Robust Mortality Prediction Approach for Identifying Vulnerable Seniors Using the Medicare Health Outcomes Survey

Using 2009 data from 238,687 seniors, we modified a validated predictive model for identifying high mortality risk seniors [Vulnerable Elders Survey (VES-13)] using items from the Medicare Health Outcomes Survey (HOS). We assessed whether the approach was robust to substitution of similar survey items and could be used in the presence of missing data.  We computed a risk score (using age, self-rated health, physical functioning limitations, and activities of daily living) at baseline using a modified predictive model that substituted new items for several unavailable items used in previously validated versions of the scoring algorithm. We predicted two-year mortality from risk score and other beneficiary characteristics using multivariate models.  We classified 38% of seniors low risk (<3 points), 29% moderate risk (3-6), and 33% high risk (>6).  Each one-point increase in the 0–10 risk score corresponded to 37% higher odds of death (p<0.001). 10.6% of moderate/high-risk seniors died within two years, compared to 2.4% of low-risk seniors, similar to relative risks reported using the original VES-13.  The 16% of HOS respondents with any missing data on the modified VES-13 instrument had a 9.5% mortality rate versus 7% for those without. Predicted mortality under median imputation matched that for cases with no missing data. This modification of the VES-13 that substitutes alternative survey questions performs as well as the original algorithm in predicting older adult mortality, accounts for missing data (a significant concern in surveys of older patients), and is robust to changes in specific survey items.