Individual- and provider-level factors associated with colorectal cancer screening in accordance with guideline recommendation: a community-level perspective across varying levels of risk

Australia’s National Bowel Cancer Screening Program

In Australia, Medicare administers the National Bowel Cancer Screening Program (NBCSP) Register [8]. Medicare selects eligible participants from either the Medicare enrolment records or the Department of Veterans’ Affairs [8]. The NBCSP Register is responsible for mailing of screening invitations and FOBT kits, the recording of participants’ details and the issuing of reminder letters [8]. As part of the NBCSP, participants are encouraged to nominate their usual primary care provider on their participants details form; however, this is not compulsory [8]. The recently re-funded NBCSP offers persons turning 50, 55 and 65 years of age via mail a one-off immunochemical Faecal Occult Blood Test (FOBT) screening invitation [9]. It is important to note that the limited format of the NBCSP (restriction of screening to persons in selected age brackets across the at-risk population), [9] is not consistent with the NHMRC recommendation of biennial FOBT screening of all Australians in the at-risk population [7]. A recent study examining the costs and outcomes from full (rather than limited) implementation of a biennial FOBT screening for adults in the at-risk population (50–74 years) identified a likely mortality reduction of 25% and saving of 500 deaths per year [10].

Screening participation in Australia

Participation rates in Australia’s NBCSP appear to have reached a plateau since the pilot program’s introduction [9, 11–13]. The pilot program received a response rate of 45%, with the roll-out of the NBCSP (with screening offered to persons 50 and 65 years of age) in 2006 receiving a slightly lower rate of 41% [12]. The rate of participation only marginally increased in 2008 to 42.9% when the program widened to offer screening to persons aged 55 years of age [11]. Most recent estimates suggest a similar rate of low participation (40.1%) [9]. The experience in other countries with national screening programs offering repeated FOBT screening to a wider section of the at-risk population suggests that much higher screening rates are achievable. For example, FOBT screening rates in the United Kingdom and Finnish screening programs are currently 52% and 71% respectively [14, 15]. In Australia, previous community- based evaluations have also indicated low rates of CRC screening, with 5 to 20% of individuals ever undertaking FOBT [16–18]. The most recent assessment in New South Wales (NSW) indicated that 18% of persons aged over 50 years had undertaken FOBT in the previous five years [19]. In relation to colonoscopy screening in Australia, two population-based assessments in NSW have suggested under screening among persons at elevated levels of risk [16, 19].

Given low rates of CRC screening, it is important to identify factors which may influence screening uptake. Previous studies have indicated that the following factors influence CRC screening behaviour: socio-demographic characteristics (e.g. older age, higher education, higher income); lifestyle factors (e.g. smoking history, chronic disease); family history (e.g. personal or family history of CRC); awareness (e.g. knowledge of CRC and perceived risk of developing CRC) and health care utilisation (e.g. usual source of care, number of GP visits, and health care coverage) [20–26]. In the Australian context, a small number of studies have explored determinants of CRC screening uptake [17–19, 27, 28] with little known about predictors of screening behaviour for persons at varying levels of risk [16]. Since the introduction of the NBCSP in 2006, no community- or population-based assessments in Australia have been conducted. Recent evidence pertaining to FOBT screening uptake and inequalities in participation have been confined to annual NBCSP monitoring reports, which report screening rates among a limited section of the at-risk population (persons 50, 55 and 65 years of age) [9, 11–13]. Further, since the NHMRC guidelines’ implementation in 1999, only one study has assessed the predictors of risk-appropriate screening for persons at each level of risk in accordance with guideline-recommendation [16]. The identification of factors associated with risk-appropriate screening is of critical importance for future planning and implementation of tailored CRC screening programs.

This study aimed to assess among a large community-based cohort of at-risk persons (aged 56–88 years), the factors associated with: (1) ever receiving any CRC testing; (2) receiving screening in accordance with screening guidelines; and (3) recent use of colonoscopy screening.

