Diagnostic performance of digital and clinical approaches for detection of podoconiosis by community health workers in Rwanda: a comparison with dermatologist diagnosis

1 Introduction

The detection of podoconiosis at the community level is crucial for preventing disease progression, reducing disability, and enabling timely interventions such as foot hygiene, use of footwear, and community education to halt further cases. The main challenge in the detection of podoconiosis lies in accurately distinguishing it from other causes of leg swelling, such as lymphatic filariasis, before irreversible lymphedema develops. Podoconiosis is a non-infectious, chronic, debilitating inflammatory condition that develops as a result of prolonged exposure to specific irritant minerals present in the red clay soil (1, 2). This disease primarily affects the lower limbs, leading to significant swelling and deformity, which substantially hampers the daily activities and quality of life of those impacted (3). The hallmark signs and symptoms associated with podoconiosis include lymphedema—persistent swelling of the lower legs—and acute episodes known as dermatolymphangioadenitis (ADLA) (4). These episodes are characterized by systemic symptoms such as fatigue, fever, chills, widespread inflammation, limb swelling, inflammation of the lymphatic vessels (lymphangitis), enlarged lymph nodes (adenitis), and eventually, peeling or shedding of the skin (4).

Podoconiosis is commonly classified into five clinical stages, ranging from early, reversible swelling to advanced, irreversible elephantiasis with significant skin changes (5). Stage 1 is characterized by reversible swelling that subsides with rest, while Stage 2 involves persistent swelling that does not fully resolve overnight. Stages 3 to 5 represent progressively advanced disease, including nodular changes, fibrosis, and mossy lesions. Early-stage podoconiosis typically refers to Stages 1 and 2, when swelling is mild and largely reversible, and when preventive and hygiene-based interventions are most effective (4).

Current global estimates indicate that approximately 4 million individuals are affected by podoconiosis across various regions, including Africa, certain parts of Latin America, and Southeast Asia (3, 6, 7). These figures highlight the significant burden of the disease in these geographic areas, where it remains a major public health concern (3). Recognizing the importance of addressing this neglected condition, the World Health Organization (WHO) formally acknowledged podoconiosis as one of the neglected tropical diseases (NTDs) in 2011, bringing greater international attention to its prevention and control (3, 8).

The disease carries profound social and economic repercussions, largely through the stigma experienced by affected individuals and the consequent loss of productivity (9, 10). In many endemic communities, individuals suffering from podoconiosis may face social exclusion, discrimination, and diminished opportunities for work, which further exacerbates the disease’s burden on their lives and livelihoods (9–11). Additionally, preventive strategies for podoconiosis focus on minimizing contact with the irritant minerals found in the soil. Consistent use of protective footwear, maintaining proper foot hygiene, and covering floors are effective preventative measures (12, 13). These interventions can significantly reduce the risk of disease development, especially in populations at high risk. Early management of the disease involves straightforward interventions aimed at alleviating the morbidity associated with lymphedema. Such management includes diligent foot hygiene, skin care routines, wound management, the use of compression techniques, physical exercises to promote lymphatic drainage, elevation of the affected limbs, and prompt treatment of acute infections, such as ADLA episodes (1, 14–16).

Apart from these approaches, the diagnosis of podoconiosis remains challenging. The disease presents with symptoms similar to other causes of lymphatic swelling, and currently, there is no reliable point-of-care diagnostic test available (9). The diagnosis is primarily clinical, relying on careful examination and a thorough exclusion of other potential causes of lymphedema, such as lymphatic filariasis, which can present with similar features (17). Effective diagnosis, therefore, hinges on a clinician’s expertise in differentiating podoconiosis from other conditions and understanding the epidemiological context in which it occurs.

