System and methods for generating predictive combinations of hospital monitor alarms

1. A system for monitoring data associated with a plurality of physiological characteristics of a patient, comprising:

a processor configured for receiving input from a plurality of independent physiological data streams from one or more monitoring devices;

wherein each of the physiological data streams comprises an individual raw alarm associated with the physiological data within the stream; and

programming executable on the processor for:

reading the input physiological data streams;

comparing the input data stream against a set of predictive super-alarm patterns associated with the plurality of physiological conditions;

the predictive super-alarm patterns comprising a temporal window of data acquired from a combination of at least two individual encoded raw alarms from independent physiological data streams that co-occur within the temporal window; and

triggering an alarm if a combination of the input physiological data matches at least a portion of a generated predictive super-alarm pattern;

wherein the triggered alarm is predictive of an adverse event associated with the physiological characteristics of a patient.

2. A system as recited in claim 1 , wherein the triggered alarm is predictive of a code-blue event.

3. A system as recited in claim 1 , wherein the set of predictive super-alarm patterns are generated by:

acquiring physiological monitoring case data from raw alarm data acquired from a plurality of patient cases; and

detecting super-alarm patterns occurring within a temporal window preceding an adverse event within each of the patient cases;

wherein the detected adverse event within the case data is correlative to the adverse event associated with the physiological characteristics of the patient.

4. A system as recited in claim 3 , wherein the set of predictive super-alarm patterns are generated by:

acquiring physiological monitoring control data comprising raw alarm data from a plurality of control patients, the control patient alarm data comprising super-alarm patterns not occurring within the temporal window of an adverse event; and

discretizing a value field for raw alarms based on the case data and the control data.

5. A system as recited in claim 4 , wherein raw control alarms and case alarms that co-occur within a specified time window are pre-processed to generate an output of an array of alarm codes, each of which identifies an individual raw alarm.

6. A system as recited in claim 4 , wherein discretizing a value field comprises dividing an alarm with a value code into sub-codes.

7. A system as recited in claim 4 , wherein the control data is used to filter out non-predictive super-alarm patterns identified for adverse event patients that have also occurred above a specified frequently for control patients.

8. A system as recited in claim 4 :

wherein a super-alarm pattern is excluded from inclusion within the predictive super-alarm pattern set if the super-alarm pattern has false positive rate (FPR) greater than a predetermined threshold; and

wherein the false positive rate (FPR) of the super-alarm pattern is a function of the acquired control data.

9. A system as recited in claim 4 , wherein the set of predictive super-alarm patterns are generated by:

generating encoding schema;

wherein encoding schema used to encode alarm codes for case alarms and control alarms.

10. A system as recited in claim 9 , further comprising:

generating encoding schema as a function of an occurrence frequency for each alarm code.

11. A system as recited in claim 1 , wherein the set of predictive super-alarm patterns are generated by:

applying closed-set filtering to remove predictive super-alarm patterns that are included as a subset of a larger predictive super-alarm pattern.

12. A system for generating a set of predictive alarm patterns associated with a plurality of physiological characteristics of a patient, the predictive alarm patterns being predictive of an adverse event associated with the physiological characteristics of a patient, comprising:

a database comprising physiological monitoring case data acquired from raw alarm data from a plurality of patient cases and a plurality of control patients that are not coded with respect to the adverse event;

a processor;

programming executable on the processor for:

evaluating raw alarm data from a plurality of patient cases and a plurality of control patients; and

generating a set of predictive super-alarm patterns associated with the plurality of physiological conditions;

the predictive super-alarm patterns comprising a temporal window of data acquired from a combination of at least two individual encoded raw alarms from independent physiological data streams that co-occur within the temporal window.

13. A system as recited in claim 12 , wherein generating a set of predictive super-alarm patterns comprises:

detecting super-alarm patterns occurring within a temporal window preceding an adverse event within each of the patient cases;

wherein the detected adverse event within the case data is correlative to the adverse event associated with the physiological characteristics of the patient.

14. A system as recited in claim 12 , wherein the alarm data from the plurality of control patients comprises super-alarm patterns not occurring within the temporal window of an adverse event, and wherein generating a set of super-alarm patterns comprises:

discretizing a value field for raw alarms based on the case data and the control data.

