INTRODUCTION: Although clinical trials indicate that oral bisphosphonates reduce osteoporotic fracture risk, compliance with bisphosphonate therapy in practice is suboptimal, with 1-year discontinuation rates exceeding 50%. METHODS: We conducted a retrospective cohort study among female members of a large integrated health care delivery system (Kaiser Permanente of Northern California), age 45 years and older, to determine their persistence with weekly alendronate (defined as continuous use, allowing for a refill gap of 60 days), predictors of discontinuation, and subsequent osteoporosis therapy. We also examined the effect of varying the refill gap from 30 to 120 days on the discontinuation rate. From 2002 through 2003, we identified 13,455 women (age 68.8+/-10.4 years) who initiated weekly oral alendronate therapy. RESULTS: Using a 60-day refill gap, the 1-year discontinuation rate was 49.6% [95% confidence interval (CI) 48.8-50.4%]; this increased to 58.0% (CI 57.2-58.8%) with a 30-day gap and decreased to 42.2% (CI 41.1-43.0%) with a 120-day gap. Among those who discontinued therapy, about one-third restarted alendronate or another osteoporosis drug within 6 months. Baseline factors associated with alendronate discontinuation included prior bone mineral density testing [adjusted odds ratio (OR) 0.64, CI 0.60-0.69], prior postmenopausal hormone therapy (OR 0.78, CI 0.73-0.84), prior high-dose oral glucocorticoid therapy (OR 1.26, CI 1.05-1.51), prior gastrointestinal diagnoses (OR 1.21, CI 1.09-1.36), and high number of therapeutic classes of prescriptions filled in the prior year (OR 1.21, CI 1.10-1.32), although the final model had limited explanatory power. CONCLUSIONS: We conclude that apparent discontinuation rates are high within 1 year after treatment initiation, although a subset of women appears to restart bisphosphonate or other osteoporosis therapy. Because intermittent use and/or poor adherence is common, discontinuation rates based on data from administrative databases are sensitive to the refill gap length. In addition, we identified no clinical factors highly predictive of discontinuation.