Identifying missing questions by clustering and outlier detection
A machine learning system may be used to suggest clinical questions to ask during or after a patient appointment. A first encoder may encode information and a second encoder may encode second information related to the current patient appointment. An aggregate encoding may be generated using the encoded first information and encoded second information. The current patient appointment may be clustered with similar appointments based on the aggregate encoding. Outlier analysis may be performed to determine if the appointment is an outlier, and, if so, which features contribute the most to outlier status. The system may generate one or more questions to ask about the features that contribute the most to outlier status.
1. A computer-implemented method, the method comprising:
training a first neural network autoencoder to encode information of a first type into a first vector representation, the first autoencoder comprising a first encoder that maps a first input vector to the first vector representation and a first decoder that maps the first vector representation to a first output vector, where the first decoder is trained to reduce the a distance between the first output vector and the first input vector;
training a second neural network autoencoder to encode information of a second type into a second vector representation, the second autoencoder comprising a second encoder that maps a second input vector to the second vector representation and a second decoder that maps the second vector representation to a second output vector, where the second decoder is trained to reduce a distance between the second output vector and the second input vector;
providing a set of medical records of patient appointments;
extracting from the medical records first portions of information of the first type and second portions of information of the second type, encoding the first portions of information with the first neural network autoencoder, and encoding the second portions of information with the second neural network autoencoder;
generating an aggregate encoding for each medical record based on the encoded first portions of information and encoded second portions of information;
clustering, using a machine learning clustering method, the medical records based on at least on the aggregate encodings to generate medical records clusters;
providing an input medical record;
extracting from the input medical record a third portion of information of the first type and a fourth portion of information of the second type, encoding the third portion of information with the first neural network autoencoder, and encoding the fourth portion of information with the second neural network autoencoder;
generating an aggregate encoding for the input medical record based on the encoded third portion of information and encoded fourth portion of information;
determining for the input medical record a closest medical records cluster based at least on the aggregate encoding of the input medical record;
determining whether the input medical record is an outlier from the closest medical records cluster;
when it is determined that the input medical record is an outlier, determining one or more features of the input medical record that cause it to be an outlier; and
generating one or more clinical questions about the one or more determined features.
2. The computer-implemented method of claim 1 , wherein the first neural network autoencoder is a variational autoencoder.
3. The computer-implemented method of claim 1 , wherein the first neural network autoencoder uses Kullback-Leibler divergence.
4. The computer-implemented method of claim 1 , wherein the second neural network autoencoder is a variational autoencoder.
5. The computer-implemented method of claim 1 , wherein the third portion of information comprises information about one or more medications and the fourth portion of information comprises information about a diagnosis.
6. The computer-implemented method of claim 1 , wherein the machine learning clustering method is k-means clustering.
7. The computer-implemented method of claim 1 , further comprising determining whether the input medical record is an outlier from the closest medical records cluster by using a z-score.
8. A non-transitory computer-readable medium, the non-transitory computer-readable medium comprising instructions for:
training a first neural network autoencoder to encode information of a first type into a first vector representation, the first autoencoder comprising a first encoder that maps a first input vector to the first vector representation and a first decoder that maps the first vector representation to a first output vector, where the first decoder is trained to reduce a distance between the first output vector and the first input vector;
training a second neural network autoencoder to encode information of a second type into a second vector representation, the second autoencoder comprising a second encoder that maps a second input vector to the second vector representation and a second decoder that maps the second vector representation to a second output vector, where the second decoder is trained to reduce the a distance between the second output vector and the second input vector;
providing a set of medical records of patient appointments;
extracting from the medical records first portions of information of the first type and second portions of information of the second type, encoding the first portions of information with the first neural network autoencoder, and encoding the second portions of information with the second neural network autoencoder;
generating an aggregate encoding for each medical record based on the encoded first portions of information and encoded second portions of information;
clustering, using a machine learning clustering method, the medical records based at least on the aggregate encodings to generate medical records clusters;
providing an input medical record;
extracting from the input medical record a third portion of information of the first type and a fourth portion of information of the second type, encoding the third portion of information with the first neural network autoencoder, and encoding the fourth portion of information with the second neural network autoencoder;
generating an aggregate encoding for the input medical record based on the encoded third portion of information and encoded fourth portion of information;
determining for the input medical record a closest medical records cluster based at least on the aggregate encoding of the input medical record;
determining whether the input medical record is an outlier from the closest medical records cluster;
when it is determined that the input medical record is an outlier, determining one or more features of the input medical record that cause it to be an outlier; and
generating one or more clinical questions about the one or more determined features.
