| CPC G16H 70/20 (2018.01) [G06N 20/00 (2019.01); G16H 10/60 (2018.01); G16H 50/30 (2018.01)] | 18 Claims |
|
1. A method for providing clinical practice guidelines by a processor, comprising:
receiving evidence data and patient data from one or more data sources, wherein receiving the evidence data and the patient data includes receiving an input set of cohort selection criteria containing pairs of discriminative variables and corresponding values descriptive of a health-related concept;
executing machine learning logic to generate a cohort model to identify statistically significant patterns and associated outcomes related to one or more cohorts detected in the one or more data sources, wherein identifying the statistically significant patterns and associated outcomes includes filtering the patient data by the discriminative variables and corresponding values using a sequential pattern mining operation to identify the one or more cohorts, and confirming the statistically significant patterns and associated outcomes using a validation table;
querying the evidence data in the one or more data sources using the identified statistically significant patterns and associated outcomes to enrich the statistically significant patterns to enrich and validate the associated outcomes with the evidence data, wherein the querying includes inputting, as the query, a tuple containing information associated with a respective outcome, the tuple including an outcome value, a concept identifier, and a semantic identifier, wherein the concept identifier and the semantic identifier are extracted from a unified medical language system (UMLS), and wherein the validating of the associated outcomes with the evidence data further includes:
identifying a data element in aggregated patient data in the one or more data sources according to the data element containing a same concept identifier and semantic identifier as the respective outcome, wherein, responsive to the query containing input terminology of a hierarchy of concepts, data elements in the aggregated patient data having a parent of the concept identifier of the respective outcome is used, and
building the validation table by determining a number of patients in the aggregated patient data that include a respective pattern and the data element containing the same concept identifier;
automatically generating, by the machine learning logic, one or more clinical practice guidelines (CPGs) having patient recommendations according to the one or more cohorts using information contained on a plurality of templates populated with the evidence data and patient data from the one or more data sources, wherein at least a first of the plurality of templates comprises a first imperative recommendation in which the information describes a first process to achieve a desirable outcome, a second of the plurality of templates comprises a second imperative recommendation in which the information describes a second process leading to an undesirable outcome, and a third of the plurality of templates comprises a descriptive recommendation in which the information links a condition found in one of the one or more cohorts to a condition found in another one of the one or more cohorts;
in conjunction with automatically generating the one or more CPGs, assembling each respective CPG into a document containing a plurality of sections, wherein the plurality of sections include:
an opening section having been populated with at least one of a list of the discriminative variables and corresponding values specified as selection criteria, and statistics and characteristics of the one or more cohorts selected by the input set of cohort selection criteria,
a body section having been populated with the patient recommendations, wherein each patient recommendation listed in the body section includes metadata inclusive of an original pattern discovered in the statistically significant patterns and associated outcomes, and the validation table, and
a closing section having been populated with a listing of all references used to provide the patient recommendations; and
collecting feedback data of an accuracy of the one or more CPGs and using the feedback data to re-train the cohort model to generate optimized CPGs.
|