Abstract
Artificial intelligence (AI) is transforming healthcare and can reduce inequalities through optimized resource allocation, timely diagnosis, and personalized care. However, over 3.5 billion people still lack access to basic health services, primarily in low- and middle-income countries (LMICs) due to infrastructural, socioeconomic, and cultural barriers. This article examines the opportunities and limitations of AI in advancing global health equity, focusing on region-specific challenges. It presents a narrative review and case-based synthesis of AI deployment in LMICs, using examples from India, Peru, Senegal, Kenya and Cambodia. Key implementation factors include infrastructure, workforce capacity, digital literacy, data availability, resource constraints, and cultural fit. A tiered AI readiness matrix is introduced to compare preparedness and support policymakers in prioritizing AI applications. Ethical risks such as algorithmic bias, data sovereignty and transparency are also emphasized. Successful AI deployment requires localized adaptation, participatory design, and supportive governance. This study provides a strategic framework for policymakers, developers, and global health actors who aim to implement AI equitably and effectively.
Keywords
1. Introduction
Global health equity (GHE) means the absence of unfair and avoidable differences in health outcomes.
1
Unlike equality, which implies identical treatment, equity recognizes that different populations require different resources and interventions to achieve comparable health outcomes. GHE is achieved when everyone, regardless of social, economic, or geographic circumstances, has a fair opportunity to achieve their full health potential. However, more than 3.5 billion people worldwide lack access to essential health services.1,2 These health inequities are particularly severe in rural areas and low- and middle-income countries (LMICs). In regions already suffering from systemic deficiencies (from understaffed clinics to inadequate infrastructure) these inequities are exacerbated during crises such as epidemics, natural disasters, and armed conflicts.3,4 (Box 1)
• • • *This article focuses on equity, recognizing that equal treatment does not necessarily produce equal outcomes.Box 1. Key terminology regarding global health equity
Artificial intelligence (AI) is being used for predictive diagnostics, workflow automation, and telemedicine, but most solutions are developed in high-income settings and often ignore the economic, infrastructural, and sociocultural realities of LMICs
Examples include a widely used US risk model that underestimates severity in Black patients by attributing “need” to prior use, allocation tools adjusted for majority population data, and pulse oximetry error masking hypoxemia in darker-skinned patients; similar risks also arise in LMICs where connectivity and workforce constraints limit data quality and model portability.5–8
Sociocultural norms also shape access. In situations where female clinicians are scarce, gender-matched teleconsultation prevents women from being referred to undesirable providers and can increase participation when platforms are designed with local norms in mind.3,9,10
This review synthesizes challenges and case studies from India, Peru, Senegal, Kenya, and Cambodia and argues that achieving global health equity with AI depends on localized data strategies, inclusive policies, and equitable governance. LMICs were selected as the primary analytical focus because they bear a disproportionate burden of health inequalities while also facing greater constraints in digital infrastructure, workforce capacity, data systems, and regulatory oversight. High-income countries were not included as a comparison group because the aim is not to benchmark AI maturity globally, but to examine how AI readiness can be assessed in environments where implementation barriers are most likely to impact equity. This focus allows the matrix to address the practical decision-making needs of policymakers and practitioners working in resource-constrained health systems.
This study follows a narrative review design combined with a case-based synthesis approach. Rather than applying systematic review methods, we purposively selected representative case studies from LMICs based on relevance, availability of published evidence, and diversity of implementation contexts. The aim was to integrate information from heterogeneous sources to identify common barriers, favorable conditions, and practical strategies for equitable AI deployment. This approach prioritizes contextual understanding and applicability rather than a comprehensive literature review. The equity challenge addressed here is not only unequal access to AI-powered tools, but also the risk that AI systems trained, managed, or deployed without local adaptation will reproduce existing care gaps. Current equity challenges include rural-urban divides, gender-based access restrictions, language exclusion, poor digital literacy, and underrepresentation of underserved populations in health datasets. Therefore, the matrix is framed as a pre-implementation equity tool rather than a general technology readiness index.
2. An overview of AI in healthcare
AI is reshaping healthcare by improving clinical decision-making, enhancing diagnostic accuracy, and relieving the burden on overburdened healthcare systems. This section provides a technical overview of AI applications in healthcare, focusing on specific technologies, their mechanisms of action, and their functional capabilities. While Section 1 addressed equity issues in AI deployment, here we examine the technical foundations that enable AI’s opportunities and limitations in various healthcare settings.
