Professional Documents
Culture Documents
Artificial Intelligence in Public Health: The Potential of Epidemic Early Warning Systems
Artificial Intelligence in Public Health: The Potential of Epidemic Early Warning Systems
Artificial Intelligence in Public Health: The Potential of Epidemic Early Warning Systems
warning systems
Abstract
The use of artificial intelligence (AI) to generate automated early warnings in epidemic surveillance
by harnessing vast open-source data with minimal human intervention has the potential to be both
revolutionary and highly sustainable. AI can overcome the challenges faced by weak health systems
by detecting epidemic signals much earlier than traditional surveillance. AI-based digital surveillance
is an adjunct to—not a replacement of—traditional surveillance and can trigger early investigation,
diagnostics and responses at the regional level. This narrative review focuses on the role of AI in
epidemic surveillance and summarises several current epidemic intelligence systems including
ProMED-mail, HealthMap, Epidemic Intelligence from Open Sources, BlueDot, Metabiota, the
Global Biosurveillance Portal, Epitweetr and EPIWATCH. Not all of these systems are AI-based,
and some are only accessible to paid users. Most systems have large volumes of unfiltered data; only
a few can sort and filter data to provide users with curated intelligence. However, uptake of these
systems by public health authorities, who have been slower to embrace AI than their clinical
1
Biosecurity Program, The Kirby Institute, Faculty of 5
Medicine, University of New South Wales, Sydney, School of Population Health, Faculty of Medicine,
Australia University of New South Wales, Sydney, Australia
6
2
College of Public Service & Community Solutions, William Harvey Research Institute, Queen Mary
Arizona State University, Tempe, United States University of London, United Kingdom
3
School of Civil and Environmental Engineering, University Corresponding author:
of New South Wales, Sydney, Australia Xin Chen, Biosecurity Program, The Kirby Institute,
4
School of Computer Science and Engineering, Faulty of Faculty of Medicine, University of New South Wales,
Engineering, University of New South Wales, Sydney, Sydney, NSW 2052, Australia.
Australia Email: jessie@epiwatch.org
Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative
Commons Attribution-NonCommercial 4.0 License (https://creativecommons.org/licenses/by-nc/4.0/) which permits
non-commercial use, reproduction and distribution of the work without further permission provided the original work is attributed
as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/open-access-at-sage).
2 Journal of International Medical Research
counterparts, is low. The widespread adoption of digital open-source surveillance and AI technol-
ogy is needed for the prevention of serious epidemics.
Keywords
Artificial intelligence, public health, epidemic intelligence, pandemic, early warning system, digital
surveillance
Date received: 23 June 2022; accepted: 3 February 2023
serious epidemics. Open-source systems gen- surveillance and the censorship of outbreak
erate large quantities of unfiltered data with reporting.3
uncertain meaning and may overwhelm users
or lead to misleading conclusions. AI can be
Early detection through AI-based
used to curate, filter and decipher such data
to provide more valid early warning signals. surveillance and monitoring
AI technology can also predict spread at a A vast array of uncurated, open-source
granular scale, guiding data-driven early data including social media and news
local responses that can be critical in the ini- reports capture the concerns and discus-
tial stages of a pandemic.6 Given the non- sions of the community. If mined regularly,
linear complex spread dynamics and the these data can provide early signals of epi-
uncertainty inherent in early epidemic evolu- demics before official detection by health
tion, complex dynamic AI-based modelling authorities.3 Public health has been slow
frameworks such as multi-agent models can to adopt AI technology and utilise AI capa-
help temporally and geospatially simulate the bility to enable rapid and early detection of
evolution of epidemics, allowing targeted and epidemic signals. AI can further address the
effective public health responses.6 In addition, issue of data censorship and the challenges
these frameworks can be used to identify the faced by weak healthcare systems in low-
most effective interventions and their impor- income countries with limited human resour-
tance in limiting spread. Finally, AI technol- ces for public health surveillance. Figure 1
ogy can overcome the challenges faced by illustrates the genesis and exponential
weak health systems and issues such as poor growth of epidemics and the importance of
Figure 1. Enhancement of epidemic detection by rapid epidemic intelligence and risk analysis.
