| Title:
|
Real-time monitoring of COVID-19 dynamics using automated trend fitting and anomaly detection |
| Abstract:
|
As several countries gradually release social distancing measures rapid detection of new localized COVID-19 hotspots and subsequent intervention will be key to avoiding large-scale resurgence of transmission. We introduce ASMODEE (automatic selection of models and outlier detection for epidemics) a new tool for detecting sudden changes in COVID-19 incidence. Our approach relies on automatically selecting the best (fitting or predicting) model from a range of user-defined time series models excluding the most recent data points to characterize the main trend in an incidence. We then derive prediction intervals and classify data points outside this interval as outliers which provides an objective criterion for identifying departures from previous trends. We also provide a method for selecting the optimal breakpoints used to define how many recent data points are to be excluded from the trend fitting procedure. The analysis of simulated COVID-19 outbreaks suggests ASMODEE compares favourably with a state-of-art outbreak-detection algorithm while being simpler and more flexible. As such our method could be of wider use for infectious disease surveillance. We illustrate ASMODEE using publicly available data of National Health Service (NHS) Pathways reporting potential COVID-19 cases in England at a fine spatial scale showing that the method would have enabled the early detection of the flare-ups in Leicester and Blackburn with Darwen two to three weeks before their respective lockdown. ASMODEE is implemented in the free R package trendbreaker. This article is part of the theme issue Modelling that shaped the early COVID-19 pandemic response in the UK. |
| Published:
|
2021-07-19 |
| Journal:
|
Philos Trans R Soc Lond B Biol Sci |
| DOI:
|
10.1098/rstb.2020.0266 |
| DOI_URL:
|
http://doi.org/10.1098/rstb.2020.0266 |
| Author Name:
|
Jombart Thibaut |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/jombart_thibaut |
| Author Name:
|
Ghozzi Stphane |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/ghozzi_stphane |
| Author Name:
|
Schumacher Dirk |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/schumacher_dirk |
| Author Name:
|
Taylor Timothy J |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/taylor_timothy_j |
| Author Name:
|
Leclerc Quentin J |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/leclerc_quentin_j |
| Author Name:
|
Jit Mark |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/jit_mark |
| Author Name:
|
Flasche Stefan |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/flasche_stefan |
| Author Name:
|
Greaves Felix |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/greaves_felix |
| Author Name:
|
Ward Tom |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/ward_tom |
| Author Name:
|
Eggo Rosalind M |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/eggo_rosalind_m |
| Author Name:
|
Nightingale Emily |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/nightingale_emily |
| Author Name:
|
Meakin Sophie |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/meakin_sophie |
| Author Name:
|
Brady Oliver J |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/brady_oliver_j |
| Author Name:
|
Medley Graham F |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/medley_graham_f |
| Author Name:
|
Hhle Michael |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/hhle_michael |
| Author Name:
|
Edmunds W John |
| Author link:
|
https://covid19-data.nist.gov/pid/rest/local/author/edmunds_w_john |
| sha:
|
273378e7640d7165ece32aecbb169b940e81df9b |
| license:
|
cc-by |
| license_url:
|
https://creativecommons.org/licenses/by/4.0/ |
| source_x:
|
Medline; PMC; WHO |
| source_x_url:
|
https://www.medline.com/https://www.ncbi.nlm.nih.gov/pubmed/https://www.who.int/ |
| pubmed_id:
|
34053271 |
| pubmed_id_url:
|
https://www.ncbi.nlm.nih.gov/pubmed/34053271 |
| pmcid:
|
PMC8165581 |
| pmcid_url:
|
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8165581 |
| url:
|
https://doi.org/10.1098/rstb.2020.0266
https://www.ncbi.nlm.nih.gov/pubmed/34053271/ |
| has_full_text:
|
TRUE |
| Keywords Extracted from Text Content:
|
UK
National Health Service
ASMODEE
breakpoints
COVID-19
Darwen
lockdown
Figure 4
Netlify
https://gitlab.com/stephaneghozzi/asmodeetrend
line
https://www.reconverse.org/
K − 1
[1] [2] [3]
farringtonFlexible
incidence2
human
pre-break
CCG Leicester City
NHS
lockdown
select_model
food-or
NHS 111/999
[9] [10] [11] [12]
Stan [34, 35] (
111-online
ASMODEE
people
SPI-M
root
Darwen
trendeval
departures/
UK
https://github.com/thibautjombart/nhs_pathways_
3-5
k.
[4] [5] [6]
CCGs
https://github.com/qleclerc/nhs_pathways_
SPI-M.
alpha
half-size
https://github.com/reconhub/trendbreaker
ASMO-DEE
−2 log(L
patient
reconverse
pointsource
brms::brm
Kivu/Ituri
CCG
http://repi demicsconsortium.org/
COVID-19 |
| Extracted Text Content in Record:
|
First 5000 Characters:One contribution of 21 to a theme issue 'Modelling that shaped the early COVID-19 pandemic response in the UK'.
