Adapting the Vulnerable Elders Survey-13 to Predict Mortality Using Responses to the Medicare Health Outcomes Survey.

Objectives: To use items from the Medicare Health Outcomes Survey (HOS) to adapt or validate a simple method for identifying community-dwelling older adults at greater risk of death and to extend the method to identify a very high-risk group.

Design: Analysis of longitudinal data.

Setting: National sample of beneficiaries from Medicare Advantage plans with 500 or more enrollees.

Participants: Medicare beneficiaries aged 65 and older responding to 2009 baseline and 2011 follow-up HOS (N = 238,687).

Measurements: Bivariate and multivariate analyses of the HOS; adaptation and validation of a previously validated Vulnerable Elders Survey-13 (VES-13) scoring system that uses age and self-reported function to predict mortality.

Results: A modified predictive model, that uses substitutes for several items in the previously validated VES-13, predicted 2-year mortality; 10.6% of those scoring 3 or more, and 2.4% of those scoring less than 3 died within 2 years (relative risk of death 4.4, similar to 4.2 for the original VES-13 sample), and 15.5% of those scoring 7 or more died within 2 years (relative risk of death (relative to scores <3) of 6.5). Sixteen percent of HOS beneficiaries were missing some data; 2-year mortality for those with missing items was 9.5%, versus 7.1% for those with no missing items (P < .001). Imputation of median values for missing VES-13 items results in valid predictions of mortality for those with partially missing data.

Conclusion: The VES-13 algorithm is robust to substitution of functional items and can be used to identify very high-risk older adults. Multiple imputation of missing items reduces loss-to-follow-up bias and increases sample size.