Systems and methods for analyzing patient health records
Techniques for analyzing patient health records are provided. Clinical documents may be received in response to a patient-initiated request, for example. In one embodiment, machine learning algorithms are used to sectionalize and extract data from clinical documents. The machine learning algorithms used may be more highly focused for analyzing text residing deeper in a clinical document hierarchy, for example. In one embodiment, extracted data is stored in a patient graph. Searches may be made against the graph to yield results to help save lives and/or improve patient outcomes.
1. A method for extracting information from clinical documents, the method comprising:
receiving, by one or more processors, a clinical document, wherein the clinical document comprises one or more sections;
identifying, by the one or more processors, a plurality of significant signatures, wherein each significant signature corresponds to a different section from the one or more sections in the clinical document, wherein the plurality of significant signatures comprises one or more of a set of predefined formatting properties for each of the different sections;
determining, by the one or more processors, a plurality of clinical section classifications based on the plurality of significant signatures;
predicting, by a first machine learning model, a clinical section classification from the plurality of clinical section classifications for each section in the one or more sections based on the significant signature of the each section and text of the each section;
identifying, by a second machine learning model, clinical entities from each section in the one or more sections based on the predicted clinical section classification; and
generating, by a third machine learning model, a clinical statement for each clinical section classification in the plurality of clinical section classifications based on the clinical entities.
2. The method of claim 1 , further comprising storing, by the one or more processors, the clinical statement for the each clinical section classification in the plurality of clinical section classifications in a structured clinical data record.
3. The method of claim 2 , wherein the structured clinical data record is stored in a searchable data structure, wherein the searchable data structure is a graph data structure comprising nodes corresponding to a plurality of clinical statements and edges corresponding to a relationship between the nodes.
4. The method of claim 1 , wherein the clinical entities are identified by a terminology service.
5. The method of claim 1 , wherein the identifying the clinical entities comprises:
for each section in the one or more sections, determining, based on the predicted clinical section classification, a natural language processing (NLP) pipeline from a plurality of NLP pipelines, wherein the NLP pipeline is configured to identify text associated with the predicted clinical section classification for the section; and
extracting, by the NLP pipeline, the clinical entities from the text.
6. The method of claim 5 , wherein the plurality of NLP pipelines comprises at least one of a patient pipeline, a document classification pipeline, a clinical pipeline, and a cancer pipeline.
7. The method of claim 1 , wherein the identifying the plurality of significant signatures is performed by a fourth machine learning model.
8. A system for extracting information from clinical documents, the system comprising:
at least one processor; and
a memory having instructions stored thereon that, when executed by the at least one processor, cause the at least one processor to:
receive a clinical document, wherein the clinical document comprises one or more sections;
identify a plurality of significant signatures, wherein each significant signature corresponds to a different section from the one or more sections in the clinical document, wherein the plurality of significant signatures comprises one or more of a set of predefined formatting properties for each of the different sections;
determine a plurality of clinical section classifications based on the plurality of significant signatures;
predict, by a first machine learning model, a clinical section classification from the plurality of clinical section classifications for each section in the one or more sections based on the significant signature of the each section and text of the each section;
identify, by a second machine learning model, clinical entities from each section in the one or more sections based on the predicted clinical section classification; and
generate, by a third machine learning model, a clinical statement for each clinical section classification in the plurality of clinical section classifications based on the clinical entities.
9. The system of claim 8 , the memory further comprising storing the clinical statement for the each clinical section classification in the plurality of clinical section classifications in a structured clinical data record.
10. The system of claim 9 , wherein the structured clinical data record is stored in a searchable data structure, wherein the searchable data structure is a graph data structure comprising nodes corresponding to a plurality of clinical statements and edges corresponding to a relationship between the nodes.
11. The system of claim 8 , wherein the clinical entities are identified by a terminology service.
12. The system of claim 8 , wherein the identifying the clinical entities comprises:
for each section in the one or more sections, determining, based on the predicted clinical section classification, an NLP pipeline from a plurality of NLP pipelines, wherein the NLP pipeline is configured to identify text associated with the predicted clinical section classification for the section; and
extracting, by the NLP pipeline, the clinical entities from the text.
13. The system of claim 12 , wherein the plurality of NLP pipelines comprises at least one of a patient pipeline, a document classification pipeline, a clinical pipeline, and a cancer pipeline.
14. The system of claim 8 , wherein the identifying the plurality of significant signatures is performed by a fourth machine learning model.
15. A non-transitory computer-readable medium (CRM) having stored therein computer-readable instructions executable to cause a computer system to perform operations comprising:
receiving, by one or more processors, a clinical document, wherein the clinical document comprises one or more sections;
identifying, by the one or more processors, a plurality of significant signatures, wherein each significant signature corresponds to a different section from the one or more sections in the clinical document, wherein the plurality of significant signatures comprises one or more of a set of predefined formatting properties for each of the different sections;
determining, by the one or more processors, a plurality of clinical section classifications based on the plurality of significant signatures;
predicting, by a first machine learning model, a clinical section classification from the plurality of clinical section classifications for each section in the one or more sections based on the significant signature of the each section and text of the each section;
identifying, by a second machine learning model, clinical entities from each section in the one or more sections based on the predicted clinical section classification; and
generating, by a third machine learning model, a clinical statement for each clinical section classification in the plurality of clinical section classifications based on the clinical entities.
16. The non-transitory CRM of claim 15 , further comprising storing, by the one or more processors, the clinical statement for the each clinical section classification in the plurality of clinical section classifications in a structured clinical data record.
17. The non-transitory CRM of claim 16 , wherein the structured clinical data record is stored in a searchable data structure, wherein the searchable data structure is a graph data structure comprising nodes corresponding to a plurality of clinical statements and edges corresponding to a relationship between the nodes.
18. The non-transitory CRM of claim 15 , wherein the clinical entities are identified by a terminology service.
19. The non-transitory CRM of claim 15 , wherein the identifying the clinical entities comprises:
for each section in the one or more sections, determining, based on the predicted clinical section classification, a natural language processing (NLP) pipeline from a plurality of NLP pipelines, wherein the NLP pipeline is configured to identify text associated with the predicted clinical section classification for the section; and
extracting, by the NLP pipeline, the clinical entities from the text.
20. The non-transitory CRM of claim 15 , wherein the identifying the plurality of significant signatures is performed by a fourth machine learning model.