Alcohol Detection Services, LLC, 4670 Somerset Court, Brookfield, WI 53045 and
1 Millennium Strategies, 418 N. Westfield Road, Madison, WI 53717, USA
Received 18 September 2000; in revised form 12 February 2001; accepted 23 February 2001
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
The ineffectiveness of traditional markers to screen for alcohol consumption in the general population has been recognized for many years. Sensitivity and/or specificity rates are far too low to propose their systematic use as screening tests in unselected medical populations. A biochemical marker with 60% sensitivity and 98% specificity rate for heavy drinking, when applied to a population with a 7% prevalence of alcohol misuse, has a positive predictive value of 0.66. This means that, if a patient has a positive test result, there is a 66% chance that this patient is a heavy drinker and a 34% chance that this patient is a false positive, rather than a true positive. Despite moderate sensitivity and specificity rates, this biochemical marker is not a good screening candidate. Tests with moderate diagnostic performance have serious drawbacks in employment, legal and insurance settings where false positive results can have serious consequences. However, if false positive results could be eliminated or greatly reduced, these tests might find greater acceptance and use in these settings.
Two ways to improve assay performance are to increase the sensitivity and specificity rates (diagnostic accuracy) of the test, or to increase disease prevalence, i.e. to use the biochemical marker in a population with a higher proportion of heavy drinkers. To increase diagnostic accuracy, several recent studies have used parallel testing by combining two or more laboratory tests to identify alcohol misuse (Sillanaukee, 1992; Hartz et al., 1997
; Harasymiw et al., 2000a
). For instance, parallel testing using CDT and
-glutamyltransferase (GGT) has been widely documented to improve the performance of either marker used alone (Anton and Moak, 1994
; VanPelt et al., 2000
). Similarly, a recent study showed the effectiveness of parallel testing using CDT and the Alcohol Use Disorders Test (AUDIT) to screen for alcohol use disorders in a routine workplace (Hermansson et al., 2000
). CDT and mean corpuscular volume (MCV) have also been used jointly in parallel testing with optimal success in females.
Statistical methods that use combinations of 10 to 40 routine laboratory tests, such as the EDAC test, constitute the essence of a different combinatorial strategy (Hartz et al., 1997; Bean, 2000
; Harasymiw et al., 2000a
). In this instance, laboratory tests are used as independent variables to predict group membership of the subjects being analysed. The widespread use of these statistical methods was limited in the past due to the archaic hardware and software packages. Both unavailable at the time to perform the statistical calculations needed to analyse and interpret the data. New developments in technology and the availability of simpler statistical packages that allow data analysis in any personal computer are changing the scope of alcohol misuse diagnosis. Linear discriminant function and logistic regression are now simple formulae that can be performed very rapidly at any laboratory, hospital or medical office with minimum training and at affordable costs.
The EDAC method uses linear discriminant function (LDF) to analyse a battery of routine laboratory tests that generate a score for each subject (Johnson and Wichern, 1998). Each score and its associated probability value translate into the likelihood of that individual being a heavy drinker or a light drinker. For instance, if the EDAC score is positive and the probability for heavy drinking (P1) is in the range of 51100%, then the individual is classified as a heavy drinker. Alternatively, if the EDAC score is negative and the probability value (P1) is <50%, then the individual is classified as a non-heavy drinker. The latter individual will bear a probability value in the range of 51100% for a non-drinker (P2).
CDT is the newest procedure available to clinicians to assess harmful alcohol consumption and the first one to obtain approval from the US Food and Drug Administration (FDA) for identification of sustained alcohol consumption. In contrast, GGT, which is also FDA-approved, is intended to detect liver damage rather than alcohol misuse. CDT uses ion-exchange chromatography to separate CDT from other transferrin isoforms in serum and quantifies CDT by immunological methods. EDAC uses LDF. Since the EDAC and CDT tests detect alcohol misuse by very different methods, we thought that a combination of these methods might enhance the ability to accurately assess alcohol misuse. In this study, we determined the efficacy of the combined use of CDT and the EDAC test in a population of males (n = 187) drinking an average of 20 drinks per day, recruited from several institutions in the midwestern USA. Heavy drinkers (n = 138) and light drinkers (n = 49) were analysed in three ways: heavy drinkers and light drinkers identified by the EDAC alone, heavy drinkers and light drinkers identified by CDT alone, and heavy drinkers identified by the EDAC and CDT combined.
