Systems and methods for extracting dates associated with a patient condition
Inventors
Gippetti, James • Phadke, Sharang • Amster, Guy • Singh, Nisha • Sridharma, Suganya • Estevez, Melissa • Ritten, John • Garapati, Sankeerth • Cohen, Aaron
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11908586-B2
Publication Date
2024-02-20
Expiration Date
Abstract
A model-assisted system for extracting patient information. A processor may be programmed to access a database storing one or more medical records associated with a patient and determine, using a first machine learning model and based on unstructured information included in the one or more medical records, whether the patient is associated with a condition. The processor may further be programmed to identify a date associated with the patient and determine, using a second machine learning model and based on the unstructured information, whether the patient is associated with the condition relative to the date. The processor may generate an output indicating whether the patient is associated with the condition and whether the patient is associated with the condition relative to the date.
Core Innovation
The invention relates to a model-assisted system for extracting patient-condition information and associated dates from unstructured information included in medical records. A first machine learning model identifies patients associated with a condition based on unstructured medical-record content, and the system identifies a subset of patients associated with the condition based on a first output from the first machine learning model.
For each patient in the subset, the system receives a user interface input of a date associated with the patient and uses a second machine learning model to determine whether the patient is associated with the condition relative to the date. Documents having a timestamp prior to a cutoff date are identified for input into the second machine learning model, where the cutoff date is based on a predetermined buffer period before or after the date, and this cutoff logic excludes documents that include one or more dates expressed in relative terms.
When excluded relative-date expressions are present, the system generates one or more pseudo-documents for input into the second machine learning model so the pseudo-documents account for one or more dates found in the unstructured information. The system then determines, based on a second output from the second machine learning model, whether the patient is associated with the condition relative to the date, and generates an output indicating whether the patient is associated with the condition and whether the patient is associated with the condition relative to the date.
Claims Coverage
The document provides three independent claims: a system claim, a method claim, and a non-transitory computer-readable medium claim. Each independent claim includes three inventive features centered on first identifying patients associated with a condition, then determining association relative to a user-provided date using a second machine learning model with cutoff/buffer logic and pseudo-documents.
Two-model patient-condition identification and relative-date determination
accesses a database storing a plurality of medical records associated with a plurality of patients; inputs unstructured information included in the plurality of medical records into a first machine learning model, the first machine learning model being trained using first training data to identify patients associated with a condition; identifies, based on a first output from the first machine learning model, a subset of the plurality of patients, the subset of the plurality of patients being associated with the condition; identifies, based on an input by a user through a user interface, a date associated with a patient of the subset of the plurality of patients; inputs unstructured information included in one or more documents into a second machine learning model, the second machine learning model being trained using second training data to indicate dates associated with the condition; determines, based on a second output from the second machine learning model, whether the patient is associated with the condition relative to the date; and generates an output indicating whether the patient is associated with the condition and whether the patient is associated with the condition relative to the date.
Cutoff-date document selection using a predetermined buffer period
identifies, within the plurality of medical records, one or more documents associated with the patient and having a timestamp prior to a cutoff date such that documents including one or more dates expressed in relative terms are excluded from the identified one or more documents, the cutoff date being based on a predetermined buffer period before or after the date.
Pseudo-document generation to account for dates in unstructured information
generates one or more pseudo-documents for input into the second machine learning model, the one or more pseudo-documents accounting for one or more dates within the unstructured information included in the one or more documents.
Across independent claim 1, 11, and 14, the core claim coverage centers on extracting patients associated with a condition from unstructured medical records using a first machine learning model, then determining whether each identified patient is associated with the condition relative to a user-provided date using a second machine learning model. The claims specifically incorporate cutoff/buffer logic to exclude documents with dates expressed in relative terms and generating pseudo-documents that account for one or more dates present in the unstructured information.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?