Cassidy, Laura D. PhD, MS; Marsh, Gary M. PhD; Talbott, Evelyn O. DrPH, MPH; Kelsey, Sheryl F. PhD
Substantial effort has gone into understanding knowledge and attitudes about cancer screening, especially breast and cervical cancer, and to a lesser extent, other cancers.1 Promoting adherence to repeated screening may require interventions different from those used to promote initial screening.2 The current knowledge of factors that influence screening behavior has not grown as rapidly as the progress in methods of early detection.3 In addition, factors associated with participation in the first round of screening have been identified to be quite different from those relating to adherence in second and subsequent rounds.
The literature has concentrated on initial adherence to a screening initiative and not on repeated screening of a high-risk population because of occupational exposures. Currently, no published literature exists examining factors affecting repeated adherence in bladder cancer screening programs in occupationally exposed workers. Limited literature focusing on adherence to continuous screening for other diseases has generally focused on second or third rounds of repeated screening.3–10
The study described here evaluated factors that affect adherence rates in an ongoing multimodality, biomarker bladder cancer screening and medical surveillance program for a high-risk cohort of 407 former employees of the Drake/Kilsdonk chemical manufacturing plant with potential exposure to the potent bladder carcinogen, β-naphthylamine.11–16 The Drake Health Registry Study provided a unique population to examine screening adherence factors because it included exposed workers who had participated in repeated screenings for up to 15 sessions at the time of this analysis.
Three screening methods were prospectively evaluated in sequential screens in the Drake Health Registry Study screening program: a standard urinalysis consisting of a single-day dipstick and microscopic analysis, Papanicolaou urine cytology, and quantitative fluorescence image analysis.17 Persons with positive screens were recommended to undergo a free diagnostic evaluation including bladder cystoscopy with biopsy. Although all three methods and cystoscopy have been used elsewhere in various combinations to screen for bladder cancer,18–26 only the study by Schulte et al26 used all three tests on an occupationally exposed cohort but it did not screen repeatedly. None of the previously reported studies evaluated adherence factors in a bladder cancer screening program.
The specific aims of this study were (1) to determine whether sociodemographic, health status and attitudinal variables (health motivation and perceived susceptibility, seriousness, benefits, barriers, and control), knowledge, risk status, social influence, experiential/demographic or health status variables are statistically significant predictors of initial and repeated adherence with bladder cancer screening in a high-risk occupationally exposed cohort; (2) to develop a composite measure of overall repeated screening adherence; and (3) to determine what combination of variables best predict initial and continued (repeated) adherence.
Screening Categories and Frequency
The paradigm for the screening protocol was defined by four screening categories: positive, monitor, negative, and repeat; the categories are a composite of the three screening tests. The details of the screening protocol and the diagnostic criteria to differentiate between positive and monitor categories are described elsewhere.15 Participants are initially placed on an annual screening schedule. Persons who have a positive or monitor result are placed on a semiannual screening cycle until they have four consecutive negative results, after which they resume an annual screening cycle. All positive screens are eligible for free diagnostic evaluation, which consists of a consult with the project physician and a cystoscopy with a random bladder biopsy.
General Health Screening Survey Data
At each screening visit, the participant supplied a urine specimen and was interviewed by the clinic coordinator. The survey consisted of questions pertaining to marital status; current and former occupations; smoking history; alcohol intake; medications and medical conditions; history of family cancer; and any symptoms he or she may be experiencing such as frequency of urination, nocturia, micturition, pain, and hematuria. Survey data collected for up to 15 screening sessions were used in this analysis.
Behavioral Health Screening Survey
In addition to the general screening survey, a validated instrument used in a study to measure factors related to colorectal cancer screening adherence for a cohort of white, blue-collar, male automobile workers was modified to focus on bladder cancer.27 Vernon et al27 used items and scales from two studies of colorectal screening adherence.5,28 Several constructs included in both research studies have a long history in research on adherence with recommended health actions, including perceived susceptibility, perceived severity, benefits/costs or perceived utility from the Health Belief Model (HBM), behavioral intention from the theory of reasoned action,29,30 and subjective norms or social influence from social cognitive theory.31
The survey included the following relevant items: (1) Salience and coherence: perceptions about the technical effectiveness, practical convenience, and personal benefit and whether or not the behavior is actively encouraged by significant others. (2) Perceived susceptibility: subjective personal risk of developing cancer. (3) Perceived efficacy of screening/benefits: perceived positive components or consequences of undertaking and completing the behavior.(4) Social influence: role of social norms and interaction with members of one's social network. (5) Intention: expressed intent to undertake a specific behavior. (6) Worries or fears/barriers: perceived negative components or consequences of undertaking and completing the behavior.
