Chronic illnesses are typically characterized by their duration, lasting at least 1 year, and by the ongoing need for medical care or limitations they impose on daily functioning [1]. Recent epidemiological data from a large-scale study identified obesity (19%), hypertension (13%), diabetes mellitus (9%), esophagitis-gastritis (5.5%), and thyroid disease (5.3%) as the 5 most prevalent chronic conditions, with significant variations in age and sex distribution across different conditions [2]. Given their increasing incidence and considerable impact, these health issues impose substantial burdens on patients, their families, and broader societal structures. In fact, chronic diseases are responsible for approximately 90% of annual health care spending in the United States, amounting to US $4.1 trillion [1].

The widespread coexistence of mental health issues with chronic physical illnesses poses one of the greatest challenges to global health care systems [3]. These complex comorbid relationships have been comprehensively documented in the scientific literature [4], with substantial evidence demonstrating that patients with chronic physiological pathologies (including persistent pain syndromes, diabetes mellitus, and inflammatory gastrointestinal disorders) frequently manifest concurrent psychopathologies, particularly depressive and anxiety disorders [5-7]. These mental health comorbidities are commonly linked to worse health outcomes, such as reduced life satisfaction and lower overall quality of life [4,8,9].

The presence of comorbidities suggests a bidirectional relationship, where the outcomes of the illness are shaped not only by its severity and progression but also by accompanying psychological dynamics. To illustrate, psychological adjustment, coping behaviors, and how patients manage their illness play a pivotal role in influencing physical health outcomes, including rates of morbidity, mortality, and complications [10-13].

Additionally, detrimental psychological processes (eg, illness-related shame, self-criticism, or rumination) constitute a barrier to implementing self-care behaviors involved in adaptive illness management (eg, treatment plan adherence, exercise, or adherence to dietary guidelines), which, in turn, could further increase distress, worsen prognosis, and diminish the quality of life [14,15]. Consequently, both self-care behaviors included in illness management as well as detrimental psychological processes have become a target of different psychological interventions aimed at promoting the quality of life of the population with chronic medical illness [16].

This is the case of compassion-based interventions (CBIs), which refer to psychological interventions aimed at enhancing compassionate and self-compassionate responses that involve the recognition of suffering and the inclination to relieve it with an act of kindness rather than criticizing, blaming, or pitying [17]. Fostering compassion and self-compassion may support illness management by encouraging acceptance of suffering as a fundamental human experience and promoting a gentler stance toward personal challenges [18].

Several individual studies have shown that CBIs can significantly enhance quality of life for individuals living with chronic health conditions [19-21]. This evidence has been corroborated by recent systematic reviews and meta-analyses, which concluded that CBIs produce improvements across multiple outcomes including depression, anxiety, self-compassion, and health-related quality of life [22-24]. In sum, being able to approach difficulties with a compassionate attitude helps individuals to feel empowered with new management strategies and promotes a sense of calm and agency to provide comfort to themselves, which facilitates the implementation of self-care behaviors and adaptive illness management [25,26].

Among the various CBIs available, attachment-based compassion therapy (ABCT) represents a particularly promising approach that has shown effectiveness among both healthy individuals and those managing chronic medical illnesses [27]. This therapy is based on attachment theory [28], which provides a framework for understanding the links between close relationships and psychopathology and includes specific practices to identify and develop a secure attachment style to promote compassion for oneself and others [27]. Specifically, the ABCT has shown its efficacy and applicability for the treatment of fibromyalgia, showing improvements in psychological outcomes such as functional status [29] and biological outcomes [30].

Despite ABCT’s demonstrated efficacy, there are specific delivery barriers that could interfere with the effectiveness of CBIs in people with chronic illnesses, such as limitations of access, mobility, or transportation. In order to tackle these limitations and to respond to the growing need of health care systems for scalability and sustainability, evidence-based interventions can benefit from adapting their delivery format through information and communications technologies (ICTs) as a complementary or alternative delivery mode [31,32]. In fact, previous research supports the notion that the psychological outcomes of people living with chronic illnesses can be improved with a self-delivered online intervention [33,34], and recent evidence shows the potential value of delivering CBIs online in the context of chronic illnesses [35,36]. Moreover, other results show that adapting CBIs to an online format may improve adherence and facilitate the involvement of chronic patients in better management of their illness [32]. Regarding the ABCT approach, an online version of internet attachment-based compassion therapy (iABCT) has been developed to be totally self-applied over the internet for Spanish speakers. The iABCT is currently assessed in a feasibility study for the general population [37]. The iABCT is designed as a stand-alone, fully self-administered psychological intervention, though the original ABCT protocol has demonstrated utility both as an independent treatment and as a complement to usual care in clinical settings [38].

