This cross-sectional survey was conducted on the basis of the CHERRIES (Checklist for Reporting Results of Internet E-Surveys) [25] (Checklist 1).

The questionnaire, created in Japanese, consisted of 21 screens and 52 items on the website and was designed to be answerable within 25 minutes (Multimedia Appendix 1). To prevent missing data, all questions were set as mandatory in the web-based questionnaire. To eliminate duplicate responses, any responses that matched IP address, sex, age, and place of residence were considered to be from the same respondent, and only the first response was included in the analysis.

Questions regarding patients with cancer information-seeking behavior were developed with reference to previous studies [26] that aimed to clarify the information needs and difficulties encountered when searching for information and using psychosocial support services among patients with ovarian cancer, as well as the patient experience survey [27] conducted by the National Cancer Center. During development, 3 researchers with experience in medical research and survey studies involving individuals with diseases examined whether each question appropriately reflected the research objectives and theoretical framework from the perspective of content validity and revised the questions for clarity of wording and to minimize respondent burden.

First, 12 items on sociodemographic information (including sex, age, place of residence, and occupation) and clinical information (including type of cancer, cancer stage, treatment status, and recurrence or metastasis) were collected using a single-choice format during the screening phase. This allowed for the collection of basic attribute information to uniquely classify each respondent. In addition, as a basic indicator related to information seeking, “device used to gather information” (1 item) was assessed using a single-choice format.

For information-seeking behavior regarding cancer, we started with the question “Have you ever searched for information about cancer?” Four multiple-choice checkbox items were used: “sources of cancer-related information,” “social networking service (SNS) or apps used for collecting cancer-related information,” and “advantages of psychosocial support and information-gathering services recommended by medical institutions.” For these questions, mutually exclusive options such as “did not search,” “do not wish to obtain information,” “do not use SNS or apps,” and “not applicable” were provided to ensure consistency in responses.

Furthermore, skip logic was introduced: if a respondent answered “did not search” to the question “Have you ever searched for information about cancer?” they were not required to answer questions about “sources of cancer-related information,” “devices used to collect information,” or “SNS or apps used for collecting cancer-related information,” and were instead directed straight to the question about “difficulties in information seeking.” Additionally, even if respondents had searched for information, if they selected either “do not wish to obtain information” or “do not use SNS or apps,” they were not asked to answer the question about the “device used to gather information.”

For the intention to use psychosocial support and information provision services recommended by medical institutions (corresponding to “behavioral intention” in the TPB), a yes or no question—“use of hospital-recommended apps and services for counseling support and information gathering”—was used (1 item). Only participants who answered “Yes” were able to proceed to the subsequent question on “advantages of psychosocial support and information-gathering services recommended by medical institutions.”

“Difficulties in information seeking” consisted of 7 items, “trusted information sources” consisted of 8 items, and “evaluation of interpersonal relationships between patients and health care providers” consisted of 9 items. To reduce the respondent burden, these were presented in a matrix format and measured using a 5-point Likert scale.

To assess personality traits, the Japanese version of the 10-Item Personality Inventory (TIPI-J) [28], which uses 10 items and a 7-point Likert scale, was used. The TIPI-J measures the Big Five personality dimensions—extraversion, agreeableness, conscientiousness, neuroticism, and openness—with 2 items each and has shown high internal consistency: extraversion (α=.92), agreeableness (α=.85), conscientiousness (α=.82), neuroticism (α=.91), and openness (α=.86) [28]. The TIPI-J has also been shown to be associated with the emotional and cognitive health of patients with brain tumors [29], making it a valid indicator for this study.

To ensure accessibility and usability, a test page simulating the survey environment was created before the actual survey to confirm that the question display and skip logic functioned accurately.

The survey was conducted privately by Asmarc Inc (a Japanese marketing research company) from December 10 to December 12, 2024. The company operates an online research panel with over 16 million registered members. After agreeing to the membership terms, registered members completed preliminary and final registration, during which they provided basic demographic information, such as age, sex, place of residence, employment status, and medical history.

The survey was conducted using the D-style web and Monitas platforms (Asmarc Inc) and was conducted using a list of available questionnaires. The eligibility criteria for this study were based on patient-reported outcome criteria for survivors with cancer 1‐5 years after diagnosis [30]. Participants were required to be between 20 and 89 years of age at the time of the survey, to have been diagnosed with cancer at least 1 year prior, and to be either currently undergoing treatment or within 5 years of completing treatment. The following exclusion criteria were applied: participants who reported their treatment status as “before cancer treatment” or “other,” those who had been diagnosed with cancer less than 1 year prior, and those for whom more than 10 years had passed since the end of treatment.

