Anna N. Baglione

Background: Breast cancer is a significant public health burden. Despite its critical role in preventing the recurrence of breast cancer, rates of long-term adherence to endocrine therapy (ET) remain low among certain breast cancer survivors. Using embedded sensors in smartphones and wearables, ecological momentary assessment data and health behavior theory may facilitate a richer understanding of the real-world context of medication-taking behaviors, which can aid in the development of personalized interventions. Objective: The objective of this paper is to describe the development of a multiscale modeling intervention (MMI) system to facilitate adherence to daily oral ET for breast cancer survivors. This represents the first phase of a larger project that aims to use machine learning to predict when breast cancer survivors are most likely to miss their ET medications in order to deploy personalized interventions. The purpose of this paper was (1) to determine the acceptability of the proposed MMI system, (2) to identify modifiable predictors of ET medication adherence among breast cancer survivors, and (3) to select surveys or items measuring constructs associated with ET adherence among breast cancer survivors for inclusion in the MMI system. Methods: Study 1 consisted of usability interviews with a cohort of breast cancer survivors (n=25) prescribed ET. For study 1, all qualitative usability interviews were conducted using a semistructured interview guide and assessed whether breast cancer survivors were willing to use various components of the MMI system. Study 2 consisted of (1) a secondary data analysis of ET adherence data from 32 breast cancer survivors using a social cognitive theory framework and (2) a review of research literature of constructs and surveys measuring constructs associated with ET adherence among breast cancer survivors using a social cognitive theory framework. The secondary data analysis included the use of randomized neural network analysis to predict factors strongly associated with medication adherence. Results: In study 1, usability interview findings suggested that participants were willing to use an ecological momentary assessment smartphone app, a smartwatch and associated smartphone app, a smart pill bottle or smart pill box and associated smartphone app, and the entire MMI system for a 6-month study period. In study 2, the randomized neural network analysis identified 104 survey items with significant contributions to 4-week medication adherence using a threshold of the 70th percentile for feature importance. After a review of peer-reviewed studies, we abstracted modifiable constructs significantly associated with adherence to adjuvant ET and identified 42 surveys used to measure these constructs. When these findings were combined, the final survey for the MMI system consisted of 32 surveys and demographic items. Conclusions: Our research highlights the use of social cognitive theory, data-driven models, and participant feedback to inform the development of a medication adherence monitoring system. Data from studies 1 and 2 were used to develop a prototype MMI system that will be deployed in a future longitudinal study with 20 breast cancer survivors over 6 months.

Background: Digital mental health is a promising paradigm for individualized, patient-driven health care. For example, cognitive bias modification programs that target interpretation biases (cognitive bias modification for interpretation [CBM-I]) can provide practice thinking about ambiguous situations in less threatening ways on the web without requiring a therapist. However, digital mental health interventions, including CBM-I, are often plagued with lack of sustained engagement and high attrition rates. New attrition detection and mitigation strategies are needed to improve these interventions. Objective: This paper aims to identify participants at a high risk of dropout during the early stages of 3 web-based trials of multisession CBM-I and to investigate which self-reported and passively detected feature sets computed from the participants interacting with the intervention and assessments were most informative in making this prediction. Methods: The participants analyzed in this paper were community adults with traits such as anxiety or negative thinking about the future (Study 1: n=252, Study 2: n=326, Study 3: n=699) who had been assigned to CBM-I conditions in 3 efficacy-effectiveness trials on our team’s public research website. To identify participants at a high risk of dropout, we created 4 unique feature sets: self-reported baseline user characteristics (eg, demographics), self-reported user context and reactions to the program (eg, state affect), self-reported user clinical functioning (eg, mental health symptoms), and passively detected user behavior on the website (eg, time spent on a web page of CBM-I training exercises, time of day during which the exercises were completed, latency of completing the assessments, and type of device used). Then, we investigated the feature sets as potential predictors of which participants were at high risk of not starting the second training session of a given program using well-known machine learning algorithms. Results: The extreme gradient boosting algorithm performed the best and identified participants at high risk with macro–F1-scores of .832 (Study 1 with 146 features), .770 (Study 2 with 87 features), and .917 (Study 3 with 127 features). Features involving passive detection of user behavior contributed the most to the prediction relative to other features. The mean Gini importance scores for the passive features were as follows: .033 (95% CI .019-.047) in Study 1; .029 (95% CI .023-.035) in Study 2; and .045 (95% CI .039-.051) in Study 3. However, using all features extracted from a given study led to the best predictive performance. Conclusions: These results suggest that using passive indicators of user behavior, alongside self-reported measures, can improve the accuracy of prediction of participants at a high risk of dropout early during multisession CBM-I programs. Furthermore, our analyses highlight the challenge of generalizability in digital health intervention studies and the need for more personalized attrition prevention strategies.

