Patient and clinician opinions on internet search data use in therapy: Insights,considerations,and guidelines for integrating a new digital phenotyping measure

Abstract

Objective

The integration of technology in mental health care has proliferated to address accessibility and symptom monitoring through smartphones and apps. While the analysis of internet search data has proved fruitful in predicting various outcomes like youth suicidality and psychosis relapse, there has been scant research assessing opinions of patients and clinicians on adopting it in outpatient care, and no work on assessing the perceived utility of visualized data. We sought to address this gap by evaluating (1) the comfort, feasibility, and utility of sharing internet search data, and (2) various data visualizations from a novel pipeline to inform how patients and clinicians might start its potential integration immediately.

Methods

Patients (n = 21) and clinicians (n = 9) were interviewed on their opinions about integrating internet search data in therapy. Thematic analysis was used to derive common themes from interviews.

Results

We revealed three common themes: varying comfort with internet search data in therapy, benefits of internet search data in therapy, and risks of internet search data in therapy. Patients and clinicians were generally comfortable integrating internet search data into therapy, but expressed reservations regarding data privacy and security, liability and complete patient autonomy. More granular search data in visualizations was considered more useful when compared to vague search frequencies.

Conclusion

Our results reveal a spectrum of comfort and utility surrounding integrating internet search data in therapy. We introduce a simple data processing pipeline for dynamic visualizations of search data at a variety of comfort levels, laying the immediate groundwork for its personalized implementation in outpatient care.

Introduction

The integration of technology into mental health care has rapidly expanded from video telehealth visits to now mobile health with myriad apps, wearables, and data.^1,2 Yet one data stream remains underutilized in mental health, despite youth reporting its “almost constant” generation ³ : internet search data. Google receives over 1 billion health-related signals per day and searching for health topics like symptoms, diagnoses, and medications has become common across the world. ⁴ These searches have been found to correlate with a variety of health outcomes such as emergency department visits, cancer diagnoses, ⁵ and even risk of impending strokes. ⁶

Mental health-related internet searches and uses are equally common, ⁷ often triggered by people's mental health concerns. ⁸ Already, analyzing internet search data has shown promise in predicting health anxiety severity, suicidal thoughts and behaviors, ⁹ and relapse in schizophrenia. ¹⁰ Given the rise of digital phenotyping to monitor and predict real-world symptoms outside of periodic clinical assessments, there is therefore potential for internet search data to be integrated as an additional digital data source to advance precision mental health care. ¹¹ However, this integration has seen little effort to date, likely due to feasibility and ethical considerations ¹² coupled with a potential lack of comfort by patients and clinicians.

While perspectives on using digital data broadly have been surveyed^12–14 studies have not focused on interviewing patients and clinicians on the routine integration of internet search data in outpatient mental health care. Our study aims to gain insight into (1) logistical, ethical, security, and privacy considerations, (2) bidirectional comfort levels of sharing and handling internet search data in therapeutic settings, and (3) the utility of processed and visualized internet search data as a tangible means toward driving purposeful clinical discussions and gauging actual interest in using the data in care.

Methods

Data collection

Patients completing care for depression and/or anxiety at Beth Israel Deaconess Medical Center (n = 21) and clinicians administering psychotherapy (n = 9) were interviewed. Clinicians consisted of one psychiatry resident and master's (n = 3) and doctoral (n = 5) level trainees. Inclusion criteria involved being above 18 years of age, speaking English and being a patient enrolled in care or a clinician employed at Beth Israel Deaconess Medical Center. Verbal informed consent was obtained from all participants as written informed consent was not required by the institutional review board (IRB). Semistructured interviews lasting for approximately 30 min were conducted online and in-person from June to August 2024 at Beth Israel Deaconess Medical Center in Boston, MA to assess general opinions on integrating internet search data in therapeutic settings (e.g., sharing comfort, risks, feasibility, clinical utility, invasiveness, etc.). Opinions on the utility of four mock visualizations of internet search data were assessed using qualitative questions and a 5-point Likert scale from “not very useful” to “very useful.” Supplemental File S1 contains the full interview questions. The code and test internet search data to create these visualizations is accessible at our Open Science Framework repository (https://osf.io/rwpgt/).

