Consensus sequence identification

• US 10,545,997 B2
• Filed: 08/27/2018
• Issued: 01/28/2020
• Est. Priority Date: 01/14/2014
• Status: Active Grant

• Current Assignee: SymphonyAI Sensa LLC (Fortive Corp.)
• Original Assignee: Ayasdi Al LLC
• Inventors: Lum, Pek Yee, Cadag, Eithon, Grahnen, Johan, Lewis, Joshua, Sexton, Harlan
• Primary Examiner(s): Rostami, Mohammad S
• Application Number:
  US16/113,809
• Publication Number:
  US 20190005114A1
• Time in Patent Office: 519 Days
• CPC Class Codes: G06F 16/285 Clustering or classification
• CPC Class Codes: G06F 16/955 using information identifiers, e.g. uniform resource locators [URL]
• CPC Class Codes: G16H 10/60 for patient-specific data, e.g. for electronic patient records
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

An example method comprises receiving historical information of episodes, constructing event sets from the historical information, categorizing each event with general labels and synthetic labels, learning an event metric on the events by using the general and synthetic labels to perform dimensionality reduction to associate a vector with each event and to determine an angle between every two vectors, determining an event set metric using distances between each pair of event sets, deriving a sequence metric on the episodes, the sequence metric obtaining a preferred match between two episodes, deriving a subsequence metric on the episodes, the subsequence metric is a function of the event set metric on subsequences of each episode, grouping episodes into subgroups based on distances, for at least one subgroup, generating a consensus sequence by finding a preferred sequence of events, and the episodes of the subgroup, and generating a report indicating the consensus sequence.

Claims

1. A method comprising:

receiving historical information of past episodes, each past episode including at least one sequence of events taken over a period of time;

constructing event sets from the historical information each of the event sets including at least one sequence of events;

categorizing each event from the historical information with general event category labels and synthetic event category labels, at least one of the synthetic event category labels being broader than the at least one general event category label, the at least one of the synthetic event category labels categorizing events which are also categorized by the at least one general event category label and categorizing at least one event not categorized by the at least one general event category label;

learning an event metric on the events by using the general event category labels and the synthetic event category labels to perform dimensionality reduction to associate a vector with each event and to determine an angle between every two vectors, each general event category label and each synthetic event category label being assigned a separate dimension;

determining an event set metric using distances between each pair of event sets using the event metric;

deriving a sequence metric on the episodes to compute distances between episodes, the sequence metric obtaining a preferred match between two episodes with respect to a cost function describing a weighting for the event set metric;

deriving a subsequence metric on the episodes to compute the distances between episodes, wherein the subsequence metric is a function of the event set metric on subsequences of each episode;

grouping episodes into subgroups based on distances obtained using the sequence metric and the subsequence metric;

for at least one subgroup, generating a consensus sequence by finding a preferred sequence of events with respect to a function of the sequence metric and the subsequence metric between the preferred sequence and the episodes of the at least one subgroup; and

generating a report indicating the consensus sequence.

2. The method of claim 1, wherein categorizing each event with general event category labels comprises retrieving an ontology in the historical information and using the ontology to determine the general event category labels.

3. The method of claim 1, wherein the preferred match between two episodes is an optimal match.

4. The method of claim 1, wherein the sequence metric is a carepath (CP) metric.

5. The method of claim 1, wherein the subsequence metric is an event set carepath metric (ESCP) metric.

6. The method of claim 1, wherein the function of the event set metric is a weighted sum.

7. The method of claim 1, wherein each subsequence is defined relative to one or more anchor points in the related episode.

8. The method of claim 1, wherein each event includes a plurality of events.

9. The method of claim 8, wherein an order of the plurality of actions of at least one of the events is not distinguishable.

10. The method of claim 1, wherein constructing event sets from the historical information comprises constructing sets of events separated by no more than a predetermined period of time.

