Phenotypic Feature Sequence Mining for Clinical Decision Support
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence-based diagnostic decision support systems face challenges such as low sensitivity and specificity, excessive false-negative determinations, interference with clinician workflow, and inability to accommodate diverse and evolving patient conditions, leading to suboptimal diagnostic and therapeutic outcomes.
Innovation Solution
A system and method for determining phenotypic findings using anticipative sequence-mining and trajectory-mining, which analyzes electronic health records to identify relevant clinical conditions and diagnoses, providing dynamic decision support without disrupting clinician workflow, and enabling predictive, preventative, and monitoring services.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional AI-based diagnostic systems are used, then diagnostic support is provided, but sensitivity and specificity are low due to inability to represent diverse patient presentations
Solution Approach 1:
The system dynamically adapts to diverse patient presentations by continuously learning from new data patterns. The machine learning models are trained on diverse clinical data and can adjust their predictions based on the specific context of each patient, allowing the system to represent and accurately diagnose a wide range of phenomena without requiring manual reconfiguration.
Solution Approach 2:
The system changes its operational parameters by adjusting model training data sources, feature weights, and prediction thresholds based on the specific clinical context. This allows the system to optimize its sensitivity and specificity for different disease presentations while maintaining overall adaptability across diverse patient populations.
2Reliability
If features are collected to improve sensitivity and specificity, then diagnostic accuracy improves, but the number of features required becomes impractically large
Solution Approach 1:
The system extracts only the most relevant and informative features from the vast amount of available clinical data. Through feature selection algorithms and machine learning models, it identifies and focuses on the critical features that contribute most to accurate diagnosis, filtering out redundant or less informative data points to reduce complexity while maintaining high diagnostic accuracy.
Solution Approach 2:
The system uses partial action by collecting and analyzing only the essential features needed for diagnosis rather than all possible features. This selective approach allows the system to achieve adequate sensitivity and specificity without requiring comprehensive collection of every possible clinical parameter, making the solution practically feasible.
3Reliability
If the system provides comprehensive diagnostic support, then diagnostic accuracy improves, but it interferes with and disrupts clinician workflow
Solution Approach 1:
The system acts as an intermediary tool that supports rather than replaces clinician decision-making. It provides diagnostic recommendations, risk assessments, and information synthesis in a way that complements the clinician's expertise, allowing healthcare professionals to maintain control over their workflow while benefiting from enhanced analytical capabilities.
Solution Approach 2:
The system incorporates feedback mechanisms that allow clinicians to review and correct AI recommendations. This feedback loop ensures that the system learns from actual clinical outcomes and adjusts its predictions accordingly, while also giving clinicians the final say in diagnostic decisions, thereby maintaining workflow efficiency and professional autonomy.
4Productivity
If the system acts autonomously to provide diagnostic conclusions, then decision speed improves, but it contradicts the fiduciary role of the clinician
Solution Approach 1:
The system performs preliminary analytical actions by pre-processing data, generating initial diagnoses, and preparing recommendations before the clinician makes the final decision. This preliminary action speeds up the overall diagnostic process by handling routine analytical tasks automatically, while the clinician retains authority for final decision-making and can quickly review and approve or modify the AI-generated conclusions.
Data Source
AI summary
Systems, methods, and computer-readable storage media are provided for determining and ascribing clinical conditions or diagnoses to patients and provide them to a caregiver, such as attending clinicians or other appropriate health services personnel. In particular, embodiments of the disclosure determine likely phenotypic findings that are salient to the decision-making context for a current human patient, based on anticipative sequence-mining and trajectory-mining. A sequential pattern mining and sequence itemset matching system is provided for determining likely, temporally-relevant concepts that are manifested in the information that is produced during the course of a patient's care. A clinician or caregiver may be provided the sequence itemset matching by generating a list or notice. In addition or alternatively, the results may be stored in an EHR associated with the patient.