1. Introduction
Although research linking health literacy and health disparities is emerging [
1,
2], there is consensus that low health literacy leads to poorer health outcomes [
3]. Health literacy is often included as a social determinant of health because of the interrelationships between education level, health literacy and health outcomes [
4,
5,
6]. Specifically, as with low educational levels, low health literacy is associated with poorer health status, lower treatment compliance, increased emergency rooms visits and decreased ability to understand instructions and participate in decision-making [
7,
8]. Intervention studies that target low health literacy have sought to decrease health disparities on a number of varied behavioral outcomes, such as increasing healthcare service access or utilization, improving patient self-management skills or implementing disease-specific self-management plans. Other health disparities interventions that target health literacy have focused on increasing knowledge, self-efficacy, health communication and quality of life as well as reducing healthcare costs [
2,
9].
According to the National Academy of Sciences Roundtable on Health Literacy “it is the responsibility of health professionals to make sure all their patients (including those with low health literacy) truly comprehend the information they are being given [
10].” Accordingly, one of the objectives of the 2020 Healthy People Initiative is to improve the health literacy of the population as a means to eliminate healthcare disparities (Objective HC/HIT-1) [
11]. This objective is included as a key issue in the health and healthcare domain [
7] and it is measured through three indicators that focus on increasing the number of people who report that their healthcare providers always (1) gave them easy-to-understand instructions, (2) asked them to explain the instructions given and (3) helped them in filling out forms. However, addressing low health literacy requires a more holistic approach that includes the complex interactions between patients, healthcare providers and healthcare systems.
It has been well documented that poor health literacy is an important barrier to cancer screening adherence [
12,
13,
14,
15]. Although studies have identified the impact of social determinants of health, such as cultural beliefs, health literacy and language on disparities in cancer screening rates among minority populations [
16,
17,
18], there is substantial evidence that health literacy is a complex, multidimensional construct that cannot be measured using generic measures that do not address different domains related to management of specific chronic diseases such as cancer [
19,
20,
21,
22,
23,
24].
Several authors have conducted systematic reviews [
25] on health literacy and cancer screening and contributed to the development of conceptual models that identify various domains of health literacy. Zarcadoolas and colleagues developed a multidimensional definition and model of health literacy that integrates and acknowledges the important roles of four types of literacy: fundamental literacy (ability to read, write, speak and work with numbers); science literacy (knowledge and abilities to understand scientific concepts including the rapid change of technology and uncertainty of science results); civic literacy (awareness of public issues, and skills needed to evaluate different positions and make decisions); and cultural literacy (abilities to recognize and understand social identity and collective beliefs and customs). The model is recommended to analyze health communication, improve interventions in health-related communications and develop assessment tools that allow the creation of profiles of people’s health literacy [
26]. Sørensen and colleagues conducted a systematic review of definitions and conceptual frameworks of health literacy and identified four competencies related to health literacy that are required to navigate the healthcare system: Access (ability to seek, find and obtain health information); understand (ability to comprehend the health information that is accessed); appraise (ability to interpret, select and evaluate the health information that has been accessed); and apply (ability to use the information to make a decision) [
27]. They applied the four competencies to build a 12-dimensional model of the impact of health literacy across three domains of the healthcare services continuum (promotion, prevention and treatment). Sørensen’s model distinguishes between distal (demographic situation, culture, language, political forces, etc.) and proximal (personal characteristics, social support, personal influences, media use, etc.) factors.
In a systematic review, Berkman and colleagues developed an analytic framework that delineates the relationships between health literacy skills, interventions and outcomes [
2]. Building on the Integrative Theory, they describe a core set of variables (e.g., attitudes, health status, social norms, patient-provider relationships and self-efficacy) that explain individual’s behavioral intention (e.g., taking medication, changing lifestyle or having screening tests), and that combined with the adequate skills (e.g., knowledge, cognitive abilities, information seeking and decision-making) and removal of barriers (e.g., access to health insurance and language services,), predicts behavior change. Although they found mixed evidence supporting their model in studies of health literacy and cancer screenings, they concluded that there was moderate evidence that lower literacy is associated with decreased utilization of Pap smear screening for cervical cancer and mammography for breast cancer, and weak evidence for colon cancer screening. In support of the model, a systematic review of attitudes toward prostate cancer among African American men found that individual (knowledge, patient-provider communication, perception of personal risk and personal/family history of cancer), cultural (threat to masculine identity, fear of cancer, mistrust of the healthcare system and religious fatalism) and social (access to preventive care, income and education) factors influenced their decision to have prostate cancer screening [
28].
