DEFINING MULTI-AGENT CHEMOTHERAPY REGIMENS USING CLAIMS DATA ANALYSIS: IS THERE A ROLE FOR DATA VISUALIZATION?

In the oncology setting, health services researchers often utilize claims-based algorithms to identify complex chemotherapy regimens and lines of therapy. The algorithms build on prior investigation of claims data and/or electronic medical records (EMR) or on prior published work. However, prior studies do not typically provide the methodological detail that permits reproduction or supports evaluation of the algorithm, which can limit opportunities to improve on a key measure for cancer studies. We show how the use of data visualization software provides longitudinal views, supports the identification of multi-agent chemotherapy regimens using claims data, and provides transparency. Recommendation: we propose incorporating data visualization during the process of algorithm development as follows: (a) based on clinical guidelines, clinician input, and relevant literature, identify the regimens and their constituent medications. (b) Using claims data, flag the occurrence of any of these medications following diagnosis. (c) On a random subset of the event-level data, characterize the longitudinal sequence of chemotherapeutic agents using data visualization software and compare actual patterns against expected patterns. (d) Define and describe regimens in terms of constituent medications, duration, and time between regimens. (e) Using [c] and [d], develop and apply a preliminary algorithm that specifies the medications, the acceptable period of time for treatment administration within the regimen, the acceptable period of time between regimens. (f) Confirm the face validity of the graphical and summarized quantitative output from [c], [d], and [e] with a practicing oncologist/specialist. (g) Modify and expand the algorithm as needed given the desired objective (e.g., to define first-line therapy or multiple lines of therapy). Conclusion: Compared to an approach that utilizes primarily claims and/or EMR data to develop algorithms for identifying treatment with multi-agent chemotherapy regimens, the incorporation of data visualization provides an efficient way to elucidate the embedded structure and assumptions of claims-based algorithms.