Introduction

Breast cancer (BC) is the most common cancer diagnosed among Canadian women and the second leading cause of death from cancer1. Age-specific BC incidence has increased over the years for all groups of women including those under 502. In Canada, most provinces and territories offer an aged-based population-level and public-funded screening program for women aged 50 to 74 every two years. Recently, some provinces have begun to allow asymptomatic women starting at age 40 to participate, primarily through self-referral3. Early detection through screening mammography remains the most effective method for reducing BC mortality by facilitating the prompt diagnosis of tumors at an early and treatable stage4. However, BC screening is also linked to potential harms, including false-positive and false-negative results, overdiagnosis, overtreatment, physical discomfort during the mammogram, and psychological distress5,6.

Furthermore, women have varying levels of risk for developing BC. Genetic, familial history, environmental, lifestyle, reproductive and hormonal factors have been associated with the risk of developing the disease7. Risk-stratified BC screening, in which personalized risk assessment based on multiple risk factors is used to tailor screening recommendations such as screening interval, starting age and test modalities, has emerged as an alternative for current aged-based programs4. Risk stratification has the potential to reduce harms for individuals at lower risk, enhance cancer detection for those at higher risk, and improve the cost-effectiveness of screening programs8. For example, individuals at higher risk may be offered screening at an earlier age and more frequently with access to more sensitive imaging modalities, while those at lower risk might undergo screening less often.

In Canada, the PERSPECTIVE I&I (Personalized Risk Assessment for Prevention and Early Detection of Breast Cancer: Integration and Implementation) study9 is evaluating the feasibility, accessibility, uptake to and cost-effectiveness of implementing risk-stratified BC screening into the healthcare systems10. However, before implementing such an approach, various economic, ethical, legal, social, and organizational factors must be carefully considered9,11. It would require the acceptability of both those delivering the intervention (e.g. healthcare professionals, stakeholders) and those receiving it (e.g. patients), which is now recognized as a critical factor in the design, evaluation, and implementation of healthcare interventions. Acceptability is a multi-faceted construct reflecting on the extent to which individuals consider an healthcare intervention to be appropriate based on anticipated or experienced cognitive and emotional responses12. Low acceptability of an intervention is more likely to lead to poor adherence, resulting in diminished benefits and less favorable clinical outcomes12. Additionally, understanding the factors that influence acceptability is crucial for enhancing the uptake of screening and reducing health inequalities related to social determinants13.

A few studies have examined whether personal characteristics (such as education level, prior screening participation, and personal or family history of cancer) are associated with different aspects of acceptability, including attitudes toward risk assessment and risk-stratified screening14,15,16,17. Although acceptability is a complex construct defined by multiple components, all studies examined the impact of personal and health characteristics in isolation using individual questions to explain each component. This approach did not account for potential mediating variables that could influence the relationship between predictors and outcomes. However, structural equation modeling (SEM), a multivariate statistical technique that estimates parameters for a system of simultaneous equations, can account for variables that are not directly measured (latent variables or constructs) by using indicators (measured variables). SEM also enables the estimation of both direct and indirect effects between explanatory variables through path analysis, as well as the relationships among latent variables through structural modeling18. SEM has been extensively used in psychology and social science and has been introduced more recently in medical research and epidemiology19.

Therefore, the aim of this study was (1) to develop and test a pathway model describing the relationship between personal characteristics and three latent concepts of acceptability i.e. attitudes and willingness towards BC risk assessment and risk-stratified screening, and comfort level with providing information for risk assessment and (2) to evaluate the direct and indirect effects of these characteristics on each latent outcome.

Methods

Study design and participants

We conducted an online, cross-sectional survey among women aged 30 to 69. The survey was distributed in the four largest provinces of Canada (Alberta, British Colombia, Ontario and Quebec) in French or English in February 2019. It was administered by Ipsos Canada, a local market research firm, to a panel of individuals from the general population who had previously consented to be contacted for research projects. The sample size estimation was based on the age-stratified distribution of women in the four t provinces, as reported in the 2018 Canadian Census, using a 95% confidence level and a 3% margin of error. Women’s opinions on BC risk assessment and screening were assessed based on hypothetical questions. Details regarding the development of the questionnaire have been reported elsewhere16. The study was approved by the Research Ethics Committees of the CHU de Québec-Université Laval and the McGill University (registration number: F9-42434). It was also performed in line with the principles of the Declaration of Helsinki and informed consent was obtained from all individual participants included in the study.