Through their ability to analyze complex datasets, AI models, particularly machine learning (ML) and deep learning (DL), can identify patterns, support real-time diagnoses, and personalize care delivery.11,12 These capabilities offer significant potential for increasing access to quality care in regions experiencing healthcare workforce shortages and resource constraints.13,14 Recent reviews also map the regulatory pathways and preventive/personalized care use cases that frame these technologies in practice. 15
2.1. Technical capabilities and applications of AI in healthcare
The application of AI has had a significant impact in many medical fields, such as radiology, pathology, and public health, by reducing human error, optimizing treatment times, and providing cost-effective solutions.7,13,16 AI-enabled telehealth platforms allow remote diagnosis and consultation, reducing the need for physical travel, which can play a critical role in addressing health issues associated with specific geographic locations and income levels. In India, Qure. ai’s chest X-ray halved the time to TB diagnosis and improved triage in rural clinics without on-site radiologists. 17 The combined evidence further highlights how implementation choices intersect with regulatory and prevention agendas, which is particularly important for the deployment contexts of LMICs. 18
Remote monitoring and telehealth reduce the burden of travel for chronic care and are particularly valuable in countries with dispersed populations and rural majorities.7,19,20 AI also streamlines operations such as scheduling and EHR management and can support capacity where specialists are scarce by reducing reliance on urban referral centers.21–23 (Box 2)
• • • • C • Box 2. Illustrative AI-enabled applications relevant to resource-constrained settings
While AI applications in healthcare encompass diagnosis, triage, workflow optimization, and health system management, their effectiveness ultimately depends on the quality, interoperability, and representativeness of the data they are trained on. These technical considerations are discussed in Section 2.2, while the broader ethical and equity implications are discussed in Section 4.
2.2. Technical limitations and implementation challenges
Although AI has demonstrated significant potential in clinical and administrative settings, its use is constrained by technical limitations. These include limited interoperability among digital health systems, inadequate data standardization, model fragility when faced with non-distributional inputs, and high computational costs that limit scalability in low-resource environments. Additional sections of this review (see Section 4) discuss how such technical limitations intersect with ethical and equity concerns; the focus here is limited to the engineering and implementation aspects that determine the reliability, accuracy, and generalizability of an AI system.
2.2.1. Computational and infrastructure requirements
Many advanced AI models, particularly deep learning systems, require significant computational resources such as graphics processing units (GPUs) and high-performance computing infrastructure that may not be readily available in rural clinics or district hospitals in LMICs.13,16 While cloud-based AI systems reduce local hardware requirements, they depend on reliable, high-bandwidth internet connectivity, which is often unavailable or intermittent in underserved areas.2,8 This connectivity dependency creates a fundamental tension between the most robust AI architectures and the realities of low-resource settings. 7
2.2.2. Data requirements and quality issues
AI algorithms require large, high-quality, structured, and representative datasets to achieve effective results. 16 In many regions, especially in LMICs, health data is often insufficient, incomplete, or poorly structured, making it difficult to train reliable models.24,25 Even when data are available, they may lack the standardization necessary for AI training; they may include inconsistent coding systems, missing values, and incompatible data formats across facilities.14,21
2.2.3. Model fragility and generalization failures
AI systems trained on specific hardware, imaging equipment, or population characteristics may fail when used in environments with varying characteristics.5,26 For example, a radiology AI model trained on high-resolution images from top-of-the-line equipment may underperform when applied to images from older, low-resolution machines common in resource-poor environments.22,23 This equipment sensitivity creates significant deployment barriers. 27
2.2.4. Interoperability and integration challenges
Legacy health information systems in many LMICs lack the application programming interfaces (APIs), data standards (such as HL7 FHIR), or technical documentation needed for AI integration. 14 Retrofitting AI systems to work with diverse and often outdated health IT infrastructures requires significant customization and technical expertise that may not be available locally.8,26 Even in settings with electronic health records (EHR), data warehouses and incompatible systems hinder the seamless integration required for effective AI deployment. 21
2.2.5. Model maintenance and degradation
AI model performance can degrade over time due to shifts in patient populations, changes in clinical practices, or the evolution of disease patterns.16,28 Maintaining model accuracy requires ongoing monitoring, periodic retraining with up-to-date data, and ongoing technical support; these resources may be scarce in environments where even basic IT maintenance is challenging.24,29 Without adequate maintenance infrastructure, AI systems that initially perform well may become unreliable or even harmful over time. 30
2.2.6. Cultural and linguistic adaptation
Beyond purely technical considerations, AI interfaces require adaptation to local languages, health literacy levels, and cultural contexts.7,8 Voice-based AI systems must adapt to different accents and dialects; text-based systems require translation that preserves medical meaning; and visual interfaces must be understandable to users with diverse educational backgrounds. 2 The technical challenge of building truly multilingual and culturally responsive AI systems is substantial and often underestimated.31,32
These technical limitations do not preclude the use of AI in resource-constrained environments, but they do require careful system design, realistic expectation setting, and investment in supporting infrastructure.13,14 Section 3 examines how these technical limitations interact with broader structural barriers to shape the feasibility of applying AI in different contexts.
