Nonparametric percentile curve estimation for a nonnegative marker with excessive zeros

Area of Science:

• Biostatistics
• Preventive Medicine
• Medical Imaging Analysis

Background:

• Percentile curves are established for infant growth metrics.
• The Agatston score, introduced in 1990, is a key marker in preventive cardiology.
• Non-negative scores like the Agatston score often exhibit an excess of zero values in healthy populations.

Purpose of the Study:

• To develop a nonparametric method for estimating percentile curves for medical markers, particularly those with zero-inflated distributions.
• To address the challenge of accurately representing skewed data, such as the Agatston score, in percentile curve analysis.

Main Methods:

• A nonparametric approach using lowess smoothing for marker-positive scores.
• Transposing smoothed curves based on estimated proportions of zero scores.
• Validation through a simulation study with varying sample sizes (N=1,000 to 10,000).

Main Results:

• The method accurately estimates 50th, 75th, and 90th percentile curves, even with exponentially distributed markers and increasing zero proportions.
• Demonstrated robustness against outliers and adherence to the non-crossing property of percentile curves.
• Successfully applied to subgroup data, showcasing applicability to highly skewed datasets.

Conclusions:

• The developed nonparametric method provides an explicit, transferable, and reproducible procedure for percentile curve estimation.
• This approach is versatile and applicable to a broad range of markers and scores across diverse scientific disciplines beyond cardiovascular medicine.
• The method effectively handles zero-inflated and skewed data, enhancing its utility in medical research and clinical practice.