Many studies have explored the association between antibiotic use and antibiotic resistance. However, methods employed in these studies to categorise prior antibiotic use (e.g. by class, by spectrum) have not been well described. The impact of using different categorisation methods on identifying risk factors for resistance is unknown. To explore these issues, we focused on extended-spectrum beta-lactamase-producing Escherichia coli and Klebsiella spp. (ESBL-EK) as a model. First, we conducted a systematic review of studies of risk factors for ESBL-EK to characterise past approaches to categorising antibiotic use. Second, we re-analysed data from a prior study of risk factors for ESBL-EK. Two separate multivariate models of risk factors for ESBL-EK were constructed: one with prior antibiotic use categorised by class and the other with prior antibiotic use categorised by spectrum of activity. Among the 20 articles that met the inclusion criteria for the systematic review, there was tremendous variability in how prior antibiotic use was categorised (e.g. by agent, class, spectrum and/or a combination of these). No study justified its choice of categorisation method. In the re-analysis of the existing data set, multivariate models of risk factors for ESBL-EK using 'class' and 'spectrum' categorisations differed substantially. In conclusion, there has been no consistent approach to categorising antibiotic use in studies of risk factors for ESBL-EK. Different categorisation schemes were shown to have a substantial impact on study results, particularly for the antibiotic exposures associated with resistance. Elucidating these issues is critical if effective strategies to curb resistance are to be designed.