3. Structural barriers in regions with health disparities
In healthcare systems with persistent shortcomings (inadequate infrastructure, workforce gaps, fragmented data, and weak institutions) AI deployment must contend with the same structural barriers that limit essential services.3,24 Workforce shortages are particularly significant: regions with a high disease burden often have the lowest clinician density, delaying diagnosis and treatment, thus increasing the appeal of AI tools operating with limited expert capacity.
33
Figure 1 highlight the stark contrast in health worker density between high- and low-income countries. Comparison of the density of health professions (dentists, doctors, midwives, nurses, pharmacists) in high- and low-income countries. The density graph (a) and map (b) display the number of health professionals per 10.000 population, in 2020. Graph A has been redrawn and Map B has been reprinted from ref. Boniol et al., 2022, an open access article distributed in accordance with the Creative Commons Attribution Non-Commercial (CC BY-NC 4.0) license.
The GBD 2019 Human Resources for Health Collaborators study revealed that the global disease burden super-regions of Sub-Saharan Africa, South Asia, North Africa, and the Middle East had the lowest densities of health professionals. According to this study, it was estimated that in order to achieve a score of 80 out of 100 on the Universal Health Coverage Effective Index, each population of 10.000 would require at least 20.7 physicians, 70.6 nurses and midwives, 8.2 dental personnel and 9.4 pharmacy personnel. Despite a steady increase in the number of health professionals globally from 1990 to 2019, significant disparities persist among all professions, both within and across regions. 33
The digital divide also significantly shapes the applicability of AI, even within the same region. For example, in South and Southeast Asia, Cambodia and Myanmar suffer from low digital literacy and internet connectivity, while neighboring Singapore has relatively higher digital literacy and technology use. 8 Therefore, the AI solution for such a region should be calibrated to operate offline or in hybrid modes, be usable by non-specialist healthcare workers, and accommodate limited data access.
Structural inequities are barriers to healthcare, but they are also barriers to AI itself. Effective, equitable AI deployment requires both technical readiness and a culturally sensitive and resource-aware approach that is adaptable to the complexities of real-world environments in LMICs. The barriers to AI deployment that form the basis of the following sections are discussed under three main headings.
3.1. Health infrastructure and technological barriers
Infrastructure shortage is a major challenge in the integration of AI into healthcare. Many rural hospitals lack reliable electricity, internet access and functional medical equipment, making the adaptation of traditional AI tools unsuitable. 7 For example, AI-powered tools have been used for early diagnosis of tuberculosis in remote clinics in Peru, while drones have been successfully adapted and implemented for medical delivery in Senegal.23,34
Another major challenge limiting the integration of AI into healthcare is the data barrier. AI algorithms require high-quality, structured, and representative datasets to deliver effective results. However, in many regions, especially in LMICs, health data are often inadequate, making training reliable models difficult.24,25 Local data integration is essential in the adoption of AI in healthcare, as models trained on foreign datasets may lead to errors in diagnosis or treatment recommendations due to differences between populations.
3.2. Cultural, gender and linguistic barriers
A severe shortage of female clinicians in parts of Afghanistan and Pakistan limits women’s access to care; gender-matched telemedicine is one practical mitigation.9,10 AI-powered telemedical platforms that match female patients with female clinicians may serve as a solution in such settings. Chatbots and voice assistants that are not linguistically localized may fail, particularly in rural settings, for both patients and non-specialist healthcare professionals.2,8
3.3. Institutional and regulatory challenges
Fragmented institutions, bureaucratic delays, and incomplete AI regulations hinder its effective and safe implementation, particularly in settings with weak health systems. 29 Sustainable use depends on clear governance, stakeholder engagement, and cross-sectoral collaboration.14,35 AI can only alleviate bottlenecks when aligned with local policy priorities and supported by institutional buy-in.
