- Email: [email protected]
Fast screening for alcohol misuse
Addictive Behaviors 28 (2003) 1453 – 1463
Short Communication
Fast screening for alcohol misuse Ray J. Hodgsona,*, Bev Johna, Tina Abbasia, Rachel C. Hodgsonc, Seta Wallerb, Betsy Thomc, Robert G. Newcombea a
University of Wales College of Medicine, Cardiff Wales, UK b Health Education Authority, London, UK c Middlesex University, London, UK
Abstract The Fast Alcohol Screening Test (FAST) has been developed from the AUDIT questionnaire [Babor, T. F., de la Fuente, J. R., Saunders, J., Grant, M. (2001). AUDIT: The Alcohol Use Disorders Identification Test: guidelines for use in primary health care. Geneva, Switzerland: World Health Organization] for use in very busy medical settings [Alcohol Alcohol. 37 (2002) 61– 66]. One feature of the FAST is its ease and speed of administration, especially since one question identifies over 50% of patients as either alcohol misusers or not. This study further explores the sensitivity and specificity of the FAST across ages, gender, and locations using the AUDIT as the gold standard. Two other quick tests are also compared with the AUDIT and the FAST, namely the Paddington Alcohol Test [J. Accid. Emerg. Med. 5 (1996) 308] and the CAGE [Am. J. Psychiatry 131 (1974) 1121]. All tests were quicker to administer than the AUDIT with the FAST taking just 12 s on average. All tests identified drinkers who would accept a health education booklet (over 70% of those identified) or 5 min of advice (over 40%). The FAST was consistently reliable when sensitivity and specificity were tested against AUDIT as the gold standard. D 2002 Published by Elsevier Science Ltd. Keywords: Alcohol screening; Audit; Fast
1. Introduction A well-researched and frequently used screening test for alcohol misuse is the Alcohol Use Disorders Identification Test (AUDIT), which was developed in a World Health Organisation * Corresponding author. Tel.: +44-29-2023-3651; fax: +44-29-2023-7930. E-mail address: [email protected] (R.J. Hodgson). 0306-4603/$ – see front matter D 2002 Published by Elsevier Science Ltd. doi:10.1016/S0306-4603(02)00246-0
1454
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
(WHO) collaborative project across six countries. (Babor, de la Fuente, Saunders, & Grant, 1989). This 10-item questionnaire has good sensitivity and specificity as a screen for all types of alcohol misuse, i.e., hazardous drinking, harmful drinking, and dependence. Hazardous drinking refers to a pattern of drinking that is associated with a high risk of psychological or physical problems in the future. Harmful drinkers are already experiencing these problems. The dependent drinker is experiencing symptoms of dependence including impaired control or a subjective experience of compulsion to drink. Henceforth, the term alcohol misuse will be used cover all of these categories. Although AUDIT has been well validated, (Allen, Litten, Fertig, & Babor, 1997; Barry & Fleming, 1993; Bohn, Babor, & Kranzler, 1995; Bradley, Bush, McDonell, Maloine, & Fihn, 1998; Isaacson, Butler, Zacharek, & Tzelepis, 1994; Schmidt, Barry, & Fleming, 1995; Volk, Steinbauer, Cantor, & Holzer, 1997) and is turning out to be a very useful screening test, there are some situations, such as very busy medical settings, where a 10-item questionnaire takes too long to administer routinely. As a consequence a number of very brief tests have been developed. The range of instruments include the CAGE (Mayfield, McLeod, & Hall, 1974), the TWEAK (Russell, Martier, Sokol, Jacobson, & Bottoms, 1991), the brief MAST (Pokorny, Miller, & Kaplan, 1972), the RAPS (Cherpitel, 2000), the five-shot test (Seppa, Lepisto, & Sillanaukee, 1998), the short AUDIT (Piccinelli et al., 1997) and the PAT (Smith, Touquet, Wright, & Das Gupta, 1996). Furthermore, Soderstrom et al. (1998) use the first two Audit questions to assess quantity and frequency of alcohol consumption and the TICS (Brown, Leonard, Rounds, & Papasouliotis, 1997) attempts to assess both alcohol and drug misuse. The main focus of this study is upon the Fast Alcohol Screening Test (FAST) questionnaire, which was developed from the AUDIT, originally for use within accident and emergency (A&E) departments, but it has also been validated in a range of other settings (Hodgson, Alwyn, John, Thom, & Smith, 2002). The FAST correlates strongly with the short AUDIT (.92) but a significant feature of the FAST is the ability of one question to identify over 50% of patients as either alcohol misusers or not. This substantially reduces the time to administer. The FAST is based upon the following AUDIT questions: 1. How often do you have six or more drinks on one occasion? 2. How often during the last year have you been unable to remember what happened the night before because you had been drinking? 3. How often during the last year have you failed to do what was normally expected of you because of drinking? 4. Has a relative or friend, or a doctor or other health worker been concerned about your drinking or suggested you cut down? The current study extends previous work on the FAST by exploring the consistency of sensitivity and specificity, using the AUDIT as the standard, across age, gender, and location. This is not to suggest that the AUDIT is an unassailable gold standard. The aim is simply to compare a shorter questionnaire with one that has proved to be useful. A subsidiary aim is to investigate the relationship between the AUDIT, the FAST, and two other measures that are
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
1455
frequently used in the UK, namely the CAGE and the PAT. The CAGE questionnaire has been widely tested and shown to be useful in identifying the more dependent drinker (Chan, Pristach, & Welte, 1994). The Paddington Alcohol Test (PAT) has also been shown to be useful as a method of identifying problem drinking, particularly within the A&E setting (Smith et al., 1996). In the current study the FAST, the CAGE, and the PAT were compared to the AUDIT in four A&E departments. Both male and female patients between 16 and 75 were involved in the study. The main issues that will be addressed are: 1. Is the FAST questionnaire consistently reliable across age, gender, and location when compared to AUDIT as the gold standard? 2. Since the FAST uses AUDIT questions the CAGE and the PAT are not expected to be as strongly related to the AUDIT but these relationships will be explored. 3. The time taken to administer the four questionnaires will be assessed. 4. The readiness to accept advice will be assessed for those patients who screen positive on the four tests.