4 Journal of International Medical Research
early detection to prevent global spread. The cases, making rapid intervention critical to
early detection of epidemics affords the mitigating spread.9 The Ebola epidemic
best prospect of preventing global spread. could have been detected in late 2013,
Currently, epidemics are detected using tra- 3 months before the World Health
ditional surveillance methods that rely on Organisation (WHO) was informed, using
disease surveillance data formally reported rapid social media-based intelligence—even
by healthcare workers and laboratories. in Guinea, which has relatively low smart-
Given the time lag between the onset of a phone penetration.10 Similarly, a retrospec-
symptom and the formal laboratory confir- tive study using open-source data on Weibo
mation report, traditional surveillance meth- detected a signal of unknown severe pneu-
ods are not rapid enough to allow for the monia in Hubei Province in mid-November
early detection of serious epidemics.3 2019, a month before the COVID-19 out-
The COVID-19 pandemic is an example: break was officially reported.7
the SARS-CoV-2 virus had already spread The most widely used outbreak alert
worldwide before the virus was first system is the Program for Monitoring
reported. The first COVID-19 cases present- Emerging Diseases (ProMED-mail), a qual-
ing severe pneumonia of unknown origin itative reporting system to which clinicians
were officially reported in China on approx- report unusual outbreaks.11 This system
imately 8 December 2019. However, a retro- relies on health professionals notifying
spective study using open-source intelligence
moderators of unusual outbreaks. While
data identified another COVID-19 case in
the system has improved the speed of tradi-
China in mid-November 2019.7 In Spain,
tional health system surveillance and has
the first COVID-19 case was officially
been the first to detect many important epi-
reported on 25 February 2020; however,
demics, it is still largely dependent on
the virus was detected in sewage water in
human reporting and does not harness the
Barcelona 41 days before that date.5 The
full capabilities of open-source data and AI.
COVID-19 pandemic demonstrates the crit-
ical need for the early detection of epidemics The ideal system must harness and process
using AI technology in public health.5 multiple sources of unstructured data and
Currently, investment in epidemic prepared- display data in a curated, filtered and struc-
ness predominantly supports the develop- tured format that can inform rapid public
ment of drugs and vaccines. Although health action.3,12
essential, drugs and vaccines tend to become Another system used for outbreak alerts is
available a considerable time after a serious Google Flu Trends, which was in use from
new infection has emerged and spread 2008 to 2015 but was terminated because
widely.8 Figure 1 highlights currently missing of errors in estimation.13 Another tool,
system capability—rapid epidemic intelli- DEFENDER,14 was an outbreak detection,
gence, recognition and risk analysis—and surveillance, forecasting and nowcasting
the potential gains of AI-based epidemic system developed as a research tool by the
intelligence. United Kingdom. The system integrated geo-
As demonstrated during the 2020 coded symptom data from Twitter with news
COVID-19 pandemic and the 2014 Ebola reports and used this information for out-
epidemic, non-pharmaceutical interventions break detection, situational awareness
are critical during epidemics, especially and nowcasting.14 DEFENDER was not a
when drugs or vaccines are unavailable.9 public system and is not currently in use.
Epidemic growth was exponential in both This review focuses on systems in current use.
