Description
- Abstract:
- The Dietary Screening Tool (DST) is a scoring system used to characterize nutritional risk in older adults. The DST evaluates intake of different food groups using a set of survey items. Based on a participant's responses to these items, a score is calculated to determine their nutritional risk. The DST has been used in numerous studies to examine the connections between nutrition and health outcomes. In some studies, participants are missing DST data. Commonly used statistical methods to address missing DST data include complete case analysis and single imputation methods. Complete case analysis may produce biased results when missingness is related to other variables, and single imputation methods underestimate the standard errors of statistics that rely on DST scores. We compare the operating characteristics of different methods that adjust for missing DST information using data from the Deliver-EE clinical trial. This study investigates the impacts of daily home-delivered meals compared to frozen, shipped, meals for older, homebound adults. The Deliver-EE trial suffers from missing DST data because of participants’ non-response and because the possible responses to several DST items were altered midway through the trial. We compare complete case analysis, mean and mode imputation, and multiple imputation with random forest and with predictive mean matching. Based on the Deliver-EE data, we simulate datasets reflecting observations that are missing under different mechanisms, and we compare estimates of the relationship between DST score and other surveyed variables. We evaluate imputation quality using percent bias, coverage rate, and width of confidence intervals. The point estimates were similar across all single and multiple imputation methods, but complete-case analysis resulted in biased estimates. Standard errors were larger when using complete case analysis compared to single and multiple imputation methods. Single imputation methods had smaller standard errors than the multiple imputation methods, but with smaller than nominal coverage. The simulation results showed that the multiple imputation methods had the smallest percent bias and nominal coverage. Studies using the DST should consider using multiple imputation to adjust for missing DST data.
- Notes:
- Thesis (Sc. M.)--Brown University, 2025
Citation
Yost, Rachel,
"Comparing Missing Data Methods for the Dietary Screening Tool"
(2025).
Biostatistics Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://repository.library.brown.edu/studio/item/bdr:2q3geuw7/
Relations
Collection:
-
Biostatistics Theses and Dissertations
Theses and Dissertations for the Biostatistics department....