Women’s preferences for testing to predict breast cancer risk – a discrete choice experiment

Abstract

Background

Risk-based breast cancer screening offers a more targeted and potentially cost-effective approach in cancer detection compared to age-based screening. This study aims to understand women’s preferences and willingness for undergoing risk assessment tests.

Methods

A discrete choice experiment (DCE) was conducted. Six attributes were selected to construct the DCE questionnaire: one-time cost of the test, methods for reducing late-stage breast cancer, annual breast cancer screening expenses, insurance coverage for early-stage breast cancer, family risk correlation, and risk communication methods. Women aged between 21 and 59 were recruited from Singapore. Latent class analysis was performed.

Results

Three hundred twenty-eight women were included in the analysis and classified into two classes: test supporters and non-supporters. Both classes prioritised test costs and screening costs. Among non-cost attributes, the potential to reduce late-stage breast cancer diagnosis was the most influential factor. Insurance coverage increased willingness to undergo testing. Risk communication methods were not significant in influencing the decision of undergoing tests. Non-supporters were less inclined to take the test if family risk correlation was high. Younger women, married women, full-time employees, and those with a history of breast disease were more likely to be supporters. Women with a family history of breast cancer were more likely to be non-supporters.

Conclusions

Financial incentives play a notable role in increasing the uptake of risk-prediction tests. However, the programme’s success depends on understanding and addressing the diverse preferences of women. While cost considerations ranked highly, additional strategies are needed to engage groups that are hesitant, particularly those with a high family risk correlation.

Background

Breast cancer is the second most common cancer worldwide in 2022 and imposes a significant disease burden in women [1]. An estimated 685,000 women died from breast cancer in 2020, making up around 16% of all women’s cancer-related deaths. Furthermore, the burden of breast cancer is on the rise, with an anticipated one million deaths annually by 2040, attributed to aging and growing populations [2]. Breast cancer screening can lower the mortality from breast cancer by facilitating early treatment [3]. From 2010 to 2021, 11 countries issued 23 guidelines on breast cancer screening, and those guidelines predominantly recommend mammographic screening as the primary method for detecting breast cancer in average-risk women aged 40 to 74 [4]. While breast cancer screening offers significant benefits, it also entails potential harms such as overdiagnosis and false-positive results which increases physical and psychological burden on women and waste healthcare resources [5, 6].

Due to individual differences, the risk of developing breast cancer varies among women [7]. Risk factors related to breast cancer include age, breast density, family history, previous biopsies, and genetic information [8]. For women with average risk of developing breast cancer, age is the primary factor used for screening recommendations [9]. However, focusing purely on age to establish a breast cancer screening strategy does not accurately identify those who need screening [10]. To better balance the benefits and drawbacks of screening, a personalized breast cancer screening strategy can be developed based on individual risk factors [11]. Depending on the risk levels, screening methods, starting and ending age for screening, and screening frequencies could vary [12]. For example, Trentham-Dietz et al. (2016) combined age and breast density to explore different screening frequency and showed that women benefit more from customized screening [13]. A study conducted by Evans DG et al. (2016) demonstrated that women can be stratified into different risk categories for personalized screening using breast density and genetic risk variations [14]. Recent advancements in breast cancer risk assessment highlight the substantial benefits of utilizing polygenic risk scores (PRS) for predictive purposes. PRS is calculated from genetic testing of single nucleotide polymorphisms linked to breast cancer, accounting for over 30% of the genetic predisposition [15]. Existing evidence indicates that PRS can increase cancer screening’s effectiveness and might be included in future cancer screening guidelines [7, 16,17,18,19,20,21].

While risk-based breast cancer screening is promising, it is unclear whether women are willing to understand their risk. Most lay people have a limited understanding of risk information, often perceiving it as an advanced abstract concept rather than mathematical probabilities. They typically view risk information as quantitative descriptions and abstract notions, closely associating it with unhealthy lifestyles and genetic predispositions [22, 23]. They also interpret risk information as indicative of a high likelihood of developing diseases, which can lead to significant psychological distress. This perception may contribute to the reluctance of many individuals to undergo risk prediction testing [24, 25]. A study indicated that women typically overestimate their probability of having breast cancer [26]. To address women’s misconceptions about risk, it is essential to implement effective health education strategies. Selecting communication methods that align with preferences, such as face-to-face, telephone, or online, can significantly enhance the effectiveness of health education [27]. Employing these preferred methods can improve women’s comprehension of risk information and facilitate risk assessment.