Web-based cognitive bias modification for interpretation (CBM-I) can improve interpretation biases and anxiety symptoms but faces high rates of dropout. This study tested the effectiveness of web-based CBM-I relative to an active psychoeducation condition and the addition of low-intensity telecoaching for a subset of CBM-I participants. 1,234 anxious community adults (Mage = 35.09 years, 81.2% female, 72.1% white, 82.6% not Hispanic) were randomly assigned at Stage 1 of a sequential, multiple-assignment randomized trial to complete five weekly sessions of CBM-I or psychoeducation on our team's public research website. After the first session, for Stage 2, an algorithm attempted to classify CBM-I participants as higher (vs. lower) risk for dropping out; those classified as higher risk were then randomly assigned to complete four brief weekly telecoaching check-ins (vs. no coaching). As hypothesized (https://doi.org/j2xr; Daniel, Eberle, & Teachman, 2020), CBM-I significantly outperformed psychoeducation at improving positive and negative interpretation biases (Recognition Ratings, Brief Body Sensations Interpretation Questionnaire) and anxiety symptoms (Overall Anxiety Severity and Impairment Scale, Anxiety Scale from Depression Anxiety Stress Scales-Short Form), with smaller treatment gains remaining significant at 2-month follow-up. Unexpectedly, CBM-I had significantly worse treatment dropout outcomes than psychoeducation, and adding coaching (vs. no coaching) did not significantly improve efficacy or dropout outcomes (notably, many participants chose not to interact with their coach). Web-based CBM-I appears effective, but supplemental coaching may not mitigate the challenge of dropout. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Objective: Web-based cognitive bias modification for interpretation (CBM-I) can improve interpretation biases and anxiety symptoms but faces high rates of dropout. This study tested the effectiveness of web-based CBM-I relative to an active psychoeducation condition and the addition of low-intensity telecoaching for a subset of CBM-I participants. Method: 1,234 anxious community adults (M_age = 35.09 years, 81.2% female, 72.1% white, 82.6% not Hispanic) were randomly assigned at Stage 1 of a sequential, multiple-assignment randomized trial to complete five weekly sessions of CBM-I or psychoeducation on our team’s public research website. After the first session, for Stage 2, an algorithm attempted to classify CBM-I participants as higher (vs. lower) risk for dropping out; those classified as higher risk were then randomly assigned to complete four brief weekly telecoaching check-ins (vs. no coaching). Results: As hypothesized (https://doi.org/j2xr; Daniel, Eberle, & Teachman, 2020), CBM-I significantly outperformed psychoeducation at improving positive and negative interpretation biases (Recognition Ratings, Brief Body Sensations Interpretation Questionnaire) and anxiety symptoms (Overall Anxiety Severity and Impairment Scale, Anxiety Scale from Depression Anxiety Stress Scales–Short Form), with smaller treatment gains remaining significant at 2-month follow-up. Unexpectedly, CBM-I had significantly worse treatment dropout outcomes than psychoeducation, and adding coaching (vs. no coaching) did not significantly improve efficacy or dropout outcomes (notably, many participants chose not to interact with their coach). Conclusions: Web-based CBM-I appears effective, but supplemental coaching may not mitigate the challenge of dropout.

