System for automated analysis of clinical text for pharmacovigilance

Patent metadata

Field	Value
Publication number	US-10614196-B2
Application number	US-201514826575-A
Country	US
Kind code	B2
Filing date	Aug 14, 2015
Priority date	Aug 14, 2014
Publication date	Apr 7, 2020
Grant date	Apr 7, 2020

Abstract

In the pharmaceutical research and development process, it may be necessary to process large amounts of medical records or clinical literature, to ensure safety of patients consuming a drug. A pharmacovigilance system may assist in this process by efficiently and automatically processing medical records to extract information and relationships contained therein and may also form a preliminary assessment regarding a medical or clinical judgment. The pharmacovigilance system may automatically generate reports based on this information, which may be validated by trained clinicians and medical experts.

First claim

What is claimed is:

1. A system comprising:
a server comprising a processor, memory coupled with the processor, a plurality of databases coupled with the processor, and a data interface;
the memory having stored thereon processor executable instructions comprising data virtualization and query instructions, ontology framework instructions, search framework instructions, natural language processing pipeline instructions, and rules engine instructions, where
the data virtualization and query instructions, when executed by the processor, cause communication with the plurality of databases, the databases including:
a medical record database, where the medical record database contains medical records,
a curated reference and clinical knowledge database, a causal and temporal model database, and a curated drug safety and case database,

wherein the ontology framework instructions cause the processor to build and manage a custom ontology;
the search framework instructions cause the processor to provide search capabilities for the system,
wherein the natural language processing pipeline instructions, the ontology framework instructions, and data virtualization and query instructions cause the processor to:
automatically detect an update to at least one of the medical records;
access unstructured textual data from the at least one of the medical records stored in one or more of the medical record database and curated reference and clinical knowledge database;
identify semantic classes for terms in the unstructured textual data by matching the terms in the unstructured textual data to attributes associated with the semantic classes, wherein the semantic classes comprise a medical substance class, a medical condition class, and a time class, the time class representative of a time value term included in the unstructured textual data, wherein a medical substance term associated with the medical substance class and a medical condition term associated with the medical condition class are in different sentences of the unstructured textual data;
obtain relationship patterns associated with semantic relationships from the curated reference and clinical knowledge database;
identify one or more semantic relationships in the unstructured textual data by matching one or more identified semantic classes to at least one of the relationship patterns;
identify one or more discourse parameters in the unstructured textual data, where the discourse parameter includes one or more temporal discourse parameters indicative of a temporal order of the medical substance term and the medical condition term in the different sentences of the unstructured textual data with respect to the time value term, wherein the discourse parameter does not include the time value term;
generate a causal and temporal model of the unstructured textual data, the causal and temporal model comprising a list of notation markers and temporal markers arranged in a structured format, the notation markers representative of the identified semantic relationships, the temporal markers indicative of the temporal order of the terms in the different sentences of the unstructured textual data, the temporal markers attached to the notation markers; and
store the causal and temporal model in the causal and temporal model database, and
wherein the rules engine instructions and the data virtualization and query instructions cause the processor to:
access the causal and temporal model from the causal and temporal model database; and
determine a clinical event based on execution of a plurality of rule instructions linked together, wherein each of the rule instructions cause the processor to:
access the list of notation markers and temporal markers of the causal and temporal model;
determine the list of notation markers and temporal markers of the causal and temporal model match a pattern of notation markers and temporal markers; and
select the clinical event from a plurality of predetermined clinical event classifications in response to the list of notation markers and temporal markers of the causal and temporal model matching the predetermined pattern of notation markers and temporal markers; and

wherein the data interface comprises intelligent text analysis platform instructions, stored in memory, that, when executed by the processor, cause the processor to, in response to detection of the update to the at least one of the medical records and determination of the clinical event:
generate an intelligent text analysis user interface, the intelligent text analysis user interface comprising a temporal chain user interface, the temporal chain user interface comprising a plurality of medical labels descriptive of the identified semantic relationships, the medical labels sequentially arranged on the temporal chain user interface based on the temporal markers of the causal and temporal model; and
transmit the user interface to a display device for a user assigned to the medical record.

2. The system of claim 1 , where the natural language processing pipeline instructions cause the processor to convert the medical information into a standard canonical format, and process the unstructured textual data of the medical information in the standard canonical format to generate the causal and temporal model.

3. The system of claim 1 , where, in order to generate the causal and temporal model, the natural language processing pipeline instructions further cause the processor to:
identify a causal discourse parameter in the unstructured textual data representative of a causal relationship between the terms in the unstructured textual data;
attach a casual marker to the at least one of the notation markers, the causal marker corresponding to the causal discourse parameter; and
determine an occurrence model for the identified semantic relationships, where the occurrence model includes a combination of the notation markers, temporal markers, and causal markers; and
generate the causal and temporal model based on the occurrence model, where the causal and temporal model comprises one or more causal chains inferred based on the combination of the notation markers, temporal markers, and causal markers included in the occurrence model.

4. The system of claim 3 , where, in order to generate the causal and temporal model, the natural language processing pipeline instructions further cause the processor to connect an antecedent semantic relationship with a consequent semantic relationship when the discourse parameter is a conditional discourse parameter.

5. The system of claim 1 , where, in order to generate the causal and temporal model, the natural language processing pipeline instructions further cause the processor to:
identify an additional discourse parameter in the unstructured textual data, where the additional discourse parameters comprise a severity parameter having a degree, a certainty parameter having a polarity, a course parameter reflecting a nature of treatment, a state parameter reflecting a current state, or a combination thereof; and
attach an additional marker to the at least one of the semantic markers, the additional marker corresponding to the additional discourse parameter.

6. The system of claim 1 , where:
the curated reference and clinical knowledge database includes medical information to be processed, a coding base contains information regarding drug classifications under one or more classification systems, and the curated drug safety and case database contains information regarding reactions experienced by patients consuming a drug;
the natural language processing pipeline instructions further cause the processor to identify at least a patient semantic class, an adverse reaction semantic class, a predicate semantic class, a drug sema

Assignees

• Accenture Global Services Ltd

Inventors

• Maitra Anutosh
• Mohamedrasheed Annervaz Karukapadath
• Jain Tom Geo
• Shivaram Madhura
• Sengupta Shubhashis
• Ramnani Roshni Ramesh
• Pathak Neetu
• Banerjee Debapriya
• Sahu Vedamati

Classifications

• G16H15/00

  ICT specially adapted for medical reports, e.g. generation or transmission thereof · CPC title

• G16H10/20

  for electronic clinical trials or questionnaires · CPC title

• G16H50/70

  for mining of medical data, e.g. analysing previous cases of other patients · CPC title

• G16H50/20

  for computer-aided diagnosis, e.g. based on medical expert systems · CPC title

• G16H70/40

  relating to drugs, e.g. their side effects or intended usage · CPC title