Extensive pre-administration piloting was conducted with a convenience sample of physicians
similar to the study population. A clear need to slim down the questionnaire emerged. Therefore, only questions concerning APC mutations were included among the knowledge items concerning the inherited forms of colorectal cancer, thus excluding questions regarding gene mutations associated with the Lynch syndrome. Other minor revisions included changes to the questionnaire item wording and format. Multiple logistic regression analysis was performed. Five models were built to identify the predictors of physicians knowledge of predictive genetic testing for breast and colorectal cancer (Models 1 and 2), attitudes (Model 3), and professional use of predictive genetic tests for breast and learn more TSA HDAC cost colorectal cancer (Models 4 and 5). For purposes of analyses, the outcome variables “knowledge” and “attitudes” in Models 1–3, originally consisting of multiple categories, were collapsed into two levels. In brief, for the variable knowledge physicians were divided in those who agreed with all correct responses versus all others, while for attitudes responders were grouped into those who showed a positive attitude in at least 70% of the questions versus all others (see Table 3 for the details of dichotomization). The Libraries following physician characteristics were initially tested in all models as predictor variables:
location; gender; age; exposure to cancer genetic testing during graduate/postgraduate courses; attendance to postgraduate epidemiology and Evidence Based Medicine (EBM) courses; knowledge of the English language; internet access in the workplace; hours per week dedicated to continuing medical education; the average number of patients treated in a typical week; patient requests for genetic tests in the previous year; the presence of genetic testing laboratories in the geographical area of professional activity; and a personal or family history of breast or colorectal cancer. The variable “adequate knowledge” was also included in the model concerning
attitudes, and the variables “adequate knowledge” and “positive attitudes” were included in L-NAME HCl the models concerning the professional use of predictive genetic tests (see Table 3 for the details of dichotomization). The model building strategy suggested by Hosmer and Lemeshow (2000) was used and included the following steps: (a) univariate analysis of each variable and inclusion if the p-value was lower than 0.25; (b) backward elimination of each variable that did not contribute to the model on the ground of the Likelihood Ratio Test using a cut-off of 0.05 level of significance; variables whose exclusion markedly altered the coefficient of the remaining variables were kept in the model; (c) testing of interaction terms using a cut-off of 0.15 level of significance.