This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR mHealth and uHealth, is properly cited. The complete bibliographic information, a link to the original publication on http://mhealth.jmir.org/, as well as this copyright and license information must be included.
Smartphone-delivered cognitive behavioral therapy (CBT) is becoming more common, but research on the topic remains in its infancy. Little is known about how people typically engage with smartphone CBT or which engagement and mobility patterns may optimize treatment. Passive smartphone data offer a unique opportunity to gain insight into these knowledge gaps.
This study aimed to examine passive smartphone data across a pilot course of smartphone CBT for body dysmorphic disorder (BDD), a psychiatric illness characterized by a preoccupation with a perceived defect in physical appearance, to inform hypothesis generation and the design of subsequent, larger trials.
A total of 10 adults with primary diagnoses of BDD were recruited nationally and completed telehealth clinician assessments with a reliable evaluator. In a 12-week open pilot trial of smartphone CBT, we initially characterized natural patterns of engagement with the treatment and tested how engagement and mobility patterns across treatment corresponded with treatment response.
Most participants interacted briefly and frequently with smartphone-delivered treatment. More frequent app usage (
The sample was small in this pilot study; thus, results should be used to inform the hypotheses and design of subsequent trials. The results provide initial evidence that frequent (even if brief) practice of CBT skills may optimize response to smartphone CBT and that mobility patterns may serve as useful passive markers of symptom severity. This is one of the first studies to examine the value that passively collected sensor data may contribute to understanding and optimizing users’ response to smartphone CBT. With further validation, the results can inform how to enhance smartphone CBT design.
The supply and demand imbalance between those who need psychological treatment and those who are able to receive it represents a serious public health concern [
Fortunately, the development of smartphone-delivered CBT treatments may help address this access gap. Compared with in-person therapy, smartphone-delivered CBT is less expensive, more widely accessible, and highly flexible (eg, it can be used anywhere and anytime patients have their phones). The potential benefits of smartphone-delivered CBT are compounded by the growth of smartphone ownership. At present, 81% of the US population own a smartphone, a rate that has more than doubled since 2011 [
Despite growing excitement, our understanding of smartphone-delivered CBT remains in its infancy, with a dramatic gap between the number of publicly available mental health apps and the paucity of scientific papers reporting on their evaluation [
Moreover, very little is known about which engagement patterns correspond with an optimal response to smartphone-delivered CBT. Understanding optimal engagement patterns can allow for the design of more potent treatments by seeking to promote the most effective patterns of CBT app use. For example, gaining information about whether one’s frequency of use or duration of use matters more in terms of treatment response can inform whether apps should be designed to promote bursts of brief engagement or longer, less frequent sessions.
Finally, little is currently known about how the mobility patterns of patients change over the course of smartphone-delivered CBT. Previous research suggests that time spent at home, measured via a GPS, can serve as a digital marker of avoidance [
Altogether, in the field’s current, early stage of developing smartphone-delivered CBT treatments, we can benefit from examining pilot engagement and mobility data, to shape how we design optimal digital services and their clinical trials in the future. Smartphones offer a unique avenue for gaining rich insights into patterns of treatment engagement and predictors of treatment response because smartphones can unobtrusively (ie, in the background, without user input) collect a wide variety of sensor-based data over the course of treatment. For example, with patient consent, smartphones can be configured to passively collect objective information about patients’ engagement with the app (ie, how often and for how long patients use the program) as well as patients’ behavioral patterns over the course of treatment (eg, where patients typically use the app, changes in mobility patterns across treatment, via GPS). Passive data offer notable strengths for learning how to optimize smartphone-delivered treatments compared with more traditional assessment methods such as clinician interviews and self-reports. Passive smartphone data are sampled at a far greater frequency than traditional clinical assessments, which, at most, might be administered weekly. Frequent assessment that is conducted as one lives daily life captures richer contextual information, has higher temporal resolution to detect changes in symptoms or severity, and reduces the influence of recall biases that arise from subjective recollection of experiences over a broad time frame [
To this end, this study exploratorily examines passive smartphone data from a 12-week open pilot trial of a smartphone-delivered CBT (
A paper by Wilhelm et al [
A total of 10 adults with a primary psychiatric diagnosis of BDD were enrolled nationally in the open pilot trial (female: n=8, male: n=2; mean age 27.6, SD 5.66 years). Other inclusion criteria required that participants had at least moderately severe BDD symptoms (defined as a Yale-Brown obsessive compulsive scale modified for BDD [BDD-YBOCS] score >20), an acuity level appropriate for an outpatient level of care and lived in the United States. Exclusion criteria prohibited participation if the individual had a current severe major depressive disorder; borderline personality disorder; substance use disorder or acute, active suicidal ideation; had a lifetime diagnosis of bipolar disorder or a psychotic disorder; had cognitive impairment or intellectual disability that would interfere with participation; had engaged in previous CBT for BDD, or did not own an iPhone that supported the app software. Participants were either unmedicated or those on medication were required to be on a stable dose for at least two months before starting the study and were instructed not to change their medication regimen during the trial.
