Health
AI: Detecting epidemics before the alarm bells ring
Daniel Ayantoye
One person develops a fever, while another reports diarrhoea. A mother searches online for the cause of her child’s persistent symptoms, and a health worker notices several patients presenting with similar complaints.
Individually, the incidents may appear insignificant. But analysed together, such signals could reveal a pattern that warrants investigation.
That is where artificial intelligence is beginning to attract attention in disease surveillance.
Rather than waiting for hospitals and laboratories to confirm enough cases to establish an outbreak, AI systems can analyse information from multiple sources, including health records, laboratory reports, news reports, social media, environmental data, climate information, mobility patterns and wastewater.
The technology is not expected to diagnose an epidemic on its own. Its emerging role is to help public health authorities identify unusual patterns earlier and determine where further investigation may be necessary.
For Nigeria, that possibility is already being tested.
Nigeria tests AI for earlier warnings
A 2026 study published in the Journal of Medical Systems examined the use of EPIWATCH, an AI-powered epidemic-intelligence platform, for infectious disease surveillance in Nigeria.
Researchers incorporated Nigerian Pidgin English into the system to help it recognise disease-related expressions used by Nigerians online.
Following the integration, EPIWATCH detected 112 outbreak reports in March 2024, compared with a median of 27 reports for March during the preceding five years, representing a 315 per cent increase in detected reports.
Signals relating to Lassa fever, meningitis, diphtheria and yellow fever also increased.
The finding does not mean Nigeria experienced 315 per cent more outbreaks.
Rather, it showed that improving an AI surveillance system’s ability to understand local language could increase the number of potential public health signals it detects.
For a country with substantial linguistic diversity, the implication is significant.
Professor of Intelligent Systems at the University of Jos, Nachamada Blamah, said the effectiveness of an AI prediction system would ultimately depend on the quality and breadth of the data available to train it.
“It all depends on the nature of data that we have available. Prediction is all about availability of data,” he said.
According to Blamah, a model must be trained on sufficient historical data and then tested against data it has not previously seen before it can be considered reliable enough for deployment.
He cautioned against simply importing models developed in other countries.
The don said, “There are advanced models that have been used in advanced countries. The only thing is that the data that exist for different cases outside Nigeria may not fit in properly for our adoption.”
Blamah said Nigeria therefore needed local datasets based on real cases recorded within the country because disease patterns could be influenced by circumstances peculiar to the Nigerian environment.
Before the official alert
Traditional surveillance remains the foundation of epidemic control.
Health workers identify suspected cases, laboratories confirm infections, and health authorities analyse reports and investigate unusual increases.
But these processes can take time.
Event-based surveillance attempts to shorten the gap by monitoring information outside routine disease reporting.
Nigeria Centre for Disease Control and Prevention operates an event-based surveillance system designed to detect and verify rumours, reports and other information relating to possible public health threats.
An evaluation of Nigeria’s national event-based surveillance programme found that over 43,000 raw signals were detected between September 2017 and June 2018. After screening, 138 were escalated for investigation, with 63 subsequently verified as public health events, including Lassa fever and cholera outbreaks.
The challenge, therefore, is not simply finding information but determining which signals matter. That is one area where AI could be useful.
A Professor of Computer Science at Ladoke Akintola University of Technology, Ogbomoso, said AI has the capacity to support disease detection when sufficient data are available.
He said, “Yes, AI has the capacity to detect epidemic and detect diseases ahead. There are aspects of AI called machine learning and deep learning. If there are data, it can predict disease and also classify the type of diseases a person has.”
He, however, stressed that research and development must receive greater government support if such technologies are to move beyond academic studies.
When social media becomes a sensor
Every day, Nigerians leave behind digital traces of what is happening around them.
People search for symptoms, discuss illnesses, report unexplained deaths and share information about health problems in their communities.
Most of these conversations have no epidemiological significance.
But some could contain early clues.
A 2025 scoping review of digital disease surveillance found growing interest in social media, search engines and other online data as sources of early-warning information.
The value of AI lies partly in its ability to process this information at a scale that would be difficult for human analysts.
An algorithm can scan thousands of posts, reports and records and identify changes in the frequency, location or combination of disease-related terms. But a signal is not an outbreak.
A sudden increase in online discussion about diarrhoea, for example, could reflect an actual cluster of illness or simply a viral news report. Human verification, therefore, remains essential.
AI-based epidemic surveillance is not limited to online conversations.
Researchers are exploring systems that combine conventional epidemiological data with climate information, human mobility, genomic surveillance, environmental measurements and wastewater monitoring.
Changes in rainfall and temperature can influence disease vectors and transmission patterns.
Population movement can affect the spread of infection between communities, while genomic surveillance can help scientists monitor how pathogens evolve.
Wastewater surveillance provides another potential source of early warning by detecting traces of pathogens circulating within a population.
A 2025 systematic review of AI-based early-warning systems identified internet searches, social media, environmental and climate data, travel and mobility information, genomic data and wastewater among the sources being investigated for infectious disease surveillance.
