“I hate Donald Trump!” “I’m exhausted and my boss doesn’t care,” “I'm jealous of my sister's new car.”
If this is what your tweets look like, then you might want to reconsider your words: research shows you may be at risk for heart disease. Most people don’t realize it, but the language they use in social media posts can be used to predict their well-being, like their risk for heart disease or other serious conditions. How can social media be used to monitor people’s health and overall well-being?
In one study, researchers evaluated 100 million tweets from 1,300 counties in the United States. The language in these tweets was analyzed and categorized as having either negative or positive sentiment. Then, the authors separated the words of each tweet into word clouds that reflected “risky” language (e.g., despise, hate, jealous, tired) or “protective” language (e.g., opportunity, strength, hope, great).
Using machine learning methods, the authors created algorithms that compared the sentiment of the tweets from each county to CDC data on causes of death. Their findings were dramatic: counties whose tweets expressed more negative emotion (e.g., tweets filled with words such as “hate”) had more heart disease–related deaths compared to counties that featured tweets with more protective language.
When looking further into the data, the methods worked very well at predicting death from hardening of the arteries (atherosclerosis), which is the leading cause of death in the United States. It’s interesting that the people who were tweeting were not the people whose deaths were measured. Instead, the overall tone of the tweets — which were from people too young to be suffering from heart problems — appeared “to have captured a snapshot of the psychology of the community at large.” The authors had therefore discovered a similar result to one that our own team found in an earlier study focusing on HIV.
The authors claim that their methods predicted death from heart disease more accurately than risk factors such as obesity, smoking, and diabetes. Moreover, the prediction accuracy remained strong even after they considered classic predictors of heart disease such as education as poverty. These claims might make you raise your eyebrows, but decades of research has shown that the words people use shows a lot about their psychology. In this case, the algorithms created by the researchers were able to predict personality traits. In fact, the authors claim the algorithms they developed predicted personality traits as well as or better than friends who filled out personality surveys about the participants!
So, how does this research impact your life? For one, an entirely new field of research, known as “digital epidemiology,” has sprung up around social media. Now that social media is widely used (65% of American adults visit social media sites regularly, and 90% of young adults use at least one service), doctors and researchers have an entirely new tool to predict communitywide well-being. Healthcare providers already monitor Google searches to forecast disease outbreaks (e.g., flu, malaria, STDs), which helps determine where resources should be allocated.
For individuals, the hope is that as more and more people use social media, health predictions will become even better. For example, future studies may ask patients to provide access to their social media accounts when they go to an emergency room or visit their primary care doctor. Language in postings and status updates could provide clues to the risk for depression, which is a major risk factor in the recurrence of many diseases. And existing studies have already helped identify post-partum depression among new mothers.
If trends are any indication, social media is going to become even more popular and integrated into tools to solve real-world problems. I’m hopeful that we’ll find ways to use social media to predict other leading causes of death in the near future, and that we’ll find an effective way to protect personal privacy at the same time.