bs_bs_banner
RESEARCH REPORT
doi:10.1111/add.12372
Usefulness of indirect alcohol biomarkers for predicting recidivism of drunk-driving among previously convicted drunk-driving offenders: results from the Recidivism Of Alcohol-impaired Driving (ROAD) study Thomas M. Maenhout1, Anneleen Poll2, Tijl Vermassen1, Marc L. De Buyzere1, Joris R. Delanghe1 & the ROAD Study Group* Department of Clinical Chemistry, Ghent University Hospital, Ghent, Belgium1 and Dienst herstelonderzoeken, Belgian Institute of Road Safety, Brussels, Belgium2
ABSTRACT Aim In several European countries, drivers under the influence (DUI), suspected of chronic alcohol abuse are referred for medical and psychological examination. This study (the ROAD study, or Recidivism Of Alcohol-impaired Driving) investigated the usefulness of indirect alcohol biomarkers for predicting drunk-driving recidivism in previously convicted drunk-driving offenders. Design, setting, participants and measurements The ROAD study is a prospective study (2009–13) that was performed on 517 randomly selected drivers in Belgium. They were convicted for drunk-driving for which their licence was confiscated. The initial post-arrest blood samples were collected and analysed for percentage carbohydrate-deficient transferrin (%CDT), transaminsase activities [alanine amino transferase (ALT), aspartate amino transferase (AST)], gamma-glutamyltransferase (γGT) and red cell mean corpuscular volume (MCV). The observation time for each driver was 3 years and dynamic. Findings A logistic regression analysis revealed that ln(%CDT) (P < 0.001), ln(γGT) (P < 0.01) and ln(ALT) (P < 0.05) were the best biochemical predictors of recidivism of drunk-driving. The ROAD index (which includes ln(%CDT), ln(γGT), -ln(ALT) and the sex of the driver) was calculated and had a significantly higher area under the receiver operator characteristic curve (0.71) than the individual biomarkers for drunk-driving recidivism. Drivers with a high risk of recidivating (ROAD index ≥ 25%; third tertile) could be distinguished from drivers with an intermediate risk (16% ≤ ROAD index < 25%; second tertile; P < 0.001) and a low recidivism risk (ROAD index < 16%; first tertile; P < 0.05). Conclusions Of all routinely used indirect alcohol markers, percentage of carbohydrate-deficient transferrin is the major predictor of recidivism of drunk-driving. The association with gamma-glutamyltransferase, alanine amino transferase and the sex of the driver could have additional value for identifying drunk-drivers at intermediate risk of recidivism. Non-specific indirect alcohol markers, such as alanine amino transferase, gamma-glutamyltransferase, aspartate amino transferase and red cell mean corpuscular volume have minimal added value to % carbohydrate-deficient transferrin for distinguishing drunk drivers with a low or high risk of recidivism. Keywords Alcoholism, carbohydrate-deficient transferrin, driver’s licence, drunk-driving, liver disease, logistic regression, transferrin. Correspondence to: Joris R. Delanghe, Department of Clinical Chemistry, Ghent University Hospital, De Pintelaan 185, 9000 Ghent, Belgium. E-mail: [email protected] Submitted 29 March 2013; initial review completed 15 May 2013; final version accepted 27 September 2013
*ROAD study group collaborators: P. Maenhout (Labo Maenhout), K. Van Poucke (Labo Van Poucke), C. Neven (LKO-LMC) and W. Top (Labo Medina). © 2013 Society for the Study of Addiction
Addiction, 109, 71–78
72
Thomas M. Maenhout et al.
INTRODUCTION Recidivism of driving under the influence (DUI) of alcohol among drunk-driving offenders is a serious concern. Repeat offenders are more likely to be involved in alcohol-related crashes than first-time offenders [1,2]. It has been shown that the rate of repeat DUI offences is greater among drunk drivers who are diagnosed with an alcohol use disorder (AUD) and report daily excessive drinking [3]. When compared to the general population, this is in correlation with the higher prevalence rates of DSM-IV alcohol dependence among drunk drivers [4]. While repeat offenders represent only a small proportion of all drivers, they contribute disproportionately to road accidents [5,6]. Predicting the risk of a repeat offence, therefore, is an important consideration for making re-licensing decisions. Re-granting a driver’s licence to a high-risk offender could pose a significant threat to the safety of public roadways [6]. In several European countries, driver’s licence re-granting programmes are used, where apprehended drivers, suspected of an AUD, are referred for medical and psychological examination [7]. Sweden, for example, has a medical review programme for DUI offenders that links reductions of alcohol markers to successful completion of their interlock programme [8]. In Belgium, a judge can propose the participation to a driver’s licence re-granting programme as a subsequent condition in order to restore the driver’s licence. Next to extensive psychological and clinical investigations, biomarkers are used to monitor the driver’s adherence to the abstinence programme [9]. Re-licensing decisions are often based on subjective expert opinions which take into consideration often poorly standardized clinical and laboratory findings.
Many subjects misusing alcohol can be identified based on clinical history and examination, combined with selfreport questionnaires; however, sensitivity is not optimal, with deliberate under-reporting being common [10]. Alcohol biomarkers are often suggested as objective measures and can give additional information, next to clinical and psychological investigation, for identifying those drunk-driving offenders showing a chronic alcohol problem, thereby identifying those drivers most at risk for recidivism of drunk-driving [5]. However, the accuracy of biomarkers used for substantiating the diagnosis of an AUD in drunk-drivers is difficult to assess, and until now there has been scarce evidence about their power to predict a drunk-driving relapse [2,11]. In the present study, we investigate the usefulness of currently widespread available indirect alcohol state markers for the prediction of recidivism of drunkdriving. Indirect alcohol markers are of particular interest for identifying chronic alcohol abuse because of their serum half-life ranging from 2 to 3 weeks, and the requirement of an excessive average daily alcohol consumption before elevation [5,10]. In Belgium, carbohydrate-deficient transferrin (CDT) testing was introduced in 2008 within the framework of driver’s licence re-instatement. It is considered the most accurate indirect biomarker for identifying sustained heavy alcohol consumption and for monitoring abstinence [12,13]. Next to CDT, the most frequently used indirect biomarkers for the detection and monitoring chronic alcohol abuse are γ-glutamyltransferase (γGT), alanine amino transferase (ALT), aspartate amino transferase (AST) and mean corpuscular volume (MCV) (Table 1). Combinations of several markers of alcohol abuse are thought to give better sensitivity, but usually show
Table 1 Description of biochemical laboratory parameters included in the Recidivism Of Alcohol-impaired Driving (ROAD) study.
