74. USING FAMILY MODELS TO ESTIMATE THE CORRELATION OF LIABILITY AND AGE AT ONSET FOR PSYCHIATRIC DISORDERS

Background Psychiatric disorders vary substantially in their age at onset (AAO) and for many disorders this variability relates to prognosis, presentation, and etiological heterogeneity. Variable AAO also complicates case-control analyses of heritability and risk factors because controls may subsequently convert to cases. This censoring can be accounted for in, e.g., cox-PH models, but these approaches include implicit assumptions about the coupling between AAO and liability. We are interested in estimating, empirically, the coupling between disease liability and AAO in a flexible framework where their relationship is not simply assumed, and where liability and AAO may be modeled as functions of the same or different covariates. Methods To examine the coupling between liability and AAO, we use an updated structural equation model developed for analysis of AAO and risk to relatives (Neale et al 1989, AmJ HumGen). The unit of analysis is a pair of relatives, each of whom has a latent disorder liability and a latent AAO. We present analyses of over 239k full-sibling pairs identified in the Danish national registers, excluding twins, with both sibs born in the 1990s. For each individual we gather registered diagnoses for eleven psychiatric traits including age at first diagnosis (used as AAO) and age at last contact. The relationship between individuals’ AAO and their relatives’ liabilities is used to estimate the correlation between liability and AAO within an individual. Models are specified in R and parameters are estimated by maximum likelihood. Results We obtain estimated couplings between liability and AAO for five traits. For MDD the correlation was r = -0.85 (SE 0.036), corresponding to beta = -5.7 (SE 0.53) years to AAO per unit liability. For ANX: r = -0.88 (0.018), beta = -6.9 (0.32); for BPD: r = -0.83 (0.070), beta = -5.4 (0.99); for AUD: r = -0.84 (0.008), beta = -4.5 (0.096); for DUD: r = -0.92 (0.006), beta = -6.5 (0.12). For SZ, OCD, BUL, ANO, ADHD and ASD the model did not stably converge to give parameter estimates due to, we believe, real data being misaligned to the model assumptions and there being low numbers of case-case concordant sibships for rare traits. Further method development is needed to improve the stability of the model to, e.g., prevalence time trends, covariates, and very rare outcomes. Discussion We have extended a framework to estimate the coupling of disorder liability and AAO, formally, using family data. We found nearly universal negative correlation, suggesting higher liability with earlier onset, but in no cases was this correlation perfect. This suggests that factors other than the disorder liability contribute to AAO.