Technology to automatically identify the most relevant health failure risk factors

1. A computing system comprising:

a processor; and

a memory coupled to the processor, the memory including a set of instructions, which when executed by the processor, cause the computing system to:

identify minority class data and majority class data in patient-level data, wherein the minority class data corresponds to patients with a health failure and the majority class data corresponds to patients without the health failure;

oversample the minority class data to obtain synthetic class data; and

automatically reduce, via a machine learning classifier, risk factor variables to a reduced set of risk factor variables based on the majority class data, the minority class data and the synthetic class data, wherein the machine learning classifier includes a multi-layer neural network configured to be trained using at least a portion of the patient-level data to perform one or more forward propagations and one or more rearward propagations until a value of a loss function is acceptable.

2. The computing system of claim 1 , wherein to oversample the minority class data, the instructions, when executed, cause the computing system to:

randomly sample nearest neighbors of instances in the minority class data; and

interpolate between the instances and the randomly sampled nearest neighbors.

3. The computing system of claim 1 , wherein the machine learning classifier includes an Extra Trees classifier configured to be trained by randomly sampling training data points when building a plurality of decision trees, considering random subsets of variables when splitting nodes of the plurality of decision trees, and testing random splits over a fraction of the variables.

4. The computing system of claim 1 , wherein the machine learning classifier includes a Random Forest classifier configured to be trained by randomly sampling training data points when building a plurality of decision trees, considering random subsets of variables when splitting nodes of the plurality of decision trees, and testing all possible splits over a fraction of the variables.

5. The computing system of claim 1 , wherein the health failure is one or more of a target lesion failure, a bleeding event, a stent thrombosis, a treatment decision with an associated level of residual risk or a heart failure.

6. The computing system of claim 5 , wherein the reduced set of risk factor variables is selected from a group consisting of total stent length, procedure duration, baseline fasting blood glucose, diastolic blood pressure, systolic blood pressure and baseline platelet count.

7. The computing system of claim 1 , wherein the instructions, when executed, further cause the computing system to provide one or more graphical user interfaces for display via a user interface, the one or more graphical user interfaces including two or more of:

a user input section that facilitates entry of parameter values associated with a patient for each risk factor variable of the reduced set of risk factor variables for evaluation by the machine learning classifier;

a model performance section that indicates performance characteristics of the machine learning classifier in evaluating the parameter values; or

a prediction probability section that provides a prediction probability indicator that indicates whether the patient is likely to encounter the health failure based on the evaluation by the machine learning classifier.

8. A computing system comprising:

a processor; and

a memory coupled to the processor, the memory including a set of instructions, which when executed by the processor, cause the computing system to:

identify minority class data and majority class data in patient-level data, wherein the minority class data corresponds to patients with a health failure and the majority class data corresponds to patients without the health failure;

oversample the minority class data to obtain synthetic class data;

automatically reduce, via a machine learning classifier, risk factor variables to a reduced set of risk factor variables based on the majority class data, the minority class data and the synthetic class data, wherein the machine learning classifier includes an Extra Trees classifier configured to be trained by randomly sampling training data points when building a plurality of decision trees, considering random subsets of variables when splitting nodes of the plurality of decision trees, and testing random splits over a fraction of the variables.

9. The computing system of claim 8 , wherein to oversample the minority class data, the instructions, when executed, cause the computing system to:

randomly sample nearest neighbors of instances in the minority class data; and

interpolate between the instances and the randomly sampled nearest neighbors.

10. The computing system of claim 8 , wherein the machine learning classifier includes a multi-layer neural network configured to be trained using at least a portion of the patient-level data to perform one or more forward propagations and one or more rearward propagations until a value of a loss function is acceptable.

11. The computing system of claim 8 , wherein the plurality of decisions trees is a first plurality of decision trees, and wherein the machine learning classifier includes a Random Forest classifier configured to be trained by randomly sampling training data points when building a second plurality of decision trees, considering random subsets of variables when splitting nodes of the second plurality of decision trees, and testing all possible splits over a fraction of the variables.

12. The computing system of claim 8 , wherein the health failure is one or more of a target lesion failure, a bleeding event, a stent thrombosis, a treatment decision with an associated level of residual risk or a heart failure.

13. The computing system of claim 12 , wherein the reduced set of risk factor variables is selected from a group consisting of total stent length, procedure duration, baseline fasting blood glucose, diastolic blood pressure, systolic blood pressure and baseline platelet count.