Although these models are addressing multi dimensions related to health literacy, Zarcadoolas and Sørensen models do not focus on cancer literacy; Berkman’s analytic framework focuses only on interventions and outcomes instead of factors influencing cancer screening; and Pederson’s systematic review is narrowed to factors associated with prostate cancer screening among African American men. Building on these models and reviews, our approach is to go beyond people’s health literacy skills and traditional narrow conceptualizations and measurement of health literacy (reading, oral and numeracy) to investigate cancer health literacy specifically, and include other factors (motivation, self-efficacy, empowerment, socio-environmental influences) that might contribute to cancer screening disparities [
29].
While no framework was found in the literature related to cancer literacy among diverse populations, in this study, we aimed to develop and test a new multidimensional framework of the effects of cancer health literacy on cancer prevention and screening behaviors among African Americans, English-speaking Latinos, Spanish-speaking Latinos and non-Latino whites that is comprehensive and includes cultural attitudes, beliefs and practices, as well as language and health literacy factors. The objective of this study is to identify the underlying structure and subscales of the Multidimensional Cancer Literacy Questionnaire (MCLQ) and test the preliminary validity of the multidimensional framework. The MCLQ focuses on predisposing factors that influence potential cancer screening mediators and outcomes. In this analysis, we focused on testing the structure and subscales of the predisposing factors only. Our rationale for focusing on this first and major portion of the framework is to apply data reduction techniques before linking these predisposing factors to potential cancer screening mediators and outcomes.
2. Materials and Methods
Based on the literature, in this study, we first, developed and described our conceptual framework, the Multidimensional Cancer Literacy Framework. Then, based on the Framework, we developed the Multidimensional Cancer Literacy Questionnaire (MCLQ). Then, we conducted a field test of the MCLQ through a self-administrated cross-sectional survey of diverse populations residing in New Orleans. Finally, we used data collected from the field test on factors associated with multiple domains related to cancer health literacy to examine the preliminary validity and internal consistency reliability of the measures and refine the Multidimensional Cancer Literacy Framework.
2.1. Development of Multidimensional Cancer Literacy Framework (MCLF)
The development of our conceptual framework, the Multidimensional Cancer Literacy Framework was informed by prior work, the Health Belief Model [
30] and Dr. Zarcadoolas’ definition of health literacy [
26]. The Health Belief Model (HBM), an individual-level framework, offers several constructs related to perceived barriers, benefits and risks that are useful and relevant for predicting cancer screening behaviors. According to the HBM model, the likelihood that individuals engage in cancer screening behaviors is influenced by their beliefs about cancer and screening risks (perceived susceptibility) and the severity of the disease and the possible harm and benefits of the screening tests (perceived seriousness).
In this study, we also apply Dr. Zarcadoolas’ broad definition of health literacy as “the wide range of skills and competencies that people develop over their lifetimes to seek out, comprehend, evaluate, and use health information and concepts to make informed choices, reduce health risks, and increase quality of life” [
26] (p. 55). This definition was used to elucidate the specific skills, competencies and use of health information related to cancer literacy that could affect cancer screening rates. The multidimensional framework (
Figure 1) expands on this prior work to include cultural factors and issues such as English language skills and trust in physicians that are especially relevant to minority communities.
