Population-based cohort development in Alberta, Canada: A feasibility study

Headnote

Abstract

In a climate of increasing privacy concerns, the feasibility of establishing new cohorts to examine chronic disease etiology has been debated. Our primary aim was to ascertain the feasibility of enrolling a geographically dispersed, population-based cohort in Alberta. We also examined whether enrolees would grant access to provincial health care utilization data and consider providing blood for future analysis. Using random digit dialling, 22,652 men and women aged 35 to 69 years, without diagnosed cancer, were recruited. Of these, 52.4 percent (N = 11,865) enrolled; 84 percent of Alberta communities were represented. Approximately 97 percent of enrolees consented to linkage with health care data, and 91 percent indicated willingness to consider future blood sampling. Comparisons between the cohort and the Canadian Community Health Survey (Cycle 1.1) for Alberta demonstrated similarities in marital status and income. However, the cohort had a smaller proportion who had not finished high school, a greater proportion of non-smokers and a higher prevalence of obesity. These findings indicate that establishment of a geographically dispersed cohort is feasible in the Canadian context, and that data linkage and biomarker studies may be viable.

Key words: Alberta, cohort studies, feasibility studies, questionnaires

Introduction

Prospective cohort studies are potentially powerful tools to examine chronic disease etiology. Because they collect exposure information prior to disease diagnosis, they are free from the potential differential biases that may occur recalling this information when cases are compared with controls. Further, because the exposure information is collected relatively contemporaneously, rather than having subjects recall distant past exposures, there is potential for increased accuracy of reporting. The high value of these research resources has resulted in the construction of a number of cohorts internationally. It has also resulted in national discussion of the potential for development of a large Canadian cohort for the study of chronic disease.

While the value of such cohorts is acknowledged, their drawbacks are equally well known. Ongoing cost, of course, is one issue that would have to be addressed in the consideration of developing a full-scale national cohort. In addition, although the cohort design minimizes recall bias, other biases are now known to exist: dietary assessment is hampered to some degree by mis-reporting,1 and there is controversy concerning the impacts of socially desirable responding on the validity of data obtained from self-administered questionnaires.25 Similarly, loss to follow-up over time may also contribute to bias.6 Moreover, quantifying the precise impact of such biases in cohort studies remains challenging. Furthermore, because cohort studies need a large sample size and extensive exposure data collection to have sufficient study power, their construction and maintenance is expensive. For etiological hypotheses, investigators must wait for a sufficient number of cases to be identified before analysis is worthwhile, delaying results and adding to the expense. This time delay of results makes cohort construction an unattractive research endeavour for investigators who live under the "publish or perish" paradigm.

In 1999, the population health research group at the Alberta Cancer Board (ACB) began to explore the feasibility of the construction of such a cohort in Canada. There were two underlying themes in the development of the cohort concept: the creation of a research legacy, identified by our team as a population health laboratory, and the maximization of both short-term and longterm potential benefits.

The research legacy term refers to the concept of developing a rich data resource that could be used by current researchers, but which would be of even more value to researchers who may enter the field in several years, when the cohort is "maturing" and disease outcomes become frequent. This resource would be more valuable with increased depth and volume of behavioural, biochemical, socio-demographic and environmental data available on each participant. Thus, collection of exposure information with detailed, validated tools, re-collection of data at reasonable time intervals, individual biological specimens and the potential to link data with complete health care utilization files were seen to be key components of such a population laboratory. While it was envisioned that the primary focus would be research into cancer etiology, many of the risk factors examined are also potentially important in the etiology of other chronic diseases, thereby ensuring that the cohort would also be valuable for research in other areas.

The long-term benefit of the cohort would be linked most closely to the complete and accurate collection of the information noted above. However, we discussed several ways in which shorter term outcomes could also be of value. Clearly, a cohort that was representative of the general population, rather than one composed of population sub-groups defined by occupation or educational status, or people recruited as volunteers, could provide more insight into general trends in cancer-related prevention and screening behaviours. While we were interested in examining the degree to which a population-based cohort could be constructed, we projected that, even if the cohort was somewhat unrepresentative with respect to demographic or behavioural characteristics, within-cohort comparisons of predictors of behaviours or behaviour change could still produce valuable insights into cancer control. In addition, such a cohort could provide an opportunity for the evaluation of "natural policy experiments" that might occur over its longitudinal course. If, for example, smoke-free public spaces were mandated in some communities and not others over the course of the study, a baseline group would already be in place with recorded smoking behaviours and other characteristics prior to the policy change. By using the rich data available to distinguish the characteristics of smokers who changed their behaviours from those who did not, we could provide excellent analysis of the predictors for success of such policies.

