The study was conducted across 2 contrasting NHS clinical settings:

First, the hospital discharge pathway (Bristol), where myCOPD was introduced as part of the acute exacerbation of COPD discharge bundle. Eligible patients were identified during follow-up clinics or virtual ward reviews within 6 weeks of hospital discharge. Delivery was primarily supported by digital health champions (DHCs) due to staffing pressures and workflow constraints.

Second, the community PR (Cornwall), where myCOPD was integrated into routine PR delivery. A digital health advisor (from my mhealth Ltd) and a respiratory care coordinator supported registration, activation, and ongoing use. Clinicians used the myCOPD dashboard to monitor progress and tailor self-management discussions.

These settings differed in purpose, workflow, and available resources, providing a natural comparison of delivery approaches.

myCOPD is a digital self-management app providing educational modules, inhaler-technique videos, PR content, symptom tracking, and action-plan support. Clinicians can view patient-reported data through a dashboard to support remote monitoring and personalized care. The intervention content was identical across sites.

Patients were recruited from the PROPEL (A Pragmatic Real-World Multicenter Observational Research Study to Explore the Clinical and Health Economic Impact of myCOPD) study cohort. During their baseline PROPEL visit, all eligible participants were informed about the qualitative component and invited to opt in or out. Those who opted in were recontacted to arrange an interview after their scheduled 3-month follow-up visit, timed to ensure adequate exposure to the intervention. The eligibility criteria at each site are outlined in Textbox 1.

A convenience sampling approach was used, with elements of purposive sampling to ensure representation across both clinical settings. Health care professionals (HCPs) involved in delivering myCOPD were identified by site leads and approached via email. A purposive sampling strategy ensured variation in professional role (DHCs, nurses, and physiotherapists) and setting (community vs hospital).

Semistructured interviews were conducted via Microsoft Teams (audio or video, depending on the technological accessibility of participants). Topic guides were tailored for patients and HCPs and refined iteratively by experienced qualitative researchers. Interview topics included experience of COPD, use of digital technology, engagement with myCOPD, and perceived barriers and facilitators. Example questions are presented in the full topic guides in Multimedia Appendix 1.

With informed consent, clinical data (eg, COPD severity, years since diagnosis, long-term conditions) were obtained from participants’ health records held by NHS clinical teams. Similarly, with consent, myCOPD usage data (eg, logins, module completion, and PR access) were obtained directly from my mhealth Ltd, which provided anonymized usage summaries for participants enrolled in the PROPEL study.

At the start of each interview, participants received a verbal recap of study aims, confidentiality, and their right to withdraw. One interviewer was present at each interview (MR conducted all patient interviews, and HCP interviews were split between BC and MR). MR (a PhD researcher in health psychology with training in qualitative interviewing and prior experience as a patient and public involvement [PPI] officer) conducted all patient interviews and facilitated PPI sessions. BC (an experienced qualitative researcher) conducted HCP interviews. Their positionalities—MR’s long-standing ties to the local community and BC’s research background—were acknowledged and reflexively considered throughout the analysis. Interviews were recorded, transcribed verbatim, checked for accuracy, and imported into NVivo 14 (Lumivero) for analysis. All participants received financial compensation (payments of £75 [US $100] for HCPs; vouchers of £25 [US $30] for patients). Finally, the data presented here include CAT scores that were taken at baseline by clinical research staff.

Coding was conducted by 2 researchers (BC and MR) for HCP interviews and by a researcher (MR) for patient interviews. HCP and patient transcripts were coded independently. During theme development, we compared the coded data to identify shared patterns, at which point the analytical process merged into a unified set of themes. Analyzing the transcripts together captured the relational aspects of self-management and reflected an overlap in coding and candidate themes across groups. While both groups discussed a particular theme, typically one group described the theme in more detail; for example, HCPs and patients both described the importance of multimorbidity, but patients described in greater detail how it shaped engagement.

Researcher positionalities were addressed through structured reflexive practices. BC and MR kept journals during data collection and coding, revisiting these reflections during theme development to check for assumptions and bias. MR also held regular debrief meetings with senior qualitative researchers (BA, KB, and LY) to discuss how background experiences might shape interpretation. These discussions helped mitigate potential limitations of single coding and promoted consistency and rigor. MR also used analytic memos and diagramming to track decision-making and consider alternative interpretations. Together, these practices provided a systematic reflexive process that helped ensure interpretations remained grounded in the data.

