Detection of misuse of a benefit system

1. A system comprising:

one or more processors; and

a memory storing instructions that, when executed by the one or more processors, in conjunction with a particular machine learning model for a subset of the instructions, cause the system to perform:

analyzing a dynamically updating database of claims;

determining a healthcare metric based on the analyses of the database of claims, the healthcare metric characterizing a relationship between one or more pharmacy events and one or more clinical events;

determining an expected pattern of the healthcare metric in relation to a factor;

comparing, to the expected pattern, an actual pattern of the healthcare metric in relation to the factor;

based on the comparison of the actual pattern of the healthcare metric to the expected pattern, automatically determining a potential source of fraud;

determining a second potential source of fraud based on the identified potential source of fraud;

in response to the potential source of fraud or the second potential source of fraud being shut down, determining a backup potential source of fraud, the determining the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud using one or more machine learning models, the one or more machine learning models comprising the particular machine learning model;

determining a similarity between the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud with a known instance of fraud;

generating, using natural language processing, an explanation indicating:

the similarity between the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud with the known instance of fraud; and

the particular machine learning model, which has a highest contribution in determining the potential source of fraud, the second potential source of fraud and the backup potential source of fraud;

appending the explanation and the reasoning to a presentation of the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud;

selecting, from among known outcomes of analyses in which veracities of previous potential sources of fraud have been determined, training datasets to train the particular machine learning model, the selecting being based on scenarios associated with the known outcomes that comprise highest uncertainty levels in the particular machine learning model and which increase an accuracy of the particular machine learning model by highest amounts following the training, wherein the training datasets comprise a first training dataset of sources verified to be fraudulent and a second training dataset of sources verified to be non-fraudulent; and

iteratively training the particular machine learning model using the first training dataset and the second training dataset.

2. The system of claim 1 , wherein determining the healthcare metric includes determining whether the one or more pharmacy events are accompanied by the one or more clinical events.

3. The system of claim 2 , wherein determining whether the one or more pharmacy events are accompanied by the one or more clinical events includes determining respective durations between occurrences of the one or more pharmacy events and the one or more clinical events.

4. The system of claim 1 , wherein the expected pattern is characterized by an expected number of clinical events associated with the one or more pharmacy events.

5. The system of claim 1 , wherein the one or more pharmacy events include one or more medical events in which a drug is prescribed or a prescription for the drug is filled.

6. The system of claim 1 , wherein the one or more clinical events include one or more medical events in which a healthcare provider assesses a patient's need for pharmaceutical treatment.

7. The system of claim 1 , wherein the factor is a weather condition.

8. The system of claim 7 , wherein the potential source of fraud identifies one or more of patients, healthcare providers, or healthcare events, and the second potential source of fraud identifies one or more of patients, healthcare providers, and/or healthcare events.

9. The system of claim 1 , wherein the iteratively training of the particular machine learning model comprises:

calculating new weights of signals associated with the particular machine learning model; and

detecting a known instance of fraud that was undetected by the trained particular machine learning model; and

incorporating the known instance of fraud into the trained particular machine learning model.

10. The system of claim 1 , wherein the instructions further cause the system to perform:

in response to verifying the potential source of fraud, the second potential source of fraud, or the backup potential source of fraud, providing a treatment to a patient.

11. A system comprising:

one or more processors; and

a memory storing instructions that, when executed by the one or more processors, in conjunction with a particular machine learning model for a subset of the instructions, cause the system to perform:

analyzing a database of claims;

determining a healthcare metric based on the analyses of the database of claims, the healthcare metric determined based on an amount of opiate doses received by a patient over a period of time;

determining an expected pattern of the healthcare metric in relation to a factor;

comparing, to the expected pattern, an actual pattern of the healthcare metric in relation to the factor;

based on the comparison of the actual pattern of the healthcare metric to the expected pattern, automatically determining a potential source of fraud;

determining a second potential source of fraud based on the identified potential source of fraud; and

in response to the potential source of fraud or the second potential source of fraud being shut down, determining a backup potential source of fraud, the determining the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud using one or more machine learning models, the one or more machine learning models comprising the particular machine learning model;