The sample size was calculated using cancer incidence statistics in Japan [31] as the reference population. Assuming a 90% CI and a 5% margin of error, the required sample size was estimated to be approximately 270 participants. This calculation was performed using the pwr package (version 1.3-0) in R. The pwr package was originally developed by Stéphane Champely and is currently maintained by Helios De Rosario. Considering a previous recommendation that at least 300 participants are desirable for conducting LCA [32], the minimum target sample size for this study was set at 300 participants.

In this study, only data for participants with responses to all questions were included, and statistical analyses were performed using R (version 4.3.2; R Foundation for Statistical Computing). Sociodemographic information was stratified on the basis of the use of psychosocial support services, specifically by responses of “Yes” or “No” to “use of hospital-recommended apps and services for counseling support and information gathering.” Descriptive statistics were calculated for each group. The sociodemographic information included sex, age, place of residence, occupation, and “device used to collect cancer information.” Cancer-related information consisted of “cancer stage,” “cancer type,” “presence or absence of recurrence,” “presence or absence of metastasis,” “current treatment status,” “time of cancer diagnosis,” and “time of cancer treatment completion.” Place of residence was recategorized into 6 categories: Hokkaido and Tohoku, Kanto, Chubu, Kinki, Chugoku and Shikoku, and Kyushu and Okinawa. To simplify the classification of cancer types, lung cancer and small cell lung cancer were grouped as lung cancer; esophageal, gastric, small intestine, and colorectal cancers were grouped as gastrointestinal cancer; liver, biliary tract, and pancreatic cancers were grouped as liver, biliary tract, and pancreatic cancer; kidney cancer and cancers of the renal pelvis and ureter and urethra, and bladder cancer were grouped as kidney, ureter, and bladder cancer; ovarian, endometrial, and cervical cancers were grouped as gynecologic cancer; leukemia, malignant lymphoma, and multiple myeloma were grouped as hematologic cancer; and brain tumors, head and neck cancer, thyroid cancer, cancer of unknown primary, skin cancer, and malignant melanoma were grouped as other. For information-seeking behaviors of patients with cancer, descriptive statistics were calculated regarding “having searched for information about cancer,” “sources of cancer information,” “SNS and apps used to gather cancer information,” and “benefits of using psychosocial support and information-gathering services recommended by medical institutions.” To understand the relational structure among items, analyses were conducted using tetrachoric correlations. Exclusive options such as “did not search” for “having searched for information about cancer,” “did not want to obtain information” for “sources of cancer information,” “do not use SNS or apps” for “SNS and apps used to gather cancer information,” and “not applicable” for “benefits of using psychosocial support and information-gathering services recommended by medical institutions” were excluded from the statistical analyses. For the TIPI-J, descriptive statistics were calculated, and the relational structure among the items was examined using polychoric correlations.

Exploratory factor analysis (EFA) was conducted according to the reporting standards proposed by Watkins [33]. EFA was chosen to identify the underlying factor structure. The measurement scales used were “behavioral barriers experienced when collecting information about cancer,” “reliable information sources,” and “relationships with doctors, medical professionals, and society.” Each item was rated on a 5-point Likert scale ranging from 1 to 5. The appropriateness of conducting EFA was confirmed using the Bartlett test of sphericity [34]. The criteria for determining the number of factors included the Kaiser criterion of eigenvalues greater than 1 [35] and the minimum average partial (MAP) method [36]. Eigenvalues exceeded 1 up to the fifth factor (fifth factor=1.26), and the MAP exhibited its lowest value (0.03) at the fifth factor; therefore, a 5-factor solution was adopted. To choose the factor extraction method, multivariate normality was examined using the Mardia test [37]. Because skewness (skewness=62.14; P<.01) and kurtosis (kurtosis=613.96; P<.01) were significant, multivariate normality was rejected, and a method that did not assume normal distribution was required. Moreover, because the items used in this study were ordinal scales based on a 5-point Likert scale, factor analysis using the minimum residual method with a polychoric correlation matrix, which is suitable for ordinal data, was adopted [38,39]. Oblimin rotation [40] was used for simplification and theoretical convergence. Because oblique rotation was used, pattern coefficients of ≥0.37 were judged to be theoretically meaningful loadings, and coefficients of ≥0.40 were considered the primary criterion [41] for substantial interpretation. Cross-loadings of ≥0.3 were examined for interpretative caution. Internal consistency was determined using Cronbach α and McDonald ω, and factors with a reliability of ≥0.7 [42,43] and theoretical significance were judged as appropriate. The analysis was conducted using the psych package (version 2.6.2) in R. The psych package was developed and is maintained by William Revelle [44].