9. The non-transitory computer-readable medium of claim 8 , wherein the first neural network autoencoder is a variational autoencoder.
10. The non-transitory computer-readable medium of claim 8 , wherein the first neural network autoencoder uses Kullback-Leibler divergence.
11. The non-transitory computer-readable medium of claim 8 , wherein the second neural network autoencoder is a variational autoencoder.
12. The non-transitory computer-readable medium of claim 8 , wherein the third portion of information comprises information about one or more medications and the fourth portion of information comprises information about a diagnosis.
13. The non-transitory computer-readable medium of claim 8 , wherein the machine learning clustering method is k-means clustering.
14. The non-transitory computer-readable medium of claim 8 , further comprising determining whether the input medical record is an outlier from the closest medical records cluster by using a z-score.
15. A computer-implemented method, the method comprising:
training a first neural network encoder to encode information of a first type into a first vector representation;
training a second neural network encoder to encode information of a second type into a second vector representation;
providing a set of medical records of patient appointments;
extracting from the medical records first portions of information of the first type and second portions of information of the second type, encoding the first portions of information with the first neural network encoder, and encoding the second portions of information with the second neural network encoder;
clustering, using a machine learning clustering method, the medical records to generate medical records clusters;
providing an input medical record;
extracting from the input medical record a third portion of information of the first type and a fourth portion of information of the second type, encoding the third portion of information with the first neural network encoder, and encoding the fourth portion of information with the second neural network encoder;
determining for the input medical record a closest medical records cluster;
determining whether the input medical record is an outlier from the closest medical records cluster;
when it is determined that the input medical record is an outlier, determining one or more features of the input medical record that cause it to be an outlier; and
generating one or more clinical questions about the one or more determined features.
16. The computer-implemented method of claim 15 , wherein the first neural network encoder is a variational autoencoder.
17. The computer-implemented method of claim 15 , wherein the first neural network encoder uses Kullback-Leibler divergence.
18. The computer-implemented method of claim 15 , wherein the second neural network encoder is a variational autoencoder.
19. The computer-implemented method of claim 15 , wherein the third portion of information comprises information about one or more medications and the fourth portion of information comprises information about a diagnosis.
20. The computer-implemented method of claim 15 , wherein the machine learning clustering method is k-means clustering.
21. The computer-implemented method of claim 15 , further comprising determining whether the input medical record is an outlier from the closest medical records cluster by using a z-score.
22. A non-transitory computer-readable medium, the non-transitory computer-readable medium comprising instructions for:
training a first neural network encoder to encode information of a first type into a first vector representation;
training a second neural network encoder to encode information of a second type into a second vector representation;
providing a set of medical records of patient appointments;
extracting from the medical records first portions of information of the first type and second portions of information of the second type, encoding the first portions of information with the first neural network encoder, and encoding the second portions of information with the second neural network encoder;
clustering, using a machine learning clustering method, the medical records to generate medical records clusters;
providing an input medical record;
extracting from the input medical record a third portion of information of the first type and a fourth portion of information of the second type, encoding the third portion of information with the first neural network encoder, and encoding the fourth portion of information with the second neural network encoder;
determining for the input medical record a closest medical records cluster;
determining whether the input medical record is an outlier from the closest medical records cluster;
when it is determined that the input medical record is an outlier, determining one or more features of the input medical record that cause it to be an outlier; and
generating one or more clinical questions about the one or more determined features.
23. The non-transitory computer-readable medium of claim 22 , wherein the first neural network encoder is a variational autoencoder.
24. The non-transitory computer-readable medium of claim 22 , wherein the first neural network encoder uses Kullback-Leibler divergence.
25. The non-transitory computer-readable medium of claim 22 , wherein the second neural network encoder is a variational autoencoder.
26. The non-transitory computer-readable medium of claim 22 , wherein the third portion of information comprises information about one or more medications and the fourth portion of information comprises information about a diagnosis.
27. The non-transitory computer-readable medium of claim 22 , wherein the machine learning clustering method is k-means clustering.
28. The non-transitory computer-readable medium of claim 22 , further comprising determining whether the input medical record is an outlier from the closest medical records cluster by using a z-score.