Choice of health conditions
In this study, we included seven health conditions with varying susceptibility to being labelled as stigmatizing. Four conditions – hypertension, diabetes, chronic depression and alcoholism – were used in a previous public opinion survey of members of the public, in which respondents were asked to imagine they had one of these health conditions.  In the previous study, hypertension and diabetes were found to be lower-stigma health conditions. Chronic depression and alcoholism were found to be higher-stigma conditions. In this study, we had the opportunity to obtain the views of people with these conditions. To these four conditions we added: HIV, to create an upper extreme category for potential stigma; breast cancer; and lung cancer. We anticipated that responses may differ between breast and lung cancer because of a greater general public support for breast cancer sufferers and a perception that lung cancer is self-inflicted through smoking.
The study proceeded in two phases. In Phase 1 (November 2006 to July 2007), we surveyed individuals with the target health conditions. In Phase 2 (July to September 2007), we held focus groups with a sub-sample of participants from Phase 1 – one group for each health condition. The chief purpose of Phase 2 was to help inform our analysis of the findings derived from Phase 1, by providing examples of the types of concerns some people took into account when making decisions around consent.
Setting and Participants
Survey participants were drawn from two sources: (1) a pre-existing cross-Canada panel of individuals with identified health conditions, maintained by Harris Interactive, a professional polling firm; and (2) patients recruited by the investigators through family physicians' offices and specialty clinics in the vicinity of Hamilton, Ontario, Canada. The reference group consisted of people recruited through Harris Interactive who had none of the target health conditions and no other serious health conditions. This group was used to approximate the response of the general public. All survey participants were 18 years or older.
Participants recruited directly by the investigators were either sent a letter from the physician's office in the mail explaining the study or handed an information brochure in the clinic by staff. In each case, information was provided for patients to contact the investigators if interested. In total, 892 brochures were mailed to patients' homes and 888 brochures provided to physicians and clinics for directly handing out to patients.
All participants recruited through Harris Interactive completed the survey over the internet. Participants recruited through local family physicians and clinics were given the choice to complete the survey via internet or by telephone. This was done to minimize refusal due to lack of access to or familiarity with the internet, particularly among older patients. Those opting to do the internet survey used the same system as the Harris participants. Those opting to complete the survey by telephone arranged a scheduled call with telephone surveyors from Harris Interactive. They were mailed a hard copy of the core questions of the survey beforehand, so as to minimize differences in response due to method of survey administration.
Sample size was calculated on the basis of the primary outcome variable – consent choice in the use of personal information for health research. This was expressed on a 5-point ordinal scale. (See Survey Data and Key Variables below.) This sample was determined with the primary goal of building a multivariable regression model to compare the overall attitudes among the seven groups and the general public controlling for several demographic and other confounding variables. Heuristics based on simulation studies indicate that at least five respondents per degree of freedom for each predictor variable are required for the stability of the model.  We have 7 predictor variables with a total of 21 degrees of freedom, which would require at least 105 participants. We aimed to recruit 1400 and obtained responses from 1137 (734 from Harris Interactive and 403 from physician offices and clinics) with the sampling stratified by health condition. We inflated our minimum sample size by a factor of over 10 to account for potential clustering of responses within a patient. Therefore, the sample size was adequate to ensure the stability of the model.
Survey Data and Key Variables
We collected information on participant demographics, attitudes about privacy, disclosure concern, and the benefits of health care and health research at improving longevity and quality of life, and the participant's health conditions. Where the participant had more than one target health condition, they were asked to answer the survey questions with only one health condition in mind. In earlier pilot work, we established an algorithm for determining which health condition would take priority, ranking the 7 health conditions according to level of disclosure concern. This algorithm placed HIV/AIDS first, followed by alcoholism, lung cancer, breast cancer, depression, diabetes and hypertension.
We presented five different scenarios involving use or linkage of personal information for health research: (1) use of health data for quality improvement; (2) use of the same data for marketing; (3) linkage of work/education/income information with health information; (4) linkage of biosamples with health information (a) assuming no profit and (b) assuming a profit element. (See additional file 1 for a more detailed description.) These scenarios were identical to those used two years earlier in a series of seven cross-Canada public dialogues.  Participants of the current study were advised that, in each scenario, names, addresses and any other information that could directly identify them were removed. Following each scenario, participants were asked which statement best described their view (words in italics not included in survey responses):
(1) There is no need for me to know. Just use it.
(2) My permission is not needed, but I want to know this is being done and a chance to say "no." [i.e. notice with opt-out]
(3) My general permission is needed. This could be for several different research studies. I could withdraw my permission in future. [i.e. broad opt-in]
(4) My permission is needed each time. [i.e. project-specific consent]
(5) My information should not be used for this purpose.