In order to overcome this problem, The World Health Organization (WHO) has developed the Skin NTDs app, a digital tool that operates offline and is designed to aid health workers in diagnosing and managing various skin-related neglected tropical diseases, including podoconiosis (18). The app features a clinical algorithm that guides users through a structured assessment process, helping to differentiate podoconiosis from other causes of lymphedema or other dermatological conditions (18). In 2017, Deribe and colleagues developed and validated a clinical algorithm based on patient history and physical examination, which can be utilized in regions suspected or endemic for podoconiosis (4).

In Rwanda, podoconiosis is recognized as an endemic disease, yet there is a significant shortage of specialized healthcare workers, particularly dermatologists and clinical experts trained in lymphatic diseases (19). Consequently, there is an urgent need to evaluate effective diagnostic methodologies. This study compared approaches to diagnosis of podoconiosis by CHWs, with dermatologists considered as the clinical reference standard for definitive diagnosis to reduce disability and improve patients’ quality of life. By assessing these approaches, the study aims to determine whether the WHO app or Clinical Algorithm can serve as reliable tools for frontline health workers in Rwanda, thereby improving case detection, enabling timely management, and ultimately reducing the disease burden in endemic communities.

2 Materials and methods

2.1 Study design

This cross-sectional study employed quantitative methods to compare the performance of the clinical algorithm (Figure 1) The flowchart illustrating the diagnostic logic underpinning the application, including key decision points based on patient history, clinical signs, and exclusion criteria used to differentiate podoconiosis from other causes of lymphedema and a representative screenshot of the application interface to demonstrate how the tool is used in practice by CHWs are shown in Figures 2a, b).

The clinical algorithm was developed through a thorough review of existing literature and adaptation from prior studies on podoconiosis diagnosis (4, 20, 21). It was designed to be a straightforward, symptom- and sign-based tool leveraging clinical history and physical examination findings to identify suspected cases (4). The algorithm’s development involved expert consultations and validation studies to ensure accuracy and ease of use in low-resource settings, particularly in endemic areas (4, 20, 21).

The WHO Skin-NTD mobile application is a clinical decision-support tool that guides users through a structured, step-by-step diagnostic process based on patient history and physical examination findings. CHWs enter key symptoms and signs into the application, which then applies a predefined algorithm to generate a presumptive diagnosis. For podoconiosis, the app incorporates criteria such as bilateral lower limb swelling, chronic progression, history of prolonged barefoot exposure to irritant soil, and the absence of systemic or infectious causes. The tool also integrates exclusion criteria to differentiate podoconiosis from other causes of lymphedema, particularly lymphatic filariasis and other dermatological or systemic conditions. Based on the combination of these inputs, the app suggests a probable diagnosis and provides guidance on case management and referral (Figure 2b). While the application includes modules for multiple skin neglected tropical diseases, Figure 2b provides an illustrative example of the interface and does not represent all diagnostic pathways available within the application. The app has been tested in various endemic regions worldwide, including Africa and Southeast Asia, demonstrating its utility in aiding CHWs with standardized disease recognition and case documentation. It provides standardized guidance, easy data recording, and supports remote consultation. It may improve diagnostic accuracy and facilitate reporting. However, it requires access to compatible mobile devices, and ongoing technical support, which could be challenging in resource-limited settings (22). It is important to note that the app supports presumptive diagnosis and is not intended to replace clinical confirmation by trained clinicians, particularly in complex or atypical cases.

2.2 Study site

Rwanda’s population has reached 13,246,394 people according to the results of the fifth Population and Housing Census (PHC) conducted in 2022. The country is located in East Africa and is bordered by Uganda, Burundi, the Democratic Republic of the Congo, and Tanzania. Rwanda is subdivided into 4 provinces (Northern, Southern, Eastern and Western) as well as the City of Kigali. Nationwide mapping estimated prevalence across Rwanda at 0.05% (6,429 people) (23). The regions with highest prevalence of podoconiosis are West-south (Nyamasheke, Rusizi and Nyaruguru districts) and North-west provinces (Musanze, Nyabihu, Burera and Rubavu districts) (19).

