OR-0106 - Development And Validation Of A Machine Learning Algorithm To Identify Anaphylaxis In Us Administrative Claims Data
ON DEMAND
Daniel C Beachler, Devon H Taylor, Mary S Anthony, Ruihua Yin, Ling Li, Catherine W Saltus, Lin Li, Alka Shaunik, Kathleen E Walsh, Stephan Lanes, Kenneth J Rothman, Catherine Johannes, Vanita Aroda, Warner Carr, Pinkus Goldberg, Andrew Accardi, J Shane O'Shura, Kristen Sharma, Juhaeri Juhaeri, and Chuntao Wu
• Boolean code-based claims algorithms for anaphylaxis are prone to false positive and/or false negative errors that can bias inferences in comparative safety studies • This study included medical record validation of possible anaphylaxis cases through clinical adjudication • The predictive model algorithm for anaphylaxis developed using machine learning techniques in this study had a positive predictive value (PPV) of 94% and with little loss of sensitivity compared to the screening algorithm (92%) • Machine learning techniques yielded an anaphylaxis algorithm that achieved a substantially higher PPV than the conventional screening algorithm while retaining a similar number of cases (true positives). • This algorithm could be considered in future safety studies using similar claims data to reduce potential outcome misclassification.