MacIntyre et al. 5
1 Epidemic events under Unusual health events in a A broad range of infec- A broad range of infec- A broad range of A broad range of New or undefined infec- Default setting is the New or undefined
surveillance broad range of infectious tious diseases tious diseases57 infectious diseases56,58 infectious diseases59 tious diseases topic of infectious infectious diseases
diseases56 A broad range of All main syndromes Began as a transport net- Syndromic surveillance60 diseases A broad range of
Clinical syndromes56 syndromes32 included, and is work modelling tool, with COVID-19,44MERS- infectious diseases
customisable open-source intelligence CoV61 A broad range of
added more recently. Other epidemics: not syndromes
mentioned
2 Types of automated or Health professionals WHO >12000 web sources Official health notifica- >240 data sources United States Tweets from 9 curated sources (e.g.,
primary sources Formal (e.g., official ProMED mail (WHO, news channels, tions (e.g., WHO) including official and Department of Defense, Twitter45 WHO, CDC, Outbreak
Validated health government websites) News media: Google news feeds, others)63 Health forums (e.g., unofficial sources59 other government agen- News Today)
sources (e.g., Informal (media News, Yahoo News, 240 twitter accounts63 ProMED mail) cies, industry, CDC, Non-curated sources
WHO, CDC, reports)62 Factiva, LexisNexis— Facebook: Sprinklr, News articles intelligence community42 (Google News using over
health blogs, etc.27 ministry of health sites Travel data 4,000 search terms)
professions) Remote sensing data64
Unvalidated web
sources
Social media (e.g.,
Twitter)
58
3 Number of languages of At least 6 languages: 15 languages 65 80 languages63 65 languages Not mentioned Not mentioned >20 languages 42 languages
information sources English, French, Spanish,
Portuguese, Russian and
Arabic21
4 Number of languages in 6 languages: 7 languages: One language: English Not mentioned One language: Not mentioned One language: One language: English
which dashboard is English, French, Spanish, English, Chinese, Spanish, English English
published Portuguese, Russian and Russian, French,
Arabic21 Portuguese and Arabic32
5 Back translation of Posts are in both original Reports are in 7 lan- Reports are in original Not mentioned Reports: brief infor- Not mentioned Tweets are in original Reports are in both
reports from original language and English guages: English, Chinese, language mation is provided in languages original language and
language into English translation56 Spanish, Russian, French, GPHIN content is English only English translation
Portuguese and Arabic32 translated
6 Frequency of item Near real- time21 Hourly66 Near real-time (check Not mentioned Near real- time39 Near real- time44 Daily Curated sources: near
collection web sources every real-time
5 minutes)33 Non-curated sources
(e.g., Google News): daily
7 Filter out irrelevant No automatic filtration21 Automatic categorisation, Automatically filtering of Use of AI for automated Use of AI for auto- Not mentioned No automatic Automatic filtration using
items filtration and integration articles for reports using filtering58 mated filtering68 filtration AI: priority algorithm
of reports with 84% AI: news article credibility improved to 88% using
accuracy27 algorithm67 machine learning
Filtering out duplicates, Human moderation: ana-
named entity recognition, lysts review all filtered
categorisation and tag- articles to send the rele-
ging/filtering out noise vant reports
(483 categories)
(continued)
Table 1. Continued.
Epidemic Intelligence from Metabiota Epidemic Global Biosurveillance
ID Parameter ProMED Mail HealthMap Open Sources (EIOS) BlueDot Tracker Portal Epitweetr EPIWATCH
Commencement year 1994 2006 2017 2013 2008 2014 2018 2016
8 Time, location, disease Minimum information Include 1) publication For COVID-19, the mini- Not mentioned Includes 1) time Not mentioned Includes 1) tweet Minimum information
information and link includes 1) time (publica- time, 2) event location, mum information includes (publication time or time, 2) event loca- includes 1) time (publica-
to original source tion time or event time), 3) disease/syndrome, 1) time (publication time event time), 2) out- tion and 3) keywords tion time or event time),
provided in all 2) outbreak location, 4) link to original article32 or event time), 2) event break location and (diseases/syn- 2) outbreak location,
reports 3) diseases/syndromes location, 3) disease: 3) diseases/ dromes) set by users 3) diseases/syndromes
and 4) link to original COVID35 and 4) link to syndrome and 4) link to original
article56 original article article
9 Mapping of reported With mapping, linked With mapping, linked With mapping, linked With mapping70 Mapping with linked Mapping44 without With mapping, no With mapping, linked
events reports and no search or reports and searchable reports and searchable Searchable function not reports, without searchable function linked tweets and, reports, and searchable
filter function56 function69 function35 mentioned searchable function39 Linked reports not no search or filter function
Linked reports not mentioned function
mentioned
10 Range of data analytics No analytics56 Number of reports over Number of COVID-19 Number of cases/deaths
Regions with low to MERS-CoV: number of Number of tweets Number of reports over
time by disease and cases for global and over time by disease and
high number of epi- cases and deaths in over time by key- time by disease syndrome
syndrome, location selected countries36 location72 demics39 South Korea61 word and location45 and location
(timeseries)69 Other epidemics-number Risk assessment Other epidemics: not Risk analysis tools:
of reports over time (produces epidemic specified EPIRISK, FLUCAST
Risk analysis tool: risk models for Predictive modelling
INFORM risk index34,71 insurance, commer- using AI45
cial and government
sectors)61,73
11 Public data access Free to public, no down- Publicly accessible COVID-19 dashboard and No public access Website (map): No public access Free to public Website: free access to
loadable dataset webpage: online view of Appears to be a paid ser- online view of the recent 1-month data,
the latest reports only35 vice for selected clients latest reports only39 online view only
Other epidemics: no System: no public System: guest user regis-
public access access tration for free, access to
recent 6 months dataset
(Excel), provides both
online view and down-
loadable data
12 Ability to customise No Not mentioned A generic event manage- Infectious disease Not mentioned Can integrate additional Data collection can The system, data collec-
ment system module67 surveillance can be software applications44 be customised by tion and analysis can be
that can be customised customised70 setting up keywords, customised based on
according to needs will be location and time45 requests (e.g., radiation)
released to EIOS
communities
The system can be
customised according
to user needs
(continued)
Table 1. Continued.
