OBJECTIVES: Conduct a systematic review of the human literature published in English from 2009-2014 to identify the prevalence of prescription opioid abuse as reported in retrospective databases where International Classification of Diseases (ICD) codes are used to measure the outcome of interest. METHODS: Sources included PubMed, Embase, OpenSIGLE (for gray literature) and others. Studies considered used a retrospective database, measured abuse using ICD codes, stated prevalence or it could be derived, and timeframe was documented. Due to study heterogeneity a meta-analysis was not performed. RESULTS: Of 5,412 citations identified, 509 were selected for preliminary inclusion/screening, 22 papers were critically appraised using criteria modified from Loney et al, and 16 papers were included for data abstraction (years 1999 – 2012) and included commercial and US government claims databases. Ten studies used ICD codes of abuse, dependence, or poisoning (all ICD codes) and 6 studies only used specific ICD codes. Eight studies limited determination of abuse to an opioid user population; 3 limited to use of specific opioids. Abuse prevalence among opioid users in commercial databases using all ICD codes varied depending on the opioid; 21/1,000 persons (reformulated ER oxymorphone; 2010-2012) to 113/1,000 (immediate-release opioids; 2010-2012). Abuse prevalence in populations with and without opioid use and using all ICD codes in commercial databases ranged from 1.15/1,000 persons (6 months 2010) to 2.66/1,000 persons (2012); Medicare 6.35/1,000 (6 months 2010); VHA 7.3/1,000 (2010); and Medicaid 8.7/1,000 (2002-2003). Studies evaluating prevalence over several years demonstrated increased prevalence over time. CONCLUSIONS: Prevalence varies depending on ICD codes used, population, timeframe and years studied. Prevalence increases when limited to opioid users. When all ICD codes are used and population includes those with and without opioid use, prevalence is highest in US government insured populations. Researchers using ICD codes to determine opioid abuse prevalence need to be aware of limitations.