Discrete Choice Experiment (DCE) is a widely utilized stated preference method in healthcare and public health to understand people’s preferences. DCE involves presenting participants with various scenario options and requesting them to select their most preferred option. This method effectively captures respondents’ preferences, enabling the adaptation of the interventions to better align with the needs of the target population and deliver greater benefits [28]. One recent systematic review examining DCEs on preferences for genetic testing to predict risk of developing hereditary cancer found that test effectiveness and detection rate were consistently important [29]. Another systematic review examining DCEs on preferences for genetic and genomic risk-informed chronic disease screening highlighted the importance of survival, test accuracy, and screening impacts [30]. Wong et al. (2018) explored women’s preference for taking single-nucleotide polymorphism gene testing to guide personalized breast cancer screening [31]. However, most existing studies focused on the characteristics of the test and its administration, with insufficient attention to other implementation factors, such as subsequent disease prevention strategies and communication methods.

In this study, we conducted a DCE to deepen the understanding of women’s preference for testing their risk of developing breast cancer. We focused on implementation-related attributes and context factors that affect women’s decision. Furthermore, we explored preference heterogeneity and factors explaining preference heterogeneity.

Methods

DCE questionnaire design

A literature review was conducted first to identify relevant attributes and levels for constructing DCE questions. The attributes and levels were subsequently presented to a group of researchers and programme administrators experienced in breast cancer screening. During the consultation, the study team members explained the study and presented the attributes and levels. Two ranking exercises were conducted. Participants ranked the attributes based on their relevance in affecting women’s decision to take tests to predict the breast cancer risk and then on their importance from an academic perspective.

Six attributes were selected to construct the DCE questions: (1) the out-of-pocket cost of a one-time test, (2) methods to reduce late-stage breast cancer diagnoses, (3) annual out-of-pocket cost for breast cancer screening, (4) insurance coverage for early-stage cancer, (5) family risk correlation, and (6) risk communication methods. Family risk correlation was defined as the likelihood that first-degree relatives are at high risk if a woman is found to have a high risk for breast cancer. Table 1 lists the attributes and levels.

Table 1 Attributes, levels, and Sample question

Sawtooth software (Advanced Teaching Suite) was used to design the DCE questionnaire. An unlabeled two-stage design was used. In the first stage, two situations were presented to the women. The women were required to choose the situation that they are more willing to do a predictive disease test to understand their risk of getting breast cancer. In the second stage, the women need to state whether they really want to do the test in real life under their selected situation in the first stage. A dummy question was included with one option clearly dominates the other option. Each woman answered ten DCE questions. Ten different versions of DCE questionnaires were generated and randomly assigned to the study participants. A sample DCE question is presented in Table 1.

Additional information collected included age, race, housing type, education, employment status, marital status, personal history of breast disease, family history of the disease, acquaintances with breast cancer, self-perceived risk of developing breast cancer, and concern about breast cancer relative to other diseases. The questionnaire is presented in Appendix A.

Sample, sample size, and data collection

We targeted women in Singapore aged 21 to 59. We stratified the sample into three strata based on age, i.e.: 21 to 39 years old, 40 to 49 years old, and 50 to 59 years old. A sample size of 100 is required for each stratum, which requires a total sample size of 300 [32, 33]. We then conducted a simulation exercise in Sawtooth to confirm the sample size is appropriate for the required statistical efficiency. Accounting for 20% incomplete and invalid responses, we targeted a sample size of 400 in the survey.

An external vendor was engaged for data collection. The online survey was conducted between November and December 2023. A pilot test was conducted on the first 30 responders. As a quality check, only surveys that were completed between four to fifteen minutes were accepted. There were no changes required for the DCE questions. The rotation design (randomized incomplete block design) was used to ensure all versions of the questionnaire were given in the different age strata.