The five response options above, hereinafter referred to as "consent choices", served as the outcome variable in a multivariable regression analysis using the following as predictor variables:
demographics (age, sex, education, marital status, employment, and income)
health condition (one of the 7 target conditions)
self-reported health (6-point scale, varying from "very poor" to "excellent")
attitudinal variables (disclosure concern and medical benefits score. These are described below. Questions used to compile these scores and the scoring scheme are found in additional file 2.)
While there are many dimensions to stigma, for the purposes of this study, we chose to focus on individuals' disclosure concern – i.e. concern that others may find out about their health condition. For this, we asked: How concerned would you be if: (a) your employer found out about any health condition(s) you have; (b) your health insurer found out about any health condition(s) you have; or (c) a friend other than those you told found out about any health conditions you have. For each of these questions, respondents replied either: "not at all concerned"; "somewhat concerned"; or "very concerned".
For the medical benefits scale, people replied on a 5-point scale ("strongly agree, somewhat agree", "neither agree nor disagree", "somewhat disagree", or "strongly disagree") to the following statements: (a) Medical treatments can improve the quality of my life; (b) medical treatments can extend my life; (c) disease prevention programs have shown me how to live a healthier life; and (d) medical research can improve my life.
We re-scaled the disclosure concern and medical benefits scores to a 0–1 scale to facilitate interpretation of relative attitudes toward disclosure and medical benefit across health conditions.
Dealing with potential sampling bias
We checked for sampling bias chiefly through two methods. Harris Interactive sampled questions from our survey in an omnibus random-digit dialled telephone survey and compared the responses from the telephone survey with those from their internet sample. In addition, we compared the consent choices of the reference group from this study regarding the five scenarios with the consent choices of the people who participated in the public dialogues in our previous study. 
Consent choices were analysed graphically across scenarios and across health conditions. We also plotted disclosure concern and medical benefit scores across health conditions using a radar graph.
To test our chief hypotheses, we used regression analysis controlling the correlation across scenarios using the method of generalized estimating equations (GEE) assuming an exchangeable correlation structure.  The results are reported as estimates of model coefficients (with corresponding 95% confidence interval) and associated p-value. Statistical computations used SAS, version 9.1.3 (Cary, NC.). In earlier work, we found the results of linear regression to yield equivalent results to multinomial logistic regression which is the more correct analysis but more difficult to interpret. 
Initially, individual predictor variables were regressed onto consent choice using univarite analysis. Variables that met the criterion of alpha = 0.20 were then included in the multivariable regression model. The chief variables of interest – health condition, disclosure concern, and medical benefit scores – were forced into the model. We tested for interaction effects of disclosure concern and medical benefits scores with research scenarios and health condition, No significant interactions were found.
At the end of the survey, those participants recruited directly by the investigators in the Hamilton area were asked if they would be willing to participate in a focus group to discuss the reasoning that participants may have used to make their consent choices. Those who agreed were contacted by the study coordinator. One focus group was convened for each health condition except for alcoholism, as the survey sample size was too small for this group. We sampled from the pool of survey participants to provide a representative sample in each focus group with regard to age, gender, and self-reported health status. When making selections, investigators were blind as to volunteers' responses to survey questions.
Six focus groups, ranging in size from 6 to 10 individuals, were conducted between July and September 2007. Focus groups were 90 minutes in length. Participants received an honorarium of $75 following the session. These meetings were moderated by two of the researchers. Upon arrival, participants completed a mini-survey comprised of the key questions from the Phase 1 survey. During the focus group, a structured interview protocol was used to ask participants to reflect on the survey findings regarding consent choice for the five different research scenarios. Participants were then asked to consider how the survey responses compared with their own responses and to consider the reasons for their responses and any possible variance from the response pattern from the Phase 1 survey. Focus group participants were also asked to comment on the relative importance of individual control over use of their information and safeguards and the relation between them. This question was addressed both in the abstract and by giving them the opportunity to rate specific controls and safeguards. Immediately following the meeting, participants again completed the mini-survey, so a comparison of the responses of focus group participants with those of the larger sample could be made. Discussions were audio recorded and transcribed for analysis.
Verbatim transcripts from all focus groups were read independently by at least two members of the research team and quotations were selected to exemplify the types of reasoning that focus group participants used to make their consent choices. These quotations are used below to help illustrate the broader survey analysis, and to provide a window on the kinds of issues some people may address when making consent choices.
The research was reviewed and approved by the research ethics boards of St. Joseph's Healthcare, Hamilton, Ontario, McMaster University Health Sciences REB, Hamilton, Ontario and the Social Sciences and Humanities REB, University of Ottawa, Ottawa, Ontario.