AutoPrognosis 2.0

AutoPrognosis 2.0 automates construction and optimization of diagnostic and prognostic machine-learning models for clinical risk prediction.

Key Features:

• Automated Machine Learning (AutoML): Leverages AutoML to search and optimize end-to-end machine-learning pipelines and hyperparameters.
• Pipeline optimization and model configuration: Identifies optimal model configurations and pipeline components for predictive performance on complex datasets.
• Model explainability: Incorporates tools to enhance interpretability of model outputs and risk scores.
• Handling complex patient covariates: Captures complex interactions among patient covariates within predictive models.
• Clinical demonstrator deployment: Facilitates deployment of trained models as clinical demonstrators in healthcare settings.

Scientific Applications:

• Prognostic risk scores: Construction and validation of prognostic risk scores, exemplified by a diabetes risk model developed using UK Biobank (prospective study of over 500,000 individuals) that showed superior discrimination versus existing expert clinical risk scores.
• Personalized diagnostics: Development of personalized diagnostic tools and individualized risk prediction models for clinical decision support.

Methodology:

Uses automated machine learning techniques to develop and optimize predictive models by automating pipeline construction, handling complex datasets, identifying optimal model configurations, and incorporating model explainability tools.