Using the Multi-Theory Model (MTM) of Health Behavior Change to Explain the Seeking of Stool-Based Tests for Colorectal Cancer Screening

Colorectal cancer is the third most common cancer worldwide and is the second leading cause of cancer-associated deaths. While colorectal cancer is on the decline in the United States (US), disparities still exist, despite the non-invasive screening modalities, such as stool-based tests have shown themselves to be effective in the detection of colorectal cancer. Many of the existing stool-based test interventions lack the use of a contemporary theory-based approach. Given the paucity of theory-based interventions intended to promote stool-based tests, this cross-sectional study utilizes the multi-theory model (MTM) of health behavior change to explain the seeking of stool-based tests for colorectal cancer (CRC) screening. An online 57-item questionnaire with an established psychometric validity was used to collect responses from the US-based sample (n = 640) of adults aged 45–75 years old. The data were analyzed using bivariate and multivariate statistical methods. Structural equation modeling (SEM) was conducted to test the construct validity of the survey instrument. In this nationwide sample, 39.2% (n = 251) of participants reported having received some form of a stool-based test. Among the participants who did not undergo stool-based CRC screening, the MTM subscales, including “participatory dialogue”, “behavioral confidence”, and “changes in the social environment”, were significant predictors of initiating screening behavior and explained 48% of the variance in the initiation among this group (R2 = 0.579, F = 5.916, p < 0.001; adjusted R2 = 0.481). The MTM may be a useful framework with which to design educational, mass media, social media, and clinical interventions for the promotion of stool-based CRC screening among adults aged 45–75 years old.


Introduction
Colorectal cancer (CRC) is among the most common cancers and sources of cancerassociated mortality worldwide [1]. Globally, an estimated 1.9 million cases of CRC were diagnosed, leading to 900,000 deaths in 2020 [1]. In 2023, approximately 153,000 individuals This was a cross-sectional study which utilized a sample of US adults aged 45-75 years who understand English language. The sample selection process has been described in Figure 1. The study samples were obtained from currently available pools of research participants, who have consented to be contacted for future studies. To avoid heavy dependence on a single segment of the population, Qualtrics pooled samples from different sources across the nation. Recruitment of respondents was conducted in an enforced quota sampling fashion to create a pool of participants representing the current US population demography. quota sampling fashion to create a pool of participants representing the current US population demography.

Sampling and Data Collection
Data for this study were collected from January to March 2023 via a web-based structured survey built in Qualtrics hosting platform (Provo, UT, USA). The Qualtrics Research Marketing team collected the data as a part of the contractual agreement set up between researchers and the Qualtrics team. The survey link was distributed through multiple avenues, including in-app notifications, listserv, among the contacts of the investigators, etc.

Ethical Considerations
The Institutional Ethics Committee of the University of Nevada, Las Vegas, granted the exempt status (protocol ID: UNLV-2022-575) on 23 December 2022. All ethical guidelines in accordance with the Declaration of Helsinki were followed. There were no personal identifiers collected to ensure the anonymity of the participants. Participants were provided with detailed information about the purpose, objectives, procedure, benefits, and risks associated with this study, which could help them make informed decisions to participate in the study. Participation in this study was completely voluntary and participants could drop out any time as they wish. To prevent multiple responses from the same participant (duplicate responses), strategies such as digital fingerprinting and the "prevent multiple responses option in Qualtrics" were used. Screening questions that were aligned with the inclusion criteria were posed at the beginning of the survey to avoid any self-selection bias. For participants who completed the survey, incentives in the form of gift cards, cash rewards, SkyMiles, or redeemable points were given per the contract between Qualtrics and panel providers.

Sampling and Data Collection
Data for this study were collected from January to March 2023 via a web-based structured survey built in Qualtrics hosting platform (Provo, UT, USA). The Qualtrics Research Marketing team collected the data as a part of the contractual agreement set up between researchers and the Qualtrics team. The survey link was distributed through multiple avenues, including in-app notifications, listserv, among the contacts of the investigators, etc.