In this study, we hypothesized that information-seeking behaviors and the use of psychosocial support services among survivors with cancer are not uniform and that multiple latent profiles exist. To identify these subgroups on the basis of statistical modeling, we used LCA [45]. Unlike traditional distance-based cluster analysis, LCA estimates latent structures using probabilistic models, allowing for objective model comparisons using information criteria, making it well suited for identifying latent profiles [46]. The analysis followed the reporting standards of Weller et al [47], and the optimal number of latent classes was determined on the basis of statistical fit indices and theoretical interpretability [48]. For the measurement variables, we recoded variables to ensure estimation stability, categorizing age, occupation, treatment status, and information-gathering devices. To examine the indicators included in the LCA, we conducted a polychoric correlation analysis and included 10 variables—sex, age, employment status, cancer type, stage, recurrence, metastasis, treatment status, time since completing treatment, and information-gathering devices—which were confirmed to have intervariable correlations of ±0.3 or higher. Model fit was assessed primarily using Bayesian information criterion (BIC), adjusted Bayesian information criterion (aBIC), and consistent Akaike information criterion (cAIC), and the model with the smallest value was selected as a candidate. Entropy was checked, with ≥0.8 considered acceptable and ≥0.9 considered ideal [48]. For the final model selection, we carefully evaluated whether each class had a sample size of n>50 or >5% of the total sample and whether each class profile was clinically meaningful. The R poLCA package (version 1.6.0.1), developed by Drew A Linzer and Jeffrey B Lewis, was used for the analysis [49].

We examined the differences in each survey item among the latent classes. The variables compared included “counseling support and information gathering through apps and services recommended by the hospital,” “survey items related to cancer,” “sources of information about cancer,” “SNS and apps used to gather cancer information,” and “advantages of using psychosocial support and information-gathering services recommended by medical institutions,” as well as the factors identified in the EFA and the TIPI-J. To assess statistical significance, we adopted nonparametric testing methods with a significance level of P<.05. For binary items, we used the chi-square test for independence, and for ordinal items, we used the Kruskal-Wallis test. For items where significant differences were observed, we conducted residual analysis for binary items and the Wilcoxon rank-sum test (P=.02) for ordinal items as post hoc analyses using the Bonferroni correction and evaluated the association between the latent classes and each survey item.

This study was approved by the Research Safety and Ethics Committee of Tokyo Metropolitan University of Industrial Technology (approval no 23020). The survey was conducted in compliance with the Declaration of Helsinki and relevant ethical guidelines, and only those who provided informed consent participated in the study. The collected data were securely stored in a password-protected database that was accessible solely to the research team. Participants were compensated with reward points according to the regulations of each platform. The allocation and exchange of points were managed exclusively by the survey company, and the research team was not informed about individual point allocations. Participants who responded via the D-style web platform received 2 points for the preliminary survey and 15 points for the main survey, totaling 17 points upon completing both surveys. Participants who responded via the Monitas platform received 2 points for the preliminary survey and 9 points for the main survey, totaling 11 points upon completing both surveys.

Of the 4183 members registered on the D-style web and Monitas platforms, 1095 individuals who indicated in their preregistration information that they “currently have cancer or have had cancer in the past” were selected for the screening survey. Of these, 627 individuals responded (participation rate: 57.3%). Among the respondents, 350 met the eligibility criteria for the main survey and proceeded to participate (transition rate: 55.8%). Accordingly, the final analysis included 350 survivors with cancer (350/1095, 31.9%; Figure 1).

