Probabilistic event classification systems and methods

• US 10,269,450 B2
• Filed: 05/21/2014
• Issued: 04/23/2019
• Est. Priority Date: 05/22/2013
• Status: Active Grant

• Current Assignee: RLDatix North America Inc. (DATIX (Holdings) Ltd.)
• Original Assignee: Quantros, Inc. (DATIX (Holdings) Ltd.)
• Inventors: Rhodes, Bryn Thomas, Childers, Randal Earl, Livaudais, Gerard Francis, Fenton, Stefanie, Thompson, Tim, Foley, Sandra, Greene, Jeffrey
• Primary Examiner(s): Cobanoglu, Dilek B
• Application Number:
  US14/284,329
• Publication Number:
  US 20140350964A1
• Time in Patent Office: 1,798 Days
• Field of Search: 705 2- 3
• CPC Class Codes: G16H 40/20 for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms

Abstract

Probabilistic event classifications systems and method are provided herein. In one embodiment, a method includes receiving an event narrative, the event narrative comprising textual content describing a safety event, parsing the textual content to identify key terms, searching a safety event database for classifications associated with the key terms, selecting a set of classifications based on the key terms using statistical analysis, the set of classifications comprising potential event types for the event narrative, and displaying the set of classifications for the event narrative via a graphical user interface.

Claims

1. A method for classifying a safety event occurring in a healthcare setting using a classification system comprising an intelligent probabilistic event classification server having a hardware processor and a memory for storing executable instructions, the hardware processor being configured to execute the instructions to perform the method, comprising:

receiving on an interactive graphical user interface an event narrative, the event narrative comprising textual content describing a safety event;

parsing the textual content to identify key terms,

the parsing comprising generating tokens from the event narrative by the hardware processor selecting a plurality of frequently-appearing tokens, the frequently-appearing tokens being tokens that appear most frequently in the event narrative,

the parsing further comprising calculating, for the event narrative, a probability for each of a plurality of classifications based on the frequently-appearing tokens, each frequently-appearing token having a corresponding probability for a classification and representing an interesting token,

the interesting token improving confidence and improving performance of the hardware processor of the intelligent probabilistic event classification server;

selecting a set of classifications, the set of classifications comprising potential event types for the event narrative, the selecting the set of classifications being based on the calculated probabilities;

the calculating probabilities comprising;

determining a number (NT) of occurrences of the token occurring in classified event narratives having a classification;

determining a number (NE) of classified event narratives having the classification;

determining a number (MT) of occurrences of the token relative to the classified event narratives not having the classification;

determining a number (ME) of events not associated with the classification; and

calculating a probability that a token is associated with a classification by;

2. The method of claim 1, the event narrative further comprising associated event metadata pertaining to one of:

a patient identity, a time of the safety event, and a place of the safety event.

3. The method of claim 2, the associated event metadata comprising at least one of patient age, time of event, care setting, and gender.

4. The method of claim 1, further comprising normalizing the event narrative, the normalizing comprising automatically identifying synonyms to the key terms, and searching a safety event database for additional classifications that are associated with the synonyms.

5. The method according to claim 1, wherein generating tokens comprises any of:

identifying each word in the event narrative as a token; and

identifying groups of adjacent words in the event narrative as a token.

6. The method according to claim 1, the key terms being more highly weighted in the probability calculation when the key terms are identified as medical terms from a medical dictionary.

7. The method according to claim 1, further comprising:

determining a probability (PK) that a token is indicative of a selected classification; and

calculating a probability that a classification has been correctly classified for the event narrative by;

8. The method according to claim 7, further comprising:

computing a set of tokens included in the event narrative for a classification; and

utilizing the equation of claim 7 to combine probabilities of tokens in the event narrative into a probability for each classification.

9. The method according to claim 1, the probability being derived from a training corpus having a frequency distribution of the frequently-appearing tokens across the plurality of classifications, further comprising verifying accuracy of the training corpus by:

obtaining a random sample of classified event narratives included in the training corpus;

reclassifying at least a portion of the classified event narratives;

comparing original classifications for the classified event narratives to the reclassifications; and

calculating a classification accuracy for the training corpus from the comparisons.

10. The method according to claim 9, further comprising:

determining a number of event narratives that could not be reclassified; and

grouping the number of event narratives into a default classification.

11. The method according to claim 9, further comprising determining a classification efficiency by:

obtaining a random sample of classified event narratives included in the training corpus;

determining a text length for each of the classified event narratives in the random sample;

reclassifying at least a portion of the classified event narratives in the random sample using the method of claim 1;

determining a classification time for each reclassification; and

based on the text length and the classification time of each classified event narrative in the random sample, determining an expected response time for classifying event narratives.

12. The method according to claim 1, further comprising analyzing one of a plurality of metrics for the event narrative, the plurality of metrics comprising a care setting, a department, a geographical location, a nature within an event type, a sub-nature within an event type, a sub-nature within an event type and a nature, an event type within a nature, and an event type within a nature and a sub-nature.

13. The method according to claim 12, further comprising adjusting a corresponding probability of a token having a classification based on the user selected classification.

14. The method of claim 1, further comprising normalizing the event narrative, the normalizing comprising automatically reducing multiple terms into a common key term, and searching a safety event database for additional classifications that are associated with the common key term.