Outcomes

Attitudes towards BC risk assessment and risk-stratified screening

The first latent construct was the attitudes towards BC risk assessment and risk-stratified screening. It was measured using three questions representing observed variables (also referred to as manifest variables or indicators) of the construct: “(A1) What do you think of the idea of using information like age, cancers in your family, having children, lifestyle factors, breast density and weight to identify women who are at high, average or low risk of developing BC?; (A2) What do you think of the idea of using results from genetic testing (i.e. analysis that checks for features in your genetic makeup) to identify women who are at high, average or low risk of developing BC? and (A3) What do you think of the idea of changing how often women are invited for breast screening based on them being at high or low risk of developing BC?”. Respondents rated their attitudes using a 5-point Likert scale defined as: very bad idea, bad idea, neither a good or bad idea, good idea, very good idea, don’t know or prefer not to answer. To maintain ordinal data, we reclassified individuals who answered “don’t know” into the neither good or bad option and excluded individuals who preferred not to answer.

Comfort level with providing information for BC risk assessment

The second latent construct was the comfort level with providing information for BC risk assessment. It was also measured using three questions from the questionnaire: “(B1) How comfortable would you feel providing personal information (e.g. information regarding your lifestyle, personal and family medical history), so you can find out your BC risk level?; (B2) How comfortable would you feel providing a small sample of blood or saliva for genetic testing (i.e. analysis that checks for features in your genetic makeup), so you can find out your BC risk level? and (B3) How comfortable would you feel having a mammogram to assess your breast density (the amount of dense tissue compared to the amount of fatty tissue in the breast) so you can find out your BC risk level?”. Respondents rated their comfort level using a 5-point Likert scale defined as: very uncomfortable, uncomfortable, neither comfortable nor uncomfortable, comfortable, very uncomfortable and prefer not to answer. We again excluded individuals who preferred not to answer.

Willingness towards BC risk assessment and risk-stratified screening

The last latent construct was willingness towards BC risk assessment and risk-stratified screening. It was measured using four questions from the questionnaire: “(C1) Would you be willing to have your BC risk level assessed using information mentioned above? This would mean being told whether you are at average, lower than average or higher than average risk of developing BC; (C2) If your estimated level of BC risk was higher than average, would you be willing to have your BC screening more often than every 2 to 3 years?; (C3) If your estimated level of BC risk was average or lower than average, would you be willing to have your BC screening less often than every 2 to 3 years? and (C4) If your estimated level of BC risk was much lower than average, would you be willing not to be offered any breast screening?”. Respondents rated their willingness using a 4-point Likert scale defined as: no, definitely not; no, probably not; yes, probably; yes, definitely, don’t know and prefer not to answer. We created a 5-point Likert scale by reclassifying individuals who answered don’t know to into a neutral category and excluded individuals who preferred not to answer.

Explanatory variables

We also collected sociodemographic variables such as age, province of residence, country of birth, ethnicity, education level, marital status, employment status and annual family income. Other variables included perception of general health, perceived lifetime risk of developing BC, personal history of BC, previous experience of mammography, personal and familial history of genetic testing for BC. Nominal explanatory variables including ethnicity, previous experience of mammography, personal and familial history of genetic testing for BC were dichotomized into either white vs. others/don’t know for ethnicity and no/don’t know vs. yes, with the most frequent category being the reference. Other nominal variables were dummy coded and the most frequent category served as the reference group, except for the province variable, where Ontario was designated as the reference. For ordinal variables, the lowest category was used as the reference group. For all variables, we excluded individuals who preferred not to answer.

Statistical analysis

We conducted descriptive analyses on sociodemographic and other explanatory variables included in the study. Proportions were reported for categorical variables, and means with standard deviations (SD) were provided for continuous variables. For the SEM analysis, the first step was to specify the models to be estimated20. We created three distinct models using path diagrams that describe the hypothesized relationships among explanatory variables. We used the findings from our previous study16 to define the potential paths and their direction in the initial full model. We also conducted a literature review to verify our hypothesis and presented the full model to experts in the field for confirmation.