Ethical Considerations
The Institutional Ethics Committee of the University of Nevada, Las Vegas, granted the exempt status (protocol ID: UNLV-2022-575) on 23 December 2022. All ethical guidelines in accordance with the Declaration of Helsinki were followed. There were no personal identifiers collected to ensure the anonymity of the participants. Participants were provided with detailed information about the purpose, objectives, procedure, benefits, and risks associated with this study, which could help them make informed decisions to participate in the study. Participation in this study was completely voluntary and participants could drop out any time as they wish. To prevent multiple responses from the same participant (duplicate responses), strategies such as digital fingerprinting and the "prevent multiple responses option in Qualtrics" were used. Screening questions that were aligned with the inclusion criteria were posed at the beginning of the survey to avoid any self-selection bias. For participants who completed the survey, incentives in the form of gift cards, cash rewards, SkyMiles, or redeemable points were given per the contract between Qualtrics and panel providers.

Questionaire
This 57-item survey tool (see Supplementary Materials) used in this study was grounded in the MTM framework [18,22]. Originally, there are two main components of the MTM, namely, "initiation", and "sustenance". However, for this study, only the "initiation" component was conceptualized. In this tool, "initiation" has four constructs, namely, "participatory dialogue", "behavioral confidence", "changes in the physical environment", and "changes in the social environment". Please see Figure 2 for a detailed overview of the MTM framework. The first construct of the "participatory dialogue" is derived from the "perceived advantages", and "perceived disadvantages" (mathematically: "participatory dialogue" score = "perceived advantages" score minus (−) "perceived disadvantages" score). There were 5 items each for the "perceived advantages" and "perceived disadvantages", which were measured on a 5-point Likert scale with a possible range of 0-20 units. Another construct was the "behavioral confidence" that was measured on the surety scale with a possible range of 0-20 units too. This construct measures confidence to initiate a certain behavior. The third construct was the "changes in the physical environment", which points to the tangible resources required to initiate a certain kind of behavior. This was measured through 3 items on a 5-point Likert scale. The last construct of "changes in the social environment" was based on the premise of social influence on initiating a behavior, which was measured on the 5-point Likert scale with 3 items.

Questionaire
This 57-item survey tool (see Supplementary Materials) used in this study was grounded in the MTM framework [18,22]. Originally, there are two main components of the MTM, namely, "initiation", and "sustenance". However, for this study, only the "initiation" component was conceptualized. In this tool, "initiation" has four constructs, namely, "participatory dialogue", "behavioral confidence", "changes in the physical environment", and "changes in the social environment". Please see Figure 2 for a detailed overview of the MTM framework. The first construct of the "participatory dialogue" is derived from the "perceived advantages", and "perceived disadvantages" (mathematically: "participatory dialogue" score = "perceived advantages" score minus (−) "perceived disadvantages" score). There were 5 items each for the "perceived advantages" and "perceived disadvantages", which were measured on a 5-point Likert scale with a possible range of 0-20 units. Another construct was the "behavioral confidence" that was measured on the surety scale with a possible range of 0-20 units too. This construct measures confidence to initiate a certain behavior. The third construct was the "changes in the physical environment", which points to the tangible resources required to initiate a certain kind of behavior. This was measured through 3 items on a 5-point Likert scale. The last construct of "changes in the social environment" was based on the premise of social influence on initiating a behavior, which was measured on the 5-point Likert scale with 3 items.

Survey Validation
The survey underwent face and content validation by six subject matter experts (of which five were university faculty with multiple expert areas) in the fields of behavioral theories (n = 3), colorectal cancer (n = 2), target population experts (n = 2), and survey validation (n = 4). Since the experts had multiple expertise, the numbers listed in the parentheses above may not sum to a total of six. The initial version included a total of 54 items. Experts provided comments to establish the face and content validity of the instrument and improve the readability/clarity of the survey, and 3 additional items were included for the construct of "changes in the social environment" to the survey. The rest of the changes entailed some wordsmithing in order to improve the readability for the target population. There was consensus among experts after three rounds of reviews to finalize a 57-item instrument for the current study. All steps of survey validation can be seen in Figure 3.

