This qualitative study explored the experiences of FHWs integrating the WHO Skin NTDs app into routine clinical workflows across 5 counties in Kenya—Baringo, Nakuru, West Pokot, Kwale, and Kajiado, purposefully selected by the Kenyan Ministry of Health due to their endemicity for skin NTDs. Participants represented FHWs across Kenya’s primary health care system, including national referral (level 5), county-level (levels 3 and 4), and community-level (level 2) facilities. Facilities were intentionally sampled across multiple levels of the health system to capture a diversity of clinical and community contexts, resource availability, and workflow structures shaping dermatological care delivery. These levels differ in staffing (eg, nurse-led [level 2] vs clinician-led [levels 3‐5] services), access to diagnostic and treatment resources, opportunities for peer consultation, and referral pathways. The aim of sampling across multiple facility levels and regions in this way was not to support any systematic formal comparison between levels. Rather, sampling across different facilities and regions provided an opportunity for particular distinguishing factors of these varied settings to be invoked by participants, as appropriate, in their explanatory accounts of usage, perceived usefulness, and impact. Data were collected through focus group discussions (FGDs) and semistructured interviews (SSIs) with participating FHWs. This qualitative component forms part of a broader assessment study on the performance of the app’s AI algorithm [30].

The WHO Skin NTDs app comprises two integrated components: (1) an educational component providing structured guidance on diagnosis and management, and (2) an AI-embedded component that generates differential diagnoses from uploaded images of skin lesions. The educational component includes interactive skin lesion localization and condition-specific information, clinical images, and management guidance adapted from the WHO training guide on skin NTDs for FHWs (Figure 1) [5]. This can be accessed during or outside clinical consultations to support learning and decision-making.

Within the “Signs & symptoms” component, users first indicate the anatomical location of the lesion using a body map interface (Figure 1A). They then select relevant signs and symptoms from a structured list (Figure 1B and C) and based on these inputs, the app generates a list of potential conditions, which users can explore further to access detailed information on diagnosis and management (Figure 1D-F).

The AI-embedded component is based on convolutional neural network models that analyze and classify uploaded skin lesion images. For each uploaded image, the model generates a small set of differential diagnoses with associated confidence levels (low, medium, and high) across 12 skin NTDs and 24 common skin conditions [21,22] (Figure 2). Given that over 80% of skin conditions encountered by FHWs are common skin conditions, their inclusion is essential for clinical usefulness [31].

The WHO Skin NTDs app is designed to be used during routine patient consultations, although no standardized protocol governs its integration within clinical workflows. In practice, use is flexible and embedded within the consultation itself, typically following initial history taking and physical examination. The anticipated workflow involves FHWs capturing an image of the skin lesion with the phone camera, uploading it to both the skin NTD and the common skin condition algorithms, and reviewing AI-generated differential diagnoses. These outputs are then interpreted alongside patient history and clinical examination findings. The app’s outputs are not prescriptive; rather, they are intended to support clinical reasoning, enabling FHWs to refine or rule out potential conditions. However, given the app’s finite disease library, the true diagnosis may not always be represented.

The algorithm was trained on a dataset comprising 5760 images of skin NTDs alongside 16,577 images of other skin conditions, 2469 images of healthy skin, and 50,000 non–skin images to support classification boundaries [30]. The training set included a diverse range of skin tones, with approximately 70% of images representing dark skin types addressing key concerns around bias in dermatology AI [30,32]. A more detailed account of the app’s design, development process, and training approach is provided elsewhere [30].

Participants were nondermatologist FHWs who routinely manage patients with skin NTDs or common skin conditions at the primary care level. Five counties were purposively selected in collaboration with the Ministry of Health, Kenya. Within each county, health facilities and affiliated FHWs were identified with support from the County Departments of Health. From this sampling frame, 50 FHWs were randomly selected and invited by telephone to participate in an initial training and deployment workshop in April 2024.

