Framing the Convergence of One Health and Digital Health in the Global South With a Gender-Sensitive Foresight Perspective: Delphi Study Using Latent Semantic Analysis
Original Paper
- Jude Kong1,2,3,4,5, PhD (https://www.jmir.org/search/searchResult?field%5B%5D=author&criteria%5B%5D=Jude+Kong) ;
- Nicola Luigi Bragazzi1, MPH, MD, PhD (https://www.jmir.org/search/searchResult?field%5B%5D=author&criteria%5B%5D=Nicola%20Luigi+Bragazzi)
1Artificial Intelligence & Mathematical Modeling Lab (AIMMLab), Dalla Lana School of Public Health, University of Toronto, Toronto, ON, Canada
2Institute of Health Policy, Management and Evaluation (IHPME), University of Toronto, Toronto, ON, Canada
3Department of Mathematics, Bahen Centre for Information Technology, University of Toronto, Toronto, ON, Canada
4Africa-Canada Artificial Intelligence and Data Innovation Consortium (ACADIC), Toronto, ON, Canada
5Global South Artificial Intelligence for Pandemic and Epidemic Preparedness and Response Network (AI4PEP), Toronto, ON, Canada
Corresponding Author:
Jude Kong, PhD
Artificial Intelligence & Mathematical Modeling Lab (AIMMLab)
Dalla Lana School of Public Health
University of Toronto
155 College Street
Toronto, ON, M5T 3M7
Canada
Phone: 1 4169783868
Email: jude.kong@utoronto.ca
Abstract
Background: The convergence of digital health and One Health represents an emergent paradigm in global health governance. While widely discussed in high-income settings, there is limited understanding of how this convergence is conceptualized in the Global South, particularly when viewed through a gender- and equity-sensitive foresight lens.
Objective: This study aimed to map and classify expert discourse on digital health, One Health, and their convergence in the Global South using latent semantic analysis, with particular attention to structural drivers, emerging issues, weak signals, and gendered patterns of anticipation.
Methods: A 3-round online Delphi survey was conducted with 45 experts from 19 countries across the Global South. Open-ended responses were analyzed using latent semantic analysis and stratified by gender. A foresight framework was applied to categorize topics as structural drivers, emerging issues, or weak signals, based on their temporal persistence, salience, and consensus.
Results: In digital health, structural drivers included the systemic integration of digital technologies into public health systems, strategic alignment, and infrastructure development. Emerging issues comprised the adoption of artificial intelligence, chronic disease management via mobile health, and concerns about digital inclusion and interoperability. Weak signals included feminist digital ethics, trust in digital systems, and relational accountability—more frequently emphasized by female experts. In One Health, structural drivers were centered on intersectoral coordination, ecological integration, and the institutionalization of health-environment frameworks. Emerging issues encompassed anticipatory risk governance, food system sustainability, and the integration of environmental and population-level data. Weak signals included indigenous knowledge systems, subnational antimicrobial resistance governance, and structural underinvestment in ecological public health, with gendered divergence in framing. In the convergence discourse (digital health and One Health), structural drivers focused on the integration of digital surveillance systems, data infrastructures, and health information platforms to operationalize One Health. Emerging issues included climate-triggered system redesign, artificial intelligence and ecological monitoring, and the governance of cross-sectoral data. Weak signals pointed to algorithmic bias in zoonotic prediction, digital sovereignty in environmental health, and feminist critiques of convergence—all thematically rich but peripheral in consensus.
Conclusions: This study revealed a multilayered and gender-influenced foresight architecture shaping the future of digital health and One Health in the Global South. Structural drivers denote maturing domains of implementation, while emerging issues and weak signals highlight latent, often overlooked opportunities and tensions. Incorporating equity-sensitive and gender-aware foresight methods is essential for crafting inclusive and anticipatory health governance strategies.
J Med Internet Res 2026;28:e78702
doi:10.2196/78702 (https://doi.org/10.2196/78702)
Keywords
Introduction
In an increasingly interconnected and rapidly changing world, the convergence of demographic aging, epidemiological transitions, and climate-related environmental threats has contributed to a multifaceted global health burden, placing immense strain on already fragile health care systems [,]. This burden is especially acute in the Global South, where both communicable and noncommunicable diseases persist amid structural inequalities and constrained resources []. Simultaneously, growing awareness of the complex, nonlinear, and interacting determinants of health has spurred a paradigmatic shift in the governance of health—away from siloed, biomedical models toward more holistic, systemic, and integrative frameworks [].