BACKGROUND Digital mental health is a promising paradigm for individualized, patient-driven health care. For example, cognitive bias modification programs that target interpretation biases (cognitive bias modification for interpretation [CBM-I]) can provide practice thinking about ambiguous situations in less threatening ways on the web without requiring a therapist. However, digital mental health interventions, including CBM-I, are often plagued with lack of sustained engagement and high attrition rates. New attrition detection and mitigation strategies are needed to improve these interventions. OBJECTIVE This paper aims to identify participants at a high risk of dropout during the early stages of 3 web-based trials of multisession CBM-I and to investigate which self-reported and passively detected feature sets computed from the participants interacting with the intervention and assessments were most informative in making this prediction. METHODS The participants analyzed in this paper were community adults with traits such as anxiety or negative thinking about the future (Study 1: n=252, Study 2: n=326, Study 3: n=699) who had been assigned to CBM-I conditions in 3 efficacy-effectiveness trials on our team’s public research website. To identify participants at a high risk of dropout, we created 4 unique feature sets: self-reported baseline user characteristics (eg, demographics), self-reported user context and reactions to the program (eg, state affect), self-reported user clinical functioning (eg, mental health symptoms), and passively detected user behavior on the website (eg, time spent on a web page of CBM-I training exercises, time of day during which the exercises were completed, latency of completing the assessments, and type of device used). Then, we investigated the feature sets as potential predictors of which participants were at high risk of not starting the second training session of a given program using well-known machine learning algorithms. RESULTS The extreme gradient boosting algorithm performed the best and identified participants at high risk with macro–F1-scores of .832 (Study 1 with 146 features), .770 (Study 2 with 87 features), and .917 (Study 3 with 127 features). Features involving passive detection of user behavior contributed the most to the prediction relative to other features. The mean Gini importance scores for the passive features were as follows: .033 (95% CI .019-.047) in Study 1; .029 (95% CI .023-.035) in Study 2; and .045 (95% CI .039-.051) in Study 3. However, using all features extracted from a given study led to the best predictive performance. CONCLUSIONS These results suggest that using passive indicators of user behavior, alongside self-reported measures, can improve the accuracy of prediction of participants at a high risk of dropout early during multisession CBM-I programs. Furthermore, our analyses highlight the challenge of generalizability in digital health intervention studies and the need for more personalized attrition prevention strategies.

In recent years, new treatments have become available which have improved survival rates in lung cancer patients. One promising treatment option is the rapidly growing field of oral targeted therapies, which employs drugs that interfere with specific molecules involved in the growth, progression, and spread of cancer. However, these therapies can cause a variety of symptoms and adverse events that can impair quality of life. mHealth technologies may help individuals with lung cancer better track their side effects and manage medications on a day-to-day basis. However, understanding patients’ attitudes toward smart devices such as smartphones, smartwatches, and smart pill bottles, as well as their specific needs when using these devices, is critical before design and deployment studies of medication adherence can be carried out. In this study, we conducted interviews with 9 individuals with stage III-IV lung cancer at a National Cancer Institute-designated comprehensive cancer center in the Mid-Atlantic region of the United States to assess the feasibility of using such devices for managing medication and medication related side-effects. We evaluated patients’ attitudes towards the design and function of smart devices and how these devices fit into their daily life. Our results may help clinicians and researchers to co-develop effective mHealth system deployments for side effect and medication management in oncology populations. Keywords mHealth smartphone medication adherence side effect cancer