The training and deployment workshop, held at a centrally located training center in Kenya, introduced participants to the WHO Skin NTDs app. The training combined short didactic sessions with practical exercises. Content included (1) step-by-step instruction on app navigation, image capture, and interpretation of AI-generated differential diagnoses, and (2) guidance on consent practices. Participants engaged in supervised practical sessions where they used the app on peers to familiarize themselves with consent practices, image capture, and result interpretation.

Following the first training workshop, participants were added to a WhatsApp group with members of the study team to facilitate ongoing technical support and troubleshooting during the deployment period. Participants were invited to integrate the app into their routine clinical practice over a deployment period of approximately 6 months. After the 6-month deployment period, a second evaluation session was convened in November 2024, again at a central location in Kenya. This session focused on understanding participant reflections on their experience using the app and comprised the main qualitative data collection activities. During this session, FGDs and SSIs were conducted to explore participants’ experiences using the app, including perceived benefits, challenges, and contextual influences on use. Of the original 50 FHWs invited, 47 attended the first training workshop and participated in the deployment phase. Thirty-six attended the second evaluation workshop and contributed to the qualitative data collection.

This study was approved by the research ethics committee at the Open University of Catalonia, with the Kenyan Medical Research Institute (KEMRI) protocol number KEMRI/SERU/CTMDR/122/4959, Open University of Catalonia reference number CE23-RC21, and the Kenya National Commission for Science, Technology and Innovation license number NACOSTI/P/24/35733. Written informed consent was obtained from all participating FHWs. The FHWs were informed that participation was voluntary and that they could withdraw from the study at any time without any consequences. All FHWs participating in the second workshop provided written consent for audio recording and the use of data for analysis and publication. In addition, FHWs obtained written informed consent from patients before using the WHO Skin NTDs app during clinical consultations during the 6-month study period. Patients were informed that participation was voluntary and that their standard of care would not be affected by refusal to participate. All data were deidentified prior to analysis to ensure participant confidentiality. No personally identifiable information is reported in this study. FHWs were not financially compensated for participating in this study, although travel and subsistence costs (11,000 Kenyan Shillings [1 Kenyan Shilling=US $0.0077 as of June 26, 2026] per FHW, based on government rates) were provided to support workshop attendance. Mobile data bundles (5000 Kenyan Shillings per FHW) were also provided to support app use within clinical workflow.

Data were collected through 15 SSIs (each lasting 30‐45 minutes) and 4 FGDs (1‐1.5 hours, 8‐10 FHW in each). The SSIs enabled a focused exploration of individual FHW perspectives, while the FGDs offered opportunity to encourage debate and consensus building, highlighting areas of agreement and contrast in perspectives on the WHO Skin NTDs app use within diverse clinical workflows. The interviewers (EQ, RN, CC, AF, and DA) used semistructured topic guides to guide FGDs and SSIs, which were reviewed by all authors prior to the workshop (Multimedia Appendices 1 and 2). All FHWs participated in FGDs, and 15 FHWs were purposively selected to participate in SSIs to ensure a balanced mix across counties.

Both SSIs and FGDs were used to enable complementary insights. SSIs allowed for in-depth exploration of personal experiences, with participants asked to recount specific instances of app use in situ. This yielded rich, contextualized accounts of how the app was integrated into individual diagnostic decision-making, including nuanced challenges and variations in use. FGDs, by contrast, enabled group-level reflection, providing a space to examine shared practices, compare perspectives, and identify points of consensus or disagreement regarding the app’s usefulness role in workflow and broader systemic challenges. This combination of methods was intended to enhance both the depth and breadth of understanding regarding app integration in routine care.

All interviews and FGDs were conducted in English (an official language in Kenya). Each FGD was cofacilitated by a local research scientist from KEMRI (RN and DA) and a researcher from the Global North (EEVQ and CC). All FGDs and SSIs were recorded on portable Dictaphones and stored on a secured university-licensed cloud service (OneDrive). Field notes were taken by all researchers during and after both FGDs and SSIs and were used to support the thematic analysis.