![]() |
SUBJECTS AND METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
To assess alcohol consumption we used the Khavari Alcohol Test (Khavari and Farber, 1978). The heavy drinkers drank an average of 248 ± 153 g of alcohol per day (range 51879 g) corresponding to a mean of 20.6 drinks per day. The 95% confidence interval (CI) for this group was 223272 g daily. The light drinkers drank on average 1.8 ± 7.1 g of alcohol daily (range 045 g); the 95% confidence interval was 03.7 g daily.
Laboratory analysis
Blood specimens (non-fasting) were obtained from each enrolled participant and sent to the laboratory for biochemical and haematological tests. Handling of specimens followed standard laboratory procedures, with daily collection by courier and overnight transportation to the testing site. Two tubes of blood were drawn for each patient, one for the blood counts and one for the chemistry panel. The tube for the chemistry panel was allowed to clot and serum separated after centrifugation at 2000 r.p.m. for 10 min. Serum was kept refrigerated until sent to the laboratory for analysis. Chemistry determinations were performed on standard equipment utilizing the Olympus AU5000 automated chemistry analyser, whereas haematology parameters were determined utilizing the Argos Cobas haematology instrument (LabCorp, Burlington, NC, USA).
The CDT test
Frozen serum specimens were shipped to Dr Sillanaukee to perform the CDT analysis. One tube of serum from each subject was thawed and subjected to the CDTect procedure (Kabi-Pharmacia, Uppsala, Sweden) as described previously (Stibler et al., 1991). Briefly, 50 µl of serum were pre-incubated with 200 µl of ferric citrate and 1 ml of elution buffer. The kit mini-columns were reconstituted and equilibrated with 2 ml of elution buffer. To each mini-column, 500 µl of iron-saturated serum sample was added and the eluate was collected and quantified by radioimmunoassay (RIA). The eluate contained CDT molecules with two, one or no sialic acid residues. The RIA was performed using rabbit anti-human transferrin antibody to bind CDT and sheep anti-rabbit antibody to precipitate the CDTantibody complex. After centrifugation, the radioactivity of the precipitate was counted in a gamma counter. The amount of CDT in duplicate samples was calculated from a 5-point calibration curve derived from the displacement of radioactive CDT from the antibody by known amounts of human transferrin. The cut-off point used was 20 arbitrary units per litre (~20 mg/l).
The EDAC test
The EDAC test uses LDF to build a predictive model of group membership, based on observed characteristics of two different samples, in this case laboratory parameters of heavy drinkers and light drinkers. The LDF analysis generates a discriminant function based on linear combinations of the predictor variables that provide the best discrimination between the two groups (Johnson and Wichern, 1998). The functions are generated from the results of the laboratory variables and demographic parameters of subjects for which group membership (heavy drinkers versus light drinkers) is known. Discriminant analysis looks like the right-hand side of a multiple linear regression equation; using coefficients a, b, c, d ... x, for the laboratory variables of the EDAC panel and demographic parameters such as gender and age, the function is:
![]() |
The D value obtained for each subject corresponds to a probability of the subject being a heavy or a light drinker, based on the subject's alcohol consumption according to self-report. In this study the prior probabilities for a subject's being a heavy drinker or a light drinker were set equal and the costs of misclassification for the two categories were also set equal. Therefore, we assumed equal costs associated with false positives and false negatives, because the prior probabilities for heavy and light drinking were set at 0.5 and 0.5, respectively. The cut-off point for the EDAC was set at a value of zero, with positive EDAC scores identifying heavy drinkers and negative EDAC scores identifying light drinkers. The derivation and validation of the discriminant equation used in this analysis has been described recently (Harasymiw and Bean, 2001).
The laboratory variables used to obtain the EDAC score were: albumin/globulin ratio, globulin, total protein, creatinine, blood urea nitrogen, blood urea nitrogen/creatinine ratio, alkaline phosphatase, triglycerides, total bilirubin, direct-reacting bilirubin, low density lipoprotein, uric acid, neutrophils, monocytes, lymphocytes and basophils.
