feel about each item. A total score ranging from 20 to 80 will be calculated, with higher scores reflecting higher levels of loneliness. Example items include ‘How often do you feel left out?’. The 6-item De Jong Gierveld Loneliness Scale will also be used to capture the two forms of loneliness – emotional and social loneliness – in our participants [38]. Example items of the two forms of loneliness include ‘I experience a general sense of emptiness’, ‘There are plenty of people I can rely on when I have problems’ (negatively worded), respectively.
Implicit loneliness. We will administer the Implicit Association Test – Loneliness (IAT-L) [31]. It is a reaction time test that implicitly measures the strength of association between our participants’ cognitive representations of target concepts (‘Self’ vs. ‘Others’) and attributes (‘Lonely’ vs. ‘Non-lonely’). The IAT-L will have seven blocks, and the first, second, and fifth blocks are the practice blocks. In each trial, labels of target concept and/or attribute categories will be presented on the left and right top panels while one item from these categories will be presented in the center of the screen, with 150 ms inter-trial intervals. Specifically, the third and fourth blocks (30 trials in each block) will have target concept and attribute category labels showing simultaneously on the screen, where participants will be instructed to sort items belonging to ‘Self’ or ‘Lonely’ to the left, and ‘Others’ or ‘Non-lonely’ to the right (Fig. 2). The sixth and seventh blocks is similar to the third and fourth blocks but with opposite pairings – participants are instructed to sort items belonging to ‘Self’ or ‘Non-lonely’ to the left, and ‘Others’ or ‘Lonely’ to the right. Example items belonging to ‘Self’, ‘Others’, ‘Lonely’, and ‘Non-lonely’ categories include ‘Me’, ‘Them’, ‘Deserted’, and ‘Cared for’, respectively [31]. A standardised D-score comparing participants’ reaction time in third and fourth blocks with that in the sixth and seventh blocks will be retrieved, with a more positive score reflecting a relatively stronger association between ‘Self’ and ‘Lonely’ [39].
Secondary outcomes
fMRI connectivity. Loneliness is highly implicated in the DMN, attention and cognitive control networks, salience and emotional networks [14, 19]. Here, resting-state fMRI data of eligible participants will be acquired to investigate their change in their functional connectivity and network centrality using a 3.0 Tesla MRI scanner. Participants will be asked to lie in the scanner and stare at a fixation cross for a 10-minute resting-state functional scan, acquired using conventional T2*-weighted gradient echo planar imaging (EPI) sequence and the following parameters: slice thickness = 3.5 mm, time-to-repetition (TR) = 2000 ms, time-to-echo (TE) = 30 ms, flip angle = 90°, voxel size = 3 × 3 × 3.5 mm3, interleaved axial slices. Their T1-weighted structural MRI data will also be acquired for co-registration and segmentation using the BRAVO sequence and the following parameters: TR = 7 ms, TE = 2.84 ms, flip angle = 8°, voxel size = 1 × 1 × 1 mm3, sagittal slices.
Emotional processing. Participants’ emotional processing and social motivation will be measured by their ratings of social/non-social, positive/negative emotional stimuli with the Emotional Processing task. Participants are required to indicate their perceived arousal (calm to aroused), valence (positive to negative), and social motivation (none to high) towards each stimulus on 5-point Likert scales. The picture stimuli were selected from The Nencki Affective Picture System, according to their valence reported in a previous study and the involvement of humans [35]. The average ratings for trials in each category, including Positive Social, Negative Social, Positive Nonsocial, Negative Nonsocial, and Neutral will be retrieved for analysis. We have already validated the ratings of the valence and social motivation of the stimuli in a separate sample of 24 adults in a pilot study. All stimuli are confirmed to match the assigned conditions (Valence: Negative > Neutral > Positive; Social motivation: Social > Nonsocial; all ps < 0.001) (Fig. 2).
Anxiety and depression. Participants’ anxiety and depression levels will be assessed by the Hospital Anxiety and Depression Scale, which is a widely used and validated questionnaire [40].
We will also assess participants’ emotional state with the state version of the Chinese Affect Scale [13], their emotional reactivity with the 21-item Emotional Reactivity Scale [41], their size of social network with family and friends, with the Lubben Social Network Scale-6 [42], and their perceived social support with the Multidimensional Scale of Perceived Social Support [43].
Planned data analysis
Behavioural data from questionnaires and tasks will be analysed with R. Missing data will be explored and imputed where appropriate. Data normality and variance will be checked and baseline demographical characteristics will be compared between the two groups using independent t-tests, chi-square tests or their non-parametric equivalent tests. Demographical characteristics that show group differences will be controlled for in the subsequent analyses where appropriate. Linear mixed models will be set up with groups (EC and control), timepoints (baseline, immediately after, and 3 months after intervention), and their interaction as fixed factors, subjects as random intercepts. Post-hoc pairwise analyses will be performed to delineate significant interaction effects. Differences in outcomes between timepoints will be obtained, and regression models will be set up to investigate their associations. Mediation analyses will be performed to explore whether changes in ratings of socio-affective stimuli mediate the intervention effects. Moderation analyses will be conducted to explore whether age, gender or baseline level of depression moderates the intervention effects. All main, interaction, moderation and mediation effects are considered significant at p < 0.05 or according to the 95% confidence intervals and bootstrapping.
Resting-state fMRI data will be preprocessed with CONN toolbox [44] and custom scripts in MATLAB. Voxel-wise, connectivity, and network measures will be analysed. Voxel-wise analyses will be performed with the general linear model using non-parametric permutation tests with 5000 permutations [45]. Model setup will be similar to that of the linear mixed models and regression models applied to behavioural data. This will identify the effects of groups, timepoints and their interaction effects on brain connectivity, as well as to investigate the associations between changes in outcomes and changes in brain connectivity. Threshold-free-cluster-enhancement and familywise-error-corrected p < 0.05 will be adopted to infer significance. Regional connectivity, and network measures will be retrieved as outcomes and will be analysed similarly to that of the behavioural data using linear mixed models, regression models, mediation and moderation analyses. Effects will be considered significant at p < 0.05 or according to the 95% confidence intervals and bootstrapping.