In this paper, we address three questions that relate to the legacy potential and the probability of collecting reasonably geographically representative data from "average" individuals. These questions are,

1. Could we enrol a cohort of randomly selected individuals across a dispersed geographic population which could adequately represent the distribution of demographics and health behaviours within a province?

2. What proportion of these individuals, in a world of increasing privacy concerns, would be prepared to give access to health care utilization files for further research?

3. What proportion of these individuals would be willing to consider providing a blood sample for storage and future analysis?

This paper reports on the findings related to these questions.

Methods

The target population for the feasibility study was men and women aged 35 to 69 years. Other enrolment criteria were as follows: 1) no known history of cancer other than non-melanoma skin cancer; 2) plans to reside in Alberta for at least one year; and 3) English speaking, to allow for collection of self-report data. Approvals to conduct the feasibility studies were obtained from the ACB and University of Calgary ethics review boards.

Subject selection and enrolment

A two-stage sampling design was used to identify eligible individuals. The first stage used a random digit dial (RDD) procedure.7 Since 97 percent of Alberta households in the year 2000 had at least one telephone line,8 a telephone-based sampling method ensured that almost all households were included in the theoretical sampling frame. The first stage of sampling selected households in the 17 regional health authorities (RHAs) extant in Alberta in 2000, and the second stage selected one eligible adult within each household. A household was defined as one or more persons, related or otherwise, who occupy the same private or collective dwelling.9 The sampling and RDD were done by an experienced social research laboratory at the University of Alberta.10

The recruitment for the feasibility component was done in four waves, in order to evaluate and, if necessary, change procedures as a result of early experience in the study. These four stages are referred to as RDDl through RDD4, respectively.

First-stage random selection

Standardized procedures were used to ensure methodological and ethical integrity of the RDD approach.7 An electronic database of randomly generated telephone numbers, mapped to RHAs, was used for calling purposes. Trained interviewers, working with a computer assisted telephone interviewing (CATI) system and standard script, called selected households to screen for eligible individuals who would be willing to consider enrolment into the Alberta Cohort Study (The Tomorrow Project®).

In order to maximize the likelihood of contacting residents in the selected households, calls for RDDl were made up to 20 times over a variety of times and days of the week before abandoning the number. Because of diminishing returns with subsequent calls, this total was reduced to 15 in RDD2 and to 12 calls in each of RDDs 3 and 4. Disproportionate sampling was done to ensure a sufficient number of participants from rural and remote regions.

Second-stage random selection

The first adult householder answering the telephone was given a description of the study purpose, eligibility criteria, conditions for participation (i.e., voluntary; long term with periodic follow-ups and repeated data requests), and examples of information asked on baseline questionnaires. In households with more than one potential study participant, the person with the most recent birthday was selected for possible enrolment, to reduce selection bias towards groups more likely to be available to answer the telephone.7

As part of our feasibility exploration, a second household member of the opposite sex was selected for possible enrolment when the first respondent was eligible and interested in considering cohort enrolment. This approach was attempted in 2527 households (RDDl) to assess the impact on rate of accrual.

At the conclusion of the RDD process, all telephone numbers/households were assigned one of the following codes: "recruited" (target respondent was eligible and interested in considering study enrolment); "ineligible"; "undetermined eligibility" (efforts at contacting the target householder were unsuccessful and/or a screening interview was not completed); or "refused".

Subject enrolment and retention

A self-administered baseline health and lifestyle questionnaire (HLQ) and a detailed consent form were sent by regular mail to individuals interested in study enrolment. Participants were classified as enrolled if they completed and returned the HLQ and the consent form. Approximately three months after enrolment into the study, two additional questionnaires concerning habitual diet and past year physical activity were mailed to participants.