11. The method of claim 1, further comprising filtering the events to remove events that happen infrequently.

12. A non-transitory computer readable medium comprising executable instructions that are executable by a processor to perform a method, the method comprising:

receiving historical information of past episodes, each past episode including at least one sequence of events taken over a period of time;

constructing event sets from the historical information, each of the event sets including at least one sequence of events;

categorizing each event from the historical information with gene-al event category labels and synthetic event category labels, at least one of the synthetic event category labels being broader than the at least one general event category label, the at least one of the synthetic event category labels categorizing events which are also categorized by the at least one general event category label and categorizing at least one event not categorized by the at least one general event category label;

learning an event metric on the events by using the general event category labels and the synthetic event category labels to perform dimensionality reduction to associate a vector with each event and to determine an angle between every two vectors, each general event category label and each synthetic event category label being assigned a separate dimension;

determining an event set metric using distances between each pair of event sets using the event metric;

deriving a sequence metric on the episodes to compute distances between episodes the sequence metric obtaining a preferred match between two episodes with respect to a cost function describing a weighting for the event set metric;

deriving a subsequence metric on the episodes to compute the distances between episodes, wherein the subsequence metric is a function of the event set metric on subsequences of each episode;

grouping episodes into subgroups based on distances obtained using the sequence metric and the subsequence metric;

for at least one subgroup, generating a consensus sequence by finding a preferred sequence of events with respect to a function of the sequence metric and the subsequence metric between the preferred sequence and the episodes of the at least one subgroup; and

generating a report indicating the consensus sequence.

13. The computer readable medium of claim 12, wherein categorizing each event with general event category labels comprises retrieving an ontology in the historical information and using the ontology to determine the general event category labels.

14. The computer readable medium of claim 12, wherein the preferred match between two episodes is an optimal match.

15. The computer readable medium of claim 12, wherein the sequence metric is a carepath (CP) metric.

16. The computer readable medium of claim 12, wherein the subsequence metric is an event set carepath metric (ESCP) metric.

17. The computer readable medium of claim 12, wherein the function of the event set metric is a weighted sum.

18. The computer readable medium of claim 12, wherein each subsequence is defined relative to one or more anchor points in the related episode.

19. The computer readable medium of claim 12, wherein each event includes a plurality of events.

20. The computer readable medium of claim 19, wherein an order of the plurality of actions of at least one of the events is not distinguishable.

21. The computer readable medium of claim 12, wherein constructing event sets from the historical information comprises constructing sets of events separated by no more than a predetermined period of time.

22. A system comprising:

at least one processor;

memory comprising instructions to configure at least one of the at least one processor to;

receiving historical information of past episodes, each past episode including at least one sequence of events taken over a period of time;

construct event sets from the historical information, each of the event sets including at least one sequence of events;

categorize each event from the historical information with general event category labels and synthetic event category labels, at least one of the synthetic event category labels being broader than the at least one general event category label, the at least one of the synthetic event category labels categorizing events which are also categorized by the at least one general event category label and categorizing at least one event not categorized by the at least one general event category label;

learn an event metric on the events by using the general event category labels and the synthetic event category labels to perform dimensionality reduction to associate a vector with each event and to determine an angle between every two vectors, each general event category label and each synthetic event category label being assigned a separate dimension;

determine an event set metric using distances between each pair of event sets using the event metric;

derive a sequence metric on the episodes to compute distances between episodes, the sequence metric obtaining a preferred match between two episodes with respect to a cost function describing a weighting for the event set metric;

derive a subsequence metric on the episodes to compute the distances between episodes, wherein the subsequence metric is a function of the event set metric on subsequences of each episode;

group episodes into subgroups based on distances obtained using the sequence metric and the subsequence metric;

for at least one subgroup, generate a consensus sequence by finding a preferred sequence of events with respect to a function of the sequence metric and the subsequence metric between the preferred sequence and the episodes of the at least one subgroup; and

generate a report indicating the consensus sequence.