Demographic variable coding

Age was classified into three categories: 21 to 39 years old, 40 to 49 years old, and 50 to 59 years old. Ethnicity was categorized as Chinese or non-Chinese. Education levels were grouped into three categories: secondary school and below, above secondary school and up to diploma equivalent, and university or higher. Employment status was divided into two categories: full-time employed and otherwise. Marital status was categorized as married or not married. Housing type was used to proxy for income level, with three categories: private housing (i.e.: condominium), government housing from housing and development board (HDB) 4 to 5 rooms, and HDB 1 to 3 rooms, reflecting decreasing expected income level.

For breast disease-related information, having a personal history of breast disease, a family history of breast cancer, knowing people with breast cancer among friends and networks, and being concerned with breast cancer were coded as dummy variables. The self-reported risk of developing breast cancer was categorized as below average, average, and above average.

Statistical analysis

Women with missing information or who failed the dummy question were excluded from the analysis. Latent class analysis was used to identify subgroups within a population based on patterns of responses [34, 35]. The latent-class analysis assumes that study participants can be assigned to unobserved (latent) classes based on their preference patterns. Each class has preference weights that are identical within the class but systematically different from other classes. Bayesian information criterion was used to determine the optimal number of classes. Dummy coding was used in the analysis, with the reference attribute levels set to a preference weight of zero. Intercepts were included to account for left-and-right bias. Demographic and breast disease-related information were used to predict class membership, which reflects the probability of a woman having a particular pattern of preferences.

Conditional relative importance (CRI) for an attribute is defined as the difference between the highest and lowest preference weights for that attribute. A higher CRI signifies that the attribute plays a more significant role in influencing women’s decisions to take the test. We used attribute-based normalization, adjusting the sum of CRI values for all attributes within a class to 100%. Uptake probabilities were then calculated for selected hypothetical scenarios.

Results

A total of 395 responses were collected. After excluding women with incomplete responses and those who failed the dummy question, the analyses included responses from 328 women. Table 2 describes the characteristics of the participants included in the analysis. The number of participants with missing data for each variable of interest is presented in Appendix B.

Table 2 Summary statistics

Table 3 shows the results of the latent class analysis. Women were classified into two groups: “supporters” and “non-supporters” for the test. This classification was based on the coefficient, or preference weight, of the “None option”, where a more negative value indicates a higher willingness to take the test, and vice versa. In total, 65% of the women were classified as supporters, and 35% as non-supporters.

Table 3 Regression results for DCE

In contrast, for the remaining coefficients, positive values indicate that women were more willing to take the test, while negative values indicate they were less willing. The coefficients of reference levels are zero. Between the two methods to reduce late-stage breast cancer, supporters (coefficient = -0.591 [95%-CI: -0.782 to -0.400]) and non-supporters (coefficient = -0.549 [95%-CI: -0.791 to -0.306]) were both less inclined to take the test if surgery was the option, compared to mammography. When insurance coverage for early-stage cancer was assessed as an attribute, insured women were more willing to take a test (supporters: coefficient = 0.427 [95%-CI: 0.315 to 0.540]; non-supporters: coefficient = 0.323 [95%-CI: 0.179 to 0.466]). We did not find significant differences in the methods of communicating risk in affecting the decision in either group to take the test.

A notable difference between supporters and non-supporters was the impact of risk correlation between themselves and first-degree relatives. If the risk correlation is more than 50%, i.e.: if a woman is found to be high risk, the probability of her first-degree relative being high risk is more than 50%, non-supporters were less likely to take the test (coefficient = -0.231 [95%-CI: -0.422 to -0.040]). However, this factor did not influence the decision of supporters (coefficient = -0.102 [95%-CI: -0.250 to 0.046]).

The relative importance of the attributes is presented in Fig. 1. The test cost and cost of the annual prevention strategy were the two most important attributes in women’s decisions to take the test. The remaining four attributes, in order of importance, were the cancer prevention method, whether they had cancer insurance, risk correlation, and risk communication methods. Cost had a greater influence on the decisions of non-supporters compared to supporters.