To guide suitability assessments, we propose a four-domain AI readiness matrix (infrastructure, digital literacy, data availability, and cultural fit risk) that scores 1–3 from nationally reported indicators synthesized from recent evidence.4,7,8,24 The matrix supports transparent cross-country comparison and prioritization of pilot sites and resources (Figure 2). Comparative AI readiness matrix assessing AI deployment readiness in selected countries representing diverse geographic and economic contexts. Countries were selected based on published evidence on AI health initiatives and represent different World Bank income classifications and geographic regions. 1 = Low readiness/high challenge, 2 = Moderate readiness/moderate challenge, 3 = High readiness/low challenge. Infrastructure level: Presence of electricity, internet, medical devices, and basic health facilities; Digital literacy: The ability of health professionals and community members to use digital tools; Data availability: Existence of structured, reliable and accessible health data; Cultural risk: Potential sociocultural incompatibilities (language, gender roles, technology trust etc.).
3.3.1. Data sources
Each dimension of the AI readiness matrix was operationalized using publicly available, nationally representative indicators. Infrastructure utilized national electrification and internet connectivity statistics (e.g., World Bank, ITU).36,37 Digital literacy captured the digital skills of the population and health workers using recent national surveys and sector reports (e.g., UNESCO, WHO).35,38 Data availability captured the existence and maturity of electronic health records and national data platforms/governance using WHO Global Observatory materials and Ministry of Health documents. Cultural adaptation risk synthesized evidence on documented gender-based access restrictions, language diversity, and trust/acceptance of digital health from peer-reviewed country assessments and reputable international reports. When multiple sources were available, we prioritized the most recent nationally representative data and used the same source type across countries whenever possible to promote comparability.
3.3.2. Operationalization
To enhance consistency and reduce subjectivity, scoring thresholds were aligned with indicative quantitative ranges where data were available. For example, infrastructure scores were determined by electrification rates (e.g., <70%, 70-90%, >90%) and internet penetration levels; digital literacy was approximated using population-level digital skills or proxy indicators (e.g., <40%, 40-70%, >70%); data availability reflected the degree of adoption of electronic health records or national health data systems (e.g., limited/no coverage, partial implementation, widespread use); and cultural fit risk was assessed against documented barriers such as language fragmentation, gender-based access restrictions, and trust in digital health systems. While precise thresholds varied slightly depending on data availability, consistent types of indicators were used across countries to support comparability.
Comparative readiness matrix assessing four critical dimensions in selected countries.
High readiness scores should not be interpreted as evidence that equitable AI deployment has already been achieved. Rather, they indicate that selected supporting conditions, such as electricity, connectivity, digital skills, and data infrastructure, are more favorable. For example, urban India might score highly in several domains because of stronger digital infrastructure and specialist availability, yet important inequalities remain across informal settlements, low-income groups, and state-level systems. Therefore, the matrix determines relative readiness for implementation, not guaranteed fairness or successful outcomes.
While these countries do not comprehensively represent all LMICs context, they provide an empirical basis for understanding how different readiness profiles shape implementation feasibility.7,39 Due to these limitations, this matrix should be interpreted with several caveats. First, the scores represent national-level estimates and inevitably mask significant variation within the country, particularly urban-rural divides.2,8 For example, India scores high (3) for infrastructure in urban areas but low (1) in many rural areas, reflecting documented inequalities in digital infrastructure across Indian states. 17 Second, the three-point scale necessarily simplifies complex, multidimensional realities; intermediate positions exist, but they are difficult to capture without excessive complexity. This methodological limitation is common in comparative health systems assessment tools. 14 Third, data availability varies significantly across countries; some assessments are based on robust national surveys, while others rely on regional estimates or smaller studies.29,39 Fourth, the rapid pace of technological and infrastructural change means that these assessments must be periodically updated to maintain relevance in decision-making processes.13,21
4. Ethical and policy considerations
AI in healthcare raises questions that go beyond model accuracy: who benefits, who is exposed to risk, and who is accountable? In environments where trust in health systems is weak, choices around data source, interface design, and governance can exacerbate rather than reduce inequalities. This section brings together four domains (bias and equity, data governance, transparency, and accountability) with an emphasis on the realities of LMICs.