Epidemic Intelligence from Metabiota Epidemic Global Biosurveillance
ID Parameter ProMED Mail HealthMap Open Sources (EIOS) BlueDot Tracker Portal Epitweetr EPIWATCH
Commencement year 1994 2006 2017 2013 2008 2014 2018 2016
13 Number of publications >1074 >1024,28,30,75–82 >1017,34,83–92 >10 <10 (7 papers)68,93– <10 (7 papers)42,60,69, <10 (3 >10104
98 99–102
to validate system papers)46,56,103
14 Timeliness of detecting ProMED identified the HealthMap detected the The EIOS system BlueDot identified Not mentioned Not mentioned Not mentioned EPIWATCH detected
early signals of early signal of the unidentified pneumonic detected the first article undiagnosed pneumonia early signals of COVID-19
COVID-19 COVID-19 outbreak on cases in Wuhan Province reporting on a pneumonia in Wuhan Province on the on the day of the WHO
30 December 2019,105 on 30 December 30 cluster in Wuhan same day that the out- announcement.
one day before the 2019,106 one day before Province on the day that break was declared by the
Chinese government the Chinese government the outbreak was WHO.38
officially acknowledged officially acknowledged declared by the WHO.32
the outbreak. the outbreak.
15 Web and Mobile App Web App developed56 Web App developed69 Web App: COVID-19 Web App (not men- Web APP Web App developed44 No Web App Web App developed
development Mobile App developed107 Mobile App developed web app accessible to tioned) developed39 No Mobile App No Mobile App Mobile App in develop-
(i.e., Outbreaks Near public.35,36 Other epi- No Mobile App No Mobile App ment, will be available in
Me)108 demics web app internally app store.
accessible
Mobile App: not
mentioned
WHO, World Health Organization; CDC, United States Centers for Disease Control; GPHIN, Global Public Health Intelligence Network; app, application.
MacIntyre et al. 11
temporally. Processing such multi- account for lags between intervention and
dimensional (i.e., geospatiotemporal) data response. Support vector machine models
for analysis and forecasting requires and transformers have also been used to
machine learning approaches that can utilise flexibly model the impact of interventions
these features to develop prediction models on pandemic growth globally; these systems
without losing salient information that car- have identified the most effective interven-
ries important signals. Convolutional neural tions employed during the SARS-CoV-2
networks,109 transfer learning, support pandemic.
vector machines, random forest,110 deep Agent-based simulation models can fur-
learning and gradient boosting machine ther provide a flexible alternative to conven-
learning have been applied with high accu- tionally used susceptible-infectious-removed
racy to these challenges in different contexts. models to model geospatial dynamics and
Used in research, these models typically uti- spread. These models can create synthetic
lise regional data on past outbreaks, envi- populations and use available granular
ronmental factors, travel data, social data on geospatial context, contact rates,
factors, vector distribution and satellite behaviour, mobility and infrastructure to
meteorological data (e.g., temperature and model the spread and impact of interven-
rainfall). These data can be highly predictive tions at a fine scale.
