Extraction, representation, and cognitive interpretation of medically relevant evidence

1. A method comprising:

receiving a corpus of documents associated with treatment of a first medical disorder;

generating, by operation of one or more processors, a plurality of relative efficacy structures for a plurality of therapies by, for at least a first respective document in the corpus of documents:

identifying and extracting, using one or more natural language processing techniques, a conclusion specified within natural language text of the respective document, wherein the conclusion includes a comparison between two or more therapies of the plurality of therapies;

determining, for each respective therapy of the plurality of therapies, a respective sentiment of the extracted conclusion by using one or more natural language processing techniques, wherein at least one determined sentiment is classified as neutral; and

identifying one or more cohorts of patients that are associated with the conclusion; and

generating a knowledge graph based on the plurality of relative efficacy structures, comprising:

identifying a first, second, and third therapy, of the plurality of therapies, specified by a first relative efficacy structure of the plurality of relative efficacy structures;

upon determining that the first therapy is not represented in the knowledge graph, generating and inserting a node for the first therapy;

upon determining that an edge does not exist between the first therapy and the second therapy in the knowledge graph, generating and inserting an edge connecting nodes corresponding to the first and second therapies, based on the first relative efficacy structure; and

upon determining that an edge exists between the second therapy and the third therapy in the knowledge graph, updating a weight of the edge based on a weight and directionality specified by the first relative efficacy structure, wherein updating the weight causes the edge between the second and third therapies to change directions.

2. The method of claim 1 , wherein receiving the corpus of documents comprises querying a larger corpus based on the first medical disorder, such that the generated corpus of documents includes documents that are clinically relevant to the first medical disorder.

3. The method of claim 1 , wherein identifying the conclusion of the respective document comprises:

identifying comparative language in the respective document; and

determining a section in the respective document where the comparative language is located, wherein generating the knowledge graph is based further on the section in the respective document where the comparative language is located.

4. The method of claim 1 , wherein extracting the conclusion of the respective document comprises:

identifying an unknown term, wherein the unknown term is either (i) an ambiguous term, or (ii) an acronym; and

analyzing the respective document to determine a meaning for the unknown term.

5. The method of claim 1 , wherein determining the sentiment of the extracted conclusion with respect to the at least one therapy comprises classifying the extracted conclusion as positive, negative, or neutral.

6. The method of claim 1 , the method further comprising determining one or more publication characteristics of the respective document.

7. The method of claim 6 , wherein generating the knowledge graph comprises:

aggregating the plurality of relative efficacy structures, wherein each of the plurality of relative efficacy structures is weighted based on the publication characteristics of the respective document from which the relative efficacy structure was generated.

8. The method of claim 1 , wherein the comparison between two or more therapies is with respect to one or more outcome types, of a plurality of outcome types, wherein the plurality of outcome types is associated with a predefined hierarchy.

9. The method of claim 8 , the method further comprising:

receiving a set of patient attributes for a first patient;

determining a cohort of the first patient;

identifying a plurality of potential therapies for the first patient by parsing the knowledge graph based on the cohort of the first patient;

ranking the plurality of potential therapies, based on a set of relative efficacy structures associated with each of the plurality of potential therapies and based further on the predefined hierarchy; and

providing the ranked plurality of potential therapies to a user.

10. A computer program product comprising a computer-readable storage medium having computer-readable program code embodied therewith, the computer-readable program code executable by one or more computer processors to perform an operation comprising:

receiving a corpus of documents associated with treatment of a first medical disorder;

generating a plurality of relative efficacy structures for a plurality of therapies by, for at least a first respective document in the corpus of documents:

identifying and extracting, using one or more natural language processing techniques, a conclusion specified within natural language text of the respective document, wherein the conclusion includes a comparison between two or more therapies of the plurality of therapies;

determining, for each respective therapy of the plurality of therapies, a respective sentiment of the extracted conclusion by using one or more natural language processing techniques, wherein at least one determined sentiment is classified as neutral; and

identifying one or more cohorts of patients that are associated with the conclusion; and

generating a knowledge graph based on the plurality of relative efficacy structures, comprising:

identifying a first, second, and third therapy, of the plurality of therapies, specified by a first relative efficacy structure of the plurality of relative efficacy structures;

upon determining that the first therapy is not represented in the knowledge graph, generating and inserting a node for the first therapy;

upon determining that an edge does not exist between the first therapy and the second therapy in the knowledge graph, generating and inserting an edge connecting nodes corresponding to the first and second therapies, based on the first relative efficacy structure; and

upon determining that an edge exists between the second therapy and the third therapy in the knowledge graph, updating a weight of the edge based on a weight and directionality specified by the first relative efficacy structure, wherein updating the weight causes the edge between the second and third therapies to change directions.

