A new study explores using remote-measurement devices like FitBits to help former soldiers readjust to civilian life.
A newly discharged American military veteran struggles emotionally to quiet memories from the battlefield. He smokes cannabis, increasingly, to fall asleep at night and to get through the day.
Whether receiving mental health treatment or not—and half of all trauma-exposed former soldiers do not, researchers say—the ex-soldiers may benefit from wearing a Fitbit or an Apple Watch around the clock in coordination with their health care provider.
Data from such remote measurement devices could help doctors and therapists see the presence, and predict the exacerbation, of serious symptoms of post-traumatic stress disorder, or PTSD, according to the new study of dozens of US veterans over three months.
In the study, a group of 74 recently demobilized veterans agreed to wear a device on their wrist 24/7 as well as self-report through a brief daily questionnaire. The goal was to observe how the use of real-time data collected passively (through remote measurement) and actively (via a survey) could help with routing trauma sufferers to behavioral health therapists and other health care specialists as needed, says Shaddy Saba, assistant professor at the NYU Silver School of Social Work and a coauthor of the study looking at PTSD and cannabis abuse among veterans.
“Unfortunately, it is hard to predict when these problems will develop or escalate,” Saba says, explaining that the broad popularity of devices offers new promise for tracking patients.
Veterans from the US wars in Iraq and Afghanistan total over 2 million Americans. While estimates of PTSD incidence vary, a national survey in 2019-20 placed it at 9.4% of the military population, exceeding the 2016 estimate of 8.1%. PTSD is the most common condition that co-occurs with cannabis use disorder, according to research cited in this study.
Saba came to the study naturally; his research bridges behavioral health, technology, and advanced analytic methods. He first became interested in working with veterans as a social worker at the Veterans Administration in Pittsburgh. At NYU Silver, he focuses on co-occurring problems among populations facing heightened perils, particularly veterans, using both theory-guided and data-driven methods to understand how conditions arise and interact. His research is informed, too, by over a decade of clinical practice in mental health and substance use treatment settings.
For this study, Saba teamed with Daniel Leightley of Kings College London, a researcher on digital epidemiology and reservist in the UK armed forces, and Jordan Davis, Saba’s former doctoral advisor at University of Southern California who is now at RAND–among other experts.
With Veterans Day approaching, Saba digs into the potential use of simple-to-use technologies to help ease veterans’ transition to post-service life:
Broadly, what major societal problem does your study address?
When veterans return from military service, they are often at heightened risk of behavioral health challenges like post-traumatic stress disorder (PTSD) and problematic cannabis use. Put another way, these behavioral health problems are more prevalent among veterans compared to their civilian counterparts. Once these problems are severe, they become hard to treat. Unfortunately, it is also hard to predict which veterans will develop these problems, making them even harder to prevent.
Participants were asked to self-report every day on things like their mood, stress, and social contact through a mobile app called MAVERICK designed by Dr. Leightley, while their wearable devices such as Apple Watch or Fitbit automatically recorded things like their heart rate, physical activity, and sleep. Our main question is whether it is feasible to use machine learning with these passive and active data sources to predict PTSD symptoms or cannabis use problems in the future.
How confident in the observational findings are you at this point?
Regarding the feasibility of remote monitoring using machine learning with smartphone and wearable device data, our experience was encouraging. Our participants provided self-reported data on their behavioral health symptoms on nearly 70% of days, which is a pretty high response rate for this kind of study.
Completion rates for the wearable data were a bit more mixed: while we collected a lot of data from some wearable streams like steps and distance traveled, other things like sleep were less consistently collected. While it still seems like we collected enough data to make accurate predictions, our experience suggests we need to think more about implementation challenges with this technology. For example, some veterans may have been removing their devices before sleep, and others may have found the three-month time frame of the study intrusive. Additionally, several veterans we recruited told us they would be concerned if their data was being shared with the VA, suggesting there might be additional barriers to implementing these tools in real-world clinical settings
Aside from this feasibility paper, our team has several main-results papers in preparation or review. In these papers, we are demonstrating that both wearable device data and daily self-report data can, in fact, help to accurately predict behavioral health problems when they are included in machine learning algorithms. We have several papers forthcoming on successful PTSD prediction with machine learning, and we are beginning analyses predicting cannabis use.
Are those analyses under way because cannabis use can indicate that someone is struggling with PTSD?
It is likely the case that some veterans use cannabis to cope with increasing PTSD, and we have shown evidence for this in some of our other work with veterans. But this was not exactly why we included cannabis in this study. Given that veterans are at risk of both problems—PTSD and cannabis use problems—we primarily wanted to see how machine learning and novel data could be used to predict each of them individually, not necessarily how PTSD and cannabis use are related to one other.
Your study notes that remote measurement is something of a new frontier.
There’s a growing trend in exploring the use of remote measurement for mental health problems, but it is still a nascent field with a lot of unanswered questions, which is why we started with a feasibility study. The study team saw a gap in exploring this technology with veterans in particular, given their unique and complex symptom profiles and the importance of assessing risk at specific critical periods, such as following their return from deployment.
Now that 4,300 days of data have been collected for this study, we are systematically developing and testing machine learning algorithms to begin to understand things like which data streams are most important for predicting PTSD or cannabis use, and which specific machine learning methods are most useful for technology like this.
The veterans were recruited for the study through social media. Those chosen met criteria such as recent cannabis use and PTSD symptoms. In the end, what did you find most surprising?
It was encouraging to see that veterans are open to using remote monitoring technology for an extended period, especially given concerns some might have about potential barriers like stigma and behavioral health privacy. I was also excited for our team to begin publishing substantive results, which provide proof of concept.
At the same time, scaling up the use of remote monitoring technology may bring new challenges. We should be cautious. Privacy concerns and possible user fatigue may become more significant when implementing these tools more broadly or in larger health systems. Those who are older, less comfortable with technology, or especially worried about privacy might experience greater barriers or have different results. We will need to better understand and address these challenges before widespread deployment.
Source: NYU
