Cocaine is the second most prevalent illicit drug used in the United States (SAMHSA, 2008
). This study demonstrates a novel application of secondary data analysis of CTN public-use data (National Institute on Drug Abuse, 2009
). It contributes new and important information on the assessment and the predictor analysis of dependence of patients seeking treatment for cocaine related problems by examining a large and geographically diverse sample of cocaine users recruited from 14 addiction treatment programs, investigating measurement bias (DIF) in assessing for cocaine dependence, applying a latent variable regression model to take into account measurement errors in the predictor analysis of cocaine dependence, and exploring the effects of DIF on diagnostic criteria. The latter has been an understudied and overlooked area of research for drug dependence, but has implications for improving the validity of analyzing self-reported diagnostic data.
MIMIC modeling revealed that (1) female cocaine users exhibited a higher likelihood of cocaine dependence than male cocaine users even after controlling for the potential confounding influences of age groups, race/ethnicity, years of cocaine use, addiction treatment history, comorbid drug dependence diagnoses, and treatment setting; (2) there was no evidence of gender-related item-response bias in the assessment of cocaine dependence; and (3) there was race/ethnicity-related DIF on the endorsement of cocaine dependence criteria. These findings have important implications not only for improving the analysis of self-reported data but also for the assessment of drug dependence, including the emerging DSM-V.
Traditionally, drug dependence is examined as a binary outcome. Such a categorical approach tends to have less statistical power and fails to consider potential measurement errors resulting from self-reported data of multiple diagnostic items. In fact, the dependence syndrome
was originally conceptualized as a continuous, dimensional construct (Edwards & Gross, 1976
). Consistent with prior studies of less geographically diverse or smaller samples (Bryant et al., 1991
; Feingold & Rounsaville, 1995
; Morgenstern et al., 1994
), the results of factor analysis support our analysis of cocaine dependence as a single dimensional construct in the MIMIC model.
Additionally, MIMIC modeling of the criteria for cocaine dependence extends earlier research by utilizing response patterns from all dependence items and providing a flexible regression framework to statistically control for potential measurement errors from self-reports. Thus, results from the MIMIC model are less biased because the construct of the outcome of interest can be more properly modeled, and potential item-response biases from self-reports can be incorporated into the analysis.
Recently, results from the 2001–2002 National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) suggest that women, Hispanics, and Blacks are more likely than men and Whites to report substance dependence but without abuse symptoms (Hasin et al., 2005
; Hasin & Grant, 2004
). The result is somewhat inconsistent with the DSM-IV’s specification because dependence is generally considered a more severe pattern of substance use than abuse (APA, 2000
), and little is known about the reasons that may explain this finding. We utilized MIMIC procedures to evaluate whether there were potential item-response biases by gender and race/ethnicity in the assessment of drug dependence. Consistent with our recent research on alcohol and marijuana dependences (Wu et al., 2009b
), we did not find gender-related DIF.
However, our analyses suggested the presence of DIF by cocaine users’ racial/ethnic backgrounds. A recent analysis of alcohol dependence data from NESARC also found evidence of DIF by race/ethnicity (Carle, 2009
). Holding constant the cocaine dependence level, Hispanics were more likely than Whites to endorse Tolerance
, but less likely to endorse Continued use despite resulting problems
; Blacks were more likely than Whites to report Inability to cut down
. Thus, Hispanics have a lower threshold than Whites to report Tolerance
, but a higher threshold to report Continued use despite resulting problems
; Blacks have a lower threshold than Whites to report Inability to cut down
. This finding suggests the need to control for race/ethnicity statistically in the analysis of cocaine dependence, and to investigate reasons accounting for measurement invariance (e.g., racial/ethnic variations in the meaning or interpretation of diagnostic items) so as to reduce its potential confounding effects on assessing for cocaine dependence. Further, accurate estimates of cocaine dependence rely on the use of a sound measure. Study findings are, therefore, consistent with recommendations that cross-cultural research is needed for DSM-V to ensure that diagnostic criteria are equivalent or valid for different racial/ethnic groups (Escobar & Vega, 2006
; Room, 2006