Rubavu District was purposively selected for this study due to its relatively high prevalence of podoconiosis, its representativeness of endemic settings in Rwanda, and the presence of an established CHW’s network, which facilitated implementation of community-based screening and evaluation of the diagnostic tools under routine field conditions.

2.3 Target population

The target population consisted of adults presenting with leg swelling of any cause within the Rubavu district. According to the population and housing census of 2022, the estimated number of people aged 18 years and above in Rubavu district (2022) was ≈ 288,000.

2.4 Eligibility criteria

Participants were eligible for inclusion if they were adults aged 18 years and above, residing in Rubavu District, and presenting with leg swelling of any cause. Individuals were identified through health facility records, outpatient registers, and CHW listings. All individuals who provided informed consent were included, regardless of the suspected cause of swelling or stage of disease. Participants who did not receive a definitive diagnosis from the dermatologist (clinical reference standard) were excluded from the final analysis.

2.5 Community engagement strategy

This study is nested within a larger implementation research project which itself is complemented by extensive community education and mobilization aimed at decreasing misconceptions about disease etiology and transmission and stimulating awareness of ways to prevent and treat podoconiosis. Among other activities, a community sensitization workshop involving CHWs, government representatives, community and religious leaders was conducted. This group of people played an important role in disseminating information about the diagnosis of podoconiosis and acted as the first level to controlling the disease in the community. Engagement with the community continued through the implementation of this study.

2.6 Sample size calculation

Sample size was calculated using Buderer’s formula (1996), a method of calculating sample size in studies of diagnostic tests by determining the precision of the expected sensitivity and specificity. We assumed the sensitivity of the mobile App to be 85%, with a 10% clinically acceptable width of the 95% CI, corresponding to 0.75 - 0.95. With 80% prevalence of podoconiosis in the target population (people with leg swelling), the required sample size according to Buderer’s formula was calculated to be:

[1.96^2 x 0.85 x (1−0.85)/0.10^2]/0.8=48.98/0.8=61.23

A separate calculation was carried out for the specificity. We assumed a specificity of 90%, and chose the width to be 8% (i.e. 0.08), corresponding to a 95% CI equal to 0.82 - 0.98. The required sample size under these assumptions was

[1.96^2 x 0.90 x (1−0.90)/0.08^2]/(1−0.8)=270.11

Thus, the required sample size was 270 participants in this study.

2.7 Sampling techniques

This cross-sectional study was conducted in Rubavu District, following a sector-by-sector approach that began with the sector reporting the highest number of individuals with leg swelling, as identified in the sampling frame. The sampling frame included individuals identified as having leg swelling based on health facility records and CHW’s registers (“identified cases”). Individuals who met the eligibility criteria, were available during the study period, and provided informed consent were enrolled as study participants (“recruited patients”). The targeted sectors included Busasamana, Bugeshi, Rugerero, Nyamyumba, Rubavu, Nyundo, Kanama, Mudende, Nyakiliba, Gisenyi, Kanzenze, and Cyanzarwe. No further selection was applied within each sector; all individuals who consented to participate were included in the study.

2.8 Data collection procedures

Data were collected using an administered structured questionnaire developed in English and translated into Kinyarwanda. Individual data on age, sex, education, occupation, place of residence, shoe wearing, foot hygiene practices and stage of disease were recorded, as well as household data on water, sanitation, and hygiene. The questionnaire was administered to the study participants by trained data collectors with support from a community engagement officer and the project research assistant. Then CHWs used the mobile App or the clinical algorithm to make a diagnosis on each patient. It took approximately 30 minutes to administer the questionnaire and perform diagnostic tests for each participant.