As part of their written consent, participants were asked for permission for data linkage with the Alberta Cancer Registry. They were also asked to voluntarily provide their Alberta Personal Health Number (PHN) and signed authorization allowing the Alberta Cohort Study to request health services utilization data held by Alberta Health and Wellness. Specifically, subjects were informed that the study would seek data from Alberta Health concerning types of health care services accessed (defined by "billing codes"), frequency of use of such services and whether services were provided in doctors' offices or hospitals. Subjects were invited to consent to periodic linkages for the duration of their participation in the study. Individuals were allowed to participate in the overall study even if they denied access to Alberta Health or to Alberta Cancer Registry data.

Participants were also asked if they would be willing to consider providing a blood sample for study purposes, should they be asked for one in the future. They were also informed that, if such a request were made, a full explanation of the blood collection's purpose would be provided and that further written consent would be required before any sample could be collected.

The final page of the consent form provided subjects with the study's contact details, and encouraged subjects to use any of several contact methods if they moved away from the address from which they were originally recruited. Specifically, we provided a "change of address" form on the study Web site (www.thetomorrowproject. org), as well as toll-free and collect-call telephone numbers, in order to ensure that subjects who moved out of province or out of Canada had access to a variety of free and convenient methods of keeping in contact. In addition, subjects were asked to provide their cellular telephone number and e-mail address (if applicable), as well as contact details for two people outside their household. These contacts would be used in the event that the subject could not be contacted using any other means. Furthermore, regular contact continues to be maintained with subjects by means of a biannual newsletter, which also serves to provide feedback on study progress and news.

Baseline data collection

Baseline information about lifestyle-related risk factors and exposures was collected using three self-administered, mailed questionnaires. The instruments were selected on the basis of 1) relevance to factors with potential high attributable risk for cancer and other chronic diseases; 2) the suitability/adaptability of the measure for self-administered surveys; and one of the following: 1) previous use in established epidemiologic studies; and/or 2) published data describing the measure's psychometric properties.

Health and Lifestyle Questionnaire (HLQ)

The HLQ is a composite of existing items used in other large studies relating to personal health and reproductive history, family history, psychosocial factors, anthropometric measures, use of cancer screening services, smoking behaviour, sun exposure and socio-demographic characteristics. Some items were developed for the Alberta Cohort Study if other sources were not available.

Items concerning personal health history, male and female reproductive information, and family history of chronic illness and longevity were adapted from questions used in the Prostate, Lung, Colorectal and Ovarian (PLCO) Screening Trial," the Women's Health Initiative (WHI) Study12 and the 2000/01 Canadian Community Health Survey (CCHS; cycle l.l).13

Items concerning Pap tests, mammograms, clinical breast examination, breast self-examination and PSA tests originated with the CCHS.13 Items about colorectal screening with digital rectal examination, sigmoidoscopy/colonoscopy and stool collection for occult blood testing were adapted from the CCHS and the California Health Interview Survey 2001.14

Questions about tobacco exposure were based on a recommended set of measures for monitoring tobacco use in Canada as developed through the Canadian Workshop on Data for Monitoring Tobacco Use.15 Sun exposure was measured using selected items recommended by the Canadian National Workshop on Measurement of Sun-Related Behaviours for monitoring sun exposure and protective behaviours.16

Social support was measured using questions from the Medical Outcomes Study (MOS)17 questionnaire. Items proposed for the CCHS 2000/2001 were used for measuring stress. Spirituality was measured with three items taken from the CCHS and one created for the baseline survey.

Subjects were also provided with detailed instructions and a 183 cm (72 inch) tape measure for obtaining accurate height, buttock and waist measures using a selfadministered method that had been tested for reliability and validity.18 Instructions were also given for recording body weight using a scale accessible to the respondent. The HLQ comprised 32 pages and took an estimated 40 minutes to complete.

Diet History Questionnaire (DHQ)

The DHQ is a cognitive-based food frequency questionnaire (FFQ) developed by the US National Cancer Institute (NCI) as a tool for assessing diet over the preceding 12-month interval.19 There is evidence that the DHQ was comparable to, or superior to other FFQs that have been used in other large cohort studies.20,21 The instrument, which takes about 60 minutes to complete, has questions about 124 food items and dietary supplements, with additional embedded questions within 44 of these items. In collaboration with the NCI, changes were made to the questionnaire and nutrient database to account for differences between the US and Canada in food availability, brand names, nutrient composition and fortification practices.22

Past Year Total Physical Activity Questionnaire (PYTPAQ)