In conclusion, although CBIs delivered through the internet show promise as a cost-effective solution and initial research has explored the feasibility and acceptability of the iABCT in a Spanish general population [37], more research is needed on their efficacy specifically in the context of chronic medical conditions. While online interventions offer numerous advantages, recent studies have identified engagement challenges that may affect their efficacy. For instance, Finlay-Jones et al [35] found that despite significant improvements in well-being and distress outcomes, program engagement and retention rates were suboptimal in their web-based self-compassion intervention for young people with chronic conditions. These findings highlight the importance of examining potential barriers to engagement and treatment acceptability in online CBIs.

Given the preliminary nature of digital CBIs for chronic illness populations, a waiting list (WL) control was chosen as the comparator to establish the initial efficacy of iABCT before conducting comparisons with active treatment alternatives, following recommended sequential evaluation frameworks for behavioral interventions [39].

Therefore, this study aims to explore the efficacy of the iABCT to improve quality of life and well-being (primary outcomes) and other psychological variables (secondary outcomes) in a population with chronic medical illness, compared to WL, as well as examine treatment acceptability and implementation barriers and facilitators. We hypothesized as follows:

No formal patient or public involvement was undertaken in the design, conduct, or reporting of this trial.

This study used a 2-arm, parallel-group randomized controlled trial (RCT) design to evaluate the efficacy of iABCT in individuals with chronic medical conditions. Participants were randomly assigned to either the intervention group (iABCT) or a WL control group. Assessments were conducted at baseline, 3-month follow-up, and 6-month follow-up. The WL group received access to the intervention after completing the 3-month assessment. As a result, between-group comparisons were feasible only at the 3-month follow-up, while 6-month assessments captured within-group maintenance for the iABCT group and postintervention effects for the WL group following delayed access.

The study adhered to the CONSORT (Consolidated Standards of Reporting Trials) 2025 statement [40]. The completed CONSORT checklist is provided in Checklist 1. The study also followed the CONSORT eHEALTH V1.6 guidelines [41], and the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) statement [42].

Two deviations from the preregistered protocol [38] occurred during trial implementation. First, 5 disease-specific illness interference measures (Diabetes Distress Scale, Roland-Morris Questionnaire, Revised Fibromyalgia Impact Questionnaire, Migraine Disability Assessment Questionnaire, and Inflammatory Bowel Disease Questionnaire) and the Illness Perception Questionnaire-Revised were included as secondary outcomes but are not reported as inferential analyses in this paper. Although data for these measures were collected and are available, subgroup sizes within each medical condition were small and highly unbalanced across groups (refer to Table S1 in Multimedia Appendix 1), precluding reliable statistical inference. Descriptive statistics are nonetheless reported in Table S1 in Multimedia Appendix 1 for transparency. Second, sensitivity analyses using multiple imputation methods, prespecified in the registered protocol, were not conducted. Given the substantial attrition rate observed across assessment points (ie, >30% missing data), additional imputation procedures were considered unnecessary. Instead, analyses were performed using mixed-effects models estimated with Restricted Maximum Likelihood (REML) procedures, which allow the inclusion of all available observations and provide unbiased parameter estimates under the missing-at-random (MAR) assumption, consistent with the result of Little MCAR test (χ²₁₅₃=150.97, P=.53). Mixed-effects models are considered a robust and recommended approach for handling incomplete longitudinal data, particularly in repeated-measures designs with moderate to high attrition, as they avoid case-wise deletion and do not require prior imputation of missing values [43].

No major bug fixes, system downtimes, or content changes to the iABCT platform occurred during the trial period.

