Identification of patient sub-cohorts and corresponding quantitative definitions of subtypes as a classification system for medical conditions
A classification method and system for medical conditions based on the concept of subtypes, which are classes of patients whose medical fact patterns as analyzed in an N-dimensional space places them closer to other patients belonging to the same subtype than to patients who belong to different subtypes and, who share similar likelihood of certain specified outcomes. A computer system processes patient data for a plurality of patients from a set of patients called a cohort. The computer system processes the patient data for the cohort to group patients into sub-cohorts of similar patients, i.e., each sub-cohort includes patients who have similar medical fact patterns in their patient data. Patients in different sub-cohorts generally, but not necessarily, have significant differences in their patient data. The computer system generates quantitative definitions, describing the patients in the sub-cohorts.
1. A computer system for encoding patient data for a plurality of patients into a respective ordered collection of features representing the medical facts for each patient in the plurality of patients, for input to a computational model, the computer system comprising:
a processing system comprising a processing device and memory, wherein the processing device processes computer program instructions to perform operations;
computer storage connected to the processing system and storing at least:
a. patient data comprising data representing a respective plurality of medical events for each of a plurality of patients, wherein data representing a medical event comprises at least one field and a respective value for each field, and
b. a library of medical instance definitions, wherein each medical instance definition comprises a respective mapping of data representing one or more medical events into data representing a respective medical instance, wherein the respective medical instance is a more general, less granular, more generic, or less specific representation of a medical fact about a patient than the one or more medical events; and
computer program instructions which, when processed by the processing system, cause the processing system to perform operations to:
access a data structure specifying, for each dimension of an N-dimensional vector, a respective medical instance definition from the library:
for each patient in the plurality of patients, convert the data representing a respective plurality of medical events for the patient into a respective N-dimensional vector of medical instances for the patient, by, for each dimension of the N-dimensional vector, applying the respective medical instance definition specified by the data structure to the data representing medical events for the patient to generate a respective medical instance for the dimension; and
transmitting the respective N-dimensional vectors for the patients as the ordered collection of features representing the patients to the computational model for training or classification.
2. The computer system of claim 1 , wherein the computer storage further stores a quantitative definition for a subtype characterizing a medically interesting sub-cohort, wherein the quantitative definition comprises data indicative of the data structure specifying the respective medical instance definitions used to generate the N-dimensional vectors for the patients and a subtype definition, wherein the subtype definition comprises a data structure that stores logic indicating:
a. an operation to be performed to process the respective N-dimensional vector for a patient to determine membership of the patient in the subtype, and
b. parameters used by the operation.
3. The computer system of claim 2 , wherein the computational model comprises computer program instructions which, when processed by the processing system, cause the processing system to perform operations to apply the operation and with the parameters specified by a subtype definition to one or more N-dimensional vectors for patients to determine whether the patients are members of the sub-cohort characterized by the subtype.
4. The computer system of claim 1 , wherein data representing a medical event includes a time for the medical event.
5. The computer system of claim 4 , wherein the respective N-dimensional vectors are generated based on data representing medical events within a specified time period.
6. The computer system of claim 1 , wherein, for at least one medical instance definition applied to one or more medical events, the medical instance is a more general representation of a medical fact about a patient than the one or more medical events.
7. The computer system of claim 1 , wherein, for at least one medical instance definition applied to one or more medical events, the medical instance is a less granular representation of a medical fact about a patient than the one or more medical events.
8. The computer system of claim 1 , wherein, for at least one medical instance definition applied to one or more medical events, the medical instance is a more generic representation of a medical fact about a patient than the one or more medical events.
9. The computer system of claim 1 , wherein, for at least one medical instance definition applied to one or more medical events, the medical instance is a less specific representation of a medical fact about a patient than the one or more medical events.
10. The computer system of claim 1 , wherein there are fewer medical instances for representing medical facts for patients than medical events.
11. The computer system of claim 1 , wherein medical facts for patients may be represented inconsistently in the patient data.
12. The computer system of claim 11 , wherein the respective N-dimensional vectors for the patients have reduced inconsistency with respect to the medical events for the patients.
