Objectives: To design a frailty index (FI) and evaluate three methods to handle missing data. Furthermore, we evaluated its construct (i.e., skewed distribution, correlation with age and sub-maximum score) and criterion validity (based on mortality risk).
Study design: We included 11,539 participants (45± years) from a population-based cohort in the Netherlands. Frailty was measured with a FI, which we constructed based on the accumulation of 45 health-related variables, related to mood, cognition, functional status, diseases and conditions, biomarkers, and nutritional status. A total FI-score was calculated by averaging the scores of the deficits, resulting in a score between 0 and 1, with higher scores indicating increasing frailty. Mean imputation, single- and multiple imputation were applied.
Main outcome measure: Mortality data were obtained by notification from the municipal administration. Median follow-up time was 9.5 years, during which 3902 (34%) participants died.
Results: The median FI for the full population was 0.16 (IQR = 0.11–0.23). The distribution of the FI was slightly right-skewed, the absolute maximum score was 0.78 and there was a strong correlation with age (Pearson correlation = 0.52;95% CI = 0.51–0.54). The adjusted HR per unit increase in FI-score on mortality was 1.05 (95% CI = 1.05–1.06). Multiple imputation seemed to provide more robust results than mean imputation.
Conclusion: Based on our results we advise to the use of at least 30 deficits from different health domains to construct a FI if data are not imputed. Future research should use the continuous nature of the FI to monitor trajectories in frailty and find preventive strategies.