Categorizing health beliefs among cancer survivors using latent profile analysis to identify targets for promoting medication adherence and quality of life Authors: Meng-Jung Wen, Daniel M. Bolt, Olayinka O. Shiyanbola Published: 2026 ABSTRACT Nearly half of cancer survivors are nonadherent to oral anticancer medications (OAM), yet no studies have systematically examined health belief profiles or identified belief-based intervention targets to improve adherence and health-related quality of life (HRQoL). This study aimed to categorize distinct health belief profiles and examine their associations with medication adherence and HRQoL among cancer survivors. This cross-sectional online survey study recruited participants from two crowdsourcing panels across the U.S. between June and July 2024. Latent profile analysis was used to categorize participants’ patterns of health beliefs. Eligibilities were self-reported who had a cancer diagnosis and were currently taking at least one OAM. Participants completed a 54-item online survey assessing cancer perceptions, beliefs about OAMs, medication adherence and HRQoL. One-way analysis of variance (ANOVA) and multivariate regression analysis were conducted to compare group differences and examine the associations between health belief profiles, adherence and HRQoL. Among 323 respondents, three distinct health belief profiles emerged, significantly associated with adherence (adjusted R2 = 0.491, F change = 27.07, p < 0.01) and HRQoL (adjusted R2 = 0.286, F change = 6.12, p < 0.01). The Optimistic and adaptive profile (13.3%) demonstrated strong understanding of OAM use, greater perceived control over recurrence, fewer negative cancer consequences, and high perceived necessity with minimal concerns about medication. The Conscious but hesitant profile (30.7%) maintained balanced perceptions of cancer and treatment but reported the lowest perceived necessity for medication. The Overwhelmed and unaware profile, the largest subgroup (56.0%), exhibited greater perceived cancer consequences, higher treatment concerns, the lowest adherence and the lowest HRQoL. Subgroup analysis revealed that specific beliefs significantly associated with nonadherence (perceived risk of recurrence) and HRQoL (cancer and treatment consequences and emotional responses), suggesting that reframing negative beliefs may improve outcomes. To support sustained medication adherence and HRQoL, interventions should focus on enhancing patient education about the role of OAMs in cancer control, addressing concerns about side effects, and enhancing access to resources that alleviate cancer-related consequences. Future research should explore the mechanisms through which health beliefs affect adherence and identify unmet emotional and informational needs among cancer survivors.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-47364-2. Introduction Nonadherence to cancer therapies poses a significant threat to patient outcomes. Approximately half of the cancer survivors are nonadherent to oral anticancer medications (OAMs) prescribed by their providers, creating a burden not only for themselves but also for their families and society1. In this study, the term ‘cancer survivors’ refers to individuals living with and beyond a cancer diagnosis, including those currently undergoing treatment. Recent evidence indicates that adherence rates for certain cancer types are as low as 33%, significantly below the widely recognized general adherence rate of 50% for patients with chronic disease2. Nonadherence has been shown to negatively impact health outcomes, leading to 20% early cancer recurrence, and four times more hospitalizations and premature death1,3,4. Additionally, nonadherence increases financial strain due to a 2.8 times increase in out-of-pocket costs compared to those who are adherent to OAMs5–7. Medication nonadherence arises from a complex interplay of multi-faceted factors8. Systematic reviews highlight individual-level barriers, particularly patient beliefs, are often more critical and modifiable through targeted interventions9–13. Notably, a recent review by Kaptein et al. emphasized that adherence among cancer survivors is shaped more by individual beliefs – such as perceptions of illness and beliefs about medicines – along with psychosocial factors than by demographic or clinical characteristics14. This underscores the need for tailored approaches to address these factors and improve adherence. To further unpack these individual-level factors, the Common-Sense Model of Self-Regulation offers a useful framework for understanding how patient beliefs—shaped by socio-cultural backgrounds and personal experiences—influence self-management decisions, including medication use15–17. Evidence continually suggests that enhancing cancer survivors understanding of their illness and fostering stronger beliefs