A computerized method of automated bias correction for computer systems, the method comprising:
- training a machine learning model with historical feature vector inputs to generate a claim prediction output, wherein the historical feature vector inputs include historical claim data structures specific to multiple patient entities, by:
- separating portions of the historical feature vector inputs into structured training data and structured test data, the structured training data and the structured test data including historical medical claim data structures including patient demographic data, procedure data, class data, provider data and employer data combined in to multiple historical feature vectors for supplying as input to the machine learning model;
- training the machine learning model using the structured training data, wherein the machine learning model includes a recurrent neural-network based model including multiple neurons arranged in at least one input layer, at least one hidden layer and at least one output layer, with each connection between neurons in adjacent layers assigned a weight which is changed during training of the recurrent neural-network based model;
- testing the trained machine learning model using the structured test data;
- evaluating results of testing the trained machine learning model;
- adjusting parameters of the machine learning model according to the results of testing the trained machine learning model;
- and
- saving the machine learning model for use in generating claim prediction outputs, in response to a determination that accuracy of the evaluated results is greater than or equal to a specified accuracy threshold;
- obtaining data corresponding to a set of multiple patient entities;
- for each patient entity in the set of multiple patient entities:
- obtaining structured claim data specific to the patient entity and structured demographic data specific to the patient entity;
- generating a feature vector input according to the structured claim data;
- processing, by the machine learning model, the feature vector input to generate the claim prediction output;
- storing the claim prediction output as a predicted future claim value specific to the patient entity;
- and
- assigning the patient entity to a first entity subset or to a second entity subset according to the structured demographic data specific to the patient entity, wherein patient entities belonging to the first entity subset have at least one demographic characteristic that differs from patient entities belonging to the second entity subset;
- obtaining a specified predicted healthcare cost threshold output value for entity intervention scheduling or entity case management;
- determining a first proportion of patient entities within the first entity subset that have a predicted future claim value exceeding the specified predicted healthcare cost threshold output value;
- determining an adjusted predicted healthcare cost threshold output value for the second entity subset according to the first proportion, such that a second proportion of patient entities within the second entity subset that have a predicted future claim value exceeding the adjusted predicted healthcare cost threshold output value is equal to the first proportion;
- obtaining an event incidence rate of one or more specified treatments for patient entities of the first entity subset;
- obtaining the event incidence rate of the one or more specified treatments for patient entities of the second entity subset;
- modifying the adjusted threshold output value according to a ratio of the event incident rates for the first entity subset and the second entity subset;
- for each entity in the first entity subset:
- determining a predicted future claim value of the entity in the first entity subset;
- and
- in response to the determined predicted future claim value of the entity in the first entity subset exceeding the specified predicted healthcare cost threshold output value, selectively populating a schedule entry for the entity in the first entity subset;
- and
- for each entity in the second entity subset:
- determining a predicted future claim value of the entity in the second entity subset;
- and
- in response to the determined predicted future claim value of the entity in the second entity subset exceeding the adjusted predicted healthcare cost threshold output value, selectively populating a schedule entry for the entity in the second entity subset.