Audio recordings were transcribed verbatim by the corresponding author (EEVQ), supporting deep familiarization with the data. Transcripts were analyzed thematically using NVivo (version 14; QSR International) software. Given that this is the first qualitative assessment of the AI component of the WHO Skin NTDs app usability and acceptability, a bottom-up inductive coding approach was selected to support the iterative construction of themes grounded in participant narratives. This exploratory approach aimed to capture nuanced insights that may inform the application of appropriate theoretical frameworks in future studies, as the evidence base on the app’s integration expands. EEVQ conducted initial coding of all transcripts. A subset of transcripts was independently double coded by AB and KO. Coding discrepancies were resolved through team discussion, and the codebook was iteratively refined. The full dataset was analyzed using reflexive thematic analysis, with codes inductively organized into 4 higher-order themes within the research team [33].

The derived themes from the analysis made reference to behavioral patterns and interdependencies that lent themselves to communication through systems-thinking visualizations. With this in mind, we constructed 2 causal loop diagrams (CLDs) grounded in FHW narratives to provide a structured way to visualize how system-level dermatological workflow factors interact in practice, both before and after app integration [34]. The intention here is not to generate formal theory but rather to offer descriptive representations of diagnostic and referral pathways and contextual factors shaping their progression as revealed through the FHW accounts. In this respect, the CLDs function as an analytic and communication tool to scaffold the interpretation and presentation of the overall thematic narrative.

The 36 participants who attended the second workshop in November 2024 comprised FHWs from 5 counties in Kenya: Nakuru (n=11), a fairly urbanized county situated in the Rift Valley; Baringo (n=8), also situated in the Rift Valley, characterized by semiurban and rural areas; Kajiado (n=6), bordering Tanzania in southern Kenya with a seminomadic population; West Pokot (n=8), a remote rural county bordering Uganda, predominantly inhabited by nomadic pastoralist communities; and Kwale (n=3), a largely rural, coastal county in southeastern Kenya [35-39]. Participant characteristics are described in Table 1.

The thematic analysis highlighted shifts in dermatological care delivery, representing the complex dynamics before and after app integration, visualized through 2 CLDs (Figures 3 and 4). The CLDs are grounded in FHW narratives, with illustrative quotes supporting each node (Multimedia Appendix 3). Yellow nodes represent patient-related factors, gray nodes denote FHW-related elements, green arrows indicate changes postapp implementation, and reinforcing loops perpetuate challenges or amplify improvements. Connections depict the relationships between these factors, as articulated by FHWs, indicating whether they have a positive or negative influence. The diagrams synthesize FHW perspectives across multiple facility levels rather than disaggregated by facility levels. However, we acknowledge that the key elements of aggregated narratives will have a differential impact across the different facility.

Before app integration, FHWs reported patterns of habitual referral of dermatological cases, where nonspecialist dermatological knowledge often limited immediate diagnostic assessment capability, combined with limited access to diagnostic resources and specialist training that would enable necessary upskilling (Figure 3). This resulted in FHWs expressing a sense of clinical detachment, reinforcing a cycle where FHWs avoided managing dermatological cases at their local facility; instead, depending upon habitual referral to higher-level facilities and specialist services. FHWs expressed that this referral dependency strained dermatology services, which they associated with diagnostic delay and subsequent patient attrition (loss to follow-up). The diagram further highlights a consequent reinforcing feedback loop—the more FHWs referred cases, the less experience they gained in managing them, further diminishing confidence and knowledge in dermatological care.

After app integration, FHWs described a shift toward more immediate dermatological case management at their local-level facility, interrupting the cycle of habitual referral to higher-level facilities and specialist services (Figure 4). The app was perceived as enhancing dermatological knowledge acquisition and development, reinforcing a sense of clinical confidence and greater feeling of ownership of dermatological cases. Local facility case management was perceived to facilitate earlier diagnosis and treatment, which FHWs linked to improved patient trust in their clinical abilities. Additionally, FHWs highlighted the app’s role in destigmatization of skin conditions, as greater diagnostic confidence reduced the need for voyeurism of collaborative deliberation over patients. FHWs spoke of patients becoming “ambassadors” within communities, actively promoting care-seeking behavior based on their positive experiences with the app. The diagram further highlights a consequent reinforcing feedback loop—as FHWs gained dermatological knowledge, their confidence in managing cases increased, fostering greater clinical ownership. This, in turn, reduced reliance on specialist referrals as FHWs managed more cases at the facility level, with potential implications for strengthened primary dermatological care.

