Real-time modelling of a pandemic influenza outbreak

Abstract

Background Real-time modelling is an essential component of the public health response to an outbreak of pandemic influenza in the UK. A model for epidemic reconstruction based on realistic epidemic surveillance data has been developed, but this model needs enhancing to provide spatially disaggregated epidemic estimates while ensuring that real-time implementation is feasible. Objectives To advance state-of-the-art real-time pandemic modelling by (1) developing an existing epidemic model to capture spatial variation in transmission, (2) devising efficient computational algorithms for the provision of timely statistical analysis and (3) incorporating the above into freely available software. Methods Markov chain Monte Carlo (MCMC) sampling was used to derive Bayesian statistical inference using 2009 pandemic data from two candidate modelling approaches: (1) a parallel-region (PR) approach, splitting the pandemic into non-interacting epidemics occurring in spatially disjoint regions; and (2) a meta-region (MR) approach, treating the country as a single meta-population with long-range contact rates informed by census data on commuting. Model discrimination is performed through posterior mean deviance statistics alongside more practical considerations. In a real-time context, the use of sequential Monte Carlo (SMC) algorithms to carry out real-time analyses is investigated as an alternative to MCMC using simulated data designed to sternly test both algorithms. SMC-derived analyses are compared with ‘gold-standard’ MCMC-derived inferences in terms of estimation quality and computational burden. Results The PR approach provides a better and more timely fit to the epidemic data. Estimates of pandemic quantities of interest are consistent across approaches and, in the PR approach, across regions (e.g. R 0 is consistently estimated to be 1.76–1.80, dropping by 43–50% during an over-summer school holiday). A SMC approach was developed, which required some tailoring to tackle a sudden ‘shock’ in the data resulting from a pandemic intervention. This semi-automated SMC algorithm outperforms MCMC, in terms of both precision of estimates and their timely provision. Software implementing all findings has been developed and installed within Public Health England (PHE), with key staff trained in its use. Limitations The PR model lacks the predictive power to forecast the spread of infection in the early stages of a pandemic, whereas the MR model may be limited by its dependence on commuting data to describe transmission routes. As demand for resources increases in a severe pandemic, data from general practices and on hospitalisations may become unreliable or biased. The SMC algorithm developed is semi-automated; therefore, some statistical literacy is required to achieve optimal performance. Conclusions Following the objectives, this study found that timely, spatially disaggregate, real-time pandemic inference is feasible, and a system that assumes data as per pandemic preparedness plans has been developed for rapid implementation. Future work recommendations Modelling studies investigating the impact of pandemic interventions (e.g. vaccination and school closure); the utility of alternative data sources (e.g. internet searches) to augment traditional surveillance; and the correct handling of test sensitivity and specificity in serological data, propagating this uncertainty into the real-time modelling. Trial registration Current Controlled Trials ISRCTN40334843. Funding This project was funded by the National Institute for Health Research (NIHR) Health Technology programme and will be published in full in Health Technology Assessment; Vol. 21, No. 58. See the NIHR Journals Library website for further project information. Daniela De Angelis was supported by the UK Medical Research Council (Unit Programme Number U105260566) and by PHE. She received funding under the NIHR grant for 10% of her time. The rest of her salary was provided by the MRC and PHE jointly.

Forecasting the 2017/2018 seasonal influenza epidemic in England using multiple dynamic transmission models: a case study

“…This is a SEEIIR model, a modified version of the model developed to reconstruct the 2009 H1N1 influenza pandemic [11,21,22]. The model uses daily (or weekly) data on the number of GP consultations for ILI from the PHE influenza surveillance dataset augmented with the RCGP's virological data to estimate the component of ILI due to influenza.…”

“…In the UK, the National Risk Register of Civil Emergencies lists an outbreak of pandemic influenza as the greatest risk faced by its population [8]. Before, during and after 2009 A/H1N1pdm, quantitative approaches for real-time modelling and forecasting burden have been developed [9][10][11]. The availability of these models, together with complementary surveillance and data collection systems including sero-epidemiology for seasonal influenza, provided the opportunity to address the challenge of predicting seasonal influenza activity in England.…”

The UK's pandemic influenza research portfolio: a model for future research on emerging infections

“…The Real-time modelling of a pandemic influenza outbreak (RTM) study has been funded to advance the state of the art of real-time modelling of influenza epidemics and to provide a tool to monitor and predict the development of an ongoing pandemic outbreak in the UK. Outputs from the RTM study include: models to produce age and region-specific epidemic forecasts [21]; algorithms, building on the latest developments in statistical computation, to allow epidemic analyses to be updated in a timely fashion as the epidemic unfolds [22]; bespoke software and relevant training of Public Health England (PHE) staff.…”

Forecasting the 2017/2018 seasonal influenza epidemic in England using multiple dynamic transmission models: a Case Study

“…Stratified Primary Care Model This is a SEEIIR model, a modified version of the model developed to reconstruct the 2009 H1N1 influenza pandemic. [11,21,22] The model uses daily (or weekly) data on the number of GP consultations for ILI from the PHE influenza surveillance dataset augmented with the RCGP's virological data to estimate the component of ILI due to influenza. In the first week of analysis, the number of ILI consultations due to influenza were simply estimated by multiplying the total number of consultations in sentinel RCGP practices by the proportion of swabs testing positive for an influenza virus, obtaining what we term ILI+.…”

“…[8] Before, during and after 2009 A/H1N1pdm, quantitative approaches for real-time modelling and forecasting burden have been developed. [9,10,11] The availability of these models, together with complementary surveillance and data collection systems including sero-epidemiology for seasonal influenza, provided the opportunity to address the challenge of predicting seasonal influenza activity in the UK. This became a pressing need, when, following a particularly intense 2017 influenza season in Australia, [12] prior to the winter season 2017/8 the National Health Service (NHS) put in place winter preparedness plans to manage potential acute pressures on the health service.…”