EMR in Ethiopia: adoption is not the hard part, use is

Ethiopia is not debating whether to digitise clinical records. The Federal Ministry of Health published a national Digital Health Blueprint in 2021 setting a ten-year direction for interoperable systems, national data standards and unique patient identification 6. The question the research has moved on to is narrower and more useful to a clinic owner: when a facility has an EMR, do the people in it use it?

The answer is: about half of them.

The measured usage rate

A systematic review and meta-analysis pooling five Ethiopian studies covering 2,439 health workers put EMR use at 51.85% (95% CI 37.14–66.55) 1. The regional breakdown is the part worth pausing on:

Percentage of health workers using an installed EMR. The intervals overlap heavily — each subgroup rests on one or two studies.

Source: Tegegne et al. 2023, meta-analysis of 5 studies, n=2,439 [1]

Show data table

	Estimate	95% CI
Northern Ethiopia	58.75%	34.74%–82.76%
Eastern Ethiopia	54.99%	30.1%–79.88%
Pooled national estimate	51.85%	37.14%–66.55%
Addis Ababa	31.7%	27.26%–66.55%

Addis Ababa — the best-connected, best-staffed, best-equipped part of the country — had the lowest recorded use. Whatever limits EMR use, it is not bandwidth or hardware budgets alone. The confidence intervals are wide, because each subgroup rests on one or two studies; read the ranking as a signal, not a league table.

Readiness measures tell a similar story from the other end. Among 384 health professionals at public hospitals in Borena and West Guji zones, surveyed in May 2024, 60.4% (95% CI 55.5–65.3) expressed readiness to use an EMR system 4 — willingness comfortably exceeds observed use.

What actually predicts use

The meta-analysis reports adjusted odds ratios for using an EMR that is already installed 1:

Adjusted odds ratios with 95% confidence intervals, log axis. Three of the four are things a clinic controls directly.

Source: Tegegne et al. 2023, meta-analysis [1]

Show data table

	Estimate	95% CI
EMR training	3.41	1.25–9.29
Age under 30	2.24	1.36–3.68
Access to an EMR manual	2.08	1.47–2.96
Managerial support	1.7	1.21–2.38

Three of the four are things a clinic controls directly, and the strongest is training. Note what is not on the list: any measure of software features. The evidence points at implementation, not product specification.

The specific things that break adoption

A qualitative study of providers in Addis Ababa names failure modes rather than attitudes 3:

• Dual workflows. Staff maintaining paper and digital records do double work, and the digital copy is what gets skipped.
• No interoperability. With no link between the EMR and DHIS2, reporting data had to be re-entered by hand — described by one provider as "a burden to us."
• Power interruptions. Unpredictable outages force reversion to paper mid-session, and the paper is rarely reconciled back.
• Under-specified hardware. Too few computers, and slow ones, so data entry cannot happen during the consultation.
• Orientation instead of training. Some providers received 25–30 minutes, with little follow-up mentorship and too few IT staff to fix problems while a patient waits.

A parallel qualitative study in Oromia reaches the same conclusion from a different angle: awareness, skills and training were low, systems were under-utilised, and although users recognised the benefits, they were sceptical about long-term sustainability 5.

The Addis providers were nonetheless clear about the benefits when it worked: better data quality and accessibility, shorter patient waiting times, better interdepartmental communication, and streamlined documentation, retrieval and billing 3.

The baseline being replaced is worse than it looks

Comparisons against "the paper system" usually imagine the paper system working. The Ethiopian measurement of that baseline is stark. Of 2,145 medical records sampled across 73 public health facilities in southern Ethiopia, only 18.4% were complete and readable 2.

Share of sampled records with each defect, against the share that were complete and readable.

Source: Endriyas et al. 2022, 73 public health facilities [2]

Show data table

	What was missing from 2,145 paper records
No date and/or signature	60.3%
Missing at least one identification field	29%
No diagnosis recorded	20.1%
No chief complaint	14.3%
No medication recorded	12.5%
At least one non-readable component	9.5%
Complete and readable (all fields)	18.4%

Health centre records were 56.8% less likely to meet the quality standard than hospital records 2.

For a clinic weighing the disruption of moving to an EMR, that is the honest comparison — not against a well-kept chart, but against a chart that four times in five is missing something a clinician, an auditor or an insurer will later need.

What the evidence implies for adoption

1. Budget training in days, not minutes. Strongest modifiable predictor of use 1; 25-minute orientations are documented failures 3.
2. Pick a switch-over date and hold it. Dual paper-and-digital running is named as a barrier, not a safe transition 3.
3. Solve power before rollout, not after. Reversion to paper during outages is where records fragment 3.
4. Keep a written manual available. Unglamorous, independently associated with use (AOR 2.08) 1.
5. Have management visibly use it. Managerial support shows up in the adoption data (AOR 1.70) 1.

What the evidence does not show

Every study cited here was conducted in Ethiopian public facilities, and the meta-analysis rests on only five cross-sectional studies with a wide interval (37.14–66.55) 1 — that interval is honest about the remaining uncertainty. Cross-sectional designs cannot establish that training causes use; they establish that the two travel together.

We found no published measurement of EMR use, abandonment or documentation quality in Ethiopian private clinics, and no Ethiopian study measuring patient outcomes before and after EMR adoption. Claims that an EMR improves clinical outcomes in this setting are extrapolated from elsewhere. The Ethiopian evidence supports the narrower statements made above: adoption is uneven, use tracks training and support, and the paper baseline is poor.