15. A system as recited in claim 14 , wherein discretizing a value field comprises dividing an alarm with a value code into sub-codes.

16. A system as recited in claim 12 , wherein raw control alarms and case alarms that co-occur within a specified time window are pre-processed to generate an output of an array of alarm codes, each of which identifies an individual raw alarm.

17. A system as recited in claim 12 , wherein the control data is used to filter out non-predictive super-alarm patterns identified for adverse event patients that have also occurred above a specified frequently for control patients.

18. A system as recited in claim 12 :

wherein a super-alarm pattern is excluded from inclusion within the predictive super-alarm pattern set if the super-alarm pattern has a false positive rate (FPR) greater than a predetermined threshold;

wherein the false positive rate (FPR) of the super-alarm pattern is a function of the acquired control data.

19. A system as recited in claim 12 , wherein generating a set of predictive super-alarm patterns further comprises:

generating encoding schema;

wherein encoding schema is used to encode alarm codes for case alarms and control alarms.

20. A system as recited in claim 19 , further comprising:

generating encoding schema as a function of an occurrence frequency for each alarm code.

21. A system as recited in claim 12 , wherein generating a set of predictive super-alarm patterns further comprises:

applying closed-set filtering to remove predictive super-alarm patterns that are included as a subset of a larger predictive super-alarm pattern.

22. A monitor for predicting an adverse event associated with the physiological characteristics of a patient, comprising:

one or more monitor inputs configured for receiving input from a plurality of independent physiological data streams generated from one or more monitoring devices;

wherein each of the physiological data streams comprises an individual raw alarm associated with the physiological data within the stream;

a processor coupled to the one or more monitor inputs; and

programming executable on the processor for:

reading the input physiological data streams;

comparing the input data stream against a set of predictive super-alarm patterns associated with the plurality of physiological conditions;

the predictive super-alarm patterns comprising a temporal window of data acquired from a combination of at least two individual encoded raw alarms from independent physiological data streams that co-occur within the temporal window; and

triggering an alarm if a combination of the input physiological data matches at least a portion of a generated predictive super-alarm pattern;

wherein the triggered alarm is predictive of an adverse event associated with the physiological characteristics of a patient.

23. A monitor as recited in claim 22 , wherein the triggered alarm is predictive of a code-blue event.

24. A monitor as recited in claim 22 , wherein the set of predictive super-alarm patterns are generated by:

acquiring physiological monitoring case data from raw alarm data acquired from a plurality of patient cases; and

detecting super-alarm patterns occurring within a temporal window preceding an adverse event within each of the patient cases;

wherein the detected adverse event within the case data is correlative to the adverse event associated with the physiological characteristics of the patient.

25. A monitor as recited in claim 24 , wherein the set of predictive super-alarm patterns are generated by:

acquiring physiological monitoring control data comprising raw alarm data from a plurality of control patients, the control patient alarm data comprising super-alarm patterns not occurring within the temporal window of an adverse event; and

discretizing a value field for raw alarms based on the case data and the control data.

26. A monitor as recited in claim 25 , wherein raw control alarms and case alarms that co-occur within a specified time window are pre-processed to generate an output of an array of alarm codes, each of which identifies an individual raw alarm.

27. A monitor as recited in claim 25 , wherein discretizing a value field comprises dividing an alarm with a value code into sub-codes.

28. A monitor as recited in claim 25 , wherein the control data is used to filter out non-predictive super-alarm patterns identified for adverse event patients that have also occurred above a specified frequently for control patients.

29. A monitor as recited in claim 25 :

wherein a super-alarm pattern is excluded from inclusion within the predictive super-alarm pattern set if the super-alarm pattern has a false positive rate (FPR) greater than a predetermined threshold;

wherein the false positive rate (FPR) of the super-alarm pattern is a function of the acquired control data.

30. A monitor as recited in claim 25 , wherein the set of predictive super-alarm patterns are generated by:

generating encoding schema;

wherein encoding schema used to encode alarm codes for case alarms and control alarms.

31. A monitor as recited in claim 30 , further comprising:

generating encoding schema as a function of an occurrence frequency for each alarm code.

32. A monitor as recited in claim 22 , wherein the set of predictive super-alarm patterns are generated by:

applying closed-set filtering to remove predictive super-alarm patterns that are included as a subset of a larger predictive super-alarm pattern.