Drawing from the literature review [
31,
32,
33,
34,
35,
36], items for the MCLQ were organized into three inter-related subdomains that fall under the domain of Predisposing Factors: The Facilitators Domain, the Barriers Domain and the Cultural Domain (
Figure 1). The Facilitators Domain includes six factors related to enablers supporting or facilitating cancer screening such as motivation to screen, access to information, English and communication skills, and trust in physicians and preferences about healthcare providers. The Barriers Domain includes three factors related to barriers or obstacles to screening including perceived barriers and symptomatic and sociocultural deterrents [
31]. The Cultural Domain includes six factors to measure participants’ beliefs about cancer as a disease, cancer treatment and cancer screening as a preventive measure; as well as participants’ beliefs influencing the way they make decisions about cancer (self-determination) and their perceptions about own risk and worriedness about having cancer.
The Outcomes Factors include items that measure participants’ screening behaviors for the more common cancers for both women and men. Additionally, the framework includes items to measure specific demographic characteristics that may influence decisions about cancer screening such as race/ethnicity, age, gender, education and type of health insurance (moderators). Factors in the Knowledge Domain include potential mediators of cancer screening outcomes including items to measure cancer literacy level and knowledge and understanding of cancer risks and early detection screening methods. These items act as mediators of the relationship between the predisposing factors and cancer screening behavior outcomes [
2].
2.2. Development of Multidimensional Cancer Literacy Questionnaire (MCLQ)
A literature review of tools measuring cancer health literacy and related constructs resulted in the identification of main domains related to cancer health literacy (
Table 1). Selected subscales and/or individual questions from these tools were adapted, as needed. Questions were organized into the preliminary domains that make up our theoretical multidimensional framework (
Figure 1) and focus on participants’ knowledge and perceptions about: cancer (causes, types, risks and symptoms); prevention (screenings and healthy behaviors); and treatments (options and access); as well as their cancer screening behaviors. Special attention was given to the operational definition of variables and constructs, the types of questions included in the questionnaire and response options. An Expert Panel was created to revise the MCLQ and determine the applicability of scales, subscales or individual items drawn from the literature to diverse populations.
The MCLQ was developed in Spanish and English and pilot-tested using cognitive interviews with 20 volunteers from each of the Latino, African American and White communities in Louisiana. Participants were recruited by members of the Community Advisory Boards (CABs) who have had a long-term relationship with the Principal Investigator (PI) in each targeted community. These interviews were conducted by the PI during community meetings specifically scheduled for this purpose. During the cognitive interviews, participants self-administered the MCLQ, and then were asked to provide feedback about the questionnaire (objective, consistency, clarity, length, applicability, etc.), the wording, response options and organization of the questions and the appropriateness of the incentive offered ($30). The MCLQ was revised to maximize comprehension based on the results of the cognitive interviews. The revised tool was reviewed again by the Expert Panel for final approval prior to the field test.
2.3. Participants and Data Collection Procedure for Field Test
This project employed community based participatory research (CBPR) methods and was approved by the Xavier University of Louisiana’s Institutional Review Board (IRB). Inclusion criteria for the field test were: ages 25 years old or older and living in Louisiana. Recruitment was stratified to obtain similar numbers by race (Latinos, African American and Whites) and gender (male/female). Efforts were made to include participants of varying educational levels (primary school or less; some high school, high school degree; some college studies; or a bachelor’s or more advanced diploma).
Flyers inviting participants to the study and explaining the purpose of the study, dates and locations of data collection events, time commitment, incentives and contact information were delivered by CAB members and other community leaders to businesses, organizations and centers serving the different communities.
At the meetings, the Principal Investigator (PI) explained the rationale and objectives of the study and used a script to obtain verbal consent of participants to complete the one-time survey. Participants meeting the inclusion criteria and willing to participate in the study were enrolled in the study. A copy of the script was given to each participant to keep. In order to maintain anonymity, questionnaires were numbered with a consecutive number and did not include any personal identifiers. Questionnaires were hand delivered by the PI and community outreach leaders, who were attentive to answer any questions participants had as they self-administered the questionnaire. Completion of the questionnaire took on average 30 min. Survey responses were entered into an Excel file by community members hired as part of the study and the Community Outreach Coordinators reviewed data entry against original questionnaires to check for accuracy.
