Objective Homeless individuals with poor health frequently use healthcare services. However, studies using self-reported data may be subject to biases. We examined health status, healthcare and drug use among mentally ill homeless individuals, comparing self-report and administrative data claims to estimate the degree of agreement between the two sources.

Methods Baseline survey data from 100 participants of the Winnipeg site of the Mental Health Commission of Canada's At Home/Chez Soi research project were linked to deidentified administrative health records stored in the Repository at the Manitoba Centre for Health Policy. Demographic characteristics, homelessness histories and health service use were analysed, as well as disease status for asthma, hypertension, arthritis and diabetes (using previously validated definitions). Participants were similarly classified using their survey responses. The degree of agreement between the two data sources was evaluated using cross-tabulations and the κ statistic.

Results There was 100% linkage of surveyed homeless people with the Repository data. In 1 year, 97% of participants had at least one ambulatory physician visit, with an age- and sex-adjusted rate of 14.82 per person-year (Manitoba rate=4.99 per person-year). 34% had an inpatient hospitalisation (adjusted hospital separation rate=491 per thousand person-years vs the Manitoba rate of 137 per thousand person-years). 95% filled at least one prescription, with 65% of drugs targeting the nervous system (majority were psycholeptics). The degree of agreement between the data sources ranged from a κ of 0.27 for arthritis to 0.57 for hypertension. Individuals were more likely to be classified as having one of the four conditions based on the administrative data than on the survey data.

Conclusions Compared with the general population, participants had high health service use, and high prescription drug use. There was poor to moderate agreement between the two data sources. Researchers studying homeless persons with mental illness should consider using multiple data sources to estimate disease prevalence and health service use.