The One Health paradigm, which acknowledges the interdependence of human, animal, and environmental health, has emerged as a critical response to zoonotic threats, ecosystem degradation, and planetary instability [,]. In parallel, the digital transformation of health—encompassed within the digital health agenda—has accelerated, leveraging data-driven technologies to enhance diagnostics, disease surveillance, and care delivery []. Their convergence, conceptualized as Digital One Health or One Digital Health, reflects an emergent field characterized by interdisciplinary cooperation and technologically mediated multisectoral coordination [,]. However, existing scholarship has largely framed such convergence through a Global North lens, overlooking the digital disparities, infrastructural deficits, and sociotechnological asymmetries that define the Global South []. Specifically, studies conducted in high-income and Global North settings have shown that the convergence of digital health and One Health is primarily operationalized through integrated surveillance systems, advanced health and environmental data infrastructures, and artificial intelligence (AI)–enabled early warning platforms, with reported gains in outbreak detection, antimicrobial resistance (AMR) monitoring, and cross-sectoral coordination. These contributions, while methodologically and technologically robust, are predominantly grounded in contexts characterized by mature digital ecosystems, stable governance arrangements, and well-resourced regulatory frameworks, thereby limiting their transferability to the structural, institutional, and equity-related realities of the Global South [-].
Moreover, a gender-sensitive or feminist lens is often lacking, thereby marginalizing the situated experiences, epistemologies, and agency of women and gender-diverse actors in shaping health and technology ecosystems [].
To address these epistemic and representational gaps, this study used a foresight-oriented text mining approach grounded in horizon scanning and the identification of weak signals, emerging issues, and structural drivers []. The goal was to map and classify the thematic architecture of the discourse on the convergence of One Health and digital health from a Global South perspective, with particular attention to gendered dynamics, equity-informed framings, and the coproduction of inclusive and pluralistic alternative futures.
Methods
Ethical Considerations
The study protocol was approved by the institutional review board of Dalla Lana School of Public Health, University of Toronto, Ontario, Canada (48350). The study was conducted in accordance with the Declaration of Helsinki, and written informed consent was obtained from all participants.
Expert Recruitment
Experts were recruited through a purposive sampling strategy aimed at capturing diverse perspectives on the convergence of digital health and One Health within the Global South. Invitations were disseminated via professional networks, relevant mailing lists, and institutional affiliations, targeting stakeholders actively engaged in health policy, digital innovation, epidemiology, veterinary medicine, environmental science, and global clinical public health practice.
Selection criteria included demonstrated expertise in at least 1 relevant domain, professional affiliation with an institution based in the Global South, and availability to complete multiple rounds of an online Delphi survey. A total of 45 experts completed all survey rounds (n=3) and were included in the final analysis. The Delphi process was conducted anonymously to encourage candid responses and reduce social desirability bias.
Survey Instrument
The study used a structured, multiround Delphi survey designed to elicit expert consensus on the convergence of digital health and One Health in the Global South. The instrument was developed to capture both sociodemographic diversity and domain-specific insights relevant to foresight-oriented analysis. The survey instrument was developed ad hoc for the purposes of this study, as no validated or standardized questionnaire currently exists to capture expert foresight on the convergence of digital health and One Health, particularly from a Global South and gender-sensitive perspective.
Participants first provided background information, including age, gender, country of birth, country/countries of study, current residence, professional role, areas of expertise, and years of experience in digital health and/or One Health. They were also asked to reflect on the interdisciplinary nature of their work environments.
Each round of the Delphi survey included open-ended prompts focused on present and future opportunities, challenges, and disruptors in both digital health and One Health domains. Specific items addressed strategic drivers, structural barriers, project involvement, and anticipated synergies between the two fields. After each round, responses were thematically analyzed, and aggregated feedback was provided to participants in subsequent rounds, allowing for reflection, refinement, and convergence of expert opinion. The final item in each round requested participants to generate an anonymous identification code to enable matching across rounds. The survey was administered via Google Forms, ensuring broad accessibility across diverse geographies. All items were open-ended to enable narrative elaboration and thematic emergence, aligning with the exploratory aims of the foresight exercise.
