Skip to main content
PMC home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2016 Mar 1.
Published in final edited form as: Addict Behav. 2014 Nov 25;0:144–147. doi: 10.1016/j.addbeh.2014.11.020

On the Invariance of the Stimulant Craving Questionnaire (STCQ) across Cocaine and Methamphetamine Users

Thomas F Northrup 1, Charles Green 2,3, Robrina Walker 4, Tracy L Greer 4, Madhukar H Trivedi 4
PMCID: PMC4272872  NIHMSID: NIHMS645363  PMID: 25462663

Abstract

Background

The rapid rise in the number of methamphetamine users, relative to cocaine users, has brought the number of each to nearly equal levels, making research on similarities and differences across these groups a needed area of exploration. Craving is postulated to play a significant role in relapse for both user types, yet group differences on observed scale scores have been reported without first assessing the prerequisite measurement equivalence (invariance) of the items, which is essential for meaningful group comparisons.

Methods/design

Baseline data from stimulant users in residential treatment (N=301; n=177 cocaine; n=124 methamphetamine) were used to assess the measurement invariance of the 10-item Stimulant Craving Questionnaire (STCQ), which was adapted from a cocaine-specific measure.

Results

The unifactorial STCQ demonstrated measurement invariance across cocaine and methamphetamine users for factor loadings (metric), common residual covariances between item pairs, and item intercepts (scalar), as determined by fit indices (RMSEA<0.05; CFI & TLI>0.95; SRMR<0.10). The latent mean, as well as 5 (out of 10) item means and the overall composite scale score, were significantly greater for methamphetamine users compared to cocaine users.

Discussion

Results indicate the STCQ is an invariant tool for the assessment of stimulant craving across the two most prevalent user types. Methamphetamine users had significantly higher levels of observed and latent craving than cocaine users, demonstrating a potentially meaningful difference in craving between users of these two stimulants. Future research will determine if treatments and statistical models need to account for craving variations across methamphetamine and cocaine users.

Keywords: Cocaine, Methamphetamine, Stimulant, Craving, Invariance

1. Introduction

In the past decade the number of methamphetamine-dependent individuals has increased dramatically (NIDA, 2012; SAMHSA, 2005) nearly equaling cocaine-dependent users (NIDA, 2010). This trend has prompted the examination of clinical differences across these two stimulant user types. Stimulant-user differences have been explored for decades (e.g., Newmeyer, 1978) and recent findings highlight pharmacokinetic differences between stimulants. For example, methamphetamine takes longer than cocaine to clear the brain (Fowler et al., 2008). Despite physiological differences, other constructs related to both groups remain underexplored.

Craving is one factor often hypothesized to play a role in stimulant use and relapse following treatment (Robinson & Berridge, 1993; Volkow et al., 2006) and is critical to evaluate for dissimilarities across users. Despite possible differences in how craving is experienced by cocaine- and methamphetamine-dependent individuals (e.g., Hilburn et al., 2011), research often assumes the construct is measured equivalently (e.g., factor loadings are invariant). Measurement invariance exists when item content has the same meaning across groups, which allows score comparison. Without establishing invariance, observed differences may simply be mean differences on the same construct or the measure could capture different constructs across groups. Ideally, one instrument could be used to assess craving equivalently across stimulant users by removing non-invariant items (Cheung & Rensvold, 1999).

A brief practical measure of cocaine craving exists (i.e., the 10-item Cocaine Craving Questionnaire-Brief [CCQ-Brief]; Sussner et al., 2006); however, the data were collected over a decade ago at one site, excluded individuals under age 26 and individuals dependent on stimulants other than cocaine, and only 3% were female. The CCQ-Brief, reworded to include more stimulant types, is ideal for psychometric evaluation with a diverse stimulant-user population. Furthermore, the modifications underscore the need for additional evaluation. The aim of this research was to psychometrically assess a new tool, the Stimulant Craving Questionnaire (STCQ), by evaluating the measurement invariance across cocaine and methamphetamine users.

