Commentary: Artificial intelligence and low-income health systems

Abstract

Commentary:  Artificial intelligence and low-income health systems

 Mutemwa R^1,2

 Affiliations:

¹School of PostGraduate Studies, University of Lusaka, Lusaka, Zambia

²School of Medicine and Health Sciences, University of Lusaka, Lusaka, Zambia

Corresponding Author:

 Name: Prof. Richard Mutemwa

Email: rmutemwa@unilus.ac.zm

Artificial Intelligence (AI) is upon us; basically because, for now, indications are that it is never going away, instead it is reaching out and evolving rapidly. Systems across the globe are scrambling resources to re-engineer themselves to accommodate and best exploit the apparently positive promise of AI. By all accounts, the dominant narrative appears to be that AI is not here to just enhance our 'lifestyle', but rather radically transform it. A recent article by Bill Gates, on the impact of AI on aspects of routine human functioning over the next 5 years, predicts a future that bears little resemblance to the present. Yet, while some AI predictions may be generic, it is equally reasonable to imagine AI-impact that will vary by context (subject to local conditions), need, and technological advancement. In that light, low- and middle-income countries (LMICs) will likely be impacted differently compared to developed economies. Thus, LMICs will be better prepared by beginning to envision their AI-dominated futures and develop useful decision-support models of those scenarios. One particular sector to prioritise in these considerations should be the health sector.

 As LMICs embark on such future-scenario modelling, a preliminary caution is perhaps imperative. That, AI should not be conceptualized as a technology to replace the traditional model in which a health system is sufficiently resourced to provide health care with adequate human resources, efficient medical supply chains, effective health information systems, and a robust health management and governance system. A progressive view is one that abhors competition between AI and the traditional health system model, and instead fosters collaboration between them. That strengthens existing health institutions and services.¹

It is necessary that each LMIC health system addresses the compound question: Can AI be used to resolve some of the challenges that perennially plague our health systems? How can we best leverage AI technology? Well, in some cases AI may help alleviate or, indeed, eliminate operational 'defects' and inefficiencies in the system, while in others the technology may fill the deficiency gap in the interim while the health system seeks to boost its resource base. AI could also enable health systems to introduce emergent medical practices (e.g., bespoke treatments) which would have taken not less than half-a-century to introduce in resource-poor contexts. AI does promise immediate benefit to some public health areas; the obvious few may be highlighted here.

First, AI-use may help moderate the ever-increasing cost of managing health care systems in these resource-poor contexts.² Many of these contexts carry high fertility rates and rapidly increasing populations mean that health systems struggle to keep up with disease-burden, growing demand for health care, and exponentially-increasing cost of providing care. AI may help lower costs through optimization of resource allocation and improved operational efficiency. 

Second, AI-use could infuse efficiency into procurement and logistics systems for medicines and medical products in the health system. Procurement delays and maldistribution of secured drugs and medical supplies are common challenges in low-income health systems. The underlying problem is almost always reliance on inefficient significantly manual procurement and logistics management systems, typically supported by less rigorous computing resources. On the other hand, AI is able to utilize more complex machine learning algorithms and greater variety of procurement, distribution and commodity utilization data to realise just-in-time procurement, distribution, and inventory management systems for medicines and medical supplies.

Third, AI could be used to mitigate the impact of low-staffing at different levels of the health system. This is more so for health systems struggling with poor health-worker retention in rural areas. Some tasks may be shifted to AI; for instance, syndromic diagnosis of disease in the absence of specialists.^3,5 There is growing evidence that AI certainly helps in health screening and early primary level diagnostics for chronic illnesses such as cancer, cardiovascular conditions, diabetes mellitus.⁶ In large complex health systems, AI promises efficiency in staff-workload and workflow analysis;^5,6 as well as staff-posting allocations to ensure equitable distribution of human resources for health across the health system.

Fourth, AI may improve the management of health management information systems (HMIS): collection of routine health information, updating the HMIS, reporting. Health management information systems are an essential feature of health systems across the Globe. However, LMIC health information systems are typically characterised by: under-staffing, poor recording, deficient information technology infrastructure, 'fragmented' data sources, and defective management. These information systems are predominantly manual and lack the capacity to perform integrated analyses of data from various sources to inform decisions. Thus, LMIC information systems tend to be defective, inefficient, and unreliable. On the other hand, AI, with its ability to analyze large quantities of data, can be used to more usefully inform public health policy decisions; as well as improve clinical case management.³      

Lastly, AI can enhance health system capacity to surveil infectious diseases, forecast their spread and guide development of public health response strategies, through predictive modeling and disease forecasting.³ AI is already being used in the control of infectious diseases⁴, including the management of seasonal or infectious disease outbreaks (such as cholera or COVID-19). Leveraging AI's ability to manage huge and diverse data sources, the future spread and impact of disease can be predicted more efficiently and accurately.

Yet, for all these potential benefits accruable from AI in a health system, some pre-requisites to adoption of AI are worth critical consideration:

1.  Quality of data: AI is a data-based technology that thrives on good quality data. An ample cadre of well-trained local computer and information scientists is the primary strategy for a functional AI-system. In the long-term, LMIC health systems, with their weak resource bases, cannot sustainably rely on AI and data expertise from the developed north. The sustainability of local AI-use significantly relies on a well-developed and robust local capacity to operate and support local AI-use.

2.  Localisation of AI systems: Successful AI technology in local health systems will depend primarily on AI systems being built on local data - that is, learning or training data collected from the context in which the AI is being installed. Implanting pre-trained AI systems developed in other settings into beneficiary country health systems would be strongly ill-advised. The modern public health has learned the hard way, since inception of primary health care (PHC) philosophy in 1979, that local knowledge and participation are critical to success of any public health intervention. Overall, fidelity to the original basic principles of PHC must be maintained. It is important that AI technology be implemented within the framework of PHC, rather than parallel to it.

