Descriptive analyses were conducted to characterize the sample demographically and clinically, and we report data on psychiatric disorders with ≥ 5% prevalence in the sample (Table 1). Based on an examination of the distributions of the cocaine-related and psychosocial impairment variables, all variables were approximately normally distributed. The primary study analyses were generalized estimating equation (GEE) linear and logistic regression analyses conducted to test the effect of psychiatric and substance dependence diagnoses and combined psychiatric comorbidity on cocaine-related impairment and service utilization. GEE is widely used as a method of dealing with correlated data in fitting the generalized linear model (Liang and Zeger, 1986). We used GEE to fit the model to account for the potential dependence among individuals within the same nuclear family. Analyses were conducted on a multivariate basis adjusting for proband versus sibling status, for demographics (i.e., gender, age, race/ethnicity, marital status, education, employment), for a history of other substance dependence diagnoses and multiple other substance dependence diagnoses, and for a history of each psychiatric diagnosis or multiple psychiatric diagnoses.
