During October through November 2005, a random sample of 4,000 persons aged 40–59 (at 1 January 2005) in the municipality of Odense (approx. 185,000 inhabitants), Denmark, was invited for an interview. Invitees were drawn from the Central Office of Civil Registration (the CPR-Office) under the Ministry for Economic and Interior Affairs.
No information from medical records was obtained, but due to age the individuals belonged to a target group of potential candidates for cardiovascular prevention therapy. Non-responders were followed up with reminders by letter and/or telephone.
The interviews took place in a university building next to the university hospital just outside the city centre. Participants were informed that the interview would be about preventive health care. For their effort the respondents would receive a gift of either two bottles of wine or one box of chocolates. Interviews were conducted in 4-hour sessions in the afternoons over a period of six weeks. Each interview lasted for about 30 minutes.
The design of the interview guide was based on intensive discussions within the research group, two groups with lay persons and one group with health personnel, and ultimately on discussions with researchers from the Danish National Centre for Social Research. A group of trained interviewers were assigned to conduct the interviews. The interview guides were structured with pre-specified options.
All participants were asked questions on sociodemographic characteristics, comprising age, gender, marital status, family income, educational attainment and occupation. They were then asked to imagine being at increased risk of cardiovascular disease (the baseline risk) and being offered medication. No medication name was mentioned, but the features of the medication (effectiveness and side effects) resembled statins.
Participants were allocated to four different levels of effectiveness, which regardless of information type was equivalent to an absolute risk reduction (ARR) of 2%, 4%, 5%, and 10%, and formed the bases for information about medication effectiveness. Even though they were initially presented with different formats of information (relative risk reduction, absolute risk reduction, number needed to treat, expected prolongation of life, as well as a pictorial presentation) [13
] they were later all presented with the same ‘complete’ effectiveness information with a combination of all four formats for medication effectiveness presented, including a pictorial presentation (see Figure presenting example of pictorial presentation). This “complete” information formed the basis of the decisions - to accept or not accept treatment - studied in this paper.
Example of pictorial presentation comprised in the ‘complete’ information.
This paper reports on data from a study aimed at exploring the different ways of explaining risk reductions to lay people [13
]. The study was powered to test hypotheses other than those tested here.
All the different effectiveness formats can be presented based on the same data, and none of the measures are more or less correct than the other, but all may affect decision; In order to minimize bias because of a specific format we presented “the full package”, and focused on the participants’ subsequent acceptance or rejection of cardiovascular preventive medication and the reasons for their decision. We refer to [13
] for detailed description of the underlying design, including details of the risk information formats, interview guide and information cards.
While the initially presented risk format led to different acceptance rates [13
], these differences were largely removed when participants had been presented with the final, “complete” information. We have therefore not allowed for an effect of the baseline risk and initial risk format, since all outcomes studied here are obtained after the final information had been provided.
We constructed the response options based on the Health Belief Model [15
]. For participants who accepted medication, the options were: “For health reasons”, “Trust in my GP” or “Responsibility towards my family”. For participants who declined medication, the options were: “Too small effect of the medication”, “Wish to avoid side-effects”, “Do not want the extra expense”, “Dislike taking medication”, “Find the information difficult to understand”, “Prefer to change lifestyle”, and “General disbelief in effectiveness of medication”. For each of the above, participants were asked to pick the most important reason for the choice made. Finally, all participants were asked about their personal and family history of hypercholesterolemia, CVD or stroke, and asked four questions to capture their numeracy skills.
According to the Act on a Biomedical Research Ethics Committee System the project was not a biomedical research project and therefore did not need the ethic committee’s approval. The study was approved by the Danish Data Protection Agency.
The basic response variable was “acceptance of medication” on a binary scale (yes/no). To identify possible associations between participants’ characteristics and medication-taking behavior we first performed simple and multiple logistic regression modeling in Stata with acceptance of medication as the dependent variable. As explanatory variables we used medication effectiveness (ARR in percent), age, gender, duration of education, household income, numeracy skills, living with a partner, personal experience with cardiovascular disease or risk factors as presence of one or more of the following conditions: previous stroke or heart attack, hypercholesterolemia, or hypertension, and whether the participant had experienced cardiovascular disease in the family or not. The variables medication effectiveness, age and household income were used as continuous covariates, i.e. the corresponding odds ratio represents the estimated relative increase in odds due to a one unit increase in the covariate (percentage point, year, 100,000 DKK, respectively. US$1.00 = DKK5.90). The other variables were binary covariates.
In addition, we performed subgroup logistic regression analyses with people’s reasons for accepting medication as response variables. The analyses included separate logistic regression analyses for each of the three different response variables “For health reasons”, “Trust in my GP”, and “Responsibility towards the family”. Secondly, we performed separate subgroup logistic analyses of reasons for declining medication with each of the response variables “Too small effect of the medication”, “Wish to avoid side-effects”, “Dislike taking medicine”, and “Rather change lifestyle”. The options “Do not want the extra expense”, “Find the information difficult to understand” and “General disbelief in effectiveness of medication” were excluded from the analysis due to low frequency (1, 2, and 1 observations, respectively). As explanatory variables for the subgroup analyses we used medication effectiveness (ARR in percent), age, gender, duration of education, household income, numeracy skills, living with a partner, personal experience with cardiovascular disease or risk factors as presence of one or more of the conditions: previous stroke or heart attack, hypercholesterolemia or hypertension, and whether the participant had experience with cardiovascular disease in the family or not. The variables were chosen a priori from considerations based on expectations and knowledge within the field. For all estimates we report odds ratios (OR), and (95% confidence intervals). We chose logistic regression modeling because the outcome variable “acceptance of medication” is binary. We chose independent variables that would be plausible predictors of the outcome. We used non-parametric smoothing to assess linearity of the response variable on the log-odds scale with respect to continuous covariates. To build the model we had a priori established a list of covariates to include, but used the size of their estimated standard errors relative to the estimated effect size to judge both their impact and statistical significance. This was the case for both continuous and categorical covariates.