in treatment efficacy could reduce concerns about OAMs, underscoring the critical role of addressing patient beliefs to improve medication adherence18–20. Person-centered categorization provides an innovative and essential approach to consider the complex interplay of patient interrelated beliefs, enabling the development of personalized interventions to improve medication adherence. Traditionally, variable-centered approaches that target single factor contributing to nonadherence among cancer survivors have proven to be less effective, as the assumptions often overlook how individuals form schemas – pattern of thinking – when making critical decisions21–23. This become important consideration as research reveals that many survivors hold ambivalence, weighing the perceived benefits of receiving treatment against the consequences of cancer and other social-structural barriers to managing their conditions24–26. This dual attitude complicates motivation and sets cancer survivors apart from patients with other chronic conditions, as they face unique challenges in balancing benefits and risks of medications amidst the uncertainty of long-term cancer management. To address these limitations, current research studying chronic diseases increasingly uses clustering approaches including latent profile or class analysis to identify subgroups of patients with shared health belief patterns, proving subgroup-specific insights and prioritizing populations for interventions27,28. The advantage of using person-centered approaches is to offer a more nuanced understanding of how patients make health decisions and also facilitate the development of tailored interventions that address specific beliefs or barriers within each subgroup. However, to date, no studies have systematically examined the types and prevalence of these health belief patterns among cancer survivors or their interactive influence on medication adherence. Therefore, this study aimed to (1) categorize cancer survivors into latent subgroups based on distinct health beliefs, and (2) examine the associations between these latent subgroups and medication adherence. By characterizing these health belief subgroups, the findings from this study can lay the groundwork for prioritizing interventions for high-risk groups and specifying behavioral targets across distinct subgroups. Methods Sample population A cross-sectional, online survey of patient health beliefs was administered to cancer survivors through two crowdsourcing panels across the United States. This study was conducted in accordance with the ethical standards of the Declaration of Helsinki. All study procedures were conducted following approval from the Institutional Review Board of the University of Wisconsin-Madison before collecting survey data and were determined to be exempt as minimal risk research. All participants provided informed consent prior to participation. Participants were eligible if they self-reported the following criteria: (1) 18 years or older, (2) a cancer diagnosis, (3) currently taking at least one oral anticancer medication, and (4) the ability to read and write in English. Participants were excluded if they did not consent to participate in the study and/or fail to pass the quality checks during survey completion. The online survey was disseminated via email through Centiment® research panel and the worker platform on Amazon Mechanical Turk®, containing a referral link to Qualtrics®. Centiment® is an online survey research panel composed of individuals who voluntarily enroll to participate in research studies, whereas Amazon Mechanical Turk® is an online crowdsourcing marketplace in which registered workers can complete tasks, including surveys, for compensation. Participants on both platforms were recruited through the existing panel infrastructure of each service, and eligible individuals who completed the survey received compensation in accordance with platform policies. The use of crowdsourcing panels is common in market research and has gradually expanded to health science to gather patient experiences within healthcare systems29,30. These web-based panels have successfully reached hard-to-reach populations and provided diverse patient characteristics across various cancer types and geographical regions30–32. A set of brief screening questions was included at the beginning of the survey to ensure that all participants met the eligibility criteria before proceeding to the main survey items. To enhance data quality, we implemented best practices for online survey data collection, including demographic verification (matching birth year and age), attention checks (one factual question) and a speeding check (excluding responses completed in less than 25% of the average completion time)30. Study measures Participants who met the inclusion criteria and consented to participate