14. The computing system of claim 8 , wherein the instructions, when executed, further cause the computing system to provide one or more graphical user interfaces for display via a user interface, the one or more graphical user interfaces including two or more of:

a user input section that facilitates entry of parameter values associated with a patient for each risk factor variable of the reduced set of risk factor variables for evaluation by the machine learning classifier;

a model performance section that indicates performance characteristics of the machine learning classifier in evaluating the parameter values; or

a prediction probability section that provides a prediction probability indicator that indicates whether the patient is likely to encounter the health failure based on the evaluation by the machine learning classifier.

15. A method comprising:

identifying minority class data and majority class data in patient-level data, wherein the minority class data corresponds to patients with a health failure and the majority class data corresponds to patients without the health failure;

oversampling the minority class data to obtain synthetic class data; and

automatically reducing, via a machine learning classifier, risk factor variables to a reduced set of risk factor variables based on the majority class data, the minority class data and the synthetic class data, wherein the machine learning classifier includes a multi-layer neural network configured to be trained using at least a portion of the patient-level data to perform one or more forward propagations and one or more rearward propagations until a value of a loss function is acceptable.

16. The method of claim 15 , wherein oversampling the minority class data includes:

randomly sampling nearest neighbors of instances in the minority class data; and

interpolating between the instances and the randomly sampled nearest neighbors.

17. The method of claim 15 , wherein the health failure is one or more of a target lesion failure, a bleeding event, a stent thrombosis, a treatment decision with an associated level of residual risk or a heart failure.

18. The method of claim 17 , wherein the reduced set of risk factor variables is selected from a group consisting of total stent length, procedure duration, baseline fasting blood glucose, diastolic blood pressure, systolic blood pressure and baseline platelet count.

19. The method of claim 15 , further comprising outputting the reduced set of risk factor variables via a user interface.

20. The method of claim 15 , further comprising providing one or more graphical user interfaces for display via a user interface, the one or more graphical user interfaces including two or more of:

a user input section that facilitates entry of parameter values associated with a patient for each risk factor variable of the reduced set of risk factor variables for evaluation by the machine learning classifier;

a model performance section that indicates performance characteristics of the machine learning classifier in evaluating the parameter values; or

a prediction probability section that provides a prediction probability indicator that indicates whether the patient is likely to encounter the health failure based on the evaluation by the machine learning classifier.

21. A method comprising:

identifying minority class data and majority class data in patient-level data, wherein the minority class data corresponds to patients with a health failure and the majority class data corresponds to patients without the health failure;

oversampling the minority class data to obtain synthetic class data; and

automatically reducing, via a machine learning classifier, risk factor variables to a reduced set of risk factor variables based on the majority class data, the minority class data and the synthetic class data, wherein the machine learning classifier includes an Extra Trees classifier configured to be trained by randomly sampling training data points when building a plurality of decision trees, considering random subsets of variables when splitting nodes of the plurality of decision trees, and testing random splits over a fraction of the variables.

22. A method comprising:

identifying minority class data and majority class data in patient-level data, wherein the minority class data corresponds to patients with a health failure and the majority class data corresponds to patients without the health failure;

oversampling the minority class data to obtain synthetic class data; and

automatically reducing, via a machine learning classifier, risk factor variables to a reduced set of risk factor variables based on the majority class data, the minority class data and the synthetic class data, wherein the machine learning classifier includes an Random Forest classifier configured to be trained by randomly sampling training data points when building a plurality of decision trees, considering random subsets of variables when splitting nodes of the plurality of decision trees, and testing all possible splits over a fraction of the variables.

23. A computing system comprising:

a processor; and

a memory coupled to the processor, the memory including a set of instructions, which when executed by the processor, cause the computing system to:

identify minority class data and majority class data in patient-level data, wherein the minority class data corresponds to patients with a health failure and the majority class data corresponds to patients without the health failure;

oversample the minority class data to obtain synthetic class data; and

automatically reduce, via a machine learning classifier, risk factor variables to a reduced set of risk factor variables based on the majority class data, the minority class data and the synthetic class data, wherein the machine learning classifier includes a Random Forest classifier configured to be trained by randomly sampling training data points when building a plurality of decision trees, considering random subsets of variables when splitting nodes of the plurality of decision trees, and testing all possible splits over a fraction of the variables.