2. Methodology A total of 62 nurses were involved in the study. They were volunteers from four A&E departments in London, Southampton, Bristol, and Cardiff. All patients seen by triage nurses for the duration of the study were recruited if they gave informed consent and were not ruled out by the following exclusion criteria: (a) inability to read English; (b) severe intoxication whether by alcohol or other drugs; (c) excessive pain or confusion. After gaining consent the nurse administered a questionnaire, which included demographic variables. This was followed by one of the short screening tests, i.e., the CAGE, the FAST, or the PAT. The full AUDIT questionnaire was then administered, followed by questions about readiness to accept advice. Questionnaires were colour coded depending upon the particular shorter screening test that they included. The order in which the three colour coded questionnaires were administered was determined by random assignment within each A&E department. A random number table was used to order the questionnaires (1 to 3). They were then placed in a box in this order and nurses were asked always to take the top questionnaire.
3. Results A total of 2185 patients completed questionnaires at four centres, London (N=503), Southampton (N=852), Bristol (N=270) and Cardiff (N=560). The overall prevalence of alcohol misuse as defined by the AUDIT instrument in the study population was 843 out of 2175 (38.8%). This will not be representative of all A&E attenders since intoxicated patients were excluded from the study.
1456
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
4. Sensitivity and specificity All three tests evaluated were substantially correlated with AUDIT. Spearman rank correlations between AUDIT and FAST, PAT and CAGE were .79, .55, and .48, respectively. Table 1 displays the sensitivity and specificity of the FAST, the CAGE, and the PAT as predictors of a positive AUDIT in all subjects receiving the test in question. Confidence intervals (CIs) were calculated by the method of Wilson (1927). Table 2 displays sensitivity and specificity (%) by age, gender, and locality. The above results indicate that the FAST has high sensitivity and specificity when compared to the AUDIT as the gold standard. The CAGE has low sensitivity but very high specificity. The PAT has moderately high sensitivity and specificity. There was substantial evidence that sensitivity and specificity were affected by centre, age, and gender. Notably, the sensitivity and specificity of PAT varied significantly between the four centres (c2=12.15 and 36.94, df=3, P<.007 and P<.001, respectively), with also a gender difference in specificity (c2=15.34, df=1, P<.001). The specificity of FAST differed significantly between centres (c2=15.39, df=3, P<.002), and its sensitivity differed significantly between younger and older age groups (c2=7.97, df=1, P<.005). Nevertheless the sensitivity and specificity of FAST were high in all localities, in men and women, and in both age groups. Fig. 1 shows the comparison in sensitivity and specificity between FAST and CAGE as predictors of the AUDIT result. This diagram, based on Newcombe (2001), depicts the relative impact of false negatives and false positives of the two tests. It assumes that the consequences of a false negative result are more serious than those of a false positive result, by a factor r. A quantity f, expressing differences in sensitivity and specificity between the two tests, is plotted against a mixing parameter l, which runs from 0 to 1; f is calculated as ldelta sens+(1l)delta spec, a weighted average of delta sens and delta spec, which denote the differences in sensitivity and specificity between the two tests; l is designed to reflect the balance of two effects: the relative consequences of the two possible types of misclassification and the prevalence of significant alcohol misuse in a population. It is calculated as l ¼ 1þ 111 =r where p denotes the prevalence. If the missed opportunity ðp Þ Table 1 Sensitivity and specificity of FAST, CAGE, and PAT tests relative to AUDIT as gold standard Estimate
95% CI
FAST Sensitivity Specificity
269/290 92.8% 404/461 87.6%
89.2% to 95.2% 84.3% to 90.3%
CAGE Sensitivity Specificity
122/303 40.3% 422/432 97.7%
34.9% to 45.9% 95.8% to 98.7%
PAT Sensitivity Specificity
175/250 70.0% 371/439 84.5%
64.1% to 75.3% 80.8% to 87.6%
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
1457
Table 2 Sensitivity and specificity percentage for FAST, CAGE, and PAT tests by age, gender, and locality FAST Male Female Age25 Age>25 Bristol Cardiff London Southampton
CAGE
PAT
N
Sensitivity
Specificity
N
Sensitivity
Specificity
N
Sensitivity
Specificity
443 305 205 543 99 181 171 300
94 89 88 97 91 96 90 92
86 90 82 89 87 82 82 95
442 291 218 516 92 186 184 273
40 41 34 45 46 37 55 31
96 99 98 98 95 96 98 99
392 296 181 507 78 190 146 275
70 69 74 67 86 78 66 59
77 91 74 86 69 71 89 94
associated with a false positive is the dominant consideration, then r tends to infinity, l tends to 1 and f reduces to delta sens. Conversely, if the prevalence is sufficiently low, the penalty associated with the preponderance of false positives predominates: as p tends to 0, with r held finite, l tends to 0 and f reduces to delta spec. The difference in sensitivity (delta sens) is plotted at l=1 to the right of Fig. 1. It is 52.5% in favour of FAST, with 95% CI from 45.8% to 58.4%. The difference in specificity (delta spec) is 10.0% in favour of CAGE and appears at the left of the diagram as 10.0%, at l=0, together with its 95% CI from 13.5% to 6.8%. These CIs are calculated by method 10 of Newcombe (1998). Both CIs exclude zero, indicating that the two tests differ significantly in both sensitivity and specificity. Here, FAST should be regarded as better than CAGE at values of l for which the diagonal line is above 0. This is for all values of l greater than 0.16. For l values greater than 0.21, where the lower curve cuts the horizontal axis, FAST is significantly better than CAGE.
Fig. 1. Comparison of sensitivity and specificity between FAST and CAGE tests. For detailed explanation, see text.
1458
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
Fig. 2. Comparison of sensitivity and specificity between FAST and PAT tests.
Conversely, for l values of 0.16 or lower, CAGE is better than FAST, significantly so for l up to 0.11. In the current study, relatively high values of l are expected. This is because prevalence is high when screening for hazardous and problem drinkers in a medical setting. Also a false negative is more problematic than a false positive. A false positive is probably a borderline case and could possibly benefit from a health promotion intervention. If we consider a false negative to be twice as serious as a false positive then, for a 10% prevalence, l is 0.18. For 20% prevalence l=0.33 and for 30% prevalence l=0.46. These values of l are all above the cut off point at which FAST is better than CAGE. Fig. 2 shows the comparison in sensitivity and specificity between FAST and PAT as predictors of the AUDIT result. FAST has a large, highly significant advantage in sensitivity, and also an advantage in specificity that is somewhat short of statistical significance. FAST is better than PAT for all values of l, and significantly better than PAT for any plausible value of 0.07 or above.