We then tested three measurement models, one for each latent construct, using confirmatory factor analysis (CFA). The relationship between the indicators and the underlying latent variables were tested against the data to confirm the factorial structure. Convergent validity which assesses whether the indicators measure the same latent construct was evaluated with the standardized factor loadings. Their value should be at least greater than 0.5 and ideally, at least 0.7. We also calculated the average variance extracted (AVE) that represents the average amount of variance captured by a latent variable from its indicators relative to the overall variance. The latent construct should explain no less than 50% of the indicators’ variance reflected by an AVE of  ≥ 0.521. Internal consistency reliability, which represents the extent to which all the items in a test measure the same concept or construct, was examined using Cronbach’s alpha22. A Cronbach’s alpha value usually ranges from 0 to 1 with a coefficient of at least 0.65 often considered as acceptable to assume that all items are interrelated with one another23.

SEM was then used to estimates the direct and indirect effects of all the explanatory variables on each latent variable. One of the principles in model selection and fit assessment in structural models is parsimony. A parsimonious model adequately explains or predicts a phenomenon with as few number of parameters as possible. That is, for two models with comparable fit, the model that has more degrees of freedom would be preferred24. Given our large set of variables and the exploratory context of the research, we selected the most relevant variables explaining each latent construct to be included in the final parsimonious SEM models with backward elimination. We began with three hypothetical full models that included all explanatory variables and specified all potential paths among them. For each model, we applied an iterative process to remove non-significant paths, eliminating the least significant path at each step (i.e., the one with the highest p-value). After each removal, the model was re-estimated until only statistically significant paths remained, using a significance threshold of p < 0.05. Overall models fit were assessed using the comparative fit index (CFI), the Tucker Lewis index (TLI) and the root mean square error of approximation (RMSEA). A value of CFI and TLI < 0.95 and RMSEA ≥ 0.06 indicated a good fit. The CFI and TLI are comparative indices comparing the specified model to the null model with values closest to 1 indicating better fit, while the RMSEA is a parsimonious measure that indicates the fit of the specified model to the population with possible values closest to 0 indicating a better fit20. The chi-square goodness of fit test and the weighted root mean square residual (WRMR) were not used to justify the fit of the data because their values are affected by sample size and not informative in large sample size25,26. Bootstrap was used to obtain the confidence intervals of direct, indirect and total effect of the final parsimonious model.

To account for the use of categorical dependant variables and indicators of a latent variable, we estimated the parameters using the weighted least squares mean and variance adjusted (WLSMV) robust estimator. The WLSMV estimator has been designed for ordinal variables and does not require the assumptions of multivariate normality and linearity of dependant variables and indicators to be respected27. Indeed, ordered categorical variables (e.g. Likert-type scale) are commonly used to assess latent constructs. When the number of categories is large (usually more than five) and are approximately normally distributed, categorical variables can be treated as continuous27. In our data, most ordinal indicators were left-skewed and treated as categorical variables. Parameters were estimated by probit regression coefficients for categorical variables and linear coefficients for continuous indicator using the theta method that fix residual means to 0 and residual variance to 1. Standardized coefficients were reported representing the variation in standard deviation of the dependant variable for one standard deviation increase of an explanatory variable. A larger beta coefficient indicates that a variable has a stronger influence on the probability of the dependant variable. Missing data were handled with pairwise deletion, which produces unbiased parameter estimates and standard errors under the assumption that the data are missing at random (MAR) or missing completely at random (MCAR)28.

Data cleaning, codification of variables and descriptive analysis were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA). CFA and SEM modeling were done using Mplus Version 8.

Results

The characteristics of the study population are presented in Table 1. The study population had an average age of 49.5 years and was predominantly white (80.6%) and Canadian-born (84.1%). Most women had a favorable socioeconomic status, with 35.4% holding a university diploma, 62.0% being married or in a common-law union, 59.2% working part- or full-time, and 32.8% reporting a total annual family income of at least $80,000. Less than 5% of the participants had a personal history of BC (4.1%), while approximately one in six women (16.2%) perceived their lifetime risk of BC as higher or much higher than that of others. Similarly, around the same proportion (18.8%) reported their current health status as fair or poor.

Table 1 Characteristics of the study population.
Full size table

Confirmatory factor analysis

The CFA models consisted of three latent variables and 10 manifest variables. Each latent construct was firstly tested separately to assess the adequacy of the measurement models.

Figure 1 shows the latent variables with the standardized factor loadings for each manifest variable. The standardized factor loadings for each latent variable were statistically significant. For panel A, two of the three factor loadings were greater than 0.7 and the AVE was equal to 0.54 supporting the convergent validity of the latent construct. For panel B, all factor loadings were greater than 0.5 and the AVE equal to 0.61 For these panels, the magnitude and significance of the factor loadings suggest that all indicators were moderately or strongly correlated with the latent variable and had acceptable convergent validity. For Panel C, two items were negatively correlated with the latent variable. Most factor loadings were smaller than 0.7 and the AVE of 0.37 was suggesting poor convergent validity.

