A cancer diagnosis is often associated with psychological challenges affecting patients and the broader support network, including spouses, family members, and friends who assume the role of caregivers [1-4]. Caregiving entails practical and supportive tasks across the continuum of care: coordinating medical visits, building relationships with clinicians, communicating with the medical team, and contributing to decision-making, adherence to treatment plans, and symptom management [2,3]. These tasks carry physical, psychosocial, emotional, and financial consequences, rendering cancer caregivers highly vulnerable to stress and mental health problems [3,5,6]. Across studies, family caregivers of patients with cancer often experience greater psychological distress than patients [2,4,7-11], and cancer caregiving is rated as more burdensome than other high-intensity caregiving scenarios [12]. Yet, caregivers’ mental health needs remain insufficiently addressed in routine cancer care, underscoring the need for targeted psychosocial interventions [13,14].

Caregiver psychological distress is common throughout the illness trajectory, with patterns that differ across diagnosis, active treatment, recurrence, and advanced disease. Between 55% and 95% of caregivers [15-17] report clinically significant distress or mental health conditions. Compared with the general US population, family caregivers of patients with advanced cancer frequently score below 30% on mental health measures [18]. Inadequate preparation for the caregiving role compounds distress, contributing to feelings of helplessness and regret as caregivers witness symptom burden [19,20]. Specific conditions are highly prevalent. Depression ranges from 16% to 67% and tends to increase as prognosis worsens [9,21-24]. Anxiety is even more common; a global meta-analysis estimates 46.5% of family caregivers experience clinically significant anxiety, with rates in advanced disease (~40%-42%) exceeding those observed in patients [8,22,25,26]. In some cancer types (eg, head and neck) up to 57% of caregivers report clinically significant anxiety and/or depression [27].

Multiple factors contribute to this heightened vulnerability. Caregiver risk factors include being female, younger age, financial strain, limited social support, high burden, low caregiving self-efficacy, and prolonged caregiving hours [22,24]. Family caregivers are more likely to experience psychological distress when the patient is younger, when the illness progresses and worsens, when the patient has limited awareness of their prognosis, or when the patient’s overall health and daily functioning decline [8,22,24]. Relational and contextual risk factors include being a spousal caregiver, insecure attachment (eg, fear of abandonment or avoidance), and family conflicts, which amplify distress and cause psychological strain among family cancer caregivers [21,22,24].

Despite this evidence, caregiver support remains underresourced and inadequate. Family caregivers frequently report unmet informational and psychological needs [4,11,26,28-31]. Traditional in-person mental health services are often inaccessible due to cost, time, and logistical barriers, limiting opportunities to address caregivers’ needs [1]. As a result, family caregivers experience sleep disruption, diminished quality of life, and reduced caregiving capacity, ultimately affecting patient care [1,12,32].

Digital mental health offers a scalable path to support cancer caregivers. Smartphone- and web-based interventions can extend reach, reduce access barriers, and deliver timely, evidence-based care. However, rigorous evaluation in oncology caregiving remains limited. To date, eHealth interventions for family caregivers of individuals living with cancer have been found to reduce depression and modestly improve quality of life versus usual care, supporting the feasibility and potential value of remote delivery for caregiver mental health. Still, effects were small and trials were few, with short follow-up and heterogeneous content, therefore underscoring the need for larger, higher-quality randomized controlled trials [1,6,33-37]. Building on this foundation, AI-enabled depression and anxiety (D&A) screening and monitoring apps have strong potential for flexible, real-time support [34,36,37]. Evidence shows that technology-based approaches can (1) improve health and psychological outcomes, (2) expand access to care, (3) be cost-effective, and (4) be delivered in self-paced, tailored formats [35,38-41]. Yet, family members remain excluded from national distress screening mandates, leaving their needs inconsistently identified and addressed in oncology care settings [42-45]. This evidence emphasizes an urgent need to design, test, and implement family caregiver-specific digital psychosocial interventions that can be integrated into routine cancer care.

Technological advancements in automated and AI-enabled speech analysis have significantly improved the ability to detect psychological symptoms [46-49]. Tools using voice data, particularly acoustic features, now offer a promising complement to traditional validated screening instruments. These tools have shown diagnostic performance comparable to widely used psychometric scales [50,51] with added advantages including reduced user burden, minimal risk of human error (eg, missing data), and the potential for greater user comfort during verbal self-expression [52-54].

Nevertheless, several limitations persist. Most tools continue to rely on either acoustic (ie, how something is said) or semantic content (ie, what is said), without combining both for improved precision [49,55-57]. Additionally, many systems have focused solely on depression, with a smaller fraction targeting anxiety and even fewer addressing both concurrently [55-57]. This gap is particularly concerning for family caregivers of patients with cancer, a population in which anxiety symptoms are both prevalent and frequently co-occur with depressive symptomatology.

