A Bayesian nonparametric method for detecting rapid changes in disease transmission (preprint)

Whether an outbreak of infectious disease is likely to grow or dissipate is determined through the time-varying reproduction number, R t . Real-time or retrospective identification of changes in R t following the imposition or relaxation of interventions can thus contribute important evidence about disease transmission dynamics which can inform policymaking. Here, we present a method for estimating shifts in R t within a renewal model framework. Our method, which we call EpiCluster, is a Bayesian nonparametric model based on the Pitman-Yor process. We assume that R t is piecewise-constant, and the incidence data and priors determine when or whether R t should change and how many times it should do so throughout the series. We also introduce a prior which induces sparsity over the number of changepoints. Being Bayesian, our approach yields a measure of uncertainty in R t and its changepoints. EpiCluster is fast, straightforward to use, and we demonstrate that it provides automated detection of rapid changes in transmission, either in real-time or retrospectively, for synthetic data series where the R t profile is known. We illustrate the practical utility of our method by fitting it to case data of outbreaks of COVID-19 in Australia and Hong Kong, where it finds changepoints coinciding with the imposition of non-pharmaceutical interventions. Bayesian nonparametric methods, such as ours, allow the volume and complexity of the data to dictate the number of parameters required to approximate the process and should find wide application in epidemiology. Highlights Identifying periods of rapid change in transmission is important for devising strategies to control epidemics. We assume that the time-varying reproduction number, R t , is piecewise-constant and transmission is determined by a Poisson renewal model. We develop a Bayesian nonparametric method, called EpiCluster, which uses a Pitman Yor process to infer changepoints in R t . Using simulated incidence series, we demonstrate that our method is adept at inferring changepoints. Using real COVID-19 incidence series, we infer abrupt changes in transmission at times coinciding with the imposition of non-pharmaceutical interventions.

Using multiple sampling strategies to estimate SARS-CoV-2 epidemiological parameters from genomic sequencing data (preprint)

SARS-CoV-2 virus genomes are currently being sequenced at an unprecedented pace. The choice of viral sequences used in genetic and epidemiological analysis is important as it can induce biases that detract from the value of these rich datasets. This raises questions about how a set of sequences should be chosen for analysis, and which epidemiological parameters derived from genomic data are sensitive or robust to changes in sampling. We provide initial insights on these largely understudied problems using SARS-CoV-2 genomic sequences from Hong Kong and the Amazonas State, Brazil. We consider sampling schemes that select sequences uniformly, in proportion or reciprocally with case incidence and which simply use all available sequences (unsampled). We apply Birth-Death Skyline and Skygrowth methods to estimate the time-varying reproduction number (Rt) and growth rate (rt) under these strategies as well as related R0 and date of origin parameters. We compare these to estimates from case data derived from EpiFilter, which we use as a reference for assessing bias. We find that both Rt and rt are sensitive to changes in sampling whilst R0 and the date of origin are relatively robust. Moreover, we find that the unsampled datasets, which reflect an opportunistic sampling scheme, engender the most biased Rt and rt estimates for both our Hong Kong and Amazonas case studies. We highlight that sampling strategy choices may be an influential yet neglected component of sequencing analysis pipelines. More targeted attempts at genomic surveillance and epidemic analyses, particularly in resource-poor settings with limited sequencing capabilities, are necessary to maximise the informativeness of virus genomic datasets.

Global predictions of short- to medium-term COVID-19 transmission trends : a retrospective assessment (preprint)

From 8th March to 29th November 2020, we produced weekly estimates of SARS-CoV-2 transmissibility and forecasts of deaths due to COVID-19 for 81 countries with evidence of sustained transmission. We also developed a novel heuristic to combine weekly estimates of transmissibility to produce forecasts over a 4-week horizon. We evaluated the robustness of the forecasts using relative error, coverage probability, and comparisons with null models. During the 39-week period covered by this study, both the short- and medium-term forecasts captured well the epidemic trajectory across different waves of COVID-19 infections with small relative errors over the forecast horizon. The model was well calibrated with 56.3\% and 45.6\% of the observations lying in the 50\% Credible Interval in 1-week and 4-week ahead forecasts respectively. We could accurately characterise the overall phase of the epidemic up to 4-weeks ahead in 84.9\% of country-days. The medium-term forecasts can be used in conjunction with the short-term forecasts of COVID-19 mortality as a useful planning tool as countries continue to relax public health measures.