Characterising pandemic severity and transmissibility from data collected during first few hundred studies

Abstract

Early estimation of the probable impact of a pandemic influenza outbreak can assist public health authorities to ensure that response measures are proportionate to the scale of the threat. Recently, frameworks based on transmissibility and severity have been proposed for initial characterization of pandemic impact. Data requirements to inform this assessment may be provided by "First Few Hundred" (FF100) studies, which involve surveillance-possibly in person, or via telephone-of household members of confirmed cases. This process of enhanced case finding enables detection of cases across the full spectrum of clinical severity, including the date of symptom onset. Such surveillance is continued until data for a few hundred cases, or satisfactory characterization of the pandemic strain, has been achieved. We present a method for analysing these data, at the household level, to provide a posterior distribution for the parameters of a model that can be interpreted in terms of severity and transmissibility of a pandemic strain. We account for imperfect case detection, where individuals are only observed with some probability that can increase after a first case is detected. Furthermore, we test this methodology using simulated data generated by an independent model, developed for a different purpose and incorporating more complex disease and social dynamics. Our method recovers transmissibility and severity parameters to a high degree of accuracy and provides a computationally efficient approach to estimating the impact of an outbreak in its early stages.

Bayesian model discrimination for partially-observed epidemic models

“…We consider diseases that have a lag between exposure of individuals and infectiousness, specifically an SE(2)I(2)R model, which is detailed in Section 4.1. This kind of model has been used in previous work on inference using early outbreak data; it has realistic features, such as non-295 exponential exposed and infectious periods, while being simple enough for inference [28,29,30,31,7]. We model the observations of symptom onset as either a transition into an exposed, infectious or recovered state for the Post, Co and Pre models respectively ( Figure 4).…”

“…While the 20 interpretability and the consistency with the Bayesian paradigm is desirable, Bayesian model selection has some issues: a large amount of data may be required before models can be distinguished; and the evidence is typically difficult to calculate as the likelihood is intractable. This is true for epidemic models, which for small populations are most naturally represented as partially-observed continuous-time Markov chains (CTMCs) 25 [6,7,8].…”

A sub-exponential branching process to study early epidemic dynamics with application to Ebola

“…Lags in time series data cause substantial problems when forecasting incidence Moss et al (2018). “First Few Hundred” (FF100) studies collect the same type of data as contact tracing and are heralded as a way to rapidly provide a characterisation of transmission dynamics Black et al (2017). While for the 2014 Ebola epidemic the time series was available before the secondary infections tree, there does not seem to be anything intrinsic to the data collection process that precludes this being reversed.…”

Infectious disease pandemic planning and response: Incorporating decision analysis

“…To this end, advances have recently been made in the design of early outbreak surveillance methods such as First Few Hundred (FF100) household transmission studies [26] and the development of novel algorithms for analyzing the resulting data [32]. FF100 studies involve the collection of data from confirmed infections and their household contacts, including the date of symptom onset and final outcome, until a satisfactory characterization of the pathogen is achieved [26].…”

“…These rapid, enhanced surveillance activities can be resource intensive but provide rich epidemiological data and overcome many quality, timeliness, and bias issues often associated with routine surveillance practices [29]. Further, when these data are analyzed with FF100-specific algorithms [32,34], estimates of pathogen transmissibility and severity are obtained, enabling timely identification of the pandemic scenario that best characterizes an actual outbreak.…”