COVID-19: Here’s how Uganda Could use AI Technologies for Public Health

By Rahman Sanya

It’s now two months and a half (from January 30 , 2020) since the World Health Organization declared the current outbreak of novel Corona Virus Disease 2019 (covid-19) a public health emergency of international concern.

Uganda, like all affected countries, is working hard to keep the  effects and impact of the disease to a minimum.

As of today April 18 , 2020 Uganda has registered

55 confirmed cases of covid-19, with 20 recoveries and zero death.

However, even with this relatively small case count there is general uncertainty as to the number of undetected, potentially infected people circulating in the population since no systematic testing is being carried out due to lack of testing kits.

A number of public health interventions are currently being implemented to achieve the [set] goal of registering low infection rates and effective case management to optimize health outcomes for those infected.

In this non-exhaustive article I try to map out some opportunities and challenges for Uganda’s use of Artificial Intelligence (AI) technologies to ramp up its fight against covid-19 or any other public health emergency.

These ideas may be used to either strengthen existing interventions or adopted as new ones. Some of the ideas will obviously be futuristic, in that the country may not currently have the means to adopt them.

Nonetheless, they will serve as something to aspire for. I use the term AI technologies here to refer to a broad range of methodologies and technologies that may fall under the sub-fields of machine learning, computer vision, natural language processing, etc, or even the broader discipline of data science which encompasses them all.

A common definition of AI is, it is the use of computers to automate decision-making to perform tasks that normally require human intelligence.

On the other hand, machine learning (ML) is a sub-field of AI that develops algorithms that give computers the ability to learn and improve on a task with experience, without being explicitly programmed.

Computer vision (CV) deals with automated methods for image-based inspection and analysis. Natural language processing (NLP) involves automated processing of large quantities of data written in human language.

AI Technology

I will categorize opportunities for Uganda’s use of AI technologies for public health under the following themes: disease prevention, disease surveillance, healthcare delivery, and medical research.

Others are optimizing resource allocation, personalized medicine, pervasive sensing, etc. I will also highlight some current challenges limiting adoption and use of AI technologies for our public health as well as identify what needs to be done to overcome them.

Strengthening measures to prevent the spread of covid-19

Given Uganda’s poorly developed healthcare system, prevention remains its best strategy against Covid-19.

Current interventions being implemented to achieve this include school closures, mandatory institutional quarantine for those who test positive for covid-19, and banning mass gatherings/crowds.

Others are, banning public transport and private vehicles, staying home, observing social distancing, guidelines for personal hygiene and sanitation practice, and imposing night curfews.

While good levels of compliance are being achieved, cases of violation of these interventions have also been reported. Daily enforcement and monitoring of compliance are therefore, important components for success of these measures.

However, enforcing and monitoring compliance on a large scale (nationwide) can be a labor-intensive and costly endeavor. It is also prone to abuse due to rampant corruption.

There are multiple opportunities for using AI technologies for monitoring compliance and enforcing adherence to guidelines in almost all the interventions listed above.

For example, drones (equipped with intelligent software and video cameras) can be used for surveillance to gather data on whether people are observing social distancing in public places such as open air markets or violating quarantine rules. They can also be

Its now two months and a half (from January 30 , 2020) since the World Health Organization deployed for monitoring, illegal cross-border movement of people, human and motor traffic on the roads, whether people are wearing masks, etc.

Communicating information to members of the public about health guidelines and other restrictions to curb the spread of covid-19, in an effective manner is key to achieving success.

Communication

In this area, NLP technologies can be used for automated translation of health messages and written speeches into different formats and local languages.

For example, we could use text-to-speech technology to convert text-based health messages into audio messages for the visually impaired, or vice versa for the deaf using speech-to-text technology.

Professional human translators on Ugandan languages are scarce, and may even be less efficient on large scale jobs.

A keen social media (WhatsApp, Facebook, etc) user in Uganda would by now have noticed that there is a lot of inaccurate information about covid-19 circulating in cyberspace.

The UN Secretary General, Antonio Guterres has termed it an informational epidemic that requires our immediate attention since it threatens to derail the fight against the covid-19 pandemic.

What we are experiencing is called disinformation, an informational disorder that involves the creation and distribution of fabricated or deliberately manipulated messages/content (e.g. audio/visual and conspiracy theories) with intent to poison the mind of members of the public.

The current disinformation with respect to covid-19 is taking place on a large scale involving massive amounts of content (text-based messages, video, etc). Given the low literacy levels, majority of Ugandans may not be able to distinguish between credible content and content designed to mislead.

The huge volumes of messages involved also causes information overload.

Due to the latter, there are high chances that useful public health messages from credible sources ‘get lost’ in the crowd of junk content. To address this informational epidemic ML can be used to monitor, detect, filter, and stop the circulation of malicious content through social media networks. Tech giants such as Google, Facebook, YouTube, Twitter, etc are already investing lots of money in research to prevent disinformation on their platforms. Since a lot of this malicious content is generated locally and presented in Ugandan languages, a local solution too is required for addressing the problem.

A dedicated, comprehensive, Ugandan government-owned source of information for covid-19 in form of an app that is updated daily would also be a good solution.