Perceived severity of disease was not included because it showed low internal consistency in previous literature5,32 and Janz and Becker33 suggested that perceived severity may be of lesser importance in predicting preventive health behaviors. A total of 30 items were measured using a four-point Likert-type format from “strongly agree” to “strongly disagree.”
Data collected via the screening survey and telephone survey were evaluated as predictor variables or as potential confounding variables. The outcome variables for these analyses were based on participation and defined as initial compliant if the subject has come in for at least one screening visit and initial noncompliant if he or she has never come in. Data were limited to general demographics and those who completed the cross-sectional survey for this outcome.
An individual's composite repeated screening adherence was measured using a calculated variable based on a person's time in months between screening sessions and their screening cycle (ie, persons on semiannual screening cycles should produce a higher number of screening sessions than persons on annual screening cycles). The measure reflected the average observed to expected months between screens (AOEM). The AOEM was dichotomized as compliant if AOEM ≤2.0 and noncompliant if AOEM >2.0. Because the screening program had been operating for more than 13 years, perfect adherence (AOEM = 1) was not necessarily considered feasible. Therefore, a cutoff of 2.0 was considered reasonable. A visual inspection of each record compared with the individuals’ calculated AOEM revealed that 2.0 was an appropriate cutoff to measure someone's overall adherence rate. This also helped to adjust for the strict criteria allowing one month for a subject to repeat a result.
The assessment of construct validity of the scales from the behavioral health survey was evaluated using factor analysis. The a priori assignment of items to constructs as defined by Vernon et al27 were evaluated by factor analysis using Stata Statistical Software (StataCorp LP, College Station, Texas).34 A standard varimax rotation was used and internal consistency reliability was estimated by Cronbach coefficient α. Generally, a coefficient of 0.70 or greater is used to identify scales with adequate internal consistency.
Initially, univariate analyses of adherence outcomes were conducted. Univariate associations between adherence and categorical independent variables were analyzed using a chi-squared test. A Fisher exact test was used when the expected value of any cell was less than five. The Wilcoxon ranked sum test was used to compare medians for data that were not normally distributed. All analyses were evaluated using a two-sided test with α ≤0.05.
Logistic regression analysis was be used to describe the relationship of the independent variables to the dichotomous dependent variable representing adherence.35 Separate models were evaluated to assess initial compliance or repeated adherence as the outcomes.
Variables associated with adherence at a P value of <0.20 at the univariate level were included in the first step of the multivariable logistic regression model. The model was fitted using stepwise logistic regression until only variables with a P value of ≤0.05 were included.
As shown in Table 1, 157 subjects completed the behavioral health survey (88% of persons contacted, 61% of eligible persons). One hundred forty-five (59.6%) of the males and 12 (75%) of the females completed the survey.
Table 2 shows the characteristics of survey respondents compared with those lost to follow-up, refusals, and unreturned mailed surveys. The distribution of the persons not contacted was similar to that of the surveyed population. There was a trend toward a higher percentage of females and higher education levels completing the survey, although these results were not statistically significant.
The perceived susceptibility items “believe chance of developing cancer is high” and “I think it is very likely that I will develop cancer is high” had a scale reliability coefficient of 0.62, which shows a reasonable internal consistency reliability. Internal consistency reliability was high (0.71) for the eight items in the salience and coherence construct. The questions “I believe that if I had a normal screening test result I wouldn't have to worry” and “I believe that when bladder cancer is found early it can be cured” reflecting the benefits of screening had a low reliability coefficient of 0.25. The social influence items “I want to do what members of my immediate family think I should do” and “Members of my immediate family think I should go through bladder cancer screening” also had a low reliability coefficient (0.25). Intention to be screened items “I do not intend to continue bladder cancer screening” and “I intend to continue to undergo bladder cancer screening” showed a surprisingly low reliability (0.29). It is possible that the reverse coding caused some confusion. Finally, the items for barriers to screening, questions “I am afraid of having an abnormal screening test result,” “I am worried that screening will show that I have bladder cancer,” and “I am bothered by the possibility that bladder cancer screening might be embarrassing” had a reliability coefficient of 0.56. The question “I am worried that screening will show that I have bladder cancer” had the lowest coefficient and when removed, items “I am afraid of having an abnormal screening test” and “I am bothered by the possibility that bladder cancer screening might be embarrassing” yielded a reliability of 0.74.
