Development and initial psychometric evaluation of the computer-based prostate Cancer screening decision aid acceptance scale for African-American men

While some technology acceptance factors are highly correlated with usability (e. g., ease of use; [27]), most technology acceptance models also include constructs that are external to the user (e. g., social influences such as subjective norms; [25]). Based on socio-ecological theory, technology acceptance models posit that an individual’s decision to adopt a specific technology is not based solely on self-identified benefits, but is the result of complex interactions between an individual and their social and physical environment [25]. One of the most widely accepted models informing the measurement of technology acceptance is the Technology Acceptance Model [28], which posits that users’ adoption of a technology for normal use is dependent on its perceived usefulness (i.e., enhances task performance) and perceived ease of use (i.e., extent to which a technology requires effort). Developed in 1989, the technology acceptance model has undergone several modifications to enhance its utility (e.g., Technology Acceptance Model 2; [29, 30]). Specifically, modified models have integrated a plethora of other socio-ecologic factors that influence technology use. [30] Since it’s introduction, the technology acceptance model has been tested with over 25 various external variables that are posited to influence the relationship between the three major constructs: perceived usefulness, perceived ease of use, and technology acceptance [30]. The most commonly tested variables include computer anxiety, self-efficacy, enjoyment, computer support, and computer-use experience [30]. To enhance the performance of the technology acceptance model, Venkatesh and Davis, the model’s creators, have also expanded the model integrating additional five variables, including job relevance (an individual’s perception regarding the degree to which the target system is applicable to his or her job), subjective norms (an individual’s perception that most people who are important to them think they should or should not perform the behavior in question), image (the degree to which use of an innovation is perceived to enhance one’s status in one’s social group), output quality (how well the system performs specific tasks), and result demonstrability (tangibility of the results of using the innovation) to explain the conditions under which individuals affect a users’ perceived ease of use [29]. Two additional variables, experience and voluntariness of use, were hypothesized to mediate subjective norms and perceived usefulness and/or an individual’s intention to use a system. Similar to the original technology acceptance model, this modified version (i.e., Technology Acceptance Model 2) has been widely adopted [25] and used in a variety of contexts, which also includes the assessment of technology-use in the healthcare environment [31, 32]. Synthesizing the Technology Acceptance Model with existing models to examine technology acceptance (e. g., Diffusion of Innovation; [33]), Venkatesh and Davis et al., [34], collaborators on the Technology Acceptance Model 2, developed the Unified Theory of Acceptance and Use of Technology. Developed in 2003, the model postulates that four factors (performance expectancy, effort expectancy, social influence, and facilitating conditions) and four moderators (i.e., age, gender, experience, and voluntariness) determine an individual’s intention to use a technology and ultimately whether they decide to adopt a technology for regular use [34]. In their development of the Unified Theory of Acceptance and Use of Technology, Venkatesh et al. [29] also examined an individual’s self-efficacy, anxiety, and attitudes and hypothesized that these factors were not causally related to technology use intention, but were fully mediated by other factors in their model (e.g., effort expectancy). Based on the Unified Theory of Acceptance and Use of Technology, Venkentash et al. [29] created a 24-item scale, the Unified Theory of Acceptance and Use of Technology Scale, that measures acceptance and use of a technology based on each of the aforementioned moderating factors and intention to use technology. In their seminal article, Venkentash et al. [29] found that factor loadings for each item were acceptable with most loadings being .70 or higher. In addition, internal consistency reliability for the full scale and subscales ranged between .77 and .94. As hypothesized by Vekentash et al. [29], self-efficacy, anxiety, and attitudes towards technology did not have a direct causal relationship with intention to use technology. However, exploratory factor analysis of the Unified Theory of Acceptance and Use of Technology Scale found that self-efficacy and anxiety are significant predictors of technology use intention [35]. Despite these differences, the Unified Theory of Acceptance and Use of Technology Scale has been validated for use across several sectors including, but not limited to, mobile banking, social networking, web-based learning environments, decisions support systems, digital learning environments, and retail [36, 37].

