Abstract:
|
BACKGROUND: Longitudinal studies of physical health, measured by the SF-36 physical component scale (PCS), are problematic if some people die, since the PCS has no code for death. Diehr et al. proposed a transformation of the PCS to the PCD, the probability of being healthy in one year conditional on the observed PCS; deaths have a PCD value of zero (Medical care 2001; 39:670-680). METHODS: We examined the longitudinal and cross-sectional performance of the PCD in the original population of veterans and in a new, very different population. FINDINGS: The PCD permitted retaining the deaths, and the dead were not outliers. Despite the unusual distribution of the PCD, the residuals were nearly normal after adjusting for baseline health problems. Results were sensitive to other missing data, which should probably be imputed, to avoid the deaths having too much influence. The transformations developed in the original population performed almost perfectly in the second population. CONCLUSION: The PCD seems a good way to account for death in studies using the PCS. The published transformation coefficients may be appropriate in other populations.
|