It is expected that in designing strategies to fight an epidemic, public health experts would evaluate multiple alternative policies/interventions before making recommendations for implementation of any or combinations of them.

Big data

Various factors may be considered when evaluating interventions including efficacy and health/socio-economic impact.

Big data analytics is a data-driven modern approach for evaluating various intervention strategies, e.g., comparing effectiveness and impact of total lock-down vs. partial lock-down, quarantine vs. mass vaccination, etc.

Enhancing the current corona virus disease surveillance program

The main purpose of any surveillance program is to detect and report occurrence of a disease in a population in numbers larger than normally expected.

Data from a disease surveillance program is also useful for predicting and understanding the likely spatio-temporal trajectory of an outbreak.

Disease surveillance provides an important application domain for AI technologies, including real- time surveillance.

An interesting example here is use of mobile phone communication data i.e., call detail records (CDR) like phone calls made by a user, SMSs sent, etc for modeling and predicting public behavior patterns such as daily movement or lack of it.

This analysis can give insight into whether people are abiding by movement restrictions. At individual level, we could use CDR to track movement of confirmed or potential covid-19 cases and for identifying and tracing their contacts.

Social network analysis based on mobile phone data can help with modeling and predicting the likely spatio-temporal trajectory of an outbreak.

Testing

As mentioned earlier, systematic testing of the population for covid-19 is not being carried out in Uganda due to lack of testing kits. According to Uganda’s Ministry of Health, it costs USD 65 (approx. UShs 250,000) to carry out a single covid-19 test.

Professors at Makerere University’s College of Health Sciences are currently undertaking research to develop a low-cost covid-19 test kit, which is expected to be ready by end of this month.

AI-based innovations could also be explored to use chest x-ray images of the lungs and upper respiratory tract to diagnose covid-19.

Since x-ray machines are readily available in many Ugandan health facilities, a low-cost AI-based technology solution for diagnosing covid-19 could be an attractive alternative.

Disease threat modeling and risk analysis is another area where AI and big data analytics could be used in the fight against covid-19.

Healthcare delivery

I already mentioned the potential to use chest x-ray images for diagnosis of covid-19. Within the realm of healthcare delivery, we can combine patient diagnostic information, clinical history, and demographic data to predict covid-19 treatment outcome. Personalized medicine and clinical decision-making support are other lines of innovation for applying AI in healthcare delivery.

There have been reports of misuse of healthcare delivery logistics such as ambulances. AI technologies can be integrated with Geographic Information System (GIS) technologies to monitor, track, and visualize ambulance movement to identify any divergences from expected normal usage patterns.

Medical research

News filtered through a few days ago that local scientists at the University of Ghana have successfully sequenced the genome (i.e., compiled genetic make-up) of strains of SARS-CoV-2, the virus that is causing the current covid-19 pandemic.

They made use of local capacity including use of the modern Next Generation Sequencing (NGS) technology that requires high performance computing systems.

This line of work falls within the domain of bio-informatics, the collection and analysis of large scale biological (i.e., genetic) data. Uganda is host to an African Center of Excellence (ACE) in Bio-informatics and Data Intensive Sciences located at the IDI-McKinnell Knowledge Center, Makerere University. This facility will be key for surveillance by identifying and tracking mutations of the strains of the SARS-CoV-2 virus in Uganda.

The race is on to find a vaccine against the SARS-CoV-2 virus. The use of AI technologies for drug design and discovery remains an open research problem.

AI-based diagnosis of covid-19 using chest x-ray images falls in the domain of automated medical image analysis. Since we have covid-19 cases in Uganda, collecting data for research purposes shouldn’t be an issue.

Challenges and solutions

In my opinion, the two most important challenges to developing innovative AI technologies for public health and medicine in Uganda are two: lack of access to data and lack of funding.

For health data, there is plenty of it sitting idle at most of our healthcare facilities that would provide the much-needed raw material for developing innovative healthcare solutions. Health data should also be available through Uganda’s Health Management Information System (HMIS).

However, administrative bureaucracy makes it difficult for researchers to access data from these sources. Secondly, a lot of health data in Uganda is of poor quality due to a number of reasons including irregular reporting to HMIS by health facilities, incompleteness, discrepancies or inaccuracies, etc.

Other kinds of data are also difficult to access for research purposes, such as CDR from telcos in Uganda.

Other challenges include lack of access/high cost of cutting edge AI development technology, especially to graduate student researchers, and ethical concerns about data privacy. Enacting relevant policies to facilitate access to health and other data for research purposes and providing access to research funding for graduate students are some of the ways in which to overcome the two major challenges.

Concerns over privacy can be addressed by anonymizing (using hashing algorithms) personally identifying data such as mobile phone communication data.

In this article, I have tried to identify and map out opportunities for Uganda’s use of AI technologies for public health. These include preventing spread of epidemics, disease surveillance, healthcare delivery, and medical research. However, lack of access to data and research funding are the major challenges hindering development of innovative AI-based technology solutions for public health use in Uganda.

Rahman Sanya

The writer is a Computer Science PhD student at Makerere University researching on applications of Machine Learning in public health.

Twitter: @rtsanya