The full set of survey items is provided in .
Overview of the Methodology: Text Mining and Foresight Exercise
To explore temporal dynamics and strategic orientations in stakeholder discourse on digital health and One Health in the Global South, we used comparative latent semantic analysis (LSA) [,] across different foresight-relevant prompts: (1) present implementation, (2) perceived opportunities, (3) perceived challenges, and (4) future-oriented opportunities or challenges. Each response set was subjected to LSA to extract latent topics, with eigenvalues and variance explained used to assess topic salience.
To interpret LSA outputs through a foresight lens, we used a 3-tiered classification framework widely used in foresight and futures research: structural drivers, emerging issues, and weak signals [], enabling differentiation between deeply embedded system forces, nascent developments gaining stakeholder traction, and low-frequency yet potentially disruptive cues.
Structural drivers were defined as topics that (1) appeared across both present- and future-oriented datasets and (2) accounted for a substantial proportion of explained variance. These themes represent widely shared, resilient priorities. Emerging issues were operationalized as topics observed predominantly or exclusively in future-oriented discourse, with a moderate share of explained variance. These topics signal directional shifts in collective attention and can anticipate inflection points in system trajectories. Weak signals were identified as topics with low variance explained, often surfacing in only 1 temporal frame (present or future) and lacking broader consensus. Although marginal, these signals are analytically significant, as they may indicate overlooked risks, latent needs, or early-stage innovations that could reshape the field.
A gendered lens was applied throughout this classification to assess how male and female respondents may differentially emphasize structural versus emergent priorities. Gender-divergent weak signals, in particular, offer critical insight into situated epistemologies—highlighting perspectives that are often sidelined in dominant discourses.
Overall, this classification supports a foresight-informed interpretation of the LSA results and strengthens the capacity to generate equity-sensitive, gender-responsive scenarios grounded in both continuity and systemic change.
Data Collection and Preprocessing
The dataset comprised the open-ended textual responses provided by the expert panel, compiled into a structured corpus for subsequent text mining. The entries varied in both length and complexity, reflecting the heterogeneous expertise and regional diversity of participants addressing digital health and One Health issues. Prior to computational analysis, the corpus underwent standard preprocessing procedures aimed at improving analytical robustness. Text normalization steps included tokenization, lowercasing, stop-word removal, punctuation stripping, and the elimination of nonalphabetical characters. Stemming and lemmatization were applied to reduce lexical variation while preserving semantic meaning. Importantly, domain-specific terminology was preserved throughout the process to maintain contextual integrity. Preprocessing was implemented using English-language stop word lists and stemming algorithms.
Text Mining Analysis
LSA [,] was used to extract conceptual structures embedded within the textual data and to systematically identify latent themes in stakeholder discourse. As a well-established natural language processing technique, LSA enables the reduction of linguistic dimensionality by analyzing patterns of word cooccurrence, making it particularly suitable for foresight-driven text mining applications. A term-document matrix was constructed using a bag-of-words representation, incorporating a minimum term frequency threshold of 2 and a sparsity cutoff of 0.975. Dimensionality reduction was performed using singular value decomposition, which facilitated the identification of principal semantic axes and coherent topic clusters. Specifically, LSA decomposes the term-document matrix into 3 matrices (U, Σ, VT) via singular value decomposition, where Σ is a diagonal matrix containing the singular values associated with each latent semantic dimension. The squared singular values correspond to the eigenvalues of the latent dimensions and quantify the amount of semantic variance captured by each extracted topic. Eigenvalues were computed and ranked in descending order, without the application of predefined cutoffs or dimensionality reduction thresholds. In this analytical context, eigenvalues represent the relative semantic weight or salience of each latent topic, reflecting the strength and coherence of word cooccurrence patterns across expert responses rather than statistical significance in an inferential sense. Topics associated with higher eigenvalues correspond to dominant and widely shared thematic structures within the discourse, whereas topics associated with lower eigenvalues represent less frequent, more weakly articulated, or emerging semantic patterns. For each topic, the percentage of explained variance was calculated as the ratio between the eigenvalue of that topic and the sum of all retained eigenvalues, multiplied by 100. Explained variance therefore expresses the proportion of total semantic variance in the corpus accounted for by each latent dimension. This measure was used to assess thematic prominence and consensus across responses, rather than to perform hypothesis testing or probabilistic inference. Eigenvalues and explained variance were interpreted within a foresight-oriented analytical framework. Topics with consistently high explained variance across both present- and future-oriented prompts were classified as structural drivers, reflecting stable and deeply embedded priorities. Topics with moderate explained variance that emerged primarily in future-oriented discourse were classified as emerging issues, indicating evolving areas of collective attention. Topics characterized by low explained variance, limited consensus, or temporal specificity were classified as weak signals. Importantly, low explained variance was not interpreted as analytical irrelevance; rather, in line with horizon-scanning and futures methodologies, such topics were considered potentially indicative of early-stage, underexplored, or disruptive dynamics that may gain relevance over time. This interpretation strategy ensured conceptual alignment between the quantitative outputs generated by LSA and the qualitative, anticipatory objectives of the foresight exercise, while enabling transparent interpretation of the eigenvalues and explained variance.