Related to decision support systems and web-based learning environments, there are a growing number of studies that have employed the Unified Theory of Acceptance and Use of Technology Scale for assessing technology acceptance in health environments [38]. These studies most often investigate the acceptance of clinical support technology, such as electronic medical records, by healthcare providers or other clinical staff [39‐43]. However, a number of studies assessed the acceptance of web-based telecare systems among patients [44‐47]. Each of these studies demonstrate the application of the Unified Theory of Acceptance and Use of Technology to healthcare settings, though the strength of hypothesized relationships among the variables has varied. According to Holden and Karsh [31], these inconsistencies in the strength of associations among variables can be largely explained by differences in the contextual operationalization of the constructs within the Unified Theory of Technology Use and Acceptance. More specifically, some studies implement the scale with general wording, but others often alter wording to be context specific to the type of technology being tested [31]. The addition of contextual relevancy is necessary to ensure that the scale, which was originally designed for use in non-healthcare settings will meaningfully translate for use in a healthcare setting [31]. Therefore, some studies also added environment-related constructs to the model which made it even more contextually relevant [31, 44, 46‐49]. For example, Ciperman et al. [46] posited that a doctor’s opinion regarding the use of telehealth technology influences performance expectancy, computer anxiety influences effort expectancy, and perceived security influences performance and effort expectancy and has a direct influence on behavioral intention. Findings show that computer anxiety has a significant negative influence on effort expectancy, while doctor’s opinion and performance expectancy had significant positive influences on performance expectancy and behavioral intention, respectively. Despite methodological differences, the aforementioned studies report acceptable to high internal consistency reliability scores (α= > .70), with some researchers conducting factor analyses [46, 50]. A few studies also found that the Unified Theory of Acceptance and Use of Technology Scale had strong convergent and divergent validity for assessing technology acceptance for health-related purposes [46, 50, 51].

Only one recent study [52] has investigated the Unified Theory of Technology Use and Acceptance in relation to cancer care or decision making. Among 300 cancer survivors, Senft et al. [52] examined whether facilitating conditions, social influence, ease of use, perceived usefulness, and security/trustworthiness was associated with eHealth use and if attitudes about the security and trustworthiness of online health services was associated with eHealth use more strongly among African-American than White cancer survivors. They discovered that facilitating conditions and perceived usefulness are associated with increased eHealth use among African-Americans and Whites and social influence did not influence e-health use among either group. Also, perceived ease of use was associated with decreased e-health activity for Whites only, whereas security/trustworthiness is associated with increased eHealth activity for African-Americans only. The authors do not report the internal consistency reliability of the Unified Theory of Technology Use and Acceptance Scale in this study. A second cancer-related study [53], proposes to use the Unified Theory of Technology Use and Acceptance Scale to understand the acceptance and use of an mHealth app for PrCA survivorship by patients, caregivers, and clinicians in the United Kingdom, but only a study protocol is currently available.

Although a number of studies report acceptable reliability and validity of the Unified Theory of Technology Use and Acceptance Scale for assessing technology use and acceptance across sectors, including healthcare, the psychometric properties of this scale have not been tested among African Americans. Having a culturally-relevant measure of technology use and acceptance is needed to measure the performance of technology-based interventions that seek to enhance decision making about PrCA screening among African-American men, who experience the highest mortality from the disease [1]. Given the need for and utility of a reliable and valid CBDA for PrCA, this study described the development and tested the psychometric properties of the Computer-Based Prostate Cancer Screening Decision Aid Acceptance Scale among African-American men.

This study included a purposive sample of 354 African-American men aged 40 and older. Eligible participants were those who (a) self-identified as African American; (b) spoke and comprehended English; (c) had no personal history of PrCA; and (d) had no self-reported history of cognitive decline. Participants were not required to have prior technology use experience. They were recruited from several social and faith-based venues in South Carolina between July 2015 to February 2016. All participants received a written consent form in-person prior to their participation in any study activities. Prior to requesting a signature on the consent form, the form was explained in detail by a member of the research team. This study was approved by the Institutional Review Board at the University of South Carolina.