AST
ALT MCV
Type of marker
Remarks
t1/2 (days)
Cytoplasmic (mitochondrial) liver enzyme Cytoplasmic liver enzyme Calculated descriptive of red blood cell volume
May be elevated in obesity, liver and biliary conditions, muscle diseases, extreme exertion
13 ± 20 [22]
May be elevated in obesity, liver and biliary conditions May be elevated in folate or B12 deficiency (including malabsorption), bleeding (leads to reticulocytosis), haematological conditions (including haemolysis, haemoglobinopathies), bone marrow disorders, liver disease, hypothyroidism, hyperglycaemia May be elevated in obesity, liver and biliary conditions (including hepatic congestion), hypertriglyceridaemia, diabetes, pancreatitis May be elevated in liver cirrhosis. Analytical difficulties encountered in transferrin variants, di-trisialo-bridging phenomena
16 ± 19 [22] NA (typical life-span of a red blood cell is approximately 120 days)
γGT
Cell surface liver enzyme
CDT
Deregulation of glycoconjugate metabolic processes
16 ± 8 [22] 16 ± 11 [7]
CDT: carbohydrate-deficient transferrin; γGT: gamma-glutamyltransferase; MCV: mean corpuscular volume; ALT: alanine amino transferase; AST: aspartate amino transferase; NA: not applicable.
© 2013 Society for the Study of Addiction
Addiction, 109, 71–78
Biomarkers and recidivism of drunk-driving
reduced specificity for detecting alcohol abuse. Despite intensive investigation, there is still no consensus on which laboratory biomarkers are satisfactory in predicting surreptitious alcohol ingestion. In 2009, the ROAD study (Recidivism Of Alcohol-impaired Driving) was set up with its major purpose investigating the predictive value of indirect alcohol biomarkers for recidivism of drunk-driving in drivers admitted to the driver’s licence re-granting programme. Being surrogate markers for chronic alcohol abuse, the predictive power of each different biomarker separately and combinations of different biomarkers for recidivism of drunk-driving will be investigated.
MATERIALS AND METHODS Study design The ROAD study is a prospective study that was performed on 517 randomly selected drunk-driving offenders who were admitted to the Belgian driver’s licence re-granting programme under the control of the Belgian Institute for Road Safety (BIVV; Belgisch Instituut voor Verkeersveiligheid—IBSR; Institut Belge pour la Sécurité Routière) in 2009. The observation time was 3 years and dynamic for each driver, which means that each driver had a time-window of 3 years for an actual relapse event. At first admittance, blood samples were collected in duplicate [one whole blood on K3ethylenediamine tetraacetic acid (EDTA) and one serum sample]. In 14 drivers, CDT could not be quantified accurately due to the presence of a genetic transferrin (Tf) variant. These cases were excluded from the analysis. A total of 108 drivers were included in the subgroup of recidivists of drunk-driving. They were arrested for a subsequent drunk-driving offence within the 3-year observation period, based on established legal limits (0.05‰ according to Belgian legislation) for driving while impaired or intoxicated with alcohol. This criterion for drunk-driving recidivism was adapted from Nochajski & Stasiewicz [14]. Drivers not meeting these criteria were included in the control group (n = 395). The study was approved by the Ethics Committee of Ghent University Hospital (EC/2012-516). Analytical methods—alcohol markers Whole blood (using K3EDTA as anticoagulant) was used to determine a complete blood count, including MCV, using an automated haematology analyser (Sysmex XE-5000). Serum was separated by centrifugation and was assayed for %CDT, γGT, ALT and AST activity. γGT activity was measured using a kinetic spectrophotometric assay (405 nm) with carboxynitroanilide as a substrate. ALT and AST activities were measured using the kinetic © 2013 Society for the Study of Addiction
73
ultraviolet (UV) spectrophotometric International Federation for Clinical Chemistry and Laboratory Medicine (IFCC) method. Serum concentrations of %CDT (i.e. percentage asialo- and disialo-Tf of the total Tf) were measured using capillary zone electrophoresis (CZE) [9], performed on the Capillarys 2™ system (Sebia, Evry, France). In this technique, after on-line iron saturation, samples are submitted to high-voltage (8200 V) zone electrophoresis in alkaline buffer (pH 8.8). Tf glycoforms are quantified by their peptide bond absorbance at 200 nm. Pherograms were validated using the manufacturer’s software [15]. Statistical analysis A logistic regression analysis was performed with different indirect alcohol biomarkers (%CDT, γGT, AST, ALT and MCV), age and sex of the driver on recidivism of drunk-driving as the dependent variable. Intentionally, no psychometric measures or intervention variables were taken into account to investigate independently the predictive value of biomarkers. The inserted variables are test results from the initial post-DUI arrest blood samples of the driver. Standard diagnostic methods for linear statistical models suggest that %CDT, γGT, MCV, AST and ALT should be analysed in logarithmic form [16]. The natural logarithm was used. We have performed an initial forward stepwise logistic regression analysis to determine which variables accounted for most of the variance when order of entry is not dictated. Afterwards, we performed a backward stepwise logistic regression. The regression equation from the latter was used to calculate the predictive index (PI). The ROAD index was calculated from the PI (ROAD index = exp(PI)/[1 + exp(PI]). To minimize the included variables in order to simplify the regression equation, a data reduction by principle component analysis was performed, and as a result serum AST activity was removed from the analysis, due to the high correlation (Pearson’s R2 = 0.75) with serum ALT activity. The Wald test was used to evaluate the contribution of each individual predictor to the model. A predictor was entered into the regression equation when the probability (P) was 0.05. Additionally, a logistic regression analysis was performed on dichotomized data to evaluate routinely applied decision limits for each indirect alcohol biomarkers. The latter were based on internationally accepted upper 97.5th limits of normal (ULN): for males, 1.6% for CDT, 61 U/L for γGT, 37 U/L for AST, 40 U/L for ALT and 96.4 fL for MCV; for females, 1.6% for CDT, 36 U/L for γGT, 31 U/L for AST, 31 U/L for ALT and 96.4 fL for MCV [9,17,18]. For the comparisons between different classification criteria, receiver operator characteristic (ROC) plots Addiction, 109, 71–78
74
Thomas M. Maenhout et al.
Table 2 Descriptive statistics of the data set used for logistic regression. Results are expressed as median and interquartile range.