This study addresses these gaps by evaluating Ellipsis Caregiver Assessment Enhancement (eCARE; Ellipsis Health, Inc), a next-generation speech-based assessment platform designed to detect both D&A symptom severity using an integrated AI model. Unlike most existing systems, eCARE combines semantic and acoustic data inputs to provide dual assessments, enhancing precision. Additionally, it offers a clinician-facing dashboard to support referrals and care coordination, promoting a person-centered model of supportive care delivery. The Ellipsis Health platform has been technically validated in multiple peer-reviewed publications demonstrating its strong machine learning performance [58-62].

eCARE is a speech-based smartphone app linked to Ellipsis Health, which is a secure, cloud-based AI infrastructure. The app processes a weekly 90-second audio recording from users who respond to prompted topics, including those related to their caregiving stress and emotional well-being. The system analyzes both the acoustic and semantic characteristics of speech using advanced deep learning techniques, including transfer learning. This approach allows for more sophisticated and accurate identification of users’ speech signals than traditional models that rely on fixed acoustic features such as pitch or volume [57,60,62].

The platform has demonstrated strong performance metrics, with reported area under the curve values of 0.85 for Patient Health Questionnaire-9 item (PHQ-9) and 0.84 for Generalized Anxiety Disorder-7 item (GAD-7) predictions (binary classification with a cutoff score of 10), root-mean-square error values of 4.25 and 4.47, and mean absolute error values of 3.13 and 3.23, respectively [63]. To date, eCARE remains the only available speech-based screening platform capable of producing concurrent D&A scores for family caregivers of patients with cancer. This dual capability is a critical feature for reducing participant fatigue during long-term monitoring and managing the acute mental health needs of this group.

The eCARE app is available for download on the Apple or the Google Play Store and can be used on iOS and Android mobile devices. After downloading the app on their smartphones, users are able to register in the app with the phone number they provided to the research team. Once a week, users will be prompted to log in to the app and record a 90-second response to a selected topic related to their mental health or caregiving responsibilities. Afterward, users will receive nonclinical feedback on their distress levels. The app also includes longitudinal tracking of responses and mental health crisis hotline information. On the provider side, eCARE features a HIPAA (Health Insurance Portability and Accountability Act)-compliant web portal that visualizes severity of distress trajectories over time, supporting timely mental health referrals and opportunities for direct integration into the cancer care continuum.

The present study will (1) expand understanding of the psychosocial issues faced by family caregivers of individuals diagnosed with cancer and (2) generate actionable evidence to refine eCARE for responsive delivery in a way that is receptive to preferences for care, format, and optimal timing of screening. Specific Aim 1: establish the feasibility and acceptability of eCARE among family caregivers of patients with cancer in a single-arm prospective cohort study, using platform completion rate (feasibility) and the acceptability of the intervention measure (AIM; acceptability) as indicators. Specific Aim 2: qualitatively evaluate facilitators and barriers to eCARE uptake among a subset of participants (n=20) to inform iterative refinement. Expected outcomes include evidence that a caregiver-driven digital approach can increase the proportion of family caregivers who are screened and monitored, as well as generating concrete recommendations to optimize eCARE for future efficacy testing and implementation in survivorship care.

This single-arm, prospective cohort study will use a mixed methods design to evaluate the feasibility and acceptability of eCARE, a digital AI-enabled tool designed to monitor psychological distress among family caregivers of individuals living with cancer. The study will:

Participants will use eCARE independently, following structured onboarding and usage instructions provided by the research team at baseline. Psychological distress will be monitored using standardized questionnaires and voice-based data collection, with app-based metrics and quantitative survey data supplemented by qualitative feedback (Table 1).

Participants will be eligible to enroll if they (1) self-identify as the primary caregiver or support person of an individual diagnosed with cancer within the past 5 years, (2) are 18 years of age or older, (3) have access to a smartphone capable of downloading and using the eCARE app, (4) are fluent in English, and (5) are able and willing to provide informed consent. To ensure adequate representation of caregiving experiences, a stratified sampling approach will be used to recruit equal numbers of caregivers who identify as non-Hispanic White (n=15), Black/African American (n=15), Hispanic and Latino (n=15), and Asian American, Native Hawaiian, and Pacific Islander (AANHPI; n=15). Participants will be excluded from the study if they (1) have a cognitive impairment; (2) have a speech impairment; and (3) have a severe mental illness that would impede the ability to provide informed consent or complete study activities. Eligibility will be self-reported through an initial study information survey. Participants will review a list of exclusionary criteria and be asked to self-select out of the study if any listed conditions apply. Eligibility will be verified as necessary via follow-up communication (eg, telephone screening). To minimize fraudulent responses, we will apply minimum time-on-task thresholds, checks for duplicate IPs or devices, and manual pattern review.

