Systems and methods to support medical therapy decisions
Inventors
Pappada, Scott M. • Feeney, John J. • DePriest, William N.
Assignees
Publication Number
US-11464456-B2
Publication Date
2022-10-11
Expiration Date
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Systems and methods for supporting medical therapy decisions are disclosed that utilize predictive models and electronic medical records (EMR) data to provide predictions of health conditions over varying time horizons. Embodiments also determine a 0-100 health risk index value that represents the “risk” for a patient to acquire a health condition based on a combination of real-time and predicted EMR data. The systems and methods receive EMR data and use the predictive models to predict one or more data values from the EMR data as diagnostic criteria. In some embodiments, the health condition trying to be avoided is Sepsis and the health risk index is a Sepsis Risk Index (SRI). In some embodiments, the predictive models are neural network models such as time delay neural networks.
Core Innovation
Systems and methods for supporting medical therapy decisions are disclosed that utilize predictive models and electronic medical records (EMR) data to provide predictions of health conditions over varying time horizons. Embodiments also determine a 0-100 health risk index value that represents the "risk" for a patient to acquire a health condition based on a combination of real-time and predicted EMR data. In some embodiments, the health condition trying to be avoided is Sepsis and the health risk index is a Sepsis Risk Index (SRI), and in some embodiments the predictive models are neural network models such as time delay neural networks.
The methods receive EMR data and use the predictive models to predict one or more data values from the EMR data as diagnostic criteria at multiple time horizons including one, two, and three hour time horizons from a current patient state. The systems determine a weighted factor value of a weighted data type and an unweighted factor value of an unweighted data type for each of the current patient state and the future patient states, normalize the weighted and unweighted factor values for each time horizon as a risk index for each time horizon, and normalize the time horizon risk indices as an overall risk index. The SRI is described as a representation of a risk for a patient to acquire sepsis and the risk index algorithm applies a penalty when data sources persist or evolve towards a risk-state over time.
The invention addresses shortcomings in optimizing the timing of antibiotic therapy and the need for early and accurate prediction of sepsis or septic shock prior to its onset. The background describes that the risk of mortality in patients with severe sepsis or septic shock may increase significantly with as little as a one hour delay in antibiotic administration and that existing institution antibiograms can often exist outside of the EMR system so key patient-specific factors are not easily integrated, while monitoring data inputs from multiple electronic and non-electronic sources can cause delays or errors in the appropriate, timely, and effective delivery of antibiotics.
Claims Coverage
This section identifies independent claims and extracts their main inventive features from the patent.
Receiving electronic medical record (EMR) data
Receiving electronic medical record (EMR) data.
Predicting EMR data values at multiple future patient states
Predicting one or more data values from the EMR data at at least two future patient states from a current patient state.
Determining weighted and unweighted factor values
Determining a weighted factor value of a weighted data type for each of the current patient state and the at least two future patient states, and determining an unweighted factor value of an unweighted data type for each of the current patient state and the at least two future patient states.
Normalizing per-horizon and overall risk indices
Normalizing the weighted factor value and the unweighted factor value for each of the current patient state and the at least two future patient states as a risk index for each state, and normalizing the risk index for each state as an overall risk index.
Using time-delay neural network models for prediction
Predicting one or more data values with a time-delay neural network model utilizing a history of EMR/patient-specific data and current data.
Using specified EMR data types as inputs
Receiving EMR data comprising a heart rate, a respiration rate, a systolic blood pressure, a mean arterial pressure, a temperature, a urine output, a blood glucose level, a white blood cell count, a creatinine level, a platelet count, and a lactate level.
Comparing predicted values to thresholds and deriving risk indices
Comparing predicted values to a threshold for each EMR data type, determining a risk index value for each data type at a time horizon, and normalizing each risk index value to an overall risk index value.
Identifying therapy risk as a sepsis risk index (SRI)
The overall risk index may be a Sepsis Risk Index (SRI) and the risk index value has a range from about 0 to about 100 representing risk for a patient to acquire sepsis.
Designating weighted and unweighted EMR data types
Designating weighted data types selected from the group consisting of heart rate, respiration rate, systolic blood pressure, mean arterial pressure, temperature, and urine output, and designating unweighted data types selected from the group consisting of blood glucose level, white blood cell count, creatinine level, platelet count, and lactate level.
Determining directive recommendations using SOFM
Determining a directive recommendation utilizing a Self-Organizing Feature Map (SOFM) wherein an input vector node based on the current patient state is mapped to a closest vector node of the SOFM to define the directive recommendation.
Displaying EMR and predictive model windows
Dynamically displaying an electronic medical record (EMR) window configured to display EMR data values and a predictive model window configured to display predicted EMR data values determined with a time-delay neural network model utilizing a history of EMR/patient-specific data and current data.
Displaying risk index and antibiogram access
Dynamically displaying a risk index window configured to display a risk index value and an access interface configured to provide access to an antibiogram of the health care facility.
The independent claims focus on receiving EMR data, predicting EMR data values at multiple future patient states using time-delay neural networks, calculating weighted and unweighted factor values and normalizing them into per-horizon and overall risk indices (including a Sepsis Risk Index of about 0-100), comparing predicted values to thresholds, determining directive recommendations via SOFM, and displaying EMR data, predictions, risk indices, and access to an institution antibiogram.
Stated Advantages
Supports earlier intervention and more effective administration of antibiotics by predicting sepsis risk before onset.
Provides a numeric risk index based on a combination of real-time and predicted EMR data as a novel diagnostic approach.
Integrates multiple pertinent EMR data sources, predictive models, and visualization into a single platform to support decision-making and trust.
Provides time horizon predictions and a penalty function so a high risk index can be assessed when predicted values exceed diagnostic criteria over time.
May contribute to reduction of mortality and ICU/hospital length of stay and improved patient outcomes based on earlier alerts and interventions.
Documented Applications
Supporting antibiotic therapy decisions, including timing and selection of antibiotics.
Predicting sepsis risk using a Sepsis Risk Index (SRI) derived from current and predicted EMR data to identify when to consider initiating broad spectrum antibiotic therapy.
Providing directive antibiotic therapy recommendations using Self-Organizing Feature Maps to suggest specific antibiotics and dosages based on patient data and outcomes.
Visualizing EMR data, predictive model outputs, and risk index values via a user interface with access to an institution antibiogram.
Simulated real-time playback of patient data and retrospective analysis of patient records for decision support and evaluation.
Interested in licensing this patent?