Subgroupa
Women Age at start Men (n) (n) study (years) CDT (%)
Control group 330 (n = 395) Drunk-driving 100 recidivist (n = 108) Overall 430 (n = 503)
ROAD index (%)
γGT (U/L)
18 (3–67)
35 (21–68) 24 (19–30) 25 (29–35) 93 (90–96)
AST (U/L)
ALT (U/L)
MCV (fL)
65
46 (36–53)
1.0 (0.8–1.5)
8
46 (37–50)
1.9* (0.9–2.8) 27* (4–65) 39 (22–74) 23 (19–33) 22 (17–33) 94 (91–98)
73
46 (36–53)
1.0 (0.8–1.8)
19 (3–67)
40 (21–69) 24 (19–31) 24 (19–35) 93 (90–97)
a The population was divided into two subgroups: when the driver was convicted more than once in 3 years following initial conviction, the subject was placed into the ‘recidivist’ subgroup; otherwise the subjects were considered as controls. *P < 0.001 for Mann–Whitney test for independent samples compared to control group. % CDT: percentage carbohydrate-deficient transferrin; γGT: gamma-glutamyltransferase; MCV: mean corpuscular volume; ALT: alanine amino transferase; AST: aspartate amino transferase; ROAD: Recidivism Of Alcohol-impaired Driving.
were constructed and the net reclassification index (NRI) was calculated for the continuous mode. Kaplan–Meier curves were constructed for recidivism of drunk-driving as the outcome and different subgroups were compared by means of the log-rank test. All P-values are two-tailed. The observation period was 24 months after completion of a driver’s licence re-granting programme of 1 year (total observation time = 3 years). Mann–Whitney U-tests were used to compare median biomarker values between different subgroups. Results are expressed as median and interquartile range (IQR). Statistical calculations were run using SPSS for Windows, version 12.0.1 (SPSS Inc., Chicago, IL, USA) in addition to R, version 2.15.1 (15 February 2013 issue; The R Foundation for Statistical Computing, Vienna, Austria).
Table 3 Unadjusted and adjusteda odds ratios (OR) and 95% confidence intervals (95% CI) for the prediction of recidivism of driving under the influence of alcohol.
RESULTS
besides ln(CDT) (P < 0.001), ln(γGT) (P < 0.01), ln(ALT) (P < 0.05) and sex (P < 0.05) were significant predictors for recidivism of drunk-driving, but with much lower P-values than ln(CDT) (Table 3). Additionally, the analysis reveals that male drivers have an odds ratio (OR) of 2.26 (1.00–5.20) over female drivers to become a recidivist of drunk-driving. The ROAD index (predicted probability of drunk-driving recidivism based on the prediction model) was calculated from the predictive index:
The medians and interquartile ranges of %CDT, γGT, AST, ALT, MCV and the ROAD index of recidivists of drunkdriving and the control group are compared in Table 2. Serum %CDT and the ROAD index were significantly higher in recidivists when comparing to one-time offenders (P < 0.001). There was no significant difference in age, liver enzyme activities and MCV. The relative number of male drivers (85.5%) was higher than the number of female drivers (14.5%).
Variable b
Adjusted a OR (95% CI)
Unadjusted OR (95% CI)
Sex Ln(CDT) Ln(γGT) Ln(ALT) Constant
2.50 (1.07–5.82) 2.17 (1.60–2.96) 1.46 (1.07–1.99) 0.57 (0.33–0.98) 0.14
2.74 (1.22–6.17) 2.24 (1.68–2.99) 1.18 (0.95–1.48) 0.89 (0.60–1.30) NA
a Binary logistic regression, adjusting for the other factors shown in the table. bVariables entered: sex of the driver, age of the driver, ln[percentage carbohydrate-deficient transferrin (CDT)], ln[gammaglutamyltransferase (γGT)], ln[alanine amino transferase (ALT)]; NA: not applicable.
predictive index ( PI ) = 0.776 × ln [CDT (%)] + 0.376 × ln [ γGT (U L )] − 0.568 × ln [ ALT (U L )] + 0.914 [SEX ]a − 1.967.
The ROAD index A logistic regression model was used to pinpoint the indirect alcohol biomarkers associated with recidivism of drunk-driving. When analysing with forward stepwise logistic regression, the variables remaining in the equation at the end of the analysis were ln(CDT) and the sex of the driver, indicating that ln(CDT) is the most important indirect biomarker for predicting drunk-driving recidivism. A backward stepwise analysis revealed that, © 2013 Society for the Study of Addiction
a
The ‘SEX’ of the driver is indicated as 0 for women; 1 for men. Effect of currently used dichotomizations of test results Ordinarily, laboratory test results are interpreted using the ULN of reference ranges. This analysis permits the OR to be calculated for each biomarker for the prediction of Addiction, 109, 71–78
Biomarkers and recidivism of drunk-driving
75
Figure 1 Kaplan–Meier curves of the probability of drunk-driving recidivism within a 3-year observation period for a given percentage carbohydrate-deficient transferrin (%CDT) (a) and Recidivism Of Alcohol-impaired Driving (ROAD) index (b) in previously convicted drunk-drivers. P-values are for the overall log-rank test
recidivism of drunk-driving within an observation period of 3 years. Serum CDT > 1.6% (97.5th ULN of %CDT applied in routine) was the only significant predictor of drunk-driving recidivism, yielding an OR of 4.02 (2.55– 6.33). Non-significant ORs were found for γGT [1.02 (0.56–1.83)], ALT [1.36 (0.67–2.64)] and MCV [1.34 (0.80–2.26)]. Kaplan–Meier curves for drunk-driving recidivism In Fig. 1a, the probability of drunk-driving recidivism is plotted over time and compared between tertiles of %CDT (P < 0.001). There is a significant difference in the probability of drunk-driving recidivism between drivers with a %CDT ≥ 1.6 (third tertile) and drivers with a %CDT < 1.6 (first and second tertiles) (P < 0.001). Figure 1b illustrates the probability of drunk-driving recidivism compared between tertiles of the ROAD index (P < 0.001). Based on the ROAD index, drivers with a high risk of recidivating (ROAD index ≥ 25%; third tertile) can be distinguished from drivers with an intermediate risk (16% ≤ ROAD index < 25%; second tertile; P < 0.001) and a low recidivism risk (ROAD index < 16%; first tertile) (P < 0.05).
Figure 2 Receiver operating characteristic (ROC) curves for predicting recidivism within a 3-year period for percentage carbohydrate-deficient transferrin (%CDT), mean corpuscular volume (MCV), gamma-glutamyltransferase (GGT), alanine amino transferase (ALT), aspartate amino transferase (AST) and the Recidivism Of Alcohol-impaired Driving (ROAD) index based on 503 observations
Re-classification improvement of the ROAD index In Fig. 2, the ROC curves are shown for %CDT, γGT, AST, ALT, MCV and the ROAD index for distinguishing drunkdriving recidivists from the control group. The area under the curve (AUC) of the ROC curve of %CDT is 0.68, which is significantly higher (P < 0.001) than the AUC of the serum transaminase activities (AST: 0.50; ALT: 0.46), γGT (0.55) and MCV (0.55). When calculating the ROAD index for each driver, the AUC under the ROC was 0.71. The NRI of the ROAD index compared to %CDT is 0.28 © 2013 Society for the Study of Addiction
(P < 0.01), indicating a significant risk classification improvement of the ROAD index for classifying drivers as being at risk of recidivism of drunk-driving in comparison to classification based solely on %CDT. ROC comparison indicates that, at a fixed specificity of 75%, %CDT has a sensitivity of 60%, while the sensitivity of the ROAD index is 65%. A sensitivity evaluation revealed a significant classification improvement in drivers aged 40 or less (AUCROAD: 0.74 versus AUC%CDT: 0.69; P < 0.001). Addiction, 109, 71–78
76
Thomas M. Maenhout et al.