5. Time, cost, and nurse appraisal Table 3 displays the mean time taken to administer each of the screening tests as well as the cost per year if 50,000 patients are routinely screened by nurses at a cost to the NHS Table 3 Mean time taken to administer each test and cost of routinely screening 50,000 patients FAST CAGE PAT AUDIT
N
Mean (S.D.) (s)
Cost per 50,000 patients (£)
67 64 48 191
12.52 (14.2) 14.37 (7.2) 42 (31.9) 78 (35.5)
1,669 1,916 5,600 10,400
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
1459
Table 4 Percentage of alcohol misusers, identified by a particular test, who would accept various interventions FAST CAGE PAT AUDIT
Booklet
5-Minute discussion
Half-hour discussion
Three counselling sessions
77 75 76 76
46 46 44 46
29 34 24 28
25 29 20 23
of £9.60 per hour (E grade nurse in UK). These data were collected in just one of the A&E departments. It should be noted that the current study explored the possibility of using PAT routinely whereas the creators of PAT advise that it should only be administered when medical problems that are known to be related to excessive alcohol use are identified. Twelve nurses were interviewed to ascertain whether any of the screening tests were preferred for any reason over any of the others. Ten nurses concluded that the FAST was the most appropriate screening test for use in a busy medical setting since it was not as intrusive as the CAGE and was quicker than the PAT. All 10 nurses concluded that the detailed calculation of units required for the PAT was the reason why it took longer to administer than the FAST or the CAGE. A key question is how many patients who are identified as alcohol misusers will accept advice. Table 4 displays the percentage of patients, identified as positive on each test, who said yes to the various forms of advice. These percentages did not differ significantly between the four tests.
6. Modifications to FAST Further research on the reliability of the FAST resulted in two very small changes that had no effect on the sensitivity or specificity of the test. The reasons for these changes are given in Hodgson et al. (2002). For men, question one was changed to eight drinks (units) instead of six (see Appendix A). The main advantage of the ‘‘eight drinks’’ version was the use of this question as a first filter for men. The ‘‘six drinks’’ question alone identified 56% of the men as hazardous or nonhazardous drinkers whereas the ‘‘eight drinks’’ question identified 63%. The accuracy of the test was not affected. Also a minor modification was made to the question: Has a relative or friend, or a doctor or other health worker been concerned about your drinking or suggested you cut down? In the AUDIT this particular question is not confined to the previous year and this could result in false positives if the concern occurred a few years ago. The question and the responses were therefore altered to focus upon the last year. In fact, this made no difference to any of the conclusions since only 5% of participants noted that concern had been expressed ‘‘but not in the last year’’. Furthermore, all remained in the same categories (alcohol misusers or not) if their response was altered to ‘‘never’’.
1460
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
7. Discussion It can be concluded that the FAST is a useful, very brief screening instrument when compared to AUDIT as the gold standard. The FAST has high sensitivity and specificity even though it is very quick to administer, and therefore cheap. A&E departments admit a high percentage of people who are misusing alcohol. By including hazardous drinking the FAST, the AUDIT, and the PAT are moving screening and brief interventions into the field of early intervention, health promotion, and prevention rather than just treatment. The FAST is particularly useful since over 50% of patients screened within a medical setting can be identified as either alcohol misusers or not by asking only the first FAST question. (Hodgson et al., 2002). Using the FAST, routine screening in an A&E department dealing with 50,000 patients per year would cost less than £2000 over the year. This does not take account of the extra costs involved in providing advice or a brief alcohol intervention. At low levels of prevalence the CAGE is probably better than the FAST. An example would be screening for more severe dependence. If prevalence is only 5% and both false positives and false negatives are equally serious then l is 0.05. At this value of l, CAGE is better than FAST. A point to be emphasised is that the usefulness of a particular screening instrument depends not only on its sensitivity and specificity, as well as prevalence, but also on the seriousness of a false negative relative to a false positive. When screening hazardous and problem drinkers to identify recipients of a brief intervention then a false negative is more serious than a false positive. The worst that can happen to those who are misclassified as hazardous drinkers is that they will receive or refuse a health promotion intervention. In a different situation, for example when more severely dependent drinkers are being screened for a long interview prior to treatment or admission to an alcoholism service, then a false positive is likely to be more serious since it will be associated with significant costs to both the individual and the service. The PAT questionnaire has a number of positive attributes that recommend its use in A&E departments. First, it is not intended to be used routinely as in the current study. It is to be used selectively for appropriate adult patients where there is a ‘‘a suspicion of alcohol misuse generated by presentations such as falls, assaults, head injuries, gastrointestinal problems, ‘unwell,’ fits, blackouts, collapse, insomnia, sweating, palpitations, chest pain, gout, rashes, depression,’’ etc. (Smith et al., 1996). Second, an attempt is made to get a very accurate estimate of peak consumption. Third, the test has been used in an A&E department for several years. This pioneering work of Dr. Touquet and his colleagues is at last beginning to generate interest in the whole field of screening and brief interventions within the A&E setting.
Acknowledgements This study was funded by the Alcohol Education & Research Council, London, UK.