LSA was first conducted on the overall corpus to extract global semantic patterns. Additional stratified analyses were performed based on participant gender. Following decomposition, the resulting topics were qualitatively interpreted and labeled by 2 independent coders (JK and NLB), who reviewed the most salient terms within each topic vector. Intercoder reliability was ensured through an iterative reconciliation process, with final topic labels established via consensus.
All LSA procedures were executed using the XLSTAT software suite (Lumivero), a commercial statistical package supporting advanced text analytics.
Results
Sample
The expert sample had a mean age of 43.1 (SD 10.3) years, reflecting midcareer professionals, and a male-to-female ratio of 2.8:1 (n=33, 73% men and n=12, 27% women). Geographically, the sample spanned 19 countries, predominantly in Africa (n=26, 58% participants), followed by Asia (n=14, 31% participants), Latin America (n=3, 7% participants), and the Middle East and North Africa region (n=2, 4% participants). Among female participants, 50% (6/12) were from Africa, 33.3% (4/12) from Asia, 8.3% (1/12) from Latin America, and 8.3% (1/12) from the Middle East and North Africa region. The distribution of participants by gender and geographic region did not differ significantly (Fisher exact test P=.77), indicating that female participants were not disproportionately concentrated in any single region despite their smaller overall number. Participants had a mean professional experience of 7.3 (SD 5.9) years, and represented diverse roles across academia, policy, investment, and practice, including directors, principal investigators, professors, consultants, and World Health Organization officers. Their professional involvement encompassed a wide array of digital health and One Health initiatives in the Global South, such as AI-driven disease prediction, electronic health records, mobile health (mHealth) applications, epidemic surveillance, and AMR monitoring in wastewater. Notably, many participants contributed to public health training, policy development, and the integration of emerging technologies to address syndemics at the human-animal-environment interface.
Digital Health in the Global South
LSA applied to responses regarding the successful implementation of digital health in the Global South revealed 10 interpretable topics. Topic 1, explaining 46.75% of the variance, reflected structural and systemic priorities—particularly integration, infrastructure, and strategic alignment. Topic 2 (7.57%) captured operational themes around patient management and service delivery, while Topic 3 (4.77%) emphasized stakeholder engagement and community outreach. Topic 4 (4.21%) focused on governance and regulation, complemented by Topic 5 (4.04%) which stressed regulatory clarity and local adaptation. Topic 6 (3.57%) addressed digital literacy and technological access, and Topic 7 (2.99%) emphasized co-design and participatory processes. Topics 8-10, collectively explaining another 6.91%, covered scalability and sustainability (2.63%), capacity building (2.21%), and context-aware implementation frameworks (2.07%), respectively. Adopting a gender lens, common themes included strategic alignment (male Topic 7: 2.68%; female Topic 6: 1.56%), capacity building (male Topic 5: 3.90%; female Topic 3: 9.22%), and patient-centered approaches (male Topic 2: 8.53%; female Topic 9: 0.83%), though to a different degree. Both groups also addressed community and infrastructure support, with the theme being more prominent in women—Topic 1 (52.95