Inclusion criteria were: (1) adults aged 18 to 70 years, (2) self-reported diagnosis of a chronic medical condition, defined as a diagnosed medical illness lasting at least 1 year and requiring ongoing medical care or imposing functional limitations (participants indicated their specific condition and its duration during screening); (3) ability to read and understand Spanish; and (4) access to a computer with internet connection and an email account. Exclusion criteria were (1) presence of a terminal disease, (2) presence of severe psychiatric disorder comorbidities (schizophrenia, substance dependence, bipolar disorder, or psychotic illness) or severe neurologic or medical condition; and (3) receiving psychological treatment or mindfulness training at the time of recruitment.

As the iABCT intervention was fully self-administered with no human care providers involved in its delivery, eligibility criteria for individuals delivering the intervention were not applicable.

This was a fully web-based trial conducted entirely online with no face-to-face contact. Participants were recruited nationally through online and offline channels, including the study website, flyers posted on social media platforms (eg, Facebook [Meta Platforms, Inc], Instagram [Meta Platforms, Inc], and LinkedIn [LinkedIn Corp]), doctors’ referrals, and associations for people with chronic conditions. All distributed materials included a dedicated email address through which participants could contact the research team.

All study procedures, including eligibility screening, informed consent, intervention delivery, and outcome assessments, were completed remotely via the Psychology and Technology web platform [44], accessed using a computer with an internet connection (as specified in the inclusion criteria). PDF files of intervention materials were available for download to allow offline review. Outcomes were self-assessed by participants through online questionnaires administered at baseline, 3-month follow-up, and 6-month follow-up. The trial was conducted without institutional affiliations being prominently displayed during recruitment.

Adverse events and harms were monitored throughout the trial on a nonsystematic basis via passive surveillance. Participants were able to report any adverse experiences at any time through the study platform or the dedicated research team email address. No formal harms assessment instrument was used. The research team reviewed all participant communications throughout the trial period for any indication of psychological distress or unintended effects potentially attributable to the intervention. Data on harms were collected across the full trial period, from randomization to 6-month follow-up.

Sample size was calculated using G*Power 3.1.9.7 (Heinrich Heine University Düsseldorf) [62]. Based on anticipated medium effect sizes (Cohen d=0.4), a statistical power of 0.80, and α=.05, the required sample was 52 participants for a repeated-measures design [63]. Accounting for an expected 30% dropout rate in internet-based interventions for chronic medical conditions [64-66], a minimum target of 34 participants per group was established, for a total minimum sample of 68 participants.

No interim analyses were planned or conducted during the trial. No formal stopping guidelines were prespecified.

Data analyses were conducted using SPSS (version 28.0; IBM Corp) [67]. Baseline equivalence between groups was assessed using t tests, ANOVAs, and chi-square analyses. To address missing data and assess the effectiveness of the intervention, an intention-to-treat (ITT) approach was applied using linear mixed-effects models without using ad hoc imputation methods [43]. This analytic strategy is particularly suitable for RCTs involving repeated measures across multiple time points and is known for its robustness to violations of normality and other distributional assumptions [68,69].

The primary estimand was the time×condition interaction effect at 3-month follow-up for the 2 coprimary outcomes (EQ-5D and PHI). Total scores were prespecified as primary targets for all outcomes; subscale-level analyses were exploratory. Following the preregistered protocol, no alpha-adjustment for multiple primary outcomes was prespecified. EQ-5D and PHI were analyzed as coprimary end points without hierarchical testing or Bonferroni correction.

Separate mixed-effects models were conducted for each outcome variable using the MIXED procedure with REML estimation. Fixed effects included time, condition, and the time×condition interaction. Models incorporated a random intercept for each participant with an identity covariance structure. Denominator degrees of freedom for F tests were calculated using the Satterthwaite approximation. The variable “time” (baseline, 3-month follow-up, and 6-month follow-up) was modeled as a within-subject factor, while the experimental condition (iABCT vs WL) served as a between-subject factor. Primary efficacy analyses tested between-group differences at the 3-month follow-up (time×condition interaction). Six-month data were analyzed as within-group comparisons (baseline to 6-month) separately for each condition to assess (1) maintenance of treatment effects in the iABCT group and (2) postintervention changes in the WL group following delayed intervention access. Where statistically significant effects were observed, follow-up pairwise comparisons were performed using Bonferroni corrections. Effect sizes (Cohen d) and corresponding 95% CIs were computed to quantify both within- and between-group differences using standardized mean differences based on observed means and SDs. A small-sample bias correction factor c(m) was applied following Hedges and Olkin [70], yielding values equivalent to Hedges’ g; given correction factors ranging from 0.98 to 0.99 across analyses, differences from uncorrected Cohen d are negligible [71,72].

