The GDS, assessed in 2008, has excellent psychometric characteristics within the age span of our population, has good validity as a continuous dimensional measure of depressive symptomology, and good sensitivity and specificity for clinical selleckbio depression when dichotomized (Sheikh and Yesavage 1986).

We therefore chose the GDS as our “anchor” assessment and scaled all other assessments against the GDS. We chose GDS-15 over CESD-10 because it was specifically developed for use in geriatric population (the mean age of NHS participants were both over Inhibitors,research,lifescience,medical 70 years old when either instrument was examined), it contained fewer somatic items. In cognitively intact patients Inhibitors,research,lifescience,medical older than 65 years, the GDS screen is the preferred instrument because the psychometric data on the CES-D are mixed in this population (Sharp and Lipsky 2002). Although the quality of the available measures used across waves differs, our approach down-weights those instruments that do not correspond well with the GDS. Our protocol

was as follows: using all NHS women with GDS scores (regardless of the availability of genetic data), we regressed the GDS score on all depression-related measures available in that wave, using a linear regression model. For example, using all measures of the depression phenotype available in 2004, we estimated the following linear regression: (1) On the basis of Inhibitors,research,lifescience,medical this linear regression, we predicted the value that the GDS score would have taken if it has been assessed in 2004. We estimated similar models for each interview wave, 1992–2006. For instruments with missing data on a few items, we used the average of nonmissing items if at least half of the items were Inhibitors,research,lifescience,medical reported and a missing indicator method for observations missing more than half of the items. In a second step, we used the regression coefficients from the initial Inhibitors,research,lifescience,medical models to predict the value of GDS for each participant at each wave, had he or she been given the GDS. In this way, all individual depression measures collected at one wave were rescaled and translated to a single common

scale (GDS-standardized score) for each participant, and these estimates could be obtained even for individuals who did not complete the GDS in 2008. The final phenotype was the average of the Cilengitide rescaled depression scores from all available questionnaire cycles (up to seven waves): (2) This approach maximized the available sample size and optimized the KPT-330 molecular weight information available on lifetime experiences of depression, because anyone with at least one wave of information with depression assessment was included. We also believe it decreased the transient component of the measure compared to using a single-wave assessment, which would strengthen our ability to detect genetic predictors. In fact, in our analytic sample 132 (1.9%) women had only one measure of depression, 136 (1.