|Home | About | Journals | Submit | Contact Us | Français|
There is a need for large scale epidemiological surveys to be a) faithful to diagnostic specifications, and b) constrain time and participant burden associated with each section of a possibly lengthy interview.
To examine whether one “gating” approach devised for recent large scale international psychiatric surveys results in a reduced number of identified cases of drug dependence, and/or biases in estimated associations with background characteristics.
Data from a recently released cross-sectional, nationally representative household survey, the United States National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) were analysed.
43,093 English speaking adults aged 18 years and over.
Dependence upon cocaine and other illegal drug dependence, defined in two ways: “ungated”, and “gated”. “Ungated” dependence included all persons meeting criteria for DSM-IV dependence, without regard for DSM-IV drug abuse clinical features. “Gated” dependence required at least one feature of DSM-IV drug abuse.
There was no statistically robust decrement in the estimated prevalence of cocaine or other drug dependence using a “gated” assessment. Patterns of association of cocaine dependence with background characteristics were not appreciably different when the gated and ungated approaches were applied.
In panoramic mental health surveys, the inefficiency of an ungated approach must be balanced against the anticipated number of cases of dependence without associated social role impairments or harm. In this study, the reduction in the number of identified cocaine dependence cases appeared to be so small that even in a sample of over 40,000 participants, attenuation in population prevalence would prove difficult to detect.
Large scale epidemiologic field surveys can provide important evidence about psychiatric disorders. They detect cases of psychiatric disorders not seen in clinical research; and cases with unusual features or more “borderline severity” may be detected, case ascertainment biases can be constrained, and the resulting estimates are more free of the influences of treatment provision (or the lack thereof) upon population estimates and the nature of “cases” (Berkson, 1946; Narrow et al., 2002; Robins and Guze, 1970; Robins, 2004).
The mandate for large scale national field surveys has expanded, such that they include the entire list of psychiatric disturbances within the current diagnostic and statistical manuals of our time, or at least a substantial subset. This panoramic perspective is constrained by a need to balance competing time demands within the interview. One way in which such demands can be juggled is to set up conditional skips based on earlier measurements during the assessment session – often termed “gating” of assessment. For drug dependence, the “gate” can be made based upon varying criteria (Hasin and Grant, 2004; Hasin et al., 2005; Narrow et al., 2002). These “gates” can be understood as a special case of mass screening in public health. A gated approach may be less sensitive than an ungated approach, whereas the ungated approach may be less specific and overly inclusive compared to the gated approach.
Those undertaking broad epidemiologic studies of psychiatric disorders must translate the prevailing psychiatric nomenclature into standardised survey assessments that may be administered by lay interviewers, in such a way as to accurately measure the constructs as presently defined (Blazer and Kaplan, 2000; Frances, 1998; Narrow et al., 2002; Pincus et al., 1998; Robins, 2004). Vigorous debate has occurred about how to juggle the time demands of panoramic psychiatric interviews with a need to remain faithful to current alcohol and other drug dependence nomenclature. For example, the issue of “gating” the assessment of drug dependence was raised in discussions in the United States about whether the Epidemiological Catchment Area (ECA) study and National Comorbidity Survey (NCS) might have overestimated the prevalence of mental disorders in the US because they did not “gate” according to a “clinical significance” criterion (Frances, 1998; Narrow et al., 2002; Regier et al., 1998; Spitzer, 1998). More recently, there has been concern about the extent and nature of the impact of a gated assessment of drug dependence upon population prevalence estimates and their correlates in several countries of the World Health Organization’s World Mental Health surveys (Grant et al., 2007; Kessler and Merikangas, 2007).
We have previously examined the impact of a “gated” assessment approach upon alcohol and cannabis dependence prevalence estimates in the US, and whether the effect of this “gating” might vary across key background variables (Degenhardt et al., 2007a; Degenhardt et al., 2007b). We found a very modest attenuation of the prevalence of past year cannabis dependence (0.26% compared to 0.32%), but no statistically robust variation in the prevalence estimate for past year cannabis use disorders (DSM-IV abuse and/or dependence) (Degenhardt et al., 2007b). This reduction is difficult to examine in the scale of most population surveys. For 12 month alcohol dependence, the gated approach produced an estimate that was not in the confidence bounds of the ungated approach; gating missed an estimated 129 cases per 10,000 participants (2.5% vs. 3.8%) (Degenhardt et al., 2007a). The relationships with demographic variables of interest remained remarkably consistent across the gated and ungated assessment approaches, however, suggesting that any attenuation of estimated prevalence was not strongly concentrated within certain subpopulations (such as age or sex groups) (Degenhardt et al., 2007a; Degenhardt et al., 2007b).