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
1461
Appendix A The final version of the Fast Alcohol Screening Test (FAST) is reproduced below. There are two slightly different methods of scoring. The first tends to result in a few more false positives and the second in a few more false negatives compared to AUDIT as the gold standard. Scoring is quick and can be completed with just a glance at the pattern of responses as follows: Question 1
FAST negative if response is Never FAST positive if response is Weekly/Daily or almost daily Only consider Q 2,3 &4 if the response to Q 1 is Less than monthly or Monthly
A.1. Scoring Method 1 Questions 2,3,4
FAST negative if responses to Q2, Q3 are Never and Q4 is No FAST positive if any other response, i.e., any hint of a problem
A.2. Scoring Method 2 Each question is scored 0 to 4 and a FAST positive is recorded if the total score for all four questions is 3 or more. Although scoring method 1 has been used in the current research the authors’ preference is for scoring method 2 since this results in slightly fewer false positives; sensitivity is reduced by 0.5% but specificity is increased by 2%. Appendix B. FAST For the following questions please circle the answer which best applies. 1 drink=1/2 pint of beer or 1 glass of wine or 1 single spirits. 1. MEN: How often do you have EIGHT or more drinks on one occasion? WOMEN: How often do you have SIX or more drinks on one occasion? Never
Less than monthly
Monthly
Weekly
Daily or almost daily
2. How often during the last year have you been unable to remember what happened the night before because you had been drinking? Never
Less than monthly
Monthly
Weekly
Daily or almost daily
3. How often during the last year have you failed to do what was normally expected of you because of drinking?
1462
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
Never
Less than monthly
Monthly
Weekly
Daily or almost daily
4. In the last year has a relative or friend, or a doctor or other health worker been concerned about your drinking or suggested you cut down? No
Yes, on one occasion
Yes, on more than one occasion Score Questions 1 to 3: 0, 1, 2, 3, 4. Score Question 4: 0, 2, 4.
References Allen, J. P., Litten, R. Z., Fertig, J. B., & Babor, T. (1997). Review of research on the Alcohol Use Disorders Identification Test (AUDIT). Alcoholism: Clinical and Experimental Research, 21(4), 613 – 619. Babor, T. F., de la Fuente, J. R., Saunders, J., & Grant, M. (1989). AUDIT: The Alcohol Use Disorders Identification Test: Guidelines for use in primary health care. Geneva, Switzerland: World Health Organization. Barry, K. L., & Fleming, M. F. (1993). Alcohol Use Disorders Identification Test (AUDIT) and the SMAST-13: predictive validity in a rural primary care sample. Alcohol and Alcoholism, 28(1), 33 – 42. Bohn, M. J., Babor, T. F., & Kranzler, H. R. (1995). Alcohol Use Disorders Identification Test (AUDIT): Validation of a screening instrument for use in medical settings. Journal of Studies on Alcohol, 56, 423 – 432. Bradley, K. A., Bush, K. R., McDonell, M. B., Maloine, T., & Fihn, S. D. (1998). Screening for problem drinking: Comparison of CAGE and AUDIT. Journal of General Internal Medicine, 13, 379 – 388. Brown, R. L., Leonard, T., Rounds, L. A., & Papasouliotis, O. (1997). A two-item screening test for abuse of alcohol and other drug problems. Journal of Family Practice, 44, 151 – 160. Chan, A. W. K., Pristach, E. A., & Welte, J. (1994). Detection by the CAGE of alcoholism or heavy drinking in primary care outpatients and the general population. Journal of Substance Abuse, 6, 123 – 135. Cherpitel, C. J. (2000). Brief screening instrument for problem drinking in the emergency room: The RAPS4. Journal of Studies on Alcohol, 61(3), 447 – 449. Hodgson, R. J., Alwyn, T., John, B., Thom, B., & Smith, A. (2002). The FAST alcohol screening test. Alcohol and Alcoholism, 37, 61 – 66. Isaacson, J. H., Butler, R., Zacharek, M., & Tzelepis, A. (1994). Screening with the Alcohol Use Disorders Identification Test (AUDIT) in an inner-city population. Journal of General Internal Medicine, 9(10), 550 – 553. Mayfield, D., McLeod, G., & Hall, P. (1974). The CAGE questionnaire: Validation of a new alcoholism instrument. American Journal of Psychiatry, 131, 1121 – 1123. Newcombe, R. G. (1998). Interval estimation for the difference between independent proportions: Comparison of eleven methods. Statistics in Medicine, 17, 873 – 890. Newcombe, R. G. (2001). Simultaneous comparison of sensitivity and specificity of two tests in the paired design: Straightforward graphical approach. Statistics in Medicine, 20, 907 – 915. Piccinelli, M., Tessari, E., Bortolomasi, M., Piasere, O., Semenzin, M., Garzotto, N., & Tansella, M. (1997). Efficacy of the Alcohol Use Disorders Identification Test as a screening tool for hazardous alcohol intake and related disorders in primary care: A validity study. British Medical Journal, 314, 420 – 424. Pokorny, A. D., Miller, B. A., & Kaplan, H. B. (1972). The brief MAST: A shortened version of the Michigan Alcoholism Screening Test. American Journal of Psychiatry, 129(3), 342 – 345. Russell, M., Martier, S. S., Sokol, R. J., Jacobson, S., & Bottoms, S. (1991). Screening for pregnancy risk drinking: TWEAKING the tests. Alcoholism: Clinical and Experimental Research, 15(2), 638.
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
1463
Schmidt, A., Barry, K. L., & Fleming, M. F. (1995). Detection of problem drinkers: The Alcohol Use Disorders Identification Test (AUDIT). Southern Medical Journal, 88(1), 52 – 59. Seppa, K., Lepisto, J., & Sillanaukee, P. (1998). Five-shot questionnaire on heavy drinking. Alcoholism: Clinical and Experimental Research, 22, 1788 – 1791. Smith, S. G., Touquet, R., Wright, S., & Das Gupta, N. (1996). Detection of alcohol misusing patients in accident and emergency departments: The Paddington Alcohol Test. Journal of Accident and Emergency Medicine, 5, 308 – 312. Soderstrom, C. A., Dischinger, P. C., Kerns, T. J., Kufera, J. A., McDuff, D. R., Gorelick, D. A., & Smith, G. S. (1998). Screening trauma patients for alcoholism according to NIAAA guidelines with Alcohol Use Disorders Identification Test Questions. Alcoholism: Clinical and Experimental Research, 22, 1470 – 1475. Volk, R. J., Steinbauer, J. R., Cantor, S. B., & Holzer, C. E. (1997). III. Alcohol Use and Disorders Identification Test (AUDIT) as screen for at-risk drinking in primary care patients of different racial/ethnic backgrounds. Addiction, 92(2), 197 – 206. Wilson, E. B. (1927). Probable inference, the law of succession, and statistical inference. Journal of the American Statistical Association, 22, 209 – 212.
