The Brief Alcohol Use Disorder Severity Scale: An Initial Validation Evaluation

Abstract

Aims

The goal of this study was to develop a standard measure of AUD severity that includes multiple dimensions and can be used in clinical settings to inform treatment selection.

Methods

A large sample (n = 1939) of moderate to heavy drinkers was amassed from six psychopharmacology studies. The severity factor was comprised of four dimensions: withdrawal, craving, AUD symptoms and alcohol-related consequences. First, a confirmatory factor analysis (CFA) was conducted to examine model fit. Next, a comprehensive item list from the four measures (i.e. CIWA, DrinC, PACs and SCID-5 AUD criteria) was reduced through exploratory factor analysis (EFA). Once the final items were merged into a preliminary assessment, an EFA was run to observe the factor structure. Initial validation of the measure was obtained via associations with clinical endpoints.

Results

The chi-square test statistic () for a single-factor model of severity demonstrated good fit. Additional goodness-of-fit indices from the CFA revealed similar support for the single-factor model of severity (i.e. SRMSR = 0.011; RMSEA = 0.011; CFI = 0.999). Next, nine items from the individual EFAs were selected based on factor loading. The final EFA conducted on the 9-item scale demonstrated that a single factor model of severity best fit the data. Analysis of the psychometric properties revealed good internal consistency (= 0.79).

Conclusions

The current study extends upon the measurement of severity and supports a brief severity measure. This brief 9-item scale can be leveraged in future studies as a screening instrument and as a tool for personalized medicine.

Short Summary: The goal of this study was to develop a standard measure of AUD severity. Analyses on the 9-item scale demonstrated good model fit and good internal consistency. This brief 9-item scale can be leveraged in future studies as a screening instrument and as a tool for personalized medicine.

INTRODUCTION

While alcohol use disorder (AUD) remains a pervasive and frequent problem in treatment settings, the effectiveness of current pharmaceutical treatments is limited. To improve the effectiveness of treatment methods available for AUD, efforts have focused on expanding the repertoire of medications while also identifying clinical characteristics of target responders to maximize uptake. One such characteristic is AUD severity, which has been shown to influence treatment response and predict other psychological disorders (Kampman et al., 2007; Boschloo et al., 2012; Falk et al., 2015; Donato et al., 2021). Severity of AUD was also found to predict the placebo response (Scherrer et al., 2021). Further, severity has been implicated as a clinically meaningful outcome in treatment studies (Kiluk et al., 2018). However, it is notable that there is little convergence among researchers or clinicians on how to operationalize severity of AUD. Within research, severity is often defined as extent of alcohol-related consequences (i.e. social, legal, physical and psychological problems) (Babor et al., 2001), levels of consumption and frequency of use (McLellan et al., 1992; Hartwell et al., 2021), and number of AUD criteria endorsed in the Diagnostic and Statistical Manual of Mental Disorders (DSM). In clinical settings severity may be conceptualized as a mix of these factors, as well as resistance to treatment, length of problematic use, presence of psychiatric comorbidities and alcohol-related medical conditions (e.g. alcohol-associated liver disease), frequency of relapse, or intensity of individual symptoms (e.g., craving, withdrawal). The lack of consensus among definitions of AUD severity can limit the translational utility of research findings by restricting implementation in clinical settings. Therefore, a universal measure of AUD severity is needed to aide clinicians in assessment of AUD, inform treatment plans and serve as a measurable clinical outcome.

AUD severity is commonly assessed by a numerical count of the DSM-5's AUD criteria. Specifically, severity is evaluated by the number of AUD criteria endorsed out of 11, with 2–3 categorized as mild, 4–5 as moderate and 6 or more as severe. While the DSM-5 AUD criteria have been shown to be reliable and valid in assessing AUD, the tri-categorized scale of measurement is purely additive, with each symptom being given equal weight in determining the presence or severity of a disorder. Researchers and clinicians alike have long acknowledged that a dimensional measure of alcohol severity would offer more information than a categorical diagnostic measure. The theory that alcohol problems compose a continuum of severity is consistently demonstrated by both latent class models and latent trait models of alcohol problems (Krueger et al., 2005). However, evidence for the added benefit of a continuous measure of severity has been mixed in the literature. Dawson et al. (2010) were one of the first to use item response theory (IRT) models to empirically test the superiority of continuous, weighted criterion measure against a simple criteria count. Their results showed little benefit to the IRT method, implying that symptom count is highly correlated with severity problems to an extent that IRT-based latent severity estimates are redundant. However, these results have been challenged for failing to consider specific diagnostic combinations (Lane and Sher, 2015). The specific configuration of diagnostic criteria endorsed has been shown to influence severity estimates, as well as comorbid conditions and treatment selection (Moss et al., 2008; Cooper and Balsis, 2009; Lane and Sher, 2015). It is not to say that the current method of assessment (i.e. criteria count) is invalid, but rather that better strides can be made towards creating a severity measure that is both valid and easier to administer in clinical settings.

