There is a calculation that haunts me. A neonate born to an HIV-positive mother needs Nevirapine Syrup — 2mg/kg/day — within 72 hours of birth. The medicine works. The science is settled. The protocol is clear.

And yet there was a period when I stood in my pharmacy store — a central supply point for over 40 surrounding health facilities — watching the shelves of Nevirapine 50mg/5ml get thinner by the week, with no resupply in sight.

We rationed first. One bottle per mother. Then the shortage deepened. So we did something I had never done before, and never want to do again: we rebottled. 60ml per child. Sometimes 100ml, calculated in real time against each infant's weight. We decanted life-saving medicine under a fluorescent light because we did not know when — or whether — the next delivery was coming.

Not one child in our network missed their dose. I am proud of that. But pride is not the right emotion. The right emotion is alarm. Because what we did was improvise brilliantly around a system failure — and no system should need that kind of brilliance to function.

 

The improvisation is not the problem. The improvisation is the evidence.

 

THE DATA WAS THERE THE WHOLE TIME

Here is what makes this story complicated: we had data. Kenya's health facilities run on stock cards, consumption logs, DHIS2 entries, monthly reports. The data infrastructure exists. It has existed for years.

But data and insight are not the same thing. No alert flagged when Nevirapine consumption trends began outpacing supply six months prior. No model anticipated the seasonal uptick in deliveries that would lean on us hardest precisely when we had least to give. The system did what it always does: it collected. It reported. It waited.

THE ARCHITECTURE PROBLEM

Take our most monitored commodity — antiretroviral therapy. Today there is genuine progress: national and county teams analyse ART stock data and track months of supply. That improvement is real and should be acknowledged.

But follow that data trail and the problem reveals itself. Figures travel upward — from facility pharmacy teams, through sub-county pharmacists, into county reports, to the national level. In lower-level facilities without posted pharmacists, non-pharmacy cadres capture the data as best they can. By the time numbers reach the top, they are aggregates — summaries of summaries, each layer adding distance from the ground.

And then the data stops. It does not come back down.

The pharmacy in-charge has no visibility into what the neighbouring facility holds. There is no shared view of which sites have three months of an ARV and which have two weeks. No alert says: you are trending toward a stockout — here is what is nearby. The sub-county pharmacist redistributes manually, by phone, by memory, by relationship. The system that collected all that data offers nothing back to the people closest to the patient.

 

We need the capacity to convert data — in real time, at facility level — into decisions: what to order, how much, by when, from where.

 

WHAT A FELLOWSHIP AND A SPREADSHEET TAUGHT ME

I am not formally trained in data science. I am a pharmacist. But a fellowship introduced me to the fundamentals of AI and data science — and it cracked something open. I taught myself Python. I got deeper into Excel — not as a filing tool, but as an analytical engine: pivot tables, dynamic stock visualisations, automated month-of-stock calculations, expiry flagging. I moved records into Google Sheets to interrogate data from any device, anywhere.

The results were immediate. When I could visualise which medicines were trending downward three months out, I could act three months out. When I flagged short-expiry stock at low-demand facilities, I initiated redistribution before wastage — not after. When I saw a particular antibiotic spike every March, I pushed the requisition in December.

These are not revolutionary tools. But used deliberately, analytically — they changed the quality of every decision I make. And I know, with complete clarity, that I am working at the ceiling of what self-teaching can give me. I am drawing trend lines where we need demand forecasting algorithms. I am making educated guesses where we need evidence-based supply chain science.

TO EVERY PHARMACIST STILL RATIONING BY HAND

If you work in a facility where stock cards are manually updated, where consumption reports go up and rarely inform what comes back down, where you have ever improvised a workaround and called it a win — I see you.

The improvisation is not the problem. It is evidence that the people closest to the patient are resourceful and dedicated, carrying far more than any system should ask. It is also evidence that those people deserve better tools.

Data science is not coming to replace clinical judgment. It is coming to carry the weight no single human being should carry alone — pattern recognition across thousands of data points, predictive signals in consumption trends, early warnings before the shelves go bare.

We already have the data. It is time we built something worthy of it.

 

 

 

About the Author

The author is a Pharmacist In Charge at a Kenyan Government sub-county hospital, overseeing pharmaceutical services for a catchment population of hundreds of thousands and coordinating medicine supply across a network of over 40 health facilities. A data science and AI fellowship recipient, she is currently building data-driven approaches to commodity management within Kenya's public health system.

Disclaimer: This article is intended solely for informational and public health advocacy purposes. It does not constitute medical, pharmaceutical, or professional advice of any kind. All procedures described were carried out by a licensed pharmaceutical professional strictly within the scope of regulated practice, under crisis conditions, and in accordance with Kenya Ministry of Health guidelines and institutional protocols. Nothing herein should be interpreted as an instruction to replicate any clinical or pharmaceutical procedure outside of proper professional authority, licensure, and regulatory oversight. The author expressly disclaims any professional, legal, or personal liability arising from the misinterpretation or misapplication of any content herein. Views expressed are solely the author's own and do not represent the position of any employer, government body, or professional association.