Data visualization creation

Given the focus of this study on the use of data in care, we created a new Google account and deliberately inputted a pseudo-random set of searches over one week, including depression-related queries. We accessed this data via Google Takeout, a free Google platform accessible to every user of Google services today at takeout.google.com. Users can select data to be exported from various Google services they have used, including Google Search, Gmail, and YouTube, and specify a preferred file destination, such as Google Drive, email, or Dropbox. To help guide discussions with clinicians and patients, and to highlight different ways data can be shared or used in care, we prepared a series of four visualizations displaying data at a variety of abstraction levels, ranging from exact search queries to frequency of search engine use (see Figure 1).

Figure 1.

Mock internet search data visualizations.

Visualization 1 (Figure 1A) was intended to provide the least amount of granular information to maintain the highest level of privacy from patient search content, representing daily search frequency over one week. Visualization 2 (Figure 1B) displays the frequency of searches grouped into broad categories that the patient and clinician can personalize and input themselves, aimed to give broad information about search content while still maintaining privacy from individual searches. For this study, we selected default categories that reflect a variety of common search topics like “travel” and “current events” and included “mental health” to show how the graph could discretely represent at least one clinically relevant category. However, patients and clinicians would also have the option to customize these categories to better tailor to the patient's care needs. To categorize individual searches into these categories, we used a zero-shot classification model, ¹⁵ an artificial intelligence approach leveraging natural language processing, to assign each query to a category based on the highest probability match. Visualization 3 (Figure 1C) was aimed to be the most granular figure by showing the time of search and exact search queries within the “mental health” category, or any category deemed relevant in care. This visualization was made available for patients and clinicians who would feel most comfortable sharing their data and less concerned with privacy. Visualization 4 (Figure 1D) was inspired by a word-cloud and meant to provide an easily interpretable and comprehensive aggregation of keywords from all searches, maintaining a level of abstraction from full searches and thus still maintaining some privacy.

Ethics approval

This study protocol was reviewed and approved by the Beth Israel Deaconess Medical Center IRB, protocol #2024P000274.

Data analysis

AD and two research assistants conducted interviews and took detailed notes. Using thematic analysis, AD and SC collaboratively generated codes and identified recurring themes, each of which were thoroughly discussed until consensus was achieved. Interviews were conducted until no new themes emerged and data saturation was achieved. General themes, subthemes, and quotations were created for patients and clinicians and merged to form common themes. Mean patient/clinician ratings and qualitative feedback for each visualization were also aggregated. This study adheres to the Consolidated Criteria for Reporting Qualitative Research where applicable. A checklist is available as supplementary material.

Results

Three themes common across trainee clinicians and patients were identified in interviews and echoed in visualization feedback: (1) varying comfort with internet search data in therapy, (2) benefits of internet search data in therapy, and (3) risks of internet search data in therapy. We reduced these more succinct themes from a comprehensive list of themes, subthemes, and supporting quotes which can be referenced in Supplemental Table S1. Patients and clinicians both found visualizations 2 and 3 the most useful, highlighting the utility of exact search queries and search frequencies grouped into tailored categories. Mean ratings for the utility of each visualization are displayed in Figure 2.

Figure 2.

Average visualization ratings.

Varying comfort with internet search data in therapy

Most patients showed mixed clarity when asked to describe internet search data, often including external components such as screen time, social media, cookies, and search behavior patterns. Despite these perceptions, patients expressed comfort in sharing internet search data with their therapist, while simultaneously stating multiple reservations and hoping that key features were in place. These features, stemming mainly from their perceived risks, consisted of patient autonomy in opting whether and when to share searches, protection of sensitive information, and receiving clear guidelines and boundaries on the storage and handling of their data. One patient stated that its implementation “is a case-by-case thing … some people feel comfortable, some might see it as an invasion of privacy, (so you) need to filter searches.” Despite possible risks, most patients, including a subset who were very uncomfortable initially sharing data, rated at least one data visualization as useful.