of the occurrence and timing of regional out-
breaks, providing a framework for early pre- The potential of AI in public
paredness and response. None of the
available epidemic intelligence systems has
health
automated capability for geospatial risk The potential of AI in public health is illus-
prediction. trated through the development and use of
the epidemic intelligence systems described
in this review. Nevertheless, AI is not
Modelling of interventions and
widely implemented at an operational level
response in the everyday practice of public health
Early warning systems can be enhanced by compared with the use of AI in clinical
modelling pandemic growth and the effec- medicine. By generating early epidemic
tiveness of interventions. This requires warnings even in low-resource settings or
modelling complex dynamic systems with in areas in which data are censored by gov-
non-linearities that can be applied to time- ernments, AI can be revolutionary. AI ena-
series data with lags between interventions bles early identification and intervention,
and responses. A variety of AI-based allowing the early management of newly
approaches have been applied to develop emergent epidemics to feasibly result in
frameworks for these data and can be auto- eradication. When added to late-stage inter-
mated, customised and added to early warning ventions such as diagnostics, drugs and vac-
systems. These approaches include long- cines, AI can considerably improve health
short term memory networks that are ideal security and the prospect of preventing pan-
for modelling temporal data trends and can demics. AI can be used to identify not only
be trained to retain memory for features that specific diseases but also clinical syndromes
are important for prediction at a given point that may predict new and emerging
in time whilst ‘forgetting’ features that infections. Innovations in novel digital syn-
are unimportant.13 These networks retain dromic surveillance systems using open-source
‘memory’, as needed, over time and can data can support the early detection of serious
12 Journal of International Medical Research
emerging infectious epidemics. The key fea- monkeypox epidemic in non-endemic coun-
tures of an optimised AI system are: tries provided a test case for using learnings
from epidemic intelligence systems’ responses
1. Rapid intelligence drawn from open- to the COVID-19 pandemic. All systems
source data to generate higher-level and reported on monkeypox; however, we are
earlier epidemic alerts compared with tra- aware of only two special initiatives.
ditional surveillance without the need for HealthMap created a monkeypox dashboard
human reporting. These alerts can be fol- with daily updates of case counts,114 and
lowed up with formal investigation and EPIWATCH created a weekly summary of
traditional surveillance methods such as syndromic surveillance for rash and fever ill-
laboratory confirmation by public health nesses that could be monkeypox misdiag-
authorities. nosed as other illnesses.115
2. The capability to rapidly and globally The past decade has seen the emergence
identify key serious syndromes that of epidemics such as the novel zoonotic
may result from new emerging infections influenza,116,117 the Middle East respiratory
or biowarfare events. syndrome coronavirus, Ebola, Zika virus
3. The ability to address the issues of cen- and SARS-CoV-2.118 These events highlight
sorship of reporting and reliance on the increasing risk of emerging infectious dis-
human reporting and the challenges eases and the need for early warning signals.
faced by weak health systems. Strategies to adopt open-source early warning
4. The capability to predict in real time the systems and provide the source code for such
likelihood of serious outcomes of identi- systems would allow for collaborative design
fied events using a suite of decision sup- on a global scale. Harnessing the creative tal-
port tools (e.g., risk analysis, modelling ents of health and software engineering
and simulation), prioritise responses and experts working collaboratively in interdisci-
determine the urgency of intervention. plinary teams could support optimising
5. Tailored user interfaces on the Web, global early warning systems. The creation
mobile applications for real-time decision of tools with free availability and user inter-
support and tools that can be adapted for faces in all major international languages can
use in health and defence across govern- increase access—including for the community
ment and non-government sectors that and local health authorities—to open-source
require early warning and intelligence on intelligence.
serious epidemics.
36. World Health Organisation (WHO). an acute disease event. Epidemiology 2020;
COVID-19 case and death count dash- 31: 90–97.
board 2022 [updated February 10, 2022; 48. Lan Z, Chen M, Goodman S, et al.
cited 2022 February 10]. Available from: ALBERT: A Lite BERT for Self-supervised
https://portal.who.int/report/eios-covid19- Learning of Language Representations.
counts/. arXiv:190911942 [cs] 2020.