The Unified Theory of Acceptance and Use of Technology Scale was adapted to examine the usability and acceptance of iDecide, a CBDA for African-American men [55]. Specifically, each question was modified to refer generally to a CBDA as opposed to generally referring to a “system”. For example, question Q28 (Table 1) was adapted to read “The system is somewhat intimidating to me” to “The CBDA is somewhat intimidating to me.” This question revision is similar to prior studies that have adapted the Unified Theory of Acceptance and Use of Technology Scale to be contextually relevant to their study environment [31]. Scale modification was conducted by the first author (O.O), an African-American male with expertise in health communications, health technology, and PrCA within the target population. Therefore, scale items were adapted for contextual relevancy while maintaining content equivalence and items were eliminated that were not contextually related to the CBDA (e.g., The senior management of this business has been helpful in the use of the system). Subscales (performance expectancy, effort expectancy, social influence, and facilitating conditions, self-efficacy, attitudes towards technology, and anxiety) and two of four original moderators (i.e., age, experience) that are hypothesized to be directly related to technology acceptance were retained. Because this scale is specific to CBDAs, the modified scale was titled, the Computer-Based Prostate Cancer Screening Decision Aid Acceptance Scale.

The Computer-Based Prostate Cancer Screening Decision Aid Acceptance Scale is comprised of 24-items with Likert scale response categories ranging from 1 (strongly agree) to 5 (strongly disagree) and 6 (does not apply). Scoring involves taking the average of all responses, excluding does not apply responses. A higher score indicates greater acceptance and use of the CBDA for PrCA informed decision making. Table 1 compares items of the Unified Theory of Acceptance and Use of Technology Scale to items of the Computer-Based Prostate Cancer Screening Decision Aid Acceptance Scale.

Based on the Unified Theory of Technology Use and Acceptance [29],the Computer-Based Prostate Cancer Screening Decision Aid Acceptance Scale emphasizes the influence of the dynamic interplay between the individual and their social and physical environment about whether an individual will adopt a CBDA. To determine the acceptance and use of a CBDA for PrCA informed decision making, seven factors are influential including: (a) performance expectancy (the degree to which an individual believes that the CBDA will lead to personal gains such as increases in PrCA knowledge and decision self-efficacy), (b) effort expectancy (the amount of effort associated with using the CBDA to retrieve PrCA information), (c) social influence (the degree to which an individual perceives that his or her social network will endorse the use of the CBDA), (d) facilitating conditions (the degree to which an individual believes that infrastructure exists to support their use of a CBDA for informed decision making about PrCA screening), (e) computer anxiety (the level of emotional fear or apprehension when an individual thinks about having to use a CBDA for finding PrCA information), (f) self-efficacy (an individual’s belief in their ability to effectively use the CBDA to find PrCA information, (g) attitudes towards technology (an individual’s positive or negative behaviors regarding use of the CBDA) and (h) behavioral intention (an individual’s intention to use a technology). Each of these factors are moderated by an individual’s age and technology-use experience [34]. Counter to the original Unified Theory of Technology Use and Acceptance, gender and voluntariness are not moderators in our conceptual framework because the CBDA is designed for an African-American male population and use of this technology is voluntary.

To assess face validity, the Computer-Based Prostate Cancer Screening Decision Aid Acceptance Scale was pre-tested with a convenience sample of two African-American men with high school or higher education. During the pre-test, the men were provided with a survey containing the full battery of 65-items that were used during the testing of the iDecide PrCA screening CBDA. These men completed a hard copy survey and noted if there were questions, words, or concepts on the survey they found difficult to interpret or that might be difficult to interpret for men with low reading levels. After survey completion, they were interviewed by the first author (O.O.) about difficulties completing the survey along with questions about survey formatting (e.g., clarity of instructions, response option formatting). Neither participant suggested changes to the survey.

Descriptive statistics described the sociodemographic characteristics of African-American men in the sample. Polychoric correlations assessed the association between factors and subscale items. Cronbach’s alpha assessed internal consistency reliability for the total scale and each subscale.

Exploratory factor analysis (EFA) is a data-driven, exploratory technique and that does not require a priori specification of the relationships between latent and observed variables [56‐58]. Thus, model specification is not required because factor structure and factor loadings are assumed to be unknown. In this study, EFA was conducted to identify the number of latent constructs (factors) and the underlying factor structure of the Computer-Based Prostate Cancer Screening Decision Aid Acceptance Scale. Performance expectancy, effort expectancy, social influence, facilitating conditions, computer anxiety, self-efficacy, and attitudes towards technology were exogenous latent variables hypothesized to be correlated with each other. Subscale items loading on each factor were also hypothesized to be correlated with each other. The number of participants to item ratio was 14:1, which is above the recommended 10:1 often used to determine a priori sample size for EFA [59].