completed a self-reported, 54-item survey with ten health belief and two outcome measures using validated measures. The outcome measure was adherence to oral anticancer medication(s). The health belief items were used as indicators to identify belief patterns and included as independent variables associated with medication nonadherence. Two questionnaire instruments were used to collect participants’ health beliefs. First, a 32-item Revised Illness Perceptions Questionnaire for Breast Cancer Survivors (IPQ-BCS) was used to assess patients’ cognitive and emotional representations of illness33. This measure was selected for its relevance to the experiences of cancer survivors and its ability to capture their unique perceptions, including past illness, cure expectations, fear of recurrence, and treatment representations. The eight domains from the measure that we adapted to fit our research questions were Cure, Coherence (understanding of cancer), Cancer consequences, Risk of recurrence, Emotional representations, Personal control, Treatment control and Treatment consequences33. Second, the Beliefs about Medicine Questionnaire–Adjuvant Endocrine Therapy (BMQ-AET) was a self-reported questionnaire comprising 10 items designed to measure two domains of patients’ beliefs about medication34. The domains included patients’ beliefs regarding the necessity of taking medication (Necessity) and their concerns about medication use (Concern). The BMQ-AET consisted of five items for necessity beliefs and five for concern beliefs, utilizing a five-point Likert scale ranging from strongly disagree to strongly agree for each item. Two outcome measures were also collected. First, the primary outcome was medication adherence, using a 12-item, self-reported Adherence to Refills and Medications Scale (ARMS) to assess participants’ adherence to OAMs35. Scores higher than 12 indicated some degree of non-adherence, with the total score ranging from 12 to 48, where a lower score reflected better medication adherence. Second, participants’ quality of life was measured by the Centers for Disease Control and Prevention’s Health-Related Quality of Life (HRQoL)36. The composite score was used to measure self-perceived health status including four core items: general self-rated health, number of physically and mentally unhealthy days, and activity limitation days over the past 30 days37. This measure is widely used in population health research to assess health-related quality of life and to monitor trends and disparities across diverse populations. Statistical analysis Latent profile analysis was applied using Mplus Version 8.10 to identify latent subgroups of cancer survivors characterized by the ten health beliefs reported in the survey. The iterative model generation process began with a single-profile model and continued by adding one profile at a time until the models were no longer interpretable. Model fit was assessed using the Akaike information criterion (AIC) and Bayesian information criterion (BIC), Adjusted Lo-Mendell-Rubin Likelihood Ratio Test (ALMR), Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (VLME-LR), and Parametric Bootstrapped Likelihood Ratio Test (BLRT) p-value38. Entropy was also computed to evaluate the confidence in the probability of individuals belonging to a particular profile. The final number of latent health belief profiles was determined based on model fit indices, the size of each latent profile, and theoretical considerations38. Profile comparisons of differences in socio-demographic and clinical characteristics were conducted using one-way analysis of variance (ANOVA) or the chi-square test, as appropriate. Hierarchical multiple linear regression was performed to examine the association between health belief subgroups/profiles and adherence to OAMs, after controlling for participant demographics. The outcome variables were medication adherence to OAMs and HRQoL, while the independent variables included health belief profiles and demographics. In the basic model (i.e., Model 1 and 3), participant socio-demographic and clinical characteristics that were significantly identified in group comparisons using ANOVA were entered as covariates. In the advanced model (i.e., Model 2 and 4), health belief profiles were added to assess the extent to which each profile contributed to the variance in medication adherence and HRQoL while controlling for socio-demographic and clinical characteristics. Subgroup analysis was conducted in subgroup(s) with the lowest adherence to examine the health belief indicators within each profile membership associated with medication adherence. The two domain scores of beliefs about medicines and the eight domain scores of beliefs about cancer in each nonadherent subgroup served as independent variables, while the dependent variable was medication adherence within the lowest