Imperial College London’s Department of Infectious Disease Epidemiology reports a model-predictive control approach that times non-pharmaceutical interventions from noisy real-time case data, generally achieving better control with lower intervention costs than preset schedules or simple thresholds.
Reacting to a public health crisis such as an outbreak of an infectious disease often relies on surveillance streams affected by reporting delays and missed infections, which can cause lags in actionable intervention decision-making. These streams consist of patients reporting symptoms, doctors ordering tests and labs returning results, and reporting of cluster event findings.
If patients do not report symptoms, doctors diagnose without testing, or labs lack regular reporting procedures, outbreaks can spread largely unseen. Conversely, knowing when an existing outbreak is sufficiently contained is hindered by the same spotty reporting data.
Missing from surveillance systems is a framework for cutting through the noise, simultaneously treating stochastic spread, incomplete case reporting, and potential intervention cost benefit ratios in real time.
In the study, “Optimal algorithms for controlling infectious diseases in real time using noisy infection data,” published in PLOS Computational Biology, researchers developed a model-predictive control algorithm to optimize when to enforce or relax tiered interventions using short-horizon projections under delayed and under-ascertained incidence.
Modeling used a renewal branching process for daily infections with generation-time distributions representative of COVID-19 and Ebola virus disease. Intervention choices were grouped as no intervention, limited social distancing, and full lockdown, implemented based on transmissibility and containment.
Surveillance imperfections were simulated by applying predictions of reporting delays and missing diagnoses. Simulations indicate that the model stabilized incidence under ideal observations, with performance degrading when delays or under-ascertainment distorted the data feeding the model. Long and relatively deterministic delays pushed action later in time, inflating peaks and widening oscillations in event impact.
Under moderate noise typical of practical surveillance, projection-based decisions generally reduced peaks and intervention time compared with threshold or cyclic intervention rules.
Simulations using Ebola virus disease showed tighter control, reflecting slower epidemic growth relative to SARS-CoV-2-like settings. Stress tests with changes to transmissibility or variable intervention effect sizes showed adaptive re-estimation and re-optimization were able to maintain spread control.
Authors conclude that earlier, faster surveillance paired with regular reviews supports well-calibrated decisions that can curb peaks, reduce intervention times and outbreak burden.
Written for you by our author Justin Jackson, edited by Sadie Harley, —this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive.
If this reporting matters to you,
please consider a donation (especially monthly).
You’ll get an ad-free account as a thank-you.
More information:
Sandor Beregi et al, Optimal algorithms for controlling infectious diseases in real time using noisy infection data, PLOS Computational Biology (2025). DOI: 10.1371/journal.pcbi.1013426
The content is provided for information purposes only.