Starting in May 2025, the WRAP data freeze LIBRA and LIBRA2 tables will include imputed values and columns for the LIBRA/LIBRA2 score with diet (previously excluded from LIBRA calculations), when available.

Out of the 11 modifiable lifestyle factors in LIBRA and 14 in LIBRA2, imputing will only occur when fewer than half (LIBRA: 6, LIBRA2: 7) of the factors are missing, additionally only when current age and genders of "male" or "female" are reported. Indicator variables and an overall count will be included to help researchers identify these imputed values.

Each subsequent data freeze will re-impute missing values, in order to allow for the donor pool (participants with full non-imputed LIBRA/LIBRA2) to update with new visits that are representative of our growing cohort. 

Due to a large proportion (>50%) of visits with missing diet information, this protective factor will not be imputed, but a LIBRA_Score_with_Diet variable will be included only for those with this information.

Imputation Method:

To evaluate the proper imputation method, we generated random pseudo-missingness within a dataset comprised of all full LIBRA visits and compared the original LIBRA score to the imputed LIBRA score from multiple methods (hot deck, last observation carried forward, mean of the previous two visits, mean of the previous three visits, mean of all visits).

We chose the hot deck imputation method after comparing bias estimates, correlation of LIBRA scores, and completeness across all methods. Our hot deck imputation included categorical variables of age (<55, 55-59, 60-64, 70-74, 75-79, 80+) and gender (male, female). This resulted in a dataset with low bias (est = -0.023), strong correlation (r = 0.87), and no missingness. 

We have included visits not eligible for imputation in the LIBRA and LIBRA2 tables. These visits will have N_Imputed = 0 and do not have calculated total scores.

References:

Andridge RR, Little RJ. A Review of Hot Deck Imputation for Survey Non-response. Int Stat Rev. 2010;78(1):40-64. doi:10.1111/j.1751-5823.2010.00103.x. PMID: 21743766; PMCID: PMC3130338.

Document History