Procedures were approved by the hospital’s institutional review board, and participants provided informed consent before beginning the study. Informed consent included a description of each type of passive smartphone data to be collected, a description of how those data were securely transmitted and deidentified before storage, the rationale for collecting those data, and a description of who would have access to the data.
Clinical assessments were conducted by reliable, independent evaluators with a Master’s degree or doctorate, who were trained in primary diagnostic and outcome measures. Assessments for this study were conducted at the screening and baseline (same visit; week 0), midpoint (week 6), and posttreatment (week 12) assessments, and participants were compensated US $25 for completing the week 6 and week 12 assessments. Clinician-administered measures were collected via secure video calls that were Health Insurance Portability and Accountability Act (HIPAA) compliant. Self-report data were collected via Research Electronic Data Capture [
In addition to providing clinical and outcome data, participants also provided qualitative feedback on the CBT app at several time points across the study. Specifically, written feedback was collected at the posttreatment assessment; oral feedback was gathered by members of the design team via separate interviews conducted shortly after the baseline, midpoint, and posttreatment clinical assessments.
Following the screening and baseline assessment, the study staff instructed eligible participants on how to download and activate the
The Mini-International Neuropsychiatric Interview (version 7.0.2) [
The BDD-YBOCS [
To quantify and analyze the patterns of engagement with
The quantity of app use was calculated as the total duration in minutes that a participant used the app. This was calculated by adding together all app
This metric measured the extent to which a participant tended to use the app frequently or infrequently, expressed as the mean duration between 2 consecutive sessions, or periods of uninterrupted use. Frequency of app use was computed for the first half (6 weeks) and the full 12 weeks of the CBT program (
Using GPS data, we calculated the percentage of time spent at home during 1-week time intervals that overlapped with baseline, midpoint, and posttreatment BDD-YBOCS assessments (including 3 days before, 3 days after, and the day of BDD-YBOCS administration). Of note, at baseline, the BDD-YBOCS was typically administered on the same day the app was installed. Therefore, GPS data were not generally available for the 3 days before the baseline BDD-YBOCS assessment. Home location was inferred as the most common location ID captured between 3 AM and 6 AM per individual. All the remaining location IDs were labeled as
Data were analyzed using Python 3.6 (Python Software Foundation).
To characterize the overall patterns of app usage, we visually inspected longitudinal patterns of usage by the participants across the 12-week treatment and we calculated the number of times the participants were engaged with the app for different lengths of time (ie, session durations). We elected not to identify subsamples based on usage (ie, clusters of users with similar engagement patterns) either visually or quantitatively, because of the small sample size.
To examine how the patterns of engagement of participants with
Two bivariate correlations were conducted, to preliminarily explore the relationships between the variables measuring (a) quantity and (b) frequency of app usage with percentage improvement in BDD-YBOCS from the baseline to week 12. Next, to initially examine the relative effect of quantity versus frequency of app use, a regression analysis of percent improvement was conducted, with both quantity and frequency of app use as independent variables. We primarily evaluated effect sizes, as opposed to statistical significance, for correlation and regression analyses, given the pilot nature of the data.
The relationship between symptom severity and mobility was explored via a bivariate correlation between BDD-YBOCS scores and the percentage of time spent at home during the week the BDD-YBOCS was measured. Note that absolute BDD-YBOCS scores were used for this analysis instead of percentage improvement, given the goal of exploring the predictive power of a GPS marker in assessing the current acuity of participants. The correlation analysis included 30 pairs of location variables and BDD-YBOCS scores (ie, 3 per participant, at baseline, midpoint, and posttreatment); thus, each participant was equally represented in the correlation analysis. Given that this analysis included multiple time points per participant, we followed up with a secondary analysis to verify that the results were not inflated based on the longitudinal nature of the data. Namely, 6000 correlation analyses were run by randomly selecting 1 of the 3 time points per participant (pre-, mid-, or posttreatment). This approach resulted in a very similar median correlation value to the analysis with 3 time points per participant; thus, secondary results are not presented. Again, we primarily evaluated the effect size, as opposed to statistical significance, for this correlation analysis, to best account for the pilot nature of the study.