The survey participants comprised 175 men and 175 women (175/350, 50%). The average age was 57.9 (SD 10.8) years. The largest age group was 60‐69 years (138/350, 39.4%), followed by 70‐79 years (89/350, 25.4%). Regarding occupation, the most common response was unemployed (108/350, 30.9%), followed by company employees (84/350, 24%). Most participants resided in the Kanto region (156/350, 44.6%). Regarding clinical characteristics, the most common types of cancer were breast cancer (93/350, 26.6%) and gastrointestinal cancer (91/350, 26%). Many patients were at an early stage, with stage 1 accounting for 156 of 350 (44.6%) participants. Regarding treatment status, more than half had completed treatment and were undergoing regular follow-up visits at the hospital (202/350, 57.7%). Smartphones (193/350, 55.1%) were the most commonly used devices to gather information about cancer. In contrast, 105 of 350 (30%) participants reported that they did not search for information. Regarding the use of hospital-recommended apps and services for counseling support and information gathering, 282 of 350 (80.6%) participants answered “yes,” and 68 of 350 (19.4%) participants answered “no.” Among participants who reported that they did not search for information in response to the question about devices used to gather information about cancer (105/350, 30%), most also reported “yes” regarding the use of hospital-recommended apps and services for counseling support and information gathering (91/105, 86.7%), with only 14 of 105 (13.3%) reporting “no.” Of those who answered “yes,” 161 of 282 (57.1%) had completed treatment and were under regular follow-up, and 72 of 282 (25.5%) were receiving outpatient or home medical care. Smartphones were the most commonly used devices to gather information about cancer (161/282, 46%; Table 1).

The mean scores reported by respondents were highest for neuroticism at 8.62 (SD 2.47), followed by openness at 8.24 (SD 2.43), extraversion at 7.65 (SD 2.6), and conscientiousness at 7.15 (SD 2.29). Agreeableness exhibited the lowest value among the 5 traits, at 5.84 (SD 2.03). A Shapiro-Wilk test [50] was conducted to assess the distribution of each scale, and the results were significant for extraversion (W=0.98; P<.001), agreeableness (W=0.95; P<.001), conscientiousness (W=0.98; P<.001), neuroticism (W=0.98; P<.001), and openness (W=0.98; P<.001), indicating that the data were not normally distributed. The correlations among the subscales were uniformly low (r=−0.43 to 0.37; Multimedia Appendix 2).

The Bartlett test of sphericity was conducted to assess the applicability of factor analysis. On the basis of the results, the null hypothesis that the correlation matrix is an identity matrix was rejected (χ²₂₁₀=5425.91; P<.001), indicating that there was a sufficiently robust correlation structure among the items. The Kaiser-Meyer-Olkin statistic was 0.81, confirming sampling adequacy and determining that factor analysis was appropriate. The number of factors was determined on the basis of eigenvalues, MAP, and theoretical consistency. Eigenvalues exceeded 1 up to the fifth factor (factor 1=6.33, factor 2=4.27, factor 3=2.56, factor 4=1.86, and factor 5=1.26), and the cumulative contribution of the 5-factor solution was 67.83%. MAP decreased with the addition of more factors and improved up to the fifth factor. In addition to these results, considering that the measurement included several conceptual domains such as “difficulties in information seeking,” “support and consultation for care,” “relationship with the physician,” “trust in information from medical institutions and specialized websites,” and “trust in information from outside medical institutions,” a 5-factor solution was adopted.

Factor 1 included items reflecting difficulties encountered in the process of seeking cancer-related information (factor loadings: 0.54-0.9) and was named “difficulties in information seeking.” Factor 2 included items related to continued employment before treatment, concerns about appearance, convalescence, precautions in daily life, and workplace considerations (factor loadings: 0.42-0.74) and was interpreted as “evaluation of support for cancer care.” Factor 3 included items on the primary physician’s knowledge and experience, ease of consultation, and trust in information provided by physicians (factor loadings: 0.67-0.81) and was named “evaluation of the relationship with the physician.” Factor 4 included trust in information from cancer-specialized or hospital or pharmaceutical company websites (factor loadings: 0.73-0.85) and was named “evaluation of the reliability of information from medical institutions and cancer-related websites.” Factor 5 included trust in sources of information outside medical institutions, such as family and acquaintances, other patients, and social networking sites (factor loadings: 0.47-0.76) and was named “assessment of the reliability of information from nonmedical institutions.”

The item “I had to visit several websites” loaded on both Factor 1 (0.60) and Factor 4 (0.33), and “the most reliable source is information from the physician” loaded on both Factor 3 (0.67) and Factor 4 (0.32). The internal consistency of the factors ranged from α=.75 to .9 and ω=0.78 to 0.92, with small interfactor correlations, the highest being between Factors 2 and 3 (r=0.47), justifying the treatment of the 5 factors as independent subconcepts (Table 3).

