Efficacy of an Internet-based self-help intervention with human guidance or automated messages to alleviate loneliness: a three-armed randomized controlled trial

The study addresses chronic loneliness, a subjective aversive experience arising from a discrepancy between desired and actual social relationships. Loneliness is common across the lifespan and linked to adverse mental and physical health (e.g., depression, social anxiety, suicidality, cardiovascular and brain health, and increased mortality). A cognitive model posits that loneliness triggers hypervigilance and negative social-cognitive biases, leading to withdrawal and perpetuation of loneliness. Prior meta-analyses suggest interventions targeting maladaptive social cognitions are effective; internet-based CBT (ICBT) can increase accessibility and reduce stigma. Earlier trials showed guided ICBT reduced loneliness versus waitlist and versus internet-based interpersonal therapy, but did not isolate the effect of human guidance. This trial tests whether ICBT reduces loneliness versus waitlist and whether human guidance adds benefit over automated messages. Hypotheses: pooled ICBT conditions reduce loneliness and secondary outcomes more than waitlist; guided ICBT yields greater improvements than automated messages.

Meta-analyses indicate psychological interventions, especially those altering social cognitions, reduce loneliness and improve social connectedness. ICBT has proven effective across disorders and can be comparable to face-to-face when guided. Prior ICBT loneliness trials reported moderate effects versus waitlist (d≈0.71–0.77) and superiority to an interpersonal therapy variant. Guidance in internet interventions increases adherence (g≈0.29) and effect sizes versus unguided or automated-support versions. Higher degrees of human contact predict better outcomes in internet interventions for depression. Unguided programs show low completion rates, though automated reminders can help. The literature highlights the need for high-quality trials to clarify the role of human guidance in loneliness interventions.

Design: Three-armed, 10-week randomized controlled trial with parallel groups comparing ICBT with human guidance (GU) and ICBT with automated messages (AM) to a waitlist control (WL). Randomization ratio 2:2:1 using block-wise randomization (Qualtrics). Ethics approval (Cantonal Ethics Committee Bern, ID: 202-01298); preregistered (NCT04655196); informed consent obtained; CONSORT-SPI 2018 followed. Participants: Adults (≥18) from German-speaking countries recruited via media and online channels between May 17, 2021 and July 31, 2022. Inclusion: UCLA-9 ≥18, German proficiency, internet access, consent, emergency contact. Exclusion: severe depressive symptoms (PHQ-9 >14), lifetime psychotic or bipolar disorder, current severe substance use disorder, or acute suicidal plans; assessed via Mini-DIPS Open Access diagnostic interview. Concomitant treatments allowed. Sample: N=243 randomized (GU n=98; AM n=97; WL n=48). Mean age 45.77 (SD=14.85), 78.6% female, 62.4% with university degree; 51.4% met criteria for ≥1 psychological disorder (most prevalent social anxiety disorder, 29.2%). Mean duration of loneliness ≈11.6 years. Interventions: SOLUS-D (German adaptation of Swedish ICBT for loneliness), nine mainly text-based modules with video/audio, mindfulness, self-compassion, and social skills. Recommended one module/week (~50 min), all modules unlocked from start; exercises and diary to shift attentional focus and compassion. Accessible via computer/smartphone/tablet; SSL encryption; anonymous login. - Guidance (GU): Weekly individualized feedback via in-program messaging from trained and supervised coaches (two master-level psychologists and ten master’s students). Semi-structured, manualized according to Supportive Accountability to enhance adherence; standardized reminders if inactive. Coaches spent on average 17.10 min per participant per week. - Automated messages (AM): Weekly standardized email messages summarizing content and motivating continued use; no human contact. Technical issues could be addressed to study team. - Waitlist (WL): Access to unguided intervention after 10-week post-assessment. Outcomes and measures: Primary outcome: loneliness (UCLA-9; 9–36 range; higher = more loneliness) at 10 weeks. Also a direct single-item loneliness and UCLA-3 administered. Secondary outcomes: depressive symptoms (PHQ-9), social anxiety (SIAS-6, SPS-6), satisfaction with life (SWLS), self-esteem (RSES), self-compassion (SOCS-S), objective social isolation/network size (SNI), maladaptive personality traits (PID5BF+), interpretation bias (IJQ_total), rejection sensitivity (A-RSQ), social avoidance behavior (CBAS), distress disclosure (DDI), authenticity (KGAI-SF), misanthropy (BVI), motivation for solitude (MSS-SF). Additional post measures: client satisfaction (CSQ-8), usability (SUS), negative effects (INEP). Program adherence: modules accessed (clicked-through pages) and time spent; coach time recorded. Procedure: Baseline assessment followed by diagnostic interview; randomization; intervention access for GU and AM immediately; WL received access after post-assessment. Assessments at baseline and 10 weeks; up to three weekly email reminders for non-response. No blinding of participants or coaches post-randomization. Statistical analysis: Intention-to-treat (ITT) included all randomized participants. Baseline group differences via ANOVA/Chi-square (with non-parametric alternatives as needed). Linear mixed-effects models (lme4 in R 4.2.1) with REML: fixed effects of time, group, and Time×Group; random intercepts for participants; time and group categorical; no random slopes due to non-convergence. Planned contrasts: pooled interventions vs WL (GU: −0.5, AM: −0.5, WL: 1) and GU vs AM (GU: −1, AM: 1, WL: 0). Effect sizes: between-group Cohen’s d computed from estimated mean differences at post divided by pooled baseline SD; within-group d from estimated pre-post change divided by pooled observed SD across timepoints; 95% CIs reported. Reliable change on UCLA-9 via Reliable Change Index using α=0.90 (external sample) and baseline SD=3.34; thresholds: improvement >2.93, deterioration <−2.93; ITT used last observation carried forward (LOCF); also per-protocol (completed baseline and post, accessed ≥4 modules). Sensitivity analyses: per-protocol; subgroups with at least one psychological disorder; concurrent psychotherapy at baseline. Power: A priori powered to detect small Time×Group interaction (f=0.10; d≈0.20) with α=0.05, power=0.80, r=0.60 pre-post correlation; targeted n=100 per intervention and n=50 WL (2:2:1; total 250). Recruitment ended at n=243 for regulatory reasons.