The study team made an initial visit to the health centre of each sector, at which we worked with health centre staff and CHWs to identify participants, explain the study and gain their consent. During this visit, CHWs received standardized training prior to data collection to ensure consistent application of the diagnostic tools. The training was conducted by members of the research team and included both theoretical and practical components. The theoretical component covered basic knowledge of podoconiosis, including its causes, signs and symptoms, staging, and differentiation from other causes of lower limb swelling. The practical component involved hands-on training in the use of the WHO Skin-NTD mobile application and the clinical algorithm, including guided case scenarios and supervised practice sessions.

CHWs were divided into two groups and trained separately based on the diagnostic method they would use (mobile application or clinical algorithm). Each training session emphasized accurate data entry, adherence to the diagnostic steps, and appropriate patient interaction. Competency was assessed informally through observation during training and supervised field practice prior to the start of data collection.

At the second visit, participants came to the health centre, where they were assessed by one CHW using the mobile App, another CHW using the clinical algorithm independent of the first and a dermatologist (clinical reference standard) for confirmation. The dermatologist collected clinical information on both legs and entered this into a tablet, including the distribution and laterality of swelling, duration and progression of symptoms, presence of characteristic skin changes (e.g., nodules, fibrosis, mossy lesions), history of acute dermatolymphangioadenitis (ADLA) episodes, footwear use and exposure to irritant soil, and features used to exclude other causes of lymphedema. The approach was repeated sector by sector, until we reached the sample size. To minimize review and observer bias, all diagnostic assessments were conducted independently under blinded conditions. The two CHWs assessed each participant independently, with one using the WHO Skin-NTD mobile application and the other using the clinical algorithm. Neither CHW had access to or was aware of the other CHW’s assessment or diagnosis. The dermatologist, who served as the reference standard, examined each participant independently and was blinded to the findings of both CHWs. Likewise, data collectors recorded the results of each assessment separately and did not have access to the results of previous assessments during data collection. Diagnostic results were only compared during the data analysis phase after completion of all participant assessments.

The sensitivity and specificity of the WHO mobile App and the clinical algorithm were compared. Data collection was performed using REDCap on tablets. At the end of each day, the researcher cross checked data for completeness. Data collectors could not proceed to the next question without completing all required fields. The study used automated validation systems to ensure high quality of data.

The clinical algorithm was applied alongside a structured clinical assessment aimed at excluding other causes of lower limb lymphedema. In addition to lymphatic filariasis, these included chronic venous insufficiency, cardiac failure, renal disease, traumatic or post-surgical lymphedema, and other systemic or dermatological conditions that may present with similar features.

Podoconiosis was classified based on established clinical criteria and staging systems described in the literature (4, 16).

2.9 Community health worker’s training and participant assessment

A total of eight CHWs participated in the assessment of study participants. Before data collection, CHWs received a one-day training that included both theoretical and practical sessions. The training covered the clinical features of podoconiosis, application of the clinical assessment algorithm, and use of the WHO Skin-NTD Mobile Application for participant evaluation. Although none of the CHWs had specific previous experience in podoconiosis diagnosis or with the assessment tools used in this study, all had more than five years of experience working as CHWs in community-based health activities.

Participant assessments were conducted by two separate groups of CHWs. One group evaluated participants using the clinical algorithm, while the second group independently assessed participants using the WHO Skin-NTD Mobile Application. The assessments were performed independently to allow comparison of the diagnostic performance of the two CHW-based assessment approaches against the clinical reference standard. Because the study design focused on evaluating the diagnostic performance of the assessment approaches rather than agreement between individual CHWs, formal inter-rater agreement assessment among CHWs was not performed.

2.10 Clinical reference standard

The clinical reference standard was established by a single medical doctor specialized in dermatology, who independently assessed all study participants. The dermatologist applied established clinical diagnostic criteria for podoconiosis, including a detailed clinical history, physical examination, and systematic exclusion of alternative causes of lower-limb lymphedema, such as lymphatic filariasis, chronic venous insufficiency, cardiac failure, renal disease, traumatic or post-surgical lymphedema, and other relevant dermatological or systemic conditions. Because all participants were evaluated by the same dermatologist, no inter-observer disagreements occurred and, therefore, no adjudication process was required.

