Machine learning systems and methods for predicting risk of renal function decline
Inventors
Cha, Theodore • Sun, Hai Po • Kipers, Chris • Fielding, Oliver • Son, Jung Hoon • Lee, Edward
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-10978176-B2
Publication Date
2021-04-13
Expiration Date
Abstract
Systems, methods and apparatuses are described herein that employ machine learning techniques to assess a likelihood or risk that one or more patients will experience an adverse outcome, such as a decline in renal function, within one or more timeframes. The embodiments may utilize patient data relating to demographics, vital signs, diagnoses, procedures, diagnostic tests, biomarker assays, genetic tests, behaviors, and/or patient symptoms, to determine risk information, such as important predictive features and patient risk scores. And the embodiments may automatically execute patient workflows, such as providing treatment recommendations to providers and/or patients, based on determined risk scores.
Core Innovation
The disclosed invention relates to a method of determining and automatically updating a risk of renal function decline for a plurality of patients via a machine-learning system. The method stores patient records in memory and transmits those records to a trained machine-learning system that has previously been trained to identify important features and calculate stored weights based on feature importance for determining a probability that a patient experiences an outcome relating to a decline in renal function within a prediction time period. The patient information includes demographics and a plurality of lab tests, and the machine-learning system calculates feature values from the stored weights and determines a current risk score for each transmitted patient record.
The lab-test inputs include first lab tests associated with biomarker variables tumor necrosis factor receptor-1 (TNFR1), tumor necrosis factor receptor-2 (TNFR2), and kidney injury molecule-1 (KIM1), as well as a second set of lab tests that includes lab test variables measured on different dates. The lab-test variables of the second set include eGFR, UACR, serum creatinine, BUN, and endostatin, and the trained machine-learning system uses important features comprising demographic features and lab test features, including a feature relating to a first lab test variable, a second lab test variable, and a time period that includes both the first date and the second date.
After receiving the current risk scores from the trained machine-learning system, the computer stores the current risk scores and determines whether each current risk score is greater than a predetermined threshold. When a threshold is satisfied, the method transmits a notification comprising at least a portion of the respective patient record to one or more providers associated with the patient. The method also continuously determines whether new patient information is available from a plurality of data sources including an EHR system, automatically ingests and preprocesses the new information using a centralized data schema, correlates the preprocessed data to stored patient records, updates the patient records, and repeats risk scoring and notification for the updated patient records.
Claims Coverage
Independent claim coverage describes an end-to-end pipeline for determining and automatically updating a machine-learning risk score for renal function decline, including continual ingestion of EHR data, preprocessing into a centralized data schema, risk scoring, threshold-based notifications, and repetition for updated patient records. Three inventive features are centered on weighted feature importance, lab-test feature inputs, and continuous updating with notification workflow.
Machine-learning risk scoring for renal function decline within a prediction time period
Determining, by the trained machine-learning system, a current risk score based on feature values calculated for important features, the current risk score relating to a probability that the patient experiences an outcome relating to a decline in renal function within the prediction time period.
Weighted important features from demographic and lab-test features including time-period features
Identifying important features from a plurality of potential features and calculating and storing weights associated with important features based on the importance of each respective feature in determining a probability, where the important features comprise a plurality of demographic features and a plurality of lab test features including a feature relating to a first lab test variable, a second lab test variable, and a time period that includes both the first date and the second date.
Lab-test feature inputs including TNFR1, TNFR2, KIM1 and second-set variables
Employing patient information comprising a plurality of lab tests, including first lab tests associated with biomarker variables TNFR1, TNFR2, and KIM1, and a second set of lab tests associated with lab test variables measured on different dates and relating to at least one of eGFR, UACR, serum creatinine, BUN, and endostatin.
Threshold-based notifications to providers based on the current risk score
Determining whether the current risk score associated with each patient record is greater than a predetermined threshold and, if so, transmitting a notification comprising at least a portion of the respective patient record to one or more providers associated with the respective patient.
Continuous updating using ingesting new EHR information, preprocessing via centralized schema, correlating to patient records, and repeating scoring
Continuously determining whether new patient information is available, automatically ingesting the new patient information, preprocessing it in accordance with a centralized data schema to create preprocessed data records, correlating each preprocessed data record to one of the patient records stored in memory to update the patient records, and repeating risk scoring and notification for the updated patient records.
The claim coverage centers on an automatically updating machine-learning risk scoring method for renal function decline that uses weighted demographic and lab-test features, including TNFR1, TNFR2, KIM1 and time-period lab features, stores resulting risk scores, and triggers provider notifications when scores exceed a predetermined threshold while continuously ingesting and preprocessing new EHR data.
Stated Advantages
Documented Applications
No documented applications found
Interested in licensing this patent?