421 Artificial Intelligence and Natural Language Processing for Automated Coding of Cervical and Lumbar Spine Surgery

INTRODUCTION: Artificial intelligence (AI) in healthcare offers substantial opportunities to improve outcomes, enhance efficiencies, and reduce cost. AI is prime to disrupt legacy healthcare processes like coding and billing where up to $175 billion of revenue is lost due to undercoding, rejected claims, and administrative cost. METHODS: Adult patients undergoing cervical or lumbar spine surgery between January 2012-December 2020 at an academic medical center were included for analysis. A 60:20 stratified split was undertaken to create training and testing populations. nNLPA coding accuracy was tested against human coders, using a third party “super coder” as a control. nNLPA performance was assessed via F1 scores, and contingency tables were used to determine sensitivity, specificity, PPV and NPV. RESULTS: 200 operative reports were assessed in this study. 192 CPT codes (88 cervical, and 104 lumbar) were identified from these operative reports. The nNLPA and human coders’ weighted mean F1 score for coding lumbar operative notes were 0.84 and 0.56 (P<0.05), respectively. Mean sensitivity, specificity, and accuracy of nNLPA coding were 0.79, 0.99, and 0.98, respectively, and 0.59, 0.97, and 0.96 (P<0.05) for human coders. The nNLPA and human coders’ weighted mean F1 score for cervical operative notes were 0.73 and 0.68 (P<0.05), respectively. nNLPA coding mean specificity and accuracy were 0.99, and 0.95 (P<0.05), respectively, and 0.89, and 0.89 (P<0.05) for human coders. CONCLUSIONS: nNLPA performance is non-inferior to humans at spine surgery medical coding. This indicates promising results for the integration of AI in spine surgery and other clinical applications. Further studies are needed to quantify cost savings and revenue uplift associated with utilizing an NLP platform for coding compared to humans.