Background: Health interventions delivered via smart devices are increasingly being used to address mental health challenges associated with cancer treatment. Engagement with mobile interventions has been associated with treatment success; however, the relationship between mood and engagement among patients with cancer remains poorly understood. A reason for this is the lack of a data-driven process for analyzing mood and app engagement data for patients with cancer.Objective: This study aimed to provide a step-by-step process for using app engagement metrics to predict continuously assessed mood outcomes in patients with breast cancer.Methods: We described the steps involved in data preprocessing, feature extraction, and data modeling and prediction. We applied this process as a case study to data collected from patients with breast cancer who engaged with a mobile mental health app intervention (IntelliCare) over 7 weeks. We compared engagement patterns over time (eg, frequency and days of use) between participants with high and low anxiety and between participants with high and low depression. We then used a linear mixed model to identify significant effects and evaluate the performance of the random forest and XGBoost classifiers in predicting weekly mood from baseline affect and engagement features.Results: We observed differences in engagement patterns between the participants with high and low levels of anxiety and depression. The linear mixed model results varied by the feature set; these results revealed weak effects for several features of engagement, including duration-based metrics and frequency. The accuracy of predicting depressed mood varied according to the feature set and classifier. The feature set containing survey features and overall app engagement features achieved the best performance (accuracy: 84.6%; precision: 82.5%; recall: 64.4%; F1 score: 67.8%) when used with a random forest classifier.Conclusions: The results from the case study support the feasibility and potential of our analytic process for understanding the relationship between app engagement and mood outcomes in patients with breast cancer. The ability to leverage both self-report and engagement features to analyze and predict mood during an intervention could be used to enhance decision-making for researchers and clinicians and assist in developing more personalized interventions for patients with breast cancer.

This study evaluated the effectiveness of different recruitment messages for encouraging enrollment in a digital mental health intervention (DMHI) for anxiety among 1,600 anxious patients in a large healthcare system. Patients were randomly assigned to receive a standard message, or one of five messages designed to encourage enrollment: Three messages offered varying financial incentives, one message offered coaching, and one message provided user testimonials. Patients could then click a link in the message to visit the DMHI website, enroll, and start the first session. We examined the effects of message features and message length (short vs. long) on rates of site clicks, enrollment, and starting the first session. We also tested whether demographic and clinical factors derived from patients’ electronic health records were associated with rates of enrollment and starting the first session to understand the characteristics of patients most likely to use DMHIs in this setting. Across messages, 19.4% of patients clicked a link to visit the DMHI website, but none of the messages were significantly associated with rates of site clicks, enrollment, or starting the first session. Females (vs. males) had a greater probability of enrollment. No other demographic or clinical variables were significantly associated with enrollment or starting the first session. Findings provide guidance for resource allocation decisions in larger scale DMHI implementations in healthcare settings.

Social anxiety disorder is associated with distinct mobility patterns (e.g., increased time spent at home compared to non-anxious individuals), but we know little about if these patterns change fol...

BACKGROUND Health interventions delivered via smart devices are increasingly being used to address mental health challenges associated with cancer treatment. Engagement with mobile interventions has been associated with treatment success; however, the relationship between mood and engagement among patients with cancer remains poorly understood. A reason for this is the lack of a data-driven process for analyzing mood and app engagement data for patients with cancer. OBJECTIVE This study aimed to provide a step-by-step process for using app engagement metrics to predict continuously assessed mood outcomes in patients with breast cancer. METHODS We described the steps involved in data preprocessing, feature extraction, and data modeling and prediction. We applied this process as a case study to data collected from patients with breast cancer who engaged with a mobile mental health app intervention (IntelliCare) over 7 weeks. We compared engagement patterns over time (eg, frequency and days of use) between participants with high and low anxiety and between participants with high and low depression. We then used a linear mixed model to identify significant effects and evaluate the performance of the random forest and XGBoost classifiers in predicting weekly mood from baseline affect and engagement features. RESULTS We observed differences in engagement patterns between the participants with high and low levels of anxiety and depression. The linear mixed model results varied by the feature set; these results revealed weak effects for several features of engagement, including duration-based metrics and frequency. The accuracy of predicting depressed mood varied according to the feature set and classifier. The feature set containing survey features and overall app engagement features achieved the best performance (accuracy: 84.6%; precision: 82.5%; recall: 64.4%; F1 score: 67.8%) when used with a random forest classifier. CONCLUSIONS The results from the case study support the feasibility and potential of our analytic process for understanding the relationship between app engagement and mood outcomes in patients with breast cancer. The ability to leverage both self-report and engagement features to analyze and predict mood during an intervention could be used to enhance decision-making for researchers and clinicians and assist in developing more personalized interventions for patients with breast cancer. CLINICALTRIAL