Short Communication
Fast screening for alcohol misuse Ray J. Hodgsona,*, Bev Johna, Tina Abbasia, Rachel C. Hodgsonc, Seta Wallerb, Betsy Thomc, Robert G. Newcombea a
University of Wales College of Medicine, Cardiff Wales, UK b Health Education Authority, London, UK c Middlesex University, London, UK
Abstract The Fast Alcohol Screening Test (FAST) has been developed from the AUDIT questionnaire [Babor, T. F., de la Fuente, J. R., Saunders, J., Grant, M. (2001). AUDIT: The Alcohol Use Disorders Identification Test: guidelines for use in primary health care. Geneva, Switzerland: World Health Organization] for use in very busy medical settings [Alcohol Alcohol. 37 (2002) 61– 66]. One feature of the FAST is its ease and speed of administration, especially since one question identifies over 50% of patients as either alcohol misusers or not. This study further explores the sensitivity and specificity of the FAST across ages, gender, and locations using the AUDIT as the gold standard. Two other quick tests are also compared with the AUDIT and the FAST, namely the Paddington Alcohol Test [J. Accid. Emerg. Med. 5 (1996) 308] and the CAGE [Am. J. Psychiatry 131 (1974) 1121]. All tests were quicker to administer than the AUDIT with the FAST taking just 12 s on average. All tests identified drinkers who would accept a health education booklet (over 70% of those identified) or 5 min of advice (over 40%). The FAST was consistently reliable when sensitivity and specificity were tested against AUDIT as the gold standard. D 2002 Published by Elsevier Science Ltd. Keywords: Alcohol screening; Audit; Fast
1. Introduction A well-researched and frequently used screening test for alcohol misuse is the Alcohol Use Disorders Identification Test (AUDIT), which was developed in a World Health Organisation * Corresponding author. Tel.: +44-29-2023-3651; fax: +44-29-2023-7930. E-mail address: [email protected] (R.J. Hodgson). 0306-4603/$ – see front matter D 2002 Published by Elsevier Science Ltd. doi:10.1016/S0306-4603(02)00246-0
1454
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
(WHO) collaborative project across six countries. (Babor, de la Fuente, Saunders, & Grant, 1989). This 10-item questionnaire has good sensitivity and specificity as a screen for all types of alcohol misuse, i.e., hazardous drinking, harmful drinking, and dependence. Hazardous drinking refers to a pattern of drinking that is associated with a high risk of psychological or physical problems in the future. Harmful drinkers are already experiencing these problems. The dependent drinker is experiencing symptoms of dependence including impaired control or a subjective experience of compulsion to drink. Henceforth, the term alcohol misuse will be used cover all of these categories. Although AUDIT has been well validated, (Allen, Litten, Fertig, & Babor, 1997; Barry & Fleming, 1993; Bohn, Babor, & Kranzler, 1995; Bradley, Bush, McDonell, Maloine, & Fihn, 1998; Isaacson, Butler, Zacharek, & Tzelepis, 1994; Schmidt, Barry, & Fleming, 1995; Volk, Steinbauer, Cantor, & Holzer, 1997) and is turning out to be a very useful screening test, there are some situations, such as very busy medical settings, where a 10-item questionnaire takes too long to administer routinely. As a consequence a number of very brief tests have been developed. The range of instruments include the CAGE (Mayfield, McLeod, & Hall, 1974), the TWEAK (Russell, Martier, Sokol, Jacobson, & Bottoms, 1991), the brief MAST (Pokorny, Miller, & Kaplan, 1972), the RAPS (Cherpitel, 2000), the five-shot test (Seppa, Lepisto, & Sillanaukee, 1998), the short AUDIT (Piccinelli et al., 1997) and the PAT (Smith, Touquet, Wright, & Das Gupta, 1996). Furthermore, Soderstrom et al. (1998) use the first two Audit questions to assess quantity and frequency of alcohol consumption and the TICS (Brown, Leonard, Rounds, & Papasouliotis, 1997) attempts to assess both alcohol and drug misuse. The main focus of this study is upon the Fast Alcohol Screening Test (FAST) questionnaire, which was developed from the AUDIT, originally for use within accident and emergency (A&E) departments, but it has also been validated in a range of other settings (Hodgson, Alwyn, John, Thom, & Smith, 2002). The FAST correlates strongly with the short AUDIT (.92) but a significant feature of the FAST is the ability of one question to identify over 50% of patients as either alcohol misusers or not. This substantially reduces the time to administer. The FAST is based upon the following AUDIT questions: 1. How often do you have six or more drinks on one occasion? 2. How often during the last year have you been unable to remember what happened the night before because you had been drinking? 3. How often during the last year have you failed to do what was normally expected of you because of drinking? 4. Has a relative or friend, or a doctor or other health worker been concerned about your drinking or suggested you cut down? The current study extends previous work on the FAST by exploring the consistency of sensitivity and specificity, using the AUDIT as the standard, across age, gender, and location. This is not to suggest that the AUDIT is an unassailable gold standard. The aim is simply to compare a shorter questionnaire with one that has proved to be useful. A subsidiary aim is to investigate the relationship between the AUDIT, the FAST, and two other measures that are
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
1455
frequently used in the UK, namely the CAGE and the PAT. The CAGE questionnaire has been widely tested and shown to be useful in identifying the more dependent drinker (Chan, Pristach, & Welte, 1994). The Paddington Alcohol Test (PAT) has also been shown to be useful as a method of identifying problem drinking, particularly within the A&E setting (Smith et al., 1996). In the current study the FAST, the CAGE, and the PAT were compared to the AUDIT in four A&E departments. Both male and female patients between 16 and 75 were involved in the study. The main issues that will be addressed are: 1. Is the FAST questionnaire consistently reliable across age, gender, and location when compared to AUDIT as the gold standard? 2. Since the FAST uses AUDIT questions the CAGE and the PAT are not expected to be as strongly related to the AUDIT but these relationships will be explored. 3. The time taken to administer the four questionnaires will be assessed. 4. The readiness to accept advice will be assessed for those patients who screen positive on the four tests.