Although computing infrastructure may be imported into specific LMIC settings, local health systems should "locally                   train" the imported AI computing system, using locally collected data and local computing expertise with the ability to perceive local nuance. 'Localisation' is the most appropriate strategy for AI adoption. AI for public health in a particular health system, is NOT necessarily infinitely compatible with other national health systems across the globe.

Most crucially perhaps, localisation of AI training data and algorithms ensures local ownership and control of the        technology - protecting LMIC health systems from the whims of powerful profit-oriented corporations which presently own Large Language Models (LLMs) of AI and AI development ecologies in the Global North.¹ Finally, localisation is likely to realise local democratisation of the technology, with the equitable participation of developers, purchasers, users, and beneficiaries - the health system clients.    

3.   Compassion, empathy & ethics:  AI should not be introduced deliberately to supplant the human health-worker or undermine the fundamental values of health services. AI cannot replace the essential human qualities that are central to the process of caregiving and healing.¹ Human health care professionals must bear the ultimate responsibility for making crucial decisions to the maximum benefit of health care recipients. AI may augment and enhance the capabilities of health care providers but not completely replace the human responsibility where human resources are available.¹ For any patient, healing is a complex and spiritual process that extends beyond the biochemical physics of the body receiving care, to somewhat tap into 'benefits' from compassion, empathy, and ethical considerations.³ That respect envisages a public health world where AI serves the greater good and prioritises the well-being of all humanity. ^1-3  

4.  Legal framework: AI brings with it a new tradition to the sector, with the need for revision of labour laws, privacy protection regulations, and technology procurement policies. There is need for an accompanying legal framework to guide the technology procurement, development, adoption and operation. The legal framework should also guide AI-assisted strategic and operational routines within the health system.^1,7

5.  Role of the Central Ministry of Health: The traditional role of the Ministry of Health is to govern and support the health system. If AI must be conceptualized as a 'new helper to the health system', then the cardinal role of the Ministry of Health is, firstly, to ensure that use of the new technology does not violate existing laws of the country, as well as facilitate the process of reconciling legal provisions that may be inimical to effective adoption and exploitation of the new technology.^8,9 Secondly, the Ministry of Health should ensure availability of ample resources to support successful adoption of AI technology in areas of public health that decide and are ready to use it. Thirdly, the Ministry of Health is responsible for both the physical and ethical safety of the population served by the health system, as AI use expands within the system. Local AI adoption cannot be at all cost. Lastly, the Ministry of Health, collaboratively with other Government agencies, should spearhead the development of AI policy to govern technology procurement, use, and local AI-capacity management in the health sector. ^8,9  

AI: Game-Changer or Paradigm-Shift?

Peel off the veil of cautionary sub-narratives about AI and what remains is the profound hope that AI brings to any beneficiary industry, including public health.^1-7 AI holds the tentative promise to 'leapfrog' the world of LMIC-health systems to the frontline thus far the exclusive abode of the developed North. LMIC-health systems may be able to do what hitherto would have been practically impossible. AI has the potential to change the efficiency and effectiveness of health policy, health management, and clinical practice. AI seems set to revolutionize the way health systems are operationalized, managed, and governed. But does that transformation amount to a game-changer or a paradigm shift? It is probably too early to definitively address that question. Game-changer? - potentially yes; Paradigm shift? - perhaps it is a wait-and-see for now.

The author is Professor of Biostatistics & Epidemiology in the School of Postgraduate Studies & School of Medicine and Health Sciences, University of Lusaka, Lusaka, Zambia.

Contributors    The article was conceptualized, implemented, and prepared by the author MR.

Funding     This article received no specific grant from any funding agency in the public, commercial or nonprofit sectors.

Disclaimer  MR affirms that the manuscript is an honest, accurate and transparent commentary on the subject matter, based on professional observation and assessment of literature; that no important aspects of the observation and assessed literature have been omitted; and that any discrepancies have been explained.

Competing interests    None declared.

Provenance and peer review    Not commissioned; externally peer reviewed.

Data availability statement      Data sharing not applicable as no datasets were generated and/or analysed for this article.

ORCID iD

Professor Richard I. Mutemwa:    http://orcid.org/0000-0002-3658-0233.

References

1. Sezgin E. (2023). Artificial intelligence in healthcare: complementing, not replacing doctors and healthcare     providers. Digital Health, July 2;9: 20552076231186520.  


2. Ikhisemojie S. (2024). Artificial intelligence in healthcare. https://punchng.com/artificial-   intelligence-in-healthcare/.


3. Olawade DB et al. (2023). Using artificial intelligence to improve public health: a narrative review. Front Public Health, 11: 1196397.


4. Parija SC & Poddar A. (2024). Artificial intelligence in parasitic disease control: a     paradigm shift in health care. Trop Parasitol, Jan-     Jun; 14(1): 2-7.


5. Barth S. (2025). AI in healthcare. https://www.foreseemed.com/artificial- intelligence-in-healthcare


6. Bajwa J et al. (2021). Artificial intelligence in healthcare: transforming the practice of medicine. Future Healthcare Journal, Jul; 8(2): e188–e194.


7. Hossain A & Akter S. (2023). Artificial intelligence in medicine: a paradigm shift.    JCMCTA, 34(1):1-2.


8. Mbewe A et al. (2026). The role of digital health and artificial intelligence in improving the reach and effectiveness of HIV prevention in Africa. The Lancet Global Health, 14(1): e131-e142.


9. Williams M et al. (2024). Ethical data acquisition for LLMs and AI algorithms in healthcare. NPJ Digital Medicine, Dec 24;7:377.