Survey Validation
The survey underwent face and content validation by six subject matter experts (of which five were university faculty with multiple expert areas) in the fields of behavioral theories (n = 3), colorectal cancer (n = 2), target population experts (n = 2), and survey validation (n = 4). Since the experts had multiple expertise, the numbers listed in the parentheses above may not sum to a total of six. The initial version included a total of 54 items. Experts provided comments to establish the face and content validity of the instrument and improve the readability/clarity of the survey, and 3 additional items were included for the construct of "changes in the social environment" to the survey. The rest of the changes entailed some wordsmithing in order to improve the readability for the target population. There was consensus among experts after three rounds of reviews to finalize a 57-item instrument for the current study. All steps of survey validation can be seen in Figure 3.

Construct Validity
Structural equation modeling (SEM) was performed by using Mplus to examine the construct validation of the measurement model. The hypothesized model included advantages, disadvantages, behavioral confidence, changes in the physical environment, and changes in the social environment as independent variables, and initiation of health behavior change as the dependent variable. The chi-square (x 2 ), comparative fit index (CFI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR) indices were reported to indicate the goodness of fit [23]. We interpreted CFI and TLI > 0.90 [24,25] and RMSEA and SRMR < 0.08 [26] as evidence of acceptable model fit.

Testing of Assumptions
Assumptions of independence of observations (i.e., Durbin-Watson statistic), linearity (e.g., scatterplot and partial regression plots), homoscedasticity, multicollinearity, and normality (i.e., P-P plot and Q-Q plot) were evaluated in this study before selecting the model. Homogeneity of variance assumption was assessed by the Levene's test. There was independence of residuals, as assessed via a Durbin-Watson statistic of nearly 2. There was homoscedasticity, as assessed via visual inspection of a plot of studentized residuals versus unstandardized predicted values. There was no evidence of multicollinearity, as assessed via tolerance values greater than 0.1. There were no studentized deleted residuals greater than ±3 standard deviations, no leverage values greater than 0.2, and values for Cook's distance above 1. The assumption of normality was met, as assessed by a P-P Plot.

Data Analysis
Univariate, bivariate, and multivariate statistical tests were used to analyze the data. Frequencies/proportions were used to represent categorical variables, whereas continuous variables were represented as the means and standard deviations given the normal distribution. For initiation, two separate models of hierarchical multiple regression analyses were used among those participants who underwent stool-based colorectal screening and those who had not undergone these screening tests. The selection of variables to be entered in the regression was assessed based on a theoretical, conceptual, and statistical basis. All tests were performed using IBM SPSS (version 28.0), with the level of significance set at 5%.
In calculating the sample size, we used the following parameters: confidence level: 95; margin of error: 5%; and population proportion: 33%. The population proportion of (100 − 67) = 33% was based on the data reported by the Centers for Disease Control and Prevention in 2018 [27], according to which 67% of US adults aged 50-75 years met the recommendations of colorectal cancer screening. The minimal sample required was 340, and after applying a 20% non-response rate, a total of n = 340 + 68 = 408 was needed. Our sample size is sufficiently larger than estimated, which allowed us to conduct structural equation modeling too [28].

