Partial summary measures of the predictiveness curve.

Area of Science:

• Biostatistics
• Epidemiology
• Medical Informatics

Background:

• Predictiveness curves are crucial for evaluating biomarker performance in risk prediction.
• Assessing classification accuracy alongside predictiveness curves is vital for risk model utility.
• Existing methods for biomarker inference can be enhanced for specific risk ranges.

Purpose of the Study:

• To propose novel partial summary measures for predictiveness curves.
• To enable comparison of biomarkers over restricted risk ranges.
• To provide inferential tools for evaluating risk prediction models.

Main Methods:

• Development of partial total gain and partial proportion of explained variation measures.
• Application of one- and two-sample inferential tools.
• Simulation studies to assess statistical power and performance.

Main Results:

• The proposed partial summary measures effectively summarize predictiveness curves over restricted risk ranges.
• Inferential tools demonstrate adequate power in simulation studies.
• The methods are applicable for comparing biomarkers and assessing risk scores.

Conclusions:

• The novel partial summary measures offer valuable insights into biomarker utility for risk prediction.
• These methods facilitate biomarker comparison, especially with existing risk stratification thresholds.
• The approach is demonstrated effectively for Alzheimer's disease risk prediction accuracy assessment.