11. The computer program product of claim 10 , wherein identifying the conclusion of the respective document comprises:

identifying comparative language in the respective document; and

determining a section in the respective document where the comparative language is located, wherein generating the knowledge graph is based further on the section in the respective document where the comparative language is located.

12. The computer program product of claim 10 , wherein extracting the conclusion of the respective document comprises:

identifying an unknown term, wherein the unknown term is either (i) an ambiguous term, or (ii) an acronym; and

analyzing the respective document to determine a meaning for the unknown term.

13. The computer program product of claim 10 , wherein determining the sentiment of the extracted conclusion with respect to the at least one therapy comprises classifying the extracted conclusion as positive, negative, or neutral.

14. The computer program product of claim 10 , wherein the comparison between two or more therapies is with respect to one or more outcome types, of a plurality of outcome types, wherein the plurality of outcome types is associated with a predefined hierarchy.

15. The computer program product of claim 14 , the operation further comprising:

receiving a set of patient attributes for a first patient;

determining a cohort of the first patient;

identifying a plurality of potential therapies for the first patient by parsing the knowledge graph based on the cohort of the first patient;

ranking the plurality of potential therapies, based on a set of relative efficacy structures associated with each of the plurality of potential therapies and based further on the predefined hierarchy; and

providing the ranked plurality of potential therapies to a user.

16. A system comprising:

one or more computer processors; and

a memory containing a program which when executed by the one or more computer processors performs an operation, the operation comprising:

receiving a corpus of documents associated with treatment of a first medical disorder;

generating a plurality of relative efficacy structures for a plurality of therapies by, for at least a first respective document in the corpus of documents:

identifying and extracting, using one or more natural language processing techniques, a conclusion specified within natural language text of the respective document, wherein the conclusion includes a comparison between two or more therapies of the plurality of therapies;

determining, for each respective therapy of the plurality of therapies, a respective sentiment of the extracted conclusion by using one or more natural language processing techniques, wherein at least one determined sentiment is classified as neutral; and

identifying one or more cohorts of patients that are associated with the conclusion; and

generating a knowledge graph based on the plurality of relative efficacy structures, comprising:

identifying a first, second, and third therapy, of the plurality of therapies, specified by a first relative efficacy structure of the plurality of relative efficacy structures;

upon determining that the first therapy is not represented in the knowledge graph, generating and inserting a node for the first therapy;

upon determining that an edge does not exist between the first therapy and the second therapy in the knowledge graph, generating and inserting an edge connecting nodes corresponding to the first and second therapies, based on the first relative efficacy structure; and

upon determining that an edge exists between the second therapy and the third therapy in the knowledge graph, updating a weight of the edge based on a weight and directionality specified by the first relative efficacy structure, wherein updating the weight causes the edge between the second and third therapies to change directions.

17. The system of claim 16 , wherein identifying the conclusion of the respective document comprises:

identifying comparative language in the respective document; and

determining a section in the respective document where the comparative language is located, wherein generating the knowledge graph is based further on the section in the respective document where the comparative language is located.

18. The system of claim 16 , wherein extracting the conclusion of the respective document comprises:

identifying an unknown term, wherein the unknown term is either (i) an ambiguous term, or (ii) an acronym; and

analyzing the respective document to determine a meaning for the unknown term.

19. The system of claim 16 , wherein determining the sentiment of the extracted conclusion with respect to the at least one therapy comprises classifying the extracted conclusion as positive, negative, or neutral.

20. The system of claim 16 , wherein the comparison between two or more therapies is with respect to one or more outcome types, of a plurality of outcome types, wherein the plurality of outcome types is associated with a predefined hierarchy, and wherein the operation further comprises:

receiving a set of patient attributes for a first patient;

determining a cohort of the first patient;

identifying a plurality of potential therapies for the first patient by parsing the knowledge graph based on the cohort of the first patient;

ranking the plurality of potential therapies, based on a set of relative efficacy structures associated with each of the plurality of potential therapies and based further on the predefined hierarchy; and

providing the ranked plurality of potential therapies to a user.