Channel-specific engagement machine learning architecture

1. A computer-implemented method comprising:

generating an intervention model by:

determining principal components for features of a training set,

associating each feature of the training set with a principal component,

selecting features of the training set most highly correlated with principal components,

performing a regression analysis on the selected features to determine a subset of the selected features that are most highly correlated with a model target,

training a machine learning model based on the subset of the selected features, and

saving the trained machine learning model as the intervention model;

obtaining data related to a user, wherein:

the data includes engagement data indicating successfulness of prior interventions with the user and

each prior intervention with the user is associated with one of multiple engagement channels;

supplying the obtained data as input to the intervention model to determine multiple channel-specific intervention expectations, wherein each channel-specific intervention expectation:

corresponds to one of the multiple engagement channels and

indicates a likelihood that the user will take action in response to an intervention being executed using the corresponding engagement channel;

determining a likelihood of a gap in care for the user; and

in response to the care gap likelihood being outside of a threshold:

identifying a highest determined value of the channel-specific intervention expectations,

selecting an intervention corresponding to the highest determined value of the channel-specific intervention expectation, and

scheduling the selected intervention for execution.

2. The method of claim 1 wherein:

at least one of the multiple engagement channels includes multiple intervention options within the engagement channel;

selecting the intervention includes selecting one of the multiple interventions within the engagement channel that has a highest intervention expectation among the intervention options; and

scheduling the intervention includes scheduling the selected one of the multiple intervention options within the engagement channel.

3. The method of claim 1 wherein:

the intervention model includes a channel-agnostic intervention model that determines a general intervention expectation indicating a likelihood that the user will take action in response to any intervention being executed using any of the engagement channels and

the method further comprises, in response to the general intervention expectation being below a specified threshold, initiating a low engagement intervention process and identifying at least one reason for low engagement of the user.

4. The method of claim 1 further comprising, in response to determining that a time elapsed since a most recent intervention for the user is less than a specified delay threshold, waiting until the specified delay threshold has elapsed prior to selecting the intervention.

5. The method of claim 1 further comprising:

identifying one or more targets relevant to the user;

determining a measure of progress toward at least one of the identified targets;

determining an engagement importance metric based on the determined measure of progress; and

weighting the channel-specific intervention expectations according to the determined engagement importance metric prior to selecting the intervention.

6. The method of claim 5 wherein scheduling the intervention includes, in response to determining that the measure of progress is less than a specified minimum threshold, scheduling an intervention corresponding to the channel-specific intervention expectation that has a highest determined value prior to weighting.

7. The method of claim 1 further comprising:

determining a cost of engagement for each of the multiple engagement channels;

determining a channel capacity for each of the multiple engagement channels; and

weighting the channel-specific intervention expectations according to the determined costs of engagement and channel capacities prior to selecting the intervention.

8. The method of claim 1 further comprising:

determining at least one user adherence factor associated with the user and

weighting the channel-specific intervention expectations according to the determined user adherence factor prior to selecting the intervention,

wherein the at least one user adherence factor includes at least one of a prescription refill cost, a pill burden, and a comorbidity condition.

9. The method of claim 8 wherein scheduling the intervention includes, in response to the comorbidity condition indicating a future adverse risk event that is higher than a specified threshold, scheduling the intervention corresponding to the channel-specific intervention expectation that has a highest determined value prior to weighting.

10. The method of claim 1 further comprising verifying the trained machine learning model with a verification set prior to saving the trained machine learning model.

11. A system comprising:

memory hardware configured to store instructions and

processing hardware configured to execute the instructions stored by the memory hardware, wherein the instructions include:

generating an intervention model by:

determining principal components for features of a training set,

associating each feature of the training set with a principal component,

selecting features of the training set most highly correlated with principal components,

performing a regression analysis on the selected features to determine a subset of the selected features that are most highly correlated with a model target,

training a machine learning model with the subset of the selected features,

saving the trained machine learning model as the intervention model;

obtaining data related to a user, wherein:

the data includes engagement data indicating successfulness of prior interventions with the user and

each prior intervention with the user is associated with one of multiple engagement channels;

supplying the obtained data as input to the intervention model to determine multiple channel-specific intervention expectations, wherein each channel-specific intervention expectation:

corresponds to one of the multiple engagement channels and

indicates a likelihood that the user will take action in response to an intervention being executed using the corresponding engagement channel;

determining a likelihood of a gap in care for the user; and

in response to the care gap likelihood being outside of a threshold:

identifying a highest determined value of the channel-specific intervention expectations,

selecting an intervention corresponding to the highest determined value of the channel-specific intervention expectation, and

scheduling the selected intervention for execution.

12. The system of claim 11 wherein:

at least one of the multiple engagement channels includes multiple intervention options within the engagement channel;

selecting the intervention includes selecting one of the multiple interventions within the engagement channel that has a highest intervention expectation among the intervention options; and

scheduling the intervention includes scheduling the selected one of the multiple intervention options within the engagement channel.

13. The system of claim 11 wherein:

the intervention model includes a channel-agnostic intervention model that determines a general intervention expectation indicating a likelihood that the user will take action in response to any intervention being executed using any of the engagement channels and

the instructions include, in response to the general intervention expectation being below a specified threshold, initiating a low engagement intervention process and identifying at least one reason for low engagement of the user.

14. The system of claim 11 wherein the instructions include, in response to determining that a time elapsed since a most recent intervention for the user is less than a specified delay threshold, waiting until the specified delay threshold has elapsed prior to selecting the intervention.

15. The system of claim 11 wherein the instructions include:

identifying one or more targets relevant to the user;

determining a measure of progress toward at least one of the identified targets;

determining an engagement importance metric based on the determined measure of progress; and

weighting the channel-specific intervention expectations according to the determined engagement importance metric prior to selecting the intervention.

16. The system of claim 15 wherein scheduling the intervention includes, in response to determining that the measure of progress is less than a specified minimum threshold, scheduling an intervention corresponding to the channel-specific intervention expectation that has a highest determined value prior to weighting.

17. The system of claim 11 wherein the instructions include:

determining a cost of engagement for each of the multiple engagement channels;

determining a channel capacity for each of the multiple engagement channels; and

weighting the channel-specific intervention expectations according to the determined costs of engagement and channel capacities prior to selecting the intervention.

18. The system of claim 11 wherein:

the instructions include:

determining at least one user adherence factor associated with the user and

weighting the channel-specific intervention expectations according to the determined user adherence factor prior to selecting the intervention and

the at least one user adherence factor includes at least one of a prescription refill cost, a pill burden, and a comorbidity condition.

19. The system of claim 18 wherein scheduling the intervention includes, in response to the comorbidity condition indicating a future adverse risk event that is higher than a specified threshold, scheduling the intervention corresponding to the channel-specific intervention expectation that has a highest determined value prior to weighting.

20. The system of claim 11 wherein the instructions include verifying the trained machine learning model with a verification set prior to saving the trained machine learning model.