Fig. 1
figure 1

Initial measurement models—CFA indicating standardized factor loadings and standard errors (in parenthesis). Letters A1 to A3, B1 to B3 and C1 to C4 represents the indicators from the questionnaire used to measure each latent variable.

Full size image

Fit of structural equation models

Model fit statistics were compared between the full a priori hypothesized models and the final parsimonious models. They are summarized for all models in Table S1. The final parsimonious model for attitudes had a good fit based on the RMSEA of 0.039, the CFI of 0.971 and TLI of 0.959. The second final parsimonious model for comfort level shown acceptable fit based on the RMSEA of 0.074, CFI of 0.935 and TLI of 0.903. The third model for willingness did not fit our data well. The initial full model had RSMEA, CFI and TLI values of 0.073, 0.737 and 0.653 respectively. When removing all non-significant paths from the model, the final parsimonious model showed an improvement in the CFI and TLI statistics going from 0.737 to 0.810 and 0.653 to 0.739. However, both values were still regarded as indicating a poor fit for the model.

Structural equation models

Three hypothetical full models were developed to illustrate the paths between all potential explanatory variables and each latent construct (Supplementary Figs. S1–S3). From these models, each non-significant path was removed one after the other to examine their effect on the models. Final parsimonious models are shown in Fig. 2. Direct, indirect and total effects were tested on each latent variables of the final parsimonious models and are presented in Tables 2, 3 and 4. Women’s perceived health and whether they had ever undergone genetic testing for BC were not found to be associated with attitudes, comfort, or willingness towards risk-stratified BC screening. The pathways involving mammography and perceived likelihood of having BC as mediating variables between age, personal history of BC, and each outcome remained consistent. Similarly, the pathways linking marital status, education level, or employment status to each outcome, with family income serving as a mediating variable, were also statistically significant across the different models.

Fig. 2
figure 2figure 2

Structural equation models for the relationship between explanatory variables and (A) attitudes towards breast cancer risk assessment and risk-stratified screening, (B) comfort level with providing information for breast cancer risk assessment and (C) willingness towards breast cancer risk assessment and risk-stratified screening. The values in the figure represent the standardized path coefficients. Thin lines are positive coefficients and thicker lines are negative coefficients. Only significant paths at a p-value < 0.05 are shown.

Full size image
Table 2 Total and indirect standardized effects from explanatory variables to attitudes towards BC risk assessment and risk-stratified screening.
Full size table
Table 3 Total and indirect standardized effects from explanatory variables to comfort level with providing information for BC risk assessment.
Full size table
Table 4 Total and indirect standardized effects from explanatory variables to willingness towards BC assessment and risk-stratified screening.
Full size table

For the Panel A, since the direct effect of age on attitudes was almost significant (p-value = 0.050), we kept this path in the final parsimonious model as age has often shown to be associated with the attitudes towards risk-stratified BC screening. In this model, living in Quebec compared to Ontario (β =  − 0.061, p = 0.003), belonging to an ethnic minority (β =  − 0.117, p < 0.001), an older age (β =  − 0.071, p = 0.050), and having a history of BC (β =  − 0.047, p = 0.025) were all associated with a less positive attitude towards BC risk assessment and risk-stratified screening. Table 2 presents the indirect standardized effect of statistically significant explanatory variables on attitudes. All specific indirect paths were statistically significant indicating that the effect of variables such as marital and employment status were completely mediated by the income.

For Panel B, belonging to an ethnic minority (β =  − 0.104, p < 0.001), an older age (β =  − 0.269, p < 0.001) and a personal history of BC (β =  − 0.056, p = 0.012) were associated with a lower comfort level with providing information for BC risk assessment. Table 3 presents the indirect standardized effect of statistically significant explanatory variables on comfort level. All specific indirect paths were statistically significant, therefore part of the effects of exogenous variables were mediated through the endogenous variables (mammography, perceived chances of BC and income). For Panel C, belonging to an ethnic minority (β =  − 0.142, p < 0.001) and an older age (β =  − 0.295, p < 0.001) were associated with less willingness towards BC risk assessment and risk-stratified screening. Table 4 presents the indirect standardized effect of statistically significant explanatory variables on willingness. All specific indirect paths were statistically significant.