The measure advanced in this study was informed by a latent factor model of severity, which included multiple dimensions of impairment, including withdrawal, craving, alcohol-related consequences and AUD symptoms (Moallem et al., 2013). The severity factor was initially validated in relation to alcohol use, mood symptomatology and motivation to change drinking behavior in a community sample of individuals with AUD. The severity factor score was recently tested in a reanalysis of a large scale, multisite clinical trial of varenicline and was a significant moderator of medication response (Donato et al., 2021). Additionally, the predictive utility of the severity factor was shown to be above and beyond that of the DSM criteria count measure (Donato et al., 2021). These findings demonstrated the clinical benefit of a multi-dimensional AUD severity construct in predicting treatment responders. The current study seeks to translate this construct into an abbreviated measure with clinical utility. While other measures exist for severity (e.g. ADS, ASI, SADQ), the proposed measure is novel in that its creation is focused on ease of implementation while maintaining a rigorous empirical and conceptual foundation.

In summary, the goal of this study is to develop a brief measure of AUD severity that can be used in clinical settings to assess patient severity and inform treatment selection. The measure will be derived directly from the original severity construct (Moallem et al., 2013; Donato et al., 2021), while doing so in an abbreviated fashion that is readily accessible to clinicians in healthcare settings. This study will be carried out in three parts: (a) a confirmatory factor analysis (CFA) will be performed to replicate and further validate the AUD severity construct in a large sample of moderate to heavy drinkers, (b) the severity construct is refined into an abbreviated form by identifying and combining the items with the top loadings (i.e. those with the highest factor loadings) for each of the four measures and (c) an exploratory factor analysis (EFA) is performed to examine the factor structure of the final 9-item measure and initial tests of reliability and validity are carried out. These steps seek to combine theoretical applications with a data-driven approach to creating the measure, while also prioritizing ease of administration/implementation.

METHODS

Data source and sample composition

The sample for this study (n = 1939) was amassed from six separate clinical and experimental psychopharmacology studies conducted in the Addictions Laboratory at the University of California, Los Angeles. All studies shared similar inclusion criteria and recruitment methods across different research objectives (e.g. examining alcohol self-administration, acute subjective response to alcohol and pharmacotherapy trials). It is important to note that all data used for this study were collected at the initial screening visit prior to any experimental procedures or pharmacological manipulations.

Participants were recruited from the greater Los Angeles area and were comprised for both treatment and non-treatment seeking moderate to heavy drinkers. If the study required heavy drinking status (n = 3), participants had to meet one of the following criteria: (a) greater than 48 drinks per month, (b) greater than 4 drinks per occasion or 7 drinks per week for females, and greater than 6 drinks per occasion or 14 drinks per week for males; (c) an Alcohol Use Disorder Identification Test (AUDIT;(Saunders et al., 1993)) score of 8 or higher or (d) a score of 2 or higher on the CAGE questionnaire (Ewing, 1984). Exclusion criteria across all six studies included: (a) current involvement in treatment programs for alcohol use or have received AUD treatment (i.e. psychotherapy, self-help groups and/or pharmacotherapy) in the prior 30 days to study participation; (b) use of non-prescription psychoactive drugs or use of prescription medications for recreational purposes; (c) history of other psychiatric disorders; (d) current use of antidepressants, mood stabilizers, sedatives, anti-anxiety medications, seizure medications or prescription painkillers; (e) history of contraindicated medical conditions (e.g. cardiac disease); (f) if female, pregnant, nursing or planning to get pregnant; (g) breath alcohol concentration (BAC) of greater than 0.000 g/dl at the time of consent; and (h) positive urine toxicology screen for any drug (other than cannabis).

