Claim analysis with deep learning
Embodiments relate to system for automatically predicting payer response to claims. In an embodiment, the system receives claim data associated with a claim. The system identifies a set of claim features of the claim data, and generates an input vector with at least a portion of the set of claim features. The system applies the input vector to a trained model. A first portion of the neural network is configured to generate an embedding representing the input vector with a lower dimensionality than the input vector. A second portion of the neural network is configured to generate a prediction of whether the claim will be denied based on the embedding. The system provides the prediction for display on a user interface of a user device. The prediction may further include denial reason codes and a response date estimation to indicate if, when, and why a claim will be denied.
1. A computer-implemented method of analyzing claims, the method comprising:
receiving one or more claims in an electronic format;
identifying a set of claim features of at least one claim of the one or more claims;
generating an embedding of tokens to provide a first input vector based, least in part, on:
application of a tokenization process to the identified set of claim features to provide first numeric values in a first sub-vector to express one or more diagnosis tokens and second numeric values in a second sub-vector to express one or more procedure tokens, the tokenization process comprising one or more machine-learning models having been trained based, at least in part, on a history of claims; and
element-wise multiplication of at least some elements of the first sub-vector with at least some elements of the second sub-vector to provide the first input vector;
executing one or more first neural networks to compute one or more suspiciousness scores and a likelihood of denial of at least at portion the at least one claim based, at least in part, on the first input vector;
communicating at least one of the suspiciousness scores to an operator/user; and
receiving inputs from the operator/user to modify at least a portion of the at least one claim to thereby affect the likelihood of denial of the at least a portion of the at least one claim.
2. The method of claim 1 , wherein at least one of the inputs comprises a correction to a previously incorrectly entered claim feature.
3. The method of claim 1 , wherein the at least one of the one or more machine-learning models has been trained according to a loss function, the loss function being based, at least in part, on a binary cross-entropy loss for a denial probability prediction, a categorical cross-entropy loss for a set of service-level denial reason code classifications or a categorical cross-entropy loss for a set of claim-level denial reason code classifications, or a combination thereof.
4. The method of claim 1 , wherein generating the embedding of tokens further comprises generating a claim sequence comprising:
a sequence of procedure tokens with numeric values;
a sequence of diagnoses tokens with numeric values; or
a sequence of single unique feature tokens with binary values, or a combination thereof.
5. The method of claim 1 , wherein the one or more machine-learning models comprise one or more second neural networks, and wherein generating the embedding of tokens further comprises:
receiving a second input vector comprising a portion of the set of claim features; and
generating the embedding of tokens to comprise a higher dimensionality than a dimensionality of the second input vector.
6. The method of claim 1 , wherein computing the one or more suspiciousness scores further comprises:
computing a gradient of a contribution of at least one claim feature of the identified set of claim features to the likelihood of denial of the at least a portion of the at least one claim.
7. The method of claim 1 , and wherein generating the embedding of tokens further comprises application of the tokenization process to the identified set of claim features to provide binary values in a third sub-vector to express demographic tokens.
8. The method of claim 1 , and further comprising determining the likelihood of denial of the at least the portion of the at least one claim based, at least in part, on a first reason code sequence, a second reason code sequence and a response date estimation, wherein the first reason code sequence includes likelihood scores for claim-level reason codes in a set of claim-level reason codes, and wherein the second reason code sequence includes likelihood scores for service-level reason codes in a set of service-level reason codes.
9. The method of claim 8 , wherein at least one of the one or more machine-learning models comprises:
a first set of task-specific output layers to generate the first reason code sequence,
a second set of task-specific output layers to generate the second reason code sequence, and
a third set of task-specific layers to generate the response date estimation.
10. The method of claim 1 , wherein at least one of the one or more suspiciousness scores reflect an impact of individual features on a prediction of a claim denial.
11. A non-transitory computer-readable storage medium comprising computer program code comprising computer-readable instructions that are executable by a processor to:
identify a set of claim features of at least one claim of one or more claims in an electronic format;
generate an embedding of tokens to provide a first input vector based, least in part, on:
application of a tokenization process to the identified set of claim features to provide first numeric values in a first sub-vector to express one or more diagnosis tokens and second numeric values in a second sub-vector to express one or more procedure tokens, the tokenization process comprising one or more machine-learning models to be determined based, at least in part, on a history of claims; and
element-wise multiplication of at least some of elements of the first sub-vector with at least some elements of the second sub-vector to provide the first input vector;
execute one or more first neural networks to compute one or more suspiciousness scores and a likelihood of denial of at least a portion of the at least one claim based, at least in part, on the first input vector;
communicate at least one of the suspiciousness scores to an operator/user; and
process inputs from the operator/user to modify at least a portion of the at least one claim to thereby affect the likelihood of denial of the at least a portion of the at least one claim.
12. The non-transitory computer-readable storage medium of claim 11 , wherein at least one of the inputs comprises a correction to a previously incorrectly entered claim feature.
13. The non-transitory computer-readable storage medium of claim 11 , wherein the at least one of the one or more first neural networks is determined according to a loss function, the loss function being based, at least in part, on a binary cross-entropy loss for a denial probability prediction, a categorical cross-entropy loss for a set of service-level denial reason code classifications or a categorical cross-entropy loss for a set of claim-level denial reason code classifications, or a combination thereof.
14. The non-transitory computer-readable storage medium of claim 11 , wherein the embedding of tokens to be further determined based, at least in part, on:
application of the tokenization process to the identified set of claim features to provide binary values in a third sub-vector to express one or more demographic tokens.
15. The non-transitory computer-readable storage medium of claim 11 , wherein the one or more machine-learning models comprise one or more second neural networks, and wherein the instructions are further executable by the processor to:
obtain a second input vector comprising a portion of the set of claim features; and
generate the embedding of tokens to comprise a higher dimensionality than a dimensionality of the second input vector.
16. The non-transitory computer-readable storage medium of claim 11 , wherein the instructions are further executable by the processor to:
determine at least one of the suspiciousness scores based, at least in part, on a computed gradient of a contribution of at least one claim feature of the identified set of claim features to the likelihood of denial of the at least a portion of the claim.
17. The non-transitory computer-readable storage medium of claim 11 , and wherein the instructions are further executable by the processor to:
determine a first set of token sequences representing procedure tokens, a second set of token sequences representing diagnosis tokens or a third set of tokens representing demographic tokens, or a combination thereof, based, at least in part on the identified set of claim features of the at least one claim.
18. The non-transitory computer-readable storage medium of claim 11 , wherein the instructions are further executable by the processor to determine the likelihood of denial of the at least a portion of the at least one claim based, at least in part, on a first reason code sequence, a second reason code sequence and a response date estimation, wherein the first reason code sequence includes likelihood scores for claim-level reason codes in a set of claim-level reason codes, and wherein the second reason code sequence includes likelihood scores for service-level reason codes in a set of service-level reason codes.
19. The non-transitory computer-readable storage medium of claim 18 , wherein at least one of the one or more machine-learning models comprises:
a first set of task-specific output layers to generate the first reason code sequence,
a second set of task-specific output layers to generate the second reason code sequence, and
a third set of task-specific layers to generate the response date estimation.