Medical report coding with acronym/abbreviation disambiguation
Techniques for coding a medical report include identifying an acronym or abbreviation in the medical report, and a plurality of phrases not explicitly included in the medical report that are possible expanded forms of the acronym or abbreviation in the medical report. From the plurality of phrases, a most likely expanded form of the acronym or abbreviation may be selected by applying to the medical report a statistical acronym/abbreviation expansion model trained on a corpus of medical reports. By applying to the medical report with the expanded acronym or abbreviation one or more statistical fact extraction models, a clinical fact may be extracted from the medical report based at least in part on the most likely expanded form of the acronym or abbreviation in the medical report, and a corresponding medical taxonomy code may be assigned to the extracted clinical fact from the medical report.
1. A method comprising:
analyzing free-form input regarding an encounter with a patient by a clinician with at least one processor to determine at least one clinical fact expressed in the free-form input, wherein the analyzing comprises analyzing the free-form input using the at least one processor and at least one statistical model to determine the at least one clinical fact, the at least one statistical model having been trained on a corpus of medical text to identify clinical facts expressed within input to the at least one statistical model, wherein analyzing the free-form input using the at least one processor and the at least one statistical model comprises:
selecting with the at least one statistical model, for a sequence of initials in the free-form input and from among a plurality of phrases that are not explicitly included in the free-form input and that are different expanded forms of the sequence of initials in the free-form input, a phrase that is likely to be an intended expanded form of the sequence of initials in the free-form input; and
analyzing the free-form input and the phrase, of the plurality of phrases, not included in the free-form input but determined by the at least one statistical model as likely to be the intended expanded form of the sequence of initials, to determine the at least one clinical fact expressed in the free-form input, wherein the analyzing comprises:
when the phrase determined by the at least one statistical model as likely to be the intended expanded form corresponds to a non-codable entity, refraining from extracting a clinical fact from a portion of the free-form input that includes the sequence of initials, and
when the phrase determined by the at least one statistical model as likely to be the intended expanded form does not correspond to a non-codable entity, extracting a clinical fact from a portion of the free-form input based at least in part on processing the phrase not included in the free-form input with the at least one statistical model implemented by the at least one processor.
2. The method of claim 1 , further comprising assigning a corresponding medical taxonomy code to the extracted clinical fact from the free-form input.
3. The method of claim 1 , wherein selecting with the at least one statistical model a phrase that is likely to be an intended expanded form of the sequence of initials in the free-form input comprises:
applying to the free-form input a statistical expansion model implemented by the at least one processor executing stored instructions and trained on a corpus of medical input to determine likely intended expanded forms of sequences of initials processed with the statistical expansion model.
4. The method of claim 1 , wherein processing the free-form input using the at least one processor and the at least one statistical model comprises:
extracting unigrams and bigrams from a specified window of text around the sequence of initials in the free-form input; and
analyzing the extracted unigrams and bigrams as features using the at least one statistical model.
5. The method of claim 1 , wherein processing the free-form input using the at least one processor and the at least one statistical model comprises:
extracting a section header of a section including the sequence of initials in the free-form input; and
analyzing the extracted section header as a feature using the at least one statistical model.
6. The method of claim 1 , wherein processing the free-form input using the at least one processor and the at least one statistical model comprises:
identifying a document type of the free-form input; and
analyzing the identified document type as a feature using the at least one statistical model.
7. A system comprising:
at least one processor; and
at least one storage medium storing processor-executable instructions that, when executed by the at least one processor, perform a method comprising:
analyzing free-form input regarding an encounter with a patient by a clinician with at least one processor to determine at least one clinical fact expressed in the free-form input, wherein the analyzing comprises analyzing the free-form input using the at least one processor and at least one statistical model to determine the at least one clinical fact, the at least one statistical model having been trained on a corpus of medical text to identify clinical facts expressed within input to the at least one statistical model, wherein analyzing the free-form input using the at least one processor and the at least one statistical model comprises:
selecting with the at least one statistical model, for a sequence of initials in the free-form input and from among a plurality of phrases that are not explicitly included in the free-form input and that are different expanded forms of the sequence of initials in the free-form input, a phrase that is likely to be an intended expanded form of the sequence of initials in the free-form input; and
analyzing the free-form input and the phrase, of the plurality of phrases, not included in the free-form input but determined by the at least one statistical model as likely to be the intended expanded form of the sequence of initials, to determine the at least one clinical fact expressed in the free-form input, wherein the analyzing comprises:
when the phrase determined by the at least one statistical model as likely to be the intended expanded form corresponds to a non-codable entity, refraining from extracting a clinical fact from a portion of the free-form input that includes the sequence of initials, and
when the phrase determined by the at least one statistical model as likely to be the intended expanded form does not correspond to a non-codable entity, extracting a clinical fact based at least in part on processing the phrase not included in the free-form input with the at least one statistical model implemented by the at least one processor.
8. The system of claim 7 , further comprising assigning a corresponding medical taxonomy code to the extracted clinical fact from the free-form input.
9. The system of claim 7 , wherein selecting with the at least one statistical model a phrase that is likely to be an intended expanded form of the sequence of initials in the free-form input comprises:
applying to the free-form input a statistical expansion model implemented by the at least one processor executing stored instructions and trained on a corpus of medical input to determine likely intended expanded forms of sequences of initials processed with the statistical expansion model.
10. The system of claim 7 , wherein processing the free-form input using the at least one processor and the at least one statistical model comprises:
extracting unigrams and bigrams from a specified window of text around the sequence of initials in the free-form input; and
analyzing the extracted unigrams and bigrams as features using the at least one statistical model.
