One hundred and fifty individual questionnaires were made available at each of the two practices. 140 completed questionnaires were collected after six weeks. Questionnaires were left in a prominent location at the reception desk, it was not possible to ascertain how many people considered participating, but then returned the survey uncompleted. All participants responded to each of the nine vignettes presented. Each vignette was presented on average 19.68 times (Range 9–32). All the vignettes were incorporated into the analysis. The majority of respondents were female (70%) and the vast majority (80%) considered themselves Australian or European. The demographics of the sample, majority older females, were consistent with those of consulting patients reported previously [14]. See Table 1, 'Demographic characteristics' in additional files.
In most vignettes, where respondents expressed a view (1202/1260), they indicated that the patient should consult a GP (1056/1202) 83.8%. The majority who answered the question (952/1228) or 77.5% recommended a consultation within a week and (215/1228) or 17.4% suggested an appointment today.
Recommending a consultation with a General Practitioner in the case as described
None of the respondent variables appeared to influence which cases would be advised to consult a GP. Of the clinical details incorporated in the vignettes, six weeks of symptoms, breathlessness or weight loss were more likely to lead to this suggestion. Table 2 displays the extent to which the variables influenced the outcome variable. Twenty-one percent of the variability could be explained from these independent variables. Area under ROC curve = 0.83. Therefore the accuracy of model as a test for respondents' views could be described as 'good'. The sensitivity of the model was 99.5%, specificity 17.8%, positive predictive value was 91.2% and the negative predictive value was 68.4%.
Recommending an appointment with a doctor within one week
For the purposes of this study we assumed that seeking an appointment within one week was a mark of significant concern for the patient described in the vignette. Respondent characteristics did not appear to have a bearing on the decision. The most significant features were six weeks of symptoms, breathlessness and or weight loss. See Table 2. Twenty one percent of the variability could be explained from these independent variables. Area under ROC curve = 0.80. Therefore the accuracy of model as a test for respondents' views could also be described as 'good'. The sensitivity of the model was 96.2%, specificity 34.5%, positive predictive value was 84.9% and the negative predictive value was 67.4%.
Identifying potential cancers
Respondents performed less well in identifying potential 'cancers' explicitly. Cancer vignettes were presented five hundred and eight times in the survey. Respondents suggested the patient is 'likely' or 'very likely' to have cancer and should make an appointment within one week in one hundred and twenty two occasions (24%). In more than half of all cases (53%) respondents were unsure of the chance that the symptoms were related to cancer. Of the cancer vignettes 55% included a history of cigarette smoking. Only 32% of cancer vignettes with a history of smoking were recognised as 'likely' or 'very likely' to have cancer. Cancer was suspected in 13.8% of all vignettes However, most cases respondents identified as cancer (77.5%) would be classified as 'high risk' on current medical guidelines. Twenty six per cent of the variability could be explained from these independent variables. Area under ROC curve was 0.85. Therefore the accuracy of model as a test for respondents' views could be described as 'good'. The sensitivity of the model was 32.5%, specificity 95.3%, positive predictive value was 52.4% and the negative predictive value was 89.8%. The most significant symptoms were duration of symptoms, weight loss, breathlessness, symptomatic person's smoking status, age and cough in that order. See Table 2, 'The impact of variables on respondents' decisions' in additional files.