“By scanning millions of public social media posts, researchers will be able to detect rare problems with vaccines more quickly.”
Social media platforms such as X/Twitter and open discussion forums such as Reddit publish vast numbers of posts from millions of users. These online conversations can be “noisy” as they contain a lot of inaccurate information and opinions, especially when the discussion turns to vaccines.
But can these discussions also be a valuable source of information about disease outbreaks and vaccine safety?
While the amount of data on social media may be too much for an individual to examine, machine learning tools can quickly scan for common themes and provide an early warning system for rare issues.
First, it’s important to know how vaccines are currently being tested and what steps are being taken to monitor safety even after a vaccine is approved.
Vaccines are administered to thousands of people in clinical trials before being made widely available to the public. This is just the first step in a safety monitoring system that encourages individuals and medical professionals to report suspected problems to regulators even after a vaccine is approved. Regulatory agencies investigate these reports to find rare adverse events that may not have been detected in clinical trials.
For more information on how regulators monitor vaccine safety, see the information box below.
Voluntary reporting of these types of adverse events remains the basis for continuous safety monitoring, but could be improved using machine learning technology, according to Dr Jim Butterly from the University of Melbourne and the Murdoch Children’s Research Institute. It is said that there is a sex.
Dr. Butterly, co-director of the Global Vaccine Data Network, is an experienced pediatrician with a long history of working on vaccine safety, a key element in maintaining public confidence in immunization systems. Early in his career, Dr. Butterly cared for many infants with meningitis. This experience shaped his approach to preventive health.
“The number of children who die or suffer permanent neurological damage from meningitis will decrease within two to three years of the introduction of the Hib vaccine (which prevents the main cause of meningitis in children under 5). It decreased dramatically,” he recalls. “From there, I became obsessed with the importance of vaccine development and deployment.”
How can machine learning help?
Most posts on X/Twitter are public, and 70% of users provide their location. Details about their location and the communities they participate in can be determined by analyzing the language they use and other information included in their posts.
People concerned about adverse events after vaccination may share their thoughts and questions on social channels. If a particular problem is mentioned by a large number of people or a significant concentration within a particular region or community, it may indicate a vaccine-related illness. Machine learning helps make sense of the vast amount of information circulating online by scanning for keywords and recognizing common themes.
Machine learning: Connecting the dots between vast numbers of data points
The most active online vaccine conversation in history occurred in the first half of 2021, when developed countries introduced COVID-19 vaccines. Although the vaccine had been tested and approved, the scale of the rollout meant more people were vaccinated that year than ever before. The Melbourne team used machine learning algorithms to detect several potential problems.
“We recognized real concerns about tinnitus, balance, and hearing problems after receiving the COVID-19 vaccine,” Dr. Butterly told Vaccines Today. “We were then able to use another dataset to test whether there was an association.”
After looking at the number of people seeking medical care, they found that while there was no increase in hearing loss after vaccination, there was an increase in reported tinnitus (ringing in the ears), leading researchers to investigate further. We contacted regulatory authorities.
Some adverse events are more difficult to detect than others, especially if there is a time lag between vaccination and the reported problem. For example, allergic reactions usually occur soon after vaccination. Other events can be picked up much later.
During the COVID-19 pandemic, social media monitoring has revealed a sharp increase in reports of menstrual disorders following vaccination. This effect is short-lived and does not affect all women, but in some cases it has led to longer menstrual cycles. Many of those affected had not made an appointment with their GP or reported it to regulators, but some did post on X/Twitter.
Although the issue did not have a major impact on the health care system, it became a hot topic at the community level. The women knew something was up. Traditional surveillance systems can be slow to identify these types of safety signals, but they can be detected more quickly by actively monitoring conversations in public spaces. In fact, if health officials can quickly identify issues that are of real concern to the community, it can make a big difference in trust in vaccines.
Platforms like X/Twitter and Reddit and open tools like Google Trends are ready-to-use sources of information. However, some platforms prohibit analysis unless researchers can secure specific permissions to access the data.
The more data researchers have, the more sensitive the system becomes. Additionally, some social media channels are particularly popular with certain demographics, so X/Twitter has a high percentage of US users. Facebook’s network is rapidly aging. TikTok and SnapChat are (currently) popular with youth, and expanding the data pool reduces the risk of bias.
There are other potential sources of bias. “Bots (automated accounts that post on social media sites) share and amplify vaccine information,” Dr. Butterly said. “It’s not all misinformation. Some bots spread positive or neutral messages, but negative posts are especially good at provoking a human response. That’s why we include them in our analysis. We apply bot detection on a regular basis.”
This rapidly changing field has the potential to shape the future of vaccine rollout and will require collaboration between doctors, vaccine experts, computer scientists, and regulators. Together, they are developing tools that can detect the signal in the noise. This not only supports the safety of the vaccine, but could also be a welcome sign of confidence in the vaccine.
read the research
How is vaccine safety monitored?
Before a vaccine is made available to the public, it will be given to thousands of people to test whether it can safely prevent serious illness. After getting vaccinated, many people roll up their sleeves and get through the rest of the day without any problems. Some people may have a sore arm or feel tired or feel generally unwell after certain vaccinations. This is usually mild and short-lived, lasting 1 to 2 days.
Independent regulatory authorities will decide whether to approve a vaccine based on the balance between these adverse events and its protective effect against severe disease. All medical interventions, from cholesterol drugs to chemotherapy and surgery, have potential downsides that patients and clinicians must consider. Removing a brain tumor comes with risks, but doing nothing may be even worse.
The standards are particularly high for vaccines, as the entire population is vaccinated, including infants, vulnerable people and healthy people. That’s why clinical trials on healthy people are closely studied before experts give a new vaccine the green light.
But once millions of people are vaccinated after approval, rare problems may occur. This means that there is a 1 in 1 million chance of an event occurring that would not have been detected in a trial of 5,000 people. This is where “post-market surveillance” comes into play. Individuals and medical professionals can report suspicious symptoms to regulatory authorities. These may include a rash or dizziness immediately after vaccination, or less common health problems several weeks after vaccination.
Regulators look for patterns in these reports and consider plausible links between vaccination and the reported problem to determine whether the vaccine could be the cause. In some cases, the vaccine may be the culprit, leading regulators to suspend its use or limit advice on who it can be offered to. In other cases, vaccination may be unrelated to the suspected problem.
Related articles
How is vaccine safety monitored?October 7, 2024
Bridging the gap: A data-driven vision for life course immunization November 10, 2023