In determining the number of latent classes, models with 2-6 classes were estimated, and BIC was used as the primary index with comparisons to aBIC and cAIC. The 3-class model had the lowest BIC (5734.74) and cAIC (5805.74) and demonstrated high classification accuracy with an entropy of 0.95. Although the 4-class model had the lowest aBIC (5465) and similarly high entropy (0.96), both BIC and cAIC worsened (BIC=5766.37 and cAIC=5861.37). For models with 5-6 classes, the information criteria further increased, suggesting model overfitting. In finalizing the model, it was confirmed that each class had a sample size of n>50 or >5% of the total and that the class profiles were clinically interpretable. On the basis of minimum BIC and cAIC values and high classification accuracy, the 3-class model was selected as the final model (Table 4).

After fitting the model, the results for the 3 classes identified by the LCA were as follows. Class 1 comprised 95 individuals, all of whom were female (95/95, 100%). Most patients in Class 1 were between the ages of 50 and 69 years, with 35 of 95 (36.8%) aged 50‐59 years and 31 of 95 (32.6%) aged 60‐69 years. Regarding treatment status, 74 of 95 (77.9%) patients were undergoing regular follow-up after completing treatment, while 21 of 95 (22.1%) were in the follow-up stage without needing to visit a medical facility. On the basis of these treatment characteristics, this class was categorized as the “group of women under observation.” Breast cancer was the most common cancer type (52/95, 54.7%), followed by gynecological cancers (19/95, 20%). In terms of cancer stage, more than half of the patients were classified as stage 1 (56/95, 58.9%). Of the 95 patients, recurrence was observed in 8 (8.4%) and metastasis in 4 (4.2%). Regarding employment status, 51 of 95 (53.7%) patients were employed. For information-seeking behavior, 43 of 95 (45.3%) patients answered, “I do not use devices to gather information about cancer.” Among those who gathered information, smartphones were used by 39 of 95 (41.1%) patients; however, overall, information-seeking behavior was limited compared with that of patients in the other classes.

Class 2 comprised 163 individuals, the majority of whom were male (141/163, 86.5%). Most patients in Class 2 were aged 60 years or older, with 144 of 163 (88.3%) patients in this range; 76 of 163 (46.6%) patients were aged 60‐69 years, and 68 of 163 (41.7%) patients were ≥70 years. Regarding treatment status, 128 of 163 (78.5%) patients were under regular observation after completing treatment, while 35 of 163 (21.5%) were in the follow-up stage without requiring hospital visits. These features led to this class being categorized as the “group of men under observation.” The most common cancer was gastrointestinal cancer (69/163, 42.3%), followed by prostate cancer (29/163, 17.8%). Most cases were classified as early stage, with 76 of 163 (46.6%) patients at stage 1 and 37 of 163 (22.7%) patients at stage 2. Recurrence occurred in 21 of 163 (12.9%) patients, and metastasis occurred in 14 of 163 (8.6%) patients. These proportions were higher than those in Class 1 but lower than those in Class 3. Regarding employment, 92 of 163 (56.4%) patients were not employed. The device most commonly used for information seeking was a smartphone (104/163, 63.8%), and 31 of 163 (19%) patients answered “I do not use devices to gather information about cancer”; this proportion was lower than that among participants in the other classes.

Class 3 comprised 92 individuals, all of whom were undergoing treatment (92/92, 100%). There were 59 (64.1%) females and 33 (35.9%) males, with the majority aged between 50 and 69 years (50‐59 years: 25/92, 27.2% and 60‐69 years: 31/92, 33.7%). Breast cancer was the most common cancer (41/92, 44.6%), followed by gastrointestinal (10/92, 10.9%) and prostate cancers (10/92, 10.9%). This class had a higher proportion of advanced cancers, with stage 4 accounting for 25 of 92 (27.2%) patients. Recurrence was observed in 33 of 92 (35.9%) patients and metastasis in 37 of 92 (40.2%) patients. Both proportions were higher in Class 3 than in the other 2 classes. Regarding information-seeking behavior, 50 of 92 (54.4%) patients used smartphones, but 31 of 92 (33.7%) patients answered “I do not use devices to gather information about cancer.” Given these features, this class was defined as the “group under treatment” (Figure 2 and Multimedia Appendix 2).

