OBJECTIVES: A patient’s temporal pattern of medication adherence may contain signals for clinical risks that are not captured by aggregate measures such as the medication possession ratio (MPR).  We used hierarchical clustering to identify natural adherence patterns to dornase alfa in patients with cystic fibrosis (CF), and assessed whether these patterns were associated with hospitalization risk.  METHODS: CF patients with a dornase alfa prescription followed by ≥1 year of health plan enrollment were identified in a national claims database (2005 - 2011). Hierarchical clustering was used to identify common patterns of adherence over time based on prescription fills. To evaluate the clinical and economic meaningfulness of the clusters, their association with CF-related hospitalization risk was assessed using Poisson regression with adjustment for MPR and other characteristics.  RESULTS: A total of 985 CF patients with ≥1 prescription for dornase alfa were included.  Average MPR was 45%.  Half of the patients experienced hospitalization. Clustering identified six adherence patterns: 1) high; 2) low→ high; 3) low→ lower; 4) high → low; 5) periodic; and 6) low.  MPR distributions overlapped between clusters.   In the Poisson regression, each 1% increase in MPR was significantly (p<0.05) associated with an approximate 1% reduction in hospitalization risk.  Including effects of cluster membership into the model with MPR significantly improved fit (P<0.001).  Even after adjusting for MPR, hospitalization risk varied significantly across clusters, and, relative to cluster 1, was increased by 97%, 115%, 52%, 286% and 57% in clusters 2 through 6, respectively.  The predictive significance of the clusters was robust to further adjustment for quadratic effects of MPR.    CONCLUSIONS:  In this study of CF patients, temporal patterns of medication adherence predicted hospitalization risk more accurately than MPR alone.  Temporal adherence patterns beyond MPR may have clinical and economic utility across therapeutic areas and warrant further study.