13. The computer system of claim 11 , wherein the respective N-dimensional vectors for the patients have improved consistency with respect to the medical events for the patients.
14. The computer system of claim 1 , wherein medical facts for patients may be represented with high dimensionality in the patient data.
15. The computer system of claim 14 , wherein the respective N-dimensional vectors for the patients have reduced dimensionality with respect to the medical events for the patients.
16. The computer system of claim 1 , wherein medical facts for patients may be represented with high specificity in the patient data.
17. The computer system of claim 16 , wherein the respective N-dimensional vectors for the patients define an outcome-oriented generality with respect to the medical events for the patients.
18. The computer system of claim 1 , wherein the respective N-dimensional vectors representing patients map the patients to respective points in an N-dimensional space.
19. The computer system of claim 1 , wherein the library of medical instance definitions comprises an operation that replaces a plurality of medical events with a single medical instance.
20. The computer system of claim 19 , wherein the operation that replaces the plurality of medical events with the single medical instance comprises an operation based on co-occurrence of the plurality of medical events in a patent history.
21. The computer system of claim 1 , wherein the library of medical instance definitions comprises an operation that replaces a first medical event with a medical instance representing a second medical event more general than the first medical event.
22. The computer system of claim 21 , wherein the first medical event comprises a first code in a hierarchy of codes and the medical instance comprises a second code more general in the hierarchy of codes than the first code.
23. The computer system of claim 1 , wherein a medical instance definition comprises an operation performed on patient data that maps a medical event to a corresponding medical instance by generalizing specific types of medical events into a more general type of medical instance.
24. The computer system of claim 1 , wherein a medical instance definition comprises an operation performed on patient data that maps a medical event to a corresponding medical instance by replacing different types of medical events that represent a same medical fact into a type of medical instance.
25. The computer system of claim 1 , further comprising:
computer program instructions which, when processed by the processing system, cause the processing system to perform operations to derive medical instances based on the patient data.
26. The computer system of claim 25 , wherein derivation of medical instances is based on co-occurrence relations.
27. The computer system of claim 25 , wherein derivation of medical instances is based on grouping codes which relate to the same condition.
28. The computer system of claim 1 , wherein the library comprises data structures specifying, for each medical instance:
a. a set of one or more medical events that are members of the medical instance;
b. one or more operations used to process the set of medical events;
c. a label or a key uniquely identifying the medical instance; and
d. a human-readable description of the medical instance.
29. The computer system of claim 1 , further comprising:
computer program instructions which, when processed by the processing system, cause the processing system to perform operations to characterize a sub-cohort of patients based on the medical instances computed for the patients.
30. A computer system, comprising:
a processing system comprising a processing device and memory, wherein the processing device processes computer program instructions to perform operations;
computer storage connected to the processing system and storing at least:
a. patient data comprising data representing a respective plurality of medical events for each of a plurality of patients, wherein data representing a medical event comprises at least one field and a respective value for each field,
b. a library of medical instance definitions, wherein each medical instance definition comprises a mapping of data representing one or more medical events into data representing a medical instance, wherein the medical instance is a more general, less granular, more generic, or less specific representation of a medical fact about a patient than the one or more medical events, and
c. a quantitative definition for a subtype characterizing a medically interesting sub-cohort of patients, wherein patients within the sub-cohort can receive a medical treatment for the sub-cohort, wherein the quantitative definition comprises:
i. a data structure specifying, for each dimension of an N-dimensional vector, a respective medical instance definition from the library for generating a value for the dimension,
ii. an operation to be performed to process a respective N-dimensional vector for a patient to determine membership of the patient in the sub-cohort, and
iii. parameters used by the operation; and
computer program instructions which, when processed by the processing system, cause the processing system to perform operations to:
access the data structure specifying the respective medical instance definitions,
for each patient in the plurality of patients, apply the medical instance definitions specified by the data structure to the respective data representing medical events for the patient to generate a respective N-dimensional vector of medical instances for the patient, and
process the respective N-dimensional vectors for the patients using the operation and parameters specified by the subtype definition to determine, for each of the patients, whether the patient is a member of the sub-cohort.