We conducted a cross-sectional survey of: a) family physicians, regarding a decision aid for menopausal women considering hormone replacement therapy (HRT) [5–7]; b) geriatricians, regarding a decision aid for individuals faced with making a decision about long-term placement of a feeding tube in a cognitively impaired older person; and c) respirologists or pulmonologists, regarding a decision aid for individuals with chronic obstructive pulmonary disease (COPD) considering mechanical ventilation at the end of their lives. These decision aids were purposefully selected to represent a range of decisions made in primary and tertiary care and to represent a spectrum of decisions, including lifestyle and end-of-life decisions, faced by both patients and substitute decision makers. Also, these decision aids have previously been shown to have beneficial effects on knowledge[6, 8], realistic expectations[5, 6, 9], decisional conflict[5, 6, 8, 9] and the proportion of patients remaining undecided about treatment choice[6, 8, 9]. The survey was administered following Dillman's total design method for mail surveys, a method designed to enhance survey response rates. One week in advance of the first mailing a letter, signed by a clinician co-investigator who was a peer of the respondent, was sent to each of the physicians. The letter was sent to explain the purpose of the study and to alert respondents to the launch of the survey. One week later, respondents were sent the decision aid, the questionnaire, an addressed stamped envelope, and an honorarium cheque for $50 (CAD). Reminders were sent to non-respondents at 3, 5, 7, 9, and 14 weeks after the initial mailing.
Three months following completion of the mail survey, respondents who had indicated on the questionnaire that they intended to use the decision aid in their clinical practice were administered a brief follow-up telephone survey and provided with a $10 (CAD) honorarium.
The sample frame for the study was drawn from MD Select, the CD ROM version of the Canadian Medical Directory. Physicians from the three practitioner groups of interest (respirologists, family physicians, geriatricians) were identified and assigned a number. Random samples were then drawn and potential respondents were contacted by phone to determine their eligibility for the mail survey. This involved confirming that the potential respondent belonged to the practitioner group of interest, was previously unaware of the decision aid, and had patients for whom the decision aid would potentially apply.
A sample size of 450 (150 from each physician group) would have allowed us to estimate the percent indicating an intention to use the decision aid well within a 10% bound on the error of estimation, with 95% confidence. Thus, allowing for a conservative response rate of 60%, a sample frame of 765 (255 practitioners from each physician group) was required. A questionnaire was mailed to random samples of respirologists (n = 255) and family physicians (n = 255). Since there were only 130 geriatricians, the entire eligible population of geriatricians (n = 130) was contacted.
Using Likert-scaled items, the two-page survey elicited physicians' perceptions of the characteristics of the decision aid and their willingness to use it. The items were based on: a) the conceptual framework of the study, the Ottawa Model of Research Use[11, 12] ; b) the work of Rogers, Grilli and Lomas, Grol et al, and Brouwers et al who have identified attributes or characteristics of innovations related to adoption; and c) a qualitative study we undertook to elicit physicians perceptions of decision aids. The 43 items assessing the characteristics of the decision aid were divided into four main areas related to: the development of the decision aid (4 items); the content and format of the decision aid (10 items); the decision aid meeting patients' needs (11 items); and, the physicians' clinical practice (18 items). The characteristics were rated on a 5-point scale ranging from 1 ('strongly agree') to 5 ('strongly disagree'). Willingness to use the decision aid was assessed by asking the physicians how comfortable they would be to use the decision aid with patients ('very uncomfortable', 'uncomfortable', 'neutral', 'comfortable', or 'very comfortable') and how likely they were to use it within the next three months ('not at all', 'very unlikely', 'somewhat likely', 'likely', or 'very likely'). Additionally, physicians were asked whether they perceived a need for a decision aid on the topic of the particular decision aid they were asked to review. Lastly, physicians were asked the number of years that they had been practicing in their specialty (all other demographic information was obtained from the Canadian Medical Directory).
The follow-up survey inquired whether the physician had carried through with their intention to use the decision aid. Those who responded 'yes' were asked with how many patients they had used the decision aid and whether the experience was positive, negative or neutral. Those who responded 'no' were asked whether they had considered using the decision aid with at least one patient and whether they were considering using it in the future.
Ethical approval for the project was obtained from the Ottawa Health Research Institute Research Ethics Board. Consent to participate in the study was assumed by the respondent returning the completed mail survey and responding to the follow-up survey.
Data management and analysis methods
All statistical analyses were conducted using SPSS 13.0. A descriptive profile of the professional characteristics of survey responders was generated and the original data set was assessed for missing data. Respondents with two or less missing data points (n = 44) had values imputed based on their respective variable mode and respondents presenting with more than two missing data points (n = 7) were removed from further analyses.
A criterion was established for participation in the follow-up survey. A participant's response to the intention question on how likely they were to use the decision aid was dichotomized into those who were 'likely' or 'very likely' versus all other responses. Only the former group was contacted for the follow-up survey. Two additional items were dichotomized: the comfort item 'How comfortable would you be offering the decision aid to your patients?' was dichotomized into those that were 'comfortable' and 'very comfortable' versus 'neutral', 'uncomfortable', and 'very uncomfortable'; and the question of whether there was need for a decision aid on the topic was dichotomized into those who felt there was a need versus those who were unsure or did not feel there was a need.
Principal components analysis with Verimax rotation was used to reduce the 43 'characteristic' Likert-scaled items to a parsimonious set of meaningful components – which were then used as independent variables in the subsequent logistic regression analysis (the mean score of all items within each extracted component was used as the composite score for each factor). Components were assigned descriptive labels that represented the items within them: Quality and Value for Patients, Value for Physicians, Decision Aid Content, and Implementation issues. Items which loaded upon multiple components were assigned to the component which made most contextual sense. Items which loaded upon multiple components and were not easily assigned to any one of them (n = 1), and items with loadings below 0.45 (n = 3) were removed from further analysis.
Logistic regression was used to examine potential factors influencing a practitioners' intention to use the decision aid. Intention was used as the dependent variable. Predictors included the dichotomized comfort and need items and the four components extracted in the principal components analysis. Given potential differences between groups, dummy variables for respirologists and family physicians (with geriatricians as the reference group) were created and included as predictors in the model. Additional demographic variables included: physician's sex, number of years in specialty, having a hospital appointment, and having a university appointment. All predictor variables that were significant at the p = 0.10 level in preliminary univariable analysis were entered into the multivariable model.