2.4. Statistical Analysis to Refine the MCLQ
Factor analysis is one of the most commonly used techniques of data reduction in social science where a large set of observed variables is reduced to a smaller set of hypothetical or latent variables (factors) [
37]. There are two different methods for factor analysis. EFA is a tool intended to help generate a new theory (theory-building) and estimate the unknown structure of the data when there is no prior theory about the factor structure of the data. On the other hand, confirmatory factor analysis (CFA) is used to test an existing theory (theory-testing), that is, to examine if an a priori model of the underlying structure of the constructs (constrained and unconstrained) fits the new data adequately [
37,
38].
Considering that the MCLQ is a new tool, exploratory factor analysis (EFA) with principal components and Varimax rotation was used as a data reduction tool to identify the underlying structure and subscales of the MCLQ and discover predominant factors (constructs) explaining each of the domains included in the multidimensional framework (
Figure 1). Recommended sample size when conducting principal component analysis depends on the number of items in the questionnaire (participant-to-items ratio). Many authors recommend a 10:1 ratio as the rule of thumb [
39]. The original MCLQ included 99 items (
Table 1), thus, the targeted sample size for this study was designed to yield an adequate participant-to-items ratio suitable for application of factor analysis techniques to explore the underlying structure of the MCLQ. We used the following criteria to determine the validity of the resulting constructs (scales): (1) total variance explained; (2) eigenvalue > 1; (3) eliminating items with low structure coefficient (loads < 0.5); (4) eliminating items that showed high cross loading (>0.4), that is, loaded significantly on more than 1 factor (discriminant validity); and (5) eliminating items with low extraction communality (h2 < 0.40) [
38]. Additionally, Cronbach alpha coefficients were used to assess the internal consistency reliability of the scales. During the reliability analysis, the following requirements also were verified: (1) whether the item-total correlation corrected for overlap (item discrimination) in new developed scales was > 0.30 [
40]; (2) whether the elimination of an item caused the alpha to increase; (3) whether a reduced range of responses was observed in an item; and, (4) whether item means were extreme. SPSS, version 13.0.1 (SPSSInc, Chicago, IL) and R were used to carry out the data analyses.
Additionally, considering that we have a large sample (
n = 1500), and as a way to test the preliminary validity of the results obtained, we conducted a secondary analysis running EFA on the first half of the data (750 randomly selected individuals) and CFA, using Diagonally Weighted Least Squares (DWLS), on the second half of the data [
41].
4. Discussion
The objective of this study was to field test the MCLQ and identify the underlying structure of the data, and whether this structure is consistent with our conceptual framework for diverse populations (
Figure 1). Results confirmed three main domains (Facilitators, Barriers and Culture) and the informed revision of the framework (
Figure 2,
Table 3). Key findings are discussed next.
The F3-
Intention to Screen construct was initially considered to be a part of the cultural domain because the items referred to beliefs regarding cancer prevention and screening that were thought to be culturally-based (
Figure 1). Upon closer inspection, the items included in this factor (
Table 4) are more related to subjective norms that function as “facilitators” for the adoption of cancer screening: “Routine cancer screening shows that people care for their health and their families”; “All people should have regular cancer screenings”; “Cancer screenings help to save lives”; and “Friends of my age have cancer screenings regularly.” These findings are consistent with those from another study that found that perceived subjective norms (“it is important for me to comply with what my close friend believes”, “it is important for me to do what my parents think is appropriate” and “the important people in my life believe colon screening can help prevent colon cancer”) predicted colon cancer screening [
43].
In our theoretical framework (
Figure 1), we expected that items in F6-
Lack Awareness and F7-
Personal Discomfort would group together considering that they all were related to personal issues that would be barriers to regular cancer screening. However, instead, they grouped into two different factors. Items in F6 focus on issues that are not under the control of the person while items in F7 focus more on personal concerns that are managed at the individual level. Similarly, we expected that items in F8-
Impediments to Screen and F9-
Lack of Resources would also group together on the same factor, considering that they relate to real barriers that participants have had to regular screening. However, items related to barriers to get the screening appointment grouped with F8 while lack of resources separated into an independent factor, F9. Interestingly, in the secondary analysis conducted using CFA, these two factors grouped together as initially expected.