EFA was conducted using preliminary estimates of communalities obtained from the square of the multiple correlation coefficient of each variable. Iterated principal factor extraction with prior communalities set to 1 was used for data extraction followed by Varimax rotation. Factor retention was assessed through parallel analysis, which has been demonstrated as a more accurate assessment of factor retention than other factor retention methods [58]. Specifically, using the K-1 method can lead to sampling error, which can overestimate the number of factors [60]. Parallel analysis produces correlation matrices from a randomly chosen simulated dataset that has a similar number of observations as the original dataset [60]. The simulated observations have the same potential sampling error as the original observations. Eigenvalues were then computed for both the simulated and original data. To determine the number of factors to retain, simulated and original data eigenvalues were compared to determine the point at which the eigenvalue in the simulated data was higher than the original data [61]. The number of factors before this transition point denoted the number of factors that were retained [61]. A scree plot was also produced to visually compare eigenvalues from simulated and original data to corroborate our determination of the number of factors to retain [59].

Factor loadings were assessed using item communalities, cross-loadings, and item statistics. A factor with less than three factor loadings was considered weak and unstable [59] and was deleted from the factor structure. Factors with three or more factor loadings were retained. An item was determined to load on a factor if the factor loading was 0.40 or greater for that factor and was less than 0.40 for other factors [62]. An item was cross-loaded if it loaded on more than one factor at 0.40 or above [59]. The standardized root mean square residual measured the difference between the observed correlation and the predicted correlation. Acceptable standardized root mean square residual estimates are less than .05 [63]. The Kaiser-Meyer-Olkin measured sampling adequacy and estimates between 0.8 and 1 were considered adequate [64].

Missing values ranged from 0.56% (n = 2) to 2.54% (n = 9) for scale items. Single and multiple imputation were used to impute missing values for 24 items. Means for each item was compared with and without imputation, which included no imputation, single imputation, and multiple imputation (n = 1000) for missing data, which were similar. Descriptive statistics were analyzed using original data (N = 354). EFA was conducted using original, single, and multiple imputation datasets given the Computer-Based Prostate Cancer Decision Aid and Acceptance Scale is a major adaptation of the Unified Theory of Acceptance and Use of Technology Scale. All data analyses were performed using SAS/STAT® statistical software, version 9.4 [65].

Table 3 reports factor loadings of the 24-item Computer-Based Prostate Cancer Screening Decision Aid Acceptance Scale using the original, single and multiple imputation datasets. Factor loadings were similar for original, single and multiple imputation datasets. Parallel analysis, scree plot (Fig. 1), and the proportion of variance explained by each factor suggested three meaningful factors. Factor 1 (F1), Technology Use Expectancy and Intention had 16 factor loadings ranging from .51 to .85 for original data, and .44 to .83 and .44 to .83 for single and multiple imputation, respectively. Factor 2 (F2), Technology Use Anxiety, had five factor loadings ranging from .52 and .92 for original data, and .41 to .93 for both single and multiple imputation. Factor 3 (F3), Technology Use Self-efficacy, had three factor loadings ranging from .67 to .88 for original and imputed datasets.

Factor-loadings varied between original and imputed datasets. For the original dataset, three items, Q11, Q12, and Q13, cross-loaded on two factors—Technology Use Expectancy and Intention (F1) and Technology Use Self-Efficacy (F3). For single and multiple imputed datasets, five items, Q9, Q10, Q11, Q12, and Q13, cross-loaded on two factors—Technology Use Expectancy and Intention (F1) and Technology Use Self-Efficacy (F3). All cross-loaded items were retained on Technology Use Expectancy and Intention (F1) because all factor loadings were highest on F1 compared to F3 (Table 3).

The Kaiser-Meyer-Olkin measure of sampling adequacy was 0.93, which is acceptable. All residuals were small and the overall standardized root mean square residual was 0.035, indicating that the factor structure explains most of the correlations. Further, the internal consistency reliability of the full scale was .91 for the original dataset and .89 for both single and multiple imputation datasets; and .95, .90, and .85 for Factors 1, 2, and 3, respectively, for all datasets (Table 4).