Measures

Four measures were used to construct the brief AUD severity measure, including: (a) the Structured Clinical Interview for DSM-5 (SCID-5) AUD Module (American Psychiatric Association, 2013); (b) the Penn Alcohol Craving Scale (PACS; (Flannery et al., 1999)); (c) the Drinker Inventory of Consequences (DrInC; (Miller, 1995)); and (d) the Clinical Institute Withdrawal Assessment- Alcohol (CIWA;(Sullivan et al., 1989)). While these are not the only assessments available for each construct, the selection of these specific measures was informed by previous validated work on an AUD severity factor (Moallem et al., 2013; Donato et al., 2021). All measures were shown to have good internal consistency, as evidenced by Cronbach's alpha values above 0.7.

The AUD Module of the SCID-5 is comprised of 11 criteria for AUD, with the threshold for diagnosis being set at the presence of two or more criteria. The PACS is a five-item self-administered instrument used to assess the frequency, intensity and duration of the patient's cravings as well as the patient's perceived ability to resist drinking (Flannery et al., 1999). The Clinical Institute Withdrawal Assessment- Alcohol, revised (CIWA-AR) is a 10-item measure used to provide a quantitative index of the severity of the alcohol withdrawal syndrome (Sullivan et al., 1989). Finally, negative drinking-related consequences were measured using the DrInC, which is a 50-item self-report assessment (Miller, 1995).

The concurrent validity of the brief AUD severity scale was evaluated by examining its relationship with existing measures of severity, including the Alcohol Dependence Scale (ADS) and the Alcohol Use Disorders Identification Task (AUDIT). The convergent validity of the brief AUD severity scale was evaluated by testing its association to other measures related to severity, such as baseline alcohol consumption and family history of alcohol-related problems. As anxiety and depression have been shown to be associated with AUD, the Beck Anxiety Inventory (BAI) and Beck Depression's Inventory (BDI) were also included in the correlation analysis to examine the relation to the proposed measure.

Data analysis

All factor analyses reported in this paper were conducted on raw data scores. First, a confirmatory factor analysis (CFA), informed by the previously proposed severity construct (Moallem et al., 2013; Donato et al., 2021), was conducted using PROC CALIS to examine model fit in a large community sample (n = 1939). Missing data were handled using a full information likelihood estimation (FIML) approach to the CFA (Enders, 2001; Köse, 2014). Summary scores were created for each of the four measures taken at an initial study visit: (a) criteria count via the SCID-5 AUD module; (b) craving per PACS, (c) withdrawal per the CIWA and (d) drinking related consequences per the DrInC. The four summary scores represented the indicator variables hypothesized to measure the latent variable, AUD severity. This important first step examined the evidence that indicator variables (i.e. craving, withdrawal, alcohol-related consequences and DSM-5 criteria count) effectively measured the underlying construct of interest, severity, and that the hypothesized measurement model demonstrated an acceptable fit to the data (i.e. standardized root mean square residual (SRMSR) < 0.08; root mean square error of approximation (RMSEA) < 0.06; comparative fit index (CFI) > 0.96).

Following results from the CFA, a comprehensive item list from the four measures (i.e. CIWA, DrinC, PACs and SCID-5 AUD criteria) was created and subsequently reduced through exploratory factor analysis (EFA) using PROC FACTOR within each individual measure to find the highest loading items. For the PACS, SCID-5 and CIWA, the top two highest loading items were chosen. Given the added length of the DrInC and inclusion of subscales, an EFA was first run on the six subscales to find the highest loading subscales. Then, an additional EFA was performed on the items within each selected subscale and the highest loading item was retained. Taken altogether, these items would be collected to form the initial brief severity measure.

Once the final items were merged into a preliminary assessment of severity, an EFA was run on the 9 items to examine the factor structure. Missing data were handled using a listwise deletion approach. The factor loadings, eigenvalues and scree plot were used to determine whether the proposed items loaded significantly onto a construct and whether a single severity factor was optimal.

To test initial psychometric properties of measure, a series of correlations were conducted to examine the severity measure's relation to clinical variables of interest and similar measures of severity. These correlations were compared to previous results reported on the severity construct and clinical variables of interest (Donato et al., 2021).