11. The system of claim 7 , wherein processing the free-form input using the at least one processor and the at least one statistical model comprises:
extracting a section header of a section including the sequence of initials in the free-form input; and
analyzing the extracted section header as a feature using the at least one statistical model.
12. The system of claim 7 , wherein processing the free-form input using the at least one processor and the at least one statistical model comprises:
identifying a document type of the free-form input; and
analyzing the identified document type as a feature using the at least one statistical model.
13. At least one non-transitory computer-readable storage medium storing computer-executable instructions that, when executed, perform a method comprising:
analyzing free-form input regarding an encounter with a patient by a clinician with at least one processor to determine at least one clinical fact expressed in the free-form input, wherein the analyzing comprises analyzing the free-form input using the at least one processor and at least one statistical model to determine the at least one clinical fact, the at least one statistical model having been trained on a corpus of medical text to identify clinical facts expressed within input to the at least one statistical model, wherein analyzing the free-form input using the at least one processor and the at least one statistical model comprises:
selecting with the at least one statistical model, for a sequence of initials in the free-form input and from among a plurality of phrases that are not explicitly included in the free-form input and that are different expanded forms of the sequence of initials in the free-form input, a phrase that is likely to be an intended expanded form of the sequence of initials in the free-form input; and
analyzing the free-form input and the phrase, of the plurality of phrases, not included in the free-form input but determined by the at least one statistical model as likely to be the intended expanded form of the sequence of initials, to determine the at least one clinical fact expressed in the free-form input, wherein the analyzing comprises:
when the phrase determined by the at least one statistical model as likely to be the intended expanded form corresponds to a non-codable entity, refraining from extracting a clinical fact from a portion of the free-form input that includes the sequence of initials, and
when the phrase designated determined by the at least one statistical model as likely to be the intended expanded form does not correspond to a non-codable entity, extracting a clinical fact based at least in part on processing the phrase not included in the free-form input with the at least one statistical model implemented by the at least one processor.
14. The computer-readable storage medium of claim 13 , further comprising assigning a corresponding medical taxonomy code to the extracted clinical fact from the medical report.
15. The computer-readable storage medium of claim 13 , wherein selecting with the at least one statistical model a phrase that is likely to be an intended expanded form of the sequence of initials in the free-form input comprises:
applying to the free-form input a statistical expansion model implemented by the at least one processor executing stored instructions and trained on a corpus of medical input to determine likely intended expanded forms of sequences of initials processed with the statistical expansion model.
16. The computer-readable storage medium of claim 13 , wherein processing the free-form input using the at least one processor and the at least one statistical model comprises:
extracting unigrams and bigrams from a specified window of text around the sequence of initials in the free-form input; and
analyzing the extracted unigrams and bigrams as features using the at least one statistical model.
17. The computer-readable storage medium of claim 13 , wherein processing the free-form input using the at least one processor and the at least one statistical model comprises:
extracting a section header of a section including the sequence of initials in the free-form input; and
analyzing the extracted section header as a feature using the at least one statistical model.
18. The method of claim 1 , wherein:
selecting the phrase that is likely to be an intended expanded form of the sequence of initials, from among the plurality of phrases that are different expanded forms of the sequence of initials, comprises:
analyzing the sequence of initials and at least a portion of the free-form input using the at least one statistical model and generating one or more confidence scores for each of the plurality of phrases, and
selecting, from the plurality of phrases, the phrase that is likely to be the intended expanded form of the sequence of initials in the free-form input based at least in part on the one or more confidence scores for each of the plurality of phrases, the phrase that is likely to be the intended expanded form being one of the plurality of phrases.
19. The system of claim 7 , wherein:
selecting the phrase that is likely to be an intended expanded form of the sequence of initials, from among the plurality of phrases that are different expanded forms of the sequence of initials, comprises:
analyzing the sequence of initials and at least a portion of the free-form input using the at least one statistical model and generating one or more confidence scores for each of the plurality of phrases, and
selecting, from the plurality of phrases, the phrase that is likely to be the intended expanded form of the sequence of initials in the free-form input based at least in part on the one or more confidence scores for each of the plurality of phrases, the phrase that is likely to be the intended expanded form being one of the plurality of phrases.
20. The computer-readable storage medium of claim 13 , wherein:
selecting the phrase that is likely to be an intended expanded form of the sequence of initials, from among the plurality of phrases that are different expanded forms of the sequence of initials, comprises:
analyzing the sequence of initials and at least a portion of the free-form input using the at least one statistical model and generating one or more confidence scores for each of the plurality of phrases, and
selecting, from the plurality of phrases, the phrase that is likely to be the intended expanded form of the sequence of initials in the free-form input based at least in part on the one or more confidence scores for each of the plurality of phrases, the phrase that is likely to be the intended expanded form being one of the plurality of phrases.
21. The method of claim 1 , further comprising training the at least one statistical model to recognize text contexts in which the sequence of initials stands for the non-codable entity, based on false positives noted in user corrections to previously coded medical reports.
22. The method of claim 2 , wherein assigning the corresponding medical taxonomy code to the extracted clinical fact comprises identifying a code, from a medical taxonomy selected from the group consisting of SNOMED, RxNorm, RadLex, ICD, CPT, E&M, MedDRA, LOINC, and NDC taxonomies of codes, corresponding to the phrase that is likely to be the intended expanded form of the sequence of initials in the free-form input.