Analysing covariates with spike at zero: a modified FP procedure and conceptual issues

Area of Science:

• Epidemiology
• Biostatistics
• Medical Research

Background:

• Risk factors in epidemiological and clinical research often exhibit unique distributions, including a 'spike at zero' (e.g., smoking, alcohol consumption) and semi-continuous distributions due to measurement detection limits.
• Accurate modeling of dose-response functions is crucial for understanding disease etiology and developing effective interventions.
• Existing methods, such as fractional polynomial (FP) approaches, may require refinement to handle these complex data distributions effectively.

Purpose of the Study:

• To propose and theoretically justify a modified fractional polynomial (FP) procedure for modeling dose-response relationships.
• To systematically derive the theoretical shapes of dose-response curves under various distributional assumptions (normal, log-normal, gamma) within a logistic regression framework.
• To evaluate the performance of the modified FP procedure and compare it with a previously suggested method.

Main Methods:

• Theoretical derivation of dose-response curve shapes using logistic regression under normal, log-normal, and gamma distribution assumptions.
• Development and application of a modified fractional polynomial (FP) procedure to handle data with a 'spike at zero' and semi-continuous distributions.
• Performance evaluation through a simulation study comparing the modified FP procedure against the existing method.

Main Results:

• The modified FP procedure demonstrates robust performance in modeling dose-response functions with complex data distributions.
• Theoretical analysis provides insights into the shapes of dose-response curves under different distributional assumptions.
• The simulation study confirms the advantages of the modified FP procedure over the previously suggested method.

Conclusions:

• The modified fractional polynomial (FP) procedure offers an improved approach for modeling dose-response relationships in the presence of zero-inflated and semi-continuous data.
• This method enhances the analysis of risk factors in epidemiological and clinical research, particularly for conditions like breast cancer.
• The findings support the use of this refined statistical technique for more accurate risk assessment and disease modeling.