Study population and site
Fisher’s et al. (1998) adapted by Mugenda and Mugenda 2003 [
28], n = z
2pq/d
2 was used determine the sample size. n = desired sample size; Z = the abbisca of normal distribution (z = 1.96); p = the proportion of the population tested for prostate cancer nationally (11%); q = 1-P (proportion not tested; d = maximum degree of error with a confidence interval of 95% = 0.05. This calculation gave a minimum sample size of 155.
Although cancer of the prostate occurs in men who are 40 years and above, this study purposely included all adult males found in the household who had some knowledge of causes and symptoms of prostate cancer. Only men who answered any one of the questions on knowledge of prostate cancer, or mentioned one symptom of prostate cancer were included in the study. Most of the youth were either in school or away from home. Those found at home during the study period happened to be aged over 25 years. We excluded those who were mentally unsound or unable to communicate.
The study was carried out in Kasikeu County Assembly, Makueni County, Kenya a rural ward. The ward was randomly selected from 30 County Assemblies in the County between October 2014 and February 2015. The population of males in this ward was 4569 while that of females was 4933 making a total population of 9502 according to the 2009 census.
Kasikeu County Assembly comprises of 37 villages consisting of 2047 households. Each village formed a primary sampling unit (PSU), while the households in the village were secondary sampling units (SSUs). The PSUs (villages) did not have the same number of SSUs (households). Thus, we selected the PSUs using Probability Proportional to Size sampling (PPS) which gave large PSUs a greater probability of occurring in the sample than small PSUs. The households were spread across the County Assembly. We moved to the first sampled household, and if the head of household had neither heard of nor knew at least one symptom of prostate cancer we moved to the nearest household until we got the eligible study subject. To get the 155 target household heads we visited 420 households.
Study design and data sources
We used an analytical, cross-sectional design to examine the relationship between cultural variables related to behavioural beliefs, and normative beliefs with perceived behavioural control. Perceived behavioural control was assessed as the indication of a person’s readiness to screen for prostate cancer within a six-month period. Behavioural beliefs (fatalism, fear, and benefits) and normative beliefs (family influence) were the independent variables, while social demographic characteristics were possible intervening variables.
Questions used in this study were adapted from the Thomas Jefferson University Prostate Cancer Screening tool [
29,
30], which drew on health behaviour models (i.e., Health Belief Model, Theory of Reasoned Action, Social Cognitive Theory). The tool was used to assess factors associated with screening frequency among African American men [
26].
We used a structured questionnaire to collect quantitative data through face to face interviews. The questions used a forced choice categorical response to obtain consistent information from all the participants.
The participants were asked whether they intended to be screened for cancer in the subsequent 6 months. This question had a 3-point response (i.e. 1 = yes, 2 = no, 3 = don’t know). “Don’t know” response was treated as a ‘No’ response.
The basis for this classification was that “Don’t know” response was equivalent to participants who were uncertain and thus their responses could not be classified as a definite yes. Treating “Don’t know” as a yes would have biased the proportion of those who were sure that they intended to be screened. Beliefs and family influence items were measured using a 5-point Likert (i.e. 1 = strongly disagree, 2 = disagree, 3 = neither agree nor disagree, 4 = agree, 5 = strongly agree). Item total scores were used to derive two categories for beliefs and family influence.
Items related to fatalism included: “If I am meant to get prostate cancer I will get it no matter what I do,” and “If I have prostate cancer I would just as well not know about it.”
Items related to fear/apprehension included: “I am bothered by the possibility that prostate cancer screening might be physically uncomfortable,” and “I am afraid that if I have a prostate screening test, the results will show that I have it.”.
Items related to perceived benefits included: “I think the benefits of prostate cancer screening outweigh any difficulty I might have in going through the tests,” and “I believe that prostate screening is an effective way to treat prostate cancer early.”
Family influence was assessed as the perceived social pressure of a family member.
Items included in family influence were: “I want to do what members of my family think I should do about prostate cancer screening,” and “Members of my family are likely to suggest I should go through prostate screening.”
One item measured screening history. The question was whether one had received a PSA test. This question had a 3-point response (i.e. 1 = yes, 2 = no, 3 = don’t know). “Don’t know” responses for PSA were treated as a ‘No’ response. The premise behind this re-categorization was that a “no” response was equivalent to participants who did not know whether or not they received a PSA test, or were not informed or given the results of the test.
Data on social demographic characteristics which included age, education, religion and marital status were collected as possible modifying variables as well as for describing the sample.
Before data collection, research assistants were trained to ensure standardization of procedures and integrity of the data. Specific practices included: a review of procedures for recruitment of the sample, training on the data collection tool, interviewing techniques, seeking consent, maintaining confidentiality, and survey administration.
After receiving the consent to participate in the study, the research assistants explained the study’s aims, the interview process, and the approximate length of time it would take to complete the interview. The respondents were also given an opportunity to ask questions on the study before being interviewed. Interviews lasted between 15 to 25 min.
Data management and analysis
Data were entered using Statistical Package for the Social Sciences (SPSS), version 20 for Windows data file which had a participant’s unique identification number. Data checking and cleaning methods included examining ranges of responses for each individual variable through frequency distributions. SPSS was used to evaluate missing data for possible oversight upon entry, normality, and outliers. Missing data were addressed through list wise deletion which excluded variables that had missing data. The advantage of this approach is that the process produced true relationship matrices. Using this procedure, we found that there were no cases excluded from analysis due to missing data in our study.
Fisher test of skewness was used to assess whether or not the continuous data were normally distributed.
Frequency distributions and percentages were used to describe the social demographic characteristics of the study participants and to summarize key study variables. Counts and proportions were used to summarize the categorical variables such as marital status, educational level, or categorized Likert scale data. The Likert scale data was treated as a continuous variable with an interval scale. Total subscale scores were created for fatalism, fear/apprehension, perceived benefits, and family influence before statistical inferences were made. Means and standard deviations were used to summarize continuous variables. Furthermore, Likert scale scores of the independent variables were categorized into two classifications using the 50th percentile of the frequency distributions.
An analytical, cross-sectional design helped to establish the associations between cultural variables related to beliefs, normative beliefs, perceived benefits, and the intent to be screened for prostate cancer. Chi Square statistic was used to test the association between categorical independent variables (social demographic characteristics, fatalism, fear/apprehension, benefits and family influence) with the intent to screen, which was the dependent variable. The null hypotheses were that there were no significant relationships between any of the independent variables and the dependent variable. Decisions for statistical significance of the findings were made using an alpha level of <0.05. Binary logistic regression analysis was used to determine which of the independent variables best explained or the intention to screen.
All variables added to the logistic regression model had significant relationships as determined through the Chi square statistic at a p value of <0.05. The variables which were found to be significant from the Chi square statistic were selected as a block (using the ‘enter ‘procedure in SPSS) in a single step into the model.