Machine learning models in location based episode prediction

1. At least one non-transitory computer-readable storage medium storing instructions, which, when executed by at least one data processor of a system, cause the system to:

train a predictive model based on training data including records of non-public healthcare insurance claims and public healthcare fee schedule data;

determine a multiplier for each of multiple healthcare facilities based on the predictive model,

wherein a first multiplier for a first healthcare facility is configured to scale the public healthcare fee schedule data such that cost estimates for healthcare services provided by the first healthcare facility are predicted based on the first multiplier; and

wherein a second multiplier for a second healthcare facility, different from the first multiplier, is configured to scale the public healthcare fee schedule data such that cost estimates for healthcare services provided by the second healthcare facility are predicted based on the second multiplier;

receive, from a consumer device, a query for a particular cost estimate of a particular healthcare service provided by the multiple healthcare facilities,

wherein the query includes constraints for a preferred geographic region or a preferred insurer, and

wherein the first and second healthcare facilities satisfy the constraints of the query;

predict a first cost estimate for the particular healthcare service at the first healthcare facility based on the first multiplier and a second cost estimate for the particular healthcare service at the second healthcare facility based on the second multiplier; and

cause the consumer device to display query results including indications of the first and second cost estimates.

2. The non-transitory computer-readable storage medium of claim 1 , wherein the predictive model processes data including multiple episodes of care, each having a frequency weighted combination of healthcare services and at least one additional healthcare service, and wherein any cost estimates are based on a normalized aggregate included in the multiple episodes of care.

3. The non-transitory computer-readable storage medium of claim 1 further causing the system to, prior to causing display of the query results:

receive an indication of the preferred geographic region,

wherein the preferred geographic region is input at the consumer device by a user of the consumer device, and

wherein the query is constrained by the preferred geographic region.

4. The non-transitory computer-readable storage medium of claim 1 further causing the system to, prior to causing display of the query results:

receive an indication of the preferred insurer,

wherein the preferred insurer is input at the consumer device by a user of the consumer device, and

wherein the query is constrained by the preferred insurer.

5. The non-transitory computer-readable storage medium of claim 1 further causing the system to, prior to causing display of the query results:

ascertain, from the records of non-public healthcare insurance claims and the public healthcare fee schedule data, multiple data points each including a value pairs for a healthcare facility and an insurer; and

perform a Bayesian regression on the multiple data points to generate the predictive model as a linear model where the slope represents a multiplier.

6. The non-transitory computer-readable storage medium of claim 1 further causing the system to:

obtain additional data points; and

performing a Bayesian regression including the additional data points to generate an updated predictive model.

7. The non-transitory computer-readable storage medium of claim 1 further causing the system to:

generate multiple predictive models including a predictive model that accounts for healthcare providers, a predictive model that accounts for patient histories, a predictive model that accounts for comorbidities, and a predictive model that accounts for secondary healthcare providers.

8. The non-transitory computer-readable storage medium of claim 1 , wherein the predictive model facilitates:

discovery of costs that have been re-negotiated by an insurer,

extrapolate missing data of the records of non-public healthcare insurance claims, or

identify outlier data points of the predictive model, fraudulent claims, negotiated cost patterns, or cost policies.

9. The non-transitory computer-readable storage medium of claim 1 further causing the system to:

cause the consumer device to display one or more visualizations indicating the first and second cost estimates that overlay a graphic map of the preferred geographic region.

10. The non-transitory computer-readable storage medium of claim 1 further causing the system to:

cause the consumer device to display one or more visualizations indicating the first and second cost estimates relative to the preferred insurer.

11. The non-transitory computer-readable storage medium of claim 1 further causing the system to, prior to causing display of the query results:

associate each of the first and second cost estimates with a visualization element indicating whether the first or second cost estimate is low, typical, or high for the geographic region.

12. The non-transitory computer-readable storage medium of claim 1 further causing the system to:

cause the consumer device to display multiple visualization elements for different cost estimates associated with different insurers; and

cause the consumer device to display additional information related to the particular cost estimate for an insurer selected from among the different insurers.

13. The non-transitory computer-readable storage medium of claim 1 further causing the system to:

cause the consumer device to display cost estimate data that is customized based on user input at the consumer device.

14. The non-transitory computer-readable storage medium of claim 1 , wherein the display includes a visualization with a range of cost estimates for an insurer as a function of multiple healthcare facilities, and wherein the cost estimates are ranked and color coded to indicate a relative cost.

15. The non-transitory computer-readable storage medium of claim 1 , wherein each record of non-public healthcare insurance claims has fields for a location, a doctor, a procedure code, and an episode, the system being further caused to:

determine, based on user input and the predictive model, a procedure code and a corresponding doctor and location associated with each cost estimate.

16. A method comprising:

training a predictive model based on training data including records of non-public healthcare insurance claims and public healthcare fee schedule data;

determining a multiplier for each of multiple healthcare facilities based on the predictive model,

wherein a first multiplier for a first healthcare facility is configured to scale the public healthcare fee schedule data such that cost estimates for healthcare services provided by the first healthcare facility are predicted based on the first multiplier; and

wherein a second multiplier for a second healthcare facility, different from the first multiplier, is configured to scale the public healthcare fee schedule data such that cost estimates for healthcare services provided by the second healthcare facility are predicted based on the second multiplier;

receiving, from a consumer device, a query for a particular cost estimate of a particular healthcare service provided by the multiple healthcare facilities,

wherein the query includes constraints for a preferred geographic region or a preferred insurer, and

wherein the first and second healthcare facilities satisfy the constraints of the query;

predicting a first cost estimate for the particular healthcare service at the first healthcare facility based on the first multiplier and a second cost estimate for the particular healthcare service at the second healthcare facility based on the second multiplier; and

causing the consumer device to display query results including indications of the first and second cost estimates.

17. The method of claim 16 , wherein the public healthcare fee schedule data includes data based on Centers for Medicare & Medicaid Services.

18. The method of claim 16 further comprising:

anonymizing data identifying patients in the records of non-public healthcare insurance claims.

19. A server computer system comprising:

a processor; and

a memory including instructions that, when executed by the processor, cause the server computer system to:

train a predictive model based on training data including records of non-public healthcare insurance claims and public healthcare fee schedule data;

determine a multiplier for each of multiple healthcare facilities based on the predictive model,

wherein a first multiplier for a first healthcare facility is configured to scale the public healthcare fee schedule data such that cost estimates for healthcare services provided by the first healthcare facility are predicted based on the first multiplier; and

wherein a second multiplier for a second healthcare facility, different from the first multiplier, is configured to scale the public healthcare fee schedule data such that cost estimates for healthcare services provided by the second healthcare facility are predicted based on the second multiplier;

receive, from a consumer device, a query for a particular cost estimate of a particular healthcare service provided by the multiple healthcare facilities,

wherein the query includes constraints for a preferred geographic region or a preferred insurer, and

wherein the first and second healthcare facilities satisfy the constraints of the query;

predict a first cost estimate for the particular healthcare service at the first healthcare facility based on the first multiplier and a second cost estimate for the particular healthcare service at the second healthcare facility based on the second multiplier; and

cause the consumer device to display query results including indications of the first and second cost estimates.