Self-reported versus health administrative data: implications for assessing chronic illness burden in populations. A cross-sectional study

Abstract

BACKGROUND: Various data sources may be used to document the presence of chronic medical conditions. This study examined the agreement between self-reported and health administrative data. METHODS: A randomly selected cohort of participants aged 25-75 years recruited by telephone from the general population of Quebec reported on the presence of 1 or more chronic conditions from a candidate list of 12 conditions: diabetes, hypertension, thyroid disorder, any cardiac disease, cancer diagnosis in the previous 5 years (including melanoma but excluding other skin cancers), asthma, osteoarthritis, rheumatoid arthritis or lupus, osteoporosis, chronic obstructive pulmonary disease, intestinal disease and hypercholesterolemia. We also used health administrative data from Quebec's universal health insurance provider to identify participants' chronic conditions. Unique identifiers allowed linkage of both data sources to the individual participant. The frequencies of the 12 conditions and the prevalence of multimorbidity (≥ 2, ≥ 3 and ≥ 4 conditions) were analyzed for each data source. RESULTS: We analyzed data for 1177 participants (mean age 53 [standard deviation 12.4] yr; 684 women [58.1%]). We found low (but varied) agreement between the 2 data sources, with the poorest agreement for hypercholesterolemia (κ = 0.04 [95% confidence interval (CI) 0.01 to 0.07]) and the best for diabetes (κ = 0.82 [95% CI 0.76 to 0.88]). Prevalence estimates of multimorbidity obtained with health administrative data were lower than those obtained with self-reported data regardless of the operational definition used. Most participants with multimorbidity were identified by self-report. INTERPRETATION: We argue for the use of self-reported chronic conditions in the study of multimorbidity, as health administrative data based on the billing system in Quebec seem to underestimate the prevalence of many chronic conditions, which results in biased estimates of multimorbidity.