DISCUSSION Our results have shown that the determination of %CDT, γGT activity and ALT activity in serum (combined in the ROAD index) of drivers at their first admittance in a driver’s licence re-granting programme can assist in the prediction of drunk-driving recidivism. The most important predicting indirect alcohol biomarker was %CDT. Indirect biochemical evidence of an AUD, when starting a driver’s licence re-granting programme, seems to be an important predictor of drunk-driving recidivism. It has been reported previously that %CDT is a very specific biomarker for enhancing detection of heavy drinking in drivers admitted to a driver’s licence programme [7,9,13]. This may explain why %CDT was found to be the key predicting indirect alcohol biomarker of recidivism in the studied population. However, there are certain pitfalls in the interpretation of a %CDT results [7,9]. Rehabilitation researchers should be aware of the grey zone of %CDT between the ULN of 1.6% and the forensic cut-off (2.3%), which takes into account measurement uncertainty [9]. The use of a forensic cut-off could significantly limit the sensitivity of %CDT in certain subgroups to identify possible drunk-driving recidivists. Also, %CDT is known to have a lower sensitivity in young drivers (aged 30 or less), subjects with a high body mass index (BMI), drivers showing an intermittent (binge) drinking pattern and females [7,9,19]. The logistic regression analysis has also indicated a small positive contribution of γGT. This implies that γGT, in addition to %CDT, has a little, yet significant, contribution in predicting drunk-driving recidivism. This confirms findings from previous reports where increased γGT activities were found in populations of drunk-drivers, concluding that a considerable proportion of problem drinkers exists in this population group [2,20,21]. However, in the general population, elevated γGT activities show up in only 30–50% of excessive drinkers [22]. Additionally, it is not considered a specific marker for chronic heavy alcohol use, because intake of certain drugs and digestive diseases, such as pancreatitis and prostate disease, can also increase γGT levels [22]. Nevertheless, a mathemathical combination of γGT and %CDT (γ-CDT) has already been proposed to discriminate social drinkers and alcohol abusers within the clinical context [23–25]. A recent study has indicated that γ-CDT emerged as a notable predictor of DUI recidivism in a group of random breath-tested drivers [2]. However, the CDT methodology (CDTect; Pharmacia & Upjohn Diagnostics, Kalamazoo, MI, USA) used for developing the combined γ-CDT marker is withdrawn from routine because it was prone to analytical interference, which could lead to false positive results in women and drivers with a high transferrin level [26]. © 2013 Society for the Study of Addiction
A third biomarker included in the ROAD index was serum ALT activity. Serum AST activity was excluded from the analysis due to collinearity with ALT (Pearson’s R2 = 0.75). The latter was selected as it is considered a more specific measure of liver injury than AST, because it is found predominantly in the liver, whereas AST is found in several organs, including the liver, heart, muscle, kidney and brain [27]. The negative β-value of ln(ALT) in the equation of the ROAD index can be interpreted as a correction term for enzyme leakage from non-hepatic tissue or non-alcoholic liver damage. When the logistic regression analysis is performed without %CDT the negative β-value for ALT remains in equation, next to a positive contribution of γGT. This could confirm that an isolated γGT increase could have a higher predictive value for drunk-driving recidivism compared to a simultaneous increase of γGT and ALT (or AST) [28]. No significant contribution to the prediction of drunk-driving recidivism was found for MCV. Although moderately specific to ethanol, MCV is far less sensitive for chronic alcohol abuse than other markers [28]. In a clinical context, increased MCV and serum ALT and AST activities may indicate alcoholic liver disease [27]; however, due to the low specificity and sensitivity, their application in re-licensing investigations should be questioned and the test results should be interpreted with caution. When applying routinely used clinical cut-off limits for these biomarkers in the dichotomized logistic regression analysis, their contribution to the prediction of recidivism disappears completely. The non-contribution of MCV and the minimal negative contribution of ALT (or AST, due to the high correlation) to the ROAD index indicate that, while they are still clinically useful, they serve no clear purpose in DUI studies. The Kaplan–Meier analysis indicated that both %CDT and the ROAD index were capable of distinguishing drunk-drivers with a low and a high risk of recidivating (Fig. 1). It seems that the ROAD index has little value above %CDT for the identification of high-risk drunk drivers. However, based on the ROAD index, which adds γGT, ALT and the sex of the driver to %CDT, an intermediate-risk group could be identified while, based on %CDT results alone, this was not possible. This might be explained by the fact that the sensitivity of the combination of CDT and γGT to detect heavy drinking exceeds the diagnostic sensitivity of both CDT and γGT without sacrificing specificity [2]. A sensitivity analysis has shown that this classification improvement was especially seen in drivers aged 40 years or less, where %CDT is known to be less sensitive [9]. A last predictor included in the ROAD index was the sex of the driver. Previous research by our group indicated that male drivers are the predominant population Addiction, 109, 71–78
Biomarkers and recidivism of drunk-driving
in the Belgian driver’s licence re-granting programme [9]. This study indicates that male drivers are also more likely to become a recidivist of drunk-driving than female drivers, with an OR of 2.26 for men for becoming a recidivist of drunk-driving. The latter is also by far the most consistent finding in the literature [29]. An important remark concerning the study is that [based on the Nagelkerke R square (0.13)] the predictors included in the ROAD index could predict only a small fraction of the drunk-driving recidivism. This indicates that other predictors, not included in the model (e.g. psychometric measures and clinical investigation), remain very important for determining a driver’s drinking pattern and predicting drunk-driving recidivism. Demographic, behavioural and psychological person-level factors have all been cited as important predictors of which DUI offenders are likely to recidivate [5]. There are significant associations between DUI offender status and age, ethnicity, education, employment status, income and marital status [29]. In the studied population, approximately 21% has relapsed within the 3-year follow-up period. Previous findings have indicated that when the follow-up period ranges between 2 and 30 years, a recidivism rate of roughly one-third