2. Methodology A total of 62 nurses were involved in the study. They were volunteers from four A&E departments in London, Southampton, Bristol, and Cardiff. All patients seen by triage nurses for the duration of the study were recruited if they gave informed consent and were not ruled out by the following exclusion criteria: (a) inability to read English; (b) severe intoxication whether by alcohol or other drugs; (c) excessive pain or confusion. After gaining consent the nurse administered a questionnaire, which included demographic variables. This was followed by one of the short screening tests, i.e., the CAGE, the FAST, or the PAT. The full AUDIT questionnaire was then administered, followed by questions about readiness to accept advice. Questionnaires were colour coded depending upon the particular shorter screening test that they included. The order in which the three colour coded questionnaires were administered was determined by random assignment within each A&E department. A random number table was used to order the questionnaires (1 to 3). They were then placed in a box in this order and nurses were asked always to take the top questionnaire.
3. Results A total of 2185 patients completed questionnaires at four centres, London (N=503), Southampton (N=852), Bristol (N=270) and Cardiff (N=560). The overall prevalence of alcohol misuse as defined by the AUDIT instrument in the study population was 843 out of 2175 (38.8%). This will not be representative of all A&E attenders since intoxicated patients were excluded from the study.
1456
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
4. Sensitivity and specificity All three tests evaluated were substantially correlated with AUDIT. Spearman rank correlations between AUDIT and FAST, PAT and CAGE were .79, .55, and .48, respectively. Table 1 displays the sensitivity and specificity of the FAST, the CAGE, and the PAT as predictors of a positive AUDIT in all subjects receiving the test in question. Confidence intervals (CIs) were calculated by the method of Wilson (1927). Table 2 displays sensitivity and specificity (%) by age, gender, and locality. The above results indicate that the FAST has high sensitivity and specificity when compared to the AUDIT as the gold standard. The CAGE has low sensitivity but very high specificity. The PAT has moderately high sensitivity and specificity. There was substantial evidence that sensitivity and specificity were affected by centre, age, and gender. Notably, the sensitivity and specificity of PAT varied significantly between the four centres (c2=12.15 and 36.94, df=3, P<.007 and P<.001, respectively), with also a gender difference in specificity (c2=15.34, df=1, P<.001). The specificity of FAST differed significantly between centres (c2=15.39, df=3, P<.002), and its sensitivity differed significantly between younger and older age groups (c2=7.97, df=1, P<.005). Nevertheless the sensitivity and specificity of FAST were high in all localities, in men and women, and in both age groups. Fig. 1 shows the comparison in sensitivity and specificity between FAST and CAGE as predictors of the AUDIT result. This diagram, based on Newcombe (2001), depicts the relative impact of false negatives and false positives of the two tests. It assumes that the consequences of a false negative result are more serious than those of a false positive result, by a factor r. A quantity f, expressing differences in sensitivity and specificity between the two tests, is plotted against a mixing parameter l, which runs from 0 to 1; f is calculated as ldelta sens+(1l)delta spec, a weighted average of delta sens and delta spec, which denote the differences in sensitivity and specificity between the two tests; l is designed to reflect the balance of two effects: the relative consequences of the two possible types of misclassification and the prevalence of significant alcohol misuse in a population. It is calculated as l ¼ 1þ 111 =r where p denotes the prevalence. If the missed opportunity ðp Þ Table 1 Sensitivity and specificity of FAST, CAGE, and PAT tests relative to AUDIT as gold standard Estimate
95% CI
FAST Sensitivity Specificity
269/290 92.8% 404/461 87.6%
89.2% to 95.2% 84.3% to 90.3%
CAGE Sensitivity Specificity
122/303 40.3% 422/432 97.7%
34.9% to 45.9% 95.8% to 98.7%
PAT Sensitivity Specificity
175/250 70.0% 371/439 84.5%
64.1% to 75.3% 80.8% to 87.6%
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
1457
Table 2 Sensitivity and specificity percentage for FAST, CAGE, and PAT tests by age, gender, and locality FAST Male Female Age25 Age>25 Bristol Cardiff London Southampton
CAGE
PAT
N
Sensitivity
Specificity
N
Sensitivity
Specificity
N
Sensitivity
Specificity
443 305 205 543 99 181 171 300
94 89 88 97 91 96 90 92
86 90 82 89 87 82 82 95
442 291 218 516 92 186 184 273
40 41 34 45 46 37 55 31
96 99 98 98 95 96 98 99
392 296 181 507 78 190 146 275
70 69 74 67 86 78 66 59
77 91 74 86 69 71 89 94
associated with a false positive is the dominant consideration, then r tends to infinity, l tends to 1 and f reduces to delta sens. Conversely, if the prevalence is sufficiently low, the penalty associated with the preponderance of false positives predominates: as p tends to 0, with r held finite, l tends to 0 and f reduces to delta spec. The difference in sensitivity (delta sens) is plotted at l=1 to the right of Fig. 1. It is 52.5% in favour of FAST, with 95% CI from 45.8% to 58.4%. The difference in specificity (delta spec) is 10.0% in favour of CAGE and appears at the left of the diagram as 10.0%, at l=0, together with its 95% CI from 13.5% to 6.8%. These CIs are calculated by method 10 of Newcombe (1998). Both CIs exclude zero, indicating that the two tests differ significantly in both sensitivity and specificity. Here, FAST should be regarded as better than CAGE at values of l for which the diagonal line is above 0. This is for all values of l greater than 0.16. For l values greater than 0.21, where the lower curve cuts the horizontal axis, FAST is significantly better than CAGE.
Fig. 1. Comparison of sensitivity and specificity between FAST and CAGE tests. For detailed explanation, see text.
1458
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
Fig. 2. Comparison of sensitivity and specificity between FAST and PAT tests.
Conversely, for l values of 0.16 or lower, CAGE is better than FAST, significantly so for l up to 0.11. In the current study, relatively high values of l are expected. This is because prevalence is high when screening for hazardous and problem drinkers in a medical setting. Also a false negative is more problematic than a false positive. A false positive is probably a borderline case and could possibly benefit from a health promotion intervention. If we consider a false negative to be twice as serious as a false positive then, for a 10% prevalence, l is 0.18. For 20% prevalence l=0.33 and for 30% prevalence l=0.46. These values of l are all above the cut off point at which FAST is better than CAGE. Fig. 2 shows the comparison in sensitivity and specificity between FAST and PAT as predictors of the AUDIT result. FAST has a large, highly significant advantage in sensitivity, and also an advantage in specificity that is somewhat short of statistical significance. FAST is better than PAT for all values of l, and significantly better than PAT for any plausible value of 0.07 or above.
