Five questions with UCSF neurologist and stroke researcher Anthony Kim about his new study on how the Internet can help to connect with people who are searching for information online and potentially reduce incidences of preventable diseases.

Anthony S. Kim, MD, MAS is assistant clinical professor of neurology at the University of California, San Francisco (UCSF) and Medical Director of the UCSF Stroke Center. His research focuses on improving the diagnosis and cost-effective management of stroke and transient ischemic attack (TIA, also called “mini-stroke”).  An estimated 800,000 new strokes occur each year in the U.S., making it the fourth leading cause of death in America. Anthony Kim believes that the Internet opens up new opportunities that will change the way we develop interventions and conduct research to improve health.

Q: Millions of Americans search online for health information each year. Scientists are using this type of data to better understand flu outbreaks, the seasonal variance of kidney stones, the demographic prevalence of stroke, and even to demonstrate the online effectiveness of health awareness campaigns. What did you learn in your latest study?

We were surprised to see that tens of thousands of people were regularly ‘asking’ a search engine about stroke-related symptoms in many cases shortly after the onset of symptoms. In fact, every month, about 100 people were finding our study website by entering the query: “Am I having a stroke?” directly into their Google search box.

One of the challenges with mini-stroke is that most people do not seek urgent medical attention because the symptoms are transitory by definition. So people don’t realize that it is a medical emergency. Even though the symptoms may have resolved, the risk of a subsequent stroke is very high—upwards of 11% within the next 90 days—with most of this risk concentrated in the first hours and days after the mini-stroke. So getting the message out there about urgent medical attention is key.

We started this study because we thought that if people who have had a mini-stroke are looking online for information on their symptoms, then rather than just listing static health information about the disease on a website, maybe we can engage them by making the website more interactive and asking them to enter some of their symptoms online. And we wondered whether we could use this information to assess whether or not it was a true TIA or stroke and then encourage them to get the urgent medical as appropriate.

One third of the people we identified hadn’t had a medical evaluation for mini-stroke yet, which is critical, because it is a medical emergency. Instead of calling a doctor or going to the emergency room, many people were turning to the Internet as the first source for health information.

Q: How did your approach work exactly?

When a person searched on Google for stroke-related keywords, a paid text advertisement “Possible Mini-Stroke/TIA?” appeared with a link to the study website (Image). The ad appeared on the search results page and on related websites with existing content about the topic.

When users clicked on the text ad link, they were directed to the study website. Those visitors who met all of the study’s entry criteria were asked to provide informed consent online. They then reported their demographic information and symptoms based on a risk score developed for use by clinicians.

We were notified in real-time as soon as someone enrolled, and then we arranged for two vascular neurologists to follow up with the patient by telephone.

Q: You tested the approach for about four months. What’s your verdict?

We definitely think that there is a lot of potential here. About 60% of U.S. adults say that their real-life medical decisions have been influenced by information they read online. This changes the way we think about providing medical care and conducting research.

With a modest advertising budget, we were able to attract more than 200 people to our study website each day from all 50 states. About one percent of them (251 out of 25,000) completed the online questionnaire, which allowed us to contact them for follow up. Although this seems low at first, it is comparable to conversion rates in other domains of online advertising.

Also, even though the people who joined the study were a highly selected group, the incremental costs for reaching an additional person were low and the potential for applying a targeted and cost-effective public health intervention in this group would still be very interesting to evaluate in the future.

Before we started, we thought that we might lose people throughout the enrollment process since we confirmed eligibility and asked for consent online, but we didn’t. For the most part, if people were interested in participating, they completed the entire online enrollment process.

During follow up calls, we learned that 38% of enrollees actually had a mini-stroke or stroke. But fully a third of them had not seen a doctor yet. Our approach made it possible to connect with these people fairly efficiently and early on in order to influence their behavior acutely.

Despite these potential advantages, Internet-based public health interventions that target people who are looking for health information online are still underdeveloped and understudied. There’s a lot for us to learn in this space.

Q: What online tools did you use to carry out your project?

We used Google AdWords and Google’s Display Network to target English-speaking adults in the U.S. During the four-month enrollment period, the tool automatically displayed our ads more than 4.5 million times based on criteria such as location, demographics, and search terms.

Ideally, to minimize ongoing costs you would want to build and optimize a website so that it ranks highly among the non-paid (organic) search results. Non-profits can also take advantage of Google Grants, a program that supports in-kind donations of advertising resources to help selected organizations promote their websites on Google.

Q: Do you have any tips for others who want to develop similar projects?

We quickly realized that it helped to work closely with our Institutional Review Board (IRB) given that this is a new and evolving area of research, and to ensure data security and safety mechanisms are in place to protect participants. I definitely recommend that.

It’s also important to be realistic about the goals and metrics of success, and not to over-interpret numbers that seem to reflect low engagement. We saw that most visitors (86%) immediately exited the website within a few seconds of arriving at the home page. This probably reflected people who were looking for something else and clicked away immediately. But the beauty of the Internet is that it is very efficient to reach people across a wide geographic area very quickly. So it is not unexpected that we would also screen visitors who may not be qualified for the study or are not interested in enrolling.

Groups interested in using this approach should think about selection bias, authentication, validation, and the “digital divide”. Even though there is some evidence that disparities in access and adoption of Internet technologies are narrowing in the U.S., depending on the goals and target for your study or intervention the reach of the Internet is not uniform.

But selection bias issues aside, for a public health intervention you may be most interested in other metrics such as the number of people reached per dollar spent, or the burden of disease averted per dollar spent, which the Internet is particularly suited to help optimize.

And, it’s definitely beneficial to bring different subject matter and methods experts to the table. Knowledge of search engine optimization, online accessibility, website and user interface design is not necessarily part of the core expertise of a traditional clinical researcher, but developing these skills and interacting with experts in these areas could become very important for the new cadre of clinical researchers and public health professionals coming down the pipeline.

The original article was published on CTSI at UCSF