A Method for Modeling Co-Occurrence Propensity of Clinical Codes with Application to ICD-10-PCS Auto-Coding

摘要

Objective. Natural language processing methods for medical auto-coding, orautomatic generation of medical billing codes from electronic health records,generally assign each code independently of the others. They may thus assigncodes for closely related procedures or diagnoses to the same document, evenwhen they do not tend to occur together in practice, simply because the rightchoice can be difficult to infer from the clinical narrative. Materials and Methods. We propose a method that injects awareness of thepropensities for code co-occurrence into this process. First, a model istrained to estimate the conditional probability that one code is assigned by ahuman coder, given than another code is known to have been assigned to the samedocument. Then, at runtime, an iterative algorithm is used to apply this modelto the output of an existing statistical auto-coder to modify the confidencescores of the codes. Results. We tested this method in combination with a primary auto-coder forICD-10 procedure codes, achieving a 12% relative improvement in F-score overthe primary auto-coder baseline. Discussion. The proposed method can be used, with appropriate features, incombination with any auto-coder that generates codes with different levels ofconfidence. Conclusion. The promising results obtained for ICD-10 procedure codes suggestthat the proposed method may have wider applications in auto-coding.