adherence subgroup. Socio-demographic and clinical characteristics significantly associated with medication adherence were included in the regression model as covariates. Descriptive statistics were also performed to summarize participants’ characteristics. Regression analysis was conducted using SPSS Version 28.0. Statistical tests were two-sided, with p < 0.05 considered statistically significant. Results A total of 1641 surveys were collected, with 323 participants being eligible and completing the study. The remaining surveys were not included due to participants not passing the initial screening for eligibility criteria. Model statistics The 3-profile model solution was identified in LPA. A summary of the model that fits statistics for potential profile models from one to six are presented in Supplementary Table 1. As the number of models increased, the values of AIC, BIC, and ABIC decreased. The BLRT p-value was significant (p < 0.001) for all six models. The entropy value was greater than 0.8 for each profile model; however, the 4-profile model had the lowest entropy value, and the size of one of the profiles was too small for meaningful interpretation. Regarding the practical interpretability of the models, although AIC, BIC and ABIC values continued to decrease, for the models with two and three profiles, the ALMR-LR and VLME-LR values were both significant (p < 0.05). Considering the complexity of cancer survivors’ health beliefs compared to other chronic diseases, the model with three potential profiles was selected for this study. Participant demographics Table 1 provides the participant socio-demographic characteristics grouped by the three profiles. The mean age of participants was 49.0 years (SD = 17.7). Most of participants were female (50.5%). The majority were White (80.2%) and Black or African American (14.2%). Breast cancer (32%) was the most common type of cancer. The average duration since cancer diagnosis was 7.32 years (SD = 10.0) and 42% had experienced cancer recurrence.Table 1Socio-demographic characteristics of three profiles.M, SD or n, %Total sample (n = 323)Profile 1: Optimistic and adaptive (n = 43)Profile 2: Conscious but hesitant (n = 99)Profile 3: Overwhelmed and unaware (n = 181)p-valueSocio-demographicsMale*159 (49%)14 (33%)45 (46%)100 (55%)0.02Age**49.00 (17.7)64.60 (14.7)51.30 (16.8)44.04 (16.5) < 0.01White259 (80%)38 (88%)74 (75%)147 (81%)0.15Degree or above265 (82%)35 (81%)83 (84%)147 (81%)0.86Employed full time**204 (63%)10 (23%)61 (62%)133 (74%) < 0.01Urbanicity55 (17%)8 (15%)20 (36%)27 (49%)0.11Health insurance316 (98%)43 (100%)95 (96%)178 (98%)0.25Clinical characteristicsCancer type0.06Breast cancer104 (32%)19 (45%)32 (32%)53 (29%)Prostate cancer47 (15%)8 (19%)16 (16%)23 (13%)Hematology cancer14 (4%)1 (2%)7 (7%)6 (3%)Cancer duration (years since diagnosis)7.32 (10.0)6.47 (8.8)6.65 (8.7)7.9 (10.9)0.51Early-stage cancer194 (61%)24 (59%)59 (60%)111 (63%)0.06Cancer recurrence**137 (42%)7 (16%)33 (33%)97 (54%) < 0.01OAM type** < 0.01Oral chemotherapy47 (17%)2 (4%)14 (30%)31 (66%)Hormone therapy87 (31%)26 (30%)31 (36%)30 (35%)Targeted therapy40 (14%)5 (13%)19 (48%)16 (40%)Multiple types106 (38%)4 (4%)19 (18%)83 (80%)OAM duration > 1 year*203 (63%)34 (17%)64 (32%)105 (52%)0.03Comorbidity**133 (41%)16 (37%)54 (55%)63 (35%) < 0.01M mean; SD standard deviation, F F-test; *p < 0.05, **p < 0.01; OAM oral anticancer medication. Table 1Socio-demographic characteristics of three profiles.M, SD or n, %Total sample (n = 323)Profile 1: Optimistic and adaptive (n = 43)Profile 2: Conscious but hesitant (n = 99)Profile 3: Overwhelmed and unaware (n = 181)p-valueSocio-demographicsMale*159 (49%)14 (33%)45 (46%)100 (55%)0.02Age**49.00 (17.7)64.60 (14.7)51.30 (16.8)44.04 (16.5) < 0.01White259 (80%)38 (88%)74 (75%)147 (81%)0.15Degree or above265 (82%)35 (81%)83 (84%)147 (81%)0.86Employed full time**204 (63%)10 (23%)61 (62%)133 (74%) < 0.01Urbanicity55 (17%)8 (15%)20 (36%)27 (49%)0.11Health insurance316 (98%)43 (100%)95 (96%)178 (98%)0.25Clinical characteristicsCancer type0.06Breast cancer104 (32%)19 (45%)32 (32%)53 (29%)Prostate cancer47 (15%)8 (19%)16 (16%)23 (13%)Hematology cancer14 (4%)1 (2%)7 (7%)6 (3%)Cancer duration (years since diagnosis)7.32 (10.0)6.47 (8.8)6.65 (8.7)7.9 (10.9)0.51Early-stage cancer194 (61%)24 (59%)59 (60%)111 (63%)0.06Cancer recurrence**137 (42%)7 (16%)33 (33%)97 (54%) < 0.01OAM type** < 0.01Oral chemotherapy47 (17%)2 (4%)14 (30%)31 (66%)Hormone therapy87 (31%)26 (30%)31 (36%)30 (35%)Targeted therapy40 (14%)5 (13%)19 (48%)16 (40%)Multiple types106 (38%)4 (4%)19 (18%)83 (80%)OAM duration > 1 year*203 (63%)34 (17%)64 (32%)105 (52%)0.03Comorbidity**133 (41%)16 (37%)54 (55%)63 (35%) < 0.01M mean; SD standard deviation, F F-test; *p < 0.05, **p < 0.01; OAM oral anticancer medication. Socio-demographic characteristics of three profiles. M mean; SD standard deviation, F