Because of the limited available data on those never screened, initial compliance measured as ever compared with never screened was evaluated univariately and based solely on information from the Behavioral Health Survey. One hundred forty-nine (95%) subjects who had ever been screened completed the Behavioral Health Survey, compared with eight (5%) subjects who had never been screened.
The median score for each question was compared between groups using the Wilcoxon ranked sum test. Frequency distributions of responses with statistically significant predictors are shown in Table 3. The question “I believe bladder cancer screening can help to protect my health” is a measure of salience and coherence. A larger percentage of the compliant than noncompliant group reported “strongly agree.” The question “I am afraid of having an abnormal screening test result” is considered a measure of barriers to screening. A larger percentage of the noncompliant group strongly disagreed to this statement than that of the compliant group. The social influence question “Members of my immediate family think I should go though bladder cancer screening” reflected stronger agreement in the compliant group than in the noncompliant group. The most striking difference pertained to the question “Finding transportation to the screening clinic is an easy thing to do.” The majority of the noncompliant group reported strong disagreement, whereas the majority of the compliant group reported strong agreement.
The composite scores for social influence and barriers were statistically significantly different between groups. The median response score to the social influence composite was “4” for the compliant group and “3.5” for the noncompliant group suggesting a stronger sense of social influence among compliers. The median composite barriers score was “3.3” for the compliant group compared with “4” for the noncompliant group indicating that barriers to screening was affecting the noncompliers.
Table 4 shows statistically significant health survey predictors of adherence based on results generated by univariate logistic regression. Subjects with diabetes are more likely to be compliant (odds ratio [OR] = 3.75) as are subjects with prostate problems (OR = 3.64). Those who reportedly did not consume alcoholic beverages were almost twice as likely to be compliant. Persons with at least one comorbidity (OR = 2.20), or at least one positive screen, were more likely to be compliant (OR = 2.37). Subjects whose age at last screening session was between 46 and 55 years were twice as likely to be screened than those aged between 36 and 45 years old (OR = 2.25) as were those in the 56- to 65-year age group, although that difference was not statistically significant (OR = 2.01).
Multivariate Analysis for Adherence by AOEM
Results of stepwise logistic regression analyses are displayed in Table 5. Two separate models were fit: “Model I” used the composite scores for salience and coherence, benefits, social influence, intent, barriers, and benefits and “Model II” used each individual question as a covariate. Salience and coherence (OR 6.2), perceived susceptibility (OR 1.5), does not consume alcohol (OR 2.8), and reporting at least one comorbidity (OR 1.8) were all positively associated with adherence as shown for Model I.
Using individual survey questions, (Model II) produced similar results. Persons who reportedly did not consume alcohol (OR 2.4), question 4 “arranging my schedule is easy” (OR 7.3), and question 16 “I am worried that screening will show that I have bladder cancer” (OR 2.2) were all positively associated with high adherence. Question 4 relates to salience and coherence, whereas question 16 represents a barrier to screening. Individual questions pertaining to perceived susceptibility were not significant in this model.
DISCUSSION AND CONCLUSIONS
Occupationally exposed subjects eligible to participate in cancer screening programs could benefit greatly from this research. Identification of significant factors (or combination of factors) that increase (or decrease) the subjects’ probability of participation in a cancer screening program can be used to develop interventions to increase adherence with screening, thus reducing morbidity and mortality of the associated disease. For example, according to the Health Belief Model,36 at least a moderate level of susceptibility and seriousness is necessary for behavior to follow. If susceptibility and seriousness for a certain disease could be measured, evaluations and interventions for belief changes about cancer screening might be instituted. Likewise, the HBM concept of benefit could be evaluated and interventions developed to help persons understand personal benefits for undergoing cancer screening. Barriers to action could be evaluated in terms of resources available, and strategies for handling barriers could be developed.37
The survey response rates were reasonable with 88% of those contacted and 61% of those eligible completing the survey. Eighty-one (31%) of those eligible were lost to follow-up. This was not a surprising result because of the general composition of the cohort. Our experience with the screening program and familiarity with the cohort over the years has shown that some of the workers live a fairly transient lifestyle.