Subject Areas: computational biology, health and disease and epidemiology, bioinformatics
As several countries gradually release social distancing measures, rapid detection of new localized COVID-19 hotspots and subsequent intervention will be key to avoiding large-scale resurgence of transmission. We introduce ASMODEE (automatic selection of models and outlier detection for epidemics), a new tool for detecting sudden changes in COVID-19 incidence. Our approach relies on automatically selecting the best (fitting or predicting) model from a range of user-defined time series models, excluding the most recent data points, to characterize the main trend in an incidence. We then derive prediction intervals and classify data points outside this interval as outliers, which provides an objective criterion for identifying departures from previous trends. We also provide a method for selecting the optimal breakpoints, used to define how many recent data points are to be excluded from the trend fitting procedure. The analysis of simulated COVID-19 outbreaks suggests ASMODEE compares favourably with a state-of-art outbreak-detection algorithm while being simpler and more flexible. As such, our method could be of wider use for infectious disease surveillance. We illustrate ASMODEE using publicly available data of National Health Service (NHS) Pathways reporting potential COVID-19 cases in England at a fine spatial scale, showing that the method would have enabled the early detection of the flare-ups in Leicester and Blackburn with Darwen, two to three weeks before their respective lockdown. ASMODEE is implemented in the free R package trendbreaker.
This article is part of the theme issue 'Modelling that shaped the early COVID-19 pandemic response in the UK'.
After a fast initial spread worldwide and large-scale epidemics in many affected countries, the trajectory of the COVID-19 pandemic is changing. In a number of severely affected countries, strong mitigation measures such as various forms of social distancing have slowed national epidemics and in many cases brought the epidemic close to control [1] [2] [3] . However, in the absence of widespread, long-lasting immunity through vaccination or natural infection [4] [5] [6] , these respites are most likely temporary, and further relapses, in the form of localized outbreaks or nation-wide resurgence, remain highly likely and a very serious threat.
In the UK, a 'lockdown' was implemented on 23 March 2020, and gradually relaxed from the beginning of June 2020, by which point about 300 000 confirmed COVID-19 cases and 40 000 deaths had been reported [7] . Unfortunately, the risk of local flare-ups was illustrated soon after, as increased case incidence in Leicester resulted in the city being put under lockdown again on 29 June 2020 [8] . Similarly, increased restrictions were imposed in Blackburn on 9 August 2020.
In order to prevent large-scale relapses, localized COVID-19 hotspots (i.e. places with high levels of transmission) need to be detected as soon as cases occur and contained as early as possible. For such detection to be optimal, COVID-19 dynamics need to be monitored at a small spatial scale, requiring daily surveillance of multiple time series of case incidence, and prompt detection of ongoing increases. Disease surveillance algorithms have been designed for such purposes [9] [10] [11] [12] , although many of them are tailored to detecting either seasonal or point-source outbreaks and may be most effective when trained on years of weekly incidence data (e.g. Farrington algorithm [13, 14] ). However, careful implementation of such algorithms has proved useful as a backbone for setting up automated disease surveillance systems for endemic diseases [15] .
Here, we introduce ASMODEE (automated selection of models and outlier detection for epidemics), an algorithm for detecting ongoing changes in COVID-19 incidence patterns. In order to characterize potentially very different dynamics in case incidence across a large number of locations, our approach implements a flexible time series framework using a variety of models including linear regression, generalized linear models (GLMs) or Bayesian regression. ASMODEE first identifies past temporal trends using automated model selection, and then uses outlier detection inspired by classical Shewhart control-charts to signal recent anomalous data points.
We used simulations to evaluate the potential of ASMO-DEE for detecting changes in incidence patterns. COVID-19 incidence dynamics were simulated using a branchingprocess model with realistic estimates of the time-varying reproduction number (Rt) and serial interval, under four scenarios: steady state (Rt close to 1), relapse, lockdown and flare-up following low levels of transmission. For comparison, we also applied the modified Farrington algorithm [14] , a standa |
| Keywords Extracted from PMC Text:
|
[1–3]
asAIC=−2 log(L)+2P
TP/(TP + FP
people
0.8.— Flare-up
NHS 111/999
pre-break
111-online
TN/(TN
k.
line
respites
human
farringtonFlexible
COVID-19
ASMODEE
scenario.—
reconverse
−2 log(L
half-size
food-
alpha
select_model
1.3.— Lockdown
brms::brm
×
lockdown
Darwen
trendeval
https://www.reconverse.org/
F1 score,5
root
non-outliers
patient
FN);2
UK
FP);3
Stan [34,35
K − 1
Kivu/Ituri
Netlify
https://covid19-nhs-pathways-asmodee.netlify.app/.
