Author's response to reviews Title: Comparing short versions of the AUDIT in a community-based survey of young people Authors:

1. Please clarify in the background section that the validity of the AUDIT-C among young people in a non-clinical setting has been evaluated: (Dawson, Grant et al. 2005). Please refer to reviewer comment 3 below. 2. Please clarify in the conclusion section that four-item AUDIT variations scored more consistently high on all evaluated statistics (as the statistics were not tests and “tests” could also refer to screening tests). This modified wording has been adopted in the abstract conclusion, and reads: “Four-item combinations scored more consistently high across evaluated statistical measures and are the preferred compromise...” (pg. 11, P. 1). In the Introduction section, please consider several revisions to Paragraph 2, sentence 6: 3. There are multiple studies that have previously validated shortened versions of the AUDIT (e.g. AUDIT-C and AUDIT Q#3) using criterion standards based on detailed interviews in general population samples of young people and should be referenced appropriately: (Dawson, Grant et al. 2005; Dawson, Pulay et al. 2010). While we acknowledge that some studies of the AUDIT and AUDIT-C in the general population have been carried out (studies referenced on pg.3, P. 2), these studies are based on adult populations aged 18 years and over. Studies that focus on young people are limited, and to our knowledge there are no studies evaluating AUDIT or its derivatives in non-clinical samples of young people aged under 18 years. Risky alcohol consumption among young people is common in Australia and associated with increased risk of injury and other adverse effects. Subsequently, they are an important group in which to monitor alcohol consumption and related harm over time. Our study provides access to a high-risk group of young people, including those aged under 18 (in this case 16 and 17 year olds), in which shortened versions of the AUDIT would be valuable, but there is inadequate evidence available to demonstrate their optimal use. 4. The reference for AUDIT-3 should be corrected—the “AUDIT-3” evaluated in Gual 2002 comprised the first three questions of the AUDIT (generally reported as the “AUDIT-C”) rather than the 3 rd question of the AUDIT (e.g. (Bradley, DeBenedetti et al. 2007). This reference has been replaced with Gordon AJ, Maisto SA, McNeil M, et al. Three questions can detect hazardous drinkers. J Fam Pract. 2001 Apr; 50(4):313-20, which included evaluation of AUDIT-3 (3 rd question of the AUDIT). 5. It should be clarified that most previous research has evaluated shortened versions of the AUDIT based on “gold-standard” criterion standards of hazardous/harmful alcohol use from detailed interviews, not as proxies for the full AUDIT. The nature of the survey setting and design precluded inclusion of a gold-standard criterion for hazardous/harmful alcohol use. This is recognised in the limitations discussion, pg. 9, P. 3. Our methods are similar to those used by McCambridge and Thomas (2009). Of note, although the AUDIT-C has generally been found to perform similarly well for identifying hazardous/harmful alcohol use compared to the full AUDIT, a recent study compared the agreement of the AUDIT-C with the full AUDIT and found that the screens identified differing patients: (Neumann, Linnen et al. 2012). We thank the reviewer for drawing this study involving a clinical population to our attention. In the introduction (pg. 3, P. 2), we only state that shortened versions may be sufficient proxies for the full AUDIT. Also, we have focused on referencing studies of AUDIT variations based on non-clinical samples because they are more relevant to our study population and performance is likely to differ considerably between populations and various clinical settings. We note that there is a need for further research to evaluate shortened version of the AUDIT, such as AUDIT-C, in non-clinical settings and particularly younger populations. This study seeks to add to this research. We found that a number of other threeor fouritem AUDIT variations performed more strongly against the full AUDIT than AUDIT-C. Under Analysis in the Methods section, consider: 6. Moving the reference to Table 1 from the first to the second paragraph since more than the 4 shortened versions of the AUDIT listed in Table 1 were scored. The sentence referring to table 1 has been amended to specify that it refers to established shorted versions of the AUDIT which were scored. 7. Describing the adaptation of the FAST that was used in this study (as mentioned in the limitations). A description of how the FAST tool was adapted has been moved from the limitations to the methods analysis section (pg. 5, P.3), and the implications of this adaptation outlined in the discussion of the study limitations (pg. 10, P. 1). 8. Clarifying what gender-comparisons were conducted (1st paragraph, last sentence). This sentence has been amended to clarify that comparisons by sex were conducted on WHO classification of hazardous drinking/high-level drinking problems/probable alcohol dependence (pg. 5, P. 2). 