Comparison of prevalence and exposure-disease associations using self-report and hospitalization data among enrollees of the world trade center health registry.

Background Although many studies have investigated agreement between survey and hospitalization data for disease prevalence, it is unknown whether exposure-chronic disease associations vary based on data collection method. We investigated agreement between self-report and administrative data for the following: 1) disease prevalence, and 2) the accuracy of self-reported hospitalization in the last 12 months, and 3) the association of seven chronic diseases (rheumatoid arthritis, hypertension, heart attack, stroke, asthma, diabetes, hyperlipidemia) with four measures of 9/11 exposure. Methods Enrollees of the World Trade Center Health Registry who resided in New York State were included (N = 18,206). Hospitalization data for chronic diseases were obtained from the New York State Planning and Research Cooperative System (SPARCS). Prevalence for each disease and concordance measures (kappa, sensitivity, specificity, positive agreement, and negative agreement) were calculated. In addition, the associations of the seven chronic diseases with the four measures of exposure were evaluated using logistic regression. Results Self-report disease prevalence ranged from moderately high (40.5% for hyperlipidemia) to low (3.8% for heart attack). Self-report prevalence was at least twice that obtained from administrative data for all seven chronic diseases. Kappa ranged from 0.35 (stroke) to 0.04 (rheumatoid arthritis). Self-reported hospitalizations within the last 12 months showed little overlap with actual hospitalization data. Agreement for exposure-disease associations was good over the twenty-eight exposure-disease pairs studied. Conclusions Agreement was good for exposure-disease associations, modest for disease prevalence, and poor for self-reported hospitalizations. Neither self-report nor administrative data can be treated as the "gold standard." Which source to use depends on the availability and context of data, and the disease under study.