The Role of Digital Phenotyping in Negative Symptoms Treatment Studies

Philip D. Harvey, PhD
Professor of Psychiatry
University of Miami Miller School of Medicine
Miami, FL

Philip D. Harvey is a Professor of Psychiatry and Behavioral Sciences at the University of Miami Miller School of Medicine. An internationally recognized expert in cognitive impairment and functional disability associated with severe mental illnesses, his research primarily focuses on schizophrenia, mood disorders, and neuropsychiatric aspects of aging. Dr. Harvey’s extensive publications have significantly advanced the understanding of cognitive deficits and their impact on daily functioning.

This video explores the use of digital phenotyping to track negative symptoms in schizophrenia. Through ecological momentary assessments (EMA) using GPS and actigraphy, the approach captures real-time behavioral data—like mobility, socialization, and activity levels. Harvey discusses its clinical relevance, especially as it enables unbiased, continuous assessment of symptom severity and treatment response, showing high adherence and reliability in capturing patient activity and mood patterns.

This presentation is part of the Emerging Approaches in Schizophrenia Editorial Focus collection from Psychiatrist.com News. The collection focuses on the latest advances in schizophrenia treatment, with an emphasis on emerging therapies that go beyond traditional dopaminergic approaches.

To learn more and watch more videos, visit our Emerging Approaches in Schizophrenia collection.

Transcript

 [00:12 – 01:06] Introduction to Digital Phenotyping in Schizophrenia

Hi, I’m Phil Harvey from the University of Miami Miller School of Medicine, and we’re going to talk about the role of digital phenotyping in negative symptoms treatment studies. There’s a lot of different ways to do digital phenotyping, and this is a field that is growing by leaps and bounds. There are both active and passive elements of digital phenotyping.

Ecological momentary assessment uses survey strategies, typically now with smartphones, where participants are asked, where are they, who are they with, what are they doing and how are they feeling, allowing for a momentary four-dimensional assessment of people’s functioning. GPS, global positioning systems, can refine the notion of home versus away, and they can validate self-report measures. People can’t walk 55 miles an hour, so if they’re going 55 miles an hour and they tell you that they’re walking, you’re getting an erroneous report.

[01:07 – 02:06] Ecological Momentary Assessment (EMA) and Its Role in Psychiatry

Measuring discrete steps is now possible, so you can actually use GPS to see where someone goes, how fast they get there, and how much effort they take in getting to where they’re going. Similarly, actigraphy, typically smart bands, can measure actual activities, including sleeping and movement, counting the number of steps and seeing whether someone is walking or running when they’re moving. So why would you want to use ecological momentary assessment for neuropsychiatric conditions? Well, self-reports of functioning seem to have real bias, particularly when recollection is required or evaluation of competence is the outcome.

EMA measures everything that’s going on at the same time, locations, moods, symptoms, activities, and reactions. It allows for concurrent and lag covariate prediction, so you can see if being happy now leads to doing more positive things later or vice versa. So there are specific relevance of digital phenotypes to negative symptoms.

[02:07 – 03:22] Behavioral Patterns in Schizophrenia: Socialization and Activity Levels

The basic constellation of negative symptoms is being home versus away, alone versus with someone, and doing unproductive or passive activities, like sitting, sleeping, resting or nothing in the middle of the day, or even failing to socialize when other people are present. How would you define improvement in a treatment study? Well, more leaving home, more being with others, a shift to socializing when others are present, and a change in the direction of positive activities when others are around. Less sedentary activities, less sleeping, less doing nothing.

So a shift in the topography of your behavior from more unproductive to more productive is really the defining case of improvement in treatment. So here’s how we do this. On the screen, it simply says, where are you? Home or away? The next screen says, are you alone or with someone? So then you can deliver a customized survey.

[03:23 – 04:28] Validating Self-Reports Through GPS and Continuous Data

And this is the home alone survey. It asks what you’re doing, all things that you can do by yourself when you’re home. You notice there aren’t any questions about socializing or interacting with other people because no one else is there.

So if you were to do a home with someone survey, you’d add in a number of socializing questions and you get to compare home alone versus home with someone to see if there’s actually any difference. What we see is the most common activities on the part of people with schizophrenia in the last hour and the last day are generally passive and unproductive. Other than for work, what we’re seeing is that 40% of all surveys are answered doing only one thing, and that is typically one single unproductive thing in the last hour.