Construct Validity
The fit indices suggested an acceptable fit of the initiation model (x 2 [195] = 767.145 (p < 0.001), CFI = 0.923, TLI = 0.908, RMSEA = 0.07, and SRMR = 0.054). First, we examined the measurement model and observed an overall pattern of statistically significant factor loadings of five latent variables: advantages; disadvantages; behavioral confidence; changes in the physical environment; and changes in the social environment. Specifically, the variable of advantages had large effects (e.g., β ranging from 0.62 to 0.91) on its indicators; the variable of disadvantages had moderate-to-large effects (e.g., β ranging from 0.44 to 0.92) on its indicators; and the variable of behavioral confidence had large effects (e.g., β ranging from 0.76 to 0.90) on its indicators as well. The variables of changes in the physical environment and changes in the social environment each had large effects on their respective indicators, with βs above 0.79 in Figure 4. These effects indicated that the scale provided a valid measurement of intended constructs. Second, we examined the construct correlations and standardized regression coefficients and found that advantages, behavioral confidence, and changes in the social environment were positively associated with (e.g., β ranging from 0.19 to 0.43, p < 0.001), and disadvantages were negatively associated with (β = −0.20, p < 0.001) the initiation of colorectal cancer screening behavior. However, the effects of changes in the physical environment on initiation were statistically insignificant. physical environment and changes in the social environment each had large effects on their respective indicators, with βs above 0.79 in Figure 4. These effects indicated that the scale provided a valid measurement of intended constructs. Second, we examined the construct correlations and standardized regression coefficients and found that advantages, behavioral confidence, and changes in the social environment were positively associated with (e.g., β ranging from 0.19 to 0.43, p < 0.001), and disadvantages were negatively associated with (β = −0.20, p < 0.001) the initiation of colorectal cancer screening behavior. However, the effects of changes in the physical environment on initiation were statistically insignificant.

Characteristics of the Sample
In a total sample of 640 participants, 251 (Group 1: 39.2%) reported having any sort of stool-based colorectal cancer screening tests, while 389 (Group 2: 60.8%) reported not having any sort of stool-based colorectal cancer screening tests (Table 1). Nearly 50% of the sample constituted Whites and Christians and were married. The sample consisted predominately of females (57.6%), non-Hispanics (80.2%), and those with health insurance (88.5%) ( Table 1). Participants who have had any form of stool-based colorectal screening tests (group 1) had a significantly lower proportion of uninsured individuals than those who have not had any form of stool-based colorectal screening tests (group 2) (2.7% vs. 8.1%, p = 0.009, Table 1). A significantly higher proportion of those who underwent any sort of stool-based colorectal cancer screening had a personal history of colorectal cancer (5.6% vs. 1.1%, p = 0.001), had a personal history of inflammatory bowel diseases (13.0% vs. 6.2%, p = 0.005), had a personal history of confirmed or suspected hereditary colorectal cancer syndrome (7.0% vs. 1.7%, p = 0.001), had a history of abdominal irradiation (9.2% vs. 2.6%, p < 0.001), and were recommended a colorectal screening by their healthcare providers (45.5% vs. 28.7%, p < 0.001, Table 2).

Correlation and Reliability Diagnostics
As shown in the Table 4, "participatory dialogue" was positively and moderately correlated with the "behavioral confidence" (R 2 = 0.51, p < 0.001), "changes in physical environment" (R 2 = 0.45, p < 0.001), and "changes in the social environment" (R 2 = 0.42, p < 0.001). Behavioral confidence had a strong positive correlation with the "changes in physical environment" (R 2 = 0.75, p < 0.001) and "changes in social environment" (R 2 = 0.68, p < 0.001). The "change in physical environment" was strongly and directly correlated with the "changes in the social environment" (R 2 = 0.77, p < 0.001, Table 4). The overall reliability (Global Cronbach's Alpha) of the tool was 0.942, and for the subscales, it ranged from 0.87 to 0.91, which indicates excellent reliability. Global Alpha = 0.942; ** p < 0.01; "participatory dialogue" was measured through "perceived advantages" and "perceived disadvantages".

