Methods and systems for calculating and using statistical models to predict medical events

1. A computer-implemented method of generating a precursor pattern detection model for detecting an occurrence of a given type of medical event, the method comprising:

obtaining annotated monitoring data by a computing device, wherein the annotated monitoring data represents measurements of physical conditions of at least one first patient;

extracting, by the computing device, a set of positive segments from the annotated monitoring data, wherein each positive segment is associated with an occurrence of a medical event of the given type;

identifying, by the computing device, a portion of the annotated monitoring data that is at least three days away from the medical events of the annotated monitoring data along a timeline;

extracting, by the computing device, a set of negative segments from the annotated monitoring data, wherein each negative segment is not associated with an occurrence of the medical event of the given type, and wherein all of the set of negative segments are extracted from the portion of the annotated monitoring data that is at least three days away from the medical events of the annotated monitoring data along the timeline;

executing, by the computing device, a first filter on the set of positive segments to generate a set of filtered positive segments;

executing, by the computing device, a second filter on the set of negative segments to generate a set of filtered negative segments;

extracting, by the computing device, a set of extracted positive features from the set of filtered positive segments and a set of extracted negative features from the set of filtered negative segments;

selecting, by the computing device, a set of selected positive features from the set of extracted positive features and a set of selected negative features from the set of extracted negative features;

calculating, by the computing device, the precursor pattern detection model based on the set of selected positive features and the set of selected negative features;

storing, by the computing device on a nontransitory computer-readable medium, the precursor pattern detection model;

sensing a signal from a second patient to form at least a portion of collected patient data;

detecting a match between a set of the collected patient data and the precursor pattern detection model; and

in response to the match, using the computing device and informing a party to seek or to provide medical care.

2. The computer-implemented method of claim 1 , wherein the annotated monitoring data includes annotations indicating occurrences of the given medical event during collection of the monitoring data.

3. The computer-implemented method of claim 1 , wherein the annotated monitoring data includes a set of signals, and the set of selected positive features includes features extracted from less than all of the set of signals included in the annotated monitoring data.

4. The computer-implemented method of claim 1 , wherein executing the first filter on the set of positive segments to generate the set of filtered positive segments includes:

executing a top-down filter on the set of positive segments to generate a set of top-down filtered positive segments; and

executing a bottom-up filter on the set of top-down filtered positive segments to generate the set of filtered positive segments.

5. The computer-implemented method of claim 4 , wherein executing the first filter on the set of positive segments to generate the set of filtered positive segments includes:

repeatedly executing the top-down filter and the bottom-up filter until a strength of the filtered positive segments does not increase further.

6. The computer-implemented method of claim 1 , wherein extracting the set of extracted positive features from the set of filtered positive segments includes performing a wavelet transform for each filtered positive segment of the set of filtered positive segments.

7. The computer-implemented method of claim 6 , wherein the wavelet transform is a Haar wavelet transform.

8. The computer-implemented method of claim 1 , wherein extracting the set of extracted positive features from the set of filtered positive segments includes determining a change in a value of each filtered positive segment from a value of a previous positive segment.

9. The computer-implemented method of claim 1 , wherein extracting the set of extracted positive features from the set of filtered positive segments includes determining a standard deviation for each filtered positive segment.

10. The computer-implemented method of claim 1 , wherein selecting the set of selected positive features from the set of extracted positive features includes selecting positive features using a wrapper procedure.

11. The computer-implemented method of claim 10 , wherein selecting the positive features using the wrapper procedure includes calculating an information gain for each extracted positive feature of the set of extracted positive features.

12. The computer-implemented method of claim 11 , wherein selecting the positive features using the wrapper procedure further includes selecting extracted positive features having an information gain greater than a predetermined threshold value.

13. The computer-implemented method of claim 1 , wherein calculating the precursor pattern detection model based on the set of selected positive features and the set of selected negative features includes providing the set of selected positive features and the set of selected negative features as input to a classifier.

14. The computer-implemented method of claim 13 , wherein the classifier is selected from a group consisting of a naïve classifier, a Bayes network classifier, a logistic regression classifier, a C4.5 decision tree classifier, a support vector machine (SVM) classifier, a voting feature: interval (VFI) classifier, and an artificial neural network (ANN) classifier.