And so in this study, we’ve validated the convergence between GPS mobility and self-reports. And as you can see in this table, we’re seeing that people with schizophrenia who tell us they’re away from home actually are. So GPS validates self-reports, making self-reports using EMA qualitatively different than subjective dispersed assessments of people telling you after the fact how good they are at things.

[04:29 – 05:39] Adherence Patterns in EMA: Implications for Long-Term Studies

  So this data is also important because GPS is continuous. So people aren’t getting paid for providing GPS signals. They just get paid for answering EMA surveys.

So the fact that the two are highly correlated really reduces the risk of bias. Things you could never learn without digital phenotyping. Identifying who is going to be adherent to their protocol on the very first day.

So let’s take a look at this. These are the correlations between adherence on the first day of a 30-day assessment protocol and then the first week and then adherence over the entire 30 days. And as you see, there’s a very strong signal of one-day adherence telling us how many surveys are going to be answered over the course of the whole 30-day period.

Further, we also see that there’s a big correlation between surveys answered in the first seven days and surveys answered over the whole 30-day period. You could say, well, maybe that’s just because all you’re doing is selecting out the people with the worst symptoms. And so by requiring adherence, you’re kicking out the people who are most symptomatic.

[05:40 – 07:32] Tracking Symptom Stability and Mood Bias in Digital Assessments

Basically, what we see is that baseline symptoms correlate zero with 30-day adherence. In both schizophrenia and bipolar depression, symptomatology at baseline in terms of severity correlate zero with 30-day adherence, meaning that symptomatology, even if it’s negative symptoms or significant depression, does not seem to be interfering with people’s ability or willingness to answer digital surveys over a 30-day period. And that’s three surveys a day for a total of 90 surveys in total.

Another thing you could never learn without digital phenotyping is that when people come to the office for a doctor’s appointment or for a clinical trial’s visit, they may be in the worst mood state of their whole week or whole month. So catching them at their worst is probably not a really good way to have your drug look like it works in a clinical trial, but let’s take a look at this data. In this study, we used GPS data that was passively and unobtrusively collected, and then we matched it to mood reports at the same time.

So the participants were not being asked, how do you feel when you leave and how do you feel when you come back? But rather, we found out when they were leaving and when they were coming back and matched their mood reports to it. So what we see is that people with schizophrenia have a negative anticipatory bias when they’re leaving home, and they feel better when they’re coming back. So what we see is that using linked, unobtrusive measurement with no possibility of bias, because the patients didn’t know what our research goals were, that you can prove that people with schizophrenia, when they come to see the doctor, are at their absolute worst over the course of the whole week that we surveyed them.

[07:33 – 08:39] Self-Reports vs. Momentary Assessment: Addressing Response Bias

So here’s another thing that you could never learn without digital phenotyping. Many studies require the recruitment of people who tell you that they’re clinically stable. That’s required for cognitive enhancement and negative symptom studies.

So what we did was we took 164 people with schizophrenia who met baseline criteria for clinical stability and saw how many psychotic symptoms they had over the course of the next 30 days. And here’s what you see. The prevalence of hallucinations, paranoia, and receiving messages from devices was surprisingly common over 30 days.

So in people who tell you they’re not hearing voices, they’re not having psychotic symptoms, they’re still having them 30 to 40% of the time over the course of the next 30 days. So patients’ self-reports at baseline, particularly if they can only sign up for a trial if they tell you certain things, are a lot less accurate than their momentary reports. The converse of that is that people will tell you when they’re having these experiences, even though they told you beforehand that they were stable.

[08:40 – 09:38] Advantages of Multidomain Digital Phenotyping for Negative Symptoms

So there’s something about ecological momentary assessment that disarms the response bias on the part of research participants. Multidomain digital phenotyping is highly relevant to negative symptoms. It’s got several strong points.

It’s feasible. There’s high adherence. It’s validated across information sources.

It’s not adversely impacted by having more severe negative symptoms. And in studies that are about to be published, it can also identify treatment changes. It can detect illness features that are impossible to capture with dispersed assessments, because as I said at the beginning, you’re capturing a four-dimensional view of the person.

Who they’re with. What are they doing. Where are they, and how are they feeling, and they don’t have to remember it.

If you ask someone two weeks later, what did you do over the last two weeks? How often were you happy when you were with other people, and how often were you sleeping when you were alone? You could never get that answer.