2. Methods

The STimulant Reduction Intervention using Dosed Exercise (STRIDE) study (conducted within the National Drug Abuse Treatment Clinical Trials Network) recruited from 9 Institutional Review Board-approved sites (Trivedi et al., 2011). Briefly, men and women (ages 18-65) who were admitted to residential treatment, used stimulants in the 30 days prior to admission, met DSM-IV criteria for stimulant abuse or dependence, and medically cleared to exercise, were eligible. Consented participants were randomized (N=302) to dosed cardiovascular exercise or health education in addition to treatment-as-usual. One participant not meeting inclusion criteria was removed from this baseline psychometric evaluation of 301 participants.

The Stimulant Craving Questionnaire-Brief (STCQ) was adapted from the 10-item unifactorial CCQ-Brief (Sussner, et al., 2006). “Cocaine” and “coke” from the CCQ-Brief were replaced with “cocaine, methamphetamine or other stimulants.” Participants were asked to indicate their feelings “right now” on a 0 (“Strongly Disagree”) to 6 (“Strongly Agree”) scale. A composite was generated by averaging all items (Table 1).

Table 1.

Standardized Factor Loadings, Item Intercepts, & Residual Covariances for the STCQ

All (N=301) Cocaine (N=177) Methamphetamine (N=124) Means Testing
Item M(SD) M(SD) FL(SE)* Intercept (SE)* M(SD) FL(SE)* Intercept (SE)* p
i1: I want [stimulants] so bad I can almost taste it 0.7(1.4) 0.6(1.4) 0.72(0.07) 0.44(0.04) 0.9(1.4) 0.84(0.04) 0.41(0.07) 0.06
i2: I have an urge for [stimulants] 1(1.6) 0.8(1.5) 0.77(0.06) 0.57(0.04) 1.4(1.7) 0.85(0.03) 0.51(0.07) <0.01
i3: I am going to use [stimulants] as soon as possible 0.3(0.8) 0.2(0.7) 0.53(0.06) 0.26(0.04) 0.4(0.8) 0.55(0.08) 0.22(0.05) <0.05
i4: I think that I could resist using [stimulants] now 1.5(2.0) 1.5(2.1) 0.21(0.06) 0.69(0.05) 1.6(1.9) 0.28(0.07) 0.76(0.06) 0.78
i5: I crave [stimulants] right now 0.9(1.6) 0.8(1.6) 0.74(0.08) 0.49(0.04) 1.2(1.6) 0.82(0.05) 0.44(0.07) <0.05
i6: All I want to use now are [stimulants] 0.5(1.3) 0.4(1.2) 0.69(0.07) 0.31(0.04) 0.8(1.3) 0.85(0.04) 0.31(0.06) <0.05
i7: I have no desire for [stimulants] right now 1.7(2.2) 1.5(2.2) 0.43(0.06) 0.67(0.05) 2.0(2.2) 0.56(0.06) 0.71(0.07) 0.12
i8: Using [stimulants] now would make things seem just perfect 0.5(1.4) 0.4(1.1) 0.56(0.08) 0.33(0.04) 0.8(1.6) 0.56(0.08) 0.26(0.05) <0.01
i9: I will use [stimulants] as soon as I get the chance 0.3(0.9) 0.2(0.8) 0.42(0.11) 0.29(0.04) 0.4(1.0) 0.43(0.10) 0.24(0.05) 0.05
i10: Nothing would be better than using [stimulants] right now 0.3(1.0) 0.2(1.0) 0.62(0.10) 0.25(0.04) 0.5(1.2) 0.68(0.07) 0.22(0.05) 0.06
STCQ Composite 0.8(0.9) 0.7(0.9) - - 1.0(1.0) - - <0.01
Residual Covariance Cocaine Estimates (SE)* Methamphetamine Estimates (SE)*
Item 3 with item 9 0.46(0.14) 0.33(0.10)
Item 1 with item 2 0.36(0.10) 0.46(0.10)
Item 4 with item 7** 0.30(0.09) -
Item 8 with item 10** 0.39(0.12) -
Open in a new tab

Note. Factor loadings and intercepts are provided for items 4 and 7 after reverse scoring. “Means testing” refers to t tests conducted for each individual item and the composite. “STCQ Composite” was computed using the CCQ-Brief scoring method of averaging the responses for all 10 items after reverse scoring items 4 and 7. [Stimulants] = cocaine, methamphetamine, or other stimulants; FL=factor loading; i=item.