2.11 Data analysis plan

Data on age, sex and the results of diagnosis of podoconiosis using mobile App or clinical algorithm were analyzed. All cases were re-examined by a clinical dermatologist for confirmation (clinical reference standard) which was used because no definitive laboratory or point-of-care diagnostic test exists for podoconiosis, making expert clinical assessment based on established diagnostic criteria and exclusion of other causes of lymphedema the accepted reference standard in endemic settings. Sensitivity, specificity and positive predictive value of the mobile App and clinical algorithm were calculated with 95% confidence intervals and compared.

As an exploratory analysis, we performed paired comparisons of the diagnostic classifications generated by the mobile application and the clinical algorithm using McNemar’s test. These analyses were undertaken to provide supplementary information on the agreement and discordance between the two methods because both tests were applied to the same participants. The study was not powered to formally compare the diagnostic performance of the two tests or to demonstrate equivalence; therefore, these analyses should be considered exploratory and interpreted with caution.

3 Results

Although the study focused on detection, participants included individuals across all stages of podoconiosis, as staging was not used as a criterion for inclusion or subgroup analysis. The distribution of individuals with leg swelling across different sectors in Rubavu district is presented in Table 1; Figure 3. A total of 308 individuals with lower-limb swelling were identified through health facility records, outpatient registers, and CHW’s listings. Of these, 216 provided informed consent and were enrolled in the study. Sixteen participants were subsequently excluded because a definitive diagnosis by a dermatologist (the reference standard) was unavailable, resulting in a study population of 200 participants. An additional four participants lacked both a diagnosis from the WHO Skin-NTD mobile application and the clinical algorithm, leaving 196 participants for the final diagnostic accuracy analyses.

Table 1. Recruitment proportions by sector in Rubavu district.

Although the final sample size was slightly smaller than originally planned, resulting in a minor reduction in precision, the estimated confidence interval width remained 0.09. This level of precision is still considered acceptable and is consistent with the study’s predefined criteria for clinically meaningful diagnostic accuracy estimates. The sector with the highest number of patients was Busasamana, with 75 cases, followed by Bugeshi with 58 cases. Other sectors such as Rugerero, Nyamyumba, and Rubavu had 30, 25, and 26 cases respectively. The Kanama sector had the fewest cases, with 19 patients, while in Cyanzarwe sector, seven individuals with leg swelling were identified in the sampling frame; however, none were recruited into the study due to unavailability during the data collection period. Overall, the data highlight the variation in leg swelling cases across sectors.

Of the 216 participants, 200 were included in the analysis. Among those assessed, 137 (68.5%) were diagnosed with podoconiosis, while 63 (31.5%) had other causes of leg swelling. The mean age of participants was 54.9 years (SD = 15.3), with those diagnosed with podoconiosis having higher mean age (mean age 56.9 years, SD = 16.0) compared to those without podoconiosis lower mean age (mean age 50.7 years, SD = 12.6). Regarding sex distribution, 50 (25.0%) were male and 150 (75.0%) female, with a higher proportion of females among those diagnosed with podoconiosis (86.9%) compared to those not (13.1%).

Education levels varied, with a majority of participants lacking formal education (70.5%), especially among those diagnosed with podoconiosis (81.8%). Most participants identified as Christian (95.0%), and the distribution of marital status showed a notable proportion of widowers in the podoconiosis group (38.7%) compared to the non-podoconiosis group (15.9%).

Over one third (37.5%) of participants were unemployed, with a higher unemployment rate among those with podoconiosis (42.3%). The majority of participants (86.0%) had outpatient medical insurance, although insurance coverage was slightly higher among those without podoconiosis (90.5%). A significant difference was observed in family history of podoconiosis, with (41.6%) of those diagnosed reporting a positive family history compared to (7.9%) with other conditions (Table 2).