Internal consistency reliability was measured for each of the six a priori constructs. Salience and coherence had the highest internal consistency reliability; therefore, it is considered a valid construct for interpretation as was perceived susceptibility. The items included in the barriers construct were also fairly reliable. The constructs comprising benefits, social influence, and intention to be screened had low internal consistency reliability. These constructs included only two items each and in some cases, the reverse coding may have caused some confusion for the survey respondents. Individual items may be better predictors of adherence for these measures than the composite scores.
Initially, “barriers to screening” was associated with adherence. Although those who were never screened reported a fear of abnormal results, they also were less willing to believe that screening would protect their health. Barriers were probably the most significant reason for workers never being screened or dropping out at the beginning of the program. They also reported difficulty in finding transportation. Nevertheless, it is important to note that the survey results for those never screened should be interpreted with caution because of the small number of respondents. Barriers to screening were also strongly associated with compliance for breast cancer screening and for colorectal cancer screening with fecal occult blood testing (FOBT).38–41
The social influence item, “subjects whose family members supported screening,” was significantly associated with enlisting in the program. Participants who did not consume alcohol, reported at least one comorbidity, were worried that screening would show bladder cancer, and felt that arranging their schedules was easy were more likely to be continuously adherent. Patients with comorbidities were more likely to see their physicians about their conditions, and previous studies on adherence to continuous colorectal cancer screening have shown a strong relation to preventive health practices, regular physician and dental visits, and screening adherence.42 When evaluating the individual survey items multivariately, subjects who think that arranging their schedule for screening is easy, are worried that screening will show cancer, do not consume alcohol, and have at least one bladder symptom were more likely to be highly adherent.
The strong association between salience and coherence and many of the adherence outcomes was the most noticeable and is supported by the FOBT screening literature as well.4 In this analysis, salience and coherence refer to perceptions about the technical effectiveness, practical convenience, and potential benefit of the screening examination. Individual items such as “arranging my schedule is easy,” “the benefits of screening outweigh the difficulty,” “bladder cancer screening is important,” “I believe that bladder cancer screening can help to protect my health,” disagreeing that “finding time is difficult,” and “bladder cancer screening is easy” were included in the scale and independently statistically significant as well. Once the participants affected by barriers and benefits dropped out of screening, salience and coherence became the strongest predictor. The importance of this variable suggests the centrality of situational decision-making in preventive health behavior.43 Ball44 described situational decision-making as considering “the sum of all recognized information so that he can engage in self-determined lines of action and interaction.”
In summary, the results of this study suggest that the factors affecting adherence with bladder cancer screening change for initial participation and according to the number of screens a participant completes (or is eligible for) and for continued adherence. Therefore, to enhance overall adherence, specific strategies should be implemented when initiating a screening program and revised accordingly over time. Barriers should be addressed at the initiation of a bladder cancer screening program. Providing additional education to the workers and stressing the success of treating bladder cancer if detected early could be an effective strategy early on. In this screening program, persons who were never screened were more likely to report transportation problems. Because barriers and benefits were the strongest predictors in early rounds of screening, stressing the benefits of early detection may also increase initial compliance. Working with families and social networks is very important when initiating a new screening program. Educating families and the community at large could have a powerful impact upon a person's willingness to initiate screening.
Salience and coherence is a very strong predictor of adherence in this screening program, especially for continuous screening. Making the screening session convenient and efficient, stressing the benefits of screening, and reinforcing the importance may help to keep the subjects returning.
A limitation of this study is that only 61% of the eligible subjects completed the behavioral health survey. Previous studies suggest that persons who complete survey instruments as part of research studies tend to be those with healthier lifestyles.45,46 Nevertheless, only 5.6% of these were refusals and 31% were lost to follow-up. Table 2 compares demographics of those who completed the survey with that of those who refused, were lost to follow-up, or did not return a mail survey (no available phone number). There were no striking differences between the groups.
Another limitation is that the Behavioral Health Survey is a cross-sectional survey that does not reflect participants’ views over time. It is quite possible that they would have answered differently in the beginning of the program. Also, only the most current general health survey data were used. The most current date of the survey is not consistent across participants. However, because this study was not planned at the beginning of the program, these were the best data we could collect. In addition, several different outcome measures were analyzed to attempt to capture as much information as possible. Future research could implement a simplified behavioral health survey at each screening visit to further assess screening behavior over time.