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
Individuals identified as heavy or light drinkers by CDT
The second objective of this study was to evaluate the use of CDT when used alone in this same population. The CDT test showed a sensitivity rate of 58% (80/138) for identifying heavy drinkers and a corresponding specificity rate of 96% as it identified 47 of the 49 light drinkers (Table 2). The false negative rate was 42% with 58 of the 138 self-reported heavy drinkers showing CDT values below the cut-off point. The false positive rate was 4% with two light drinkers showing positive CDT results.
|
The combined use of EDAC and CDT
The analysis of the combined use of EDAC and CDT was done in two ways. First, we assumed the clinician would have access to analyse the results from both tests simultaneously. The criteria for diagnosis were as follows. If both CDT and EDAC tests were positive or if only one of them was positive, then the clinician would diagnose heavy drinking. Only when both tests were negative, did the clinician assume lack of heavy drinking. When EDAC and CDT were analysed simultaneously the overall sensitivity was 92%, which means that 127 of the 138 heavy drinkers by self-report show at least one positive result for EDAC and/or CDT (Table 3). The corresponding specificity rate was 94% with 46 of the 49 light drinkers showing negative results for both tests (Table 3
). The PPV for this combination of markers when used in a population with 10% prevalence of alcohol misuse was 0.63 (92/146), similar to the one obtained when using only CDT.
|
|
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Similar to most biochemical markers, additional sources of variation for the EDAC test include differences in the amount, duration and type of alcohol consumed, subjects' drinking patterns and nutritional or health characteristics that affect individual responses to alcohol. The analysis of a larger population of subjects representing less extreme groups showed that the EDAC worked best when classifying populations according to gender and age (Harasymiw et al., 2000b).
The CDT test showed a low PPV when used hypothetically as a screening tool even though this biochemical marker has been tested successfully for its ability to detect heavy drinking in males. A well-known drawback of the CDT kit used in this study relates to the fact that it measures absolute CDT concentrations without taking into consideration the contribution of total serum transferrin. In this scenario, low serum transferrin could lead to false negative results for a poor overall performance of the test. Newer CDT tests with better performance, such as the %CDT Turbidimetric immunoassay (Axis Shield ASA, Norway) recently approved by the FDA may be better-suited for generalized use.
Aside from optimizing the sensitivity and/or specificity rates of any single test, another way to increase the positive predictive value is by combining markers in either parallel testing or reflex testing. Biomarkers that are poorly correlated should be selected for parallel testing and biomarkers that are highly correlated should be selected for reflex testing. For instance, AUDIT and CDT both have value for identifying a different segment of the high risk drinking population and, therefore, have been considered as complementary instruments for parallel testing during alcohol screening (Hermansson et al., 2000). In contrast, the insurance industry in the USA uses reflex testing when it screens its applicants using high density lipoprotein (HDL)-cholesterol and liver enzyme tests. Samples from subjects with abnormal HDL undergo further testing by CDT and subjects with abnormal liver enzymes undergo further testing by haemoglobin-associated acetaldehyde (HAA). Thus, CDT and HAA are used as reflex tests (Daniel, 1997
). When using reflex testing, the medical professional needs to combine tests that are highly correlated with each other so that a subject who tests positive in the screen would also test positive in the reflex test.
In our study, we combined EDAC and CDT in two ways. When both tests were used in parallel, sensitivity increased, but specificity decreased, compared to the use of each marker alone. This loss of specificity resulted in a concomitant decrease in PPV to resemble the one obtained when using only CDT. In contrast, when the EDAC-positive samples were submitted to reflex testing by the CDT test, we maximized specificity as evident by the elimination of all false positives. This preliminary study shows that EDAC and CDT may react independently to alcohol intake and that they can be combined for maximum diagnostic accuracy.
Alcohol misuse is a major factor in morbidity and mortality worldwide. It was estimated in the early 1990s that in the USA alone the costs associated with the consequences of alcohol misuse surpassed $148 billion per year (Lewin Group, 1992). Rates of frequent heavy drinking and alcohol-related problems between 1984 and 1995 remained especially high among African American and Hispanic men, suggesting that men of these two ethnic groups should be specifically targeted for renewed prevention efforts (Caetano, 2000
). However, diagnosis and treatment usually occur only after the onset of either physical morbidity or a personal crisis resulting from mounting psychosocial problems (Bucholz et al., 1992
). Most hazardous and harmful drinkers can be identified with simple verbal screening, but some will not acknowledge heavy drinking. Denials occur more frequently in settings where there may be negative consequences due to diagnosis. Indeed, misuse of alcohol creates many situations in which individuals may find themselves in conflict with their employers, spouses or the courts. The costs associated with lost productivity, family distress and legal problems are substantial (Lewin Group, 1992
). A cost-effective biological marker that is also accurate in identifying heavy alcohol consumption would facilitate timely assessment of the patients and promote earlier interventions for optimal clinical benefits and treatment results.
