Tech to Improve Stroke Outcomes; Impact of a Nation’s Social Media Ban

TTHealthWatch is a weekly podcast from Texas Tech. In it, Elizabeth Tracey, director of electronic media for Johns Hopkins Medicine in Baltimore, and Rick Lange, MD, president of Texas Tech Health El Paso, look at the top medical stories of the week.

This week’s topics include technology to improve stroke outcomes, adult use of social media and health decisions, the impact of a social media ban on Australian youth, and a large language model (LLM) to augment clinical management in a low-resource setting.

Program notes:

0:46 Social media for health information among adults in U.S.

1:46 85% sharing personal or health information

2:46 Commonly used and asked about

3:46 Patient-specific factors

4:17 Artificial intelligence (AI)-enabled clinical support system in a low-resource setting

5:17 Using the record with or without LLM

6:18 What are appropriate outcomes?

7:20 Australia social media restrictions for teens

8:21 Time spent stable before and after

9:00 Immersive virtual reality and stroke recovery

10:00 Virtual reality, synchronous sensory stimulation

11:00 Conventional rehab focused on motor function

12:49 End

Transcript:

Elizabeth: Can abundant technology improve outcomes after stroke?

Rick: How do U.S. adults use social media for health information?

Elizabeth: Can generative AI help clinical decision making in low-resource settings?

Rick: And do age-based restrictions on social media change use among teenagers?

Elizabeth: That’s what we’re talking about this week on TTHealthWatch, your weekly look at the medical headlines from Texas Tech University Health Sciences Center in El Paso. I’m Elizabeth Tracey, a Baltimore-based medical journalist.

Rick: And I’m Rick Lange, president of Texas Tech Health El Paso.

Elizabeth: Rick, the social media things have been much in the news, so I’m going to say pick your poison. Which of those two, either the BMJ or JAMA, would you like to start with?

Rick: Well, let’s start with JAMA: “[Use of] Social Media for Health Information Among U.S. Adults.” Patterns of health-related engagement with social media among U.S. adults with and without chronic conditions. And it was conducted between February and March of 2026. And they used data from the 2024 Health Information National Trends Survey. And they asked adults how they used social media and social networking sites, things like Facebook or LinkedIn, over the preceding 12 months. And they looked at four different types of health-related social media engagement. Did the individual share general or personal health content? Did they participate in online communities? Did they use it to make health decisions? And finally, their perceptions of misinformation.

Over 7,200 respondents that represent 262 million U.S. adults. They noted that social media use was reported by 88%. And among those social media users, engagement with health-related content was common, about 85% sharing personal or general health information. About 70% were involved in online communities. And one in five reported making health-related decisions based upon the social media content.

Here’s the thing that’s perplexing. About 80% of the users reported believing that the health information encountered on social media was false or misleading. Older and Hispanic social media users were significantly more likely to make health decisions based on information from social media. Those that had a higher educational attainment, higher household income, and those that had distrust in social media were less likely to use it for health information.

Elizabeth: The corollary to this, of course, is, as you know, I record these TIMS files — This is My Story — with patients. And just this last Saturday, I had a patient who said to me in answering one of the questions — which was, “What does the team need to know about you to care for you best?” — related the fact that he deeply appreciated the fact that when he sat there as an inpatient and scrolled about various parameters of his clinical condition on Google and then asked the team about them when they came in, he was deeply appreciative of the fact that they were OK with that, that they didn’t disparage the fact that he had come up with this information by that mechanism. We see it all the time. All right. How harmful is it if you act on some of the stuff that is available on these platforms?

Rick: As you mentioned, the majority of U.S. adults that use social media, it’s almost nine in 10, report some engagement with health-related information. And I think it’s healthy that 80% of adults have some mistrust for it. It’s a piece of information, and if you use it and bring it to your healthcare provider, they can say, yes, that’s great information, or B is I wouldn’t trust it.

Elizabeth: I would also note that sometimes what the healthcare provider says is, yes, indeed, for many people with this condition, this is an appropriate strategy, but you have these additional factors that we need to consider. And therefore, in your case, this is not an actionable piece of information.

Rick: You’re right because social media does not have the context. We need to look at approaches that actually enhance the accuracy of the content on these social media, and we need to have effective ways of countering misinformation.

Elizabeth: And I suspect we’re going to see it, the day, very shortly, where intake questionnaires are going to ask people about their social media use so that clinicians can be apprised ahead of time.

Rick: Yeah.

Elizabeth: Let’s turn to Nature Medicine. Let’s take a look at this generative AI-enabled clinical decision support system in a primary care setting.

There’s a lot of hype, as we know, about the implementation of these large language models and their ability to inform clinical decision making. And that’s in many venues, not just in low-resource settings. Clearly, however, it’s attractive to consider implementing them in low-resource settings because they are, by their nature, they have fewer clinicians. The clinician experience and education is not at such a high level. It seems intuitive that the likelihood for misdiagnosis or mismanagement might be higher.

In this case, we’re talking about a low-resource setting in Africa. They looked at the performance of a large language model in Kenya. There, employ folks that are called clinical officers and they were randomized to use the electronic medical record for their patients with or without LLM assistance. Their primary outcome was composite treatment failure experienced within 14 days of enrollment, and they also gathered information on hospitalization and death. They had just shy of 10,000 patients enrolled and 103 clinical officers, about half in each of the arms.