F-test; *p < 0.05, **p < 0.01; OAM oral anticancer medication. Group differences in health beliefs Table 2 presents the mean score and standard deviation (SD) of each health belief domain across the three profiles. A clear pattern of negative perceptions of cancer emerged in the post-hoc test of mean scores across three profiles. Participants in Profile 3, labeled as Overwhelmed and unaware (n = 181, 56.0%) had significantly higher mean scores for Cancer consequences, Risk of recurrence and Emotional representations, followed by those in Profile 2, and Profile 1. However, a trend emerged where Profile 1, labeled as Optimistic and adaptive (n = 43, 13.3%) had higher perceived understanding (Coherence) and Personal control over cancer compared to other two profiles. A similar trend emerged in negative beliefs about OAMs, where cancer survivors in Profile 3 had significantly higher mean scores for Treatment consequences and Concerns. For positive perceptions, cancer survivors in Profile 1 perceived greater control over cancer than those who in Profile 2, followed by Profile 3. Interestingly, in terms of the perceived necessity of taking OAMs, patients in Profile 2, labeled as Conscious but hesitant (n = 99, 30.7%) had the lowest mean score compared the other two profiles.Table 2Scores on measures of health beliefs, outcome measures and psychosocial factors of three profiles.DomainTotal sample (n = 323)Profile 1: Optimistic and adaptive (n = 43)Profile 2: Conscious but hesitant (n = 99)Profile 3: Overwhelmed and unaware (n = 181)FProfile comparisonBeliefs about Cancer (M, SD)Cure10.50 (3.6)11.26 (3.6)9.85 (3.1)10.68 (3.8)2.85–Cancer Consequences**14.54 (3.6)8.42 (2.9)13.89 (2.7)16.35 (2.2)183.973 > 2 > 1Risk of Recurrence**13.85 (3.7)10.19 (3.9)12.95 (3.5)15.22 (2.9)47.683 > 2 > 1Emotional Representations**15.08 (4.2)8.72 (3.8)13.72 (2.9)17.34 (2.8)160.433 > 2 > 1Coherence**14.40 (3.1)16.02 (3.1)14.99 (2.8)13.70 (3.0)13.351 = 2 > 3Personal Control*12.92 (3.5)14.14 (4.4)12.96 (3.2)12.61 (3.3)3.451 > 2 = 3Beliefs about OAMs (M, SD)Treatment Control**13.64 (3.2)15.58 (3.3)14.23 (3.0)12.86 (3.0)16.611 > 2 > 3Necessity**20.22 (3.4)20.14 (3.5)18.92 (3.2)20.96 (3.2)12.673 = 1 > 2Treatment Consequences**13.92 (4.2)6.77 (2.2)11.85 (2.6)16.75 (1.9)418.863 > 2 > 1Concerns**16.90 (5.1)8.88 (2.9)14.72 (3.7)20.00 (3.1)232.823 > 2 > 1Outcome measure and psychosocial factors (M, SD)Medication nonadherence**18.91 (8.0)14.12 (2.7)17.23 (6.4)20.98 (9.0) < 0.013 > 2 = 1Health-related quality of life**17.51 (8.6)24.46 (7.8)20.15 (7.9)14.45 (7.8) < 0.011 > 2 > 3Health literacy**11.68 (2.9)13.77 (1.7)12.25 (2.8)10.87 (2.9) < 0.011 > 2 > 3M mean; SD standard deviation, F F-test; *p < 0.05, **p < 0.01; OAM oral anticancer medication. Table 2Scores on measures of health beliefs, outcome measures and psychosocial factors of three profiles.DomainTotal sample (n = 323)Profile 1: Optimistic and adaptive (n = 43)Profile 2: Conscious but hesitant (n = 99)Profile 3: Overwhelmed and unaware (n = 181)FProfile comparisonBeliefs about Cancer (M, SD)Cure10.50 (3.6)11.26 (3.6)9.85 (3.1)10.68 (3.8)2.85–Cancer Consequences**14.54 (3.6)8.42 (2.9)13.89 (2.7)16.35 (2.2)183.973 > 2 > 1Risk of Recurrence**13.85 (3.7)10.19 (3.9)12.95 (3.5)15.22 (2.9)47.683 > 2 > 1Emotional Representations**15.08 (4.2)8.72 (3.8)13.72 (2.9)17.34 (2.8)160.433 > 2 > 1Coherence**14.40 (3.1)16.02 (3.1)14.99 (2.8)13.70 (3.0)13.351 = 2 > 3Personal Control*12.92 (3.5)14.14 (4.4)12.96 (3.2)12.61 (3.3)3.451 > 2 = 3Beliefs about OAMs (M, SD)Treatment Control**13.64 (3.2)15.58 (3.3)14.23 (3.0)12.86 (3.0)16.611 > 2 > 3Necessity**20.22 (3.4)20.14 (3.5)18.92 (3.2)20.96 (3.2)12.673 = 1 > 2Treatment Consequences**13.92 (4.2)6.77 (2.2)11.85 (2.6)16.75 (1.9)418.863 > 2 > 1Concerns**16.90 (5.1)8.88 (2.9)14.72 (3.7)20.00 (3.1)232.823 > 2 > 1Outcome measure and psychosocial factors (M, SD)Medication nonadherence**18.91 (8.0)14.12 (2.7)17.23 (6.4)20.98 (9.0) < 0.013 > 2 = 1Health-related quality of life**17.51 (8.6)24.46 (7.8)20.15 (7.9)14.45 (7.8) < 0.011 > 2 > 3Health literacy**11.68 (2.9)13.77 (1.7)12.25 (2.8)10.87 (2.9) < 0.011 > 2 > 3M mean; SD standard deviation, F F-test; *p < 0.05, **p < 0.01; OAM oral anticancer medication. Scores on measures of health beliefs, outcome measures and psychosocial factors of three profiles. M mean; SD standard deviation, F F-test; *p < 0.05, **p < 0.01; OAM oral anticancer medication. Group differences in medication adherence, psychosocial and demographic characteristics Medication adherence and psychosocial characteristics differed significantly across the three profiles (Table 3). The results showed that Profile 3 had the lowest adherence (mean = 20.98, SD = 9.0), with statistically significant differences compared to the other two groups. Health-related quality of life (HRQoL) and health literacy were highest in Profile 1, followed by Profile 2 and