For qualitative data, participants’ responses regarding barriers to study enrollment (nonparticipants) and barriers to intervention engagement (dropouts) were explored using qualitative content analysis with a coding and categorization approach, including word frequency counts with ATLAS.ti software (v.23; ATLAS.ti Scientific Software Development GmbH). One researcher (MM-S) conducted initial open coding to identify facilitators and barriers. Codes were then reviewed by a second researcher (MDV) with experience in qualitative analysis. Discrepancies were discussed until consensus was reached. Additionally, thematic analysis following Braun and Clarke’s [73] methodology was conducted to identify common themes and patterns related to participant opinions on different aspects of the intervention program. Main findings from both analyses are presented in the “Results” section, while complete qualitative analysis including all tables, participant quotes, and detailed statistical breakdowns are provided in Multimedia Appendix 1.

This trial was conducted in compliance with the study protocol [38], the Declaration of Helsinki, and good clinical practice. Ethical approval for this trial was obtained from the Ethics Committee for Human Research of the University of Valencia (UV-INV_ETICA-1564960). The trial was prospectively registered in March 2021, prior to the enrollment of the first participant (study start: July 8, 2021).

Informed consent was obtained online before any study procedures. The online consent form included details of the trial, an explanation of potential risks and benefits, and contact information for the research team. Participants were informed that participation was completely optional and could be discontinued at any time.

Participants did not receive any financial compensation for their participation. Access to the intervention was provided free of charge to all participants.

All study data were collected and stored anonymously and deidentified. Participants were assigned unique identification codes, and no personally identifiable information was linked to study responses. Data were stored on secure servers with restricted access limited to the research team. This study did not involve the collection of images, photographs, or any visual material that could identify individual participants. All figures and tables included in the paper and Multimedia Appendix 1 contain only aggregated or anonymized data, and no identification of individual participants is possible.

Initially, 276 participants were interested in the study, and 130 were excluded because they did not fulfill eligibility criteria (Figure 1). Thus, 146 participants were randomized (iABCT=72; WL=74). Of these, 53 participants did not access the platform (iABCT=24; WL=29), resulting in 93 participants who completed the baseline assessment (iABCT=48; WL=45). Little MCAR test confirmed that missing data were completely at random (χ²₁₅₃=150.97, P=.53).

Dropout rates from baseline to 3-month follow-up were 43.75% for the iABCT group (21/48) and 42.22% for the WL group (19/45), resulting in 27 and 26 participants, respectively. Between 3-month and 6-month follow-up, additional dropout rates were 33.33% (9/27) for the iABCT group and 53.85% (14/26) for the WL group, with 18 and 12 participants completing the final assessment, respectively.

Recruitment took place from April 2021 to April 2022. The last participant completed the 6-month follow-up assessment in October 2022. Each participant’s total follow-up duration was 6 months from randomization.

The iABCT intervention was fully self-administered via the online platform, with no human therapist involved in delivery. Of the 72 participants randomized to iABCT, 24 (33.3%) did not access the platform and therefore received no intervention content. Systematic data on individual module completion were not collected. Adherence was monitored indirectly through platform login records, which triggered automated email reminders when participants had not logged in for 1 week. For the WL group, 29 of 74 (39.2%) randomized participants did not access the platform following delayed intervention access.

Table 2 shows participants’ sociodemographic data for each condition. Groups were comparable on all demographic variables.

Regarding meditation experience, 37.7% (55/146) of participants had previous meditation experience, with similar proportions in both groups. Among experienced meditators, groups showed comparable frequency, duration, lifetime practice, and context of practice. Complete meditation experience data are provided in Table S3 in Multimedia Appendix 1.

Regarding clinical outcomes (Table 3), groups showed comparable scores on most measures. Descriptively, the WL group had higher self-compassion scores, while the iABCT group showed numerically higher self-criticism. All other outcomes showed similar baseline values across groups.

No deaths occurred during the trial. No participants withdrew from the study due to adverse events or harms. No serious adverse events were identified through passive surveillance over the full trial period (randomization to 6-month follow-up) in either the iABCT or WL group.

