AI Surveillance of Epidemics

Time is of the essence in preventing a disease outbreak. Prof Liu Jiming, Chair Professor of the Department of Computer Science at HKBU, and his team developed an active surveillance system with machine learning and data-driven modeling, which can predict the spread of disease and inform policy decisions.


Self Photos / Files - AI Surveillance of Epidemics 1

Prof Liu Jiming


Malaria is a mosquito-borne infectious disease that is widespread in Southeast Asia and Africa. According to the World Health Organization (WHO), in 2018, there were an estimated 228 million cases of malaria worldwide, which led to about 405,000 deaths. In response to the call of the WHO and the United Nations to fight malaria, the Government of China in 2010 set a target to eliminate the disease within its borders by 2020. With quite a few indigenous cases reported for several years, China turned its attention to cross-border infection, especially in provinces bordering countries in the Mekong region. “The last mile is the most difficult,” says Prof Liu. “If imported cases are not detected in a timely fashion, then local infections may rebound and malaria elimination by 2020 will not be accomplished.”


Prediction based on socio-economic and ecological factors 

In 2011, Prof Liu and his team joined hands with the National Institute of Parasitic Diseases (NIPD), a national organisation under the Chinese Center for Disease Control and Prevention (China CDC) to tackle malaria. Prof Liu says that understanding the malaria problem specific to different regions is the key to finding a solution. According to observations by the disease control specialists, most cases of infection in Yunnan are imported from Myanmar through cross-border activities, which makes it extremely difficult to trace the spread of the disease. The mountainous terrain, which also implies a lack of resources, doctors and disease control specialists, adds extra challenges to disease control.


Instead of tracing infected individuals, Prof Liu’s team takes a more proactive approach. They applied data-driven modeling to predict the spread of malaria based on available information. The reason Yunnan villagers cross the border is to work on the other side. Thus, their movement can actually be determined by socio-economic factors, i.e. their income status and distance from the border. On the other hand, the transmission of malaria is related to the life cycle of the mosquitoes that can harbour and transmit the disease. This is based on several ecological factors, including the amount of rainfall, temperature, and distance from water.


Self Photos / Files - AI Surveillance of Epidemics 2

Malaria is a mosquito-borne infectious disease


Self Photos / Files - AI Surveillance of Epidemics 3

Prof Liu’s team members go on field trips to understand the local malaria problem


Self Photos / Files - AI Surveillance of Epidemics 4

Yunnan's mountainous terrain poses special difficulties for resource allocation


Effective resource allocation is vital

By analysing these socio-economic and ecological factors via data- driven modeling, the malaria transmission risk of different villages can be predicted and ranked. Once this is done, disease control specialists can be deployed to villages considered high risk before an outbreak and nip it in the bud. Prof Liu says the so-called “active surveillance” is particularly effective in the mountainous areas of China near the Mekong region, since one single city in Yunnan is already five times the size of Hong Kong or even larger, and a city might consist of more than 200 villages. Fewer than 10 disease control specialists are stationed in such a vast area, so effective allocation of human resources is the key to fighting malaria.


For Prof Liu, the most difficult part was the beginning. They had to identify the key factors behind malaria transmission in certain areas. To do this, the team had to work closely with the disease control specialists. Team members also went on field trips to understand the problem. Through their efforts, the prediction tool they developed is over 90 percent accurate.


Self Photos / Files - AI Surveillance of Epidemics 5

Self Photos / Files - AI Surveillance of Epidemics 6

The data-driven model predicts and ranks the malaria transmission risk of different villages based on socio-economic and ecological factors


Future plans in the Mekong region

Prof Liu’s AI tools were deployed in Tengchong, a city in Yunnan with a long history of malaria. The city’s malaria elimination policy was changed as a result of the team’s research. In 2016, Tengchong was declared the first city in the China-Myanmar border to eliminate malaria, four years before the national target. NIPD and the China CDC then applied the same technology to tackle malaria in 20 cities along the border. Owing to the positive impact of the AI tools, Prof Liu received the Yunnan Health Science and Technology Award 2019. His team is continuing the joint research with NIPD in the hope of accomplishing the goal of ending the risk of malaria nationwide.