Fig. 1

Conditional relative importance

The uptake probability under three hypothetical scenarios is presented in Table 4. Scenario 1 represents the situation where non-supporters were most likely to take the test, while Scenario 3 represents the situation where they were least likely to take it. Scenario 2 is a realistic scenario. For non-supporters, under the scenario that the cost of the one-time test and cost of the annual screening are $499; surgery to remove breast is used as the prevention method; insurance coverage for early-stage breast cancer is not available; family risk correlation is high; and the risk information is conveyed through breast cancer specific risk report, the uptake rate is 2%. The uptake rate increases to 58% under the scenario that the cost of the one-time test and cost of the annual screening are $0; other imaging method is used for prevention; insurance coverage for early-stage breast cancer is available; family risk correlation is low; and the risk information is conveyed through medical professional using a breast cancer specific risk report. For supporters, the uptake rates are 81% and 98% under the two scenarios.

Table 4 Uptake probability for selected scenarios

Figure 2 illustrates the factors influencing whether women were supporters or non-supporters of the test. Younger women, married women, and full-time employees were more likely to be supporters. Women with a history of breast disease, those who knew people with breast cancer, or those who expressed higher concern for breast cancer compared to other diseases were also more likely to be supporters. Chinese women were more likely to be non-supporters compared to non-Chinese women. Women with higher social economic status, indicated by higher education and living in better housing (i.e. condominiums), were more likely to be non-supporters. Women with a family history of breast cancer were more likely to be non-supporters.

Fig. 2

Impact of demographic factors on being supporters for the test. Notes HDB Housing and Development Board; ITE Institute of Technical Education; Poly Polytechnic; sec secondary school; BC Breast cancer

Discussion

In this study, we examined Singaporean women’s preferences for tests that assess their risk of developing breast cancer. Among the non-cost attributes, methods that reduce the chances of being diagnosed with late-stage breast cancer were the most influential in shaping women’s decisions. There were no statistically significant differences in preferences for mammography, other imaging methods and medication. Insurance coverage for early-stage breast cancer, arguably a cost-related attribute, ranked just after late-stage cancer prevention methods in importance. Family risk correlation affected the decision of only non-supporters, while risk communication methods did not significantly influence women’s decisions.

The cost of the test was a significant factor influencing women’s decisions to undergo the test. For supporters, the one-time out-of-pocket cost of the test and the annual screening cost are similarly weighted. In contrast, non-supporters are more sensitive to the one-time test cost than the annual screening cost. This heightened sensitivity suggests that the initial expense of the test is a significant barrier for them, more so than the recurring annual costs. Therefore, financial incentives or subsidies should focus on reducing the one-time cost of the test. By subsidizing the one-time test, it is likely to address the primary concern for non-supporters and could effectively encourage them to participate. In contrast, subsidizing annual screenings might not have the same impact, as it does not address the initial financial barrier that non-supporters find more daunting. This targeted approach could lead to higher test uptake rates among those who are less inclined to participate.

Among the non-cost attributes, methods to reduce the risk of being diagnosed with late-stage breast cancer were the most influential in shaping women’s decisions regarding testing. This influence was primarily due to their aversion to preventive surgery, reflecting a significant discomfort with or negative view of surgical interventions. In contrast, preferences for mammography, other imaging methods, and medication showed no statistically significant differences, indicating that these factors did not significantly affect women’s decisions about undergoing the test. The potential use of chemoprevention (e.g. low-dose tamoxifen), along with more frequent screenings or alternative imaging techniques, could offer alternatives to address women’s aversion to preventive surgery while still effectively reducing the risk of late-stage breast cancer [36,37,38].

Our results on the traits of supporters and non-supporters highlighted several unexpected findings. Women with a personal history of breast disease, a family history of breast cancer, or who know someone diagnosed with breast cancer are generally expected to be more concerned about their risk and more likely to participate in screening programs [39]. Indeed, our original hypothesis was that a high family risk correlation would encourage women to take the test, but the results showed the opposite. We found that while women with past experiences of breast disease or acquaintances with breast cancer were more likely to be supporters, those with a family history of breast cancer were more likely to be non-supporters. Women with personal experiences of breast disease or who know others with breast cancer might be more curious and thus be more inclined to support taking the test. In contrast, those with a family history might perceive the risk differently (they might feel that they already know they are at high risk) or might feel overwhelmed by the familial history, leading to lower support for additional testing [40].