Discussion

In this study, we developed and tested three hypothetical pathway models describing the relationship between Canadian women’s sociodemographic and self-reported health characteristics and three latent variables including attitudes and willingness towards BC risk assessment and risk-stratified screening and comfort level with providing information for BC risk assessment. Based on the theoretical framework of acceptability of healthcare interventions, attitudes towards BC risk assessment and risk-stratified screening which tends to measure the extent to which the intervention is perceived as likely to achieve its purpose and comfort level which refers to how an individual feels about an intervention represent different component constructs of acceptability. Those constructs could be predictors of the willingness to participate in an intervention which is defined as the amount of effort an individual is planning to exert12,29. These constructs applied together will provide a clearer evaluation of an intervention acceptability that is crucial for individuals’ engagement.

Similarly to other behavioral theories such as the Health Belief Model30, our main assumption was that other factors such as socioeconomic factors, age and ethnicity modified the acceptability of BC risk assessment and risk-stratified screening. When building our hypothetical full models, one rationale was that each explanatory variable had a direct effect on each outcome. In our hypothesized models, personal history of mammography, perceived lifetime risk of BC, and family income served as mediating variables between other sociodemographic and health characteristics and each latent variable. Age was also a key determinant of multiple variables such perceived risk of BC, history of mammography and perceived general health as highlighted by previous evidence31,32,33. For example, older women perceived they have a higher risk of BC which can influence their mammography uptake.

Therefore, we assessed the direct and indirect effects of sociodemographic and self-reported health factors on each latent outcome. All indirect pathways through these mediating variables were statistically significant. The most influential factor associated with a positive attitude, greater comfort, and higher willingness was the personal history of mammography. This finding suggests that women who have participated in current BC screening programs are more likely to adapt to changes within these programs. On the other hand, the strongest factor associated with a negative attitude, lower comfort, and reduced willingness was belonging to an ethnic minority group. Several studies have already highlighted the lower participation of ethnic minorities in BC screening programs in Canada34,35,36,37. Therefore, implementing a novel approach such as risk-stratified screening would require targeted strategies to engage ethnic minorities and prevent exacerbating existing healthcare disparities.

Some studies have quantitatively explored different concepts of acceptability towards risk-stratified BC screening among women such as attitudes and preferences regarding risk-based screening modalities as well as comfort level and willingness to provide personal information for risk assessment14,15,16,17,38,39,40. Ghanouni et al. (2020) conducted a cross-sectional survey in England on women’s attitudes towards risk-stratified BC screening39. In their study, they asked if women would be willing (or unwilling) to have their risk of BC estimated using multifactorial algorithms and to have (i) more frequent screening, (ii) less frequent screening or (iii) no screening based on their risk status. Loft et al. (2024) also explored women’s attitudes towards risk-stratified BC screening in Danmark17. The authors used several questions including what women’s thoughts were about (i) knowing their personal risk in order to be offered a mammography more or less often, (ii) being offered a mammography every year if they are at high risk and (iii) being offered a mammography every four years if they are at low risk of developing BC. Therefore, we based the selection of indicators on available research and develop alternative models to assess the model impact on our findings. Our research still confirms the findings of other studies about the association of women’s personal characteristics and their views on risk stratified screening.

Furthermore, while all the other studies examined each indicator individually, SEM analyses allowed to combine indicators referring to the same concept. As attitudes, comfort level and willingness towards new health interventions are complex constructs that cannot be entirely explained by a single indicator, the use of latent constructs provides a more comprehensive measure of these concepts. Our study also hypothesizes a pathway to illustrate the relationships between the personal characteristics among themselves and with the outcomes. Certain characteristics, such as ethnicity, whether blood relatives have undergone genetic testing, and age, consistently had a direct impact on attitudes, comfort level, and willingness. The effect of other variables was mostly indirect through crucial variables. For example, the indirect effect of marital status, employment and education were always significant through income but their direct effects varied depending on the outcome. Indeed, those factors are often used to determine an individual socioeconomic status which is a multidimensional construct in itself41.