CCGs
CCG
SPI-M.
SPI-M
TP/(TP
NHS
incidence2
https://github.com/reconhub/trendbreaker
TP
method.— |
| Extracted PMC Text Content in Record:
|
First 5000 Characters:After a fast initial spread worldwide and large-scale epidemics in many affected countries, the trajectory of the COVID-19 pandemic is changing. In a number of severely affected countries, strong mitigation measures such as various forms of social distancing have slowed national epidemics and in many cases brought the epidemic close to control [1–3]. However, in the absence of widespread, long-lasting immunity through vaccination or natural infection [4–6], these respites are most likely temporary, and further relapses, in the form of localized outbreaks or nation-wide resurgence, remain highly likely and a very serious threat.
In the UK, a 'lockdown' was implemented on 23 March 2020, and gradually relaxed from the beginning of June 2020, by which point about 300 000 confirmed COVID-19 cases and 40 000 deaths had been reported [7]. Unfortunately, the risk of local flare-ups was illustrated soon after, as increased case incidence in Leicester resulted in the city being put under lockdown again on 29 June 2020 [8]. Similarly, increased restrictions were imposed in Blackburn on 9 August 2020.
In order to prevent large-scale relapses, localized COVID-19 hotspots (i.e. places with high levels of transmission) need to be detected as soon as cases occur and contained as early as possible. For such detection to be optimal, COVID-19 dynamics need to be monitored at a small spatial scale, requiring daily surveillance of multiple time series of case incidence, and prompt detection of ongoing increases. Disease surveillance algorithms have been designed for such purposes [9–12], although many of them are tailored to detecting either seasonal or point-source outbreaks and may be most effective when trained on years of weekly incidence data (e.g. Farrington algorithm [13,14]). However, careful implementation of such algorithms has proved useful as a backbone for setting up automated disease surveillance systems for endemic diseases [15].
Here, we introduce ASMODEE (automated selection of models and outlier detection for epidemics), an algorithm for detecting ongoing changes in COVID-19 incidence patterns. In order to characterize potentially very different dynamics in case incidence across a large number of locations, our approach implements a flexible time series framework using a variety of models including linear regression, generalized linear models (GLMs) or Bayesian regression. ASMODEE first identifies past temporal trends using automated model selection, and then uses outlier detection inspired by classical Shewhart control-charts to signal recent anomalous data points.
We used simulations to evaluate the potential of ASMODEE for detecting changes in incidence patterns. COVID-19 incidence dynamics were simulated using a branching-process model with realistic estimates of the time-varying reproduction number (Rt) and serial interval, under four scenarios: steady state (Rt close to 1), relapse, lockdown and flare-up following low levels of transmission. For comparison, we also applied the modified Farrington algorithm [14], a standard method designed for the detection of point-source outbreaks and used in many public health institutions [9]. We computed a variety of scores such as the probability of detection, sensitivity and specificity for two configurations of ASMODEE.
We used our approach to design automated surveillance pipelines that monitor changes in potential COVID-19 cases reported through an online and telephone hotline used in England, the National Health Service (NHS) Pathways system, which includes calls made to 111/999 as well as reports made through the 111-online system. We conducted the analysis at the level of Clinical Commissioning Groups (CCGs), small area divisions used for healthcare management in England's NHS, with an average of 226 000 people each. One advantage of the NHS Pathways system is that reports occur with little delay, because no confirmatory diagnostic tests are involved. The downside is that this system suffers from the usual sensitivity and specificity issues of a syndromic surveillance system: some reported 'potential cases' will not be actual COVID-19 cases, and some actual cases will be unreported. We show that when applied to NHS 111/999 calls data, ASMODEE would have enabled the early detection of the flare-ups in Leicester and Blackburn with Darwen in June and July 2020, respectively. We propose that ASMODEE may be a useful, flexible complement to existing outbreak detection methods for designing disease surveillance pipelines, for COVID-19 and other diseases.
All source code necessary to run ASMODEE, analyse and visualize results, implement the NHS Pathways pipeline and reproduce the analyses is open and freely available under MIT license (see §2d).
ASMODEE is designed for detecting recent departures/aberrations from past temporal trends in univariate time series. The response variable typically represents case counts, but it can readily accommodate other response variab |
| PDF JSON Files:
|
document_parses/pdf_json/273378e7640d7165ece32aecbb169b940e81df9b.json |
| PMC JSON Files:
|
document_parses/pmc_json/PMC8165581.xml.json |
| G_ID:
|
real_time_monitoring_of_covid_19_dynamics_using_automated_trend_fitting_and_anomaly |
