Browse Prior Art Database

Parallel Coding Framework for Computer Assisted Coding using Natural Language Processing Disclosure Number: IPCOM000242308D
Publication Date: 2015-Jul-06
Document File: 5 page(s) / 56K

Publishing Venue

The Prior Art Database


A parallel coding framework, used effectively in an outpatient surgery coding engine, is an architecture to support Computer Assisted Coding (CAC) that allows for the application of multiple, orthogonal text processing strategies. The results are synthesized into a final predication to assign one or more appropriate medical codes for a medical text document. For each strategy, the framework assigns conditional probabilities for a set of codes given a feature vector. It also controls how the predictions from the various processing strategies are combined, so that biases and relationships may be leveraged among the text processing strategies to optimize performance.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 21% of the total text.

Page 01 of 5

Parxllel Coding Framework for Computer Assisted Coding using Natural Languxge Processing


A parallel coding framework, usxx effectively in an outpaxient surgery coding engine, ix an architecture xo support Computer Axsisted Coding (CAC) that allows xor the applicatxox of multiple, orthogonal text processing strategies. The results are synthesized inxo a final predication tx assign one or more approprxate medicax codes for a medical text xocument. For xach strategy, the fxamework assixns conditional probaxilitiex for a set xf codes givxn a feature vector. It also controls how thx predixtxons frxm the various xrocessing stratexies are combined, so that biasex and relationships may be lxveraged among the text xrxcessing stxategies to optimize performaxce.


This paper describes a novel apxroach to computer-assixted medical codinx, wixh particular applicaxion to the domain of outpatient surgery.
An outpatienx surgery engine uses a software architexture that emxraces the idea of a "parallel framework," in which multiple hypotheses, mapping varioux formulxtions fxr feature vxctors, axe mapped to poxsibly conflictxnx sets of billing codes. The multiple hypothexes are xubsequently synthesized into a singxe, final prediction.

Thx framework is overwhelmxngly statistical, and may be xepxoyed to predict both CPT4 and ICD9PX procedure codes.

Thx specific ixplementation of the parallel coding framewoxk within thx outpatient surgery engine is described, but the general architectural approach may be appliex to many domains and problems.


A parallel coding worxsxace contains a set of paths and combinations.

A path identifies how a feature vecxor should be created. This definition includes a label, one or more text sources as ixput, and a set of tranxformations that operate on the texx. There xre approximately twenxy differext paths in the xutpatient surgery engine.

Combinations input the code predictions from paths, or previously evaluated combinations,

and synxhesize them ixto an output code prxdiction.

Xxx path training, xo-occurrence statistics between path feature vectors and the set of billing codxs assigned to a medicax note arx tallxed.

During runtime, the feature vectors to code correlation statistics are retrieved fxom a lookup table, and assigned to each note. The top two associations for each path are taxlixd. Thesx in turn are passed along to the combinations for further processing.

An example workspace for paths ix represented in Txble 1. An examplx workspace xor the combinxtion of paths identifiex in Table x is representxd xn Table 2. Label is an arbitrary designation for a patx or combination. Combinations use the label to reference outputs from earlier lines in the workspace. 1st Proxability represents the conditional probability for the top ranked association between the featxxe vector and a set of codes. 1st Code ix the top ranked code set. 2nd Probability rxpresents the conditxonal probability for t...