Online Program

Network analyses of prescription switching among bisphosphonates (302472)

 View Presentation

Keywords: Prescription switching, network analysis, bisphosphonate

Network analysis is a technique that combines computer science and network theory to describe and summarize relational structures and uncover powerful images of reality. While network analysis has gained increasing popularity in many disciplines, its application in describing prescription utilization is novel. Prescription switching patterns, which typically involve large and informative medication-to-medication interconnections, can be constructed and described using network structures. This study considers bisphosphonate prescriptions for osteoporosis as one of the many possible medications that can be extracted from a large claims database. Description of utilization patterns is demonstrated through the use of three classic centrality measures (degree, betweeness, and closeness centrality), quadratic assignment procedure (QAP) regressions, and network graphics. All networks were constructed in and analyzed with UCINET 6 (v. 6.385, Borgatti et al. 2006) with NetDraw package. Frequencies of prescription-to-prescription switching patterns were recorded by using 9 x 9 matrices, representing 9 different prescription drugs. The major finding was that for patients taking a bisphosphonate and switching to another bisphosphonate, Actonel, Alendronate Sodium, and Boniva are most likely to be the “winner” not only because they have a net gain in patients in the long term, but also acting as both transitional and highly interchangeable drugs. QAP regression was used to test the drug-specific attributes (dose frequency, copay amount, whether the drug was generic and the proportion of the prescription drugs on the formulary list) with the frequencies of switching. While the lack of explanatory variables in the data prevent us from identifying key factors that contribute to the overall decision pattern of switching, QAP regression is considered as a beneficial technique when more comprehensive data were available. Lastly, the network graphs were used to visualize and enhance our recognition of the relational structure and position of the prescription drugs. Our result shows that drugs with similar chemical ingredients would appear together as a “clique” more often than drugs that have different molecular formula, suggesting that the brand-name drugs are highly interchangeable with their generic counterparts. These methods have added a great value in analyzing asymmetric relational data which could not have been more straight-forward in other conventional statistical tools. In this study, we demonstrated that analogy can be drawn between the former definitions of network analysis and data interpretation. The results of this study can also be used to evaluate patients’ drug exposure, physicians’ prescribing patterns and facilitates interpretations for further researches. In addition, the visualization tool can aid as a powerful tool to facilitate communications among scientific and industrial communities. It is conceivable that the area of network analysis is increasing envisioned in future research methods as the methodology derived from network and graphical theory is known for its versatility to accommodate researches for various purposes.