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 a computer system, machine-readable versions of said clinical documents;
sectionalizing the clinical documents based on a plurality of machine learning (ML) algorithms, wherein one or more first ML algorithms classify different sections of the clinical documents based on significant signatures corresponding to the different sections, wherein the significant signatures comprise one or more of a set of predefined formatting properties for each of the different sections, and wherein one or more second ML algorithms verify said classifications of the different sections based on text associated with respective sections of the different sections;
transforming text associated with the sections into a plurality of structured clinical data records, wherein different words or phrases of text for each section are classified using a plurality of third ML algorithms selected based on said classification of each section, and wherein different combinations of classified words or phrases of the text for each respective section are mapped to a particular structured clinical data record; and
storing information from the structured clinical data records in a searchable data structure.
2. The method of claim 1 wherein receiving said clinical documents is in response to a patient-initiated request.
3. The method of claim 1 wherein sectionalizing comprises selecting the one or more second ML algorithms from a plurality of ML algorithms based on the classification by the one or more first ML algorithms.
4. The method of claim 3 wherein sectionalizing further comprises processing text associated with different sections having different classifications using different natural language processing (NLP) algorithms.
5. The method of claim 3 wherein the one or more ML algorithms selected are one or more of: a clinical entity natural language processing (NLP) algorithm, a cancer entity NLP algorithm, a demographic entity NLP algorithm, a demographic ML model, a provider entity NLP algorithm, and a provider ML model.
6. The method of claim 1 wherein the one or more second ML algorithms comprise a combination of at least one ML model and at least one natural language processing (NLP) algorithm.
7. The method of claim 1 wherein the one or more clinical documents include one or more sections comprising a plurality of subsections, wherein one or more of the plurality of third ML algorithms classify different subsections of the clinical documents based on subsection significant signatures corresponding to the different subsections, and wherein one or more fourth ML algorithms verify said classification of the different subsections based on text associated with each respective subsection.
8. The method of claim 1 , wherein text associated with sections having a first classification is classified using a first set of natural language processing (NLP) algorithms of the plurality of third ML algorithms and wherein text associated with sections having a second classification is classified using a second set NLP algorithms of the plurality of third ML algorithms.
9. The method of claim 1 , the plurality of third ML algorithms selected based on said classification of each section comprising one or more of: a clinical entity natural language processing (NLP) algorithm, a cancer entity NLP algorithm, a demographic entity NLP algorithm, a demographic ML model, a provider entity NLP algorithm, and a provider ML model.
10. The method of claim 1 , the plurality of third ML algorithms selected based on said classification of each section comprising a first natural language processing (NLP) algorithm for recognizing clinical entities comprising one or more of: medications, conditions, labs, procedures, diagnosis, vitals, allergies, treatments and genomic alterations.
11. The method of claim 1 , wherein mapping comprises sending the classified words or phrases of the text for each respective section to a terminology service to determine medical codes.
12. The method of claim 1 wherein the structured clinical data records comprise one or more medical codes retrieved from a data store based on one or more classifications of the words or phrases of the text, and wherein the one or more medical codes are associated with one or more of the words or phrases of the text.
13. The method of claim 1 wherein one or more natural language processing (NLP) algorithms for recognizing clinical entities is included in said one or more second ML algorithms and said plurality of third ML algorithms.
14. The method of claim 1 wherein the searchable data structure is a graph data structure.
15. The method of claim 14 wherein the graph data structure comprises nodes corresponding to patients and nodes corresponding to one or more of: a patient, a procedure, a condition, a provider, and a medication.
16. The method of claim 14 wherein the graph data structure comprises edges describing a relationship between a patient node and another node corresponding to a clinical concept or clinical domain of a plurality of predefined clinical concepts or clinical domains.
17. The method of claim 14 wherein the graph data structure comprises edges having at least one probability attribute.
18. The method of claim 14 further comprising modifying the probabilities of a plurality of edges based on analysis of edges between a same type of nodes.
19. A non-transitory computer readable medium having stored thereon program code executable by a computer system, the program code comprising:
code that causes the computer system to receive machine-readable versions of said clinical documents;
code that causes the computer system to sectionalize the clinical documents based on a plurality of machine learning (ML) algorithms, wherein one or more first ML algorithms classify different sections of the clinical documents based on significant signatures corresponding to the different sections, wherein the significant signatures comprise one or more of a set of predefined formatting properties for each of the different sections, and wherein one or more second ML algorithms verify said classifications of the different sections based on text associated with respective sections of the different sections;
code that causes the computer system to transform text associated with the sections into a plurality of structured clinical data records, wherein different words or phrases of text for each section are classified using a plurality of third ML algorithms selected based on said classification of each section, and wherein different combinations of classified words or phrases of the text for each respective section are mapped to a particular structured clinical data record; and
code that causes the computer system to store information from the structured clinical data records in a searchable data structure.
20. A computer system comprising:
at least one processor; and
a memory having stored thereon program code that, when executed by the at least one processor, causes the processor to:
receive machine-readable versions of said clinical documents;
sectionalize the clinical documents based on a plurality of machine learning (ML) algorithms, wherein one or more first ML algorithms classify different sections of the clinical documents based on significant signatures corresponding to the different sections, wherein the significant signatures comprise one or more of a set of predefined formatting properties for each of the different sections, and wherein one or more second ML algorithms verify said classifications of the different sections based on text associated with respective sections of the different sections;
transform text associated with the sections into a plurality of structured clinical data records, wherein different words or phrases of text for each section are classified using a plurality of third ML algorithms selected based on said classification of each section, and wherein different combinations of classified words or phrases of the text for each respective section are mapped to a particular structured clinical data record; and
store information from the structured clinical data records in a searchable data structure.