The PYTPAQ was based on a questionnaire developed to measure lifetime total physical activity (LTPAQ). The LTPAQ is an interviewer -administered questionnaire that provides a reliable lifetime measure of occupational, household and recreational activities from childhood to present.23 Frequency, duration and intensity of all types of activity (i.e., occupational, household, recreational) are recorded to yield measures of expended energy within each type of activity area and an overall measure of the energy cost of physical activity. To produce the PYTPAQ, the LTPAQ was adapted for s elf- administration and the reference period was changed from lifetime to the 12-month period preceding questionnaire completion. A separate study to evaluate the reliability and validity of the PYTPAQ was also conducted.24

All questionnaires are available on request.

Data handling and analysis

TeleForm® software (TeleForm V8.1; Verity, Sunnyvale CA USA) was used for automated optical scanning and data capture of HLQ and DHQ data, while PYTPAQ data were entered using Blaise® software (Westat, Rockville, MD USA). Routine quality checks were performed before and after data entry, and telephone follow-up was used to clarify ambiguous data. HLQ, DHQ and PYTPAQ data were linked by subject identification number, and no subject identifiers were stored with questionnaire data. To ensure security, all electronic data were stored on servers with limited access, and all files were password protected and backed up on a daily basis.

Data cleaning and analyses were done using the SAS® statistical software program (SAS V9 2003; SAS Institute Ine, Cary NC USA).

Results

Recruitment and enrolment

The four waves of telephone calling were carried out between October 2000 and June 2002, resulting in 77,327 randomly selected households being contacted. A screening interview to identify eligible residents was not completed in 38.9 percent of households; in most of these cases, the person answering the telephone could not be engaged in the interview. A screening interview was completed in 61.1 percent of selected households, and an eligible individual was recruited for possible study enrolment (i.e. willing to consider study participation) in 47.9 percent of these households. The remainder were ineligible and thus excluded for reasons of age outside the target range (89.4 percent); history of cancer (7.2 percent); expecting to move away from Alberta within the following year (2.9 percent); and, unable to understand and complete the study material in English (0.5 percent).

In 2,527 households in RDDl, we attempted to select a second participant of the opposite sex in households where a first person was successfully recruited. As a result, 711 subjects were recruited as "second in household". Of these, 384 (54 percent) enrolled in the cohort; this double recruitment strategy required, on average, two additional telephone calls to the household. The combined response from eligible first and second contacts was 56.7 percent compared to a response of 47 percent in households where only one person was recruited.

Of the 22,652 eligible individuals who were recruited, 52.4 percent (N= 11,865) enrolled in the cohort between February 25, 2001 and June 30, 2003. It is estimated that the enrolled sample represents about 32 percent of all potential participants; exact percentages cannot be given as we do not know the eligibility of those who did not complete the screening interview.

The enrolled sample of 11,865 represents about one per cent of the Alberta population aged 35 to 69 (based on the population estimate for 2002) and 84 percent of Alberta communities and municipalities are represented. Figure 1 shows regional study enrolment; enrolment outside the major metropolitan areas ranged from 44 percent to 58.6 percent and in the urban areas it was similar at 48.2 percent and 54.0 percent in the Edmonton and Calgary regions, respectively (Table 1). Non-urban participants were selectively overrepresented as planned.

Of those enrolled, approximately 88 percent returned completed DHQs and PYTPAQs.

Baseline characteristics of cohort participants

The cohort was made up of 4,907 men (41.4 percent), 6,956 women (58.6 percent) and two transgender individuals.

In order to examine whether or not the cohort was similar to the Alberta population, a comparison was made between the Alberta cohort respondents and Alberta respondents from the CCHS (cycle l.l)25 carried out between September 2000 and November 2001. The latter survey has a response rate of about 85.1 percent in Alberta,26 and is commonly used to reflect population-based estimates of health and behaviours.

Following exclusion of the transgender subjects and those recruited as "second in household", the cohort data were weighted to the CCHS population frequency estimate, stratified by sex, age and RHA of residence at the time of recruitment. As shown in Table 2, the cohort sample was comparable to the CCHS sample in terms of marital status and annual household income below and above $50,000; the median family income in Alberta in 2001 was $60,100.27 The proportion of the samples with post-secondary education was similar, but there were fewer individuals in the cohort with less than high school education.