Profile 3. Significant differences were also identified in gender, age and employment status. No significant differences were found in other characteristics.Table 3Multiple hierarchical linear regression of latent health belief profiles and medication nonadherence (n = 278).Standardized coefficients (β)RAdjusted R2F ChangeChange in R2Model 1Model 2Health belief profiles0.3350.10617.35**0.112Profile 1: Optimistic and adaptive(ref)(ref)Profile 2: Conscious but hesitant0.17* − 0.02Profile 3: Overwhelmed and unaware0.44**0.003Socio-demographics and clinical characteristics0.7140.49127.07**0.398Age − 0.18*Gender (ref: female) − 0.12*Employment (ref: unemployed)0.05Cancer recurrence (ref: no) − 0.11*OAM type0.06OAM duration − 0.04Comorbidity0.16**Health literacy − 0.43** Table 3Multiple hierarchical linear regression of latent health belief profiles and medication nonadherence (n = 278).Standardized coefficients (β)RAdjusted R2F ChangeChange in R2Model 1Model 2Health belief profiles0.3350.10617.35**0.112Profile 1: Optimistic and adaptive(ref)(ref)Profile 2: Conscious but hesitant0.17* − 0.02Profile 3: Overwhelmed and unaware0.44**0.003Socio-demographics and clinical characteristics0.7140.49127.07**0.398Age − 0.18*Gender (ref: female) − 0.12*Employment (ref: unemployed)0.05Cancer recurrence (ref: no) − 0.11*OAM type0.06OAM duration − 0.04Comorbidity0.16**Health literacy − 0.43** Multiple hierarchical linear regression of latent health belief profiles and medication nonadherence (n = 278). Multiple hierarchical liner regression of health belief profiles and medication nonadherence Multivariate analysis revealed that health belief profile assignment had a relationship with medication nonadherence and HRQoL (Table 4). The results of the regression analysis in Model 1 showed 10.6% of the variance (adjusted R2 = 0.106) in medication nonadherence was explained (F change = 17.35, p < 0.01). Adding covariates to the regression (Model 2) significantly improved the model’s ability to explain variations (adjusted R2 = 0.491, F change = 27.07, p < 0.01) in nonadherence, increasing the explained variance by 38.5%. The results of regression analysis in Model 3 explained 18.0% of the variance (adjusted R2 = 0.180) in HRQoL (F change = 31.42, p < 0.01). Adding covariates to the regression (Model 4) significantly improved the model’s ability to explain variations (adjusted R2 = 0.286, F change = 6.12, p < 0.01) in HRQoL.Table 4Multiple hierarchical linear regression of latent health belief profiles and health-related quality of life (n = 278).Standardized coefficients (β)RAdjusted R2F ChangeChange in R2Model 3Model 4Health belief profiles0.4310.18031.42**0.186Profile 1: Optimistic and adaptive0.38**0.23**Profile 2: Conscious but hesitant0.32**0.22**Profile 3: Overwhelmed and unaware(reference)(reference)Socio-demographics and clinical characteristics0.5590.2866.12**0.126Age0.21*Gender (ref: female) − 0.03Employment (ref: unemployed)0.07Cancer recurrence (ref: no)0.02OAM type − 0.06OAM duration0.15*Comorbidity − 0.06Health literacy0.20**p < 0.05, **p < 0.01. Table 4Multiple hierarchical linear regression of latent health belief profiles and health-related quality of life (n = 278).Standardized coefficients (β)RAdjusted R2F ChangeChange in R2Model 3Model 4Health belief profiles0.4310.18031.42**0.186Profile 1: Optimistic and adaptive0.38**0.23**Profile 2: Conscious but hesitant0.32**0.22**Profile 3: Overwhelmed and unaware(reference)(reference)Socio-demographics and clinical characteristics0.5590.2866.12**0.126Age0.21*Gender (ref: female) − 0.03Employment (ref: unemployed)0.07Cancer recurrence (ref: no)0.02OAM type − 0.06OAM duration0.15*Comorbidity − 0.06Health literacy0.20**p < 0.05, **p < 0.01. Multiple hierarchical linear regression of latent health belief profiles and health-related quality of life (n = 278). *p < 0.05, **p < 0.01. Subgroup analysis of multiple regression on health beliefs and medication nonadherence in profile 3 Given that Profile 3: Overwhelmed and unaware had the lowest adherence rates, a subgroup analysis was conducted to identify specific health beliefs and risk factors associated with nonadherence as well as HRQoL. Table 5 showed the standardized coefficients (β) for the association of health beliefs and the two outcome measures. Multivariate analysis showed that Risk of recurrence was significantly associated with medication nonadherence while Treatment consequences, Cancer consequences and Emotional representations were significant factors influencing HRQoL, when controlling for demographic variables (R = 0.72, adjusted R2 = 0.46). Additionally, patient socio-demographics, including age (β = −0.19, p < 0.05) and health literacy (β = −0.36, p < 0.01), and comorbidity (β = 0.18, p < 0.01) significantly associated medication nonadherence, suggesting that younger individuals with lower health literacy and more comorbidities were likely not to adhere to OAMs.Table 5Multiple hierarchical linear regression of health beliefs and medication nonadherence as well as HRQoL in Profile 3: Overwhelmed and unaware, lowest adherence and HRQoL group (n = 180).