RESULTS

Confirmatory factor analysis of severity construct

Based on an exploratory factor analysis previously reported by Donato et al. (2021), a single-factor model of severity was proposed for the confirmatory factor analysis. The chi-square statistic was examined to determine how close the model-implied covariance matrix matched the observed covariance matrix. The chi-square test statistic () demonstrated that the model provided good fit, as evidence by the small chi-square value and large P-value. Other contemporary goodness-of-fit indices revealed similar support for the single-factor model of severity. The standardized root mean square residual (SRMSR; Bentler, 1995), an absolute index of model fit, was substantially below the fit index cutoff value of 0.08 or less (SRMSR = 0.011) (Hu and Bentler, 1999). The root mean square error of approximation (RMSEA; Steiger, 1980), a parsimony index, was also below the criteria (i.e. 0.06 or less; Browne and Cudeck, 1992) for good fit (RMSEA = 0.011, 90% CI = 0.000–0.048). Lastly, the comparative fit index (CFI; Bentler, 1995), an incremental index, was above the established cutoff of 0.96 or larger (CFI = 0.999) (Hu and Bentler, 1999). The CFA indicated moderate to high factor loadings for the indicator variables defining the severity factor (i.e. CIWA = 0.42; DSM-5 Criteria Count = 0.90; PACS = 0.68; DrInC = 0.76). All four standardized factor loadings were significant (P-values <0.0001), indicating that each observed variable significantly contributed to the measurement of the severity factor. Additionally, indicator reliability estimates were in the favorable range (i.e. R² > 0.39) for all indicator variables except the CIWA (DrInC R² = 0.58; PACS R² = 0.51; DSM-5 Criteria Count R² = 0.81; CIWA R² = 0.21). Taken together, these findings provided general support for the singular severity factor model and its chosen indicators.

Exploratory factor analysis of brief severity measure

Following the results of the CFA, individual exploratory factor analyses were run within each of the four indicator variable measures (i.e. CIWA, PACs, DSM-5 criteria count and DrInC) to select the highest loading items. Refer to the Supplementary Material for the factor loading results for each measure. In total, nine items were selected from the four measures.

After identifying the highest loading items from each indicator variable, the nine items were combined to form an initial prototype of the brief severity measure to be used for analytic purposes. An EFA conducted to examine the factor structure of the measure revealed that a single factor model of severity was still found to best fit the data, as determined by the eigenvalue greater than one ( (Kaiser 1960), scree test (i.e. clear "break" in scree plot; see Supplementary Material) (Cattell, 1966), proportion of variance accounted for (42.8%) and interpretability criterion (Hatcher and O'Rourke, 2013). All nine items loaded significantly onto the severity factor, indicated by factor loadings of 0.40 or greater. Refer to Table 1.

Table 1.

Factor pattern and final communality estimates from exploratory factor analysis of brief AUD Severity Scale

Factor loadings	Communality estimates (h2)	Items
0.43	0.18	Do you feel sick to your stomach? Have you vomited?”)
0.44	0.19	Do you feel nervous?
0.75	0.56	At its most severe point, how strong was your craving during the past week?
0.76	0.58	Rate your overall average alcohol craving for the past week
0.66	0.43	Have you spent a lot of time drinking, being drunk or hung over
0.69	0.47	Have you had to give up or reduce the time you spent at work or school, with family or friends or on things you like to do because you were drinking or hungover
0.72	0.52	Because of my drinking, I have not eaten properly
0.65	0.43	A friendship or close relationship has been damaged by my drinking
0.69	0.48	I have failed to do what is expected of me because of my drinking

Note. N = 324. Missing data handled using listwise deletion.

Initial assessments of psychometric properties

Cronbach's coefficient alpha was 0.79 for the nine items included in the analysis. This coefficient exceeds the recommended minimum value of 0.70 (Nunnally, 1994) and nearly meets the ideal range of 0.80 ≤ α ≤ 0.90, indicating the measure has good internal consistency. Reliability analysis on the items showed all items appeared to be worthy of retention, as evidenced by a decrease in the alpha if deleted. The brief severity scale also demonstrated strong correlations between similar measures of severity (e.g. ADS, AUDIT) and clinical variables of interest (e.g. family history of alcohol-related problems, BAI, BDI, baseline alcohol consumption). These results (refer to Supplementary Material) indicated preliminary evidence for the convergent validity and concurrent validity of the measure.