Study population

The Hunter Community Study (HCS) is a longitudinal community cohort aged 55–85 years at baseline in the Hunter Region, New South Wales (NSW), Australia. The sampled population, Hunter Community Study participants are at-risk of developing CRC with clinical practice guidelines recommending all persons commence CRC screening from the age of 50 years [7]. HCS participants were randomly selected from the NSW State electoral roll between December 2004 and December 2007. A modified Dillman recruiting strategy [29] was used for this study, whereby two letters of introduction and an invitation to participate were posted to selected persons. Persons not speaking English and living in a residential aged-care facility were deemed ineligible. Following consent to participate, individuals were asked to complete two self-reported questionnaires and were asked to return these when they attended the HCS data collection centre, at which time a series of clinical measures were obtained. For further details on the HCS cohort and an exhaustive list of measures obtained from HCS participants, see the cohort profile [30]. The HCS cohort profile reflects state and national profiles for gender and marital status but is slightly younger in age [30]. For the current study, a randomly selected sub-sample of the HCS (n = 1592) aged 56–88 years (reflective of age since HCS data collection point) were mailed a questionnaire during November, 2009. The sample size of 1592 was chosen to investigate delay in seeking medical advice for primary symptoms of CRC in additional studies using the HCS cohort [31, 32] in order to achieve a power of 80%, so that the proportion of individuals who delay seeking medical advice for symptoms or treatment for CRC will be estimated with 95% confidence intervals.

Questionnaire

Predictors of CRC screening behaviour

Based on existing literature, the following items selected from the HCS databank, assessed at HCS baseline were investigated: Socio-demographic characteristics: age, gender, education, marital status, country of birth, household income, retirement, private health insurance status, tobacco or alcohol use; Clinical characteristics: number of general practice visits over the past 12 months, previous cancer diagnosis (excluding CRC), body mass index, and presence of a chronic health-condition; and Psychosocial characteristics: physical health (SF-36) and mental health (K-10 Kessler Scale). Predictors ascertained at survey completion included: risk category, discussion of family history of CRC with doctor and notification of any “increased risk” (Never discussed/ discussed and informed of possible “increased risk”/ discussed and not informed of any possible “increased risk”), and ever received screening advice from doctor (Yes/No).

Statistical analysis

Descriptive statistics including frequencies, percentages, means, and standard deviations were used to describe the characteristics of participants. Chi-squared testing was used to assess any responder bias in relation to age, gender, and level of education. “Ever received CRC testing” was calculated by combining the proportion of respondents indicating “Yes” to undertaking any of the following CRC tests (i.e. FOBT, sigmoidoscopy or colonoscopy) divided by the total number of respondents. Respondents who had not provided any information on any of these CRC tests were excluded from analysis. The proportion of respondents screened in accordance with screening guidelines (Yes/ No) was assessed by risk category [7] as follows: “at or slightly above average risk” (FOBT within two years or sigmoidoscopy preferably flexible within five years) and “moderately increased risk/ potentially high risk” (colonoscopy within 5 years). Self reported family history of CRC was used to allocate respondents to a risk category in accordance with screening guidelines. Respondents with missing values relating to type or number of relative(s) diagnosed or age at their diagnosis were excluded from analysis. Multiple logistic regression modelling was used to assess the association between socio-demographic, clinical and psychosocial characteristics and each study outcome: ever received CRC test, screened in accordance with guidelines for level of risk (“at or slightly above average risk” and “moderately/ potentially increased risk”), and recent colonoscopy screening (within previous five years). Variables with a p value <.25 following simple logistic regression analysis (see Additional file 2) were entered into a stepwise multiple logistic regression analysis. Forward and Backward stepwise regression was used on complete cases so that the likelihood ratio test could be used to determine the best fit from one iteration to the next. Both the forward selection and backward elimination regression analysis produced equivalent results. Results from the backward stepwise elimination model were reported. Variables that met the significance cut point (p value < .05) were included in the final model. Data were analysed using STATA 11 (STATA, Texas, USA).

