As physicians, we are taught to diagnose accurately and act quickly. Clinical guidelines assume that if we identify the problem, the health care system will provide the necessary tools to solve it. But what happens when the diagnosis is clear and the treatment simply does not exist?
Recently, while working as an intern physician in South Sudan, I cared for a 9-year-old boy who had traveled from the outskirts of Juba with high-grade fever, shortness of breath, lethargy, and profound pallor. His mother reported a history of a blood transfusion three years earlier and a chronic productive cough.
On examination, he was tachycardic and febrile but maintaining adequate oxygen saturation. A complete blood count, the only laboratory investigation immediately available, revealed a hemoglobin level of 4.1 g/dL, with microcytic indices and mild leukocytosis. The result left little doubt. Whatever the underlying cause, the child had life-threatening severe anemia and required urgent blood transfusion.
I requested a cross-match. Then the reality of practicing medicine in a resource-limited setting became painfully clear. His father had traveled to Uganda. His mother was ill and could not donate blood. The hospital had no blood bank with readily available units.
Real physician voices, twice a week
Free, and one click to unsubscribe.
The medical decision was straightforward. The system could not deliver it.
This experience has stayed with me, not only as a clinician but also as someone deeply interested in the future of artificial intelligence in medicine. Much of today’s discussion about clinical AI focuses on diagnostic accuracy. Can a model detect pneumonia from a chest X-ray? Can it identify leukemia from laboratory data? Can it outperform physicians on board-style examinations? These are important questions.
But in many parts of the world, another question may be even more important: What should AI recommend when the ideal intervention cannot be implemented?
Imagine an AI system reviewing this child’s case. It correctly identifies severe anemia and advises immediate transfusion. From a technical perspective, it has succeeded. From the perspective of the clinician standing at the bedside, however, the recommendation is incomplete if there is no blood available.
Health care is not practiced in a vacuum. Decisions are shaped by infrastructure, supply chains, staffing, geography, and economics. An AI designed only for resource-rich hospitals may produce recommendations that are evidence-based yet impossible to execute elsewhere.
This is why I believe the next frontier in clinical AI is not only intelligence but context awareness. A truly useful clinical decision support system should understand the environment in which it operates. It should adapt recommendations to local realities, acknowledge constraints, prioritize feasible options, and help clinicians navigate difficult situations when ideal care is unattainable.
That requires more than better algorithms. It requires listening to the physicians who practice every day in settings where scarcity is routine.
As global conversations about AI governance continue, frontline clinicians from resource-limited health systems deserve a place at the table. Our experiences reveal challenges that may never appear in benchmark datasets but profoundly influence patient care.
The future of clinical AI should not be built solely around what medicine looks like in the best-equipped hospitals. It should also reflect what medicine looks like when the blood bank is empty.
The patient scenario described here has been anonymized and certain details have been modified to protect privacy while preserving the educational and policy implications of the case.
Buga Charles George Kenyi is a physician in South Sudan.