Table 2. Socio-demographic characteristics of the respondents.

Among the 200 participants with full data, the majority (83.0%) reported consistent shoe-wearing outside the home, with 88.9% of those without podoconiosis and 80.3% of those diagnosed with podoconiosis indicating they wore shoes every day. Nearly all participants (98.5%) reported washing their feet, with 93.5% washing them daily regardless of diagnosis status (Table 3). Regarding footwear use during specific activities, most participants used their shoes when walking (95.0%), at home (90.5%), and while going to market (93.0%). However, a small proportion (16.8%) did not consistently wear shoes when farming or when walking outside the home (5.1%). The majority had been wearing shoes for more than 20 years, with no significant difference between groups. The frequency of foot washing was high across groups, and no significant differences were observed in foot hygiene practices related to podoconiosis diagnosis. Overall, protective behaviors such as consistent shoe-wearing and foot washing were prevalent among participants, regardless of diagnosis status.

Table 3. Health and behavioral characteristics of participants.

3.1 Comparative diagnostic accuracy of mobile app and clinical algorithm

The mobile application and the clinical algorithm produced different, but overlapping, proportion classified positive estimates. The mobile application estimated a proportion classified positive of 66.3% (95% confidence interval: 59.2%–72.9%), while the clinical algorithm estimated a proportion classified positive of 71.4% (95% confidence interval: 64.6%–76.2%) (Table 4). The sensitivity of the Mobile App was 89.8% (83.4%–94.3%), slightly lower than that of the Clinical Algorithm 96.4% (91.7%–98.8%), indicating that both were effective at correctly identifying positive cases, with the clinical approach being marginally more sensitive. Specificity measures are also comparable, with the Mobile App at 88.1% (77.1%–95.1%) and the Clinical Algorithm at 86.4% (75.0%–94.0%), demonstrating good ability to correctly exclude negatives. The overall diagnostic accuracy, assessed by the Sensitivity + Specificity/2, was high for both: 0.89 (0.84–0.94) for the Mobile App and 0.91 (0.87–0.96) for the clinical algorithm, indicating excellent discrimination between positive and negative cases. Both methods also exhibited high positive predictive values—94.6% (89.2%–97.8%) for the Mobile App and 94.3% (89.1%–97.5%) for the clinical algorithm—reflecting a strong likelihood that positive test results are true positives. The negative predictive value was higher for the clinical algorithm at 91.1% (80.4%–97.0%) compared to 78.8% (67.0%–87.9%) for the Mobile App, suggesting that the clinical approach is slightly, though not significantly, more reliable in ruling out disease when results are negative (Table 4). Overall, both diagnostic tools performed well, with the clinical algorithm showing a slight advantage in sensitivity and negative predictive value, and both demonstrating comparable accuracy across other metrics.

Table 4. Diagnostic performance of mobile app and clinical algorithm approaches for podoconiosis detection: sensitivity and specificity relative to dermatologist diagnosis.

Pr(A) is the true proportion classified positive of Podoconiosis diagnosis as defined by dermatologist diagnosis.

Pr(B) is the proportion classified positive of Podoconiosis diagnosis as defined by Mobile App diagnosis.

Pr(C) is the proportion classified positive of Podoconiosis diagnosis as defined by Clinical Algorithm diagnosis.

4 Discussion

This study presents a comprehensive comparison of two promising diagnostic approaches for detection of podoconiosis by CHWs in Rwanda: the WHO Skin-NTD mobile application and a validated clinical algorithm. The findings demonstrate that both tools exhibit high diagnostic accuracy, with slight differences in performance metrics, and highlight the potential for integrating digital health solutions with conventional clinical assessments to improve case detection, which is critical in resource-limited endemic settings.

