Browse Prior Art Database

Hierarchical Support Vector Machines (SVM) for One-vs-All Computer Assisted Coding

IP.com Disclosure Number: IPCOM000243455D
Publication Date: 2015-Sep-22
Document File: 5 page(s) / 143K

Publishing Venue

The IP.com Prior Art Database

Abstract

Medical codesets for assigning diagnoses or procedures typically have an inherent hierarchical structure in which topically related codes are grouped logically within the codeset. A modified architecture is presented that breaks the code predication into two steps to model these insights.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 27% of the total text.

Page 01 of 5

Hierarchical Support Vector Machines (SVM) for One-vs-All Computer Assisted Coding

Abstract

Medical codesets for assigning diagnoses or procedures typically have an inherent hierarchical structure in which topically related codes are grouped logically within the codeset. A modified architecture is presented that breaks the code predication into two steps to model these insights.

Introduction

While it is typical for multiple diagnosis and procedure codes to be assigned to a given document, it is frequently the case that within a hierarchical grouping, only one code is ever appropriate. Conventional "one-vs-all" (OVA) architectures treat each code label as an independent class, and thus are unable to exploit the inherent code hierarchical structure.

The modified architecture presented here comprises two steps. The first "multi-label" step (in which one or more labels can be assigned) identifies a set of "hierarchical category groups" that are appropriate for the document. The second "multiclass" step (in which only one label can be assigned) predicts the appropriate billing code for each predicted category group from the first step.

Further presented is a method that uses codeset structure along with code co- occurrence statistics in order to infer an effective two level hierarchy for use within this hybrid architecture which can be applied to almost any medical codeset.

Medical codesets generally have hierarchical structure that roughly proceeds in layers of added specificity. For instance, perhaps a codeset may start with very general distinctions like "disease" vs. "symptom", then if "disease" is chosen it may further break up codes by body system, then by whether the agent is a virus or bacteria, and so on, until a very specific condition is described. As an example, Table 1 depicts how the ICD9CM codeset handles "viral hepatitis."

Table 1 - ICD9CM Viral Hepatitis Codeset

Code Description

Incomplete Code? (1=Yes, 0=No)

070 Viral hepatitis

1


070.0 Viral hepatitis A with hepatic coma 0


070.1 Viral hepatitis A without mention of hepatic coma 0


070.2 Viral hepatitis B with hepatic coma 1

070.20 Viral hepatitis B with hepatic coma, acute or unspecified, without mention of hepatitis delta 0

070.21 Viral hepatitis B with hepatic coma, acute or unspecified, with hepatitis delta 0

070.22 Chronic viral hepatitis B with hepatic coma without hepatitis delta 0


Page 02 of 5

070.23 Chronic viral hepatitis B with hepatic coma with hepatitis delta 0


070.3 Viral hepatitis B without mention of hepatic coma 1

070.30 Viral hepatitis B without mention of hepatic coma, acute or unspecified, without mention of hepatitis delta 0

070.31 Viral hepatitis B without mention of hepatic coma, acute or unspecified, with hepatitis delta 0

070.32 Chronic viral hepatitis B without mention of hepatic coma without mention of hepatitis delta 0

070.33 Chronic viral hepatitis B without mention of hepatic coma with hepatitis delta 0


070.4 Other specified viral hepatiti...