As part of this study, we are collecting a comprehensive set of variables to capture the multifaceted experiences of caregiving.

Quantitative and qualitative data will be analyzed using a mixed methods approach to evaluate the feasibility and acceptability of the eCARE tool among family caregivers of individuals diagnosed with cancer. Descriptive statistics will be used to summarize participants’ sociodemographic, clinical, and caregiving characteristics, including means, SDs, ranges, and 95% CIs. Graphical methods (eg, histograms, boxplots, and scatterplots) will be used to examine variable distributions and identify the need for transformations. For comparisons, we will use ANOVA for continuous variables and chi-square tests for categorical variables. Differences in psychological distress, caregiving burden, relational closeness, perceived responsiveness, and motivation to provide care will also be explored. Feasibility, the primary outcome of Aim 1, will be assessed based on the number of participants completing at least 4 of 6 total weekly eCARE assessments during the 8-week study period. Participants exceeding 6 sessions due to more frequent use will be classified as having met the feasibility threshold, consistent with those completing at least 4 of 6 sessions. The study will be considered feasible if 60% or more of enrolled participants (≥36 of 60) achieve this end point. Acceptability, the secondary outcome, will be assessed using the AIM [76]. The mean and SD of AIM scores will be calculated, with an average score of ≥4.0 (on a 5-point scale) interpreted as evidence of acceptability. Exploratory latent growth curve modeling will be used to examine changes in depression and anxiety symptoms over time (measured by PHQ-9 and GAD-7). These analyses are considered exploratory due to the limited statistical power associated with the sample size (n=60).

For Aim 2, qualitative data from semi-structured interviews with a purposive subsample of 20 family caregivers (including both completers and noncompleters) will be analyzed using thematic content analysis in ATLAS.ti [96]. Interview transcripts will be transcribed verbatim and independently coded by trained research staff. An initial coding scheme will be developed inductively and applied iteratively. Two independent coders will review all transcripts, with discrepancies resolved through discussion until consensus is achieved [97]. Axial coding and thematic mapping will be used to identify higher-order categories related to eCARE’s perceived strengths and limitations, usability, relevance, and reported facilitators or barriers to engagement. Analysis will continue until thematic saturation is reached, defined as the point at which no new codes or themes emerge [79,97,98].

This study was reviewed and approved by the IRB at the University of Houston (Protocol# STUDY00003186) and adheres to the ethical standards outlined in the Declaration of Helsinki. All study procedures will be conducted in accordance with the approved protocol and applicable institutional requirements for human subjects’ research.

Participants will provide informed consent prior to initiating any study procedures. Consent will be documented in writing at enrollment either via Qualtrics e-signature functionality or by signing a printed consent form when enrollment occurs in person. For the optional qualitative interview component, participants will complete an additional oral consent procedure prior to the start of the interview, and the study team will request a waiver of written documentation for that interview-specific consent, as applicable.

Participant privacy and confidentiality will be protected through multiple safeguards. Direct identifiers (eg, name, email, phone) will be collected solely for recruitment, scheduling, and compensation purposes and will be stored separately from research data, linked only via a study ID. App-based and survey data will be maintained on encrypted, password-protected platforms with access restricted to authorized study personnel; downloaded data will be stored on password-protected, encrypted university devices. While participants’ interest in the qualitative component will be captured as part of the informed consent process, oral informed consent will be obtained again prior to the commencement of the qualitative interviews. Qualitative interviews will be audio and/or video-recorded only with participant permission; recordings will be transcribed and then deleted following transcription. Interview transcripts will be labeled with study IDs. Only approved members of the research team will have access to identifiable information, and the master linkage file will be maintained by the principal investigator and destroyed at study completion per protocol.

Participants will receive incentives in the form of electronic Amazon gift cards. To minimize the risk of undue inducement, gift cards will be distributed at three predetermined time points, including US $30 upon completion of a baseline consent and pretest assessment survey, US $30 at the midpoint assessment (T4) contingent on at least one completed check-in with eCARE, and US $30 upon completion of the posttest assessment (T7). Participants who complete the qualitative interview will receive an additional US $50 e-gift card.