Prior to comparing health behaviours, the cohort sample was further weighted by educational and income levels (Table 3). Even following this adjustment, the cohort group had more non-smokers than the CCHS group and a higher prevalence of obesity (body mass index > 30). In both groups, the majority of women had had at least one Pap smear. For women over 50, a greater proportion from the cohort reported having had at least one mammogram (94.4 percent versus 85.6 percent). Similarly, for men over 50, prostate specific antigen history was higher in the cohort group (54.0 percent versus 43.3 percent).

These comparative analyses were based on Statistics Canada's Canadian Community Health Survey, Cycle 1.1, Public Use Microdata File, which contains anonymized data collected in the year 2000/2001. All computations on these microdata were carried out by staff employed by the Division of Population Health and Information at the Alberta Cancer Board, and the responsibility for the use and interpretation of these data is entirely that of the authors.

Consent for health file linkage

As part of their written consent, participants agreed to periodic data linkages with the Alberta Cancer Registry to identify incident cases of cancer. The consent also specifically asked participants for authorization to allow the Alberta Cohort Study to request health services utilization data held by the provincial health ministry (Alberta Health and Wellness); if they agreed, they were asked to provide their PHN. The majority of men (95.8 percent) and women (98.1 percent) consented to this aspect of the study and provided their PHN (Table 4).

Willingness to participate in blood collection studies

A separate form included in the enrolment package asked participants to indicate their willingness to be contacted in the future to consider providing a blood sample for study purposes. Approximately 91 percent of men and women gave a positive response to this proposal (Table 4).

Discussion

One of our primary questions was whether Canadian individuals not affiliated with any particular profession, association or known registries would agree to be part of a long-term prospective study. While cohort studies of this type have been initiated in Europe, there was considerable question as to whether North Americans, in an environment of increasing concern over privacy in the 21st century, would be willing to participate. Our results have shown that enrolment of such a cohort is indeed possible, and that the response rate obtained (32 percent), although lower than would be desirable for simple crosssectional studies, is comparable to cohort studies elsewhere in the world, recruited in earlier time periods. For example, a Swedish population-based cohort reported 40 percent participation,28 the Utrecht EPIC study reported a 34.5 percent participation rate,29 and the German EPIC study reported enrolment of 22.7 percent in Potsdam and 38.3 percent in Heidelberg.30 Among single-sex cohorts, a national sample of Dutch women had a response rate of 35.5 percent,31 the Iowa Women's Health Study reported 42 percent enrolment,32 57.1 percent participation was reported in a women's cohort in Norway33 and 51.3 percent was reported in Sweden.33

Since the main rationale for most cohort studies is the investigation of etiologic hypotheses, such response rates are not of concern for internal validity.34 In fact, even when restricted populations, such as the cohort of women in the Nurses' Health Study, are used as the enrolment sampling frame, about 51 percent of the letters sent resulted in enrolment.35 This has not precluded using these data for the investigation of etiologic hypotheses.

The lower response rates for cohort studies, as opposed to case-control or crosssectional studies, are hardly surprising given the far higher degree of commitment asked of cohort participants. In our study, we asked participants to be willing to be followed until the age of 85 or until death, and similar commitments are expected in other prospective studies. In fact, we believe that the intensity of the questionnaire process used in this first enrolment phase of the study was a useful study component, not unlike the "run-in" period used in long-term randomized controlled trial designs.36·37 Other cohort investigators have also noted this potential advantage,30 since those who do enrol are more likely to continue the follow-up over a number of years. Indeed, this possibility has now been corroborated by the results of our first follow-up survey, in which approximately 92 percent of those fully enrolled did complete and return the questionnaire. Such a response bodes well for future follow-up.

However, thought has to be given to what degree the cohort data can be used to determine answers to population-based questions. Our data indicate that some caution would need to be used in attempting to use cohort data to reflect prevalence of health behaviours. There is a slight tendency towards "healthy enrôlée" effects in a long-term study group, and our group had higher non-smoker rates and slightly higher "ever use" of screening tests. However, the fact that the cohort group had a higher prevalence of obesity than the CCHS group indicates the presence of more than a simple bias. It is possible that some, although not all, of the difference between the two groups reflects the later secular time period for collecting the cohort data; smoking rates have been decreasing in Alberta, while use of screening tests and obesity rates have been increasing. It is interesting to note that the differences between the two groups in health practices reflect the same differences in secular trend.

