AI-Enabled Early Prediction of Lassa Fever: A Clinical Decision Support System for Endemic Settings

In the warm, humid regions of West Africa, a silent and dangerous game of medical triage plays out every day in crowded clinics. When a patient arrives with a fever, a headache, and a general feeling of sickness, the doctor faces a difficult choice. These symptoms are common to many illnesses, from malaria to typhoid, but they are also the first signs of Lassa fever, a viral hemorrhagic disease that can be fatal if not treated quickly. The trouble is that confirming a case of Lassa fever requires a specialized laboratory test that is often far away, expensive, or simply unavailable in remote areas. By the time a lab result returns, a patient might have already spread the virus to others in the waiting room, or their own condition might have worsened beyond repair. For decades, doctors have had to rely on their own judgment and a set of general rules to decide who needs to be isolated immediately and who can wait. This uncertainty creates a bottleneck where the most dangerous cases might be missed, or where limited resources are wasted on patients who are actually safe.

A team of researchers from Nigeria and international partners has built a new digital tool to help doctors navigate this uncertainty before a lab test is even possible. They created a computer system that acts as a second pair of eyes for clinicians, analyzing the information available at the very first moment a patient walks through the door. Instead of waiting for a lab, the system looks at who the patient is, how long they have been sick, and exactly which symptoms they are showing. By learning from a massive collection of past medical records, the system can spot patterns that are too subtle for the human eye to catch consistently. It does not claim to diagnose the disease with certainty—that still requires a lab—but it can tell a doctor which patients are most likely to have Lassa fever and therefore need urgent attention. This approach shifts the focus from guessing to calculating, offering a way to catch the disease earlier in places where every minute counts.

The researchers trained this system using a decade of real-world data from the Irrua Specialist Teaching Hospital in Nigeria, one of the largest centers for treating Lassa fever in the world. They gathered records from over 31,000 patient visits between 2015 and 2025, a period that covered many different outbreaks and changes in how the disease behaved. To make sure the tool would work in the real world, they were very strict about what information they allowed the computer to see. They only fed it data that a doctor would have at the very start of a visit: the patient's age, how many days had passed since symptoms started, and a list of symptoms like fever, vomiting, abdominal pain, or red eyes. They deliberately left out any information that would only be known after a lab test was done, ensuring the tool could be used exactly when it is needed most.

The team tested several different ways of teaching the computer to recognize the disease, eventually settling on a method called XGBoost, which is a powerful type of learning algorithm that builds a complex decision-making structure by combining many simple rules. They tested this system not just on random chunks of data, but on the most recent years of records, simulating a future where the disease might be less common than it was in the past. This was a crucial test because the prevalence of confirmed cases in their data dropped significantly over time, from nearly 11 percent in the earlier years to just over 6 percent in the most recent period. The system had to prove it could still find the sick patients even when they were harder to find.

The results showed that the tool was exceptionally good at its primary job: not missing anyone who might be sick. On the most recent data, the system correctly identified nearly 98 percent of the patients who were later confirmed to have Lassa fever. This high level of sensitivity is vital in an outbreak setting, where the cost of missing a single case is the continued spread of a deadly virus. However, the researchers were honest about the trade-off. Because the system is tuned to catch almost every possible case, it also flagged many people who did not actually have the disease. In the most recent test group, only about 6 percent of the people the system flagged actually had the virus. While this sounds like a high number of false alarms, the researchers explain that in a triage situation, it is far better to over-prepare and isolate a few healthy people than to miss a single infected one.

To make the tool useful for busy doctors, the team did not just give a simple "yes" or "no" answer. Instead, they organized the computer's predictions into three clear categories: low risk, moderate risk, and high risk. This allowed the system to sort patients into groups that matched different levels of urgency. For example, in the most recent test data, the "high risk" group contained only about 11 percent of all the patients, yet it included nearly 30 percent of all the confirmed Lassa fever cases. This means the tool successfully concentrated the most dangerous cases into a small group, allowing hospitals to focus their limited isolation rooms and lab tests on the people who needed them most. The "moderate risk" group captured another 22 percent of the confirmed cases, ensuring that a large majority of potential threats were escalated for further review.

The system also provides a reason for its decision, showing the doctor which specific factors pushed the risk score up or down. For instance, if a patient is older and has had symptoms for only a few days, the system might flag them as higher risk. If they have red eyes or general weakness, those symptoms might add to the risk score. This transparency helps doctors trust the tool, as they can see the logic behind the suggestion rather than treating it as a black box. The entire system was built to be lightweight and accessible, running on a simple web page that can be opened on a standard computer or tablet without needing powerful servers or expensive internet connections. This design ensures that the technology can be used in the very clinics where it is needed most, even in areas with limited infrastructure.

The study concludes that while the tool is not a replacement for a laboratory test or a doctor's judgment, it is a powerful aid for the critical first moments of patient care. By using a decade of real data to learn the subtle patterns of the disease, the system offers a way to make faster, more consistent decisions in a chaotic environment. The researchers acknowledge that the tool was tested in a single location and that its performance might vary in other places or under different conditions. They also note that the high number of false alarms is an expected part of a safety-first strategy in a low-prevalence setting. Nevertheless, the ability to catch nearly every confirmed case while sorting patients into clear risk groups suggests that this kind of digital support could become a standard part of how doctors fight Lassa fever, turning the tide from reactive treatment to proactive containment.