Patient characteristics included sociodemographic characteristics, optimism, and social support. Clinical characteristics measured at baseline included comorbidities, depressive symptoms, trait anxiety, symptom severity, controllability, consequences, and distress. Sociodemographic factors (age, education, marital status, and employment) were assessed with the Center for Research in Chronic Disorders Sociodemographic Survey [34]. Optimism was measured by the reliable and validated Revised Life Orientation Test [35], an 8-item scale. Items were rated from 0 (agree a lot) to 4 (disagree a lot). The scores were aggregated to a total score. Social support was measured by the validated Interpersonal Support Evaluation List 12-item short form [36]. Items were rated from 1 (definitely false) to 4 (definitely true) and summed up for a total score. Comorbidities were assessed using the validated Charlson Comorbidity Index, a 10-item self-report comorbidity index with each comorbidity weighted based on 1-year mortality risk [37]. Trait anxiety was assessed with the validated 20-item trait anxiety subscale of the Spielberger State-Trait Anxiety Inventory [38], which evaluates how a participant feels on a 4-point scale (1=“Almost Never” to 4=“Almost always”) and is summed for an overall score. Depressive symptoms were assessed with the Brief Center for Epidemiologic Studies Depression scale [39]. Ten items were rated on a 4-point scale (0=“rarely or none of the time” to 3=“All of the time”) and summed for a total score. Symptom severity, consequences, distress, and controllability were assessed with the Symptom Representation Questionnaire, which is a validated scale of symptom representations in women with ovarian cancer [40]. First, participants completed a 28-item symptom inventory reporting symptom severity (at its worst) in the past week from 0 (did not experience the symptom) to 10 (as bad as I can imagine). Second, participants identified 3 symptoms they wanted to improve or control. Another 3 subscales assessed consequences (5 items), distress (3 items), and controllability (5 items) for each target symptom on a scale of 0 (strongly disagree) to 4 (strongly agree). The mean scores for symptom representation items across 3 “target” symptoms at baseline were used for the statistical analysis in aim 1. Additionally, for aim 2, we also assessed the mean score for symptom controllability, our primary intervention outcome variable, at the 8-week mark. This measurement involved averaging the total scores of 5 items related to each of the 3 “target” symptoms. These items included statements like “I have a significant degree of control over this symptom” and “My actions can influence whether this symptom improves or worsens.”

Table 1 presents the engagement analysis framework, which includes 6 measures across socioaffective, cognitive, and behavioral dimensions, as developed in our prior work [30]. Consistent with this framework, engagement was conceptualized as a multidimensional construct rather than as equivalent to intervention dose or completion. Behavioral indicators, such as question completion and care plan completion, were included as one established dimension of engagement; however, the engagement classification also incorporated socioaffective and cognitive indicators derived from message-board communication. Thus, engagement in this study reflected not only the amount of intervention activity completed but also the nature and quality of participants’ interaction with the intervention content and the nurse interventionist.

Participants were classified into high, moderate, and low engagement groups using k-means clustering based on the 6 engagement measures shown in Table 1. The clustering analysis identified 3 distinct engagement groups: high engagers (n=13), moderate engagers (n=17), and low engagers (n=38). Because the engagement measures were nonnormally distributed, medians and IQRs were used to describe engagement levels within each group. Additional details on the development of the engagement framework, measures, and clustering approach are provided in our prior work [30].

Preliminary bivariate analyses were conducted to identify candidate baseline patient and clinical variables associated with engagement outcomes. A screening threshold of P≤.20 was used to identify potential variables for multivariable model building. These bivariate analyses were used only for variable screening and were not interpreted as primary inferential findings.

Candidate variables were entered into multivariable models, and final models were selected using the minimum Akaike Information Criterion. Ordinal regression was used to examine baseline patient and clinical characteristics associated with overall engagement level, categorized as high, moderate, or low. Multiple linear regression was then used to examine baseline patient and clinical characteristics associated with engagement across the socioaffective, cognitive, and behavioral dimensions, as represented by the 6 engagement measures. Statistical significance was initially evaluated at P<.05.

To address multiple testing, the Benjamini-Hochberg false discovery rate (FDR) correction was applied within each aim to the P values from the final models. For aim 1, the correction was applied to the P values for predictors retained in the final ordinal regression model for overall engagement level and the final linear regression models for the 6 engagement measures. For aim 2, FDR correction was applied to the P values for engagement predictors examined in relation to perceived symptom controllability at 8 weeks. For both aims, unadjusted and FDR-adjusted P values were examined, and findings were interpreted in light of the adjusted results.

We conducted a content analysis of 68 participants’ posts on the message boards to summarize the challenges to engagement mentioned by participants without nurse prompts during the intervention. Additionally, for the 39 participants who discontinued participation on the message boards during the intervention, 2 investigators (YW and R DiLello) examined the specific intervention elements participants were working on before they stopped responding. Interrater reliability was assessed based on the posts from these 39 participants, with a Cohen κ of 0.98. The intervention elements and goals are available in Multimedia Appendix 1.

