How a tablet is changing neonatal care in Zimbabwe

"The doctor has just admitted twelve babies in three hours. She knows each baby's name and gestational age. But the clinical history for seven of them was recorded on paper scraps that won't survive the week, and the early warning signs she's watching for are held entirely in her memory." — The problem Neotree was built to solve

Neonatal mortality remains one of the most persistent and tractable problems in global health. Sub-Saharan Africa carries a disproportionate share of the burden: in Zimbabwe, neonatal deaths account for roughly 40% of all under-five mortality. The causes — sepsis, hypothermia, birth asphyxia, low birth weight — are largely preventable or treatable with timely, appropriate clinical intervention. The gap is not, primarily, one of medical knowledge. It is one of information.

This is the problem that Neotree addresses.

What Neotree actually does

Neotree is a digital health platform deployed in neonatal units in Zimbabwe (Sally Mugabe Central Hospital, Harare) and Malawi (Kamuzu Central Hospital, Blantyre). At its core, it is a structured data capture tool — a tablet-based form that guides a clinician through the systematic recording of a baby's admission history, clinical signs, management plan, and outcomes.

But Neotree is more than an electronic form. As the clinician enters data, the platform checks for internal consistency, flags clinical red flags according to evidence-based protocols, and generates a structured admission record that can be reviewed, audited, and analysed. Every baby admitted through Neotree generates a data point. Over time, those data points become a dataset — one that, in a low-resource neonatal unit, simply did not exist before.

The data science behind the platform

My role within Neotree covers several distinct but connected workstreams. The first is interrupted time series (ITS) analysis — an epidemiological method for evaluating the impact of an intervention on a time-ordered outcome measure. We are using ITS to quantify whether — and by how much — the introduction of Neotree changed neonatal outcomes at the sites where it was deployed. This work is in preparation for publication.

The second strand is machine learning for clinical prediction. Working with colleagues at UCL and Imperial College, we are developing models to predict early-onset sepsis and hypothermia from the structured admission data that Neotree generates. Hypothermia — defined as a temperature below 36.5°C — is a strong independent predictor of neonatal mortality and is measurable on admission. A model that can flag high-risk babies early, before deterioration becomes clinically apparent, gives clinicians something they currently lack: time.

The third strand is data infrastructure and governance. The Neotree dataset exists in two environments: a local version managed at the hospital site, and a version held in the NHS Digital Services Hub (DSH), a Trusted Research Environment in the UK. Building and maintaining the pipelines that move data from point of care into a form suitable for analysis — cleanly, consistently, and in compliance with data governance requirements — is unglamorous work. It is also foundational to everything else.

What makes this hard

Clinical decision support in low-resource settings is technically and logistically different from the same work in a well-resourced hospital system. The infrastructure assumptions that underpin most clinical AI — reliable internet connectivity, standardised electronic health records, large labelled training datasets — do not hold. Neotree was designed from the ground up for intermittent connectivity and a tablet-first workflow. The machine learning models we are developing are trained on the Neotree dataset itself, which means they are built on real data from these specific clinical environments rather than generalised from populations elsewhere.

None of this makes the work less rigorous. It makes it more constrained — and more directly useful to the clinicians and babies it is designed to serve.