These system-level causal-loop representations (Figures 3 and 4) offer high-level synthesis of FHW-identified factors influencing the transformation of care and referral pathways. Building on this, we present a deeper dive into the phenomenological accounts of FHW-reported app use and the perceived transformation of practice experienced arising from its adoption.

The integration of the WHO Skin NTDs app into diagnostic workflows was seen to facilitate a shift from habitual referral to local facility management. FHWs described this transition as improving patient retention and fostering wider community engagement (Figures 3 and 4).

Findings suggest that the shift from habitual referral to local facility case management was driven by FHWs’ perception of improved dermatological knowledge, in turn enhancing clinical confidence (Figures 3 and 4). The majority of FHWs suggested that the app did not replace clinical judgment but seemed to reinforce it—functioning as an assistive tool that encouraged active engagement with differential dermatological diagnoses.

Findings show that the integration of the WHO Skin NTDs app into clinical workflows appeared to shift diagnosis from a clinician-driven process to a more participatory model, actively involving patients in their care. Initial narratives from FHWs indicate that the app seemed to function as both a clinical decision support tool and a patient engagement mechanism.

FHWs envisioned the WHO Skin NTDs app evolving beyond its current role as an assistive tool for capacity building within the individual clinical encounter. Their recommendations highlighted its potential to function as part of an interconnected health system, linking clinical dermatological care with broader health system infrastructure.

This study presents a comprehensive understanding of FHWs’ experience and perception of the WHO Skin NTDs app in Kenya. The majority of FHWs described the app as facilitating the following: (1) validating their clinical judgment, which reduced dependency on referral pathways and improved patient retention and community engagement; (2) incremental knowledge acquisition and growing confidence in managing dermatological conditions independently; and (3) enhanced patient engagement and perceived trust in the diagnostic process.

These findings contribute to a growing body of evidence on how mHealth tools are being integrated into resource-limited frontline health systems and the ethical and pedagogical challenges this raises, particularly given the app’s intended function as an educational and capacity-building tool [21,30,40]. By contrasting FHWs perceptions of pre- and postapp clinical workflows (Figures 3 and 4) and integrating FHW narratives (Multimedia Appendix 3) with thematic analysis, we identify potential system-level challenges associated with dermatological care and specific mechanisms of change facilitated by the app.

These insights inform both app refinement and broader implementation strategies to maximize pedagogical value and ensure ethical integration of mHealth tools in frontline health care systems. Taken together, they suggest that the app is having a positive impact on dermatological care within the studied context and is, in many cases, being used in alignment with its educational remit. However, there remains a need to examine edge use cases and potential risks—particularly where usage may extend beyond the app’s original design parameters [23].

This study highlights the central role of clinical improvisation—real-time adaptation of clinical workflows and care strategies—as a core component of dermatological care in resource-limited settings [41,42]. In sub-Saharan Africa, FHWs routinely navigate complex social, political, infrastructural, and economic constraints, necessitating clinical improvisation [41]. FHWs identified multiple systemic barriers impacting dermatological care, including limited diagnostic resources, inadequate training, scarce dermatology expertise, and high patient volumes (Figure 3). In response, they described improvisational practices such as habitual referral, informal WhatsApp consultations, and public deliberation of clinical cases (Figure 3). Crucially, these practices are not merely indicative of knowledge gaps or health system limitations; rather, they reflect locally adapted strategies developed under constraints, described elsewhere as the “craft” of care under constraint [41].

However, while such strategies may enable care delivery where formal systems fall short, our findings suggest that some may also inadvertently limit opportunities for building and maintaining dermatological knowledge and case management skills. In this study, FHWs linked habitual referral and reliance on external consultation to diagnostic delays, patient attrition, and stigma–patterns observed across similar African contexts [43,44]. These findings suggest that some adaptive strategies may unintentionally reinforce ongoing dependency on specialist services by reducing opportunities for local dermatological knowledge development and case management experience.