*

All factor loadings, intercepts, and residual covariances were significant at the p<0.01 level.

**

Only included in the Cocaine group's model.

3. Statistical Analyses

A Confirmatory Factor Analysis (CFA) with Maximum Likelihood estimation and a mean-adjusted chi-square test (MLM) obtained standard errors and the Satorra Bentler chi-square test (χ2SB), which are robust to non-normality (Curran, West, & Finch, 1996; Muthen, 2009). The Comparative Fit Index (CFI; >0.95 considered good), Tucker-Lewis Index (TLI; >0.95 considered good), Root Mean Square Error of Approximation (RMSEA; ≤0.05 considered good), and Standardized Root Mean Residual (SRMR; <0.10 considered good) were also used to evaluate model fit (Kline, 2005).

Model fitting and invariance testing followed recommendations by Byrne (Chapters 4 & 7; 2012) for MPlus, version 7. Generally, invariance testing proceeded by freely estimating parameters for each group, then constraining them (i.e., factor loadings, shared residual covariances, intercepts, and latent means) sequentially to assess model-fit decrements as evidence of non-invariance. Specifically, we began with a CFA to test the hypothesized unifactorial validity for each group (cocaine and methamphetamine) and optimize the model for each group separately before invariance testing. Modification indices (Muthen & Muthen, 1998-2012) guided theoretically justified modifications. Invariance testing proceeded by estimating configural invariance (i.e., factor loadings' patterns), metric invariance (i.e., factor loadings' magnitudes), common residual covariance invariance, scalar invariance (i.e., items' intercepts), and finally latent means. The χ2SB difference test (Δ χ2SB) was used to compare nested models for each successive step following suitable mathematical calculations for MLM estimation (MPlus, 2014; Muthen & Muthen, 1998-2012). Stepwise evaluation of improvements (following modifications) were significant if the MLM-adjusted chi-square difference test value was ≥3.84 (p<0.05).

4. Results

Mean age was 39.0 years (SD=10.8) and 39.9% of the sample were female. The sample was 45.2% White, 43.2% Black, 10.3% Hispanic/Latino, and 4.0% Other. Fifty-nine percent (n=177) were cocaine dependent only, 30.6% (n=92) were methamphetamine dependent only, and 10.6% (n=32) were dependent on methamphetamine and one or more other stimulants (polystimulant group). The methamphetamine only and polystimulant groups were combined (n=124).

The entire sample's craving composite was <1.0 (Table 1), suggesting relatively low craving. The number of days since last stimulant use was not different (t=1.50, df=298, p=0.13) across the cocaine-only (M=17.3, SD=9.2) and methamphetamine groups (M=15.7, SD=8.7). Also, the correlation between days since last stimulant use and STCQ scores was not different from zero (r=-0.08, p=0.14), and therefore not included as a covariate. Chronbach's alpha (unstandardized) was 0.83, suggesting good internal consistency reliability.

Model Optimization

Each group's unifactorial model resulted in significant improvements by allowing residual covariances between items 1 and 2 and between items 3 and 9. The cocaine group model also allowed residual covariances between the two reverse-scored items (i.e., 4 and 7) and items 8 and 10. These modifications allowed the models to account for overlapping item content and similar methodology, as recommended by Byrne (p. 233; 2012). Model fit indices were generally excellent for the cocaine group (χ2SB=42.6, df=31, p=0.08; CFI=0.96; TLI=0.94; RMSEA=0.05; SRMR=0.05) and methamphetamine group (χ2SB=42.0, df=33, p=0.13; CFI=0.98; TLI=0.97; RMSEA=0.05; SRMR=0.04).

