Investigation of sources of potential bias in laboratory surveillance for anti-microbial resistance.

PURPOSE There are a number of biases that may influence the validity of laboratory-based surveillance for antimicrobial resistance. Our objective was to evaluate the potential magnitude of bias in reporting of etiologic agents and their resistance rates associated with inclusion of multiple patient samples and non-random timing and location of sampling. METHODS All urine cultures submitted to a regional laboratory in the Calgary Health Region during 2004 and 2005 were studied. Comparisons were then made using either the overall cohort or different subgroups compared with the "reference" or gold standard population where only the first isolate per patient per year per species was included. RESULTS Overall 56,897 organisms were cultured at > or =104 cfu/mL from 53,548 samples from 35,890 patients; 39,835 organisms were included in the reference cohort. Escherichia coli was reported in 37,246 (65.5%) of overall cohort and 28,257 (70.9%) of the reference cohort. Therefore, the overall cohort resulted in a relative underestimation of the importance of E. coli as the principal cause of urinary tract infections by 8%. Similarly, reported rates of resistance to antimicrobial agents most notably ciprofloxacin [6,480/52,544 (12.3%) vs. 2,647/37,086 (7.1%)], gentamicin [2,991/48,070 (6.2%) vs. 1,567/34,608 (4.5%)], and ceftriaxone [1,737/44,922 (3.9%) vs. 889/32,745 (2.7%)] were higher in the overall than in the reference cohorts. There were large differences in both the distribution of organisms and rates of resistance associated with sampling during different times of the day, week, and year as well as from acute care hospitals and outpatient clinics (P< or =0.001). CONCLUSIONS Reports from laboratory-based surveillance studies may be biased depending on the population studied and method of sampling employed. Care must be taken in interpreting results of surveillance studies that do not protect from these major sources of bias.