Hierarchical Regression
Among the participants who underwent stool-based CRC screening, the MTM subscales, including "participatory dialogue", "behavioral confidence", and "changes in the social environment", were significant predictors of initiating screening behavior and explained 61% of the variance in the initiation among this group (R 2 = 0.625, F = 15.730, p < 0.001; adjusted R 2 = 0.609) (model 4, Table 5). Another significant predictor was the urban residence that had a conditional mean of (0.601 (intercept) + 0.358 (slope) = 0.959)). In other words, an average participant living in an urban area has a 0.358 points higher score of initiation than those living in rural areas.   Among the participants who did not undergo stool-based CRC screening, the MTM subscales, including "participatory dialogue", "behavioral confidence", and "changes in the social environment", were significant predictors of initiating-screening behavior and explained 48% of the variance in the initiation among this group (R 2 = 0.579, F = 5.916, p < 0.001; adjusted R 2 = 0.481) (model 4, Table 6). Another significant predictor was the AAPI race, which had a conditional mean of (1.171 (intercept) + 0.726 (slope) = 1.897)). In other words, on average, the AAPI race has a 0.726 points higher score of initiation than the Black race.

Discussion
The study aimed to explain the correlates of seeking stool-based screening tests for colorectal cancer among adults aged 45-75 years using the modified initiation theoretical paradigm of the multi-theory model (MTM) of health behavior change. The interpretation of the results is divided into two groups: (1) those who had stool-based CRC screening tests; (2) those who did not have stool-based CRC screening tests. Both of these interpretations are complementary to each other.
Since MTM is about health behavior change, we will first interpret the results for those who did not have any form of stool-based CRC screening tests. It is worth noting that the constructs of participatory dialogue, behavioral confidence, and changes in the social environment, along with being of Asian American and Pacific Islander (AAPI) racial denomination, were positive and statistically significant explanatory variables; these accounted for nearly 48% of the variance in the intent to seek stool-based CRC screening tests. This amount of explanation is considered to be fairly high in behavioral and social sciences [22]. It is evident from these findings that being convinced of advantages (such as early detection of CRC, increased chances of cure, having peace of mind, reduction in worries, chances of living longer, and so on) over disadvantages (participatory dialogue) is crucial in swaying the decisional balance for those who have not yet taken stool-based CRC screening tests. Similarly, the ability to proactively seek these CRC screening tests at prescribed intervals with conviction despite difficulties (behavioral confidence) is also vital in motivating individuals who have not yet taken the stool-based CRC screening tests to take the tests. Likewise, support from family, friends, healthcare professionals, social media, and so on (changes in the social environment) is also essential for motivating people to take stool-based CRC screening tests. These findings are in consonance with previous work based on other theories such as HBM, TTM, and TPB [11]. In demographic variables, our findings indicated that AAPI were more inclined to take stool-based CRC screening tests than Blacks. AAPIs have the lowest incidence and mortality due to CRC among all races [2]. Perhaps the lower mortality could be explained due to the early detection of polyps in this racial group. Health insurance status was also a predictor of obtaining a stool-based CRC screening test which aligns with previous findings [29]. Increased efforts to ensure equitable access to health are needed. Our study did not find changes in the physical environment or any other demographic variables to be significant predictors of the intent to seek stool-based CRC screening among adults aged 45-75 years who had not yet taken these tests in the final model.
As mentioned earlier, the findings from those who had stool-based CRC screening in the past were found to be complementary to the findings explained earlier. Once again, the MTM constructs of participatory dialogue, behavioral confidence, and changes in the social environment were statistically significant and accounted for almost 61% of the variance in the intent to seek stool-based CRC screening when due. As pointed out earlier, previous evidence from work in this area using HBM, TTM, and TPB is supportive of our findings [11]. The only difference in the findings among those who had CRC screening tests in the past that was noted with regard to demographic variables was that race was not significant. However, the predicted value for urban areas was found to be higher than for rural areas in the final model. This can perhaps be explained due to higher awareness of CRC and its screening in urban areas compared to rural areas.
Findings from the participants' mean MTM scores revealed that neither group reached the highest possible scores, suggesting room for improvement. This can be conducted through educational interventions which can be used to increase scores in the MTM constructs. Further, all of the MTM construct's mean scores were higher in directionality for those who had any form of stool-based CRC screening compared to those who did not. These results also support the inferential findings; these present a credible case for designing interventions using MTM to promote stool-based CRC screening tests. The findings for pre-existing conditions and their association with stool-based screening tests are also worth noting. The proportion of individuals with a personal history of CRC, a history of IBD, a history of hereditary CRC syndrome, a history of radiation to the abdomen, recommendations from healthcare providers, and a recent visit to a healthcare provider were all significantly higher among those who had a history of stool-based CRC screening compared to those who did not. These findings point to increased awareness and motivation among those who have personal adverse experiences or have been motivated by healthcare providers. This can be capitalized by future interventions to promote stool-based CRC screening tests among adults aged 45-75 years.

