Test-retest reliability of a questionnaire used in a hospital-based case control study of nasopharyngeal carcinoma (NPC) in Hong Kong : a pilot study

Abstract

Background: From 1980’s, there is a no comprehensive, large-scale epidemiological study investigating Nasopharyngeal carcinoma (NPC) in Hong Kong. Thus, a hospital-based case control study organized by the Centre of Nasopharyngeal Carcinoma Research (CNPCR), Area of Excellence (AoE), was conducted. A questionnaire, which was designed with life course perspective, is used to elucidate the relationship between the dietary, familial and lifestyle risk factors and NPC risk. Method: The test-retest reliability of the newly-designed questionnaire was evaluated among subjects a hospital-base, case-control study investigating nasopharyngeal carcinoma in Hong Kong. Thirty-four cases were interviewed using the same computer assisted self-administrated application with an average of 24.4 weeks (range, 2 to 56 weeks) after the first interview at the Queen Mary Hospital (QMH). Cohen’s Kappa, quadratic-weighted Cohen’s kappa, and intraclass correlation coefficient (2-way, random effect model) were used to estimate the agreement and the consistency between test and retest responses for all items in the questionnaire. Result: Kappa ranged from 0.04 to 1.00 among case subjects with most estimates fall between 0.4 and 0.8. The ICC estimates for ordinal and continuous variables ranged from -0.03 to 1.00 as well with most estimates fall between 0.4 and 0.7. The questions on body size, visualized by the body image chart, had the highest Kappa (0.84 to 1.00). In general, there is no obvious difference in coefficients across different life periods, except the questions on sunlight exposure and skin tone (in which more recent recall life period it was, the higher Kappa and ICC the estimates were). Conclusion: The present study shows that this questionnaire is generally reliable for the ongoing case control study, and further study is warranted to estimate the test-retest reliability with a larger size of study population, and to estimate the effect on the estimated association (away from/toward null value) between exposures and NPC risk and the validity of the questionnaire.