DISCUSSION

The purpose of the present study was to empirically derive a brief measure of AUD severity (i.e. the Brief AUD Severity Scale (BASS)) that could be implemented in clinical settings. The foundation for this measure is based on the severity model first proposed by Moallem et al. (2013) and later supported in a treatment-seeking sample (n = 200) (Donato et al., 2021). Confirmatory factor analysis was employed to investigate whether the previously established dimensionality and factor-loading pattern (Moallem et al., 2013; Donato et al., 2021) would fit a new sample of moderate to heavy drinkers (n = 1939). The goodness of fit indices (i.e. SRMSR, RMSEA and CFI) represented very good fit between the measurement model and the data. Additionally, the percent of variation in the indicator variables that is explained by the factor that it is supposed to measure (i.e. AUD severity) was examined. These estimates of indicator reliability (R² values) were above 0.39 for all indicators except the CIWA. The CIWA has consistently been found to be the lowest loading item (Donato et al., 2021); therefore, it is unsurprising that it had the lowest R² value as R² values are derived from standardized factor loadings. Given the overall goodness of fit indices and the consistency of results found across the three studies, it appears that the proposed latent construct of severity, as indicated by CIWA, PACS, DrInC and DSM-5 criteria count, is an effective and reliable measurement model.

Following the results from the CFA, we sought to derive a theoretically (i.e. capture the purported constructs) and empirically (i.e. has the highest loadings) meaningful item list. A simple, structured measure was the primary goal of this approach in order to optimize the comprehension and delivery of the measure. Results from the individual EFAs determined nine items total to comprise the measure. These items were then analyzed in an EFA to examine the factor structure of the measure. All items loaded significantly onto the measure, and the scree plot, eigenvalues and proportion of variance estimates demonstrated that the items represented a single severity factor, as hypothesized. The Cronbach alpha indicated acceptable internal reliability of the measure. It is important to take into consideration that including a smaller number of items in a measure, as in the case of the BASS, can result in a smaller α. It is also preferable to be below 0.90 as high α values could indicate redundant questions on the test. Evidence of preliminary convergent and concurrent validity was also provided through examination of the relation of the BASS with existing measures of AUD severity, as well as variables theorized to be related to AUD severity. Significant associations in the expected direction were observed for a majority of the variables tested. These correlations indicate that the measure is related to similar measures of severity, including the ADS and the AUDIT. The measure was shown to be positively correlated to clinical variables of interest such as family history of AUD, depression, anxiety and alcohol consumption. This is a notable strength of the measure as it both aligns with our theoretical understanding of AUD as well as provides foundation for the possible predictive utility of the measure.

While four previously validated and well-established measures (i.e. CIWA, PACS, DrInC and DSM-5 criteria count) were used to derive the original model of the BASS, the final measure differs from previous work in several ways. The measurement properties of the BASS were selected with the intent of providing clinically useful information about an individual's AUD severity in a brief and simplified manner. The items chosen for the measure were reworded to specifically address the observer's perspective (i.e. self-report) in order to eliminate the need for a formal clinical interview by a trained professional. Item responses were formatted so that clinicians could compute a meaningful final severity score using little to no technology. Careful consideration was also given to the timeline of certain questions (e.g. past year, lifetime, last two weeks) and adjustments were made to promote clarity and utility of the measure. It is important to note that despite the strong correlations with alternative measures of AUD and overlap with the DSM AUD criteria, the proposed measure represents an important contribution in that it will save time and resources while providing a similar, if not better, measure of AUD severity. Additionally, the AUD severity construct that the measure is based upon has been shown to have predictive utility above that of the DSM-5 criteria count in research settings (Donato et al., 2021).

While the items selected for the BASS were derived empirically, they have a strong theoretical basis and align with the current literature. Given the lack of consensus on how best to operationalize severity, some researchers may argue that degree of alcohol consumption is a necessary aspect of AUD severity and should thus be included in clinical measures of severity. However, the necessity of quantifying alcohol consumption for the purpose of indexing AUD severity is not supported in the literature and only shows a moderate overlap with AUD (Saha et al., 2007; Rehm et al., 2013). In line with current literature, it was important that the measure represent a continuum of severity (Saha et al., 2006; Liu, 2017; Boness et al., 2019). Numerous studies using IRT models to estimate the severity of AUD criteria indicate that the criteria differ in terms of AUD severity; reducing activities and withdrawal were often at the highest end of AUD severity (Langenbucher et al., 2004 (Langenbucher et al., 2004; Kahler and Strong, 2006; Saha et al., 2006; Gilder et al., 2011; McCutcheon et al., 2011), while craving was in the moderate range of severity (Casey et al., 2012) and time spent drinking fell in the low severity range (Kahler and Strong, 2006). Given that these items are all represented in some capacity in the BASS, we hypothesize that the measure can accurately represent a wide continuum of severity, given both the range of response items and the items included.