Additionally, we expected that items related to English skills (F10) were not going to stay together as a strong factor. Our assumption was that these items were important only for Latino immigrant participants who may have poor English skills and needed interpreter services. However, during the analysis items related to the need (barrier) and use (facilitator) of interpreters were deleted because they were cross loaded with this factor. Poor English skills may not be only a concern for Latino immigrants. A report about adult literacy in the U.S. found that “U.S.-born adults make up two-thirds of adults with low levels of English literacy skills” [
44] [p. 2], and the American College Testing (ACT), one of the major admission tests for college in U.S., found that only 59% of the high school class of 2019 reached the minimum college readiness benchmark in English [
45].
Initially, we included items in F12-
Low Locus of Control as cultural aspects influencing individual’s capacity to make decisions (C4-Self-determination,
Table 1). However, considering that they separated themselves into an independent factor and when looking at the statements (“I would offend my doctor if I were to make my own decisions about my health”, and “I don’t know enough to make my own medical decisions”), it makes sense that they are included in the barriers domain instead (
Table 5).
During the literature review, it was not clear about the different impact that perceived cancer risk (F18) and worriedness about cancer (F19) would have on cancer screening (
Table 6). While some authors choose one construct over the other, we decided to keep both in order to be able to select the one that best matches the model. Interestingly, both concepts grouped as independent and strong factors. These results support that patients’ perceptions about risk are different than cancer worry. However, items related to risk and worry of men-related cancer (prostate) and women-related cancers (breast and cervical) had to be deleted during the analysis because they were not applicable to the entire sample. According to Klein and Stefanek [
46], individual decisions such as avoiding known risk factors, utilizing screening tests, undergoing cancer testing and making treatment decisions are “tied inextricably to comprehension and perceptions of personal risk.” (p. 147) Specifically, they argue that people’s perception of risk or worriedness of getting the disease is influenced by: “the frequency (e.g., how many people smoke); the covariation (e.g., how many smokers get lung cancer); the similarity (e.g., how many of my smoking friends have gotten lung cancer); and normativeness (e.g., how unusual is a smoker to die from something else than lung cancer)” [p. 151]. Interestingly, when looking at the perception of risk, people usually estimate their risk to be lower than average. However, when looking at the worriedness about having the disease, Klein and colleague explain that people usually focus more on the 8 people who died because of the disease than on the 92 who did not, and their personal behaviors (e.g., If I smoke, how much I smoke, what kind of cigarettes I smoke, etc.) [
46]. In summary, worriedness is a function of both how people view their own risk and how they compare their risk with that of others [
47].
When conducting factor analysis, it is recommended that factors with two items be interpreted with caution, especially when the variables have a low correlation with each other (r < 0.70) [
38,
39]. In our case, four factors (F6-
Lack Awareness, F12-
Low Locus of Control, F13-
Stigmas about Cancer and F15-
Beliefs about Treatment) had only 2 items each (
Table 5 and
Table 6). Although these items contribute little to the total variance explained by the model (
Table 3), all of them, except F12, have factor loadings of over 0.5, meeting criteria for inclusion in the revised framework. Although these factors were not confirmed in the CFA analysis, considering that our objective is to examine the preliminary validity of the conceptual framework and the importance of these factors in individuals screening behaviors, instead of deleting the items in these four factors, we decided to keep them for further testing.
Important limitations of this study need to be considered. As this study is a preliminary exploration of a new developed conceptual framework and its respective questionnaire, results need to be interpreted with caution. In addition, three of the subscales, F12-Low Locus of Control, F15-Beliefs about Treatment and F17-Self-Determination demonstrated marginal internal consistency reliability (Cronbach alphas). Based on the conceptual importance of these constructs with respect to cancer screening, especially among diverse populations, we recommend that researchers continue to develop and test the conceptual and psychometric adequacy of these measures. Although the tool is lengthy (82 items), the subscales could be used independently; however, caution is recommended when using those scales for which validity was not confirmed in the secondary analysis of the data. We suggest that researchers continue to refine and test the subscales and items of the MCLQ to improve their psychometric properties. This future work could include known-groups validity testing to help determine if MCLQ scores discriminate among subgroups of diverse populations known to differ on some of these constructs.