Coding followed an abductive thematic analysis [27] workflow that combined the MRC process evaluation framework [26] with inductive code generation. Transcripts were first read repeatedly to support familiarization. A deductive coding structure based on MRC domains (context, implementation, engagement, mechanisms of impact) [26] served as the overarching structure, and data were coded line-by-line according to the relevant categories. Alongside this, inductive codes were generated to capture participant-led insights that did not fit the MRC framework. The coding framework was refined iteratively, with codes merged, repositioned, or expanded as patterns became clearer. Coded data were then reviewed within and across domains to identify candidate themes, with the MRC structure providing the overarching scaffold and inductive codes supplying the detailed content. Themes were developed through team discussion, diagramming, and matrix mapping to examine relationships across settings and between patient and HCP accounts. Reporting of qualitative methods and findings was informed by the COREQ (Consolidated Criteria for Reporting Qualitative Research) checklist (Checklist 1). This iterative movement between empirical data and implementation theory supported the identification of mechanisms, such as how delivery approaches shape engagement trajectories, that are central to understanding implementation and align with calls for qualitative methods [28].

PPI activities were integrated throughout the PROPEL study and used to refine qualitative findings. The PPI group for the work outlined here consisted of Windrush Generation Elders (pioneering individuals who immigrated to the United Kingdom from Caribbean countries and other Commonwealth nations between 1948 and 1971). They all had experience of living with or caring for someone with asthma or COPD. This group highlighted the importance of intersectionality before data collection and therefore supported this focus. One-to-one sessions were held with stakeholders (professionals based in the NHS and at the digital provider) who were experienced in delivering myCOPD. Work with the community-based group adopted a participatory approach to activities (facilitated by MR). These activities were designed to assess the relevance, clarity, and real-world applicability of emerging themes. Feedback informed the refinement and interpretation of themes.

The PPI group was not told the results at the beginning of the workshop (Multimedia Appendix 2 provides further detail using GRIPP2 [Guidance for Reporting Involvement of Patients and the Public, version 2] form); they were asked to individually draw journey maps of their experiences with COPD (in some cases, asthma). We then discussed their maps, with the chance for them to respond to each other and adapt their own maps. Finally, the results were introduced, and we discussed factors that overlapped with the results and those that were different, reasons for differences, and the importance of each factor. Consent was not obtained for PPI activities, and therefore these do not feature as a facet in the “Results” section. PPI highlighted that the results were relevant, how they related to lived experience, and a few aspects that were missing from the results (eg, the role of carers).

Ethical approval was granted by the NHS Research Ethics Committee (REC reference: 23/SC/0155) and the University of Southampton (ERGO: 84797). All participants provided informed consent. Clinical and usage data were anonymized or deidentified before analysis. Confidentiality was maintained throughout, and participants received financial compensation for their time.

The qualitative interview data generated and analyzed during this study are not publicly available due to confidentiality agreements and the absence of participant consent for data sharing.

Thirty interviews were conducted, 16 with patients and 14 with HCPs. Patient interviews lasted 14 to 35 (mean 23.7, SD 6.3) minutes, and HCP interviews lasted 23 to 57 (mean 36.8, SD 7.9) minutes. Sixteen of the 37 patients who initially agreed completed interviews; there was a range of reasons for nonparticipation: no response to invitations (n=15), unavailable contact details (n=2), technical failures (n=2), or withdrawal (n=2). Patient characteristics are provided in Table 1, and HCP characteristics are provided in Table 2. Participant usage data (Table 3) indicated higher engagement among Cornwall patients, who had a longer period since activation (107‐148 days) and more frequent logins. Across both sites, 81.25% (13/16) accessed educational modules and 68.75% (11/16) accessed PR content.

Five themes were developed, each with subthemes (Table 4). Read sequentially, the themes describe a pathway from patient characteristics to delivery context, engagement patterns, perceived barriers, and perceived benefits. This structure highlights how individual and structural factors interact to shape engagement with myCOPD. As mentioned in the “Methods” section, the analysis of the transcripts was often complementary between HCPs and patients. For example, both HCPs and patients could describe the health care setting as affecting engagement with myCOPD, though HCPs typically provided more detail.

Three characteristics (multimorbidity, perceived digital ability, and socioeconomic context) operated as “layers of vulnerability” influencing engagement. Importantly, it was not the presence of a characteristic alone but how patients interpreted and negotiated that characteristic that shaped their engagement trajectory.

This theme describes how patients perceived engagement with myCOPD, a perception seemingly affected by contextual factors described in previous themes. Two subthemes are used to describe different types of engagement. The contrast in engagement across the 2 sites highlights that engagement itself is a layer of vulnerability rather than a neutral process, one that does not rely solely on design and individual capacity.