In fact, some would argue that the use of such population-based information can be extrapolated further, to the calculation of population attributable risk and preventable proportions, provided known exposure rates in the general population are applied to the cohort data in question.28·30 It should be noted that even obtaining such point estimates in current cross-sectional studies is becoming more difficult; the participation rates for such cross-sectional studies as the Behavioural Risk Factor Surveillance Study, used to gauge health behaviours in the USA, has participation rates of 42.4 percent overall and as low as 24.0 percent in some states.38 Thus, any surveillance of population-based risk factors is likely to be subject to increased selection pressure in future.

One of the intended applications of this cohort is the ability to observe the outcome of "natural experiments". By this we mean that we will have the ability to observe how local changes in policy or environmental conditions affect cohort participants in the affected environment as compared to the "controls" in the stable environments. This particular application of population data will not be affected by the population selection pressure since one will be able to either select matched controls or control for potential effect modifiers within the entire design; the same arguments about internal validity apply as when one is using the cohort data to examine etiologic hypotheses. In addition, the fact that we were able to successfully enrol over the entire geographic area of the province, with similar uptake rates around the province, predicts that we will be able to apply the cohort results to monitor such effects in future.

Other recruitment strategies, such as collaborating with Statistics Canada's CCHS or using health care insurance files, were explored but found to be not feasible. The CCHS-affiliated method would have introduced a selection bias, as consent for cohort enrolment could only be asked for at the conclusion of CCHS interviews, and then only when the interview was concluded face to face. Because Statistics Canada indicated that many interviews were concluded on the telephone, it was decided to try a method where at least the first approach to the individual was carried out via random selection. Timeliness issues argued against attempting to use health care insurance files. Thus, random digit dialing (RDD) became our primary recruitment method. Because RDD does not depend on telephone directory listings, all households with telephone lines had an equal chance of being called. Since 97 percent of households had at least one telephone line at the time, using a telephonebased, RDD sampling method meant that almost all households were included in the theoretical sampling frame. However, we are now starting to become concerned that the RDD telephone-based recruitment approach is likely to become less effective in the near future. In the USA, there is growing disquiet concerning declining response rates to RDD methods of contacting potential subjects.39·40 Declining response rates not only raise concerns about the ability of the resulting cohort to reflect even broadly the characteristics of the general population, but may also have severe fiscal consequences as the length of time and amount of effort required to achieve the desired sample numbers increases. It is therefore possible that other recruitment methods may be evaluated against the RDD experience before future waves of enrolment are undertaken in the Alberta Cohort Study.

While the double recruitment strategy was somewhat effective in increasing the number of potential study subjects, the potential "cost" of this approach outweighed the benefit. That is, a relatively high portion of the RDDl sample (28.1 percent) shared a household with another person recruited for the study (most often a spouse). If a large proportion of these individuals along with their household partners were to enrol in the study, the potential of having a high degree of correlation present in the study data, especially on measures of exposure would outweigh the advantages of using the "second person" strategy. Despite the added efficiency, the method was not used in subsequent RDD recruitment waves.

There had been considerable media discussion about information privacy concerns and the potential ethical challenges surrounding the collection of biologic samples for long-term storage and use prior to and during our study enrolment. In fact, the Office of the Information and Privacy Commissioner was established in Alberta in 1995, presumably- and in part- to address public concerns in this area.

However, the consent for use of personal health information and to be contacted for biologic specimens was extremely high. Undoubtedly, some of this is due to the research-oriented nature of individuals who choose to participate in such a study; those who were uncomfortable with this request may have elected not to participate in the cohort at all. Furthermore, the somewhat abstract notion that they may be contacted in the future to consider providing a blood sample may have encouraged some subjects to respond positively to the request without thinking it through. In order to estimate how many people would provide a sample if asked, we subsequently conducted a small pilot study, which demonstrated that approximately two thirds of those approached would give a 50 mL sample of blood for banking. These samples (N = 769) have been processed to provide multiple aliquots of serum, plasma, red blood cells and buffy coat, which are being stored at -85 0C in mechanical freezers. The planned collection of further blood samples will necessitate expansion of the existing bio-repository and establishment of protocols to guide the granting of access to the samples for further research. All further research on the samples will be subject to full ethical approval.

