skip to content

Department of Computer Science and Technology

What opportunities are there for AI methods to contribute to healthcare in Ethiopia?

Although health record systems are not a primary focus of my work in Bahir Dar, one goal of the project here was to explore application of AI to maternal health. A specific opportunity came about through chance meeting with the dynamic duo of Australian midwives Atlanta and Elisa ( who are doing amazing work during a 6 week volunteer residency at the Vision Maternity Care centre in Bahir Dar.

They suggested I should meet the enthusiastic "IT guy" Minichil Azene, who runs the network, installs the servers, administers the database, trains the users, and (it turns out) even customises the health record software. He is a computer science graduate from Bahir Dar university, who has been trained under a CDC-funded initiative from Tulane University to use their SmartCare system for Electronic Medical Records.

This was a fascinating opportunity to compare IT infrastructure in Ethiopia to some of my own previous experience, when working on Electronic Health Records with Alain Vuylsteke at Papworth hospital and Matthew Jones in the Judge Business School. Over more than 10 years of research, we’ve looked at the implications of deploying customisable systems in hospitals, and at the potential for the recorded data to be used in improving clinical practice with AI-based visualisation methods.

Minichil tells me that he is personally responsible for patient record systems at 100 hospitals across the Bahir Dar region, and another 800 health centres. This seems a huge load, even for a very energetic 25 year-old, so I explored some of the background.

It seems that SmartCare was originally developed by Tulane University in Zambia, then selected for deployment in Ethiopia by the Federal Ministry of Health in Addis Ababa. However a Masters thesis in 2011 noted that the software was a closed application, owned by TUTAPE (Tulane University’s Technical Assistance Program for Ethiopia), and that the Ministry doesn’t have direct access to make modifications or improvements. The student at that time reported that there wasn’t any available documentation, user manual or technical description of the system, and suggested that Tulane University might train a couple of Ethiopian programmers and give them the right to maintain the system. The conclusion of that thesis was that the Ministry should perhaps start again with an open source package.

It looks as though Minichil's training, together with 9 other Ethiopian graduates, was a response to that situation - and has led to the current status in which he is indeed able to make local customisations. However, the ambition of the rollout is way beyond what a single sysadmin/programmer could hope to handle. While the Bahir Dar health centre has 42 networked workstations, and is keeping full clinical records for all its patients, most of the other 100 hospitals have only a single (non-networked) computer, and use it only to register patients and record basic demographic data in their "card room" (paper file store), with no clinical data at all.

Minichil arranged for me to make a tour of the main hospital in Bahir Dar, together with his colleague Geremew Simachew, and Dr Abiy, the director of the hospital. A small server room is under construction, and some network cables have been pulled, but the next stage of deployment needs funding for about 100 computers. The initial implementation will allow Accident and Emergency triage to access patient data online (rather than waiting for paper records to be carried from the "card room" across the hospital grounds), for laboratory results to be recorded electronically, and for stock records to be updated in the A&E pharmacy.

Even these provisions would only be a small step toward the kind of patient record systems that I have worked with in Cambridge. But resource prioritisation is a real challenge, given the many other limitations that I saw in the A&E department - lack of beds, poor physical access, no paramedics, and extremely limited stocks even of the pharmaceuticals necessary for routine A&E work. There may be some potential for AI to assist, once data starts to be recorded, but the technical challenges here are mainly elsewhere.