4.1. Algorithmic bias and equity
Algorithmic bias arises when training data, labels, features, or distribution contexts encode structural inequalities, leading to unequal performance across subgroups defined by gender, race/ethnicity, language, socioeconomic status, or geography. Well-documented examples include a widely used population health risk model that underestimates disease severity in black patients by comparing need with prior use, and pandemic-era allocation tools calibrated to majority population data.5,6 In LMICs, data scarcity and portability gaps exacerbate the risk: Models trained on high-end devices or narrow populations may fail when applied to older equipment, different disease prevalence rates, or different clinical workflows.7,8,26 Therefore, equity by design requires representative data or open-field adaptation, disaggregated performance reporting by distinct subgroups, and context-aware deployment (e.g., gender-matched teleconsultation where norms restrict access). Without these safeguards, AI may divert scarce resources to already best-served populations.
4.2. Data governance, privacy and sovereignty
The development of AI depends on large, interconnected datasets, while human protection depends on due process, confidentiality, consent, and control over cross-border data flows.32,40 Data protection capacity varies significantly among LMICs. While some upper-middle-income countries, such as Brazil and South Africa, have GDPR-compliant protections and supervisory authorities, others are still drafting legislation or lack the resources to implement it.14,29,35 This heterogeneity means that general statements about LMICs are misleading: protections range from robust to minimal.
Operational risks increase when cloud-based or vendor-managed AI moves data beyond national control, creating uncertainty about sovereignty, benefit-sharing, and redress in the event of violations or misuse. Where regulators are nascent, power asymmetries allow external actors to extract data value without providing sufficient local benefits. Practical measures are feasible and commensurate with capacity: purpose-limited data use and explicit consent options; privacy-by-design measures such as de-identification, access controls, and audit trails; local copies or residency for sensitive datasets; gradual standards compliance to ensure secure interoperability (e.g., HL7 FHIR); and public records of AI systems used in maintenance. These measures support innovation while protecting rights and strengthening institutional trust.
4.3. Explainability and transparency
Trust and accountability require interpretable outputs, traceable data pedigrees, and documented performance, particularly where literacy barriers, multilingual contexts, and workforce shortages hinder informed use.16,26,31 Transparency is best addressed as a suite of practical measures: model cards and data sheets that explain training data, intended use, and known limitations; clinician-focused justifications appropriate to the task (e.g., key features or counterfactuals); patient-focused explanations in local languages and plain language; post-commissioning monitoring with subgroup performance dashboards; and independent audits for safety-critical applications. In LMICs, pragmatic explainability emphasizes actionable signals, such as why a triage alert was triggered, rather than full algorithmic introspection combined with workforce training and co-designed interfaces. Recent syntheses of multimodal AI for clinical decision support emphasize that explainability and auditability are prerequisites for safe adoption. 18
4.4. Accountability and liability
Clinical AI disperses responsibility among developers, suppliers, data managers, institutions, and clinicians, raising unresolved questions about who is responsible for harm and under what jurisdiction.28,30 Some regulators now treat certain AI tools as medical devices subject to pre-market evaluation and post-market surveillance, while professional organizations are exploring how standards of care should evolve when clinicians use decision support. 29 However, in many settings, legislation is lacking, or enforcement capacity is weak, creating uncertainty that can deter adoption or shift risk to under-resourced providers.
A workable approach combines several elements. Risk stratification provides stronger evidence and oversight for safety-critical functions such as diagnosis and triage. Responsibilities should be clearly delineated: developers are responsible for design, data source, and performance claims; institutions are responsible for validation and secure implementation; and clinicians are responsible for contextual use. Event logging and audit logs should link AI outputs to clinician actions and outcomes to enable research and learning. Local validation should be conducted on representative data and equipment before deployment, and reassessment should occur as populations or workflows change. Government purchasers should negotiate contracts that prevent the transfer of overall responsibility to providers and guarantee ongoing support, updates, and model maintenance. Clearer rules reduce uncertainty, ensure responsible adoption, and protect patients who otherwise have limited recourse.
