Original Coverage & Source Attribution: bioengineer.org
Antimicrobial resistance continues to pose a serious threat to global public health, and African healthcare systems carry a disproportionate share of the burden. In the WHO African Region, bacterial antimicrobial resistance was associated with an estimated 1.05 million deaths in 2019, and in Western sub-Saharan Africa alone it was directly responsible for 27.3 deaths per 100,000 population in the same year. Recent reports show that sub-Saharan Africa records some of the highest mortality rates linked to resistant infections globally. Without effective interventions, resistance-related mortality is projected to increase substantially by 2050. A new commentary published in Discover Public Health argues that artificial intelligence could help the continent respond, but it also delivers a sober assessment of how far that promise currently extends in practice.
The commentary, authored by Suzan Fikayo Animashaun, examines the opportunities, barriers and recommendations for integrating artificial intelligence into antimicrobial stewardship programmes across Africa. Antimicrobial stewardship, as defined by the World Health Organization, comprises coordinated interventions aimed at improving antimicrobial prescribing and treatment outcomes. Such programmes have been recommended as an important strategy for slowing the emergence of resistance, yet most countries in sub-Saharan Africa still lack national stewardship programmes. Weak regulatory frameworks have contributed to widespread antimicrobial misuse in both healthcare and agricultural settings, and progress has been slowed by shortages of infectious disease specialists, microbiologists, pharmacists and trained stewardship personnel. Delayed diagnosis, empirical treatment and weak surveillance systems continue to worsen the situation.
Surveillance gaps illustrate the scale of the problem. Participation in the WHO Global Antimicrobial Resistance and Use Surveillance System is limited in the WHO African Region, with only 27 of 47 countries, or 57.4 percent, reporting antimicrobial resistance data in 2023. This has contributed to weak surveillance systems and poor availability of resistance data across much of the region. The commentary notes a growing need for solutions that can function effectively in settings with limited specialist workforce capacity and weak health information systems. Artificial intelligence and machine learning have shown promising potential in this respect, through clinical decision support, resistance prediction, automated surveillance, and data extraction from imperfect health information systems.
However, the current status of artificial intelligence in African antimicrobial stewardship is best described as emerging rather than operational. Evidence of artificial intelligence systems in routine use for stewardship specifically remains limited across the continent, and where primary studies exist they have generally focused on retrospective validation or proof of concept rather than prospective evaluation of interventions. One of the clearest examples is machine learning based resistance prediction. Nsubuga and colleagues in 2024 evaluated the generalizability of models developed using whole-genome sequencing data from England by validating them on an independent dataset from Uganda, Nigeria and Tanzania, assessing resistance in Escherichia coli to ciprofloxacin, ampicillin and cefotaxime. In the African validation dataset, logistic regression achieved 94 percent accuracy for predicting ampicillin resistance, while the best-performing models for ciprofloxacin and cefotaxime reached only 50 percent and 45 percent respectively. The dataset contained up to 170 samples and was affected by substantial class imbalance, underscoring that models built on high-income data may not generalise consistently to African populations.
Emerging African research has also explored machine learning resistance prediction using E. coli and Salmonella isolates from countries including Kenya, Nigeria and South Africa, demonstrating the technical feasibility of locally grounded prediction, although such studies remain limited compared with Europe and North America. A 2025 review by Ayesiga and colleagues examined the potential of artificial intelligence and machine learning for resistance biosurveillance in sub-Saharan Africa, including approaches that could analyse clinical, microbiological, epidemiological and genomic information to detect, track and predict resistant organisms. That review highlighted the potential of these tools but identified data scarcity, infrastructural limitations and ethical concerns as important barriers. The available evidence, the commentary stresses, demonstrates potential applicability rather than widespread routine deployment of artificial intelligence-enabled surveillance across African healthcare facilities.
Related work in African infectious disease diagnostics shows that artificial intelligence-supported clinical technologies can be developed and evaluated within African health systems, although these applications should be distinguished from direct stewardship interventions. Computer-aided detection technologies have been evaluated for tuberculosis screening using chest radiographs in Zambia, South Africa and Kenya, and automated approaches have been investigated for malaria diagnosis. In Kenya, a national tuberculosis prevalence survey found that computer-aided detection achieved 95 percent sensitivity and 83.3 percent specificity across nearly 62,000 chest radiographs, while a separate evaluation in Lusaka, Zambia, reported an area under the curve of 0.81 for a comparable system. These tools do not themselves constitute artificial intelligence-enabled stewardship and should not be presented as evidence that artificial intelligence has improved antimicrobial prescribing, but they may provide useful implementation experience for future applications.
The most compelling demonstrations of clinical impact come from high-income settings and are presented as benchmarks rather than evidence of African performance. In the United States, Sepsis Watch, a deep-learning early-warning model, has been in continuous clinical use across the Duke University Health System since November 2018, and its external validation at Summa Health in Ohio demonstrated robust performance across four emergency departments, with AUROC values ranging from 0.906 to 0.960. In an Israel-based study, Yelin and colleagues reported in 2019, using a decade of linked community health-fund data, that machine learning algorithms reduced inappropriate treatment for urinary tract infections by 42 percent compared with physician-prescribed therapy. In a UK-based randomised study, an artificial intelligence clinical decision support system for intravenous-to-oral antibiotic switching significantly influenced clinician behaviour, with clinicians receiving a recommendation against switching markedly less likely to proceed, with an odds ratio of 0.13. A systematic review and meta-analysis of 80 studies reported that machine learning algorithms achieved a pooled accuracy of 74.97 percent and sensitivity of 76.89 percent across stewardship settings, though this evidence is drawn predominantly from outside Africa and should be interpreted cautiously when extrapolated.
Several barriers restrict large-scale implementation in Africa. Resistance data are often scattered across laboratories, hospitals and surveillance programmes, collected with different testing methods, frequently incomplete, and too short in coverage to train models reliably. Electronic health record systems remain limited in many facilities, and some institutions still rely on paper-based documentation unsuitable for data integration. African pathogen genomic data are underrepresented in global sequence repositories, and nearly 97 percent of bacterial genome assemblies from the East African Community are reportedly analysed outside the continent. Digital infrastructure is another constraint: a 2018 survey of bioinformatics researchers in Tanzania found that 96.4 percent relied on personal computers for analyses, with only about 10 percent having access to institutional high-performance computing. Workforce shortages, ongoing brain drain, underdeveloped regulatory frameworks, unresolved questions of algorithmic bias and transparency, scarce external validation in independent African cohorts, limited healthcare budgets, dependence on donor support, and the need for periodic model retraining all compound the challenge, leaving many projects stalled at pilot stage.
The commentary proposes a set of recommendations for responsible implementation. Investments should focus on interoperable data infrastructure, with standardised antimicrobial resistance and use platforms guided by FAIR principles of findability, accessibility, interoperability and reusability, alongside electronic health record and laboratory information systems that can securely integrate microbiological, prescribing and patient data. Expanding participation in GLASS and improving surveillance reporting quality should remain priorities. Machine learning models should be developed and validated using locally representative clinical and microbiological datasets, and pilot programmes with phased implementation can evaluate performance and feasibility before wider deployment. Incorporating bias assessments and explainable artificial intelligence methods may improve transparency and clinician confidence, while decision support systems should deliver clear, actionable recommendations that preserve human oversight and clinical accountability. Training programmes, collaboration among clinicians, pharmacists, microbiologists, data scientists and policymakers, and coordination by bodies such as Africa CDC could support adoption, and policymakers are urged to allocate dedicated budget lines for surveillance digitisation, mandate minimum interoperability standards for publicly procured laboratory systems, and negotiate pooled licensing arrangements to lower per-facility costs.
The overall picture is one of genuine opportunity tempered by an evidence gap. Prospective clinical evidence that artificial intelligence-enabled stewardship interventions actually improve antimicrobial prescribing, patient outcomes or resistance patterns in African healthcare settings remains scarce, and claims that such tools are already improving prescribing across the continent would be premature. Realising the potential will require further development and validation of locally representative models, prospective clinical evaluation, implementation research, and assessment of clinical and economic outcomes before widespread adoption can be justified. The commentary concludes that artificial intelligence should strengthen and complement existing stewardship programmes and national antimicrobial resistance action plans rather than operate separately from them, and that coordinated efforts, adequate investment and long-term commitment will be essential to translating encouraging early findings into routine clinical benefit.
Subject of Research: The potential and current limitations of artificial intelligence for antimicrobial stewardship in Africa
Article Title: Advancing antimicrobial stewardship in Africa through responsible artificial intelligence implementation
Article References: Animashaun, S. F. (2026). Advancing antimicrobial stewardship in Africa through responsible artificial intelligence implementation. Discover Public Health, 23(1), Article 1712. https://doi.org/10.1186/s12982-026-03022-6
Keywords: antimicrobial resistance, antimicrobial stewardship, artificial intelligence, machine learning, Africa, public health, surveillance, clinical decision support, AMR prediction, digital health, genomic surveillance, health systems