After listwise deletion, 66 participants were included in the analysis for aim 2. Two participants were excluded because their symptom controllability score at the end of the program was missing; no other analysis variables had missing data. Multiple linear regression was used to examine the relationship between perceived symptom controllability at 8 weeks and engagement across the 3 dimensions, represented by the 6 engagement measures, after controlling for baseline symptom controllability and other retained patient and clinical covariates. Patient and clinical covariates were determined based on preliminary bivariate analyses with perceived symptom controllability at 8 weeks, using a screening threshold of P≤.20. Statistical significance was initially evaluated at P<.05.

For aim 2, the Benjamini-Hochberg FDR correction was applied to P values for engagement predictors examined in relation to perceived symptom controllability at 8 weeks. For both aims, unadjusted and FDR-adjusted P values were examined, and findings were interpreted in light of the adjusted results.

Tables 3 and 4 present the ordinal and multiple linear regression models examining patient and clinical characteristics associated with engagement. After FDR correction, education was no longer significantly associated with overall engagement level (odds ratio [OR] 1.22, unadjusted P=.04; FDR-adjusted P=.08).

In the socioaffective dimension, higher education was associated with a greater word count, and this association remained at the FDR-adjusted significance threshold (B=1467.8, unadjusted P=.02; FDR-adjusted P=.05). Additionally, higher education (B=2.44, unadjusted P=.07; FDR-adjusted P=.09) and the absence of comorbidities (B=13.32, unadjusted P=.06; FDR-adjusted P=.09) showed exploratory trends toward more frequent socioaffective engagement activities.

In the cognitive dimension, education was positively associated with cognitive engagement and remained at the FDR-adjusted significance threshold (B=1.89, unadjusted P=.01; FDR-adjusted P=.05). The absence of comorbidity showed an exploratory trend but was not statistically significant after correction (B=7.13, unadjusted P=.07; FDR-adjusted P=.09). No patient or clinical factors were associated with the average question completion percentage.

In the behavioral dimension, education remained associated with completing more symptom care plans (OR 1.23, unadjusted P=.02; FDR-adjusted P=.05) and care plan reviews (OR 1.29, unadjusted P=.01; FDR-adjusted P=.05).

Across all engagement levels, the most common challenges to engaging in the intervention, reported by 54.4% (37/68) of participants, were related to deteriorating medical conditions and treatments. These included new or worsening symptoms, infections, surgeries, hospitalizations, cancer treatments, medical appointments, and clinical trials. The second significant challenge, cited by 33.8% (23/68) of participants, was being occupied with family activities such as vacations, visits, celebrations, reunions, funerals, and sports. The third challenge, noted by 29.4% (20/68) of participants, involved difficulties using the website or message boards, such as locating documents or reading nurses’ posts. Figure 2 depicts the most frequently endorsed challenges for engagers as a whole and within each group. Other less prevalent challenges reported by engagers encompassed limited computer literacy and internet and hardware issues.

Considering engagement challenges by group, high engagers faced the same top 3 challenges, albeit in a different order. Besides the same common challenges, busy work schedules (6/17, 35.3%) and other social activities (6/17, 35.3%) were frequently cited for moderate engagers. Compared to high engagers, a significant portion of moderate engagers (5/17, 29.4%) felt confused about the intervention process and timeline and mentioned challenges related to family responsibilities.

Low engagers who mentioned challenges, apart from declining health (18/38, 47.4%) and family activities (9/38, 23.7%), also reported confusion about the intervention (8/38, 21.1%), similar to some moderate engagers. Table 5 presents the intervention activities completed by all 38 low engagers. Among these 38 low engagers, 24 (63.2%) stopped participating on the message boards before the intervention ended. In contrast, none of the high engagers stopped responding, and only 1 moderate engager stopped responding when asked by the nurse to select symptom management strategies.

This study examined patient and clinical characteristics associated with engagement in Nurse WRITE, engagement challenges, and the relationship between engagement and perceived symptom controllability. Three main findings emerged. First, education was the most consistent patient characteristic associated with specific engagement measures. Second, qualitative findings contextualized engagement challenges, including intervention complexity, written communication demands, symptom burden, competing responsibilities, and early disengagement. Third, higher overall engagement and greater socioaffective and cognitive engagement were associated with improved symptom controllability at the end of the intervention.

Quantitative findings identified education level as the most consistent predictor of specific engagement measures, consistent with a prior review [12]. Qualitative analysis further explained this association, suggesting that participants with lower education levels may have struggled with the length and complexity of nurses’ messages, which is the primary mode in Nurse WRITE. For example, 20% to 30% of less engaged participants, compared to high engagers, reported confusion about the intervention process, even though nurses had explained it at the outset. These challenges may also reflect differences in health literacy, communication preferences, or other linguistic factors, which could influence participants’ ability to engage with written intervention materials [41]. Future research should further explore how these factors intersect with education level to affect engagement in DHIs.

Compared to SD WRITE, where participants interacted with self-directed modules without nurse involvement, the presence of nurses in Nurse WRITE may have influenced how less well-educated patients engaged with the intervention. Participants might have felt self-conscious about their writing when communicating with nurses and faced challenges such as website navigation and confusion regarding the intervention process. Previous studies have linked computer literacy and DHI engagement positively [42-44], but none, including ours, adequately measured baseline computer literacy. Therefore, future provider-guided DHIs should incorporate this measure into their designs to confirm its impact.