Treatment failure occurred in 2.2% of the intervention arm and in 2.0% of the control arm. So, clearly, there was no benefit to using the large language model in this setting. They didn’t see any adverse events related to it and they determined, ultimately, that the large language model assistance was safe, but did not reduce this treatment failure within 14 days, and that any benefit that was experienced here was modest at best.

Rick: As we use artificial intelligence, large language models, generative AI, we’re going to have to define what we consider to be the appropriate outcome. What they didn’t look at is how often was the correct diagnosis made, whether the person received treatment according to standard guidelines.

They did look at documentation to show that the documentation was better with large language models than it was by the individual clinical officer. Geographically, the conditions they’re seeing here are very different than what you see in a primary care center in the United States. Sixty percent of these were in febrile or infectious diseases, whereas in the U.S., it’s more likely to be cardiovascular, hypertension, diabetes, obesity, chronic diseases. AI and LLMs are going to continue to get better. Will it be beneficial or not? Still up in the air as far as I’m concerned.

Elizabeth: I’m going to say maybe on that, because I think we are starting to see data emerge on the limitations of LLMs in clinical settings, and I think that’s irrespective of whether it’s a low-resource setting or not.

Rick: I think your point is well taken, as we think that generative AI and LLM is going to solve a lot of issues. But until we actually prove it in a clinical setting, it still remains up for grabs.

Elizabeth: Let’s turn to the BMJ.

Rick: I was unaware that in December of 2025, Australia introduced a national policy that required specific social media platforms to restrict access for individuals under the age of 16. OK. How effective was that?

They took a look at 408 kids immediately before the act and 3 months afterwards. More than 85 of the participants under age 16 reported continued social media platform use even after the act. Many of these social media acts had self-reported age. Sometimes they asked individuals to load a picture or a selfie, but that wasn’t terribly effective. Others used a fake account or social media access via a private browser.

When they looked at did it really decrease social media use, between 2% to 9% among 12- to 13-year-olds and 14- to 15-year-olds. The time spent using social media on a daily basis was relatively stable from before to after the act. They looked at, in this particular study, 12- and 13- and 14- and 15-year-olds that had already had long-time exposure to social media.

Elizabeth: The other thing I think it brings up for me, anyway, is how do you validate age when it comes to these kinds of platforms? And I’m not sure we really have a good way of doing that at this point.

Rick: You’re right. If you ask the kids to self-report, ask them to take a picture, it doesn’t help. Are there other ways to do it? Yep. For example, you may have to get the parents to verify what’s the kid’s age. Just enacting an act like Australia did, not very effective.

Elizabeth: Back to the drawing board.

Finally, let’s turn back to Nature Medicine. And this study is taking a look at what’s called immersive virtual reality with synchronous neurostimulation for upper limb recovery after stroke.

We know that stroke affects 15 million people annually and about 5 million of them experience permanent disability as a result. In what the authors describe as their chronic phase, which is greater than 3 months after the event, patients also often experience persistent sensory motor deficits and altered body representation. And rehabilitation at that point can be very partial, inconsistent, and doesn’t really result in improvement. So these folks decided to take immersive technologies and noninvasive neurostimulation in synchrony with each other to evaluate how well this might help people who had had deficits in the upper limb after having suffered a stroke.

They created this thing called MultiSensy integrating virtual reality and synchronous transcutaneous sensory neurostimulation. They had two different metrics in this, the Fugl-Meyer Assessment Upper Extremity and Action Research Arm Test, in comparison to the control group worked really well. Both of those scores improved significantly, almost twice as much with one assessment and three times as much with the other assessment when this MultiSensy platform was utilized. The authors say it’s possible that we might be able to even do this at home to help people improve their improvement after having suffered a stroke.

Rick: They took individuals that had a stroke anywhere from 3 to as long as 20 years previously. Most of the recovery from stroke occurs within the first 3 months, but there’s obviously some recovery that can occur later, some neuroplasticity. Conventional rehabilitation focuses on motor function. It doesn’t integrate sensation and audiovisual stimulation, and motor function as well. That’s what this multimodality did, using virtual reality to do so. Very creative and, as you mentioned, very effective. It looks very promising.

Elizabeth: And it’s actually also a fairly modest intervention. They had both groups who had 3 weeks of dose-matched upper limb rehabilitation, corresponding to 12 sessions, four per week, of active intervention, and then the MultiSensy group used their platform, completing six to eight rehabilitation games per session, each lasting at least 4 minutes. And I like the aspect of it that it’s like a game. I think people probably find that a little bit more engaging, although they don’t really assess that.

Rick: It’s intuitive. It was engaging. They can use these same movements to actually monitor real time how well they’re doing it. The avatar-based movements, along with the simulation, give them spatial orientation that allowed them to evaluate real time how they’re doing. All the participants engaged well, probably engaged better with that than in routine physical therapy.

Elizabeth: Maybe it’s going to be ultimately some combination of things that’s going to end up being individualized, I suspect, for people.

Rick: Yeah. They hypothesize that the next generation of digital therapies will be accessible, personalized, and scalable for at-home neurorehabilitation.

Elizabeth: We’ll look forward to it.

On that note then, that’s a look at this week’s medical headlines from Texas Tech. I’m Elizabeth Tracey.

Rick: And I’m Rick Lange. Y’all listen up and make healthy choices.

Please enable JavaScript to view the comments powered by Disqus.