- Primary outcome (UCLA-9): Significant Time×Group interaction, F(2,191.98)=8.22, p<0.001. Pooled interventions vs WL showed reduced loneliness at post: t(241)=3.13, p<0.002, d=0.57 (95% CI [0.25; 0.89]). Guided vs automated at post favored guidance: t(193)=2.38, p=0.02, d=0.42 (95% CI [0.13; 0.70]). Within-group effects: GU d=1.02 (95% CI [0.71; 1.31]); AM d=0.73 (95% CI [0.43; 1.02]); WL d=0.28 (95% CI [−0.12; 0.68]). - Secondary outcomes (significant Time×Group interactions and contrasts for pooled interventions vs WL): Depressive symptoms (PHQ-9): t(241)=2.89, p=0.004, d=0.52; Social phobia symptoms (SPS-6): t(240)=2.30, p=0.02, d=0.37; Social avoidance behavior (CBAS): t(241)=1.98, p=0.048, d=0.32; Rejection sensitivity (A-RSQ): t(240)=2.22, p=0.03, d=0.38. Self-compassion (SOCS-S) and misanthropy (BVI) showed significant interactions but no significant pooled intervention vs WL contrasts (p=0.09–0.18; d=0.23–0.29). No significant differences between GU and AM at post on secondary outcomes (p=0.05–0.92; d=0.02–0.31). - Intervention usage: Accessed all nine modules: GU 42.9% (n=42), AM 40.2% (n=39). Accessed ≥4 modules: GU 84.7% (n=83), AM 72.2% (n=70); χ²(1)=4.53, p=0.03, V=0.15; post-hoc residuals non-significant. Mean modules accessed: GU 6.77 (SD=2.62), AM 6.07 (SD=3.16); t(186)=−1.67, p=0.10. Time in program: GU 563.28 min (SD=543.86), AM 370.44 min (SD=338.36); t(193)=−2.97, p=0.003, d=−0.43. - Dropout: 25.9% did not complete post (GU 28; AM 33; WL 2). Non-completion higher in interventions vs WL (GU vs WL: χ²(1, n=146)=11.75, p<0.001, V=0.28; AM vs WL: χ²(1, n=145)=15.63, p<0.001, V=0.33); non-completers younger (t(241)=−2.62, p=0.009). - Reliable change (UCLA-9): ITT reliable improvement did not differ significantly (GU 48.0%; AM 41.2%; WL 31.3%; χ²(2)=3.73, p=0.15). Deterioration differed: GU 2.0%, AM 5.2%, WL 12.5%; χ²(2)=6.97, p=0.03, V=0.17; GU vs WL OR=0.15 (95% CI 0.02–0.73), p=0.02. Per-protocol reliable improvement: GU 60.9%, AM 52.6%, WL 30.4%; χ²(2)=10.46, p=0.005, V=0.25; GU vs WL OR=3.50 (95% CI 1.60–7.96), p=0.002; AM vs WL OR=2.51 (95% CI 1.12–5.82), p=0.03. Per-protocol deterioration lower in GU vs WL: p=0.02, OR=0.11 (95% CI 0.00–0.71). - Satisfaction and negative effects: CSQ-8 satisfaction high (GU M=3.18, SD=0.61; AM M=3.02, SD=0.55); t(123)=−1.59, p=0.12. Usability (SUS): GU M=80.17, SD=15.68; AM M=79.44, SD=16.02; t(124)=−0.26, p=0.80. Negative effects: mean number similar (GU M=0.32, SD=0.81; AM M=0.35, SD=0.88; t(120)=0.18, p=0.86); at least one negative effect reported by 21.0% (GU) and 25.0% (AM); most frequent were prolonged feeling bad and increased suffering from past events. - Sensitivity analyses: Results robust in per-protocol and subgroups (with psychological disorder; concurrent psychotherapy). GU vs AM remained significant or borderline (in disorder subgroup t(99)=1.90, p=0.06, d=0.50).