(n = 180)Medication adherenceHRQoLStandardized coefficients (β)SigStandardized coefficients (β)SigHealth belief domainsTreatment consequences − 0.010.520.15*0.05Risk of recurrence − 0.17*0.01 − 0.010.90Cancer consequences0.320.910.20*0.02Personal control0.200.36 − 0.010.95Treatment control − 0.080.33 − 0.080.42Coherence − 0.020.75 − 0.050.58Emotional representations − 0.010.890.16*0.04Necessity0.110.13 − 0.0020.99Concerns0.090.22 − 0.060.49CovariatesAge* − 0.190.02 − 0.24**0.01Gender − 0.120.07 − 0.060.46Cancer recurrence0.110.10 − 0.060.42OAMs duration0.010.91 − 0.18**0.01Comorbidity**0.18 < 0.010.030.64Health literacy** − 0.36 < 0.01 − 0.24**0.01HRQoL Health-related quality of life; Sig. statistically significant; *p < 0.05, **p < 0.01; OAM oral anticancer medication. Table 5Multiple hierarchical linear regression of health beliefs and medication nonadherence as well as HRQoL in Profile 3: Overwhelmed and unaware, lowest adherence and HRQoL group (n = 180).(n = 180)Medication adherenceHRQoLStandardized coefficients (β)SigStandardized coefficients (β)SigHealth belief domainsTreatment consequences − 0.010.520.15*0.05Risk of recurrence − 0.17*0.01 − 0.010.90Cancer consequences0.320.910.20*0.02Personal control0.200.36 − 0.010.95Treatment control − 0.080.33 − 0.080.42Coherence − 0.020.75 − 0.050.58Emotional representations − 0.010.890.16*0.04Necessity0.110.13 − 0.0020.99Concerns0.090.22 − 0.060.49CovariatesAge* − 0.190.02 − 0.24**0.01Gender − 0.120.07 − 0.060.46Cancer recurrence0.110.10 − 0.060.42OAMs duration0.010.91 − 0.18**0.01Comorbidity**0.18 < 0.010.030.64Health literacy** − 0.36 < 0.01 − 0.24**0.01HRQoL Health-related quality of life; Sig. statistically significant; *p < 0.05, **p < 0.01; OAM oral anticancer medication. Multiple hierarchical linear regression of health beliefs and medication nonadherence as well as HRQoL in Profile 3: Overwhelmed and unaware, lowest adherence and HRQoL group (n = 180). HRQoL Health-related quality of life; Sig. statistically significant; *p < 0.05, **p < 0.01; OAM oral anticancer medication. Discussion This study is one of the first studies that used latent profile analysis to identify subgroups of cancer survivors with distinct patterns of health beliefs, examining the associations with medication adherence and HRQoL. The results confirmed that health belief profiles are associated with adherence, highlighting the importance of personalized approaches in supporting medication adherence among cancer survivors. The three health belief profiles revealed clear differences in perceptions and treatment-related behaviors. Profile 1: Optimistic and adaptive had the strongest sense of personal control over cancer and treatment, reporting fewer concerns about medication, and demonstrating the highest adherence rates and HRQoL. Profile 2: Conscious but hesitant exhibited moderate health beliefs but expressed the lowest necessity beliefs, making them more hesitant about treatment. Profile 3: Overwhelmed and unaware, the largest subgroup, had the most negative health beliefs, experienced more consequences from cancer and OAM use, and exhibited the lowest adherence rates. Additionally, individuals in Profile 3 had lower health literacy and HRQoL compared to the other groups. ​Recognizing more diverse health belief profiles facilitates the identification of specific behavioral targets for interventions aimed at improving health outcomes among cancer survivors. This study identified three health belief profiles that were similar to the exploration of illness perceptions from studies of Ma. et al. and Valentine et al., with a mixed perception of illness presented among patients with cancer23,39. Uniquely compared to other chronic conditions40–43, cancer survivors with more positive perceptions were found to be less common while patients with more risky beliefs and physical and psychological symptoms are the most prevalent. This study also expands the understanding of how subgroups of cancer survivors with more threatening beliefs are more likely to be nonadherence to medication and experience a declining quality of life. Patients in Profile 3: Overwhelmed and unaware may perceive their condition as uncontrollable or unpredictable, leading to low motivation for adherence, reduced trust in treatment efficacy, and emotional distress that interferes with self-management. Additionally, greater concerns about treatment side effects, uncertainty about long-term benefits, and psychological burden may contribute to disengagement from prescribed regimens. Another important implication from the multivariate analyses is that clinical characteristics such