Clinicians showed mixed clarity regarding internet search data, considering social media use and analyzed patterns of search behavior as included. They also tended to be more cautious about receiving their patient's data, raising concerns like “…how do I use the data? Do I talk to them about it? Do I see sensitive information?.” Most, however, regained comfort if legal, privacy, and patient autonomy concerns were addressed. Almost all clinicians expressed interest in seeing as much search data as the patient was willing to share, exemplified by one saying, “the patient should choose the level of privacy … it's less about what I want, I could find value in about anything they shared.” Clinicians showed generally positive attitudes about the utility of visualizations, preferring granular data like topics of searches, keywords and exact queries. They also found the visualization that displayed exact mental health searches and the time of search most useful (Visualization 3) and displaying unlabeled search frequencies the least useful (Visualization 1). After seeing data visualized, some clinicians expressed increased comfort in the idea of integrating internet search data into their care.

Benefits of internet search data in therapy

Patients envisioned many clinical benefits, such as being more transparent with their clinician, portraying a more comprehensive view of themselves, reducing recall burden, and helping guide therapeutic discussions. Some expressed that seeing visualizations “helps me understand my moods and feeling during that (search) time” and provides “more objective and holistic information,” further reinforcing its utility for increased self-awareness through categorizing overall search activity. Patients found the ability to personalize the categories searches can be broken up within Visualization 2 very beneficial as it enables clinicians to focus on the most relevant categories for their needs, while also limiting invasiveness on irrelevant and sensitive information. Even among patients who initially expressed discomfort with sharing their data, viewing broad categories of searches in Visualization 2 was seen as useful and did not feel invasive.

Among clinicians, there was a consensus about the perceived utility of internet searches, including more objective, holistic, and granular data than what a patient might share in therapy, hence helping guide discussions in session. One clinician justified that “people feel more at ease asking Google than a real person.” These attitudes were reflected in visualization feedback, where most clinicians found benefits for guiding therapeutic discussions around searched topics, monitoring searches within certain categories (e.g., anxiety, rumination, self-harm), and seeing patients’ exact, real-world thoughts. The variety of visualizations enabled clinicians to envision different levels of granularity as more beneficial depending on the severity of a patient's symptoms or situation, without having to continuously invade the patient's privacy. Given the spectrum of granularity and invasiveness, clinicians found practical cases for each of the visualizations and expressed comfort in seeing more granular data despite its potential invasiveness, expressing they were “more useful than envisioned.” Visualizations of internet search data could help clinicians develop a quick overview of a patient's retrospective symptoms “used almost like diary cards” before a therapy session.

Risks of internet search data in therapy

Patients perceived multiple risks of sharing their data, including data breaches, identity theft, data misuse and clinician liability for reporting alarming searches. To protect themselves from data leak risks, some patients preferred more stringent data security mechanisms like restricted access to just their therapist, HIPAA protection, and weekly consent for sharing the previous week's data. Clinically, patients generally feared sharing internet search data might fracture their therapeutic relationship and be too invasive. They also hoped that certain searches that contain sensitive information would not be shared, and that data handling procedures and risk mitigation would be clearly outlined in the informed consent process. One patient said, “there needs to be boundaries, sometimes it's personal searches or bank account information” and that they “should be disclosed what data is being monitored.” Seeing visualizations, patients shared similar concerns regarding the absence of context for search terms, fearing clinicians may misinterpret intent and affect the therapeutic alliance. A subset of patients further remained dubious of drastic clinical utility, with some citing that they do not use search engines a lot, and that clinicians would not find anything more useful than what is shared during a therapy session.

Clinicians were concerned about having a clear clinical rationale, the patient being comfortable and in control of sharing, data being secure and private, and liability for alarming or illegal searches. Clinicians also shared worries about the potential for breaching trust and damaging the therapeutic alliance. Within visualizations, clinicians were especially concerned with the lack of context and detailed search information for visualizations that only showed frequencies, keywords and general categories, with one stating it “would be better with context, but unhelpful by itself.” Further risks included feeling overwhelmed from seeing many searches listed, searches irrelevant to therapy, and unactionable keywords. To mitigate these concerns, clinicians suggested to include exact searches and times on most visualizations, or at the very least general themes and keywords, fluctuations of search terms and categories over time, and correlations between search categories, mood and other digital phenotyping features if available. Despite these improvements, some clinicians remained skeptical of its practical utility for the average patient aside from monitoring addiction-related behaviors due to an increase of redundant and irrelevant information, at the cost of a privacy invasion.

