Early detection can make emerging epidemics easier to contain, but existing surveillance systems may fail when a disease produces symptoms that are unexpected, evolve over time or vary between patients. An article published by Nature on 2 November 2026 argues that health systems need to improve how they recognize these clinical signals.
Outbreak monitoring currently relies on event-based surveillance, genomic sequencing and syndromic surveillance. These methods use reports from communities and health workers, laboratory confirmation of pathogens, and the identification of people with characteristic symptoms. However, atypical presentations and co-infections can obscure diagnoses and delay public-health action.
The article points to several examples. During the early 2020 spread of SARS-CoV-2, fever and respiratory distress were initially emphasized, while loss of smell took months to become widely recognized as an indicator. Later variants, including Omicron, were more often associated with cold- and influenza-like symptoms. In the 2022 mpox outbreak, some patients developed genital or anal sores instead of the generalized rash commonly associated with the disease. Nipah virus infections in Bangladesh and India can also involve changing patterns, including brain inflammation, confusion and seizures, or breathing difficulties.
To address these gaps, the authors propose five connected steps under the term “symptom preparedness”. The approach is intended to detect clinical signals, validate them, incorporate them into surveillance and diagnostic guidance, and return updated information quickly to front-line clinicians.
The first step is to establish formal mechanisms for reporting clusters of unusual cases. Clinicians could document what the article calls “clinical drift” through existing digital surveillance platforms, such as District Health Information Software 2. A senior clinician or hospital epidemiologist would verify that a cluster is genuinely atypical, while statistical tools could help determine whether it is unlikely to reflect routine seasonal illnesses. Verified reports could then trigger localized alerts for national public-health intelligence centres to assess.
The second proposed measure is using artificial intelligence to identify atypical cases. Machine-learning systems trained on clinical data labelled by time and location could flag patients whose symptoms differ from patterns normally seen in their communities. The systems would not diagnose diseases, but would identify cases requiring further investigation and support, rather than replace, clinical judgement.
The authors note that these tools remain at an early stage and would require reliable connectivity, good-quality data, local testing, continuous monitoring and privacy protections. In settings without the facilities to run AI models, anonymized information could be submitted to a centralized national system through mobile applications, SMS or existing electronic registries.
The third measure is to adopt tiered and dynamic case definitions instead of relying solely on fixed clinical criteria. Public-health authorities could distinguish suspected, probable and confirmed cases, with specialist committees updating the criteria as new clinical information becomes available. The article argues that such a system might have allowed localized genital sores to be added more rapidly to the suspected mpox case definition in 2022.
The source text identifies two further elements of the proposed approach: validating emerging clinical signals and rapidly translating them into surveillance and diagnostic guidance, followed by feedback to clinicians. Together, these steps are intended to help health systems respond when pathogens do not initially resemble the diseases they are known to cause.
Sources: Nature.