Implications for Practice
MTM seems to be a useful framework with which to design educational, mass media, social media, and clinical interventions promoting stool-based CRC screening among adults aged 45-75 years. For educational interventions, the advantages of stool-based CRC screening need to be facilitated through face-to-face settings or asynchronous/synchronous online settings. These advantages include highlighting early detection of CRC through these tests, increased chances of a cure if detected early, developing peace of mind, reduction in worries, chances of living longer, and so on. Moreover, in educational interventions, behavioral confidence can be built by demystifying the screening tests and identifying and remediating possible barriers to obtaining them. To influence changes in the social environment, family support, friends' support, and healthcare professionals' support should be actively encouraged.
In mass media and social media campaigns, messaging is very important. Succinct and clear messages based on MTM constructs can be crafted for dissemination. For example, a single message about an advantage of stool-based Cologuard can be, "Convenience of home-based detection of colorectal cancer now within reach! Visit your primary care provider". This can be supplemented with a visual appeal.
Finally, interventions can be designed to be delivered by clinicians in clinical settings. Once again, these need to be succinct, brief, and to the point in order that the providers can easily incorporate these into their busy schedules. These can consist of a crib sheet consisting of 2-3 lines to be emphasized in communication with the patient that is readily accessible to the provider as a reminder to be read at each visit. Table 7 below summarizes a possible script for suggesting Cologuard to an eligible patient.

Advantages
Cologuard is a non-invasive, home-based stool test for the early detection of polyps and colon cancer for people 45 years of age and older.
Behavioral confidence It is very easy to use and comes with all detailed instructions.
Changes in the social environment Should you decide to use it, our nurse will be able to explain all preliminary details to you.

Strengths and Limitations of Our Study
Ours is the first study using the fourth-generation framework of MTM in the conceptualization and synthesis of data about the correlates of stool-based screening tests for CRC in adults aged 45-75 years. Our study was able to explain a substantial amount of the variance in the intent to take stool-based CRC screening tests in our target population. Our study was able to provide definitive implications for designing interventions in order to promote stool-based CRC screening tests in our target population. However, our study did have some limitations. First, we did not analyze the data pertaining to visual (structural) tests such as colonoscopy, virtual colonoscopy, or flexible sigmoidoscopy. Second, we used a cross-sectional study design, which gave us quick results but precludes us from making causal inferences due to a lack of established temporality. Third, we relied on self-reports that are amenable to several biases. However, our study was primarily based on identifying attitudes, and there could be no other means by which to do so. Next, as part of the contractual agreement with Qualtrics, partial responses were not provided, which limited our ability to analyze the characteristics of the non-responders, which is important for understanding the sources of non-response bias. Finally, while we established the face content and construct validity of our instrument, along with the internal consistency reliability, we could not establish test-retest reliability. This is something future researchers should attempt before embarking on intervention research.

Conclusions
This study reified the constructs of the modified initiation framework of the multitheory model (MTM) of health behavior change in order to explain the correlates of seeking stool-based screening tests for colorectal cancer among adults aged 45-75 years and found it to be cogent. Specifically, the constructs of participatory dialogue, behavioral confidence, and changes in the social environment were highly predictive of seeking stool-based CRC screening tests. These constructs need to be epitomized in the next phase of subsequent research endeavors in this direction.  Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to ethical reasons.