Predicting medicine stock-outs to protect neonatal care in East Africa

In sub-Saharan Africa, the neonatal mortality rate stands at 27 deaths per 1,000 live births^[1], and the region accounts for roughly 43% of all newborn deaths worldwide^[2]. Many of these deaths are preventable. Conditions like birth asphyxia, sepsis, and complications of prematurity are treatable when the right medicines and supplies are available at the point of care. But across East African health facilities, stock-outs of essential neonatal and maternal medicines remain common: studies in Tanzania have found stock-out rates above 30% for maternal health medicines, and in some regions fewer than 10% of facilities had sufficient stocks of unexpired, properly stored medications. When a facility runs out of a critical drug, the delay in care can be fatal for a newborn who cannot wait.

Global Strategies works to improve health outcomes for mothers and newborns in East Africa through Noviguide, a platform that supports clinical decision-making and tracks medicine stock levels across healthcare facilities. Revolution Analytics partnered with Global Strategies through Tech To The Rescue, a platform connecting technology companies with NGOs for social impact projects. The goal was to build a stock-out risk prediction system focused on high-priority neonatal and maternal drugs, the medicines that matter most when a newborn’s life depends on immediate treatment.

The challenge

Noviguide was already collecting stock data from facilities, but the information only told teams what had already happened. By the time a facility reported a stock-out, the shortage was a fact and the window for intervention had closed. For neonatal care, where treatment delays of even hours can be the difference between life and death, reactive monitoring was not enough.

The data itself reflected the realities of healthcare operations in resource-limited settings. Facilities reported at different frequencies. Some updated stock levels daily, others went weeks between reports. Stock behaviour for a given medicine varied significantly across facilities depending on patient volume, supply chain reliability, and local prescribing patterns. A simple threshold rule like “alert when stock is low” would only tell teams what they already knew, and it would generate so many alerts that the signal would be lost in the noise.

What was needed was a system that could learn from each facility’s stock patterns over time, estimate the probability of a stock-out before it happened, and surface that risk in a way that was actionable and trustworthy enough for teams to act on.

What we built

We designed and delivered an end-to-end prediction and alerting system that estimates seven-day stock-out risk for every facility and medicine, converts those predictions into severity-graded alerts, and explains why each alert was raised.

Stock data from Noviguide flows into the system in near real time, replacing the periodic manual exports the team had relied on. An event-driven pipeline on Google Cloud captures, deduplicates, and normalises stock activity as it happens across the facility network, giving the model a continuous and up-to-date view of stock levels everywhere.

The model estimates the probability that a given facility will run out of a specific medicine within the next seven days. Rather than a simple yes/no, each prediction is mapped to four severity levels with a confidence rating, so teams can distinguish between a high-risk alert backed by strong data and one flagged under uncertainty. Every alert comes with a human-readable explanation of the specific drivers behind the risk: whether that is recurring shortages, a recent drop from normal stock levels, or a long gap since the facility last reported. When a coordinator can see exactly why a facility was flagged, they can make faster decisions about where to direct limited resources.

The system feeds into a dashboard that gives supply teams a real-time view of stock-out risk across all facilities and medicines, with stock history charts, alert timelines, and facility-level drill-downs. Built-in alert management prevents fatigue through daily caps and cooldown logic, and a full audit trail tracks every alert from creation to resolution.

The result

The system moved Noviguide from reactive stock monitoring to proactive risk detection for the medicines that matter most to neonatal survival. Instead of learning about a stock-out after a newborn has already been denied treatment, supply teams now have a seven-day window to intervene, whether that means redirecting stock from a neighbouring facility, expediting a resupply, or escalating to a regional coordinator.

For a region where nearly three-quarters of neonatal deaths occur in the first week of life and where around two-thirds of emergency obstetric facilities are not fully functional, the ability to predict and prevent medicine shortages is not an operational convenience. It is a direct line to keeping newborns alive. The system is running in production and continues to process stock signals daily as Global Strategies expands Noviguide’s reach across East Africa.