15. The computer-implemented method of claim 1 , wherein executing the first filter on the set of positive segments to generate the set of filtered positive segments includes:

for each positive segment of the set of positive segments,

determining a strength of the positive segment;

subdividing the positive segment into multiple subsegments based on data granularity;

for each subsegment of the multiple subsegments, determining a strength of a remaining portion of the positive segment with the subsegment removed;

identifying a subsegment of the multiple subsegments that has a largest impact on the strength of the remaining portion of the positive segment with the subsegment removed; and

removing the identified subsegment according to the strength of the remaining portion of the positive segment with the identified subsegment removed.

16. The computer-implemented method of claim 1 , wherein at least one positive segment of the set of positive segments are extracted from a portion of the annotated monitoring data before an annotated medical event of the annotated monitoring data.

17. The computer-implemented method of claim 1 , wherein at least one negative segment of the set of negative segments is extracted from a portion of the annotated monitoring data associated with the least one first patient that does not exhibit or has not exhibited the medical event of the given type.

18. The computer-implemented method of claim 1 , further comprising, in response to the match, causing a medication to be automatically administered, or causing an amount of monitoring to be changed.

19. A nontransitory computer-readable storage medium having instructions stored thereon, wherein the instructions are executable by at least one processor to cause the processor to perform operations comprising:

obtaining annotated monitoring data, wherein the annotated monitoring data represents measurements of physical conditions of at least one first patient;

extracting a set of positive segments from the annotated monitoring data, wherein each positive segment is associated with an occurrence of a medical event of a given type;

identifying a portion of the annotated monitoring data that is at least three days away from the medical events of the annotated monitoring data along a timeline;

extracting a set of negative segments from the annotated monitoring data, wherein each negative segment is not associated with an occurrence of the medical event of the given type, and wherein all of the set of negative segments are extracted from the portion of the annotated monitoring data that is at least three days away from the medical events of the annotated monitoring data along the timeline;

executing a first filter on the set of positive segments to generate a′ set of filtered positive segments;

executing a second filter on the set of negative segments to generate a set of filtered negative segments;

extracting a set of extracted positive features from the set of filtered positive segments and a set of extracted negative features from the set of filtered negative segments;

selecting a set of selected positive features from the set of extracted positive features and a set of selected negative features from the set of extracted negative features;

calculating a precursor pattern detection model based on the set of selected positive features and the set of selected negative features;

sensing a signal from a second patient to form at least a portion of collected patient data;

detecting a match between a set of the collected patient data and the precursor pattern detection model; and

in response to the match, informing a party to seek or to provide medical care.

20. A system comprising:

at least one processor and a memory connected to the processor, the memory storing instructions executable by the processor to cause the processor to

obtain annotated monitoring data, wherein the annotated monitoring data represents measurements of physical conditions of at least one first patient;

extract a set of positive segments from the annotated monitoring data, wherein each positive segment is associated with an occurrence of a medical event of a given type;

identify a portion of the annotated monitoring data that is at least three days away from the medical events of the annotated monitoring data along a timeline;

extract a set of negative segments from the annotated monitoring data, wherein each negative segment is not associated with an occurrence of the medical event of the given type, and wherein all of the set of negative segments are extracted from the portion of the annotated monitoring data that is at least three clays away from the medical events of the annotated monitoring data along the timeline;

execute a first filter on the set of positive segments to generate a set of filtered positive segments;

execute a second filter on the set of negative segments to generate a set of filtered negative segments;

extract a set of extracted positive features from the set of filtered positive segments and a set of extracted negative features from the set of filtered negative segments;

select a set of selected positive features from the set of extracted positive features and a set of selected negative features from the set of extracted negative features;

calculate a precursor pattern detection model based on the set of selected positive features and the set of selected negative features;

sense a signal from a second patient to form at least a portion of collected patient data;

detect a match between a set of the collected patient data and the precursor pattern detection model; and

in response to the match, informed a party to seek or to provide medical care.