Purpose
To develop bridging algorithms to score the Veterans Rand-12 (VR-12) scales for comparability to those of the SF-36® for facilitating multi-cohort studies using data from the National Cancer Institute Surveillance, Epidemiology, and End Results Program (SEER) linked to Medicare Health Outcomes Survey (MHOS), and to provide a model for minimizing non-statistical error in pooled analyses stemming from changes to survey instruments over time.
Methods
Observational study of MHOS cohorts 1–12 (1998–2011). We modeled 2-year follow-up SF-36 scale scores from cohorts 1–6 based on baseline SF-36 scores, age, and gender, yielding 100 clusters using Classification and Regression Trees. Within each cluster, we averaged follow-up SF-36 scores. Using the same cluster specifications, expected follow-up SF-36 scores, based on cohorts 1–6, were computed for cohorts 7–8 (where the VR-12 was the follow-up survey). We created a new criterion validity measure, termed “extensibility,” calculated from the square root of the mean square difference between expected SF-36 scale averages and observed VR-12 item score from cohorts 7–8, weighted by cluster size. VR-12 items were rescored to minimize this quantity.
Results
Extensibility of rescored VR-12 items and scales was considerably improved from the “simple” scoring method for comparability to the SF-36 scales.
Conclusions
The algorithms are appropriate across a wide range of potential subsamples within the MHOS and provide robust application for future studies that span the SF-36 and VR-12 eras. It is possible that these surveys in a different setting outside the MHOS, especially in younger age groups, could produce somewhat different results.
