Browse Prior Art Database

Confidence Score Assignment for Section-Level Evaluation and Management (E/M) Coding

IP.com Disclosure Number: IPCOM000243506D
Publication Date: 2015-Sep-28
Document File: 4 page(s) / 142K

Publishing Venue

The IP.com Prior Art Database

Abstract

A method is presented for displaying multiple overlapping and interacting levels of confidence scores in automated processing results (based on, for example, machine learning or natural language processing) in a way that focuses a human evaluator's attention on the relevant section of the documentation. As an example of the process, Evaluation and Management medical coding ("E/M Coding") is examined in detail, paying attention to the confidence of sub-tasks, the interaction of those sub-tasks in determining the top-level result, and an implicit deficiency analysis of the automatic processing.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 30% of the total text.

Page 01 of 4

Confidence Score Assignment for Section-Level Evaluation and Management (E/M) Coding

Abstract

A method is presented for displaying multiple overlapping and interacting levels of confidence scores in automated processing results (based on, for example, machine learning or natural language processing) in a way that focuses a human evaluator's attention on the relevant section of the documentation. As an example of the process, Evaluation and Management medical coding ("E/M Coding") is examined in detail, paying attention to the confidence of sub-tasks, the interaction of those sub-tasks in determining the top-level result, and an implicit deficiency analysis of the automatic processing.

Introduction

Evaluation and Management (E/M) medical coding is more complex than, for example, outpatient radiology coding. The precise details of E/M coding are known in the industry and thus in this paper, a somewhat simplified model of E/M coding is used. There are several sections that are scored independently based on various criteria, and each section is a assigned a "section score" from 1 to 5 representing, roughly, the thoroughness or complexity of the examination and documentation. The individual scores are cumulative, in that a score of 5 subsumes the complexity of a score of 4, and so on. The final score for the document is selected as the minimum of the individual section scores.

The criteria used in the various sections range from simply counting indications of body parts,
to a fairly subjective evaluation of complex risk. As a result, some sections are more amenable than others to very accurate automated processing. This, coupled with the relatively high monetary value of E/M codes (and the difference in payment between different final code levels), results in a situation where the likelihood of an entire document being accurately coded to be processed without review by a human coder is fairly low. By contrast, with outpatient radiology 50-80% of documents may be auto-coded without human review.

The goal of automated processing isn't necessarily to have, say, 50% of documents auto-coded without human review, but rather to reduce the workload of human coders by 50%. In outpatient radiology, these are more or less the same perspectives. In E/M coding, the workload reduction goal can be accomplished by saving the human coder 50% of the work in each note.

The approach described for displaying the computed confidence with respect to the results of the automatic coding of an E/M note is not to report the top-level confidence that the whole note is correct, or even the confidence in the individual section scores, but rather to report on the individual levels within each section score, and to illuminate the relevant interactions between them so that the human coder can review the evidence where deficiencies in the automated coding are likely to matter most.

Implementation

Assume the following values:


1. 85% confidence is "confident"

2. 95% conf...