The current deadly Ebola outbreak in Africa is the latest in a series of disease outbreaks to make headlines in recent years, following coronavirus disease 2019 (COVID-19), mpox and hantavirus. The earlier an outbreak is detected and contained, the easier it is to control. Yet many emerging, evolving or unfamiliar diseases evade existing surveillance systems because they can produce unusual symptoms that are difficult to recognize.
Better diagnosis could have limited this Ebola outbreak
Epidemics are tracked in three main ways. Event-based surveillance monitors local sources, including media reports, social-media posts, community alerts and reports from health-care workers, for early signs of an outbreak. Genome sequencing is then used to confirm the identity of the pathogen.1,2 Symptomatic surveillance looks for people at risk of infection by identifying and monitoring individuals with characteristic symptoms.
These approaches do not fully account for the possibility that emerging pathogens will cause unusual or atypical symptoms. Symptoms can change as a pathogen evolves3, and they can differ from person to person depending on levels of immunity. Co-infections that produce similar symptoms can further obscure the diagnosis.4
This gap in surveillance has caused problems repeatedly.
During the SARS-CoV-2 outbreak in early 2020, initial diagnosis relied on symptoms such as fever and difficulty breathing.5,6 In many countries, it took months for loss of smell to be widely accepted as an indicator of COVID-19 infection. During that time, some infected people could bypass isolation protocols. Later variants also caused different symptoms: Omicron was more often associated with cold- and flu-like symptoms than with loss of smell or taste.7
The origins of the coronavirus: What we know and what we don’t know
During the 2022 mpox outbreak, which primarily affected Europe and North America, some infected people developed sores on their genitals and anus rather than the widespread rash associated with classic mpox. These cases could initially be mistaken for other conditions, including sexually transmitted infections, potentially delaying recognition and public-health measures.8
Nipah virus has similarly been difficult to diagnose in clinics in Bangladesh and India because its symptoms can vary from inflammation of brain tissue, which can cause confusion and seizures, to difficulty breathing.9
Even the most advanced tracking systems will not respond unless clinicians recognize unusual symptoms as worthy of investigation. Five steps could improve health systems’ ability to recognize emerging pathogens. Together, they form a loop in which clinical signals are detected, validated, translated into monitoring and diagnostic guidance, and rapidly fed back to frontline clinicians. This approach is called “symptomatic preparedness.”
Flag clusters of unusual cases
Health systems need a formal mechanism that allows clinicians to act on the impression that they are seeing something unusual, such as several people presenting with the same previously unrecognized set of symptoms. National public-health authorities and ministries of health should establish procedures for documenting this type of “clinical drift” through existing digital-surveillance tools, such as the online platform District Health Information Software 2 (DHIS2).
Credit: Tony Karumba/AFP via Getty
To reduce false-positive alerts, a second senior clinician or hospital epidemiologist should confirm that the cluster is atypical. Clinicians can then run the relevant algorithm in DHIS2 or an equivalent tool and use statistical checks to determine whether the symptoms are unlikely to have been caused by routine seasonal illnesses. Once reports are verified, localized digital alerts can be sent automatically to the country’s public-health intelligence center. Epidemiologists can then cross-reference local data before formal surveillance measures are triggered.10
Use AI to flag special cases
Machine-learning models trained on clinical data labeled by time and location could help doctors detect atypical cases. By identifying people whose symptoms differ from patterns typically observed in the same population, such models could highlight cases that require further testing. The models would not provide a diagnosis; instead, artificial-intelligence tools would support rather than replace clinical judgment.11,12
Such AI systems are being developed, but they are currently at an early stage.13 Implementation requires reliable internet connectivity, high-quality data, testing and evaluation in the intended local setting, continuous performance monitoring and privacy protections.
Low-resource settings might lack the infrastructure needed to run AI models. However, health-care workers could send anonymized clinical information to a national, centralized AI system through a phone app, an SMS-based reporting system or an existing electronic registration system. The AI system could then return alerts when needed.
Adopt a step-by-step case definition
Public-health authorities should move away from relying solely on standardized clinical criteria provided by organizations such as the World Health Organization to classify disease cases. Instead, they should adopt a dynamic, step-by-step case definition that distinguishes among suspected, probable and confirmed cases.
Don’t let the AI biohazard hype distract from the real risks we currently face
A panel of epidemiological experts could stratify diagnostic criteria into stages and update them weekly or at other regular intervals as real-time clinical data emerge.10 For example, standard mpox symptoms have traditionally included a generalized rash. If the dynamic diagnostic system had been available in 2022, public-health officials could have expanded the definition of a “suspected case” to include localized genital ulcers within days of recognizing the atypical cluster. This rapid adaptation might have enabled misdiagnosed individuals to be identified sooner.
Source: www.nature.com