ICBT substantially reduced loneliness compared with waitlist, aligning with prior Swedish trials. For the first time, human guidance within ICBT demonstrated superiority over automated messages in reducing loneliness, potentially via relational factors (validation, feeling understood) and/or increased adherence/accountability; the latter reflected in significantly greater time spent in the program. The guidance effect fits broader evidence that human contact enhances outcomes in internet interventions. Despite reductions in loneliness, social network size did not change, underscoring that quality of relationships and self-related attitudes may be more salient than network quantity. Secondary benefits included reductions in depressive symptoms, social anxiety, social avoidance, and rejection sensitivity, suggesting that alleviating loneliness may reduce broader psychopathology. Differences between indirect and direct measures of loneliness were observed: significant indirect measure differences at post between interventions and waitlist, but no between-group differences on the direct single-item measure due to improvements across all groups, highlighting measurement considerations. Heterogeneity in response emphasizes the need to identify moderators and tailor approaches to individual causes and contexts of loneliness (e.g., small network vs perceived inadequacy). Operationalizing adherence requires careful consideration, as module completion did not differ while time-on-task did, informing future design and evaluation of eHealth interventions.

A 10-week internet-based CBT self-help program (SOLUS-D) effectively reduced loneliness compared to waitlist. Human guidance enhanced the reduction in loneliness relative to automated messages, demonstrating the added value of human support in ICBT for loneliness. The intervention also lowered depressive and social anxiety symptoms, social avoidance, and rejection sensitivity, indicating broader mental health benefits. Future work should: evaluate long-term outcomes (6- and 12-month follow-ups); clarify mechanisms of guidance (e.g., therapeutic alliance, adherence) and indirect effects on depression/anxiety; refine measurement strategies for loneliness (direct vs indirect); identify predictors/moderators to personalize interventions; and assess applicability to more diverse populations, including those with severe depression.

- Exclusion of individuals with severe depressive symptoms limits generalizability to highly depressed populations. - Self-selected, motivated sample; findings may not generalize to less motivated individuals or the broader lonely population. - Predominantly female and highly educated participants; limited representativeness. - Short-term outcomes only; follow-up effects pending. - Reliance on self-report measures may introduce bias (e.g., social desirability), though loneliness is inherently subjective. - Higher non-completion in intervention arms; potential attrition bias, although mixed models and sensitivity analyses were used. - Participants and coaches were not blinded post-randomization. - Recruitment slightly below target (n=243 vs intended 250), possibly affecting power for some comparisons.