among drunken drivers is seen [2]. Of course, data on frequency of recidivism are only preliminary, as many offenders are never caught and an unknown proportion of DUI offenders continue to drink and drive. Nevertheless, this could indicate that, irrespective of the follow-up period of drunk-drivers, they are a special population who seem to relapse continually. Therefore, it is important to have as many tools as possible available to identify the high-risk offenders and keep them out of traffic. We have shown that indirect alcohol biomarkers have a small, yet significant contribution to the prediction of drunk-driving recidivism Finally, in addition to indirect alcohol biomarkers, non-oxidative ethanol metabolites (direct alcohol biomarkers) should be taken into account in re-licensing decisions. Direct markers of alcohol intake [phosphathidyl ethanol (PEth) in whole blood, fatty acid ethyl esters (FAEE) in hair, ethyl glucuronide (EtG) and ethyl sulphate (EtS) in hair or urine] have been found to be strongly predictive of driver-related risks, but these laboratory measurements are not routinely available [10,30,31]. Due to their high sensitivity, high specificity and intermediate normalization rates, they could fill the gap between direct ethanol measurement and chronic alcohol biomarkers such as CDT and γGT [10]. Direct alcohol biomarkers could significantly increase specificity and sensitivity for detecting AUD in drivers and, thereby, probably assist in predicting the risk of drunk-driving recidivism. © 2013 Society for the Study of Addiction
77
Implications The ROAD study revealed that, of all routinely used indirect alcohol markers, %CDT is the major predictor of recidivism of drunk-driving. Additionally, we have indicated that non-specific indirect markers of alcohol intake, such as γGT, AST, ALT and MCV, have minimal or no added value for distinguishing drunk drivers with a low and a high risk of recidivating. The ROAD index reveals that taking into account the γGT activity, ALT activity and the sex of the driver next to the %CDT result could have an additional value over %CDT for distinguishing drunk drivers with a low and an intermediate risk of recidivating. Future research should focus on investigating the value of the ROAD index as an interpretative tool for indirect alcohol markers test results in addition to direct alcohol markers, questionnaires, patient history, social status and clinical examination to estimate the risk of a drunk driver to become a drunk-driving recidivist. Declaration of interests None. Acknowledgements None. References 1. Janitzek T. Drink driving: Young drivers and recidivist offenders. European Transport Safety Council (ETSC). 2008. Available at: http://www.etsc.eu/documents/ETS_004-08.pdf (accessed 10 July 2012). (Archived at http://www .webcitation.org/6KsWRPKaa on 4 November 2013). 2. Portman M., Pentillä A., Haukka J., Eriksson P., Alho H., Kuoppasalmi K. Predicting DUI recidivism of male drunken driving: a prospective study of the impact of alcohol markers and previous drunken driving. Drug Alcohol Depend 2010; 106: 186–92. 3. Christophersen A. S., Beylich K. M., Bjørneboe A., Skurtveit S., Mørland J. Recidivism among drunken and drugged drivers in Norway. Alcohol Alcohol 1996; 31: 609–11. 4. Caetano R., Raspberry K. Drinking and DSM-IV alcohol and drug dependence among white and Mexican-American DUI offenders. J Stud Alcohol 2000; 61: 420–6. 5. Schell T. L., Chan K. S., Morral A. R. Predicting DUI recidivism: personality, attitudinal, and behavioral risk factors. Drug Alcohol Depend 2006; 82: 33–40. 6. Bishop N. J. Predicting rapid DUI recidivism using the Driver Risk Inventory on a state-wide sample of Floridian DUI offenders. Drug Alcohol Depend 2011; 118: 423–9. 7. Delanghe J. R., De Buyzere M. L. Carbohydrate-deficient transferrin and forensic medicine. Clin Chim Acta 2009; 406: 1–7. 8. Bjerre B. Primary and secondary prevention of drinkdriving by the use of alcolock device and program. Swedish experiences. Accid Anal Prev 2005; 37: 1145–52. 9. Maenhout T. M., Baten G., De Buyzere M. L., Delanghe J. R. Carbohydrate deficient transferrin in a driver’s license regranting program. Alcohol Alcohol 2012; 47: 253–60. Addiction, 109, 71–78
78
Thomas M. Maenhout et al.
10. Maenhout T. M., De Buyzere M. L., Delanghe J. R. Nonoxidative ethanol metabolites as a measure of alcohol intake. Clin Chim Acta 2013; 415: 322–9. 11. Korzec A., Bar M., Korter M. W. J., de Kieviet W. Diagnosing alcoholism in high risk drinking drivers: comparing different diagnostic procedures with estimated prevalence of hazardous alcohol use. Alcohol Alcohol 2001; 36: 594– 602. 12. Stibler H. Carbohydrate-deficient transferrin in serum: a new marker of potentially harmful alcohol consumption reviewed. Clin Chem 1991; 37: 2029–37. 13. Helander A. Biological markers of alcohol abuse. J Neural Transm 2003; 66: 15–32. 14. Nochajski T. H., Stasiewicz P. R. Relationship of initial depression with 18-month follow-up substance related outcomes for court-mandated DUI offenders. Proceedings of the International Conference on Alcohol, Drugs and Traffic Safety. Seattle: ICADTS; 2007. 15. Schellenberg F., Girre C., Nalpas B., Plat A., Tome A., Pagès J. C. Analytical evaluation of a new capillary electrophoresis method for carbohydrate deficient transferrin measurement. Clin Chim Acta 2007; 382: 48–53. 16. Carrol R. J., Ruppert D. Transformation and Weighting in Regression. Monographs on statistics and applied probability. New York: Chapman and Hall; 1988, 241 pp. 17. Schellenberg F., Wielders J. P. M. Evaluation of capillary electrophoresis assay for CDT on SEBIA’s Capillarys system: intra and inter laboratory precision, reference interval and cut-off. Clin Chim Acta 2010; 411: 1888–93. 18. Thomas L. Reference ranges for adults and children. Labor und Diagnose (English translation: Laboratory and Diagnosis), 5th edn. Frankfurt, Germany: TH Books; 1998, p. 83. 19. Whitfield J. B., Fletcher L. M., Murphy T. L., Powell L. W., Halliday J., Heath A. C. et al. Smoking, obesity, and hypertension alter the dose–response curve and test sensitivity of carbohydrate-deficient transferrin as a marker of alcohol intake. Clin Chem 1998; 44: 2480–9. 20. Pikkarainen J., Penttilä A. Drinking, driving and road-side studies—the Finnish experience. In: Utzelmann H.-D., Kroj G., Berghaus G., editors. Alcohol, Drugs and Traffic Safety T92. Cologne: TUV-Akademie Rheinland GmbH; 1992, pp. 1582–602. 21. Dunbar J. A., Penttilä A., Pikkarainen J. Random breath testing: an effective means of reducing drinking and driving
© 2013 Society for the Study of Addiction
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