Family caregivers play a central role in cancer care, assuming extensive practical, emotional, and coordination responsibilities that contribute to substantial psychological, physical, and financial strain [1,2,11]. Across cancer types and illness stages, caregivers frequently experience clinically significant anxiety and depression, often at rates exceeding those observed in patients themselves, yet their mental health needs remain insufficiently addressed [4,5,7-9,22]. This single-arm, prospective pilot study is among the first to evaluate the feasibility and acceptability of using artificial intelligence and speech-based analytics through a mobile app (eCARE) to monitor the severity of depression and anxiety among a sample of community-based family caregivers of individuals living with cancer. Traditional methods, such as clinic-based assessments or paper forms, often present challenges with high participant burden and low engagement, which can compromise data accuracy and limit opportunities for timely intervention. It is essential to explore innovative strategies offering more flexible, scalable, and caregiver-centered approaches to distress monitoring. To this end, we hypothesize that (1) eCARE will be feasible and acceptable, and (2) that participants will provide in-depth information related to barriers and facilitators influencing its uptake, in addition to feedback for future refinement and implementation.

Prior research has demonstrated that eHealth interventions for cancer caregivers can reduce depressive symptoms and modestly improve quality of life; however, existing studies are few, heterogeneous, and often limited by short follow-up periods and narrow outcome focus [6,12,34,35,39,99]. Moreover, most digital interventions rely on self-report questionnaires, which can introduce response burden and missing data. By requiring a brief speech sample to generate predictions of depression and anxiety, eCARE represents a minimally burdensome pathway for mental health monitoring. The mobile application integrates natural language processing to analyze spoken words and acoustic modeling to assess vocal features, leveraging an accessible, multimodal approach well-suited for smartphones [37,39,59,99,100]. As Schoenberg [101] highlighted, digital tools that are frictionless and embedded in daily life are more likely to be adopted and sustained in real-world environments.

Findings from this pilot study will provide preliminary data on key indicators of feasibility (eg, completion rates) and acceptability (eg, satisfaction ratings). This information will be instrumental in helping to determine whether a speech-based tool can be successfully integrated into the complex caregiving context, where time constraints, emotional burden, and competing demands often limit participation in mental health initiatives. To further this goal, the study also incorporates qualitative feedback, allowing for a user-informed refinement of the application’s content, usability, and perceived value. While eCARE does not currently involve provider feedback loops, future iterations may benefit from exploring how clinicians can access and respond to data trends emerging from the app, potentially enhancing care coordination and timely referral to supportive services.

Should eCARE demonstrate feasibility and acceptability, these findings will directly inform the design of a larger, multisite randomized controlled trial. Such a trial would be designed to robustly compare eCARE with established screening methods, like the National Comprehensive Cancer Network (NCCN) Distress Thermometer or standard electronic symptom checklists, and to evaluate its efficacy on family caregivers’ mental health.

Finally, since symptoms of depression and anxiety are among the most prominent predictors of quality of life and long-term health outcomes, effective and early identification is critical. If validated, speech-based assessments like eCARE could serve as a foundation for future interventions aimed at improving caregiver resilience, reducing emotional burden, and ultimately enhancing the caregiving experience.

This study has several limitations that should be considered. First, as a single-arm feasibility and acceptability study, it is not designed to evaluate the efficacy of eCARE in reducing depression or anxiety symptoms, nor to support causal inferences. Second, generalizability may be limited, as participation requires access to a smartphone and a minimum level of comfort with mobile technology, potentially excluding family caregivers with lower digital literacy or limited access to digital resources. Third, self-selection bias may be present, as caregivers who choose to enroll may be experiencing higher distress, may be more motivated, or may be more receptive to digital mental health tools, which could inflate estimates of feasibility and acceptability. Finally, the relatively short monitoring period affects the ability to elaborate on sustained engagement and longer-term acceptability, in addition to barriers and facilitators influencing eCARE uptake over time.

This study will generate foundational evidence on the feasibility, acceptability, and perceived utility of a mobile, speech-based, AI-enabled tool for monitoring depression and anxiety among family caregivers of individuals living with cancer. Findings will inform the design of a fully powered clinical trial and guide the optimization of remote mental health screening approaches tailored to the caregiving experience. As a scalable, low-cost, and low-burden intervention, eCARE has the potential to support timely identification of psychological distress, enhance access to supportive care services, and improve coordination of psychosocial care across oncology settings. Beyond cancer caregiving, the underlying model holds promise for adaptation to other chronic or life-limiting conditions and to non–English-speaking populations, pending appropriate validation. Future studies should compare eCARE with established screening tools and examine long-term clinical outcomes, user engagement, and effects on both caregiver and patient well-being, while also evaluating strategies for integrating AI-derived screening results into clinical workflows without increasing provider burden. Dissemination of findings will include sharing results with academic, clinical, and community stakeholders to support continued refinement and implementation of digital mental health solutions for caregivers.