Further, in addition to women, we found that young adults had increased odds of exhibiting a high level of cocaine dependence. Years of cocaine use, episodes of prior drug abuse treatment, as well as dependence on alcohol, marijuana, and opioids were all positively associated with cocaine dependence. These results are in line with prior findings suggesting that abusing one drug increases the likelihood of abusing others, which could be influenced through a shared vulnerability (Tsuang et al., 1998
), greater opportunities for substance use (Wilcox et al., 2002
), or direct effects from substance use (APA, 2000
). For example, cocaine and opioids are often used together to enhance their subjective reinforcing effects (Leri et al., 2003
), and alcohol is reportedly used by cocaine users to moderate the discomfort following cocaine use (Magura & Rosenblum, 2000
Given that dependence on other substances was positively associated with the level of cocaine dependence in this study, women’s greater odds for exhibiting cocaine dependence is a cause of concern. Recently, we utilized latent class analysis to examine cocaine users in treatment, and found that women were more likely than men to be classified in the most severe subgroup that was characterized by reporting five or more dependence symptoms (Wu et al., 2009d
). Another study on marijuana dependence among clinical patients also found greater odds of having it among women as compared with men (Wu et al., 2009b
). Taken together, this and other studies suggest that treatment-seeking cocaine-using women may be disproportionately more severe in their substance use patterns and related problems than treatment-seeking men (Denier et al., 1991
; Hernandez-Avila et al., 2004
; Weiss et al., 1997
; Wu et al., 2009d
Lastly, amphetamine dependence in this study was associated with decreased odds of cocaine dependence, suggesting that dependence on both cocaine and amphetamines is infrequent. Our recent study of cocaine dependence among treat-seeking stimulant users also found a low rate (7%) of co-use of cocaine and amphetamines within a 12-month period among stimulant users (Wu et al., 2009d
). It has been suggested that infrequent co-use of cocaine and amphetamines among regular stimulant users may be related to unique differences in demographic characteristics of users and pharmacological effects (see Wu et al., 2009d
). For example, cocaine users were older and primarily Black, and amphetamine users were younger and predominantly White (Wu et al., 2009d
). Differences between cocaine and amphetamines in their duration of effects, accessibility, and distribution networks may explain unique racial variations in the use of cocaine and amphetamines (Sexton et al., 2005
4.1. Study limitations and strengths
These findings should be interpreted within the context of the following limitations. First, these findings are based on treatment-seeking drug users who participated in CTN trials. The study sample is not necessarily representative of all outpatient cocaine-using patients. Another limitation is reliance on participants’ self-reported information, which is subject to recall errors, exaggeration, or underreporting. Third, cocaine abuse was not examined due to the fact that dependence was assessed first, and abuse was not assessed among those who met the criteria for dependence. According to the DSM-IV (APA, 2000
), abuse is not recognized among those with a dependence diagnosis. In this sample, 80% of cocaine users were classified as having cocaine dependence.
The present study also has several strengths. It extends prior studies of correlates of cocaine dependence by investigating a large, geographically diverse sample of users and employs MIMIC methods to understand item-level measurement errors in the risk factor analysis for cocaine dependence. Generalizability to clinical patients is improved because study participants were recruited from 14 major outpatient programs across the nation. These findings also demonstrate a novel use of CTN data to shed light on the quality of a drug dependence measure, the latent construct of cocaine dependence criteria, and predictors of cocaine dependence while simultaneously incorporating measurement errors into the analysis. Thus, this approach not only enhances our understanding of the drug-using population, but also provides important empirical information for future research on the quality and validity of diagnostic assessments for major diagnostic systems. Considering that substance use diagnoses are determined almost exclusively by self-reports and that the quality of self-report measures affects the validity of research and clinical findings, the need for the use of latent variable approaches to take into account measurement errors from self-reports is clear (Wu et al., 2009b
Within the context of multiple community-based addiction treatment settings, this study demonstrates that women and young adults are more likely than men and older adults to exhibit cocaine dependence according to DSM-IV criteria, and that race/ethnicity is not associated with cocaine dependence. These findings suggest that addiction treatment research needs to evaluate further gender- and age-related differences at treatment entry, and investigate how these differences may affect study participation, retention, and treatment response in order to better serve these populations.