5. Solutions for equitable AI deployment
Digital health innovations, when aligned with local realities, can address access and capacity constraints. In LMICs, the most effective levers include: 1) Task-focused AI tools that operate with intermittent connectivity and limited hardware; 2) Telemedicine platforms that extend scarce specialist expertise; and 3) Logistics and supply chain applications, such as drones, for time-critical deliveries. Taken together, these approaches target different layers of vulnerability in the system (clinical, informational, and infrastructural) and, when adapted to local conditions, can strengthen service coverage.17,34,41
In government-supported tuberculosis programs, Qure. ai algorithms shortened time to diagnosis and reduced radiology backlogs, demonstrating that context-specific AI can augment, rather than replace, existing pathways. 17 Similarly, drone-based vaccine and blood transport networks have shortened delivery times in remote areas, while low-bandwidth remote monitoring applications have enabled non-specialist workers to manage chronic diseases remotely.17,23
Technology should adapt to sociocultural preferences; gender-matched teleconsultation models can increase service uptake where women prefer female providers. Incorporating such cultural parameters into AI planning or triage systems can prevent unintentional exclusion. Local data strategies (increased electronic health record coverage, data standardization, and privacy policies by design) are prerequisites for reliable and replicable AI.2,24 Parallel workforce skills development initiatives and user-centered interfaces (multilingual, low-literacy compatible) increase acceptance and performance in resource-limited environments.8,32
Equitable scaling requires stronger local ownership of data assets and governance. Governments can establish national data protection standards, ethical review frameworks for algorithmic use, and public-private partnerships that mandate capacity transfer. Investments in interoperable digital infrastructure (cloud-edge hybrids, local data centers, and open-source standards) reduce reliance on single vendors and protect data sovereignty.14,35
The AI readiness matrix provides a structured way to link observed gaps to targeted interventions. Low infrastructure scores indicate priorities such as reliable electricity, broadband, and device supply; limited digital literacy highlights workforce training needs. Poor data availability points to lack of registries and standardized data collection tools, while a high risk of cultural fit requires participatory design and gender-balanced teams.
Summarized representative AI application case studies results by regions of inequality.
5.4. Integrating AI into broader health system strengthening
Key barriers and suggested solutions for communities facing disparities when implementing AI in healthcare.
Where possible, barriers are expressed using measurable or proxy indicators to improve comparability and reduce subjectivity; for example, workforce constraints may be interpreted using doctor or nurse density per 10,000 population, while connectivity limitations may be assessed using national internet penetration rates.
5.5. Summary
AI can augment scarce expertise, streamline workflows, and narrow geographic and socioeconomic gaps. However, this may happen only when deployment strategies respect local capacity, culture, and regulations. The readiness matrix and related solutions provide a practical roadmap for prioritizing equitable and sustainable AI implementation in healthcare.
6. Strengths and limitation of this study
This study integrates technical, institutional, ethical, and sociocultural perspectives into a single preparatory framework and bases its analysis on published case materials from various low- and middle-income country contexts. However, it is a narrative review and case-based synthesis rather than a systematic review. Country selection is purposive and dependent on the availability of published evidence, which may lead to a bias toward environments with stronger research visibility. Additionally, the preparatory matrix reduces complex facts to national-level scores and may obscure sub-national and urban-rural differences. These limitations mean that the framework should be interpreted not as a definitive preparatory measure, but as a practical heuristic method for comparative assessment.
7. Conclusion
Equitable AI deployment requires more than just technical readiness; it relies on contextual fit, transparent governance, and sustainable institutional capacity. The AI Readiness Matrix proposed here can be used by governments, donors, and health system planners as a pre-development screening tool to determine whether the baseline conditions for AI deployment are sufficiently developed and which gaps require investment prior to pilot launch. In practice, the matrix can guide site selection, prioritization of infrastructure and workforce investments, and the design of locally acceptable deployment models. Future work should validate the matrix against real-world implementation outcomes, refine indicators using more standardized cross-country data, and test its applicability at sub-national levels where urban-rural disparities are often significant. Through iterative refinement and field-based validation, the matrix can evolve into a more robust decision-support framework for equitable AI adoption in diverse low- and middle-income country settings.
Footnotes
Author Contributions
RD and NA designed the study, drafted the manuscript, and supervised and edited it. NA wrote the manuscript. Both authors read and approved the final version.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interest
The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: RD is the president of the INVAMED Institute for Medical Innovation. NA is a volunteer consultant for Med-International UK Health Agency Ltd.
Data Availability Statement
All supporting data is included in the article.