9. Adding at the start of the second paragraph that all possible combinations of three and four AUDIT items were evaluated. A statement to this effect has been added to methods, pg. 5, P.2. 10. Clarifying in a separate sentence that R2 statistics were calculated for each individual question (as well as for all possible combinations of three or four AUDIT items and established AUDIT variations, to which the sentence pertains). The wording has been amended to clarify the meaning of the sentence, such that R2 statistics were calculated to measure the total variance explained in the overall AUDIT-10 score by each individual item and novel combination of items (pg. 6, P.1) 11. Removing the reference to Table 2 results. The reference to table 2 has been removed. In the Results section, consider: 12. Rephrasing the first sentence so that all parts are past-tense. The sentence has been rephrased into past tense. 13. Clarifying that Table 2 shows the ten top performing novel combinations. It has been clarified that table 2 shows the top ten performing combinations of both threeand fouritem combinations, as well as established AUDIT variations (pg. 7, P.2). 14. Clarifying that the novel combination x, x, and x is the novel combination of questions x, x, and x, at least when you first introduce this terminology. This has been clarified in the results, pg. 7, P.3, where the terminology is first introduced. 15. Clarifying under “variance explained” that this refers to variance in the AUDIT-10. This wording has been amended to specify results refer to variance explained in the AUDIT-10 score (pg . 7, P.4). In the Discussion section, consider: 16. Clarifying in the first sentence that results capture much of the information available from the full AUDIT, suggesting that the shortened versions of the AUDIT might be as effective a screening tool for measuring hazardous or harmful alcohol consumption (since the study doesn’t evaluate this directly). As suggested, this sentence has been modified to match the scope of the study and avoid overstating the findings (pg. 8, P. 2). 17. Clarifying in the last sentence of the 1st paragraph that novel AUDIT versions may be more effective...in community-based populations of young Australians. This sentence has been amended (pg. 8, P. 2). 18. Clarifying in the 3rd paragraph that correlation with the full AUDIT is also likely high because the screens are subsets of AUDIT questions and not independent measures. This is an important point which may explain the high correlation, and has been added to the discussion, pg. 9, P. 1. However, we believe it still holds true that the inclusion of all three AUDIT domains may account for the high correlation because some shortened versions, such as AUDIT-C, are still a subset of the AUDIT but exhibited comparatively lower correlation to AUDIT-10. 19. Adding mention of the limitation of agreement measures that agreement is not a measure of “correctness” (e.g. two tests could agree and both be wrong). This was alluded to in the sentence stating “As such, we are only able to determine the performance of the shortened AUDIT variations in comparison to the full AUDIT, rather than as a predictor of harmful drinking per se”. To further emphasise this limitation, we have added that agreement does not infer accuracy of either the shortened version of AUDIT 10 as a predictor of hazardous drinking or high-level drinking problems (pg. 10, P. 1). 20. Clarifying in the limitations paragraph that the study data (or the study survey) did not include a gold-standard measure of harmful/hazardous alcohol use or alcohol-related problems. It was initially written that the study did not include a gold-standard or comparative clinical diagnosis of alcohol-related problems, such as DSM-IV. This has been amended to also state that the study did not have a gold standard for harmful/hazardous alcohol use as per the reviewer’s suggestion (pg. 9, P. 3). 21. Clarifying in the limitations paragraph that the problem with the shortened versions of AUDIT not being independent of the AUDIT is that the diagnostic accuracy (i.e. AUCs) of the short versions is likely to be overstated to the degree the diagnostic error of the AUDIT is correlated with the error of the screen being evaluated. (Note: the assumption of independence necessary for inference based on linear regression is among residuals – errors associated with one observation should not be correlated with the errors of any other observation). The AUROC analysis has been removed from the manuscript. In the conclusion section, consider: 23. Clarifying that the study identified shorter versions of the AUDIT that were predictive of high-level alcohol problems as identified by AUDIT scores ≥ 16. Removal of AUROCs from the analysis means that the study no longer evaluates shortened versions of the AUDIT that were predictive of high-level alcohol problems and this sentence has been removed from the conclusion (pg. 10, P.3). 24. Clarifying in the last sentence that four-item combinations scored more consistently high across multiple measures. This sentence has been amended to read ‘consistently high’ (pg. 11, P. 1).