In conclusion, the results of this feasibility study suggest that cohort development in the Canadian context is feasible, and that the potential for future studies using biological samples to determine prevalence of biomarkers- or correlations between reported exposures or disease outcomes with biomarkers- appears to be high. It is our hope that the information in this feasibility study will be of use to the much more extensive discussions that will need to take place in proposing the ultimate design, funding and administration of a full-scale cohort.

Acknowledgements

This project was funded by the Alberta Cancer Board's New Initiative Program. Christine Friedenreich is supported by a CIHR New Investigator award and an AHFMR Scholar award. The authors gratefully acknowledge the input of their colleagues, Ilona Csizmadi, Elizabeth McGregor and Linda Cook, on aspects of study design and review, and Karen Kopciuk and Penny Brasher for statistical advice on weighting of samples. Thanks also to Gwynne Rees and Will Rosner for undertaking the statistical analyses.

Copies of the research instruments are available on request.

References

References

1. Livingstone MBE, Black AE. Markers of the validity of reported energy intake. J Nutr 2003;133 Suppl 3:895S-920S.

2. Moti RW, McAuley E, DiStefano C. Is social desirability associated with selfreported physical activity? Prev Med 2005;40(6):735-9.

3. Adams SA, Matthews CE, Ebbeling CB, et al. The effect of social desirability and social approval on self-reports of physical activity. Am J Epidemiol 2005;161(4):389-98.

4. Hebert JR, Peterson KE, Hurley TG, et al. The effect of social desirability trait on selfreported dietary measures among multiethnic female health center employees. Ann Epidemiol 2001 ;11(6):41 7-27.

5. Abba K, Clarke S, Cousins R. Assessment of tli e potential effects of population changes in attitudes, awareness and beliefs on selfreporting of occupational ill-health. Occup Med (Lond) 2004;54(4):238-44.

6. Kristman V, Manno M, Cote P. Loss to followup in cohort studies: how much is too much? Eur J Epidemiol 2004;19(8):751-60.

7. Lavrakas PJ. Telephone Survey Methods Sampling, Selection, and Supervision. 2nd ed. Sage Publications, Inc. 1993.

8. Statistics Canada. Residential Telephone Service Survey (RTSS). 2001

9. Statistics Canada. Definitions, data sources and methods. 2003. Available from: URL: http: //www. statcan. ca/english/concepts/ stat-unit-def.htm (last accessed 19th December 2005).

10. University of Alberta. Population Research Laboratory, University of Alberta. Available from: URL: http://www.uofaweb.ualberta. ca/prl/ (last accessed 19"' December 2005).

11. National Cancer Institute. Prostate, Lung, Colorectal & Ovarian Cancer Screening Trial (PLCO). Available from: URL: http:// www3.cancer.gov/prevention/plco/index. html (last accessed 21st December 2005).

12. Women's Health Initiative Study Group. Design of the Women's Health Initiative Clincal Trial and Observational Study. Control Clin Trials 1998;19:61-109.

13. Statistics Canada. Canadian Community Health Survey (CCHS), Questionnaire for Cyde 1.1. Available from: URL http://www. statcan.ca.english/sdds/instmment/3226_Ql_ Vl_E.pdf Oast accessed 21st December 2005).

14. UCLA Center for Health Policy Research, California Department Health Policy Research, Public Health Institute. 2001 California Health Interview Survey Adult Questionnaire.

15. Mills C, Stephens T, Wilkins K. Summary report of the workshop on data for monitoring tobacco use. Health Rep 1994; 6(3):377-87.

16. Lovato C, Shoveller J, Mills C. Canadian national workshop on measurement of sun-related behaviours [Workshop report] . Chronic Dis Can 1999;20(2):96-100.

17. Rand Health. Medical Outcomes Study Questionnaire. Available from: URL: http:// www.rand.org/health/surveys/core/ (last accessed 21s1 December 2005).

18. Kushi LH, Kaye SA, Folsom AR, Soler JT, Prineas RJ. Accuracy and reliability of self-measurement of body girths. Am J Epidemiol 1988; 128(4) :740-8.

19. National Cancer Institute. Diet History Questionnaire: Available from: URL http://riskfactor.cancer.gov/DHQ/forms/ canadian/index.html (last accessed 14th July 2006).