Wilhelm et al [
We visually examined the longitudinal patterns of engagement with
Individual patterns of engagement at a daily level. Usage is displayed as an aggregated sum of total minutes per day.
Despite the diversity in app usage across participants, several common usage patterns also emerged. First, most participants used the app with higher and lower intensities in the first and last weeks of the treatment, respectively (
Moreover, unlike in-person therapy, most interactions with
Means, standard deviations, and bivariate correlations of the percentage improvement in the BDD-YBOCS, the quantity of app usage, and frequency of app usage are provided in
Descriptive statistics and correlations between patterns of engagement with smartphone-delivered cognitive behavioral therapy for body dysmorphic disorder and treatment response.
App usage patterns | Midtreatment, mean (SD) | Posttreatment, mean (SD) | Correlation with percentage improvement in BDD-YBOCSa (baseline to week 12)b | |
Percentage improvement in BDD-YBOCS (%) | 32.3 (23.8) | 45.3 (14.7) | N/Ac | N/A |
Quantity of app use | 293.9 (276.4) | 398.0 (310.25) | –0.084 | .82 |
Frequency of app used | 6.24 (0.7) | 6.44 (0.73) | –0.57 | .08 |
aBDD-YBOCS: Yale-Brown obsessive compulsive scale modified for body dysmorphic disorder.
bCorrelations calculated for quantity and frequency of app use were derived from full 12-week treatment.
cN/A: not applicable.
dFrequency of app use values are log-transformed. Nontransformed means at midtreatment and posttreatment are 512 min (SD 1 255 to 1032 min) and 626 min (SD 1 301 to 1299 min), respectively.
The quantity of app usage was uncorrelated with percentage improvement in the BDD-YBOCS, whereas the frequency of app usage correlated strongly with treatment response and trended toward significance. The strong, negative relationship between mean (log) length of breaks between sessions (ie, frequency of app use) and improvement in the BDD-YBOCS initially suggests that shorter breaks between sessions corresponded with greater improvements (
To follow up on patterns elucidated in bivariate correlations, we used regression analysis to preliminarily examine whether the frequency of app usage corresponded with treatment response more so than the quantity of app usage. When the primary outcome (percentage improvement in the BDD-YBOCS) was entered as a dependent variable, the frequency of app usage (ie, mean (log) duration between 2 consecutive sessions) predicted percentage improvement in the BDD-YBOCS with a small effect (beta=–0.13;
We used a scatterplot to visually inspect the relationship between time spent at home (based on GPS data) and symptom severity (measured with the BDD-YBOCS;
Body dysmorphic disorder severity and time spent at home across treatment. BDD-YBOCS: Yale-Brown obsessive compulsive scale modified for body dysmorphic disorder.
Although enthusiasm for smartphone-delivered CBT is growing rapidly, there has not yet been substantial research on ways to enhance smartphone treatment. Before the widespread development and deployment of smartphone CBT treatments, it is important to first examine pilot data that characterizes the natural engagement patterns of users with smartphone-delivered CBT and identifies which usage and mobility patterns may optimize treatment. Such pilot data will provide timely information to researchers about variables and hypotheses of focus, in advance of larger, more costly validation trials, and can elucidate how we may explore enhancing smartphone-delivered CBT for optimum response in larger trials.
In particular, passively collected usage and sensor data from smartphones offer a unique, low-burden approach for gaining these important insights. Although a variety of passive data (eg, typing speed, activity level, phone usage, acoustic level) can be collected by smartphones [
Although app usage patterns varied substantially across participants, visual examination and descriptive analysis of usage data revealed several common patterns of engagement in our sample. First, participants tended to use the app more frequently and for a greater overall duration at the beginning of the 12-week treatment, with considerably lower usage later in treatment. This result is not surprising and may reflect that early on, participants required more time on the app to learn new information and skills. Later in the treatment, the participants may have transitioned to practicing greater applied skills, offline and in the real world [
Second, descriptive results highlighted that participants typically used the app at home during the first two-thirds of treatment; later, the participants tended to use the app more when out of the house. This within-person pattern of increased usage outside of the house over time is consistent with the hypothesis that as participants gained CBT skills across treatment, they may have transitioned to using those skills offline and in the real world.