and identifying problem drinkers. In: Valverius M., Roszbach R., editors. Proceedings 10th International Conference on Alcohol, Drugs and Traffic Safety. Amsterdam: Elsevier; 1987, pp. 517–20. Conigrave K. M., Davies P., Haber P., Whitfield J. B. Traditional markers of excessive alcohol use. Addiction 2003; 98: 31–43. Sillanaukee P., Olsson U. Improved diagnostic classification of alcohol abusers by combining carbohydrate-deficient transferrin and γ-glutamyltransferase. Clin Chem 2001; 47: 681–5. Hietela J., Koivisto H., Antilla P., Niemela O. Comparison of the combined marker γGT–CDT and the conventional laboratory markers of alcohol abuse in heavy drinkers, moderate drinkers and abstainers. Alcohol Alcohol 2006; 41: 528–33. Marques P., Tippetts S., Allen J., Javors M., Alling C., Yegles M. Estimating driver risk using alcohol biomarkers, interlock BAC tests and psychometric assessments: initial descriptive. Addiction 2010; 105: 226–39. Anttila P., Järvi K., Latvala J., Blake J. E., Niemelä O. Diagnostic characteristics of different carbohydrate-deficient transferrin methods in the detection of problem drinking: effects of liver disease and alcohol consumption. Alcohol Alcohol 2003; 38: 415–20. Halvorson M. R., Campbell J. L., Sprague G., Slater K., Noffsinger J. K., Peterson C. M. Comparative evaluation of the clinical utility of three markers of ethanol intake: the effect of gender. Alcohol Clin Exp Res 1993; 17: 225–9. Mundle G., Munkes J., Ackermann K., Mann K. Sex differences of carbohydrate-deficient transferrin, gammaglutamyltransferase, and mean corpuscular volume in alcohol dependent patients. Alcohol Clin Exp Res 2000; 24: 1400–5. Nochajski T. H., Stasiewicz P. R. Relapse to driving under the influence (DUI): a review. Clin Psychol Rev 2006; 26: 179–95. Albermann M. E., Musshoff F., Madea B. Comparison of ethyl glucuronide (EtG) and fatty acid ethyl esters (FAEEs) concentrations in hair for testing abstinence. Anal Bioanal Chem 2011; 400: 175–81. Muskovich M., Haag-Dawoud M. Alcohol consumption among drivers subject to the Swiss license restriction of zero tolerance when driving. Traffic Inj Prev 2012; 13: 537–43.
Addiction, 109, 71–78
RESEARCH REPORT
doi:10.1111/add.12372
Usefulness of indirect alcohol biomarkers for predicting recidivism of drunk-driving among previously convicted drunk-driving offenders: results from the Recidivism Of Alcohol-impaired Driving (ROAD) study Thomas M. Maenhout1, Anneleen Poll2, Tijl Vermassen1, Marc L. De Buyzere1, Joris R. Delanghe1 & the ROAD Study Group* Department of Clinical Chemistry, Ghent University Hospital, Ghent, Belgium1 and Dienst herstelonderzoeken, Belgian Institute of Road Safety, Brussels, Belgium2
ABSTRACT Aim In several European countries, drivers under the influence (DUI), suspected of chronic alcohol abuse are referred for medical and psychological examination. This study (the ROAD study, or Recidivism Of Alcohol-impaired Driving) investigated the usefulness of indirect alcohol biomarkers for predicting drunk-driving recidivism in previously convicted drunk-driving offenders. Design, setting, participants and measurements The ROAD study is a prospective study (2009–13) that was performed on 517 randomly selected drivers in Belgium. They were convicted for drunk-driving for which their licence was confiscated. The initial post-arrest blood samples were collected and analysed for percentage carbohydrate-deficient transferrin (%CDT), transaminsase activities [alanine amino transferase (ALT), aspartate amino transferase (AST)], gamma-glutamyltransferase (γGT) and red cell mean corpuscular volume (MCV). The observation time for each driver was 3 years and dynamic. Findings A logistic regression analysis revealed that ln(%CDT) (P < 0.001), ln(γGT) (P < 0.01) and ln(ALT) (P < 0.05) were the best biochemical predictors of recidivism of drunk-driving. The ROAD index (which includes ln(%CDT), ln(γGT), -ln(ALT) and the sex of the driver) was calculated and had a significantly higher area under the receiver operator characteristic curve (0.71) than the individual biomarkers for drunk-driving recidivism. Drivers with a high risk of recidivating (ROAD index ≥ 25%; third tertile) could be distinguished from drivers with an intermediate risk (16% ≤ ROAD index < 25%; second tertile; P < 0.001) and a low recidivism risk (ROAD index < 16%; first tertile; P < 0.05). Conclusions Of all routinely used indirect alcohol markers, percentage of carbohydrate-deficient transferrin is the major predictor of recidivism of drunk-driving. The association with gamma-glutamyltransferase, alanine amino transferase and the sex of the driver could have additional value for identifying drunk-drivers at intermediate risk of recidivism. Non-specific indirect alcohol markers, such as alanine amino transferase, gamma-glutamyltransferase, aspartate amino transferase and red cell mean corpuscular volume have minimal added value to % carbohydrate-deficient transferrin for distinguishing drunk drivers with a low or high risk of recidivism. Keywords Alcoholism, carbohydrate-deficient transferrin, driver’s licence, drunk-driving, liver disease, logistic regression, transferrin. Correspondence to: Joris R. Delanghe, Department of Clinical Chemistry, Ghent University Hospital, De Pintelaan 185, 9000 Ghent, Belgium. E-mail: [email protected] Submitted 29 March 2013; initial review completed 15 May 2013; final version accepted 27 September 2013
*ROAD study group collaborators: P. Maenhout (Labo Maenhout), K. Van Poucke (Labo Van Poucke), C. Neven (LKO-LMC) and W. Top (Labo Medina). © 2013 Society for the Study of Addiction
Addiction, 109, 71–78
72
Thomas M. Maenhout et al.
INTRODUCTION Recidivism of driving under the influence (DUI) of alcohol among drunk-driving offenders is a serious concern. Repeat offenders are more likely to be involved in alcohol-related crashes than first-time offenders [1,2]. It has been shown that the rate of repeat DUI offences is greater among drunk drivers who are diagnosed with an alcohol use disorder (AUD) and report daily excessive drinking [3]. When compared to the general population, this is in correlation with the higher prevalence rates of DSM-IV alcohol dependence among drunk drivers [4]. While repeat offenders represent only a small proportion of all drivers, they contribute disproportionately to road accidents [5,6]. Predicting the risk of a repeat offence, therefore, is an important consideration for making re-licensing decisions. Re-granting a driver’s licence to a high-risk offender could pose a significant threat to the safety of public roadways [6]. In several European countries, driver’s licence re-granting programmes are used, where apprehended drivers, suspected of an AUD, are referred for medical and psychological examination [7]. Sweden, for example, has a medical review programme for DUI offenders that links reductions of alcohol markers to successful completion of their interlock programme [8]. In Belgium, a judge can propose the participation to a driver’s licence re-granting programme as a subsequent condition in order to restore the driver’s licence. Next to extensive psychological and clinical investigations, biomarkers are used to monitor the driver’s adherence to the abstinence programme [9]. Re-licensing decisions are often based on subjective expert opinions which take into consideration often poorly standardized clinical and laboratory findings.