Our findings suggest that integrating the WHO Skin NTDs app into clinical workflows may help reshape improvisational practices by facilitating more reflective diagnostic reasoning and greater confidence in local facility case management. While some of these benefits may also arise through other forms of clinical decision support, the AI-enabled nature of the app allows FHWs to compare image-based differential diagnoses with their own clinical assessments in real time.

In the broader literature, AI-embedded tools are ideally situated within a model of human-AI symbiosis: a collaborative relationship where the clinician’s reasoning complements the algorithm’s suggestions [45,46]. Most FHWs in our study described using the app in these terms—referring to it as a “partner” [Participant 15], akin to a second reader that supports and moderates, rather than supplants, their clinical judgment. FHWs reflected that this perceived partnership enabled more confident and independent patient management at the local-level facility, reducing reliance on improvisational workarounds such as habitual referral. This could be interpreted through the lens of self-efficacy theory, which posits that an individual’s engagement in a behavior is related to their belief in their ability to perform such tasks [47]. In this context, the app appears to function as a mechanism for enhancing FHWs’ perceived self-efficacy in dermatological care, supporting their ability to act on clinical decisions locally. This reflects wider literature on how AI-embedded clinical decision support systems can be, and should be, designed to augment rather than replace human reasoning [46,48].

However, our findings also suggest that when the app’s output did not align with the FHW’s clinical judgment, FHWs often reverted to referral to specialist dermatologists. While this can be understood as an appropriate response to clinical uncertainty, it also highlights an important opportunity for future design refinement. AI-embedded decision support tools may benefit from incorporating structured support mechanisms to guide arbitration in such cases, ensuring that the burden of resolving diagnostic discrepancies does not fall solely on the FHW. Such mechanisms could further enhance the app’s pedagogical value by facilitating reflection and learning through structured resolution pathways, rather than relying on referral as a default form of risk mitigation.

Our study raises a potential ethical concern with respect to how some FHWs may inadvertently use the app beyond its intended remit as an educational clinical decision support tool, potentially engaging in care beyond their clinical competence [23]. One FHW, for example, explicitly described deferring entirely to the app’s diagnostic suggestions. This pattern may reflect elements of automation bias, whereby users place undue trust in algorithmic outputs over their own clinical judgment [49]. Given the limited dermatological knowledge reported by FHWs, such potential overreliance may raise medicolegal and ethical concerns [23].

Some scholars caution that as AI systems are perceived to outperform human judgment, they may begin to displace human epistemic authority [50]. This risk is echoed in broader literature, which highlights that AI tools may contribute to clinician deskilling, diminished critical reasoning, and reduced capacity to navigate complex or atypical cases through overreliance [48-53]. While often discussed as a future-facing concern, our findings suggest that elements of this dynamic may already be emerging in contexts where dermatological expertise is limited [50]. In such cases, for some FHWs, the app’s output may be seen not as suggestive but as directive—an inversion of its intended role.

Meaningful human-AI collaboration depends on users having sufficient knowledge to question and validate algorithmic suggestions—a process that relies on training, confidence, and the time and institutional space to exercise clinical judgment [46]. Many FHWs in our study described using the app in this way, comparing AI-generated differentials with their own clinical assessments and reporting perceived knowledge gains through continued app use. Such experiences may, over time, strengthen their capacity to engage critically with AI outputs. Conversely, where foundational dermatological knowledge is limited, the same system may increase the risk of inappropriate reliance on algorithmic suggestions.

This presents a paradox in the positioning of the WHO Skin NTD app. The tool is explicitly designed to support capacity building among FHWs in response to limited dermatological expertise [21,22,30]. Yet, for the app to be used responsibly, FHWs must possess sufficient baseline knowledge to critically interrogate and contextualize its outputs. As our results suggest, this prerequisite cannot be assumed and is likely to vary significantly across different health system contexts, facility levels, and individual FHWs. To help mitigate this risk, explainability of AI outputs is critical: presenting outputs in transparent, interpretable ways can better support FHWs in understanding and validating diagnostic suggestions [54]. Other human-centered design principles, such as offering contextual guidance and supporting stepwise reasoning, could further assist FHWs in maintaining epistemic authority while engaging with the app. Features already present in the app, such as graded confidence levels associated with differentials, may help surface diagnostic uncertainty in ways that encourage critical engagement [54-56].

