Can recursive partitioning empirically derive a pulmonary function test interpretive algorithm

Abstract

Background: Electronic decision support (EDS) tools are time consuming to develop and validate. A classification and regression tree technique (CART) known as recursive partitioning (RP) has proven beneficial in the development of EDS tools. In this study, we wanted to determine whether RP could be used to validate the logic of an EDS tool for Pulmonary Function Test (PFT) interpretation, ePFT. Objectives: Determine whether Recursive Partitioning (RP) can be used to empirically derive a PFT interpretive algorithm. Compare the derived decision tree to the current published guidelines, on which ePFT was based. Methods: 4,906 PFTs obtained as part of clinical care at Intermountain and their ePFT interpretations were analyzed using RP. We selected 55 input variables from the PFT database. The output variable was the PFT interpretation by ePFT. Results: The preferred tree exhibited the same logic as the published ATS/ERS PFT interpretation schema. Additionally, it reflects real-world clinical PFT interpretations and accounted for PFT interpretations where TLC was not measured and restriction could not be definitively diagnosed. Conclusions: Recursive partitioning can derive PFT interpretation logic based solely on input PFT variables and the final PFT interpretation. This approach may facilitate development and validation of EDS tools.