5. Time, cost, and nurse appraisal Table 3 displays the mean time taken to administer each of the screening tests as well as the cost per year if 50,000 patients are routinely screened by nurses at a cost to the NHS Table 3 Mean time taken to administer each test and cost of routinely screening 50,000 patients FAST CAGE PAT AUDIT
N
Mean (S.D.) (s)
Cost per 50,000 patients (£)
67 64 48 191
12.52 (14.2) 14.37 (7.2) 42 (31.9) 78 (35.5)
1,669 1,916 5,600 10,400
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
1459
Table 4 Percentage of alcohol misusers, identified by a particular test, who would accept various interventions FAST CAGE PAT AUDIT
Booklet
5-Minute discussion
Half-hour discussion
Three counselling sessions
77 75 76 76
46 46 44 46
29 34 24 28
25 29 20 23
of £9.60 per hour (E grade nurse in UK). These data were collected in just one of the A&E departments. It should be noted that the current study explored the possibility of using PAT routinely whereas the creators of PAT advise that it should only be administered when medical problems that are known to be related to excessive alcohol use are identified. Twelve nurses were interviewed to ascertain whether any of the screening tests were preferred for any reason over any of the others. Ten nurses concluded that the FAST was the most appropriate screening test for use in a busy medical setting since it was not as intrusive as the CAGE and was quicker than the PAT. All 10 nurses concluded that the detailed calculation of units required for the PAT was the reason why it took longer to administer than the FAST or the CAGE. A key question is how many patients who are identified as alcohol misusers will accept advice. Table 4 displays the percentage of patients, identified as positive on each test, who said yes to the various forms of advice. These percentages did not differ significantly between the four tests.
6. Modifications to FAST Further research on the reliability of the FAST resulted in two very small changes that had no effect on the sensitivity or specificity of the test. The reasons for these changes are given in Hodgson et al. (2002). For men, question one was changed to eight drinks (units) instead of six (see Appendix A). The main advantage of the ‘‘eight drinks’’ version was the use of this question as a first filter for men. The ‘‘six drinks’’ question alone identified 56% of the men as hazardous or nonhazardous drinkers whereas the ‘‘eight drinks’’ question identified 63%. The accuracy of the test was not affected. Also a minor modification was made to the question: Has a relative or friend, or a doctor or other health worker been concerned about your drinking or suggested you cut down? In the AUDIT this particular question is not confined to the previous year and this could result in false positives if the concern occurred a few years ago. The question and the responses were therefore altered to focus upon the last year. In fact, this made no difference to any of the conclusions since only 5% of participants noted that concern had been expressed ‘‘but not in the last year’’. Furthermore, all remained in the same categories (alcohol misusers or not) if their response was altered to ‘‘never’’.
1460
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
7. Discussion It can be concluded that the FAST is a useful, very brief screening instrument when compared to AUDIT as the gold standard. The FAST has high sensitivity and specificity even though it is very quick to administer, and therefore cheap. A&E departments admit a high percentage of people who are misusing alcohol. By including hazardous drinking the FAST, the AUDIT, and the PAT are moving screening and brief interventions into the field of early intervention, health promotion, and prevention rather than just treatment. The FAST is particularly useful since over 50% of patients screened within a medical setting can be identified as either alcohol misusers or not by asking only the first FAST question. (Hodgson et al., 2002). Using the FAST, routine screening in an A&E department dealing with 50,000 patients per year would cost less than £2000 over the year. This does not take account of the extra costs involved in providing advice or a brief alcohol intervention. At low levels of prevalence the CAGE is probably better than the FAST. An example would be screening for more severe dependence. If prevalence is only 5% and both false positives and false negatives are equally serious then l is 0.05. At this value of l, CAGE is better than FAST. A point to be emphasised is that the usefulness of a particular screening instrument depends not only on its sensitivity and specificity, as well as prevalence, but also on the seriousness of a false negative relative to a false positive. When screening hazardous and problem drinkers to identify recipients of a brief intervention then a false negative is more serious than a false positive. The worst that can happen to those who are misclassified as hazardous drinkers is that they will receive or refuse a health promotion intervention. In a different situation, for example when more severely dependent drinkers are being screened for a long interview prior to treatment or admission to an alcoholism service, then a false positive is likely to be more serious since it will be associated with significant costs to both the individual and the service. The PAT questionnaire has a number of positive attributes that recommend its use in A&E departments. First, it is not intended to be used routinely as in the current study. It is to be used selectively for appropriate adult patients where there is a ‘‘a suspicion of alcohol misuse generated by presentations such as falls, assaults, head injuries, gastrointestinal problems, ‘unwell,’ fits, blackouts, collapse, insomnia, sweating, palpitations, chest pain, gout, rashes, depression,’’ etc. (Smith et al., 1996). Second, an attempt is made to get a very accurate estimate of peak consumption. Third, the test has been used in an A&E department for several years. This pioneering work of Dr. Touquet and his colleagues is at last beginning to generate interest in the whole field of screening and brief interventions within the A&E setting.
Acknowledgements This study was funded by the Alcohol Education & Research Council, London, UK.