Although nurses were encouraged to tailor their responses to participants’ capacities and needs, they faced challenges adhering to intervention protocols. For instance, while some participants preferred phone communication, nurses were restricted to asynchronous message boards. Additionally, nurses were required to follow structured intervention phases, even when some participants wanted to skip directly to solutions, bypassing representational assessments or discussions of concerns (eg, opioid addiction). While the relationship between lower education and lower engagement warrants further investigation, identifying education level as a potential risk factor for reduced engagement with specific intervention activities could help target participants who may need additional support.

Quantitative findings showed an exploratory pattern, suggesting that patients with higher symptom severity may have completed more care plan reviews, which could reflect motivation to revisit and refine symptom management strategies. However, qualitative findings indicated that deteriorating health and ongoing treatments posed significant challenges, often requiring hospitalization or medical appointments that limited participants’ ability to remain engaged. Together, these findings suggest that symptom burden may both motivate and hinder engagement.

Additionally, quantitative findings showed exploratory trends, suggesting that comorbidities, such as hypertension and diabetes, may pose additional challenges for patients with recurrent ovarian cancer and may affect participation. These findings align with previous research on engagement in DHIs [12]. Complementing these results, qualitative data revealed that participants often prioritized managing their health over engaging in symptom management activities on the message boards. Beyond health-related challenges, competing responsibilities, including demanding work schedules, family obligations, and other social commitments, also reduced participants’ time and energy for learning and managing symptoms. Future research with larger sample sizes is needed to confirm the impact of these factors.

Lastly, qualitative findings indicated that over half of the low engagers stopped participating after the introduction and representational assessment phase. This early disengagement may reflect challenges in establishing trust and rapport, underscoring the importance of relationship building during the orientation phase [45]. Early video conferencing for introductions and orientation allows nurses to gather verbal and nonverbal cues related to symptomatology [46], facilitating a quick understanding of the participant’s condition and the development of effective rapport.

Our study highlighted the importance of socioaffective engagement in improving symptom control. This type of engagement involves expressing positive emotions, valuing nurse input, using the nurse’s name, sharing personal experiences, showing vulnerability, and seeking ongoing interactions. These actions help build trust and foster a strong nurse-patient relationship, which is essential for effective online communication, learning, coordination, and symptom management.

The Nurse WRITE program supported this engagement through features such as nurse bios, photos, personalized web pages, and empathetic communication training. Future provider-guided DHIs could further enhance these efforts by accommodating participants’ communication preferences, such as voice-to-text conversion, and synchronous options, such as chat rooms, phone calls, text messages, and emails, to facilitate continuous interactions.

In the cognitive dimension, higher participant engagement significantly improved symptom control through effective coordination with nurses. Engagement included actions such as responding to questions, fulfilling requests, expressing agreement or disagreement, seeking information, and asking clarifying questions. This collaboration fostered knowledge sharing and the personalization of care plans, ultimately boosting patient confidence in managing symptoms post intervention.

Our study revealed that participants completing more care plans managed their symptoms better. This process, involving 6 out of 7 intervention elements, enhanced patients’ symptom understanding, corrected misconceptions, and built knowledge and confidence in effective symptom management. While such step-by-step processes may yield greater control and engagement [47], this method requires patience and effort, and failing to find a solution can lead to frustration and disengagement. Our qualitative analysis showed that the majority of low-engagement participants disengaged after the introduction phase but before completing a care plan. Previous studies have also suggested a connection between DHIs that allow participants to choose their interaction style and increased engagement [48-50]. Therefore, future provider-guided DHIs should aim for a harmonious blend of protocol adherence and flexibility, such as allowing early care plans, collaborative adjustments, and visual progress tracking on the homepage to enhance meaningful behavioral engagement and prevent frustration.

Although nurses worked to tailor communication to patients’ needs, strict protocol adherence posed challenges. Future DHIs should address these limitations by introducing greater flexibility, simplifying content, and utilizing diverse formats and technologies, such as visual aids, serious games, AI chatbots, or self-learning materials, to enhance comprehension and sustain engagement.

High completion rates of intervention questions and requests in participant posts indicated effective collaboration with nurses, with comprehensive responses addressing multiple questions. On the other hand, low completion rates led to recurring queries and extended communication, causing participant frustration and delaying symptom management. In Nurse WRITE, some nurses enhanced interactions by numbering questions and providing personalized feedback tailored to participants’ progress, goals, and challenges. Future DHIs could enhance participant attention and efficiency by incorporating formatting elements like color or italics to highlight key questions. Additionally, AI tools could prompt participants to address unanswered questions or requests before submitting their posts, fostering more efficient and continuous interactions.

These findings should be interpreted within the context of a structured, provider-guided intervention. In broader real-world settings, variations in provider availability, workflow constraints, and access to technology may influence engagement differently. Future research is needed to examine how these engagement patterns and influencing factors translate to less controlled clinical, community-based, and rural environments.