Currently, several international studies including MyPeBS in Europe42, WISDOM in the United States43 and PROCAS in the United Kingdom44 are evaluating the impact of risk-stratified BC screening programs showing the growing interest of this approach by many healthcare systems. In the Canadian context, the feasibility of implementing this healthcare intervention is being evaluated at multiple levels, including the efficacy of the PRS45, cost46, feasibility10 and acceptability for both health professionals and beneficiaries16,47,48. In this regard, our study provides strong evidence that will help guide the implementation of this approach. While women’s acceptability of risk-based screening varies based on various factors, our study has identified a limited number of factors that can be targeted to improve women’s uptake of screening. These actions could involve targeted public health messaging, improved invitation methods, and expanded educational activities to help women make informed decisions.

This study also presents some limitations. Since no theoretical models currently exist to explain the factors influencing attitudes, comfort level, and willingness toward risk-stratified BC screening, we developed our hypothetical models based on a literature review and expert opinion. Therefore, it is possible that some paths were missing or not clearly established by previous evidence. Moreover, alternative models or other configurations could have fit the data better or equally well. To address this, we tested alternative models for each outcome. Additionally, we used a data-driven backward selection approach to determine the final parsimonious models. While our findings are mostly well-suited to our data, they may not be generalizable to other populations. To minimize the impact of using a data-driven variable selection method, we employed a bootstrap technique with 500 replications, which allows for resampling a single dataset to generate multiple simulated samples. We addressed missing data using pairwise deletion, acknowledging that this approach is not the most robust when working with latent constructs. Pairwise deletion can lead to inflated standard errors due to varying sample sizes across regression analyses. As a sensitivity analysis, we also applied listwise deletion (i.e., complete case analysis), which is another common method for handling missing data. Given the small proportion of missing values, the estimates obtained using listwise deletion were comparable to those from the primary analysis.

Our results also suggested that the indicators selected for two of the latent variables, attitudes and comfort level, represented the concepts well. For those two outcomes, the Cronbach alpha value indicated an acceptable internal consistency and the models were adequately fitting the data. However, for willingness to have BC risk assessment and risk-stratified screening the indicators showed a small agreement between them. A high Cronbach alpha value indicates that responses for each participant are consistent based on a set of questions. If participants give a high response (i.e., a high Likert score) to one item, they are more likely to provide high responses to other items within the same concept. In our questionnaire, two items related to willingness asked a positive question and showed an acceptable agreement (α = 0.67) and two items with negative questions showed poor agreement (α = 0.57). This discrepancy suggests that willingness may represent a multidimensional construct—i.e., a single overarching concept measured through multiple, potentially distinct subcomponents rather than directly observed indicators49. It is therefore possible that the positively and negatively framed items captured different aspects of willingness and may have been more appropriately analyzed as separate constructs. Nonetheless, CFA usually requires a minimum of three indicators when specifying a model to avoid identification problems50 making it difficult to create two separate latent variables with our set of indicators. The results of this panel should be interpreted taking these limitations into account.

In addition, our study included women who voluntarily completed the questionnaire, which may have introduced potential selection bias. However, when comparing the characteristics of our sample to the data of the Canadian census, we observed that it was comparable in terms of age distribution and education level. Still, the representativeness of our sample may be limited to Canadian women with similar demographic profiles and vary depending on healthcare access. As many epidemiological studies, we were limited with cross-sectional data and cannot interpret the findings as definitive mediation or causation. Despite these limitations, this was an exploratory study that allowed to generate and test our hypothesis regarding factors that could influence attitudes, comfort level and willingness towards risk-stratified BC screening and to explore the use of SEM in epidemiological studies.

This study also has several notable strengths. It was conducted with one of the largest samples to evaluate women’s acceptability of risk-stratified screening, which enabled the use of more advanced methods to explore the relationships between explanatory variables and attitudes, comfort level, and willingness toward this approach. To our knowledge, no other studies have developed hypothetical models of various concepts of acceptability related to risk-stratified BC screening and examined the direct and indirect effects of sociodemographic characteristics and self-reported health variables on these concepts. Thus, the evidence generated from this study could support the implementation of risk-based stratified screening in Canada and other countries.

Conclusion

This population-based survey identified several key factors that may explain Canadian women’s attitudes, comfort level, and willingness toward risk-stratified BC screening. Previous participation in screening programs was strongly associated with a positive attitude, higher comfort level, and greater willingness toward this approach, highlighting the importance of efforts to increase screening participation in current programs. Additionally, the study emphasizes the need to develop strategies that will reach specific subgroups, including individuals with lower socioeconomic status, ethnic minorities, and older women, when implementing risk-stratified screening to prevent exacerbating health disparities.