Statistics

The associations between non-recognition of persistent cough or hoarseness and unexplained bleeding as cancer symptoms, number of NCBs about cancer and long anticipated patient intervals for persistent cough and rectal bleeding, respectively, were examined in two separate logistic regression analyses. Odds ratios (ORs) were calculated as measures of association. In the adjusted analyses, both non-recognition of symptom and number of NCBs were included as independent variables and sex, age, marital status, highest level of education, smoking status, experience of cancer and country were included as co-variables. As a sensitivity analysis, the logistic regression analyses were repeated with the answer “don’t know” to the questions about whether persistent cough or hoarseness and unexplained bleeding could be signs of cancer coded as “non-recognition” instead of missing. To test for a moderating effect of NCBs on the association between non-recognition of the symptoms and length of anticipated patient intervals, two interaction terms were computed: non-recognition of persistent cough or hoarseness as a cancer symptom multiplied with the number of NCBs and non-recognition of unexplained bleeding as a cancer symptom multiplied with the number of NCBs. These two interaction terms were included in two separate logistic regression analyses which also included independent and demographic variables. The likelihood-ratio (LR) test was used to determine whether the models with and without the interaction terms were statistically significantly different. In case of a statistically significant interaction, predictive margins would be used to visualise the effect and to assist interpretation (https://www.cscu.cornell.edu/news/statnews/stnews84.pdf). P-values of 5% or less were considered statistically significant. Data were analysed using STATA 13.1.