Many subjects misusing alcohol can be identified based on clinical history and examination, combined with selfreport questionnaires; however, sensitivity is not optimal, with deliberate under-reporting being common [10]. Alcohol biomarkers are often suggested as objective measures and can give additional information, next to clinical and psychological investigation, for identifying those drunk-driving offenders showing a chronic alcohol problem, thereby identifying those drivers most at risk for recidivism of drunk-driving [5]. However, the accuracy of biomarkers used for substantiating the diagnosis of an AUD in drunk-drivers is difficult to assess, and until now there has been scarce evidence about their power to predict a drunk-driving relapse [2,11]. In the present study, we investigate the usefulness of currently widespread available indirect alcohol state markers for the prediction of recidivism of drunkdriving. Indirect alcohol markers are of particular interest for identifying chronic alcohol abuse because of their serum half-life ranging from 2 to 3 weeks, and the requirement of an excessive average daily alcohol consumption before elevation [5,10]. In Belgium, carbohydrate-deficient transferrin (CDT) testing was introduced in 2008 within the framework of driver’s licence re-instatement. It is considered the most accurate indirect biomarker for identifying sustained heavy alcohol consumption and for monitoring abstinence [12,13]. Next to CDT, the most frequently used indirect biomarkers for the detection and monitoring chronic alcohol abuse are γ-glutamyltransferase (γGT), alanine amino transferase (ALT), aspartate amino transferase (AST) and mean corpuscular volume (MCV) (Table 1). Combinations of several markers of alcohol abuse are thought to give better sensitivity, but usually show
Table 1 Description of biochemical laboratory parameters included in the Recidivism Of Alcohol-impaired Driving (ROAD) study.
AST
ALT MCV
Type of marker
Remarks
t1/2 (days)
Cytoplasmic (mitochondrial) liver enzyme Cytoplasmic liver enzyme Calculated descriptive of red blood cell volume
May be elevated in obesity, liver and biliary conditions, muscle diseases, extreme exertion
13 ± 20 [22]
May be elevated in obesity, liver and biliary conditions May be elevated in folate or B12 deficiency (including malabsorption), bleeding (leads to reticulocytosis), haematological conditions (including haemolysis, haemoglobinopathies), bone marrow disorders, liver disease, hypothyroidism, hyperglycaemia May be elevated in obesity, liver and biliary conditions (including hepatic congestion), hypertriglyceridaemia, diabetes, pancreatitis May be elevated in liver cirrhosis. Analytical difficulties encountered in transferrin variants, di-trisialo-bridging phenomena
16 ± 19 [22] NA (typical life-span of a red blood cell is approximately 120 days)
γGT
Cell surface liver enzyme
CDT
Deregulation of glycoconjugate metabolic processes
16 ± 8 [22] 16 ± 11 [7]
CDT: carbohydrate-deficient transferrin; γGT: gamma-glutamyltransferase; MCV: mean corpuscular volume; ALT: alanine amino transferase; AST: aspartate amino transferase; NA: not applicable.
© 2013 Society for the Study of Addiction
Addiction, 109, 71–78
Biomarkers and recidivism of drunk-driving
reduced specificity for detecting alcohol abuse. Despite intensive investigation, there is still no consensus on which laboratory biomarkers are satisfactory in predicting surreptitious alcohol ingestion. In 2009, the ROAD study (Recidivism Of Alcohol-impaired Driving) was set up with its major purpose investigating the predictive value of indirect alcohol biomarkers for recidivism of drunk-driving in drivers admitted to the driver’s licence re-granting programme. Being surrogate markers for chronic alcohol abuse, the predictive power of each different biomarker separately and combinations of different biomarkers for recidivism of drunk-driving will be investigated.
MATERIALS AND METHODS Study design The ROAD study is a prospective study that was performed on 517 randomly selected drunk-driving offenders who were admitted to the Belgian driver’s licence re-granting programme under the control of the Belgian Institute for Road Safety (BIVV; Belgisch Instituut voor Verkeersveiligheid—IBSR; Institut Belge pour la Sécurité Routière) in 2009. The observation time was 3 years and dynamic for each driver, which means that each driver had a time-window of 3 years for an actual relapse event. At first admittance, blood samples were collected in duplicate [one whole blood on K3ethylenediamine tetraacetic acid (EDTA) and one serum sample]. In 14 drivers, CDT could not be quantified accurately due to the presence of a genetic transferrin (Tf) variant. These cases were excluded from the analysis. A total of 108 drivers were included in the subgroup of recidivists of drunk-driving. They were arrested for a subsequent drunk-driving offence within the 3-year observation period, based on established legal limits (0.05‰ according to Belgian legislation) for driving while impaired or intoxicated with alcohol. This criterion for drunk-driving recidivism was adapted from Nochajski & Stasiewicz [14]. Drivers not meeting these criteria were included in the control group (n = 395). The study was approved by the Ethics Committee of Ghent University Hospital (EC/2012-516). Analytical methods—alcohol markers Whole blood (using K3EDTA as anticoagulant) was used to determine a complete blood count, including MCV, using an automated haematology analyser (Sysmex XE-5000). Serum was separated by centrifugation and was assayed for %CDT, γGT, ALT and AST activity. γGT activity was measured using a kinetic spectrophotometric assay (405 nm) with carboxynitroanilide as a substrate. ALT and AST activities were measured using the kinetic © 2013 Society for the Study of Addiction
73
ultraviolet (UV) spectrophotometric International Federation for Clinical Chemistry and Laboratory Medicine (IFCC) method. Serum concentrations of %CDT (i.e. percentage asialo- and disialo-Tf of the total Tf) were measured using capillary zone electrophoresis (CZE) [9], performed on the Capillarys 2™ system (Sebia, Evry, France). In this technique, after on-line iron saturation, samples are submitted to high-voltage (8200 V) zone electrophoresis in alkaline buffer (pH 8.8). Tf glycoforms are quantified by their peptide bond absorbance at 200 nm. Pherograms were validated using the manufacturer’s software [15]. Statistical analysis A logistic regression analysis was performed with different indirect alcohol biomarkers (%CDT, γGT, AST, ALT and MCV), age and sex of the driver on recidivism of drunk-driving as the dependent variable. Intentionally, no psychometric measures or intervention variables were taken into account to investigate independently the predictive value of biomarkers. The inserted variables are test results from the initial post-DUI arrest blood samples of the driver. Standard diagnostic methods for linear statistical models suggest that %CDT, γGT, MCV, AST and ALT should be analysed in logarithmic form [16]. The natural logarithm was used. We have performed an initial forward stepwise logistic regression analysis to determine which variables accounted for most of the variance when order of entry is not dictated. Afterwards, we performed a backward stepwise logistic regression. The regression equation from the latter was used to calculate the predictive index (PI). The ROAD index was calculated from the PI (ROAD index = exp(PI)/[1 + exp(PI]). To minimize the included variables in order to simplify the regression equation, a data reduction by principle component analysis was performed, and as a result serum AST activity was removed from the analysis, due to the high correlation (Pearson’s R2 = 0.75) with serum ALT activity. The Wald test was used to evaluate the contribution of each individual predictor to the model. A predictor was entered into the regression equation when the probability (P) was 0.05. Additionally, a logistic regression analysis was performed on dichotomized data to evaluate routinely applied decision limits for each indirect alcohol biomarkers. The latter were based on internationally accepted upper 97.5th limits of normal (ULN): for males, 1.6% for CDT, 61 U/L for γGT, 37 U/L for AST, 40 U/L for ALT and 96.4 fL for MCV; for females, 1.6% for CDT, 36 U/L for γGT, 31 U/L for AST, 31 U/L for ALT and 96.4 fL for MCV [9,17,18]. For the comparisons between different classification criteria, receiver operator characteristic (ROC) plots Addiction, 109, 71–78
74
Thomas M. Maenhout et al.