37. World Health Organization (WHO). 49. Devlin J, Chang MW, Lee K, et al.
Coronavirus news as reported through BERT: Pre-training of Deep Bidirectional
publicly available websites: Epidemic Transformers for Language Understanding.
Intelligence of Open Sources (EIOS); 2020 arXiv:181004805 [cs] 2018.
[updated 30 Jan 2022. Available from: 50. Tay Y, Dehghani M, Bahri D, et al.
https://portal.who.int/eios-coronavirus- Efficient Transformers: A Survey. ACM
newsmap/. Comput Surv 2022; 55: 1–28.
38. BlueDot 2022 [updated Jan 2022. Available 51. Rasmy L, Xiang Y, Xie Z, et al. Med-
from: https://bluedot.global/. BERT: pretrained contextualized embed-
39. Metabiota. Metabiota Epidemic Tracker dings on large-scale structured electronic
2017 [updated 30 Jan 2022. Available health records for disease prediction. NPJ
from: https://www.metabiota.com/epidem Digit Med 2021; 4: 1–13.
ic-tracker. 52. Moa A, Muscatello D, Chughtai A, et al.
40. Singer AW. The Evolution of Parametric Flucast: A Real-Time Tool to Predict
Insurance. Risk Management 2019; 66: Severity of an Influenza Season. JMIR
32–36. Public Health Surveill 2019; 5: e11780.
41. Metabiota 2017 [updated Jan 2022. 53. Lesmanawati DA, Veenstra P, Moa A,
Available from: https://metabiota.com/. et al. A rapid risk analysis tool to prioritise
42. Anderson JA, Rosenzweig CN, Roos J, response to infectious disease outbreaks.
et al. The Global Biosurveillance Portal: BMJ Global Health 2020; 5: e002327.
Biosurveillance for the Department of 54. Chen X, Chughtai AA and MacIntyre CR.
Defense. Online J Public Health Inform Recalibration of the Grunow–Finke
2015; 7: e63. Assessment Tool to Improve Performance
43. Ruchkin I, Dwivedi V, Garlan D, et al. in Detecting Unnatural Epidemics. Risk
Architectural Modeling of Ozone Widget Anal 2019; 39: 1465–1475.
Framework End-User Compositions 55. German RR, Lee LM, Horan JM, et al.
(CMU-ISR-14-108). 2014. Updated guidelines for evaluating public
44. JPL CBRN Information Management/ health surveillance systems: recommenda-
Information Technology. Global tions from the Guidelines Working
Biosurveillance Portal [cited 2022 Group. MMWR Recomm Rep 2001; 50:
February 12]. Available from: https:// 1–35; quiz CE1–7.
www.jpeocbrnd.osd.mil/Portals/90/jpm- 56. Trotta A, Marinaro M, Cavalli A, et al.
fact-sheet-gbsp-final.pdf. African Swine Fever—How to Unravel
45. European Centre for Disease Prevention Fake News in Veterinary Medicine. Animals
and Control. Epitweetr tool 2020 [updated (Basel) 2022; 12: 656.
October 1, 2020; cited 2021 October 28]. 57. EIOS Newsletter July 2021 2021 [cited 2022
Available from: https://www.ecdc.europa. April 12]. Available from: https://cdn.who.
eu/en/publications-data/epitweetr-tool. int/media/docs/default-source/eios/eios-jul
46. Espinosa L, Wijermans A, Orchard F, et al. y-2021-newsletter.pdf?sfvrsn=16a5cb51_3
Epitweetr: Early warning of public health &download=true.
threats using Twitter data. Euro Surveill 58. AWS. Powered by AWS, BlueDot Uses
2021; 27: 2200177. Available at SSRN Machine Learning to Detect and Respond
3811673. to Infectious Disease Risks 2020 [cited 2022
47. Joshi A, Sparks R, McHugh J, et al. April 12]. Available from: https://aws.ama
Harnessing tweets for early detection of zon.com/solutions/case-studies/bluedot/.
16 Journal of International Medical Research
78. Hossain N and Househ MS, editors. Using COVID-19. Nature Machine Intelligence
HealthMap to Analyse Middle East 2020; 2: 295–297.