The perceived benefits of engagement with myCOPD appear to be affected by the first 3 themes. The first subtheme captures direct benefits associated with myCOPD. For HCPs, this meant positive alignment with existing care guidelines; for patients, this meant that the app was a tangible resource for improving specific abilities. The second subtheme describes wider benefits of myCOPD.

This study explored how myCOPD was delivered and used across 2 contrasting NHS settings and how contextual, organizational, and individual factors shaped engagement. Across both sites, patients described how multimorbidity, perceived digital ability, and socioeconomic context influenced their capacity to engage with the intervention. These characteristics did not operate in isolation; rather, they interacted with local delivery practices to shape engagement trajectories.

A key finding was the marked difference in delivery approaches between settings. Community PR HCPs adopted a proactive, iterative model of support using myCOPD to structure conversations, reinforce self-management strategies, and monitor progress. In contrast, hospital HCPs described a more passive, onboarding-focused approach shaped by workflow pressures, limited staffing, and the prioritization of discharge. These delivery styles appeared to influence whether patients engaged in single-aspect engagement (targeted, short-term usage) or multifaceted engagement (routine usage).

Both HCPs and patients identified barriers and facilitators to engagement. Barriers included usability concerns, competing health priorities, limited digital confidence, and challenges integrating the app within in-person care. Facilitators included personalized support from clinicians, clear explanations of app features, and opportunities to discuss content during follow-up contacts. PPI contributors confirmed the relevance of these findings, emphasizing the importance of timely support, clear communication about the purpose of the app, and recognition of the challenges posed by multimorbidity and limited digital confidence.

A key strength of this study is the inclusion of both patients and HCPs across 2 contrasting NHS pathways, enabling a nuanced understanding of how delivery context shapes engagement. The use of abductive thematic analysis allowed us to integrate theoretical insights with inductively generated themes, and PPI involvement strengthened the relevance and clarity of the findings. Collectively, these findings highlight that engagement with myCOPD is not simply a function of individual motivation or app design but emerges from the interaction between patient characteristics and the delivery environment, an insight that aligns with and extends implementation theories emphasizing context-mechanism interactions [27].

Our findings align with implementation research demonstrating that digital interventions are not inherently effective; their impact depends on how they are introduced, supported, and embedded within clinical workflows [11,26]. Prior COPD digital evaluations [22-25] similarly reported high initial interest in inhaler videos and symptom tracking but noted rapid declines in use. Our findings extend this work by showing why this decline occurs, specifically, that passive onboarding in hospital settings limits opportunities for relational reinforcement.

More specifically, our observation that proactive, relationship-based support in community PR settings facilitated sustained engagement echoes the work of Pinnock et al [5,6] on supported self-management and aligns with the theorization of implementation mechanisms by Lewis et al [28], which emphasizes the importance of relational reinforcement and iterative feedback loops. Our data provide concrete examples of how these mechanisms operate within COPD pathways, illustrating how clinicians used myCOPD to scaffold conversations, tailor recommendations, and maintain continuity, practices that have been theorized but rarely described empirically in COPD digital implementation studies.

Conversely, the passive onboarding approach observed in hospital settings reflects challenges documented in acute care environments, where digital tools are often introduced during periods of high patient burden and limited staff capacity [10,14]. This is consistent with the findings of O’Connor et al [29], which indicate that workflow pressures and resource constraints in acute care limit opportunities for relational support and follow-up. Our findings extend this literature by showing how these constraints shape not only uptake but also the depth and trajectory of digital engagement, with hospital patients more likely to adopt single-aspect, short-term use.

Our findings also contribute to emerging evidence on digital inequalities. Multimorbidity, perceived digital ability, and socioeconomic context shaped how participants navigated myCOPD, consistent with the literature highlighting the intersection of health and digital disparities [13-16]. Informed by the concept of “layers of vulnerability” proposed by Luna [19,20], our analysis suggests that these vulnerabilities are dynamic and relational, shaped not only by individual circumstances but also by the support available within clinical settings. This was evident in patients who described shifting their self-management priorities in response to exacerbations or competing long-term conditions, relying on family members or clinicians to navigate digital tasks when confidence fluctuated, and adjusting their engagement depending on whether services provided opportunities for discussion, troubleshooting, or reinforcement.

