With COVID cases rising dramatically in India, the country faces another enemy when trying to slow the infection rate: misinformation. As this article states well, “Experts believe that people's low trust in news media and a weak public service media, coupled with a fragmented audience and high social media use, have been responsible for the rapid and wide spread of misinformation.” WhatsApp seems to be the primary culprit, with sharing of conspiracy theories and hoax stories about vaccines dominating communication in the country.

And this issue is obviously not limited to India; here in the United States, vaccination rates are slowing down dramatically as we continue to fight against false news about easily available vaccines. The Biden administration has shifted their distribution strategy, moving away from mass vaccination sites and focusing more on directly reaching those who are hesitant or unmotivated. And while there are multiple reasons for why someone may choose to not get vaccinated, how do we confront the looming issue of those individuals who simply refuse to get vaccinated due to information they’ve received from Facebook or other misinformation groups?

The pandemic has clearly highlighted the challenge of dealing with false information in medicine, especially during a public health crisis where our risk is increasingly determined by the status of individuals in our surrounding community. But these challenges have plagued our society for ages now, especially in the past decade when social media platforms have allowed for the proliferation of this speech without consequence. In this article I’ll focus on how misinformation has hidden yet devastating consequences for healthcare, and what approaches are being researched to counter this surging tide.

Patient Empowerment

Ask anyone and they’ll tell you that empowering patients with knowledge and information is a good thing. Patients know more about their lives outside of the doctor’s office, and thus they provide valuable insight to make an informed decision with their physician. As Eric Topol writes about in his book ‘The Patient Will See You Now”, the patient democratization process is already underway, and advances in digital health and open-sourced medical knowledge are enabling a fundamental shift in how healthcare is and will be delivered. And this book was written 7 years ago! 

But the flip side of this patient autonomy is that patients will sometimes feel that they know better than their physician, and will override guidelines and recommendations based on their own knowledge sources. The future of healthcare involves shared decision-making between the patient and the physician, but what does research tell us about how this information can change patient decisions?

One paper took a look at the harmful effects of celebrities’ cancer stories on patient behavior. One example is Angelina Jolie’s decision to “undergo the preventative measure of removing both her breasts and by encouraging awareness of familial risk for breast cancer and exploration of therapeutic options in order to make informed choices.” The authors noted that this led to an increase in patients seeking information regarding risk-reducing double mastectomy, and that while this may make sense in the case of Jolie (who had an 87% lifetime risk from breast cancer due to a genetic mutation), it may be an unnecessarily aggressive surgical option for others.

They note the reality that “...patients find scientific evidence generated by rigorous clinical trials less compelling than anecdotes by or about celebrities.” Humans are always looking for patterns or stories that entrance us, and we use any evidence of that information to build a case for a certain decision. This paper, along with most others that address this topic, advise physicians to be proactive in their response to such decisions. Monitoring public information and trends is important, along with actively engaging patients in the community and in their office visits. By working with patients to navigate their questions and concerns appropriately, physicians can build trust and converge on an appropriate course of care.

The Impact of Media

The role of the media in forming our opinions has been in the spotlight lately, mainly in relation to the political division and polarization that exists in the United States. And this past year, as many of us have been stuck inside due to the pandemic, we’ve turned to our televisions and smartphones to stay connected and learn more about what was going on in the world. But when it comes to health information, was this a good thing?

Early on during the pandemic, researchers from Pennsylvania conducted a cross-sectional online survey in their state to find out which sources were most trusted for COVID-related information. They found that most trusted internet government websites (CDC/NIH/WHO; 42.8%, N = 2547), followed by television news (27.2%, N = 1620) and health system communications (9.3%, N = 550). They then followed up with knowledge questions about COVID-19, and then stratified the probability of getting a question right based on the single most trusted response from that respondent. 

Overall, they found that respondents who trusted a government health website were more likely to answer a knowledge question correctly, while those who relied on television or Facebook sources were less likely to answer correctly. Obviously, the generalizability and selection bias in this study are important considerations, as the researchers themselves note, but it is an interesting finding on a large sample. 

The concerns about the effects of Facebook on health misinformation are well known, and with more patients getting information from conspiracy theorists, how should physicians combat this rising tide? Some healthcare workers are using their voice to create content and promote health literacy, but as with any content, there are also others who promote false information.

So then does it fall upon platforms like Facebook and Twitter to monitor and regulate their content? We’ve been down that road before with mixed results. Twitter added fact verification flags on certain tweets, but we obviously can’t expect them to catch everything. Facebook has always taken a more hands-off approach, and although recent events have caused them to make some minor changes, this is still largely true. COVID-19 has shown the holes in our public health infrastructure, and if most individuals are getting information online, then we need our government officials to do more in regulating and monitoring these sources of false information. Some of this is happening in other areas, but health bodies should get involved with a focus on the nuance of misinformation in healthcare.

As for trust, it will be a continual battle but one worth fighting. Especially in minority groups that have been (and continue to be) discriminated against and silenced, trust in local physicians and healthcare staff is fundamental to decision making. There are tons of examples of physicians who are leaders in this space, and with our generation growing up as digital natives with a hunger for social justice, I’m optimistic about the future.

The Role of Machine Learning

Automatically detecting the factfulness of a claim is an incredibly complex and nuanced Natural Language Processing (NLP) task. On top of that, measuring what ‘truth’ is makes it even more complicated; traditional approaches have just compared claims to sources that were labelled as ‘trusted’, but what if these sources change/don’t adapt to new evidence? We saw this first-hand during the pandemic, where research and knowledge were constantly evolving and guidelines were changed accordingly. This led to confusion for a lot of people, and even a preference from some for consistent information and recommendations, even if that information was unfounded.

A new dataset called VitaminC was recently created to infuse this concept into models. The authors created this dataset by creating examples with contrastive knowledge, where there is a change in the context that requires the model to adapt and still identify the fact. They provide the follow example:

They created the dataset by annotating and modifying articles from English Wikipedia, and then focused on 4 tasks, which are outlined below on a COVID example:

Overall, they were able to show that in comparison to current datasets, classifiers trained on VitaminC improved sensitivity to subtle changes in evidence, as well as increased robustness to adversarial examples (purposefully created false claims). This is a cool example of how human creativity and knowledge about information sources can help us train machines to separate fact from fiction. 

To be clear, I don’t think we’re going to be able to build a massive neural network model that learns to automatically flag and remove all misinformation in order to resolve this issue. But with the vast amount of false information coming from online sources, we do have access to a ton of data that can be used to measure behavior and provide feedback. We’re a long ways away from automated fact-checking, and there are additional ethical and societal considerations that we need to think about before deploying something like that, but this is a rapidly growing area of research that has huge ramifications for the future of healthcare.

What’s Next

While we’ve only scraped the tip of the iceberg in this article, in the future I’m hoping that we can dive into some more technical examples of misinformation at V2 Labs, as well as talk more about its clinical impact on quality and satisfaction metrics. If there’s something you’d like to see us touch on in this field, feel free to drop a comment! 

Thanks for reading,

Viggy