Systems and methods for management of clinical queues
Inventors
Nida, Dean • Brown-Hayes, Michael • Chung, James • Espinoza, Carolyn • Emerson, Michael • Larson, John • Tschirpke, Jennifer
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12080395-B2
Publication Date
2024-09-03
Expiration Date
Abstract
Systems and methods for generating and actively improving clinical queues are provided. In one embodiment, a method includes generating, with a plurality of predictive models, a plurality of risk scores for each individual a plurality of individuals based on medical claims of each individual, transmitting a clinical queue to a client device associated with a healthcare provider, the clinical queue comprising a list of individuals prioritized for healthcare intervention based on the plurality of risk scores, receiving feedback from the client device regarding the clinical queue, and updating at least one predictive model of the plurality of predictive models based on the feedback. In this way, members may be identified for intervention in a timely manner, multiple assessment methods may be combined to generate clinical queues, and the assessment methods may be actively improved over time.
Core Innovation
The invention evaluates individuals using multiple predictive models based on medical claims collected from one or more healthcare databases, where each medical claim is associated with one of a plurality of individuals. In a first stage, the system generates risk scores in different categories for the plurality of individuals using different machine learning models, and uses these risk scores to determine a clinical queue for a client device via a first machine learning model. The clinical queue is transmitted to a client device associated with a healthcare provider and includes a list of individuals prioritized for healthcare intervention based on the risk scores.
In a second stage, the invention receives feedback from the client device regarding the clinical queue, where the feedback includes a first label indicating whether a case in the clinical queue is opened and a second label indicating whether the case was reviewed. The invention creates a second training set comprising the feedback and determines reviewed and unopened cases so that the second label is displayed without showing the first label for those reviewed and unopened cases, and inclusion of the one or more cases in the clinical queue is negative only responsive to the reviewed and unopened condition. The invention retrains the first machine learning model using feature vectors constructed from patient data, mapping feature vectors to labels, and forms an updated model.
The invention then evaluates a performance of the updated model compared to the performance of the original model. If the performance is improved, the updated model is replaced in a model library; if performance has not improved, the updated model is discarded or disabled, or another model is selected from the library for use. Additionally, the invention describes an intelligent dashboard platform that aggregates patient data from multiple sources and orders modular display modules using an actionability prediction model trained from user interactions and preferences.
In the dashboard, the system provides a graphical user interface with modular dashboard modules, where multiple display modules are ordered based on actionability and relevance predictions. It includes an actionability feedback module that uses user-adjusted display order to update module ordering and corresponding models or user profiles. The dashboard supports selectable modules such as a pharmaceutical timeline module and a smart search, with controls for reordering display modules in the user interface.
Claims Coverage
The independent claims are directed to a two-stage clinical-queue generation and model-updating approach, together with an intelligent dashboard platform. Across the independent claims, the inventive approach centers on generating risk scores from medical claims using predictive models, transmitting a prioritized clinical queue to a client device, receiving opened/reviewed feedback labels, retraining and evaluating updated models using patient-data feature vectors mapped to labels, and applying performance-based replacement, disablement, discard, or selection of models in a model library.
Two-stage clinical queue generation with risk scores from medical claims
collecting medical claims from one or more healthcare databases, generating risk scores using predictive models based on medical claims of each individual, creating a first training set comprising the risk scores, determining a clinical queue for a client device via a first machine learning model, and transmitting the clinical queue comprising a list of individuals prioritized for healthcare intervention based on the risk scores
Model retraining from client feedback labels mapped to feature vectors
receiving feedback from the client device regarding the clinical queue with a first label indicating whether a case is opened and a second label indicating whether the case was reviewed, creating a second training set comprising the feedback, determining reviewed and unopened cases so that the second label is displayed without showing the first label for those cases, retraining the first machine learning model using feature vectors constructed from patient data, and mapping the feature vectors to labels
Performance-based updated model replacement or discard/disable/selection
evaluating performance of the updated model compared to performance of the original model, replacing the model in a model library if performance is improved, and discarding or disabling the updated model if performance has not improved, or selecting another model from the library for use
Client-server clinical queue system with interactive feedback display
providing a client device and a server to collect medical claims, generate risk scores with predictive models, determine and transmit a clinical queue prioritized for healthcare intervention, receive feedback labels from the client device, determine reviewed and unopened cases for negative inclusion and for displaying the second label without showing the first label for those cases, retrain and evaluate updated model performance, and either replace the updated predictive model or select another model from the library for use
Actionability-based modular dashboard ordering
aggregate patient data from multiple sources and order multiple modular display modules based on actionability and relevance using an actionability prediction model trained from user interactions and preferences, including an actionability feedback module and user-adjusted display order used to update module ordering and corresponding models or user profiles
Graphical user interface presentation with pharmaceutical timeline and smart search modules
display through a display subsystem a graphical user interface showing a clinical queue with an intelligent dashboard per individual, where the dashboard includes a selectable pharmaceutical timeline module and supports smart search and reordering controls
Across the independent claims, the core claim coverage centers on determining a prioritized clinical queue from medical-claims-derived risk scores using predictive models, updating models via client feedback labels through feature-vector construction and label mapping, and using performance evaluation to replace, disable, discard, or select models from a model library. The claims coverage also encompasses client-server presentation of the clinical queue with label-based feedback display and an intelligent dashboard platform that orders modular display modules using actionability prediction and user feedback.
Stated Advantages
Prioritize a list of individuals for healthcare intervention based on risk scores derived from medical claims.
Improved updated model performance can replace an existing model in a model library.
If updated model performance is not improved, the updated model can be discarded or disabled, or replaced by selecting another model.
Order modular dashboard display modules based on predicted actionability and relevance informed by user interactions and preferences.
Indicate negative inclusion in the clinical queue responsive to cases being reviewed and unopened, based on client feedback labels.
Documented Applications
Generating clinical queues for healthcare providers on a client device by prioritizing individuals for healthcare intervention based on risk scores from medical claims.
Providing an intelligent dashboard platform that aggregates patient data from multiple sources and orders modular display modules, including selectable pharmaceutical timeline modules and smart search, based on actionability and relevance predictions using an actionability prediction model trained from user interactions and preferences.
Interested in licensing this patent?