Strategies for handling missing data that improve Frailty Index estimation and predictive power: lessons from the...

Area of Science:

• Gerontology
• Biostatistics
• Epidemiology

Background:

• Missing data are common in aging research, potentially affecting health outcome assessments.
• The Frailty Index (FI) is a key measure in aging studies, but its accuracy can be compromised by missing data.

Purpose of the Study:

• To investigate the impact of real and simulated missing data on the Frailty Index (FI) and survival analysis.
• To evaluate the effectiveness of various imputation strategies for mitigating missing data biases in aging studies.

Main Methods:

• Combined NHANES 2003/2004 and 2005/2006 data (N=9307) to analyze missing data patterns.
• Simulated missing data to assess bias in FI and hazard rates (HR).
• Compared imputation methods: multivariate imputation by chained equations (MICE) using CART, CART with auxiliary variables (CART+Aux), and default MICE (PMM/logreg).

Main Results:

• Observed significant hazard rate differences between missing and present data blocks.
• Ignoring missing values introduced bias (0.0112 ± 0.0008) to the mean FI.
• CART+Aux imputation effectively corrected bias and improved FI predictive power and HR reliability.
• Default MICE models (PMM/logreg) exacerbated FI bias.

Conclusions:

• Handling of missing data critically impacts FI calibration as a mortality predictor.
• Ignoring missing FI values may suffice for rough clinical predictions of adverse outcomes.
• For cross-study/population comparisons, careful imputation (e.g., CART+Aux) is essential for reliable and precise statistical conclusions.