Invariance Testing

Excellent fit indices (χ2SB=84.9, df =64, p=0.04; CFI= 0.97; TLI=0.96; RMSEA=0.05; SRMR=0.05) for the configural invariance step demonstrated the factor loadings' pattern was invariant across groups. No model adjustments were made based on modification indices. Next, constraining factor loadings to test metric invariance did not result in a significant decrement to model fit, (Δχ2SB=9.1, p=ns) and fit indices were excellent, with the exception of the SRMR (0.11), and no theoretically meaningful changes were suggested by the modification indices. The third step constrained residual covariances between item pairs 3 and 9 and items 1 and 2. The Δχ2SB value was not significant, no model changes were made, and all fit indices (except the SRMR=0.11) were excellent. This result implies that the method effects of these item pairs operate similarly across groups. Next, scalar invariance was assessed by constraining the item intercepts across groups. The Δχ2SB was not significant and the model fit indices remained in the excellent range with the exception of SRMR (0.12).

In the final step, the cocaine group's factor mean was fixed to zero and freely estimated in the methamphetamine group (for model identification; Byrne, 2012), which provides the factor mean for the methamphetamine group relative to the cocaine group. This final step retained excellent fit indices (χ2SB=96.0, df =84, p=0.18; CFI= 0.98; TLI=0.98; RMSEA=0.03; SRMR=0.07). The standardized latent mean of +0.33 (SE=0.10) for the methamphetamine group was significant (z=3.34, p <0.01) and the Δχ2SB was significant (p<0.01). This result implies that individuals in the methamphetamine group (relative to the cocaine group) have higher levels of latent craving, which is supported by a scatter plot (fit with separate regression lines) of latent factor scores by observed STCQ composites (Figure 1) and the comparisons of observed individual item means and overall STCQ composite scores (Table 1). Specifically, 5 (of 10) item means and the STCQ composite were significantly higher for the methamphetamine group (1.0 vs. 0.7, p<0.01). The final model's standardized factor loadings, item intercepts, and residual covariances are provided (Table 1).

Figure 1.

Figure 1

Open in a new tab

Factor (latent) scores by STCQ composite (observed scores) fitted separately for cocaine and methamphetamine. The lines are linear regression fitlines with 95% confidence intervals and prediction limits by group.

Post-Hoc Sensitivity Analysis

Consolidating 32 polystimulant and 92 methamphetamine users could have biased the findings in favor of invariance. The invariance analyses were repeated without the polystimulant users and the model fit indices for each step, individual parameters, and conclusions were highly similar. Therefore, the results including polystimulant users were retained.

5. Discussion

The STCQ demonstrated a high level of invariance across a relatively large residential-treatment sample composed of the two most prevalent stimulant-user groups. These results indicate cocaine and methamphetamine users respond similarly to individual craving items. The significantly higher latent mean for the methamphetamine group supports earlier work (Hilburn, et al., 2011) that this group experiences more intense craving. Latent mean differences are commonly found and still allow meaningful group comparisons.

One other difference stemmed from the cocaine group having two extra residual covariances estimated. This suggests cocaine users may have stronger method-related effects from these item pairs. However, these extra residual covariance estimations were unlikely to have substantively affected the results.

Using CFA to compare latent means (rather than composite [observed] scores), and knowing that latent means are not attenuated by measurement error (Vandenberg & Lance, 2000), offered robust statistical evidence of group differences. Despite methodological rigor, psychometric evaluations can be highly sample dependent; therefore, it is important to compare this study and work reported for the CCQ-Brief (Sussner, et al., 2006). The present sample was recruited from 9 U.S.-based residential treatment programs and included individuals under 26 years of age and significantly more females, which were absent or underrepresented in the single-site veterans treatment center sample. Further, with recent trends for significantly higher proportions of younger adults (18-25 years) to report methamphetamine and cocaine use (NIDA, 2010, 2012; SAMHSA, 2005) this sample may better reflect this growing population. Also, this study only excluded opioid-dependent patients, whereas Sussner et al. did not include individuals dependent on substances other than cocaine. The heterogeneous stimulant-using population poses craving measurement challenges and invariant tools are necessary.

Limitations include an inadequate sample to allow for cross-validation, potentially increasing the capitalization on chance variation. Furthermore, merging polystimulant and methamphetamine users may have introduced a small amount of error, but the post-hoc sensitivity analyses suggested the consolidation was acceptable. The relatively low craving levels reported while this sample was in residential treatment might differ markedly from craving measured in an outpatient setting. Further, stimulant craving levels may change over time (Wang et al., 2013), which future work could assess.

