Methods for a similarity measure for clinical attributes based on survival data analysis

Area of Science:

• Computational biology
• Medical informatics
• Biostatistics

Background:

• Case-based reasoning (CBR) relies on past cases for decision support.
• The accuracy of CBR systems hinges on effective similarity measures.
• Clinical case comparison, especially using survival data, requires specialized similarity metrics.

Purpose of the Study:

• To propose a collection of methods for similarity measures tailored for clinical case comparison.
• To enhance decision support in scenarios with longitudinal survival data.
• To provide a robust similarity approach for randomized clinical trials.

Main Methods:

• Developed local similarity measures for nominal and numeric attributes using survival functions.
• Implemented an attribute weight calculation and feature selection method.
• Defined a global similarity measure based on the area between survival function curves.

Main Results:

• The proposed similarity measure demonstrated higher accuracy (0.909-0.998) in silico compared to existing methods (e.g., 0.657-0.831).
• Biomarker attributes received significantly higher weights (6.59-6.95 times) than non-biomarker attributes.
• Overall survival data proved effective for similarity calculations, especially when therapy-based measures are inapplicable.

Conclusions:

• The developed similarity measures are suitable for case-based reasoning applications utilizing survival data.
• This approach offers improved accuracy and robustness for clinical decision support.
• The methods are particularly valuable for analyzing data from clinical trials with survival endpoints.