as cancer recurrence and medication duration did not emerge as significant contributors to nonadherence, consistent with a previous study that showed psychosocial factors, such as health beliefs and health literacy may play a more critical role in adherence14. The identified three distinct belief profiles provide important implications of tailoring adherence interventions to patients’ needs. Patients with Overwhelmed and unaware beliefs, may benefit from education that enhances their understanding of their condition and addresses both the psychological and physical consequence of their disease10,20. Targeted interventions are especially needed to empower patients with self-management skills, such as side effects management, proactive discussions about cancer recurrence and prevention, and referral to mental health support, to help reduce emotional burden and improve adherence. Cancer survivors with Conscious but hesitant beliefs displayed more defined patterns, particularly with a better understanding of cancer. This finding may be attributed to our sample’s longer duration of cancer diagnosis and OAMs use, aligning with previous findings that cancer survivors develop a greater understanding over time. However, an unexpected finding was that this profile demonstrated the lowest perceived needs for treatment, which may indicate a unique challenge regarding medication adherence. These results are somewhat surprising, given that previous studies suggest that patients with a better understanding of their disease tend to perceive greater benefits from treatment, which should inherently lead to better adherence17,20. One possible explanation is that survivors in this group may feel stable in their survivorship or symptom-free, leading them to perceive less need for medications, especially if they are not experiencing immediate consequences from nonadherence. Further research is needed to explore the psychological and behavioral drivers of hesitancy in taking medication and its possible association with comorbidities within the context of the cancer survivorship journey. Limitations Although this study offers valuable insights into the health beliefs and medication adherence patterns of cancer survivors across the U.S., capturing diverse types of cancer and reaching participants nationwide, this study has several limitations that should be considered. The use of online survey panels may limit generalizability, as participants were predominantly younger, White, more educated, and insured, with fewer rural respondents represented44,45. As one of the first exploratory studies, future research should employ quota-based recruitment strategies to enhance racial ethnic and geographic diversity46. Additionally, while our study identifies associations between patient beliefs and nonadherence, we cannot definitively establish causality. A particular challenge is the likely presence of variables, such as cancer type, disease severity, and variability in treatment effects across different cancer types. Although the heterogeneity in our sample enhances the generalizability of our findings, it also limits our ability to make strong causal inferences, as unmeasured factors related to this heterogeneity may account for the observed associations. Future research with more targeted or controlled designs is needed to clarify causal pathways. The reliance on self-reported data introduces potential for social desirability and recall bias, and selection bias may have led to overrepresentation of more health-engaged individuals47. To mitigate this effect, we included normalizing statements before each section (e.g., “Many people find it difficult to adhere to their medications consistently”) to encourage honest responses. The cross-sectional design restricts interpretation to a single time point, limiting insights into how beliefs and adherence evolve over time. Lastly, the sample primarily included participants who were in the early stages of cancer or those with long-term OAM use, which may not reflect patients’ experiences of newly diagnosed or advanced stage. Conclusions This study identified three distinct health belief profiles among cancer survivors using oral anticancer medications, offering a person-centered approach to understanding variations in health beliefs and their impacts on medication adherence. These findings suggest that belief-based profiles can help guide the development of tailored interventions to support adherence. The study highlights the importance of addressing both cognitive and emotional aspects of survivors’ experiences. Future research should validate these profiles in diverse populations and examine how belief-based categorization can inform the design and implementation of personalized adherence support strategies. Supplementary Information Additional file1 (PDF 423 KB) Additional file1 (PDF 423 KB)