The emerging model is, therefore, not simply an AI system watching social media.
It is a system capable of combining multiple signals and identifying patterns that may warrant human investigation.
The growing use of AI creates an important misconception: that algorithms can predict an epidemic with certainty. They cannot.
AI identifies patterns based on the data available to it. Epidemiologists must determine whether those patterns represent a genuine public health threat.
Blamah’s warning about data is particularly relevant here. A sophisticated algorithm trained on incomplete or poorly representative Nigerian data could produce predictions that appear precise but are fundamentally unreliable.
A false signal could also arise from misinformation, a viral news story or an unusual but harmless increase in online discussion.
That is why AI should function as an additional layer of epidemic intelligence rather than a substitute for laboratory confirmation, field investigation or professional judgement.
The machine can raise the alarm. Humans must establish whether there is a fire.
The Nigerian data challenge
AI cannot detect what the data fail to capture. Where health facilities do not report cases promptly, laboratory capacity is limited or digital records are incomplete, an algorithm may receive an incomplete picture of disease activity.
This is particularly important for Nigeria, where health information systems and digital surveillance capacity remain uneven.
The challenge is not merely to acquire more data but to build reliable, representative and locally relevant datasets.
As Blamah explained, data from another country may not adequately reflect Nigeria’s disease patterns, population characteristics or environmental conditions.
That makes the development of Nigerian datasets critical to any serious attempt to build AI systems for epidemic prediction.
Better algorithms alone will not compensate indefinitely for missing laboratory results, weak reporting systems or under-represented communities.
The price of getting it wrong
The promise of faster detection must also be balanced against the risks.
AI systems can reproduce biases in the data used to train them. They can generate false signals, overlook communities that produce little digital information and raise privacy concerns when analysing health-related data.
The World Health Organisation has also warned that artificial intelligence in health raises concerns involving bias, privacy, cybersecurity, misinformation and patient safety and has called for strong governance and human oversight.
For epidemic surveillance, the consequences of a poorly designed system could be serious.
A false negative could delay an outbreak response. A false positive could trigger unnecessary anxiety, investigations or the diversion of scarce public health resources.
The objective should, therefore, not be to build an AI system that makes decisions without people. It should be to build one that helps public health professionals make better decisions earlier.
From research to policy
The two Nigerian professors point to another challenge: moving promising research from academic institutions into practical public health systems.
The LAUTECH professor said limited research funding and the relatively small number of industries capable of supporting the development and deployment of research products remained obstacles.
He argued that AI-based disease surveillance should form part of government policy so that promising research does not remain confined to universities.
This could be crucial for Nigeria.
Developing an effective epidemic-intelligence system would require more than purchasing software.
It would require sustained investment in data infrastructure, laboratory networks, computing capacity, cybersecurity, trained epidemiologists and AI specialists.
It would also require mechanisms for testing algorithms against real Nigerian outbreaks before they are trusted to inform public health decisions.
Nigeria’s opportunity
Nigeria already has important pieces of the infrastructure required for this transition.
The country has an established disease-surveillance system, event-based surveillance, digital epidemic-management tools and a growing technology sector.
More importantly, researchers have begun testing AI against Nigerian surveillance challenges.
The EPIWATCH Nigerian Pidgin study offers an important lesson: technology designed around Nigeria’s linguistic and epidemiological realities may perform differently from technology simply imported from elsewhere.
The next step could be to connect more sources of information.
Imagine an early-warning platform detecting an unusual increase in diarrhoeal symptoms in a local government area while simultaneously identifying heavy rainfall, flooding, increased online searches and reports from health facilities.
None of those signals would prove an outbreak.
Together, however, they could tell epidemiologists that the area warrants immediate attention.
That is where AI’s greatest public health value may lie.
The early-warning advantage
In epidemic response, time can determine the scale of an outbreak.
An earlier warning can give health authorities more time to investigate cases, deploy surveillance teams, strengthen laboratory testing, communicate with communities and prepare healthcare facilities.
AI offers a way of processing large volumes of information and identifying patterns that might otherwise be difficult to detect quickly.
For Nigeria, the question is no longer simply whether artificial intelligence can contribute to disease surveillance. It can.
The more important question is whether the country can build the data systems, technical expertise, governance safeguards and human capacity required to use it responsibly.
The next epidemic may not announce itself with a dramatic surge in hospital admissions.
It may begin as a collection of seemingly unrelated signals: a few unusual symptoms, a cluster of community reports, a change in environmental conditions or an increase in online searches.
AI may be able to connect those signals before they become an obvious pattern.
But the final warning will still belong to people. The machine may spot the signal.
The epidemiologist must determine what it means and decide when to act.
Copyright PUNCH
All rights reserved. This material, and other digital content on this website, may not be reproduced, published, broadcast, rewritten or redistributed in whole or in part without prior express written permission from PUNCH.
Contact: [email protected]