Discussion

Our study sought to directly assess the comfort, feasibility, and utility of integrating internet search data in therapy by directly interviewing patients and trainee clinicians. We derived three main themes from interviews, backed by feedback from a variety of novel data visualizations: varying comfort with internet search data in therapy, benefits of internet search data in therapy, and risks of internet search data in therapy.

Patients and clinicians formed a spectrum of comfort and utility surrounding integrating internet search data in therapy, predicated on various perceived benefits and concerns. Most patients were comfortable sharing data with their therapist, but wanted data privacy and security, clear communication of these policies and other data handling procedures, and most importantly, complete autonomy in deciding when and what to share. While some questioned its invasiveness and added value, many envisioned an objective complement to daily thoughts and behaviors, highlighting the potential for internet search data to beneficially contribute to self-awareness and therapeutic discussions. These findings are in line with broader positive patient attitudes toward data sharing in digital health and indications for addressing privacy and security concerns with novel digital technologies.^16,17

Clinicians further reinforced a spectrum from uncomfortable seeing any specific search data to comfortable seeing the most granular data the patient is willing to share. Most clinicians concurred on utility for cases of internet addiction, rumination, anxiety searches, and suicidal ideation, but envisioning added utility outside of these use cases was more challenging. All clinicians agreed that sharing any data should be completely in the control of the patient and that privacy, security, and data handling should be taken seriously and communicated clearly with patients. For care implementation, we, therefore, recommend clinicians and clinics to (1) use HIPAA-compliant platforms for storing and accessing patient search data ¹⁸ and (2) develop deliberate consent workflows outlining both benefits/risks of using search data.

Patients and clinicians also varied in comfort around the granularity of data visualizations, but most found broad clinical utility, consistent with attitudes about integrating data visualizations in care. ¹⁹ Despite potential invasiveness, both preferred more granular search data in the form of specific categories and queries, while also seeking more temporal and contextual data. Some, however, remained doubtful of the added value at the expense of patient privacy. Both patients and clinicians also expressed worry about the lack of context from searches and potential misinterpretations, particularly within visualizations with high levels of abstraction from exact searches. Hence, we emphasize the inclusion of (1) a short description of each visualization and examples of potential use cases in therapy sessions, as exemplified in the section below, and (2) patient contextualization and thorough discussion around searches, as opposed to its use as stand-alone data.

Our results show the perceived feasibility of internet search data for improving a wide range of care. We further reveal positive attitudes toward concrete visualizations and emphasize a comfort spectrum for the potential integration of internet search data, particularly regarding the variety of presentation options that may enable a personalized match to the patient and clinician's comfort level. Taken together, these findings highlight the potential of visualized internet search data to function as a complementary dynamic and transdiagnostic tool for advancing tailored mental health care.

Example use case of internet search data in therapy

A patient suffering from major depressive disorder and substance use disorder is moderately comfortable sharing general themes about their internet search data, but does not feel comfortable sharing every individual search with their clinician. After the clinician presents the patient with all data visualization options, the patient decides that Visualization 2 and 4, shown in Figure 3, would be most appropriate for balancing privacy protection with their sharing comfort. Given the customizability of search categories in Visualization 2, the patient is pleased with most of the default categories. Through further discussion, the clinician asks if the patient would be comfortable including a more precise category related to their substance use to better monitor their thoughts and behaviors about it throughout the week. The patient agrees under the condition that detailed information about those searches will not be revealed outside of the chosen visualizations.

Figure 3.

Example use case visualizations.

In the next session, the clinician asks the patient about their substance use this past week and if they were able to manage their cravings successfully. After an initial discussion, the clinician suggests looking at the past week's internet search data to explore if any search behavior was related. The clinician explains the first bar graph to show that the “current events” category was searched the most frequently, followed by substance use-related searches. This prompts a discussion about how the patient often uses substances to regulate their affect after political events, which they have followed extensively for the past few weeks. The clinician then presents Visualization 3, modeled after a word cloud, showing how the words “depression,” “news,” and “symptoms” were among the most searched key words. The clinician prompts the patient to reflect on whether/how these insights complement the previous graph's information or accurately represent their previous week. The patient confirms the associations of viewing news with increased depressive symptoms; however they also mention that presence of the “symptoms” key word is likely due to searching information about their spouse who recently got the flu—emphasizing the need for contextualization when interpreting this data.