The current study should be considered in terms of its strengths and limitations. Additional strengths of this study include the large sample size and diverse range of clinical presentations. Limitations include the high percentage of missing data in the sample, which limited the final sample size for the EFA on the measure. Additionally, IRT studies have suggested that although using empirical estimates of severity (e.g. IRT estimates of latent severity) represent an incremental improvement over severity grading (i.e. in the case of DSM-5 AUD), these approaches may still be problematic if they still assume equal weighting among criteria. The current measure includes different response options for each question; therefore, certain items hold the potential to contribute more to the overall severity score if the items were weighted equally. To combat this potential problem, the final severity score will be calculated using specific item weights informed by factor structure. This scoring technique is pending further refinement in future crowdsourcing studies. As an additional note, the severity of AUD captured by this measure must be considered in terms of the time of functioning assessed. For example, it is important to note that the withdrawal items (informed by the CIWA-AR) are designed to assess current withdrawal syndrome, and therefore will not capture withdrawal symptoms that may have already subsided.

Another aspect of the measure that warrants discussion is the nature of the measure as self-report. There is debate as to whether diagnostic criteria from the DSM-5 AUD module can be accurately adapted into self-report formats for the purpose of diagnosis, and efforts to do so have been critiqued in the literature (Campbell and Strickland, 2019; Baggio and Iglesias, 2020). Numerous researchers have pointed out the limitations of self-report measures as they often lead to high rates of false positives or false negatives (Wiseman and Heithoff, 1996; Connors and Volk, 2003; Maraz et al., 2015). While self-report is a useful tool in terms of ease of implementation, it is not as rigorous of a diagnostic assessment such as a clinical interview. At this point, we cannot make claims about the accuracy of the measure in detecting AUD pathology, nor is this our goal. The purpose of our efforts is to propose a measure to calculate a continuous variable indicating severity of one's AUD, rather than a categorical measure. The ability of our measure as a diagnostic classification tool has yet to be studied. However, current efforts are being made to implement the measure into ongoing lab studies at the UCLA Addictions Laboratory. All the laboratory studies currently require a clinical interview (i.e. DSM-5 SCID) with a trained clinician, which offers the opportunity to test the diagnostic accuracy of the assessment. While the BASS is not meant to be a diagnostic tool, it is important to study the sensitivity, specificity and predictive value of the measure to make more accurate recommendations on how it can be implemented and explore the possibility of establishing clinically meaningful cutoff scores. These cutoff scores could further inform treatment selection, such as matching to pharmacotherapies or triaging to appropriate resources.

To maximize the utility of the BASS and to address the limitations noted above, additional studies of the psychometric properties of the instrument are needed. Future analyses are planned to empirically test how the measure relates to previous work. It remains necessary to verify the dimensional structure of the instrument in a new, diverse sample of drinkers and establish its psychometric properties in terms of reliability, construct validity, predictive validity and responsiveness. Subsequent studies should also examine the influence of demographic variables (e.g. gender, age, socioeconomic status) on performance.

In conclusion, the results of the present study suggest that the novel brief AUD severity scale has the potential to be a useful instrument and that further psychometric investigations are warranted. Past research has emphasized need for a valid and reliable instrument that measures AUD severity. This brief 9-item scale can be leveraged in future studies as a screening instrument for AUD severity and as a tool for personalized medicine (i.e. treatment matching).

DATA AVAILABILITY

The data underlying this article cannot be shared publicly to protect the privacy of individuals that participated in the study. The data will be shared on reasonable request to the corresponding author.

FUNDING

This research was supported by grants from the National Institute on Alcohol Abuse and Alcoholism (NIAAA) [K24AA025704 to L.A.R and F32AA029288 to S.J.N] and a training grant from the National Institute on Drug Abuse (NIDA) [5T32DA024635 to S.D.]

CONFLICT OF INTEREST STATEMENT

All authors report no financial relationships with commercial interests.