From the newsletter
The Africa CDC in partnership with the African Society for Laboratory Medicine and the European Union have launched a four-year surveillance initiative to generate drug resistance data across the continent. Fackswell Mateyo, a pharmacist, explains how AI can use this data to prescribe antibiotics and fight drug resistance.
“Investment in local microbiology laboratory capacity and structured resistance surveillance is not optional infrastructure. It is the prerequisite without which AI-guided prescribing cannot deliver accurate guidance in Africa.” Says Mateyo.
He adds that AI-guided prescribing could transform how antibiotics are used. It can make treatment more precise, improve patient outcomes, reduce resistance and support data-driven decisions in healthcare.
More details
By Fackswell Mateyo, an antimicrobial pharmacist from Zimbabwe.
Imagine this, the antibiotics you prescribe being guided not just by your training or intuition but by a machine that has analysed thousands of antibiotic treatment regimens and resistance patterns specific to your patient’s history, your local pathogen profile and your institution prescribing profile. This is no longer a hypothetical.
Across Israel, the United States, Germany, and beyond, hospitals are deploying AI-driven clinical decision support systems to guide empiric antibiotic selection with measurable reductions in treatment failure, inappropriate prescribing and resistance-driven drug switches. For Africa, the continent carrying the world’s highest antimicrobial resistance (AMR) mortality burden, the question is no longer whether AI-guided prescribing is relevant, but whether we can afford to be left behind.
AI-guided antibiotic selection is simply using data to predict the most likely effective antibiotic before laboratory results are available. These systems analyse patient characteristics such as age, history and comorbidities, local resistance patterns, hospital antibiograms, previous infections and treatments, and clinical presentation. From this, AI generates treatment recommendations or probabilities, not decisions.
The most extensively documented live deployment comes from Maccabi Healthcare Services (MHS), Israel’s second-largest Health Maintenance Organisation, serving over 2.6 million members. Maccabi deployed the ‘UTI Smart-Set’ (UTIS), a machine learning model that generates personalised antibiotic recommendations for urinary tract infections, integrating AI-predicted resistance patterns, patient clinical history and institutional prescribing guidelines. The results were compelling.
Following deployment, clinicians recorded a 35 per cent reduction in the need to switch antibiotics due to bacterial resistance. A more granular analysis found a 31.8 per cent reduction in measured antibiotic resistance among patients who followed UTIS recommendations, alongside a 47 per cent cumulative improvement in adherence to clinical guidelines. At Stanford, AI-guided selection achieved an infection coverage rate of 85.9 per cent, comparable to the clinician benchmark of 84.3 per cent.
At Boston, the AI achieved 90.4 per cent coverage against a clinician benchmark of 88.1 per cent. In addition, the KINBIOTICS project in three German hospitals, funded by the German Federal Ministry of Health, piloted an AI-based clinical decision support system specifically for antibiotic therapy in sepsis patients in intensive care, which is the highest-stakes application of AI prescribing guidance, where the wrong antibiotic choice can be fatal within hours.
In Africa, empiric prescribing is most common because microbiological diagnostics remain inaccessible at the point of care. While WHO recommends clinical bacteriology services at the primary hospital level, in practice these are limited to tertiary facilities and are poorly integrated with clinical workflows. The risks are clear, including wrong antibiotic choice, overuse of broad-spectrum agents, treatment failure, delayed effective treatment and increased resistance pressure.
AI models are only as good as the data they learn from, meaning good quality data equals good quality output. Here lies Africa’s most fundamental challenge: only 11 of 44 sub-Saharan African countries had National Action Plans on AMR, with just 32 per cent performing routine AMR surveillance on clinical pathogens and only 2 per cent on veterinary pathogens.
The implication is clear: investment in local microbiology laboratory capacity and structured resistance surveillance is not optional infrastructure, but the prerequisite without which AI-guided prescribing cannot deliver accurate guidance in African contexts.
Despite the promise, successfully integrating AI into Africa’s health systems faces formidable obstacles, including limited human capacity and AI literacy among clinical and pharmacy workforces, widespread public and professional distrust of automated clinical guidance, insufficient funding for health technology adoption beyond donor-driven pilots, fragmented data infrastructure and lack of interoperability between health systems, and weak regulatory and policy enforcement frameworks for both AI tools and antibiotic dispensing.
Moving forward, strengthening microbiology laboratory capacity at district hospitals and systematically linking that data to central repositories is the prerequisite investment before AI can deliver locally accurate guidance, meaning capital investment in diagnostic infrastructure, not just software deployment.
Given the infrastructure realities across most of sub-Saharan Africa, AI stewardship tools built for the continent must work on smartphones, function with intermittent connectivity and require no specialist hardware.
Clinician and pharmacist buy-in also requires that AI recommendations come with clear, accessible reasoning, because a prescriber who cannot understand why the AI system is recommending amoxicillin over ciprofloxacin will override the recommendation or ignore the tool entirely.