6. Conclusion

Craving can persist up to one year after cessation, particularly for methamphetamine users (Wang, et al., 2013), and is critical to measure. The 10-item STCQ should be less prone to measurement error than single-item measures (Sayette et al., 2000) and more sensitive to change throughout treatment. Future research may determine whether craving is an accurate predictor of relapse as suggested by Sussner et al. (2006). Additional analyses with this measure may also demonstrate key differences across other stimulant user types, as higher levels of latent craving for individuals dependent on methamphetamine were detected. In closing, the STCQ is an invariant instrument for continued study with stimulant users. The clinical utility of this difference is unknown at this stage of research; however, it could be important to account for when designing interventions or performing statistical modeling with a heterogeneous stimulant-user population.

Highlights.

  • An invariant measure of craving is needed across methamphetamine and cocaine users.

  • The Stimulant Craving Questionnaire demonstrated invariance across both user types.

  • During residential treatment, methamphetamine users had greater latent craving.

  • Future research will determine the clinical utility of these craving differences.

Acknowledgments

The authors wish to thank Ira Bernstein, Ph.D., and Thomas Carmody, Ph.D. who assisted in preparation of data for this manuscript.

Statement 1: Role of Funding Sources: STRIDE (CTN-0037) was supported by the National Institute on Drug Abuse of the National Institutes of Health under Award Number U10DA020024 (PI: Trivedi). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Dr. Greer has received research funding from NARSAD and consulting fees from H Lundbeck A/S and Takeda Pharmaceuticals International, Inc. Madhukar H. Trivedi is or has been an advisor/consultant to: Alkermes, AstraZeneca, Bristol-Myers Squibb Company, Cephalon, Inc., Cerecor, Concert Pharmaceuticals, Inc., Eli Lilly & Company, Evotec, Forest Pharmaceuticals, GlaxoSmithKline, Janssen Global Services, LLC/Janssen Pharmaceutica Products, LP/Johnson & Johnson PRD, Lundbeck, MedAvante, Medtronic, Merck, Mitsubishi Tanabe Pharma Development America, Inc., Naurex, Neuronetics, Otsuka Pharmaceuticals, Pamlab, Pfizer Inc., Phoenix Marketing Solutions, Rexahn Pharmaceuticals, Ridge Diagnostics, Roche Products Ltd., Sepracor, SHIRE Development, Sunovion, Takeda, Vivus, and Wyeth-Ayerst Laboratories. In addition, he has received research support from: Agency for Healthcare Research and Quality (AHRQ), National Alliance for Research in Schizophrenia and Depression, National Institute of Mental Health, and National Institute on Drug Abuse

Footnotes

Statement 2: Contributors: Dr. Northrup designed and wrote the manuscript, in consultation with Drs. Walker, Greer, and Trivedi. Dr. Northrup conducted the preliminary statistical analyses. Dr. Green, as a senior biostatistician provided consultation and additional statistical oversight; Dr. Green edited several versions of the manuscript as well. Drs. Walker, Greer, and Trivedi provided substantive edits on drafts of the manuscript. All authors contributed to and have approved the final manuscript.

Statement 3: Conflict of Interest: Drs. Northrup, Green, and Walker declare that they have no conflicts of interest.

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.