The sensitivity of the AI mobile app was 89.8%, while the clinical algorithm achieved a higher sensitivity of 96.4%. Sensitivity is vital in screening contexts because it reflects the ability to correctly identify true positive cases. The slightly lower sensitivity of the AI tool suggests that, although effective, it may miss some podoconiosis cases (Stages 1 and 2), which may be a concern considering the importance of detection to prevent disease progression and irreversible disability. Conversely, the high sensitivity of the clinical algorithm emphasizes its reliability as a frontline tool.

In terms of specificity, both tools demonstrated good performance, with the mobile app at 88.1% and the clinical algorithm at 86.4%. Specificity is crucial for ruling out non-cases and preventing unnecessary treatment or referrals. These comparable values suggest that both methods are capable of excluding individuals without podoconiosis reliably. High specificity reduces false positives, which minimizes the burden on health systems and ensures that resources are directed towards genuine cases.

Both methods achieved high overall diagnostic accuracy, with Sensitivity + Specificity/2of 0.89 and 0.91, respectively. These values indicate excellent discrimination between cases and non-cases, reinforcing their potential utility in field-based integrated screening. High positive predictive values further support their reliability, implying that positive assessments are highly likely to be true cases. Notably, the clinical algorithm had a higher negative predictive value (91.1%) than the AI app (78.8%), suggesting that the clinical approach is more reliable in confidently ruling out non-cases. This is particularly relevant for resource-limited settings where unnecessary referrals could overwhelm healthcare facilities.

The clinical algorithm, developed from expert consensus and validated previously, relies on physical examination and patient history, making it contextually adaptable, cost-effective, and feasible in low-resource settings (4, 20, 21). Its high sensitivity and negative predictive value demonstrate robustness in community screening. However, it depends on clear training of CHWs, which, if limited, might affect the consistency of diagnosis.

The WHO Skin-NTD offers standardized guidance, data recording, and potential for remote consultation. Its performance, comparable to the clinical algorithm, affirms that AI-based tools can effectively support disease detection (18). Nevertheless, challenges such as dependence on mobile device availability, and technical support in remote areas remain significant hurdles. Additionally, digital literacy among CHWs and sustainability of technological infrastructure are crucial considerations.

Other advantages of the WHO Skin-NTD mobile application include standardized decision support and digital data capture, with the potential for integration into surveillance systems where adequate technological infrastructure exists (18, 22). Evidence from digital health implementation studies suggests that such tools may strengthen community-level diagnostic support in appropriate settings (24, 25).

Given that both tools demonstrated comparable diagnostic performance, their use in public health practice should be guided by context-specific resource availability. The clinical algorithm, which requires no digital infrastructure, is particularly suitable for low-resource and remote settings and can be implemented widely with minimal cost, consistent with its original development and validation in endemic field conditions (4, 20, 21).

Rather than applying both tools concurrently to the same population, a flexible, context-dependent approach is more appropriate. For example, the mobile application may be deployed in areas with functional digital infrastructure to support screening, reporting, and data aggregation, while the clinical algorithm may be prioritized in settings where digital tools are not feasible or sustainable. Either tool may also be used for supportive supervision, training reinforcement, or validation of community-level screening outputs (24, 25). This approach aligns with broader digital health integration strategies in low-resource settings (22, 26) and ensures efficient use of limited resources while maintaining high diagnostic performance. Our study findings underscore the importance of structured and repeated training of CHWs in both digital and conventional diagnostic approaches. Digital tools such as the WHO Skin-NTD application require initial and ongoing training to ensure correct use and interpretation, while clinical algorithms require reinforcement in clinical reasoning and case differentiation skills (24, 25). Supervision and refresher training are critical to maintain diagnostic consistency and minimize inter-observer variability in community-based screening programmes (24, 25). Digital health training interventions have been shown to improve CHW performance and acceptability of mobile diagnostic tools in similar settings (24, 25). A key limitation of this study is that participants were not stratified by podoconiosis stage when evaluating diagnostic performance. Although early disease (Stages 1 and 2) is the primary target for prevention of progression, individuals across all stages were included in the analysis. However, given small numbers of advanced stage cases, it was not possible to determine whether the WHO Skin-NTD mobile application and the clinical algorithm perform differently in early versus advanced disease. This limits interpretation of their effectiveness specifically for detection. Future studies should incorporate stage-specific analyses to better evaluate the ability of these tools to identify early, potentially reversible cases.

An additional limitation is that participants were recruited from health facility records, outpatient registers, and CHWs listings of individuals already presenting with lower-limb swelling, rather than through population-based community screening. This recruitment strategy may have introduced selection and spectrum bias by increasing disease proportion classified positive and preferentially including more clinically apparent cases, which could have influenced the estimated diagnostic performance of both tools and may limit the generalizability of our findings to broader community screening populations.

Early detection of podoconiosis remains an important component of disease control because it enables timely interventions, including limb hygiene, wound management, and behavioural change communication, which may help prevent disease progression and reduce morbidity and disability. Community-based screening using effective diagnostic tools may also improve case identification and provide epidemiological information that can support public health planning.

Although the present study did not evaluate health system outcomes, the findings suggest that the WHO Skin-NTD application could be considered for integration into community-based podoconiosis screening programmes in Rwanda. If implemented alongside existing clinical algorithms, digital tools may improve diagnostic consistency, particularly in underserved and rural settings where specialist expertise is limited. They may also support CHWs by providing standardized decision support and facilitating digital data collection.

Scaling up the use of AI diagnostic applications, however, requires several operational and logistical considerations to be addressed. These include ensuring devices are available and functional in remote settings, establishing robust data management and privacy protocols, and providing ongoing technical support and maintenance. Integration with existing health information systems is essential for seamless data sharing and program coordination.

Further implementation research is needed to determine whether integration with national surveillance systems, automated reporting, and real-time monitoring is feasible and beneficial within the Rwandan health system. Similarly, studies evaluating long-term sustainability, user acceptability, and cost-effectiveness are required before conclusions can be drawn regarding large-scale implementation or contributions to national podoconiosis elimination efforts. Stakeholder engagement, including community leaders, local health authorities, and policymakers, will also be essential to assess feasibility and support successful implementation.

Further research should focus on longitudinal studies assessing the impact of digital tools on disease elimination outcomes, monitoring false-positive and false-negative rates over time, and exploring the integration of more advanced AI algorithms, such as image recognition from photographs or videos. Incorporating machine learning techniques could enhance diagnostic precision and adapt to regional variations in disease presentation. Developing multilingual and culturally sensitive interfaces will also promote usability across diverse populations.

In conclusion, both the WHO Skin-NTD mobile application and the clinical algorithm demonstrated high and comparable diagnostic accuracy for podoconiosis in community settings in Rwanda. These findings support their potential role in strengthening case detection by CHWs. However, their implementation should be guided by context-specific health system capacity rather than simultaneous use in all settings. The clinical algorithm is particularly suitable for low-resource and offline environments due to its simplicity and minimal infrastructure requirements, while the mobile application offers additional advantages in standardized decision support, digital data capture, and potential integration into surveillance systems where digital infrastructure is available. A flexible, context-adapted deployment strategy rather than concurrent application of both tools may therefore represent the most efficient and sustainable approach to improving podoconiosis detection and supporting elimination efforts in endemic settings.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Hugues C. Nana Djeunga, Translational Research and Development Foundation (TREND Foundation), Cameroon

Reviewed by: G. Takop Nchanji, TroDDIVaT Initiative, Cameroon

Dr Sunny Doodu Mante, African Filariasis Morbidity Project, Ghana

Mateus Henrique Dias Guimarães, International Epidemiological Association (IEA), United States

Portia Mira, University Medical Center Phoenix, United States