Natural Language Processing Technologies for Public Health in Africa: Scoping Review

Songbo Hu, Abigail Oppong, Ebele Mogo, Charlotte Collins, Giulia Occhini, Anna Barford, Anna KorhonenJournal of Medical Internet Research 2025 · DOI

TL;DR

This scoping review maps the current landscape of Natural Language Processing (NLP) technologies for public health in Africa, revealing significant potential but also substantial gaps in language inclusivity, deployment, and impact evaluation. The findings highlight the urgent need for more context-aware, culturally relevant, and linguistically diverse NLP solutions to truly benefit African health systems and communities.

What Problem Does This Paper Address?

Africa faces immense challenges in achieving good health and well-being, exacerbated by a shortage of healthcare professionals, weak surveillance systems, and limited resources. While Artificial Intelligence (AI), particularly NLP, offers promising avenues to bolster these systems, its application in African health contexts is hindered by the continent's vast linguistic diversity and a scarcity of relevant digital data. Existing research often focuses on high-resource languages and lacks a comprehensive understanding of how NLP can be tailored to address Africa's unique public health needs and existing infrastructure. This paper aims to bridge this gap by systematically reviewing the current state of NLP technologies for public health in Africa, identifying challenges, opportunities, and areas for future development.

Methods

This study employed a scoping review methodology, adhering to the PRISMA-ScR guidelines. A systematic search was conducted across five major academic databases (MEDLINE via PubMed, ACL Anthology, Scopus, IEEE Xplore, and ACM Digital Library) and supplemented with gray literature searches. The review focused on studies published between January 1, 2013, and October 3, 2024, to capture recent advancements in NLP, particularly those following the introduction of neural language models. Inclusion criteria focused on NLP technologies promoting public health with relevance to Africa, while exclusion criteria filtered out non-NLP applications, non-African contexts, and studies lacking full-text availability. Data extraction involved characterizing NLP technologies, their contributions to Essential Public Health Functions (EPHFs) and Sustainable Development Goals (SDGs), deployment stages, accessibility, and reported impacts. A narrative synthesis was used to analyze the findings due to the heterogeneity of study designs and outcomes.

Datasets

The paper does not introduce new datasets. Instead, it analyzes studies that utilized various datasets for training and evaluating NLP technologies. These datasets primarily came from:

  • General domain data: Social media, general text.
  • Clinical domain data: Clinical notes, patient interactions.
  • Research domain data: Research articles, professional materials.

The modality of data used was predominantly text-based (98% of studies), with limited exploration of audio or image data.

Key Insights and Ideas

  • The "Digital Divide" in Language: The most striking insight is the profound lack of linguistic inclusivity. While English, Arabic, and French dominate, the vast majority of Africa's over 2000 languages remain unsupported, creating a significant barrier to equitable access and utilization of NLP technologies.
  • Focus on Technical Metrics Over Impact: A major disconnect exists between the technical performance of NLP systems and their actual impact on public health outcomes. The overwhelming majority of studies report technical metrics (accuracy, precision, etc.), with very few attempting to measure real-world health improvements or behavioral changes.
  • Nascent Deployment Stage: Most NLP applications are stuck in the "design and prototyping" phase. Only one fully deployed chatbot addressing vaccine hesitancy was identified, highlighting a significant gap between research and real-world implementation.
  • Expert-Centric Design: Current NLP technologies are largely designed for expert users (researchers, policymakers, healthcare providers), with limited focus on public-facing applications that could empower general communities to manage their health.
  • Cross-Sectoral Collaboration is Key: The success of commercial and NGO-led initiatives, often partnering with telecom providers, underscores the importance of cross-sectoral collaboration to bridge the gap between academic research and practical, scalable public health solutions.

Key Results

  • 54 Studies Included: After screening 2186 citations, 54 studies were included in the review.
  • Language Coverage: Only 59 languages are supported across the 54 studies, a fraction of Africa's linguistic diversity. Widely spoken languages like Kiswahili and Zulu have limited support, and most indigenous languages are entirely absent.
  • Geographic Concentration: South Africa is the primary target country (46% of studies), followed by Kenya (26%) and Nigeria (17%), indicating an uneven distribution of NLP technology development.
  • EPHF Coverage: Existing NLP technologies address 9 out of 12 Essential Public Health Functions (EPHFs), with EPHF 3 (stewardship), EPHF 8 (community engagement), and EPHF 12 (access to health products) remaining unaddressed.
  • SDG Contribution: All reviewed technologies contribute to SDG 3 (good health and well-being), with some also supporting SDG 10 (reduced inequality) and SDG 9 (industry, innovation, and infrastructure).
  • Deployment Status: Only 1 (2%) study reported a fully deployed system (a chatbot for vaccine hesitancy). 88% are in the design/prototyping phase, and 10% are in the validation stage.
  • Accessibility: Over half (58%) of NLP applications are publicly accessible, with 24% being open-source. However, many require significant technical expertise.
  • Impact Evaluation: Only 15% of studies reported health-related evaluations, and a mere 4% demonstrated positive public health outcomes (e.g., improved mood, increased vaccine intentions).

Limitations and Future Work

Limitations:

  • English Language Bias: The search strategy primarily used English terms, potentially excluding studies published in other languages without English annotations.
  • Database Coverage: Reliance on specific databases might have missed relevant studies published in other venues, especially in the highly interdisciplinary field of NLP.
  • Gray Literature Scope: The gray literature search was not exhaustive, potentially missing some commercial products or NGO initiatives.
  • Heterogeneity: The diverse methodologies, applications, and evaluation metrics made a formal meta-analysis impossible, necessitating a narrative synthesis.

Future Work and Recommendations:

  • Expand Language Coverage: Prioritize the development of NLP technologies for underrepresented African languages, leveraging multilingual NLP and cross-lingual transfer learning.
  • Contextualize Solutions: Design NLP applications that are culturally relevant and embedded within local health systems and practices.
  • Focus on Health Outcomes: Shift evaluation from technical metrics to measurable health impacts and behavioral changes, utilizing frameworks like the WHO's "Monitoring and Evaluating Digital Health Interventions."
  • Enhance Deployment and Integration: Move technologies from prototype to operational use by ensuring accessibility, usability, and seamless integration into existing health infrastructures.
  • Promote Participatory Design: Engage local communities and domain experts throughout the development process to ensure technologies meet actual needs and are accepted.
  • Strengthen Cross-Sectoral Collaboration: Foster partnerships between academia, industry, NGOs, and government bodies to accelerate the translation of NLP research into impactful public health solutions.
  • Establish Reporting Standards: Develop standardized protocols for reporting NLP technologies to improve comparability and facilitate future research.

Summary updated

Abstract

RR2-doi:10.1101/2024.07.02.24309815.
BibTeX
@article{hu2025natural,
  author = {Songbo Hu and Abigail Oppong and Ebele Mogo and Charlotte Collins and Giulia Occhini and Anna Barford and Anna Korhonen},
  title = {{Natural Language Processing Technologies for Public Health in Africa: Scoping Review}},
  journal = {Journal of Medical Internet Research},
  year = {2025},
  doi = {10.2196/68720},
}
Figure