OBJECTIVES: The objective was to describe the disease state (DS) transition patterns of gastroesophageal reflux disease (GERD) by applying a multi-state model assuming a Markov process. METHODS: This retrospective study utilized administrative claims for subjects with GERD from a large Midwest USA healthcare plan. Subjects were tracked for six, six-month time periods (baseline and five followups). Within each time period, ICD-9 diagnosis codes were used to categorize subjects’ GERD into four DS plus a non-symptomatic state: DS0 {no GERD diagnosis}, DS1 {mild esophagitis}, DS2 {reflux esophagitis}, DS3 {esophageal ulceration}, and DS4 {strictures and complications}. GERD transition probabilities and patterns were analyzed in a five-state Markov framework. Disease regressions and progressions were allowed. The effects of patient and provider covariables on transition probabilities were modeled using logistic regression techniques. RESULTS: A total of 7575 subjects with GERD were analyzed. In the five followup periods combined, 79% of the subjects were in GERD DS0, 6% in DS1, 8% in DS2, 2% in DS3, and 5% in DS4. For all initial DS, the most frequent transition path was to regress to DS0 (becoming non-symptomatic) and the second most common was to stay in the initial disease state. For subjects ending a time period in DS1, 89% regressed to DS0 in the next time period, while 6% stayed in DS1, 3% progressed to DS2, 1% progressed to DS3, and 1% progressed to DS4. Multivariate modeling of risk factors influencing transitions showed that progressing from DS1 is associated with age >70, a proton pump inhibitor prescription, and absence of a diagnostic procedure. CONCLUSIONS: The Markov analysis showed that subjects with GERD commonly have their symptoms regress, with only a small percent progressing. The Markov model is a useful methodology to research disease states in a retrospective database setting within a healthcare plan.