Machine learning techniques for identifying opportunity patients
Systems and techniques are disclosed for using machine-learning to identify potential opportunity patients that are more likely to adjust his/her preference for a healthcare provider or service. In some implementations, integrated patient data is obtained. A patient sequence feature vector, a provider sequence feature vector, and a set of entity-specific feature vectors are generated. A set of opportunity patients is identified. A notification is transmitted to the set of opportunity patients about a second treatment plan.
1. A computer-implemented method, the method comprising:
obtaining patient sequence information identifying treatment events associated with a first treatment plan and provider sequence information identifying providers that have participated in a treatment event associated with the first treatment plan;
generating a plurality of feature vectors, including:
generating a patient sequence feature vector corresponding to the patient sequence information based on applying a first model to the patient sequence information, wherein the first model is trained to output patient sequence feature vectors for respective patients with different patient sequences;
generating a provider sequence feature vector corresponding to the provider sequence information based on applying a second model to the provider sequence information, wherein the second model is trained to output provider sequence feature vectors for respective providers with different provider sequences;
training a combining model to classify a patient as being likely to adjust the first treatment plan;
providing at least some of the plurality of feature vectors as an input to the combining model;
classifying, based on the combining model, a set of patients as being likely to adjust the first treatment plan; and
transmitting a notification to the set of patients about a second treatment plan.
2. The method of claim 1 , wherein:
the first treatment plan represents treatment of a disease condition using a pharmaceutical product;
the patient sequence information identifies, for each patient, a first treatment event indicating diagnosis of the disease condition, a second treatment event indicating a prescription of the pharmaceutical product, and a third treatment event indicating a treatment follow-up; and
the set of patients represents patients that are classified by the combining model as being likely to withdraw from the first treatment plan based on data associated with the first treatment event, the second treatment event, or the third treatment event.
3. The method of claim 1 , wherein generating the plurality of feature vectors comprises generating a set of entity-specific feature vectors corresponding to a set of entity-specific attributes based on applying a third model to the set of entity-specific attributes, wherein the third model is trained to output sets of entity-specific feature vectors for respective sets of entity-specific attributes.
4. The method of claim 3 , wherein:
the first model and the second model each comprise recurrent neural networks; and
the third model and the combining model each comprise deep neural networks.
5. The method of claim 3 , wherein:
the set of entity-specific attributes comprises provider-level attributes, patient-level attributes, and plan-level attributes; and
the set of entity-specific feature vectors comprises a provider-level feature vector, a patient-level feature vector, and a plan-level feature vector.
6. The method of claim 1 , comprising:
training the combining model to classify a provider as being likely to adjust the first treatment plan.
7. The method of claim 1 , wherein the set of patients comprises one or more patients classified by the combining model as being likely to adjust a pharmaceutical product associated with the first treatment plan.
8. The method of claim 1 , wherein the set of patients comprises one or more patients classified by the combining model as being likely to adjust a provider associated with the first treatment plan.
9. A system comprising:
one or more computing devices; and
a non-transitory computer-readable storage device storing instructions that are executable by the one or more computing devices to perform operations comprising:
obtaining patient sequence information identifying treatment events associated with a first treatment plan and provider sequence information identifying providers that have participated in a treatment event associated with the first treatment plan;
generating a plurality of feature vectors, including:
generating a patient sequence feature vector corresponding to the patient sequence information based on applying a first model to the patient sequence information, wherein the first model is trained to output patient sequence feature vectors for respective patients with different patient sequences;
generating a provider sequence feature vector corresponding to the provider sequence information based on applying a second model to the provider sequence information, wherein the second model is trained to output provider sequence feature vectors for respective providers with different provider sequences;
training a combining model to classify a patient as being likely to adjust the first treatment plan;
training a combining model to classify a patient as being likely to adjust the first treatment plan;
providing at least some of the plurality of feature vectors as an input to the combining model;
classifying, based on the combining model, a set of patients as being likely to adjust the first treatment plan; and
transmitting a notification to the set of patients about a second treatment plan.
10. The system of claim 9 , wherein:
the first treatment plan represents treatment of a disease condition using a pharmaceutical product;
the patient sequence information identifies, for each patient, a first treatment event indicating diagnosis of the disease condition, a second treatment event indicating a prescription of the pharmaceutical product, and a third treatment event indicating a treatment follow-up; and
the set of patients represents patients that are classified by the combining model as being likely to withdraw from the first treatment plan based on data associated with the first treatment event, the second treatment event, or the third treatment event.
11. The system of claim 9 , wherein generating the plurality of feature vectors comprises generating a set of entity-specific feature vectors corresponding to a set of entity-specific attributes based on applying a third model to the set of entity-specific attributes, wherein the third model is trained to output sets of entity-specific feature vectors for respective sets of entity-specific attributes.
12. At least one non-transitory computer-readable storage device storing instructions that are executable by one or more computing devices to perform operations comprising:
obtaining patient sequence information identifying treatment events associated with a first treatment plan and provider sequence information identifying providers that have participated in a treatment event associated with the first treatment plan;
generating a plurality of feature vectors, including:
generating a patient sequence feature vector corresponding to the patient sequence information based on applying a first model to the patient sequence information, wherein the first model is trained to output patient sequence feature vectors for respective patients with different patient sequences;
generating a provider sequence feature vector corresponding to the provider sequence information based on applying a second model to the provider sequence information, wherein the second model is trained to output provider sequence feature vectors for respective providers with different provider sequences;
training a combining model to classify a patient as being likely to adjust the first treatment plan;
providing at least some of the plurality of feature vectors as an input to the combining model;
classifying, based on the combining model, a set of patients as being likely to adjust the first treatment plan; and
transmitting a notification to the set of patients about a second treatment plan.
13. The one non-transitory computer-readable storage device of claim 12 , wherein:
the first treatment plan represents treatment of a disease condition using a pharmaceutical product;
the patient sequence information identifies, for each patient, a first treatment event indicating diagnosis of the disease condition, a second treatment event indicating a prescription of the pharmaceutical product, and a third treatment event indicating a treatment follow-up; and
the set of patients represents patients that are classified by the combining model as being likely to withdraw from the first treatment plan based on data associated with the first treatment event, the second treatment event, or the third treatment event.
14. The one non-transitory computer-readable storage device of claim 12 , wherein generating the plurality of feature vectors comprises generating a set of entity-specific feature vectors corresponding to a set of entity-specific attributes based on applying a third model to the set of entity-specific attributes, wherein the third model is trained to output sets of entity-specific feature vectors for respective sets of entity-specific attributes.