The empirical findings for other drug dependence may not be the same. Cocaine is the second most common form of illegal drug use in the United States, and its use has, at times, been highly prevalent according to certain population characteristics such as race-ethnicity (Golub and Johnson, 1994; Johnson et al., 2000; Miech et al., 2005). Alcohol and cannabis, in contrast, are less concentrated within subgroups groups (such as those of a particular age, sex, socioeconomic status or race-ethnicity). The lack of impact of a gated assessment upon associations of dependence with age, sex and other characteristics for alcohol and cannabis may therefore not the true for cocaine and other less commonly used drugs. For these reasons it is prudent to conduct a separate examination of the potential impact of a “gated” approach with respect to cocaine and other drug dependence.
The current study aimed to:
This study is based upon US data from NESARC, a population based face-to-face survey of 43,093 participants aged 18 years and older. The sample was recruited from a prior US Census Supplementary Survey sample (C2SS) that had been interviewed in 2000–2001. Participant selection and interviewing were conducted by US Census Bureau interviewers (Stetser et al., 2002). Recruitment for NESARC took place between August 2001 and April 2002, between 5–18 months after the prior C2SS contact (Stetser et al., 2002). The response rate for eligible housing units in the NESARC has been reported as 81%. We have previously summarized the details of the sampling methodology (Degenhardt et al., 2007a; Degenhardt et al., 2007b) (see also (Stetser et al., 2002)); further details on the interview, method and characteristics of the NESARC sample can be found elsewhere (Grant et al., 2001; Grant et al., 2003; Grant et al., 2004).
The main response variable under study was case status with respect to recently active drug dependence defined by DSM-IV-TR criteria, with both drug use and a clustering of at least three manifestations of dependence within the 12 months prior to assessment. The primary focus in this study was upon cocaine, but we have also provided the estimates obtained using “ungated” and “gated” assessment approaches for the prevalence of 12 month dependence upon the following drugs: cannabis (for reference; further details published in (Degenhardt et al., 2007b), opioids (excluding heroin), stimulants (excluding cocaine), hallucinogens, heroin, inhalants, sedatives, and anxiolytics.
NESARC assessments were via computer-assisted personal interviews (CAPI) with the “ungated” approach. We simulated the “gated” approach by recoding drug dependence cases as non-cases if the participant did not also report at least one symptom of DSM-IV “drug abuse”. Sociodemographic characteristics were also assessed via CAPI standardized survey questions. In the NESARC, all participants were asked if they had ever used the drugs concerned; those who had were asked if they had used those drugs in the past 12 months, and for those who had used such drugs in the past 12 months, DSM-IV drug dependence and abuse for the past 12 months. For this paper, the following “dependence” categories were generated:
Weighted prevalence estimates and their 95% confidence intervals were derived using Taylor series linearisation with SUDAAN Version 9.0, which accommodates field survey samples with self-representing PSUs and accounts for other features of the complex survey sampling design (Research Triangle Institute, 2005). Prevalence estimates (and their 95% confidence intervals) were made according to sex, age group (18–24 years, 25–34 years, 35–44 years and 45 years and older), and race-ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, and other); and annual personal income (<USD$20,000; USD$20,000–34,999; USD$35,000 and above). Note that some group categories had to be collapsed due to the very small raw sample sizes for older age groups and some race-ethnicity sub-groups. Multiple logistic regressions, which included all background variables in the models, were run using SUDAAN 9.0, to derive estimates of strength of association (odds ratios). The agreement of estimates produced using gated vs. ungated assessments was compared statistically using a simple Kappa coefficient.
Table 1 shows the prevalence and 95% confidence intervals of cocaine dependence according to “ungated” and “gated” assessment approaches. There was no statistically robust difference in the estimated prevalence of dependence using the two approaches: 0.13% vs. 0.12%, with very broadly overlapping confidence intervals.
The direction and strength of association with background variables was remarkably similar for the “gated” and “ungated” assessment approaches (Table 1). In other words, the gated approach did not distort estimated relationships with background variables according to whether cocaine-related social role impairments or other abuse criteria were required.
Figure 1 shows the prevalence estimates for other drug dependence using gated and ungated assessment approaches. No reliable prevalence estimates could be obtained for heroin or inhalant dependence due to very small raw numbers meeting diagnostic criteria (n = 4 and 2, respectively, with no impact of gating of assessment). As can be seen, estimates for each drug dependence category were very similar (95% confidence intervals overlapped in all cases) regardless of whether an ungated or gated assessment approach was simulated.
