Thick Data Analytics (TDA): An Iterative and Inductive Framework for Algorithmic Improvement.

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Artificial Intelligence in Healthcare

Background:

• A significant gap exists between developing predictive models and their safe, effective deployment in real-world, high-stakes scenarios.
• Human expert reasoning is crucial for identifying model limitations and ensuring patient safety during clinical decision-making.
• Existing statistical models often fail to utilize the full spectrum of real-world data available beyond observable quantities.

Purpose of the Study:

• To introduce a Thick Data Analytics (TDA) framework for incorporating expert human insight into model evaluation.
• To address the limitations of purely data-driven approaches by leveraging qualitative expert knowledge.
• To improve the safety, actionability, and acceptability of risk prediction models in clinical practice.

Main Methods:

• Proposed a Thick Data Analytics (TDA) framework to elicit and combine expert insights with model predictions.
• Developed a sampling procedure to identify informative cases for in-depth expert review.
• Utilized expert feedback on problem formulation and data interpretation to refine model development and deployment strategies.

Main Results:

• Demonstrated how expert insights can identify richer information sources beyond standard statistical inputs.
• Showcased the value of expert re-framing and re-evaluation of prediction problems for real-world applicability.
• Illustrated the iterative improvement of risk prediction models through integrated expert feedback.

Conclusions:

• Thick Data Analytics (TDA) provides a structured approach to incorporate essential human expertise into predictive model evaluation.
• Integrating expert insights leads to safer, more actionable, and clinically acceptable risk prediction models.
• This framework facilitates iterative model development, ultimately enhancing decision-making in critical care settings.