Characteristics of the sample

Of the 1592 mailed surveys, 1117 respondents completed and returned a survey (response rate = 70%). Respondents previously diagnosed with CRC (n = 24) or reporting they had undergone major abdominal surgery (n = 8) were excluded from analysis, leaving a total sample of 1085 participants with data. Responder bias was assessed on the following key characteristics: gender, age (56–64, 65–74, 75–88), and education (secondary schooling completed, secondary schooling not competed, trade qualification or TAFE, university or other tertiary education and other). Pearson’s χ2 test found no significant differences relating to gender (χ2 = .09; df =1 ; p = .76 ) or education (χ2 = 8.75; df = 4; p = .07). However, for age, respondents were significantly more likely to be of a younger age compared to non-respondents (χ2 = 11.12; df = 2; p < .01).

Inclusion and exclusion for CRC screening

Ever received CRC testing

CRC screening in accordance with national screening guidelines

Recent colonoscopy screening irrespective of level of risk

Factors associated with CRC screening behaviour

Limitations

It should be acknowledged that responder bias place some limitations on the interpretation of findings and their generalisability. For the most part, the HCS is a representative sample of the Hunter Region and NSW in comparison to Census data on a number of demographic characteristics, with the exception of age [30]. Analysis of responders’ bias identified that responders were more likely to be younger than non-responders from the HCS cohort. In addition, persons living in residential care facilities and those from non-English speaking background were excluded in this study. Previous literature has indicated that non-English speaking persons are significantly less likely to undertake CRC screening [11–13, 19]. Future population based CRC screening programmes could benefit from strategies to improve CRC screening in non-English speaking persons and migrant groups [19]. For respondents undertaking FOBT, type of test undertaken (guaiac or immunochemical) and the outcome of testing (negative/ positive) was not assessed, and this should be considered when interpreting the rate of colonoscopy screening (previous five years), as it may be an independent predictor for seeking a colonoscopy.

While it must be acknowledged that self-reported screening behaviour may have some recall bias, it has been shown to have reasonable agreement with physician report and relatively high levels of sensitivity and specificity [54, 55]. Although two new studies and a recent meta-analysis have indicated high levels of sensitivity and specificity for self-reported screening behaviour, [56–58] it is important to recognise that accuracy of self-reported screening varies across patient characteristics and test modalities [20]. A meta-analysis of validation studies on self-reported CRC cancer screening use in the United States found that self-report versus documented history of screening had a high level of specificity (endoscopy .90 and FOBT .78) and sensitivity (endoscopy .82 and FOBT .79) [56].

To enhance recall, the current study adopted descriptions of CRC tests to increase the accurate recall of CRC testing. Self-reported family history of CRC was used to assign level of risk. Previous studies have indicated self-reported family history of CRC has relatively high levels of sensitivity and specificity for reporting of affected first-degree relatives [59–61]. However, the level of accuracy is substantially reduced for affected second-degree relatives [59–61]. The accuracy of reporting of age at diagnosis in CRC- affected relatives has not been identified in the literature. Further, the extent to which misleading family history information reorients risk classification for relatives following verification is largely unknown [61]. Therefore, it is possible that the classification of risk may be inaccurate for a small proportion of study respondents given the limitation of no objective verification of self reported family history. For respondents at “moderately increased/ potentially high risk”, level of risk may have been overestimated for a small minority of respondents. For respondents indicating that both a first-degree relative and second-degree relative were diagnosed with CRC, an assumption was made that both relatives were diagnosed on the same side of the family. This assumption may have over-estimated the level of risk for some respondents.

Finally, the current study is subject to limitations associated with stepwise regression analysis including: parameter estimation, error rate estimation and reliance on a single best model [62–65]. Conducting stepwise regression models with a large number of independent variables and a small sample size, may over-estimate model fit [63, 64]. However, this limitation is difficult to overcome when applying statistical modelling techniques to identify predictors associated with rare occurrences in the population i.e. persons at increased familial risk of disease. It is conceivable that future large scale population-based studies examining CRC screening behaviour across varying levels of familial risk will assist in further defining the predictors of risk-appropriate CRC screening.