Method and apparatus for refining an automated coding model

1. A method for refining an automated coding model for secondary diagnosis prediction, the method comprising:

operating a deep learning model including an artificial neural network to perform operations as follows:

for each respective candidate secondary diagnosis from a set of candidate secondary diagnoses, predicting a probability of the respective secondary diagnosis being contained in a medical chart;

selecting one of the candidate secondary diagnoses as being contained in the medical chart based upon the probability that is predicted and removing the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses;

following removal of the candidate secondary diagnosis that has been selected from the set of secondary diagnoses, repeatedly predicting the probability of a respective secondary diagnosis being contained in the medical chart, selecting one of the candidate secondary diagnoses as being contained in the medical chart based upon the probability that is predicted and removing the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses; and

determining a categorical crossentropy loss based upon the probabilities that are predicted for the candidate secondary diagnoses that are selected;

adjusting one or more parameters of the artificial neural network, which implements the automated coding model, based upon the categorical crossentropy loss,

wherein the set of candidate secondary diagnoses also includes a null element, and wherein repeatedly predicting, selecting and removing is concluded upon selecting the null element based upon the probability of the null element that is predicted.

2. A method according to claim 1 wherein selecting one of the candidate secondary diagnoses comprises selecting the candidate secondary diagnosis having the greatest probability of being contained in the medical chart.

3. A method according to claim 1 wherein predicting the probability of a respective secondary diagnosis being contained in a medical chart comprises predicting the probability such that a cumulative probability of the plurality of candidate secondary diagnoses from the set of candidate secondary diagnoses equals 1.

4. A method according to claim 1 wherein determining the categorical crossentropy loss comprises:

for each of the candidate secondary diagnoses that are selected, determining a product of a label representative of a respective candidate secondary diagnosis that is selected and the probability of the respective candidate secondary diagnosis; and

summing the products for each of the candidate secondary diagnoses that are selected to determine the categorical crossentropy loss.

5. A method according to claim 1 further comprising adjusting the one or more parameters of the automated coding model based upon the categorical crossentropy loss to reduce the categorical crossentropy loss.

6. A method according to claim 1 wherein removing the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses comprises removing the candidate secondary diagnosis that has been selected and one or more additional candidate secondary diagnoses that have a defined relationship to the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses.

7. An apparatus for refining an automated coding model for secondary diagnosis prediction, the apparatus comprising:

a processor; and

a memory coupled to the processor and comprising computer readable program code embodied in the memory that is executable by the processor to operate a deep learning model including an artificial neural network to perform operations comprising:

predicting, for each respective candidate secondary diagnosis from a set of candidate secondary diagnoses, a probability of the respective secondary diagnosis being contained in a medical chart;

selecting one of the candidate secondary diagnoses as being contained in the medical chart based upon the probability that is predicted and remove the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses; and

following removal of the candidate secondary diagnosis that has been selected from the set of secondary diagnoses, repeatedly predicting the probability of a respective secondary diagnosis being contained in the medical chart, select one of the candidate secondary diagnoses as being contained in the medical chart based upon the probability that is predicted and remove the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses; and

determining a categorical crossentropy loss based upon the probabilities that are predicted for the candidate secondary diagnoses that are selected;

adjusting one or more parameters of the artificial neural network, which implements the automated coding model, based upon the categorical crossentropy loss,

wherein the set of candidate secondary diagnoses includes a null element, and wherein repeatedly predicting, selecting and removing is concluded upon selecting the null element based upon the probability of the null element that is predicted.

8. An apparatus according to claim 7 wherein selecting one of the candidate secondary diagnoses comprises selecting the candidate secondary diagnosis having the greatest probability of being contained in the medical chart.

9. An apparatus according to claim 7 wherein predicting the probability of a respective secondary diagnosis being contained in a medical chart comprises predicting the probability such that a cumulative probability of the plurality of candidate secondary diagnoses from the set of candidate secondary diagnoses equals 1.

10. An apparatus according to claim 7 wherein determining the categorical crossentropy loss comprises:

for each of the candidate secondary diagnoses that are selected, determining a product of a label representative of a respective candidate secondary diagnosis that is selected and the probability of the respective candidate secondary diagnosis; and

summing the products for each of the candidate secondary diagnoses that are selected to determine the categorical crossentropy loss.

11. An apparatus according to claim 7 wherein the operations further comprise:

adjusting the one or more parameters of the automated coding model based upon the categorical crossentropy loss to reduce the categorical crossentropy loss.

12. An apparatus according to claim 7 wherein removing the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses comprises removing the candidate secondary diagnosis that has been selected and one or more additional candidate secondary diagnoses that have a defined relationship to the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses.

13. A computer program product, comprising:

a non-transitory computer readable storage medium comprising computer readable program code embodied in the medium that is executable by a processor to perform operations comprising:

operating a deep learning model including an artificial neural network to perform operations as follows:

for each respective candidate secondary diagnosis from a set of candidate secondary diagnoses, predicting a probability of the respective secondary diagnosis being contained in a medical chart;

selecting one of the candidate secondary diagnoses as being contained in the medical chart based upon the probability that is predicted and removing the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses;

following removal of the candidate secondary diagnosis that has been selected from the set of secondary diagnoses, repeatedly predicting the probability of a respective secondary diagnosis being contained in the medical chart, selecting one of the candidate secondary diagnoses as being contained in the medical chart based upon the probability that is predicted and removing the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses; and

determining a categorical crossentropy loss based upon the probabilities that are predicted for the candidate secondary diagnoses that are selected;

adjusting one or more parameters of the artificial neural network, which implements the automated coding model, based upon the categorical crossentropy loss,

wherein the set of candidate secondary diagnoses also includes a null element, and wherein repeatedly predicting, selecting and removing is concluded upon selecting the null element based upon the probability of the null element that is predicted.

14. A computer program product according to claim 13 wherein selecting one of the candidate secondary diagnoses comprises selecting the candidate secondary diagnosis having the greatest probability of being contained in the medical chart.

15. A computer program product according to claim 13 wherein predicting the probability of a respective code being contained in the medical chart comprises predicting the probability such that a cumulative probability of the plurality of candidate secondary diagnoses from the set of candidate secondary diagnoses equals 1.

16. A computer program product according to claim 13 wherein determining the categorical crossentropy loss comprises:

for each of the candidate secondary diagnoses that are selected, determining a product of a label representative of a respective candidate secondary diagnosis that is selected and the probability of the respective candidate secondary diagnosis; and

summing the products for each of the candidate secondary diagnoses that are selected to determine the categorical crossentropy loss.

17. A computer program product according to claim 13 wherein the operations further comprise adjusting the one or more parameters of the automated coding model based upon the categorical crossentropy loss to reduce the categorical crossentropy loss.

18. A computer program product according to claim 13 wherein removing the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses comprises removing the candidate secondary diagnosis that has been selected and one or more additional candidate secondary diagnoses that have a defined relationship to the candidate secondary diagnosis that has been selected from the set of candidate secondary diagnoses.