All results were obtained from post hoc analyses. Regarding “use of hospital-recommended apps and services for counseling support and information gathering,” no significant differences were found. As shown in Figure 3, regarding “things I’ve researched about cancer,” the results revealed that “information about other patients” was significantly lower in Class 2 than in the other classes (χ²_₂=12.22; P=.002, post hoc test: Class 1 > Class 2; P=.009, and Class 2 < Class 3; P=.001). Regarding “life during treatment” (χ²_₂=7.22; P=.03 and post hoc test: Class 2 < Class 3; P=.01), Class 3 was significantly higher than Class 2. Regarding “sources of cancer information,” “Cancer Consultation Support Center” (χ²_₂=17.31; P<.001 and post hoc test: Class 2 < Class 3; P<.001) and “National Cancer Center website” (χ²_₂=7.59; P=.02 and post hoc test: Class 2 < Class 3; P=.01) were significantly higher in Class 3 compared with Class 2. Regarding “SNS and apps used to gather cancer information,” “Instagram” (χ²_₂=12.57; P=.002 and post hoc test: Class 2< Class 3; P=.001) and “X (Twitter)” (χ²_₂=9.55; P=.008 and post hoc test: Class 2 < Class 3; P=.002) were significantly higher in Class 3 compared with Class 2. Regarding “blog” (χ²_₂=5.82; P=.054), no significant differences were found between classes after the post hoc test. For “benefits of using psychosocial support and information-gathering services recommended by health care institutions,” regarding “facilitate communication with doctors and medical staff” (χ²_₂=12.57; P=.02 and post hoc test: Class 1 < Class 2; P=.006), Class 2 was significantly higher than Class 1 (Figure 3 and Multimedia Appendix 2).

According to Figure 4, for “Factor 1: difficulties in information seeking” (χ²_₂=11.9; P=.003 and post hoc test: Class 2 < Class 3; P=.007), Class 3 showed a significantly higher value than Class 2. For “Factor 3: evaluation of the relationship with the physician” (χ²_₂=12.2; P=.002 and post hoc test: Class 2 > Class 3; P=.002), Class 2 had a significantly higher value than Class 3. For “Factor 5: assessment of reliability of information from nonmedical institutions” (χ²_₂=12.3; P=.002 and post hoc test: Class 1 > Class 2; P=.002), Class 1 showed a significantly higher score than Class 2. Regarding the TIPI-J, agreeableness (χ²_₂=6.7; P=.04) and neuroticism (χ²_₂=8.8; P=.01) did not show significant differences after the post hoc test (Figure 4 and Multimedia Appendix 2).

The results revealed that age distribution and cancer type composition among the respondents in this study were generally consistent with trends in Japan’s national cancer statistics [31]. In addition, more than 80% of respondents indicated that they would like to use psychosocial support services or apps if recommended by medical institutions. Furthermore, in the results of the LCA, the “female under observation” group showed a high level of trust in “Factor 5: assessment of the reliability of information from nonmedical institutions,” whereas the “male under observation” group was less proactive in gathering information but placed a high value on “Factor 3: evaluation of the relationship with the physician” and “facilitate communication with doctors and medical staff,” indicating that they prioritize relationships with health care professionals. The “under treatment” group was the most active in collecting information, frequently using “information about other patients,” “life during treatment,” “Cancer Consultation Support Center,” and “National Cancer Center website.” However, the “Factor 1: difficulties in information seeking” score was also high, suggesting that treatment status and gender may influence barriers to gathering information and differences in trusted sources.

This study suggests that integrating the 3 factors from the TPB (ie, “attitude,” “subjective norms,” and “perceived behavioral control”) with the “relationship function” and “task-oriented function” from the patient-provider relationship model is associated with a stronger intention to use psychosocial support services and hospital-recommended apps. Within the TPB framework, hospital recommendations may consistently shape behavioral intentions as a form of subjective norm. Previous studies have also reported that recommendations from health care providers strongly influence information-seeking behavior and support service usage among patients with cancer [51], and the results of this study are consistent with these findings.

Looking at each group, the male group under follow-up showed a high evaluation of their relationship with physicians and a strong trust in physician-provided information, with subjective norms in the TPB being particularly dependent on physicians. Previous research has also shown that male patients with cancer place a high value on trustworthiness and sincerity in their interactions with health care providers [52]. Furthermore, in the male follow-up group, the relationship function based on trust in physicians appears to play a central role, possibly serving as a prerequisite for accepting psychosocial support services and information provision.