No subgroup analyses were prespecified or conducted in this trial. Sensitivity analyses using multiple imputation methods were prespecified in the registered protocol but were not conducted (see the “Changes to Trial Protocol” section). All other analyses reported were prespecified in the registered protocol, except for the exploratory subscale-level analyses, which were post hoc.

Several limitations warrant consideration. First, the waitlist control design rather than an active control condition may have contributed to effect overestimation, as waitlist controls have been shown to function as nocebo conditions in psychotherapy trials [85,86]. Second, high attrition rates, particularly at 6-month follow-up, may have compromised statistical power and the interpretability of long-term effects. Third, the identity covariance structure used in the mixed models, while necessary to ensure model convergence given the modest sample size and high attrition, may not adequately capture residual correlations between repeated measures. Misspecification of the covariance structure can bias SEs and affect statistical inference, particularly under the MAR assumption [102,103]. Fourth, the transdiagnostic approach, while beneficial for addressing comorbid conditions, prevented illness-specific analyses and may have limited intervention tailoring [104]. The heterogeneity of the sample and the unbalanced number of participants per chronic condition also prevented analysis of illness interference as a moderator, which has been shown to be relevant in CBIs [29]. Fifth, no multiplicity adjustment was prespecified for the dual primary outcomes or the multiple secondary outcomes, increasing the risk of type I error. Relatedly, between-group comparisons should be prioritized as the most rigorous test of intervention efficacy, and within-group improvements without corresponding between-group differences should be considered exploratory. Sixth, cost-effectiveness was not evaluated, limiting health care implementation assessments despite the considerable socioeconomic burden of chronic diseases [1,105]. Seventh, the absence of between-group comparisons at 6-month follow-up due to ethical considerations restricts conclusions about long-term maintenance of effects. Regarding the qualitative analyses, only around half of contacted individuals completed the questionnaire, formal interrater reliability was not calculated, and thematic saturation was not prospectively assessed, which limits the generalizability of those findings. Finally, the predominantly female, Spanish-speaking, and highly educated sample limits the generalizability of findings to broader chronic illness populations.

This preliminary trial provides initial evidence that a fully self-guided, iABCT may improve overall quality of life in adults with chronic medical conditions, with effects maintained at 6-month follow-up. Significant reductions in preoccupied attachment were also observed. However, given the absence of alpha adjustment for dual primary outcomes, the high attrition rates, the modest sample size, and the waitlist control design, these findings should be interpreted cautiously as exploratory and require replication in adequately powered confirmatory trials with active control conditions and appropriate multiplicity correction. Improvements in both the iABCT and waitlist groups for some outcomes suggest potential contributions from assessment effects, natural symptom fluctuation, or regression to the mean, which further limits causal attribution. Despite high treatment satisfaction and acceptability, qualitative analyses revealed critical engagement barriers emphasizing the need for human support, personalized content, and user-centered design in digital CBIs.

This study makes several novel contributions to the field. To our knowledge, it provides the first empirical evidence from an RCT that a fully self-guided, internet-delivered attachment-based compassion intervention may produce improvements in quality of life across diverse chronic medical conditions, extending ABCT beyond its original face-to-face [29,30] and single-diagnosis [29] formats. Unlike existing digital compassion programs, iABCT uniquely integrates attachment security as a theoretical change mechanism alongside self-compassion cultivation, offering a theoretically grounded framework that may be particularly relevant for chronic illness populations whose psychological difficulties are often rooted in early relational experiences. Importantly, by targeting transdiagnostic psychological processes common across chronic medical conditions, iABCT is positioned as a potentially universal intervention capable of addressing shared mechanisms of suffering across the full spectrum of chronic illness. The scalable format offers potential to address unmet psychological needs in populations facing geographic, economic, or mobility-related access barriers [76-78], with promise for reducing health care burden [1,105]. Future research should prioritize adequately powered confirmatory trials with active control conditions, hierarchical testing procedures for multiple outcomes, and alternative covariance structures to ensure valid inference. Incorporating human support elements such as therapist guidance or peer support, developing condition-specific adaptations, investigating mechanisms of change, and conducting cost-effectiveness analyses are also key priorities to optimize both the reach and effectiveness of digital CBIs in real-world health care settings [32,98,106].