Women employed full-time were more likely to be supporters for the test than those who were unemployed or retired. This could be due to their financial independence, as well as the potential larger impact on their financial status if they develop late-stage cancer. Typically, individuals with higher socioeconomic status are expected to be more proactive about their health and more likely to participate in preventive health measures [39, 41, 42]. However, while full-time employment (as a proxy for financial independence) was associated with greater support for taking tests in our study, higher education and better housing (as a proxy for higher socioeconomic status) are associated with less support, which is unexpected based on typical assumptions about socioeconomic status and health behavior [43]. This contradiction might be explained by several factors. Women with higher socioeconomic status might perceive less need for additional tests due to better access to healthcare and trust in their existing healthcare routines. They could be more sceptical about new tests, concerned about overdiagnosis, and weigh the cost-benefit ratio more critically [44]. Furthermore, cultural and societal norms might influence health behavior, and in many contexts, certain ethnic groups are expected to have higher participation rates in preventive health measures [39, 45, 46]. Our finding that Chinese women are more likely to be non-supporters go against what we have previously observed [39].

Despite the literature indicating that young women generally perceive a low susceptibility to developing breast cancer, lack confidence in proper breast self-checking behaviors, and are disengaged with breast awareness, our results found that younger women were more likely to be supporters of the test [47, 48]. This discrepancy could be due to younger women’s greater acceptance of new technologies and a desire to proactively manage their health risks.

We initially aimed to determine the preferred method for communicating test results. It is surprising that the method of result communication does not significantly impact the recipient’s decision to take the test. However, risk communication method is still important as it could affect women’s perception about risk and their decisions in the subsequent disease management. Public preferences for program features, including communication methods, typically affect the acceptability and effectiveness of screening programs [30]. While communication methods might not significantly influence decisions, factors such as perceived risk, program accuracy, and survival benefits may be more impactful [30, 49,50,51].

Our study advances the field of cancer screening by examining women’s preferences for breast cancer risk assessment tests, highlighting the importance of considering individual preferences. However, our study has several limitations. Firstly, the nature of DCE limits the number of attributes that can be included. Other important factors could include the risk classification accuracy, the need for re-testing, and potential health benefits from taking the test. We selected attributes based on a ranking exercise with domain experts and programme administrators and found that five out of the six attributes influenced women’s preferences for taking the test. Secondly, this study focused specifically on women’s preferences for breast cancer risk prediction tests, so the conclusions may not apply to men or other types of cancer. Third, while Singapore is a high-income country with multi-ethnic population, we need to be mindful when generalising results to other similar settings. Preferences are affected by cultural and social norms. Studies considering local context should be conducted to understand preferences of local women. Nevertheless, our study can guide the development of similar studies in local context where risk-based cancer screening is under consideration.

Despite these limitations, this study provides valuable initial insights into patients’ perspectives in Singapore and underscores the importance of incorporating local viewpoints in healthcare and public health policy development. Future studies could consider recruiting a larger sample, gathering more comprehensive data, conducting in-depth qualitative interviews, and employing alternative data analysis methods. Due to the restricted participant pool, the findings may not fully represent the diversity of the Singaporean population, particularly across different ethnic, socioeconomic, and educational backgrounds. Future studies could benefit from broader recruitment strategies to better reflect Singapore’s multicultural demographic. Additionally, while the questionnaire aimed to capture key aspects of participants’ opinions, it may not have fully accounted for varying levels of healthcare literacy and understanding of medical processes, which could have influenced some responses. Tailoring future questionnaires to assess participants’ baseline knowledge and aligning them with the local healthcare context could enhance the reliability of the findings. Moreover, qualitative interviews could be employed to delve deeper into women’s decision-making processes and to explore discrepancies between our study results and the initial hypotheses.

Alternative statistical and data analysis methods could be employed to further explore important patterns in the data. Our study primarily focused on main effects, without delving into potential interaction effects, such as those between different attributes or demographic factors. With a larger sample size, future study could apply less restrictive methods, such as random coefficient model or graph analysis [52,53,54], to investigate and establish additional key patterns in women’s preferences for understanding their breast cancer risk.

Conclusion

Our study identified key factors affecting women’s willingness to undergo breast cancer risk assessment. The study offers insights for designing effective risk-based breast cancer screening programs.