Respiratory Syndrome (MERS) Data. 89. Laxmi Lydia E, Moses Gummadi J, Ranjan
ICIMTH; 2016. Pattanaik C, et al, editors. Interdependence in
79. Chen H, Zeng D and Yan P. HealthMap. Artificial Intelligence to Empower Worldwide
Infectious Disease Informatics: Springer; COVID-19 Sensitivity. International
2010; 183–186. Conference on Communication, Computing
80. Brownstein JS, Freifeld CC, Reis BY, et al. and Electronics Systems; 2021: Springer.
Surveillance Sans Frontieres: Internet- 90. Impouma B, Wolfe CM, Mboussou F,
based emerging infectious disease intelli- et al. Use of electronic tools for evidence-
gence and the HealthMap project. PLoS based preparedness and response to the
Med 2008; 5: e151. COVID-19 pandemic in the WHO
81. Brownstein JS and Freifeld C. HealthMap: African region. Lancet Digit Health 2020;
the development of automated real-time 2: e500–e502.
internet surveillance for epidemic intelli- 91. Fanelli A, Awada L, Caceres-Soto P, et al.
gence. Weekly releases (1997–2007) 2007; Sensitivity of an international notification
12: 3322. system for wildlife diseases: A case study
82. Thomas MC, Kite-Powell A, Atrubin D, using the OIE-WAHIS data on tularemia.
et al. Evaluating the utility of HealthMap Zoonoses Public Health 2022; 69: 286–294.
as a supplementary surveillance tool. 92. De Lusignan S, Liyanage H, McGagh D,
Online J Public Health Inform 2015; 7. et al. COVID-19 surveillance in a primary
83. Williams GS, Impouma B, Mboussou F, care sentinel network: in-pandemic devel-
et al. Implementing epidemic intelligence opment of an application ontology. JMIR
in the WHO African region for early detec- Public Health Surveill 2020; 6: e21434.
tion and response to acute public health 93. Badker R, Miller K, Pardee C, et al.
events. Epidemiol Infect 2021; 149: e261. Challenges in reported COVID-19 data:
84. Wark W. Building a better global health best practices and recommendations for
security early-warning system post-COVID: future epidemics. BMJ Glob Health 2021;
The view from Canada. International 6: e005542.
Journal 2021; 76: 55–67. 94. Gallivan M, Oppenheim B and Madhav
85. Ng V, Rees EE, Niu J, et al. Artificial intel- NK. Using social media to estimate Zika’s
ligence in public health: Application of nat- impact on tourism:# babymoon, 2014–
ural language processing algorithms for 2017. PLoS ONE 2019; 14: e0212507.
extracting information from news articles 95. Allam Z. The rise of machine intelligence in
in event-based surveillance. Can Commun the COVID-19 pandemic and its impact on
Dis Rep 2020; 46: 186–191. health policy. Surveying the COVID-19
86. Mercier A. An integrated approach to epi- Pandemic its Implications 2020: 89.
demic intelligence: the importance of pluri- 96. Alhasan M and Hasaneen M. Digital imaging,
disciplinary collaborations in the framework technologies and artificial intelligence applica-
of the development of methods to collect tions during COVID-19 pandemic. Comput
and analyze health data: Universite Paris- Med Imaging Graph 2021; 91: 101933.
Est; 2020. 97. Jamison DT, Lau LJ, Wu KB, et al.
87. Mboussou F, Ndumbi P, Ngom R, et al. Country performance against COVID-19:
Infectious disease outbreaks in the rankings for 35 countries. BMJ Glob
African region: overview of events reported Health 2020; 5: e003047.
to the World Health Organization in 2018. 98. Meadows AJ, Oppenheim B, Guerrero J,
Epidemiol Infect 2019; 147: e299. et al. Estimating infectious disease under-
88. Luengo-Oroz M, Hoffmann Pham K, reporting at the country level: a model and
Bullock J, et al. Artificial intelligence coop- application to the COVID-19 pandemic.
eration to support the global response to Available at SSRN 3706059. 2020.
18 Journal of International Medical Research
99. Cho HW and Chu C. A Joint Exercise including outbreak predictions based on
against Intentional Biothreats. Osong machine-learning. Geospat Health 2019; 14.