In addition to this epistemic tension, it is critical to recognize that the dynamics of AI use in clinical practice are not shaped by individual knowledge alone. Broader contextual factors, including patient load, time constraints, and infrastructural limitations, also influence how AI outputs are interpreted and acted upon [45]. This reinforces the importance of designing and adapting AI tools and implementation strategies in ways that explicitly protect and promote clinical autonomy, both in individual decision-making and across broader health care delivery [45,57,58].

Three areas of further work are needed. First, direct observation of app use in situ will be critical to understanding how it is conceptualized in practice, how broader contextual factors shape its use and whether it is functioning as intended (as a clinical decision support tool used for educational purposes within the clinical encounter), or whether some usage patterns risk reclassification. This will also be critical for elucidating how patients perceive the app's role in the collaborative context of FHW-patient interaction. Second, targeted design adaptations, such as embedded disclaimers or periodic training modules, should be explored to reinforce the app’s role as a nondirective support tool and bolster FHWs’ dermatological knowledge to safeguard clinical autonomy. Third, development of a context-specific professional ethical guidance framework for health care AI use (PEG-AI) may be necessary to ensure the safe, responsible, and ethically robust integration of the app into frontline care [23]. PEG-AI offers a context-sensitive governance approach for AI-embedded tools, essential for ensuring ethical implementation in resource-limited settings and safeguarding clinical autonomy, regardless of medical device classification [23].

Maintaining the use of the app within the boundaries of clinical competencies is therefore not solely a function of the tool itself but depends on the interaction between design features (eg, explainability, confidence indicators, and embedded guidance) and broader system-level supports, including training, workflow integration, and context-specific ethical governance.

In our study, only 1 FHW had ever heard of the WHO manual “Recognising Neglected Tropical Diseases Through Changes on the Skin: A Training Guide for Frontline Health Workers” [5]—a training guide specifically developed to build dermatological capacity among FHWs in resource-limited settings. Interestingly, the entire contents of the app are derived from this manual. However, our findings suggest that the format in which the information is both accessed and delivered shapes engagement and learning. While the manual offers comprehensive guidance, its didactic, decontextualized format perhaps separates learning from real-world clinical application. By contrast, the app delivers this same content in ways that are tailored to individual clinical cases, offering real-time feedback and follow-up information that are directly tied to patient encounters. Embedded in the relational space of the consultation, the app appears to support experiential, situated learning [59]. This aligns with well-established principles from the situated learning literature which emphasizes that learning is most effective when embedded within authentic contexts of practice [60-62].

In optimal conditions, the affordance of a continuously connected mobile device likely enhances this mode of learning. However, in our study, limited internet connectivity emerged as a barrier to consistent app use. Ensuring that the app’s AI algorithm functions offline will be essential for equitable and sustained integration—an improvement the WHO Department for NTDs is currently exploring.

Existing literature suggests that over 80% of dermatological cases encountered by FHWs at the primary care level are common skin conditions, which are generally simpler to diagnose and manage [31]. In our study, FHWs reported that the app supported their ability to identify and manage these common skin conditions, often confirming their clinical judgment and enabling local facility management. However, when confronted with complex cases—particularly suspected skin NTDs—our findings show that FHWs remained more likely to defer to specialist referral, especially when the app’s diagnostic output could not be corroborated by their own limited dermatological knowledge. This pattern aligns with a key dimension of the PEG-AI framework which emphasizes that AI-embedded tools should be used within the boundaries of clinical competencies [23].

In this context, the WHO Skin NTDs app appears to strengthen local dermatological care by enabling FHWs to manage common skin conditions more confidently, potentially reducing unnecessary referrals. This may help alleviate pressure on overburdened dermatology services, allowing specialists to focus on more complex or rare presentations, particularly skin NTDs.

