Optimizing Psychiatric Care Through Artificial Intelligence, Wearables, and Mobile Apps
Digital technologies have had a significant impact on many areas of healthcare, including improved access to care through telemedicine and data analytics along with artificial intelligence tools that help researchers uncover new patterns and insights.
Increasingly, the field of psychiatry is employing many of these technologies to predict psychiatric conditions and states. Wearable devices and mobile apps are just two of the tools that psychiatrists can use to monitor and optimize the treatment of patients. Researchers at Mass General Brigham are at the forefront of developing many of these technologies, which use network analysis, large language models, and machine learning to track patients' activities and help provide the best care.
"The whole diagnostic framework in mental health is based on what the patient tells you and what you observe as a clinician," says Daniel Barron, MD, PhD, director of the Pain Intervention and Digital Research Program at Brigham and Women's Hospital. "Other fields are able to use more quantitative measures, like EKGs in cardiology and molecular tests in oncology. In psychiatry, we're still trying to learn which measurements are most relevant and what we can do with that data once we've measured it."
The following is a review of some areas toward which Mass General Brigham investigators are directing their efforts.
Wearables and Apps Help Monitor Depressive Symptoms
Paola Pedrelli, PhD, associate director at the Depression Clinical and Research Program at Massachusetts General Hospital, is focused on developing new technologies to assess symptoms of depression. Among her projects, she is working with the MIT Media Lab's Rosalind Picard, ScD, to find ways to employ wearable devices such as wristbands and smartphone sensors to assess and monitor patients' long-term levels of depression.
"If someone is depressed, they are likely to have low energy and be less active," Dr. Pedrelli says. "We can use smartphones and wearable devices to actually measure how active they have been by examining how many steps they have taken and how often they have left their home. We can also use apps on their phones to detect how social they are by examining whether they have been making calls or sending texts."
The ultimate goal of this work is to create models that can more objectively detect and monitor patient symptoms.
Dr. Pedrelli conducted a pilot study examining the feasibility of using wearables and apps to monitor depressive symptoms. In that study, patients agreed to be monitored for two months using wearables as well as a phone app that monitors socialization levels. Additionally, participants completed weekly interviews with a clinician, during which their depression was assessed with the 17-item Hamilton Depression Rating Scale (HDRS-17).
Machine learning models were developed that combined input from the wearable sensors and the app, and the data was correlated with the HDRS-17. With machine learning, the concept of "ground truth" refers to the reality of what's actually happening—something that's important when using things like GPS data to measure and interpret behavior. Here, the HDRS-17 was used as ground truth. Findings from the study, which suggested that this was a promising approach, were published in 2020 in Frontiers in Psychiatry.
Subsequently, Drs. Pedrelli and Picard were funded by the National Institute of Mental Health to conducted a follow-up study. In this study, the technology was used to evaluate patients with depression who recently had a change in their treatment regimen, such as starting a new medication, starting therapy, or changing medication dosage.
"The pilot study showed that using these technologies is feasible, but because it included primarily patients who were chronically depressed, it didn't allow us to say with confidence whether our machine learning models could accurately detect changes in their symptoms," Dr. Pedrelli explains. "We just completed the data collection for the second study, where we observed that many participants experienced changes in symptom severity. We plan to start data analyses soon and hope that we will be able to develop a model that will prove useful for the management of patients."
The ultimate goal is to create an automated detection system in which the patient and/or doctors would receive a notification if the patient's symptoms were worsening. "We are not there yet, but we hope that eventually there will be algorithms that can be applied in this area," Dr. Pedrelli concludes.
Investing in Digital Tools for Better Diagnosis
In his research, which is funded by the National Institute on Aging, Dr. Barron is developing tools to measure changes in relation functional status (an index of mobility, emotion, and sociability) in older patients with chronic pain conditions. He is the author of the book Reading Our Minds: The Rise of Big Data in Psychiatry as well as many papers on the topic. He is also looking at the value of wearables to monitor things like patients' activity levels and sleep.
"Sleep is a low-hanging fruit, but we know it can tell us a lot," he says. "With pretty much any psychiatric condition, you have some level of sleep disturbance, whether patients are sleeping too much or not enough. This is a quantitative measure that we can trace over time and combine with other lab values that have clinical relevance."
Gathering data is only the first step; a parallel line of research is developing data management strategies and models to make sense of the data. Ideally these decision models will deploy algorithms that can learn about an individual patient's baseline behavior and be able to help identify when there have been changes from that baseline.
"Digital tools can help psychiatry by making it easier to quantify data already present from standard clinical interactions and by allowing decision models to be operationalized and evaluated," Dr. Barron says. "They can also help us test whether new forms of data, such as what we are able to collect from these wearable devices, might have value."
Dr. Barron acknowledges the challenges of developing such tools, including the financial resources needed to evaluate these models and build infrastructure. "The actual, real-world implementation of digital tools is not something that's going to be—or should be—funded by research grants," he notes. "We need deeper investment from the Centers for Medicare & Medicaid Services as well as from major insurance companies."
Such investment could provide a lot of value, Dr. Barron argues. He adds that these tools could be especially useful in providing care for people living in rural areas who don't have regular access to in-person clinic visits as well as for older patients and those with chronic pain and other chronic medical conditions. "If we can detect a problem sooner, we can act on it sooner, and ultimately that could save money," he says.
Much of this research falls under the field of digital phenotyping, which involves using data patterns to determine the best intervention for each individual patient. This includes identifying patterns in wearable device or smartphone use and connecting those patterns to specific health behaviors.
"If we can collect enough information from a patient, we can reasonably say whether they fit into one of these data-driven patterns," Dr. Barron says. "The challenge right now is developing the tools for designing and testing these patterns."
Finding Ways for Psychiatry to Become More Proactive
Brigham emergency medicine physician Peter Ray Chai, MD, MS, focuses many of his efforts on the use of digital technology in psychiatry and in medicine more broadly.
One area that he has studied is how ingestible electronic sensors can be used to monitor medication adherence. This technology so far has been applied largely in the HIV setting, but he sees value in using it to monitor whether psychiatric patients are regularly taking their medications. The technology employs pills labeled with a radio sensor that emits a signal that's detectable by a wearable device.
"We may be able to connect this to some kind of app that would remind patients to take their medications if they forget," Dr. Chai says.
Using digital phenotyping, it may also be possible to link data about medication adherence to other measures that can be tracked via a smartphone. "We can get a GPS track of an individual's typical day and potentially sense deviations that could contribute to phenotypic risks for changes in mental health or substance abuse issues," he says. "We're hoping to build out these technologies."
Digital phenotyping is also useful in syndemics, or evaluating the interactions of co-occurring social and health conditions that can contribute to worse health outcomes. It has particular applications in the context of substance abuse, in which GPS data could track patients going to specific locations that have been linked with past substance abuse. Wearable devices can also detect biometric changes associated with substance abuse, including changes in heart rate and body temperature.
Dr. Chai explains that digital phenotyping generates massive data sets, which require the development of AI models that can interpret these data and distinguish between usable data and noise. In this work, too, ground truth is important.
"In psychiatry, the challenge is in understanding how to get close to that ground truth," he says. "This is where digital phenotyping can be so valuable. There are many models that can allow us to explore how these things eventually get integrated into clinical care."
One anticipated benefit of this work, Dr. Chai notes, is to help the field of psychiatry become more proactive. "A lot of what we do in mental health and behavioral science is reactive," he says. "We wait for patients to do something, and then we respond. A big thrust of our digital phenotyping work is to be able to predict or anticipate when patients are likely to have problems and to intervene before it's too late.”