Public Health Res Perspect 2018; 9: 1–2. 110. Ong J, Liu X, Rajarethinam J, et al.
100. Toffoli CA. Optimizing Mission-Specific Mapping dengue risk in Singapore using
Medical Threat Readiness and Preventive Random Forest. PLoS Negl Trop Dis
Medicine for Service Members. US Army 2018; 12: e0006587.
Med Dep J 2018; 49–54. 111. Center for Disease Control and Prevention
101. Rhee C, Burkom H, Yoon C, et al. Military (CDC). CDC Stands Up New Disease
Real-time Syndromic Surveillance System Forecasting Center 2021 [updated August
for Biosurveillance Portal in Korea. 18, 2021; cited 2022 February 10].
Online J Public Health Inform 2015; 7: e48. Available from: https://www.cdc.gov/
102. Tak S, Jareb A, Choi S, et al. Enhancing media/releases/2021/p0818-disease-forecast
‘whole-of-government’ response to biological ing-center.html.
events in Korea: able response 2014. Osong 112. GOV.UK. PM announces plan for ‘Global
Public Health Res Perspect 2018; 9: 32–35. Pandemic Radar’ 2021 [updated 21 May
103. S€okmen S, Sevinc F and Yolal M. COVID- 2021; cited 2022 February 8]. Available
(Mis) Infodemic as a Nascent Concept for from: https://www.gov.uk/government/
Tourism and the Case of Epitweetr. news/pm-announces-plan-for-global-pan
Co-Editors 2021: 136. demic-radar.
104. EPIWATCH. Curated data at your finger- 113. World Health Organisation (WHO).
tips [cited 2022 April 12]. Available from:
WHO, Germany open Hub for Pandemic
https://www.epiwatch.org/publications.
and Epidemic Intelligence in Berlin 2021
105. ProMED. PRO/AH/EDR> Undiagnosed
[updated 1 September 2021 cited 2022
pneumonia – China (HU): RFI 2019
February 8]. Available from: https://www.
[cited 2022 April 19]. Available from:
who.int/news/item/01-09-2021-who-germa
https://scholar.harvard.edu/files/kleelerner/
ny-open-hub-for-pandemic-and-epidemic-
files/20191230_promed_-_undiagnosed_
intelligence-in-berlin.
pneumonia_-_china_hu-_rfi_archive_
114. Global. health Map. Monkeypox Line List
number-_20191230.6864153.pdf.
106. O’brien M and Larson C. Can AI flag dis- Cases [cited 2022 September 19]. Available
ease outbreaks faster than humans? Not from: https://map.monkeypox.global.
quite 2020 [cited 2023 February 1]. health/country.
Available from: https://apnews.com/arti 115. EPIWATCH. EPISCOPE-SEP 13 2022
cle/united-nations-us-news-ap-top-news- [cited 2022 September 19]. Available
international-news-virus-outbreak- from:5https://www.epiwatch.org/media/
100fbb228c958f98d4c755b133112582. EPISCOPE_SEPTEMBER_13_2022.pdf.
107. Apptricker. Download ProMED-mail App 116. Yang Y, Halloran ME, Sugimoto JD, et al.
for IOS & Android Phones, How to [cited Detecting human-to-human transmission
2022 April 12]. Available from: https://app of avian influenza A (H5N1). Emerg
tricker.in/download-promed-mail-app-for- Infect Dis 2007; 13: 1348–1353.
ios-android-phones-how-to/. 117. Muscatello DJ, Cretikos MA and
108. Chunara R, Andrews JR and Brownstein MacIntyre CR. All-cause mortality during
JS. Social and news media enable estima- first wave of pandemic (H1N1) 2009, New
tion of epidemiological patterns early in the South Wales, Australia, 2009. Emerg Infect
2010 Haitian cholera outbreak. Am J Trop Dis 2010; 16: 1396–1402.
Med Hyg 2012; 86: 39–45. 118. Bernstein AS, Ando AW, Loch-Temzelides
109. Anno S, Hara T, Kai H, et al. T, et al. The costs and benefits of primary
Spatiotemporal dengue fever hotspots asso- prevention of zoonotic pandemics. Sci Adv
ciated with climatic factors in Taiwan 2022; 8: eabl4183.