However, this shift in clinical workflow introduces algorithmic implications. As FHWs increasingly retain patients at the facility level, the accuracy of the algorithm becomes more consequential. In such cases, it may be preferable for the algorithm to prioritize sensitivity over specificity (ie, to favor false positives over false negatives) to minimize the risk of missed cases, particularly as reliance on referral as a default pathway has declined (Figures 3 and 4).

This shift also raises important system-level considerations. While strengthening that facility-level management may improve access to care and reduce pressure on referral systems, it may also introduce new risks. If only suspected skin NTDs cases are escalated—while structural barriers to referral (such as distance, cost, and dermatologist shortages) remain—patients with complex dermatological conditions may still face delayed diagnosis and discontinuity of care, particularly in rural and underserved areas.

Assessing the full impact of AI-embedded tools such as the WHO Skin NTDs app—both intended and unintended—requires moving beyond the individual clinical encounter to examine how such tools reshape referral pathways, clinical workflows, and health system capacity. A systems-level process evaluation will be essential to understand these dynamics and their implications for patient outcomes, including health economic considerations [63,64]. This aligns with broader calls to evaluate mHealth interventions not only at the point of care but across the full health system to surface both enabling factors and unintended consequences [65].

A key challenge frequently reported by FHWs was the lack of immediate management and follow-up support following the app's diagnostic suggestions. While the app provides AI-generated diagnostic suggestions instantaneously, the accompanying disease management guidance—although present—is in a separate section on the app and not seamlessly integrated into the clinical decision support workflow. This design choice may inadvertently reinforce a fragmented view of care, privileging diagnosis as an endpoint rather than as part of a broader continuum that includes, for example, management, follow-up, and psychosocial support [41]. This is particularly relevant, given the chronic course and complex illness trajectories associated with many common skin conditions and skin NTDs [66].

Enhancing the app’s capacity to comprehensively support FHWs across the full illness trajectory—by embedding management guidance directly after diagnostic suggestion—could significantly broaden its value. The section of the app describing how to diagnose and treat each skin disease incorporates links to targeted training modules or embedded decision pathways (such as those aligned with WHO Academy resources) to better equip FHWs to deliver holistic, sustained care that responds to the complex and chronic illness trajectories of dermatological conditions [67].

This study has several limitations. First, findings reflect the perspectives of FHWs only; patient perspectives on the app’s influence on care, agency, and experience were not explored. This is a key aim that will be addressed in forthcoming ethnographic fieldwork. Second, FHW responses may have been influenced by social desirability bias, given some of the research team’s involvement with the app’s development. However, independent facilitation was used to mitigate this risk [68,69]. Third, reported shifts in care may reflect short-term novelty effects rather than sustained practice change [70].

While this study aimed to generate a systems-level understanding of the app’s implementation, future research should complement this with detailed patient pathway mapping and comparative studies across diverse contexts to assess transferability [62]. Of the 50 FHWs initially enrolled, 47 participated in the training and deployment phase, and 36 FHWs attended the follow-up workshop where qualitative data collection occurred (72% follow-up). This is notable, given the wide geographical spread of participants and the logistical barriers involved in attending a centralized workshop. All 5 counties were represented among those who attended. However, it is possible that nonattendees had lower engagement with the app during the deployment phase, although no definitive reasons were provided for their absence. This potential attrition bias should be considered when interpreting findings.

This study offers preliminary insights into the ways FHWs perceive the AI component of the WHO Skin NTDs app is reshaping frontline dermatological care in a complex, resource-limited health system. FHWs reported that the app supported more confident local-level facility management, disrupted habitual referral patterns, and fostered incremental knowledge acquisition. Importantly, our findings highlight the ethical and pedagogical complexities of integrating AI-embedded tools into frontline care. While FHWs reported that the app often strengthened clinical reasoning and learning, risks of overreliance and unintended shifts in clinical autonomy remain, particularly where baseline dermatological knowledge is limited. Future research should explore these dynamics through in situ observation and pathway-level evaluations. With continued attention to both pedagogical value and ethical governance, the WHO Skin NTD app has strong potential to advance the WHO’s vision of equitable, person-centered skin health for all.