Finally, we observed that overall, the participants tended to use the app in brief and frequent sessions. In fact, most app sessions lasted <5 min each. This pattern reflects the way in which most people use smartphones in general: engaging with them often during short moments of downtime throughout the day [
On the other hand, this pattern of brief and frequent sessions is notably distinct from how patients engage with face-to-face CBT or ICBT. Given the distinctive pattern of engagement we observed compared with better-established CBT modalities, it is critical to examine whether the naturally brief usage patterns of participants with smartphone-delivered CBT are effective or whether longer sessions are needed for response. Interestingly, preliminary correlation and regression results suggest that more frequent app usage, as opposed to greater duration of app usage, correlated strongly with treatment response—and trended toward statistical significance—in our (albeit small) sample. Consistent with these results, a previous review showed that overall time spent on web-based treatments for depression does not typically correlate with response to treatment [
It is possible that frequent doses of practice help with learning CBT skills, as regular reinforcement of skills across broad contexts may enhance consolidation and generalization [
In addition to usage patterns, we also examined mobility patterns from GPS data that correspond to BDD severity. Preliminary results showed that across treatment, the proportion of time spent at home—a potential digital marker of avoidance [
Altogether, strong initial GPS results underscore one variable where gains of data collection may outweigh costs; researchers designing upcoming smartphone-delivered CBT trials should consider measuring time spent at home, to further validate this potential unobtrusive marker of clinical severity. With further validation, detecting changes in one’s time spent at home could enhance smartphone-delivered CBT by unobtrusively triggering just-in-time interventions—a promising yet underdeveloped area of research [
Results from this study should be interpreted, bearing in mind its limitations. Most notably, this pilot study had a small sample size. Thus, it is possible for 1 or 2 participants’ outlying usage patterns to unduly influence the results. That said, Kazdin [
In addition to a small sample, this pilot trial specifically focused on smartphone-delivered CBT for BDD. It is possible that insights will not generalize to smartphone-delivered CBT treatments for other disorders. However, given the core similarities between CBT for BDD and many other psychiatric conditions, such as anxiety disorders, obsessive-compulsive–related disorders, and eating disorders, we anticipate that findings will be relevant in the design of smartphone-delivered CBT treatments for related conditions. Finally, our strong initial GPS results should be interpreted, bearing in mind the high degree of missing GPS data (683/1134, 60.23% of the days) in our sample. Although the specific reasons for missing GPS data in our study are unknown, a high rate of missing geolocation data in mobile research is typical (eg, ranging from 40% to 90% missing) [
This study also had several notable strengths. First, whereas many existing smartphone-delivered CBT trials use nonclinical or convenience samples, we used a clinical sample that was diagnosed and assessed via gold-standard, clinician-administered measures. Participants were recruited nationally, which may enhance the generalizability of our initial findings. Finally, the correlation results for app usage and GPS patterns were robust despite our small sample, suggesting that these novel insights have strong potential to enhance costly, well-powered future trials.
Altogether, the results suggest that as researchers design efficacy trials to test smartphone-delivered CBT, it is worthwhile to collect data on patterns of use (with a focus on frequency versus quantity of use) and time spent at home. Novel study results suggest that these variables may correspond meaningfully with the response to treatment and, with further validation, may inform how to enhance smartphone-delivered CBT interventions.
Distribution of session durations aggregated across the sample.
body dysmorphic disorder
Yale-Brown obsessive compulsive scale modified for body dysmorphic disorder
cognitive behavioral therapy
Health Insurance Portability and Accountability Act
internet-based cognitive behavioral therapy
The authors would like to thank Remko Vermeulen, Anna Hiltunen, Thomas McCoy, Berta Summers, and Ilana Ladis for their work on this study. SW served as the principal investigator and designed the trial, with HW and JG. HW and AM conceptualized this paper. HW and AM wrote the manuscript. AM and RC conducted the statistical analyses. All authors read, edited, and approved the manuscript before submission. This work was supported by Telefonica Innovation Alpha.
The sponsor and investigators from Massachusetts General Hospital collaborated in the development of the Perspectives digital service, and they collaborated on the analysis and writing of this manuscript. HW, JG, and SW received salary support from Telefonica Innovation Alpha and are presenters for the Massachusetts General Hospital Psychiatry Academy in educational programs supported through independent medical education grants from pharmaceutical companies. SW has received royalties from Elsevier Publications, Guilford Publications, New Harbinger Publications, and Oxford University Press. SW has also received speaking honoraria from various academic institutions and foundations, including the International Obsessive Compulsive Disorder Foundation and the Tourette Association of America. In addition, SW received payment from the Association for Behavioral and Cognitive Therapies for her role as Associate Editor for the Behavior Therapy journal as well as from John Wiley & Sons, Inc for her role as Associate Editor on the journal Depression & Anxiety. OH, RG, and AM are employees of Telefonica Innovation Alpha.