determining a similarity between the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud with a known instance of fraud;

generating, using natural language processing, an explanation indicating:

the similarity between the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud with the known instance of fraud; and

the particular machine learning model, which has a highest contribution in determining the potential source of fraud, the second potential source of fraud and the backup potential source of fraud;

appending the explanation and the reasoning to a presentation of the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud;

selecting, from among known outcomes of analyses in which veracities of previous potential sources of fraud have been determined, training datasets to train the particular machine learning model, the selecting being based on scenarios associated with the known outcomes that comprise highest uncertainty levels in the particular machine learning model and which increase an accuracy of the particular machine learning model by highest amounts following the training, wherein the training datasets comprise a first training dataset of sources verified to be fraudulent and a second training dataset of sources verified to be non-fraudulent; and

iteratively training the particular machine learning model using the first training dataset and the second training dataset.

12. The system of claim 11 , wherein determining the healthcare metric includes converting the amount of opiate doses received by the patient over the period of time into a morphine equivalent.

13. The system of claim 11 , wherein the healthcare metric is adjusted based on a size of the patient.

14. The system of 11 , wherein the healthcare metric includes a patient healthcare metric, the patient healthcare metric characterizing the amount of opiate doses received by the patient over the period of time.

15. The system of claim 11 , wherein the healthcare metric includes a healthcare provider healthcare metric, the healthcare provider healthcare metric characterizing the amount of opiate doses received by the patient and one or more other amounts of opiate doses received by one or more other patients over the period of time.

16. A system comprising:

one or more processors; and

a memory storing instructions that, when executed by the one or more processors, in conjunction with a particular machine learning model for a subset of the instructions, cause the system to perform:

analyzing a database of claims;

determining a healthcare metric based on the analyses of the database of claims, the healthcare metric characterizing a billing pattern of one or more healthcare providers;

determining an expected pattern of the healthcare metric in relation to a factor;

comparing, to the expected pattern, an actual pattern of the healthcare metric in relation to the factor;

based on the comparison of the actual pattern of the healthcare metric to the expected pattern, automatically determining a potential source of fraud;

determining a second potential source of fraud based on the identified potential source of fraud;

in response to the potential source of fraud or the second potential source of fraud being shut down, determining a backup potential source of fraud, the determining the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud using one or more machine learning models, the one or more machine learning models comprising the particular machine learning model;

determining a similarity between the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud with a known instance of fraud;

generating, using natural language processing, an explanation indicating:

the similarity between the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud with the known instance of fraud; and

the particular machine learning model, which has a highest contribution in determining the potential source of fraud, the second potential source of fraud and the backup potential source of fraud;

appending the explanation and the reasoning to a presentation of the potential source of fraud, the second potential source of fraud, and the backup potential source of fraud;

selecting, from among known outcomes of analyses in which veracities of previous potential sources of fraud have been determined, training datasets to train the particular machine learning model, the selecting being based on scenarios associated with the known outcomes that comprise highest uncertainty levels in the particular machine learning model and which increase an accuracy of the particular machine learning model by highest amounts following the training, wherein the training datasets comprise a first training dataset of sources verified to be fraudulent and a second training dataset of sources verified to be non-fraudulent; and

iteratively training the particular machine learning model using the first training dataset and the second training dataset.

17. The system of claim 16 , wherein the healthcare metric is determined using a dependence of billings of the one or more healthcare providers on patients seen by the one or more healthcare providers.

18. The system of claim 17 , wherein the dependence of billings of the one or more healthcare providers on the patients indicates a level of similarity of tests or treatments among the patients.

19. The system of claim 16 , wherein determining the healthcare metric includes determining a level or a periodicity of the one or more healthcare providers' billings that are independent of external factors.

20. The system of claim 16 , wherein the expected pattern comprises an expected decrease of the healthcare metric during a period having a weather condition;

the actual pattern of the healthcare metric comprises an increase during the period having the weather condition; and

in response to comparing the actual pattern of the healthcare metric to the expected pattern, automatically identifying a potential source of fraud.