Limitations

While previous work has asked patients and clinicians about opinions on sharing broad digital data in therapy ¹² and from specific domains like social media or health apps in therapy,^13,14 our study was the first to directly assess attitudes about internet search data in depth while providing tangible data visualizations to patients and clinicians. Results were limited to patients and clinicians within one hospital who volunteered to be interviewed, not capturing attitudes across larger samples and from diverse geographic and socioeconomic backgrounds. Another limitation is that our study did not capture participant demographic data. Therefore, future work is needed to generalize our findings across larger, diverse samples and assess the time/support staff necessary to create these visualizations. Patients in our sample were mainly suffering from depression and anxiety, necessitating further work to determine comfort levels and feasibility in treatment for serious mental illness. Our study evaluated perceptions about sharing data and assessed the perceived utility of data visualizations without applying them in a treatment setting with a clinician, further leaving the possibility of discrepancies between theoretical and practical attitudes. Future work should assess feasibility and utility in real-world care, revealing whether theoretical and practical attitudes match for both patients and clinicians, emergent challenges that may have been overlooked, and improvements to be made.

Conclusion

Our results reveal the psychological feasibility of integrating internet search data into mental health treatment directly from patients and clinicians. Given the spectrum of comfort levels and perceived utility from both parties, future implementation should keep patient autonomy as a precondition and constant throughout treatment. Stringent data privacy and storage, alongside clear data handling procedures outlined in a mandatory consent should also be ensured. The visualizations we created elicited generally positive attitudes for interpretability and utility, satisfying a spectrum of privacy and comfort levels for both patients and clinicians and showing feasibility for use in care. Our novel pipeline for preprocessing and visualizing internet search data is available for use by researchers and clinicians around the globe seeking to interpret internet search data while accommodating varying levels of patient comfort. Future work will need to embed this solution in a platform that addresses security and patient autonomy requirements. Ultimately, its implementation could expand upon analyzing retrospective internet search data, enabling prospective symptom monitoring from real-time searches to launch adaptive interventions and prevent high-risk patients from self-harm or relapse.

Supplemental Material

sj-docx-1-dhj-10.1177_20552076251327044 - Supplemental material for Patient and clinician opinions on internet search data use in therapy: Insights, considerations, and guidelines for integrating a new digital phenotyping measure

Supplemental material, sj-docx-1-dhj-10.1177_20552076251327044 for Patient and clinician opinions on internet search data use in therapy: Insights, considerations, and guidelines for integrating a new digital phenotyping measure by Alex Dhima, Soumya Choudhary, Keris Myrick and John Torous in DIGITAL HEALTH

Supplemental Material

sj-docx-2-dhj-10.1177_20552076251327044 - Supplemental material for Patient and clinician opinions on internet search data use in therapy: Insights, considerations, and guidelines for integrating a new digital phenotyping measure

Supplemental material, sj-docx-2-dhj-10.1177_20552076251327044 for Patient and clinician opinions on internet search data use in therapy: Insights, considerations, and guidelines for integrating a new digital phenotyping measure by Alex Dhima, Soumya Choudhary, Keris Myrick and John Torous in DIGITAL HEALTH

Supplemental Material

sj-pdf-3-dhj-10.1177_20552076251327044 - Supplemental material for Patient and clinician opinions on internet search data use in therapy: Insights, considerations, and guidelines for integrating a new digital phenotyping measure

Supplemental material, sj-pdf-3-dhj-10.1177_20552076251327044 for Patient and clinician opinions on internet search data use in therapy: Insights, considerations, and guidelines for integrating a new digital phenotyping measure by Alex Dhima, Soumya Choudhary, Keris Myrick and John Torous in DIGITAL HEALTH

Supplemental Material

sj-pdf-4-dhj-10.1177_20552076251327044 - Supplemental material for Patient and clinician opinions on internet search data use in therapy: Insights, considerations, and guidelines for integrating a new digital phenotyping measure

Supplemental material, sj-pdf-4-dhj-10.1177_20552076251327044 for Patient and clinician opinions on internet search data use in therapy: Insights, considerations, and guidelines for integrating a new digital phenotyping measure by Alex Dhima, Soumya Choudhary, Keris Myrick and John Torous in DIGITAL HEALTH