The next step for Prof Liu’s team is to extend the research project to other countries in the Greater Mekong subregion and Southeast Asia which are affected by malaria and lack resources, such as Cambodia and Laos. The research team has presented their tools and findings to the WHO and other Greater Mekong subregion countries at the International Workshop on AI-enabled Malaria Control and Prevention. The WHO recognises the usefulness of the AI tools in assessing the malaria transmission risk in a timely manner for anti- malaria resource allocation. It also supports malaria control officers and field practitioners to learn how to use the tools. Prof Liu has also received collaboration plans from national-level Department of Health in Cambodia, Bhutan, and Sri Lanka to help fight malaria.


Among countries in the Mekong region, Cambodia has the most serious problem of malaria because infection cases there have been found to be resistant to antimalarial drug artemisinin. Prof Liu’s team found that the key factors of malaria transmission may vary from region to region. For instance, the species of mosquito that carries malaria in Cambodia is different from the mosquitoes in Yunnan, and the forest, rather than bodies of water, is their natural habitat. Thus, different ecological factors should be used in the analysis to take into account these differences.


Self Photos / Files - AI Surveillance of Epidemics 7

The tools and findings are introduced to the WHO and other Greater Mekong subregion countries in an international workshop


Self Photos / Files - AI Surveillance of Epidemics 8

Prof Liu’s team analyses COVID-19 transmission patterns


Unveiling COVID-19 transmission patterns

In addition to his work on malaria, Prof Liu also contributed to the battle against the recent coronavirus (COVID-19) outbreak in mainland China. He led a new research on characterising and quantifying the underlying transmission patterns of COVID-19 among different populations using a data-driven modelling approach, in collaboration with NIPD and the Chinese Academy of Sciences. One key feature of the computational model is that it is able to estimate the number of infections per day. According to literature, COVID-19 has an average incubation period of six to seven days. Thus, the number of confirmed cases might not necessarily reflect the actual threat of the disease during that period of time. The computed infection numbers provide the disease control department a clearer picture of the situation. The estimate also aligns with the confirmed cases a few days later, which proves its accuracy.


To analyse the risk of transmission through interpersonal contact among different age groups, the computational model divided the population into seven age groups with their own specific social circles, gathering places and activity patterns. Four representative social contact settings, namely households, schools, workplaces, and public places, were considered. A contact matrix was inferred to describe the contact intensity between different age groups for each of the four settings, and this enabled the dynamics of disease transmission to be computed. The results revealed that age groups in public places and households are more scattered, which means it is easier for the virus to spread among different age groups in these places.


Self Photos / Files - AI Surveillance of Epidemics 9

Social-distancing and lockdown measures are implemented in many cities to contain the spread of COVID-19


Prospective analysis of work resumption plans

With the COVID-19 pandemic resulting in lockdowns in different parts of the world, there have been intense debates on when and how we can safely resume ‘normal’ life. The computational model was used to analyse six cities in mainland China—Wuhan, Beijing, Tianjin, Hangzhou, Suzhou and Shenzhen (located in three major economic zones in China)—on a case-by-case basis to estimate the disease transmission risk and the impact of different work resumption plans on local GDP growth. Confirmed cases, population sizes, and the intervention measures taken by each city were considered when building the model. The data-driven computational model provides a science-based analytical solution, enabling policymakers to design plans that can achieve both the containment of disease transmission risk and the safe, gradual reopening of affected cities.


The research was published in The Lancet's EClinicalMedicine journal. Prof Liu says their data-driven computational modelling and analytical tools have been openly shared with public health policymakers and researchers around the world. This will allow them to capitalise on the AI tools for decision making using their domestic epidemiological data and cope with the current situation.