In doing so, our study provides empirical grounding for the conceptualization of vulnerability as layered rather than categorical, as proposed by Luna [19,20], demonstrating how multimorbidity, digital confidence, and socioeconomic constraints accumulate and interact within real-world COPD pathways. This complements the call for intersectional approaches to digital health disparity research made by Husain et al [11] and aligns with the argument of Veinot et al [30] that digital inequalities are structurally produced rather than individually determined. Our findings also resonate with the findings of Greenhalgh et al [31], specifically the finding that multimorbidity often complicates digital adoption, particularly when tools are not embedded within supportive relational contexts.

Although this study offers important insights into how delivery context shapes engagement with myCOPD, several limitations should be acknowledged when interpreting the findings. First, participants were recruited from within the PROPEL study cohort, which may limit transferability to settings where myCOPD is introduced outside of a research context. Second, interviews were conducted remotely, which may have influenced participation among individuals with limited digital access. Third, while we examined differences across 2 settings, the sample size did not allow for detailed subgroup analysis (eg, by digital literacy, severity, or SES). Finally, usage data provided by my mhealth Ltd offered helpful context but did not capture qualitative nuances of engagement. These limitations underscore the need for future work that examines delivery models across a wider range of organizational contexts and includes more granular measures of digital literacy and multimorbidity burden, as recommended by Venoit et al [30].

Our findings highlight the importance of aligning digital self-management interventions with local workflows and capacities. Proactive, iterative support (such as checking usage data, discussing content during follow-up, and tailoring recommendations) appears to facilitate more meaningful engagement. Services adopting myCOPD may benefit from clarifying roles (eg, between clinicians and DHCs), ensuring adequate staffing, and embedding digital support within routine contacts. These implications arise directly from the contrast observed in our study: PR clinicians were able to provide iterative support because roles were clear and time was protected, whereas hospital clinicians lacked these structural conditions. More specifically, our data suggest that proactive delivery models may require explicit resourcing, including protected time for clinicians to review usage data and integrate digital discussions into routine care. This aligns with the argument of Greenhalgh et al [32] that digital health implementation depends on relational work and organizational readiness. Hospital settings, where workflow pressures limit such opportunities, may benefit from dedicated DHCs or structured onboarding protocols that extend beyond discharge.

At the patient level, recognizing the influence of multimorbidity, digital confidence, and socioeconomic context is essential for equitable implementation. Tailored onboarding, opportunities for hands-on support, and clear communication about the purpose and benefits of the app may help address barriers. These findings reinforce the value of attending to layers of vulnerability when designing and delivering digital self-management support. In practice, this may involve prioritizing hands-on demonstrations for patients with low digital confidence, offering multimorbidity-sensitive guidance that helps patients identify which app features align with their immediate priorities and ensuring that digital support is available beyond initial onboarding. These approaches are consistent with the suggestions of Vorrink et al [33] to improve digital self-management interventions.

For policy and commissioning, these findings suggest that digital interventions should not be implemented as stand-alone tools but as components of broader self-management pathways that account for organizational readiness and patient diversity. Commissioners may therefore need to incorporate workflow assessments, staffing considerations, and digital literacy support into procurement and implementation plans, ensuring that digital tools are embedded within coherent, adequately resourced pathways rather than added as isolated components. This aligns with Normalization Process Theory as outlined by Murray et al [34], which emphasizes the importance of integration within existing workflows.

Future research should examine how different models of delivery influence long-term engagement and clinical outcomes, including whether proactive support can be feasibly scaled. Further work is also needed to explore how digital interventions can be tailored for individuals with multimorbidity, low digital confidence, or limited access to technology. Mixed methods evaluations that integrate qualitative insights with detailed usage analytics may help identify mechanisms of engagement and inform more equitable implementation strategies.

Our findings point to several specific avenues for investigation, including comparative studies of proactive versus passive delivery models, evaluations of DHC roles within COPD pathways, and intersectional analyses that examine how layers of vulnerability shape engagement trajectories over time. Integrating relational measures of support with usage analytics may be particularly valuable for identifying mechanisms that drive sustained engagement, as suggested by the work of Lewis et al [28] on implementation mechanisms.

Delivery context and local workflows play a central role in shaping how patients engage with myCOPD. Proactive, relationship-based support appears to facilitate more integrated and sustained engagement, whereas passive onboarding may limit the intervention’s potential. Considering patient characteristics, particularly multimorbidity and perceived digital ability, alongside organizational capacities may support more equitable and effective implementation of digital self-management tools in COPD care. The concept of layers of vulnerability provides a useful lens for understanding how individual and structural factors combine to shape engagement trajectories.

By illustrating how delivery models interact with patient characteristics to shape engagement, this study contributes to a more nuanced understanding of digital self-management implementation and highlights the need for context-sensitive, equity-oriented approaches to digital health in COPD care.