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
1461
Appendix A The final version of the Fast Alcohol Screening Test (FAST) is reproduced below. There are two slightly different methods of scoring. The first tends to result in a few more false positives and the second in a few more false negatives compared to AUDIT as the gold standard. Scoring is quick and can be completed with just a glance at the pattern of responses as follows: Question 1
FAST negative if response is Never FAST positive if response is Weekly/Daily or almost daily Only consider Q 2,3 &4 if the response to Q 1 is Less than monthly or Monthly
A.1. Scoring Method 1 Questions 2,3,4
FAST negative if responses to Q2, Q3 are Never and Q4 is No FAST positive if any other response, i.e., any hint of a problem
A.2. Scoring Method 2 Each question is scored 0 to 4 and a FAST positive is recorded if the total score for all four questions is 3 or more. Although scoring method 1 has been used in the current research the authors’ preference is for scoring method 2 since this results in slightly fewer false positives; sensitivity is reduced by 0.5% but specificity is increased by 2%. Appendix B. FAST For the following questions please circle the answer which best applies. 1 drink=1/2 pint of beer or 1 glass of wine or 1 single spirits. 1. MEN: How often do you have EIGHT or more drinks on one occasion? WOMEN: How often do you have SIX or more drinks on one occasion? Never
Less than monthly
Monthly
Weekly
Daily or almost daily
2. How often during the last year have you been unable to remember what happened the night before because you had been drinking? Never
Less than monthly
Monthly
Weekly
Daily or almost daily
3. How often during the last year have you failed to do what was normally expected of you because of drinking?
1462
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
Never
Less than monthly
Monthly
Weekly
Daily or almost daily
4. In the last year has a relative or friend, or a doctor or other health worker been concerned about your drinking or suggested you cut down? No
Yes, on one occasion
Yes, on more than one occasion Score Questions 1 to 3: 0, 1, 2, 3, 4. Score Question 4: 0, 2, 4.
References Allen, J. P., Litten, R. Z., Fertig, J. B., & Babor, T. (1997). Review of research on the Alcohol Use Disorders Identification Test (AUDIT). Alcoholism: Clinical and Experimental Research, 21(4), 613 – 619. Babor, T. F., de la Fuente, J. R., Saunders, J., & Grant, M. (1989). AUDIT: The Alcohol Use Disorders Identification Test: Guidelines for use in primary health care. Geneva, Switzerland: World Health Organization. Barry, K. L., & Fleming, M. F. (1993). Alcohol Use Disorders Identification Test (AUDIT) and the SMAST-13: predictive validity in a rural primary care sample. Alcohol and Alcoholism, 28(1), 33 – 42. Bohn, M. J., Babor, T. F., & Kranzler, H. R. (1995). Alcohol Use Disorders Identification Test (AUDIT): Validation of a screening instrument for use in medical settings. Journal of Studies on Alcohol, 56, 423 – 432. Bradley, K. A., Bush, K. R., McDonell, M. B., Maloine, T., & Fihn, S. D. (1998). Screening for problem drinking: Comparison of CAGE and AUDIT. Journal of General Internal Medicine, 13, 379 – 388. Brown, R. L., Leonard, T., Rounds, L. A., & Papasouliotis, O. (1997). A two-item screening test for abuse of alcohol and other drug problems. Journal of Family Practice, 44, 151 – 160. Chan, A. W. K., Pristach, E. A., & Welte, J. (1994). Detection by the CAGE of alcoholism or heavy drinking in primary care outpatients and the general population. Journal of Substance Abuse, 6, 123 – 135. Cherpitel, C. J. (2000). Brief screening instrument for problem drinking in the emergency room: The RAPS4. Journal of Studies on Alcohol, 61(3), 447 – 449. Hodgson, R. J., Alwyn, T., John, B., Thom, B., & Smith, A. (2002). The FAST alcohol screening test. Alcohol and Alcoholism, 37, 61 – 66. Isaacson, J. H., Butler, R., Zacharek, M., & Tzelepis, A. (1994). Screening with the Alcohol Use Disorders Identification Test (AUDIT) in an inner-city population. Journal of General Internal Medicine, 9(10), 550 – 553. Mayfield, D., McLeod, G., & Hall, P. (1974). The CAGE questionnaire: Validation of a new alcoholism instrument. American Journal of Psychiatry, 131, 1121 – 1123. Newcombe, R. G. (1998). Interval estimation for the difference between independent proportions: Comparison of eleven methods. Statistics in Medicine, 17, 873 – 890. Newcombe, R. G. (2001). Simultaneous comparison of sensitivity and specificity of two tests in the paired design: Straightforward graphical approach. Statistics in Medicine, 20, 907 – 915. Piccinelli, M., Tessari, E., Bortolomasi, M., Piasere, O., Semenzin, M., Garzotto, N., & Tansella, M. (1997). Efficacy of the Alcohol Use Disorders Identification Test as a screening tool for hazardous alcohol intake and related disorders in primary care: A validity study. British Medical Journal, 314, 420 – 424. Pokorny, A. D., Miller, B. A., & Kaplan, H. B. (1972). The brief MAST: A shortened version of the Michigan Alcoholism Screening Test. American Journal of Psychiatry, 129(3), 342 – 345. Russell, M., Martier, S. S., Sokol, R. J., Jacobson, S., & Bottoms, S. (1991). Screening for pregnancy risk drinking: TWEAKING the tests. Alcoholism: Clinical and Experimental Research, 15(2), 638.
R.J. Hodgson et al. / Addictive Behaviors 28 (2003) 1453–1463
1463
Schmidt, A., Barry, K. L., & Fleming, M. F. (1995). Detection of problem drinkers: The Alcohol Use Disorders Identification Test (AUDIT). Southern Medical Journal, 88(1), 52 – 59. Seppa, K., Lepisto, J., & Sillanaukee, P. (1998). Five-shot questionnaire on heavy drinking. Alcoholism: Clinical and Experimental Research, 22, 1788 – 1791. Smith, S. G., Touquet, R., Wright, S., & Das Gupta, N. (1996). Detection of alcohol misusing patients in accident and emergency departments: The Paddington Alcohol Test. Journal of Accident and Emergency Medicine, 5, 308 – 312. Soderstrom, C. A., Dischinger, P. C., Kerns, T. J., Kufera, J. A., McDuff, D. R., Gorelick, D. A., & Smith, G. S. (1998). Screening trauma patients for alcoholism according to NIAAA guidelines with Alcohol Use Disorders Identification Test Questions. Alcoholism: Clinical and Experimental Research, 22, 1470 – 1475. Volk, R. J., Steinbauer, J. R., Cantor, S. B., & Holzer, C. E. (1997). III. Alcohol Use and Disorders Identification Test (AUDIT) as screen for at-risk drinking in primary care patients of different racial/ethnic backgrounds. Addiction, 92(2), 197 – 206. Wilson, E. B. (1927). Probable inference, the law of succession, and statistical inference. Journal of the American Statistical Association, 22, 209 – 212.