20. Subar AF, Thompson FE, Kipnis V, et al. Comparative validation of the Block, Willett, and National Cancer Institute food frequency questionnaires : the Eating at America's Table Study. Am J Epidemiol 2001;154(12):1089-99.

21. Thompson FE, Subar AF, Brown CC, et al. Cognitive research enhances accuracy of food frequency questionnaire reports: results of an experimental validation study. J Am Diet Assoc 2002; 102 (2):21 2-25.

22. Csizmadi I, Kahle L, Ullman R, et al. Adaptation and evaluation of the National Cancer Institute's Diet History Questionnaire and nutrient database for use in Canadian populations. Public Health Nutr (in press).

23. Friedenreich CM, Courneya KS, Bryant HE. The lifetime total physical activity questionnaire: development and reliability. Med Sci Sports Exerc 1998;30(2):266-74.

24. Friedenreich CM, Courneya KS, Neilson HK, et al. Reliability and validity of the Past Year Total Physical Activity Questionnaire. Am J Epidemiol 2006; 163(10): 959-70.

25. Statistics Canada. Canadian Community Health Survey Cycle 1.1. 2001. Public Use Microdata File.

26. Béland Y, Dufour J, Hamel M. Preventing nonresponse in the Canadian community health survey. Proceedings of Statistics Canada's Symposium, Achieving Data Quality in a Statistical Agency: a Methodological Perspective. 2001. Ottawa, Ontario, Statistics Canada.

27. Census Operations Division, Statistics Canada. Income of Canadian Families. 2001 . Available from: URL: http://www23.statcan. ca/english/census01 /Products/ Analytic/ companion/inc./provs.cfm Oast accessed 21st December 2005).

28. Manjer J, Carlsson S, Elmstahl S, et al. The Malmo Diet and Cancer Study: representativity, cancer incidence and mortality in participants and non-participants. Eur J Cancer Prev 2001;10(6):489-99.

29. Boker LK, van Noord PA, van der Schouw YT, et al. Prospect-EPIC Utrecht: study design and characteristics of the cohort population. Eur J Epidemiol 2001;17(11): 1047-53.

30. Boeing H, Korfmann A, Bergmann MM. Recruitment procedures of EPICGermany. European Investigation into Cancer and Nutrition. Ann Nutr Metab 1999;43(4):205-15.

31. van den Brandt PA, Goldbohm RA, van't Veer P, Volovics A, Hermus RJ, Sturmans F. A large-scale prospective cohort study on diet and cancer in The Netherlands. J Clin Epidemiol 1990;43(3):285-95.

32. Limburg PJ, Anderson KE, Johnson TW, et al. Diabetes mellitus and subsitespecific colorectal cancer risks in the Iowa Women's Health Study. Cancer Epidemiol Biomarkers Prev 2005;14(l):133-7.

33. Gram IT, Braaten T, Terry PD, et al. Breast cancer risk among women who start smoking as teenagers. Cancer Epidemiol Biomarkers Prev 2005;14(l):61-6.

34. Szklo M. Population-based cohort studies. Epidemiologic Reviews 1998;20(l):81-90.

35. Barton J, Bain C, Hennekens CH, et al. Characteristics of respondents and nonrespondents to a mailed questionnaire. Am J Public Health 1980;70(8):823-5.

36. Lang JM. The use of a run-in to enhance compliance. Stat Med 1990;9(l-2):87-93.

37. Buring JE, Hennekens CH. Cost and efficiency in clinical trials: the U.S. Physicians' Health Study. Stat Med 1990;9(l-2):29-33.

38. Centers for Disease Control. 2003 Behavioral Risk Factor Surveillance System Summary Data Quality Report. Available from: URL: http://www.cdc.gov/brfss/ technical_infodata/2003 QualityReport. htm Oast accessed 21st December 2005).

39. Allen M, Ambrose D, Halfpenny G, Simmie T Telephone refusal rates still rising: Results of the 2002 Response Rate Survey. Available from: URL: http://www.pmrsaprm.com/specialresponse/article01.html Oast accessed 21s' December 2005).

40. Curtin R, Presser S, Singer E. Changes in telephone survey nonresponse over the past quarter century. Public Opinion Quarterly 2005;69(l):87-98.