The Kappa coefficients produced for each pair of drug dependence estimates are also presented in Table 2. As can be seen, there was good agreement between the gated and ungated assessment methods for all drugs examined; the lowest kappa was for sedative dependence but the upper limit of the 95% confidence interval for the kappa value was in the range commonly accepted for good agreement (0.53; 0.33, 0.72).
In this contrast of ungated versus gated methods of assessing cocaine dependence (whereby participants are assessed for dependence only if they have exhibited “clinically significant” manifestations of maladaptive use of drugs, or drug-related social role impairments) there appears to be no appreciable reduction of population prevalence estimates. The estimated prevalence of dependence upon other drug types was even smaller than that for cocaine, and there was insufficient power to detect any difference at a general population level, even with this sample of over 40,000 persons.
The relationship of cocaine dependence with background characteristics did not differ appreciably when a gated assessment of cocaine dependence was simulated compared to the ungated assessment method. With the other drugs examined here, the number of cases of drug dependence was too small to warrant detailed probing into the background associations.
There has been considerable discussion of the meaning, utility and importance of the concept of “clinical significance” within the current DSM-IV classification system (Beals et al., 2004; Blazer and Kaplan, 2000; Frances, 1998; Narrow et al., 2002). There has also been debate about whether data from some of the surveys conducted as part of the World Mental Health Surveys may have resulted in attenuated or biased estimates of prevalence and associations of drug dependence because they used a gating criterion as was used in this study (Grant et al., 2007; Kessler and Merikangas, 2007). This study is unlikely to resolve what appears to be significant disagreement about how and why to use different assessment approaches in large scale survey research on the epidemiology of psychiatric disorders, but it certainly appears to be the case that the prevalence estimates of dependence for most drug types will not be appreciably attenuated when a gated assessment approach is used. Moreover, a gated approach may be preferable when the focus of enquiry requires balancing the competing time demands for assessment of a large number of psychiatric disorders. There is not the need for such balancing, of course, when surveys are more highly focused upon drug dependence specifically. Further, panoramic studies may also be used to inform further development of the classification system; for this purpose, gating should often be minimised.
The NESARC data upon which this study was based are subject to limitations. It is possible that the NESARC underestimated illegal drug use and DSM-IV use disorders. This has been suggested in a recent comparison of NESARC data with that from the US National Survey on Drug Use and Health (NSDUH) (Grucza et al., 2007). This could have been related to the way in which the study was conducted – the NESARC was conducted face to face, compared to a self-complete method as used in the NSDUH (Grucza et al., 2007). Previous research has been consistent with the possibility that concerns about anonymity result in lower levels of disclosure of drug use (Aquilino and LoSciouto, 1990).
A further possibility might be related to the NESARC recruitment and sample identification method: the sample was recruited from a prior US Census Supplementary Survey sample (C2SS) that had been interviewed in 2000–2001, and participant and interviewing were conducted by the US Census Bureau (Stetser et al., 2002). It is possible that participants in the NESARC had greater concerns for privacy and anonymity, given that the interviews were conducted by government employees and identifying details from the Census interviews had been used to target them for the NESARC in the first place.
When time, cost, participant burden and loss of attention are immaterial, a gated approach may not be useful in surveys of drug dependence. However, when such limits do exist, a gated approach to assessment of drug dependence may serve well, according to the findings of the present empirical study. Here, the gated approach did not result in a robust reduction in the population prevalence estimates for dependence, and there appeared to be no differential distribution across the background variables examined here. It would appear that most cocaine dependent individuals will also have features of cocaine abuse. In a general population survey even of the size of the NESARC (over 40,000 individuals), there was insufficient power to detect and/or investigate the nature of the relatively fewer cases who do not. This was even more markedly the case for dependence upon other illegal drugs, where numbers were even smaller. Similar work examining this issue in other countries would provide important additional information on the extent to which gating affects dependence prevalence estimates in other settings.
The work was supported by NIH/NIDA research project award R01DA016558 and a NIDA KO5 Senior Scientist award (K05DA015799) to the last author (JCA), as well as funds from Michigan State University and the Australian Government Department of Health and Ageing. KM’s apprenticeship is supported under an NIH/NIDA institutional training program award, T32DA021129. Data reported herein come from national survey data collected under the auspices of the US National Institute on Alcohol Abuse and Alcoholism under the direction of Dr. Bridget Grant and colleagues. We are grateful for the public use dataset created by the NIAAA research.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.