In contrast, the female group under follow-up showed relatively higher trust in “information from other patients” and nonmedical sources, suggesting that peer-derived experiential knowledge plays an important role in forming attitudes in the TPB. Earlier studies, particularly those focusing on patients with breast cancer, have reported that peer support contributes to acquiring information and alleviating anxiety about recurrence [53-55], which is consistent with the current findings. For this group, it is also possible that task-oriented functions, such as managing uncertainty and decision support, are supplemented outside health care institutions. By providing opportunities for exchange with other patients or sharing experiential information, health care providers may help translate behavioral intentions into action.

The “under treatment” group displayed the highest level of active information-seeking behavior among the 3 groups, but also the greatest difficulty in seeking information. In terms of the TPB framework, while “behavioral intention” to seek information or use psychosocial services was high, “perceived behavioral control” appeared to be low. From the perspective of the patient-provider relationship model, it was suggested that the task-oriented functions for exchanging information and managing uncertainty, and the relationship functions providing reassurance and emotional support, may not be sufficiently aligned with patients’ needs. Survivors with cancer undergoing treatment have strong information needs and clear intentions to seek information but may lack sufficient self-efficacy to understand the information and make decisions, leading to increased difficulty. Prior studies have also reported that patients with cancer in advanced or treatment phases experience heightened anxiety from online information [56], lack of social support beyond the internet, and negative treatment experiences [57]. Such negative experiences can decrease patients with cancer’s self-efficacy and trust in their digital skills, making information seeking a heavy burden. Therefore, for survivors with cancer undergoing treatment, clarifying reliable information sources and having health care providers organize and supplement information is crucial for alleviating the psychological burden associated with information-seeking.

The strength of this study lies in its analysis of information-seeking behaviors of survivors with cancer and usage of psychosocial support services, grounded in the 3 components of TPB, as well as the communication functions of relationship-oriented and task-oriented functions in the patient-provider relationship model. By using EFA to identify factors and subsequently using LCA to classify multiple subgroups, this study systematically captured the behavioral intentions underlying information-seeking activities. This approach identified diverse needs among subgroups—needs that are often difficult to capture by a simple comparison of attributes—regarding trusted sources of information and communication tendencies among survivors with cancer. In particular, differences in the levels of trust in information sources between men and women during the follow-up period highlight the importance of personalized information provision, offering practical insights for future patient support service design. However, this study involved several limitations that should be considered. First, although the target sample size was achieved and trends were found to partially align with national cancer incidence statistics, the sampling was not stratified by age or cancer stage. Consequently, there was a bias toward survivors in stages 1 and 2, whereas the proportion of survivors in stages 3 and 4, or those who experienced recurrence or metastasis, was relatively low. Given the imbalances in cancer type, age group, and stage, caution is warranted regarding the generalization of the study findings to all survivors with cancer. Second, this was a cross-sectional study aimed primarily at identifying tendencies in types of information-seeking behaviors and thus cannot establish causal relationships related to various factors and backgrounds associated with each survey item. Detailed factor analyses and longitudinal studies are necessary to elucidate how information-seeking behaviors and perceptions evolve over time. Third, because this study focused on differences by treatment status and gender, socioeconomic factors such as income, education, and digital literacy were not included in the survey. However, these factors may also influence information-seeking behaviors and the use of support services, suggesting that future research should comprehensively examine these variables. Fourth, the questionnaire used in this study was developed on the basis of previous research and theoretical frameworks. However, because no formal pilot testing or test-retest reliability verification was conducted before the survey, the temporal stability of the measurement could not be assessed. This represents a limitation, and caution is required when interpreting the results. In the future, a comprehensive psychometric evaluation, including test-retest reliability and measurement invariance, will be necessary.

This study clarified how information-seeking behaviors among survivors with cancer differ depending on treatment status and sex. By applying the TPB and patient-provider relationship models and using EFA, LCA, and multiple comparison analyses, we identified subgroups of patients with cancer with different behavioral characteristics. Our findings provide valuable insights for designing necessary approaches to promote personalized psychosocial support services and information provision for patients, as well as for improving the user interfaces of products and services that support patients psychologically and socially. Future research should focus on the realities of information provision and communication between patients with cancer and health care professionals, as they relate to individualized support for middle-aged and older survivors with cancer. Further exploration from multifaceted perspectives, including health care providers’ recognition of information delivery and patients’ acceptance of information, is expected to lead to the development and improvement of personalized and patient-centered communication approaches in the future.