A graphical method for assessing risk factor threshold values using the generalized additive model: the multi-ethnic...

Area of Science:

• Biostatistics
• Epidemiology
• Health Outcomes Research

Background:

• Continuous variable dichotomization is common in health research for simplifying analysis.
• Existing threshold selection methods (e.g., median splits) are often ad-hoc and can introduce bias.
• This bias may lead to underestimation of risk factor effects on health outcomes.

Purpose of the Study:

• To propose an improved method for threshold selection in continuous variable dichotomization.
• To account for mixture distributions in health outcomes and adjust for covariates.
• To provide empirically derived risk factor thresholds for clinical interpretation.

Main Methods:

• A semi-parametric statistical method is proposed for threshold selection.
• Visualization techniques are incorporated to aid in threshold identification.
• The method accounts for outcome mixture distributions and adjusts for confounding covariates.

Main Results:

• The proposed method offers an alternative to arbitrary threshold selection.
• Empirically based thresholds can be identified, potentially offering more accurate risk factor effect estimation.
• Visualization aids in understanding the relationship between risk factors and outcomes.

Conclusions:

• The suggested semi-parametric approach improves threshold selection in variable dichotomization.
• This method can mitigate bias and potentially lead to more accurate health outcome effect estimation.
• Empirically derived thresholds may offer valuable clinical insights into risk factor impact.