Applying sequence clustering techniques to explore practice-based ambulatory care pathways in insurance claims data

Background: Care pathways are a widely used mean to ensure well-coordinated and high quality care by defining the optimal timing and interval of health services for a specific indication. However, evidence on common sequences of services actually followed by patients has rarely been quantified. This study aims to explore whether sequence clustering techniques can be used to empirically identify typical treatment sequences in ambulatory care for heart failure (HF) patients and compare their effectiveness. Methods: Routine data of HF patients were provided by a large statutory sickness fund in Germany from 2009 until 2011. Events were categorized by either (i) the specialty of the physician, (ii) the type of service/procedure provided and (iii) the medication prescribed. Similarities between sequences were measured using the 'longest common subsequence' (LCS). The k-medoids clustering algorithm was applied to identify distinct subgroups of sequences. We used logistic regression to identify the most effective sequences for avoiding hospitalizations. Results: Treatment data of 982 incident HF patients were analyzed to identify typical treatment sequences. The cluster analysis revealed three distinct clusters of specialty sequences, four clusters of procedure sequences and four clusters of prescription sequences. Clusters differed in terms of timing and interval of physician visits, procedures and drug prescriptions as well as comorbidities and HF hospitalization rates. We found no significant association between cluster membership and HF hospitalization. Conclusions: Sequence clustering techniques can be used as an explorative tool to systematically extract, describe compare and analyze treatment sequences and associated characteristics.