Study design, period, and setting: An institution-based cross-sectional study was conducted from May 20-June 15/2022. This study was conducted at the University of Gondar, which is a higher education institution located in the Amhara National Regional State approximately 740 kilometers away from Addis Ababa, the capital city of Ethiopia. University of Gondar has five campuses, namely, the College of Medicine and Health Sciences, the Atse Tewodros Campus, the Maraki Campus, the Atse Fasil Campus, and the Teda Campus. The university is organized under six colleges, two institutes, two faculties, and one school. These are College of Medicine and Health Science(CMHS), College of Natural and Computational Science(CNCS), College of Business and Economics(CBE), College of Social Science and Humanity(CSSH), College of Veterinary Medicine and Animal Science(CVMAS), College of Agriculture and Environmental Science(CAES), Institute of Technology(IOT), Institute of Biotechnology(IOB), Faculty of Informatics(FOI), Faculty of Education(FOE), and school of law(law)[73]. In the 2021/22 academic year, the university had 9,810 regular undergraduate students. Among those, 2910 were female students (College of Medicine and Health Science=933 and 1977 from Non-health Science Colleges) [74].
Populations
Source population: All female regular undergraduate students registered for 2021/22 at the University of Gondar.
Study population: All female regular undergraduate students registered for 2021–22 years and were active students at the time of data collection.
Sample size determination and sampling procedure
The sample size was calculated according to the planned analysis method, which is structural equation modeling (SEM). SEM was used for this study because it provides a flexible and powerful means of simultaneously assessing the quality of measurement of constructs and examining both direct and indirect relationships among constructs[75]. This technique requires a large sample. It is impossible to determine what constitutes a “large enough” sample size in SEM. Sample size estimation is flexible, and there is no “one-size-fits-all” rule[76]. As a general rule of thumb, some researchers recommend a minimum sample size of 200 patients, while others suggest a minimum of 5 or 10 patients per indicator in the model[76]. For this study, the sample size was calculated using an online calculator provided by Soper [77] for a priori sample size for SEM. A reasonable lower bound of the sample size is calculated when the proposed model is used, the number of latent variables, the number of observed variables and the number of hypothesized causal paths are known[75]. By setting the medium-to-waitant effect = 0.3, power = 80%, number of latent variables = 10, number of observed variables = 82, and type one error (α) = 0.05, the required minimum sample size was calculated to be 772.
The nonresponse rate was adjusted by adding 10% of the 772 participants; thus, the final required sample size for the study was computed to be 849.
Study participants were selected using a stratified simple random sampling technique. Because of the possible difference in BSE awareness and practice across colleges and years of study [36, 37], the sample size was proportionally allocated for health science students from CMHS and non-health science students from other colleges, institutes, faculties, and schools; subsequently, the samples were proportionally allocated to years of study. Subsequently, by obtaining a list of students from the assistant registrars of each collage, a simple random sampling technique was used for the selection of study units.
Variables of the study
Endogenous observed variable: BSE behavior
Endogenous latent variables: fear of breast cancer and protection motivation
Exogenous observed variables: age, year of study, previous residence, department, family history of breast cancer, father educational status, mother educational status, ever noticed a lump in breast, ever discussed with someone on BSE, ever heard about BSE, and knowledge on BSE
Exogenous latent variables: Perceived severity, perceived vulnerability, maladaptive response reward, response efficacy, self-efficacy, response cost, and attitude toward BSE;
Study variable measurements
BSE behavior: BSE behavior was assessed as self-reported BSE within the past month[78]. An index with ten indicators, such as “Did you perform BSE in the last month?”, was used to assess BSE behavior and was adapted from the literature[32, 79]. Each of the items was scored 1 if conducted by the participant and 0 if not practiced. To calculate the BSE behavior score for each participant, practice questions were computed and combined; higher scores indicate greater BSE behavior. For descriptive purposes, a demarcation threshold formula was used to categorize participants as having regular BSE behavior or not.
Cutoff value= (total highest score-total lowest score/2) +total lowest score
= ((10-0)/2) +0=5
Based on this, those who scored 5 or above were considered to have regular BSE behavior.
PMT constructs: were examined using measures adapted from previous literature [26, 80-83] and based on the recommendation of Norman et al., 2015[49]. Separate multi-item subscales were used to assess PMT constructs; individual items were weighted equally and ranked on 5-point Likert scales. For all PMT constructs, subscale scores were generated by summing component items after reverse coding of negatively worded items; higher scores indicate higher levels of the corresponding constructs.
Perceived severity: Nine items [26] were used to collect information about the perceived severity of breast cancer (ps1-ps9). (E.g., breast cancer can lead to mastectomy (removal of the entire breast)).
Perceived vulnerability: five items (pv1-pv5) [80-82] were used to assess vulnerability E.g., I feel I will get breast cancer sometime during my life
Maladaptive reward: Five items (mr1-mr5) [26] were used to assess mal adaptive reward. (E.g., early diagnosis of a breast mass is not important for treating breast cancer)
Response efficacy: Six items (re1-re6) [26] were used to collect data on beliefs about the potential advantages of BSE. (E.g., doing self-examinations saves the costs that would be incurred by cancer treatment)
Self- efficacy: Six items (se1-se6) [26] were used to measure perceived capability to undergo BSE given certain obstacles (e.g., I can perform breast self-examination, even if it’s time consuming).
Response cost: Five items [26, 83] were used to assess perceptions of potential disadvantages resulting from BSE. (E.g., remembering the timing of self-examinations is difficult for me)
Fear: Eight items adopted from the Breast Cancer Fear Scale [84] were used to assess fear of breast cancer (e.g., the thought of breast cancer scares me).