References

  1. Byrne BM. Structural Equation Modeling with Mplus: Basic Concepts, Applications, and Programming. New York: Routledge; 2012. [Google Scholar]
  2. Cheung GW, Rensvold RB. Testing factorial invariance across groups: A reconceptualization and proposed new method. Journal of Management. 1999;25(1):1–27. doi: 10.1016/S0149-2063(99)80001-4. [DOI] [Google Scholar]
  3. Curran PJ, West SG, Finch JF. The robustness of test statistics to nonnormality and specification error in confirmatory factor analysis. Psychological Methods. 1996;1(1):16. doi: 10.1037//1082-989X.1.1.16. [DOI] [Google Scholar]
  4. Fowler JS, Volkow ND, Logan J, Alexoff D, Telang F, Wang GJ, Apelskog K. Fast uptake and long-lasting binding of methamphetamine in the human brain: comparison with cocaine. Neuroimage. 2008;43(4):756–763. doi: 10.1016/j.neuroimage.2008.07.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Hilburn C, Nejtek VA, Underwood WA, Singh M, Patel G, Gangwani P, Forster MJ. Is serum brain-derived neurotrophic factor related to craving for or use of alcohol, cocaine, or methamphetamine? Neuropsychiatric disease and treatment. 2011;7:357. doi: 10.2147/NDT.S18953. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Kline RB. Principles and Practice of Structural Equation Modeling. 2nd. New York: The Guilford Press; 2005. [Google Scholar]
  7. MPlus. Chi-Square Difference Testing Using the Satorra-Bentler Scaled Chi-Square. 2014 Retrieved March 26th, 2014, from http://www.statmodel.com/chidiff.shtml.
  8. Muthen LK. MLR Acronym? 2009 Retrieved March 26th, 2014, from http://www.statmodel2.com/discussion/messages/11/2156.html?1176562358.
  9. Muthen LK, Muthen BO. Mplus User's Guide. Sixth. Los Angeles, CA: Muthén & Muthén; 1998-2012. [Google Scholar]
  10. Newmeyer JA. The Epidemiology of the Use of Amphetamine and Related Substances Journal of Psychoactive Drugs. 1978;10(4):293–302. doi: 10.1080/02791072.1978.10471533. [DOI] [Google Scholar]
  11. NIDA. Cocaine: What is the scope of cocaine use in the United States? 2010 Retrieved March 26th, 2014, from http://www.drugabuse.gov/publications/research-reports/cocaine/what-scope-cocaine-use-in-united-states.
  12. NIDA. Methamphetamine. 2012 Retrieved March 26th, 2014, from http://www.drugabuse.gov/drugs-abuse/methamphetamine.
  13. Robinson TE, Berridge KC. The neural basis of drug craving: an incentive-sensitization theory of addiction. Brain Research Brain Research Reviews. 1993;18(3):247–291. doi: 10.1016/0165-0173(93)90013-P. [DOI] [PubMed] [Google Scholar]
  14. SAMHSA. The NSDUH Report: Methamphetamine use, abuse, and dependence: 2002, 2003, and 2004. 2005 Retrieved March 26th, 2014, from http://www.samhsa.gov/data/2k5/meth/meth.htm.
  15. Sayette MA, Shiffman S, Tiffany ST, Niaura RS, Martin CS, Shadel WG. The measurement of drug craving. Addiction. 2000;95 Suppl 2:S189–210. doi: 10.1046/j.1360-0443.95.8s2.8.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Sussner BD, Smelson DA, Rodrigues S, Kline A, Losonczy M, Ziedonis D. The validity and reliability of a brief measure of cocaine craving. Drug and Alcohol Dependence. 2006;83(3):233–237. doi: 10.1016/j.drugalcdep.2005.11.022. [DOI] [PubMed] [Google Scholar]
  17. Trivedi MH, Greer TL, Grannemann BD, Church TS, Somoza E, Blair SN, Nunes E. Stimulant reduction intervention using dosed exercise (STRIDE) - CTN 0037: study protocol for a randomized controlled trial. Trials. 2011;12:206. doi: 10.1186/1745-6215-12-2061745-6215-12-206[pii]. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Vandenberg RJ, Lance CE. A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods. 2000;3(1):4–69. doi: 10.1177/109442810031002. [DOI] [Google Scholar]
  19. Volkow ND, Wang GJ, Telang F, Fowler JS, Logan J, Childress AR, Wong C. Cocaine cues and dopamine in dorsal striatum: mechanism of craving in cocaine addiction. Journal of Neuroscience. 2006;26(24):6583–6588. doi: 10.1523/jneurosci.1544-06.2006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Wang G, Shi J, Chen N, Xu L, Li J, Li P, Lu L. Effects of length of abstinence on decision-making and craving in methamphetamine abusers. PLoS ONE. 2013;8(7):e68791. doi: 10.1371/journal.pone.0068791. [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES