C72-26 An Automated Software Application Compared With Clinician Interpretation of Cardiopulmonary Exercise Tests

Abstract Rationale There are well-recognized inconsistencies in the performance of cardiopulmonary exercise testing (CPET) in terms of protocol design, data processing and reporting. Furthermore, data collation and interpretation of results can take a physiologist and a physician up to 40 minutes. We evaluated a novel software application (enhancedCPETanalytics™, Los Angeles, USA) that processes raw data from CPET and generates an organized report in less than 10 seconds and compared its determinations with clinician interpretations. Methods We selected 151 CPET studies performed sequentially in Cambridge. They were divided into seven diagnostic categories according to previous clinical interpretations. Data were collected using a metabolic cart (Vyntus™ CPX) and exported via the diagnostic software platform (SentrySuite™, Jaeger Medical GmbH, Hochberg, Germany) as spreadsheets with prespecified format. The data were blinded according to diagnostic category and processed in Los Angeles using the novel software application. Studies were analyzed using the two methods by blinded investigators. Outliers, notably those values with differences between the two centers outside 95% CI, were reviewed by blinded investigators and the need for corrections was identified. Key numerical values were compared using Bland-Altman statistics. Key categorical data were compared using Cohen’s kappa statistics. This investigation was approved under a data sharing agreement between our two universities. Results Subject age was 51.1 (17.6) years (mean, SD) and 55% were female. A metabolic threshold could not be detected by clinical or automated interpretation in three studies. Outliers corrected by more than what we considered the Minimal Clinically Important Difference for the variable were: Cambridge 53, enhancedCPETanalytics 17. Statistical comparisons are shown in Table 1. For numerical data we saw significant correlation and small mean bias by Bland-Altman analysis between the two analytical methods, more so for maximal values than for slopes and thresholds. Identification of normal metabolic efficiency, functional impairment, and cardiovascular limitation showed substantial to near perfect agreement by Cohen’s kappa. Conclusions Close agreement was seen for all numerical variables and key categorical variables when comparing the novel software application with clinician interpretation. The application minimizes analytical subjectivity, standardizes interpretation, and improves reporting time-efficiency. Such qualities are essential for the wider acceptance and recognition of the clinical usefulness of CPET. This abstract is funded by: none