Knowledge of BSE: This is the awareness and understanding of the steps involved in examining one’s own breast for any lumps or changes in size, shape, or texture. It was measured using 11 items adapted from a review of the literature [32, 79]. Example item: “How can a woman perform breast self-examination? What should be looked for during breast self-examination? Optional answers, including ‘don’t know’ as one option, were provided to prevent participants from selecting the correct answer by chance. Each correct response was scored as 1, and each wrong and ‘don’t know’ response was scored as 0. The scores for each participant’s knowledge questions were pulled together, and higher scores of knowledge indicated higher levels of knowledge. For descriptive purposes, the demarcation threshold formula was used to categorize participants as having good BSE knowledge or not.
Cutoff value= (total highest score-total lowest score/2) +total lowest score
= ((11-0)/2) +0=5.5
Based on this, those who scored 6 or above were considered to have good BSE knowledge.
Attitude: People’s overall positive or negative evaluation of performing monthly BSE was measured using six items, with the stem “for me to perform monthly BSE would be”, followed by the adjectives “good”, “bad”, “wise”, “folish”, “beneficial”, harmful”[85].
Data collection, validity and reliability of the questionnaire
The data were collected using a self-administered questionnaire. The questionnaire had 5 parts, including socio-demographic and related variables, BSE behavior, PMT constructs, BSE knowledge and BSE attitude. S4 Questionnaire
The questionnaire was initially prepared in English and subsequently translated into Amharic and returned to English to check for consistency. In the questionnaire, the items were presented in mixed order so that their conceptual assignment to factors was not transparent to the respondents. The reliability of the questionnaire was assessed by the internal consistency method (Cronbach’s alpha coefficient), and values equal to or greater than 0.7 were considered acceptable[86]. All the constructs had Cronbach’s alpha coefficients above 0.7 except for response cost (0.53) (Table 3).
To ensure validity, the questionnaire was checked for face and content validity by nine experts, including 7 health promotion and behavioral science experts, 1 reproductive health expert, and 1 Gyn/OB resident (R4). The item-level content validity index (I-CVI) and scale-level content validity index (S-CVI) were computed, and 0.78 was taken as the cutoff[87]. All the items had acceptable I-CVIs and S-CVIs (S1 Supplementary file 1), and necessary amendments to the questionnaire were made based on the facial validity findings.
Based on face validity, double-barrel questions were modified. For example, an item saying “Development of breast cancer results in the loss of a feminine appearance and beauty” was changed to “Development of breast cancer results in the loss of beauty”. The leading questions were adjusted; for example, “Do you perform BSE by touching your entire breast?”, was changed to “How do you perform breast self-examination?”, and ambiguous words were substituted; for example, “Have you accomplished BSE in the last month?”, was modified to “Have you performed BSE in the last month?”. Additionally, regarding face validity, some items were removed, e.g., “By conducting a monthly self-examination, I can prevent breast cancer,” because BSE cannot prevent breast cancer but rather is good for the early detection of any abnormalities. Variables such as family education, previous residence, and ever detected a lump were included from the expert suggestion. A pretest of the instrument was subsequently performed on 5% of the participants (42 students) at Gondar College of Teacher’s Education and Teda Health Science College. Some of the changes made after the pretest were word modifications, e.g., The Amharic version of the term “Aladnim is changed to Alkotbim” meant that I did not save much in terms of cost, the print was changed to a separate page and portrait layout, and instructions and important words were highlighted and underlined. Moreover, items for measuring attitudes were written separately.
Data processing and analysis
The data were entered into Epi Data version 4.6 and exported to STATA version 14 and AMOS 26 for further data management and analysis. The data were checked for missing cases, and outliers and appropriate amendments were made. Variable coding and transformations were performed to ensure that the dataset was ready for analysis. The multivariate normality and multivariate outlier test results were checked; the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity were computed. Descriptive analysis was performed, and the frequencies, medians and interquartile ranges (IQRs) were calculated.
Confirmatory factor analysis was used to assess convergent and discriminant validity and the fitness of the measurement model. Convergent validity was ensured by the strength of standardized factor loadings (0.5 and above), composite reliability (0.7 and above), and average variance extracted (AVE) (0.5 and above) [76]. Since the AVE is a strict measure and difficult to measure for some constructs, 0.4 was also taken as acceptable[88]. Discriminant validity was assessed by using the Fronell–Larcker table, which assumes that the square root of the AVE for every latent variable should be greater than that of other correlation values among the latent variables[89]. Misspecifications in the fitted model were assessed based on modification indices. The path coefficient and the causal relationship between the variables were tested by SEM using bootstrapping techniques that offer no distributional assumptions[90]. To verify the fitness of the proposed model, χ2, goodness-of-fit index (GFI) > 0.9, adjusted GFI (AGFI) > 0.9, comparative fit index (CFI) > 0.9, root mean square error of approximation (RMSEA) < 0.08, standardized root mean residual (SRMR) < 0.05, and chi-square/df < 5.0 were used. A p value less than 0.05 was considered to indicate statistical significance [76]. The writing of the results was guided by guidelines for reporting SEM-based research findings[91].
Ethical consideration
Ethical approval was obtained from the University of Gondar, College of Medicine and Health Science, Institute of Public Health Ethical Review Committee (Ref.no: IPH/2121/2014). Study participants were assured that all the information provided would be kept in strict confidence, that the data would be used only for research purposes, and that any personal identifiers (ex. name) will not be asked. Written consent was subsequently obtained from the participants after informing them that participation in the study was voluntary; they could refuse to participate or withdraw from the study at any time, and there was no incentive or payment for their participation in the study.