Biomarker Machine Learning for Diabetic Kidney Decline
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current clinical measurements and standard prognostic tools are inadequate for early identification of rapid kidney function decline in diabetic kidney disease, particularly in individuals with African ancestry, leading to delayed care and high incidence of kidney failure.
Innovation Solution
A machine learning model trained on biomarker data and human subject data, including sTNFR-1, sTNFR-2, KIM-1, and demographic factors, to predict progressive decline in kidney function, using a random forest algorithm and data preprocessing to harmonize and normalize data for accurate risk stratification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If standard clinical measurements (eGFR and uACR) are used for risk stratification, then the diagnostic approach is simple and widely available, but the ability to identify patients with rapid kidney function decline is insufficient
Solution Approach 1:
The patent combines multiple data sources including biomarker data (sTNFR-1, sTNFR-2, KIM-1), electronic health record data, and demographic information into a unified machine learning model. This integration allows the system to leverage the predictive value of each individual data type while achieving superior identification of rapid kidney function decline compared to any single measurement alone.
Solution Approach 2:
The patent replaces traditional mechanical/statistical risk calculation methods with a machine learning model that automatically learns complex patterns from data. The model substitutes conventional clinical scoring systems with an adaptive computational approach that can identify non-linear relationships and interactions among multiple variables, significantly improving prediction accuracy.
2Reliability
If machine learning models with multiple biomarkers are used, then the prediction accuracy improves, but the computational complexity and data processing requirements increase
Solution Approach 1:
The patent performs extensive data preprocessing and feature engineering before model training, including handling missing values, normalizing variables, and selecting relevant features. This preliminary preparation reduces the dimensionality and complexity of the input data, making the subsequent machine learning computation more efficient while maintaining high predictive accuracy.
Solution Approach 2:
The patent extracts and focuses on the most predictive features from a large set of available variables. By identifying and utilizing only the most relevant biomarkers and clinical parameters, the model achieves high accuracy without requiring computation across all possible variables, thereby reducing unnecessary computational complexity.
3Measurement precision
If comprehensive biomarker panels are implemented, then the identification of high-risk patients improves, but the cost and accessibility of the test decreases
Solution Approach 1:
The patent implements a tiered approach where the machine learning model can operate with varying levels of input data quality and completeness. The model can provide useful predictions even when not all biomarkers are available, allowing gradual implementation where clinics can start with available markers and add others over time, reducing upfront costs while maintaining predictive capability.
Data Source
AI summary
A non-transitory processor-readable medium stores code to be executed by a processor of a first computer device. The code causes the processor to receive, from a second computer device remote from the first computer device, a trained machine learning model. The code causes the processor to receive biomarker data and HSD of a diabetic human subject. The biomarker data indicates a level of at least one of the following biomarkers: sTNFR-1, sTNFR-2, KIM-1, and/or ratios to one another of any of the preceding. The HSD includes a metabolic factor, a health-related factor, or a demographic-related factor. The code causes the processor to execute the trained machine learning model to generate an indication of whether the diabetic human subject will experience a progressive decline in kidney function over a period of time.