OBJECTIVES: To assess frequency of proxy definitions for health conditions in retrospective studies querying existing databases.  METHODS: Systematic literature review (PubMed and ISPOR Scientific Presentations Database, searched 1/1/2013-12/31/2015). Eligible studies focused on cost/utilization burden of any disease, published in English or presented at ISPOR annual meetings. Studies with populations explicitly limited by treatment(s) received were excluded. We used publicly available information (abstract, poster, and/or full text) to extract data on study methodology used to identify patients with the condition of interest. The primary endpoint was the proportion of studies using proxy definitions of illness, e.g. treatments received or other algorithms besides diagnosis codes, to qualify patients for inclusion. P values were computed using Fisher's exact test. RESULTS: CONCLUSIONS: ISPOR annual meeting abstracts were not more likely than published papers to use proxy definitions of disease conditions. Most proxy definitions involved treatments received, which can dwarf the number eligible by diagnosis code. Cost of illness estimates may be influenced by patient selection using proxy definitions.