Based on this study, the EDAC test may represent a practical candidate as a screening tool, because of the broad availability of the routine laboratory panel, optimal performance and reduced costs. Ongoing studies are currently evaluating CDT as an additional independent predictor variable in the discriminant analysis. The addition of CDT to the EDAC panel may further improve the performance of the EDAC and it may allow the use of a reduced number of routine laboratory tests.
![]() |
ACKNOWLEDGEMENTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
FOOTNOTES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
REFERENCES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Bean, P. (1999) Carbohydrate-deficient transferrin in the assessment of harmful alcohol consumption: diagnostic performance and clinical significance. Addiction Biology 4, 151161.[ISI]
Bean, P. (2000) Choosing the proper laboratory procedure to diagnose alcohol abuse in the new millennium. American Clinical Laboratory 19, 610.
Bucholz, K. K., Homan, S. M. and Helzer, J. E. (1992) When do alcoholics first discuss drinking problems? Journal of Studies on Alcohol 53, 582589.[ISI][Medline]
Caetano, R. (2000) Drinking patterns and alcohol problems in the U.S. household population. Presented at the symposium Alcohol use and HIV pharmacotherapy, National Institute of Alcohol Abuse and Alcoholism, Bethesda, MD, April 2628, 2000.
Conigrave, K. M., Saunders, J. B. and Whitfield, J. B. (1995) Diagnostic tests for alcohol consumption. Alcohol and Alcoholism 30, 1326.[Abstract]
Daniel, P. (1997) Correlation of results of two alcohol marker tests: CDT and HAA, with other laboratory test results. On the Risk 13, 6771.
Harasymiw, J. and Bean, P. (2001) Identification of heavy drinkers by using the Early Detection of Alcohol Consumption score. Alcoholism: Clinical and Experimental Research 25, 228235.[ISI][Medline]
Harasymiw, J., Vinson, D. and Bean, P. (2000a) The early detection of alcohol consumption score in the identification of heavy and at-risk drinkers from routine blood tests. Journal of Addictive Diseases 19, 4359.
Harasymiw, J., Soofi, E. and Bean, P. (2000b) Identification of heavy drinkers by the EDAC score based on gender, race and age. Presented at the 31st Annual Medical-Scientific Conference of the American Society of Addiction Medicine, April 1316, 2000, Chicago, IL, USA (abstract # 19).
Hartz, A. J., Guse, C. and Kajdacsy-Balla, A. (1997) Identification of heavy drinkers using a combination of laboratory tests. Journal of Clinical Epidemiology 50, 13571368.[ISI][Medline]
Hermansson, U., Helander, A., Huss, A., Brandt, L. and Ronnberg, S. (2000) The alcohol use disorders identification test (AUDIT) and carbohydrate-deficient transferrin (CDT) in a routine workplace health examination. Alcoholism: Clinical and Experimental Research 24, 180187.[ISI][Medline]
Johnson, R. A. and Wichern, D. W. (1998) Applied Multivariate Statistical Analysis, 4th edn, pp. 116128. PrenticeHall, New Jersey.
Khavari, K. and Farber, P. (1978) A profile instrument for the quantification and assessment of alcohol consumption. Journal of Studies on Alcoholism 39, 15251539.[ISI][Medline]
Lewin Group (1992) National Institute on Alcohol Abuse and Alcoholism. The economic costs of alcohol and drug abuse in the United States 1992. NIH Publication No. 98-4327, September 1997.
Sillanaukee, P. (1992) The diagnostic value of a discriminant score in the detection of alcohol abuse. Archives of Pathology and Laboratory Medicine 116, 924929.[Medline]
Stibler, H., Borg, S. and Joustra, M. (1991) A modified method for the assay of carbohydrate-deficient transferrin in serum. Alcohol and Alcoholism 26 (Suppl. 1), 451454.[ISI][Medline]
VanPelt, J., Leusink, G. L., vanNierop, P. W. M. and Keyzer, J. J. (2000) Test characteristics of carbohydrate-deficient transferrin and -glutamyltransferase in alcohol-using perimenopausal women. Alcoholism: Clinical and Experimental Research 24, 176179.[ISI][Medline]