Table 2 Descriptive statistics of the data set used for logistic regression. Results are expressed as median and interquartile range.
Subgroupa
Women Age at start Men (n) (n) study (years) CDT (%)
Control group 330 (n = 395) Drunk-driving 100 recidivist (n = 108) Overall 430 (n = 503)
ROAD index (%)
γGT (U/L)
18 (3–67)
35 (21–68) 24 (19–30) 25 (29–35) 93 (90–96)
AST (U/L)
ALT (U/L)
MCV (fL)
65
46 (36–53)
1.0 (0.8–1.5)
8
46 (37–50)
1.9* (0.9–2.8) 27* (4–65) 39 (22–74) 23 (19–33) 22 (17–33) 94 (91–98)
73
46 (36–53)
1.0 (0.8–1.8)
19 (3–67)
40 (21–69) 24 (19–31) 24 (19–35) 93 (90–97)
a The population was divided into two subgroups: when the driver was convicted more than once in 3 years following initial conviction, the subject was placed into the ‘recidivist’ subgroup; otherwise the subjects were considered as controls. *P < 0.001 for Mann–Whitney test for independent samples compared to control group. % CDT: percentage carbohydrate-deficient transferrin; γGT: gamma-glutamyltransferase; MCV: mean corpuscular volume; ALT: alanine amino transferase; AST: aspartate amino transferase; ROAD: Recidivism Of Alcohol-impaired Driving.
were constructed and the net reclassification index (NRI) was calculated for the continuous mode. Kaplan–Meier curves were constructed for recidivism of drunk-driving as the outcome and different subgroups were compared by means of the log-rank test. All P-values are two-tailed. The observation period was 24 months after completion of a driver’s licence re-granting programme of 1 year (total observation time = 3 years). Mann–Whitney U-tests were used to compare median biomarker values between different subgroups. Results are expressed as median and interquartile range (IQR). Statistical calculations were run using SPSS for Windows, version 12.0.1 (SPSS Inc., Chicago, IL, USA) in addition to R, version 2.15.1 (15 February 2013 issue; The R Foundation for Statistical Computing, Vienna, Austria).
Table 3 Unadjusted and adjusteda odds ratios (OR) and 95% confidence intervals (95% CI) for the prediction of recidivism of driving under the influence of alcohol.
RESULTS
besides ln(CDT) (P < 0.001), ln(γGT) (P < 0.01), ln(ALT) (P < 0.05) and sex (P < 0.05) were significant predictors for recidivism of drunk-driving, but with much lower P-values than ln(CDT) (Table 3). Additionally, the analysis reveals that male drivers have an odds ratio (OR) of 2.26 (1.00–5.20) over female drivers to become a recidivist of drunk-driving. The ROAD index (predicted probability of drunk-driving recidivism based on the prediction model) was calculated from the predictive index:
The medians and interquartile ranges of %CDT, γGT, AST, ALT, MCV and the ROAD index of recidivists of drunkdriving and the control group are compared in Table 2. Serum %CDT and the ROAD index were significantly higher in recidivists when comparing to one-time offenders (P < 0.001). There was no significant difference in age, liver enzyme activities and MCV. The relative number of male drivers (85.5%) was higher than the number of female drivers (14.5%).
Variable b
Adjusted a OR (95% CI)
Unadjusted OR (95% CI)
Sex Ln(CDT) Ln(γGT) Ln(ALT) Constant
2.50 (1.07–5.82) 2.17 (1.60–2.96) 1.46 (1.07–1.99) 0.57 (0.33–0.98) 0.14
2.74 (1.22–6.17) 2.24 (1.68–2.99) 1.18 (0.95–1.48) 0.89 (0.60–1.30) NA
a Binary logistic regression, adjusting for the other factors shown in the table. bVariables entered: sex of the driver, age of the driver, ln[percentage carbohydrate-deficient transferrin (CDT)], ln[gammaglutamyltransferase (γGT)], ln[alanine amino transferase (ALT)]; NA: not applicable.
predictive index ( PI ) = 0.776 × ln [CDT (%)] + 0.376 × ln [ γGT (U L )] − 0.568 × ln [ ALT (U L )] + 0.914 [SEX ]a − 1.967.
The ROAD index A logistic regression model was used to pinpoint the indirect alcohol biomarkers associated with recidivism of drunk-driving. When analysing with forward stepwise logistic regression, the variables remaining in the equation at the end of the analysis were ln(CDT) and the sex of the driver, indicating that ln(CDT) is the most important indirect biomarker for predicting drunk-driving recidivism. A backward stepwise analysis revealed that, © 2013 Society for the Study of Addiction
a
The ‘SEX’ of the driver is indicated as 0 for women; 1 for men. Effect of currently used dichotomizations of test results Ordinarily, laboratory test results are interpreted using the ULN of reference ranges. This analysis permits the OR to be calculated for each biomarker for the prediction of Addiction, 109, 71–78
Biomarkers and recidivism of drunk-driving
75
Figure 1 Kaplan–Meier curves of the probability of drunk-driving recidivism within a 3-year observation period for a given percentage carbohydrate-deficient transferrin (%CDT) (a) and Recidivism Of Alcohol-impaired Driving (ROAD) index (b) in previously convicted drunk-drivers. P-values are for the overall log-rank test
recidivism of drunk-driving within an observation period of 3 years. Serum CDT > 1.6% (97.5th ULN of %CDT applied in routine) was the only significant predictor of drunk-driving recidivism, yielding an OR of 4.02 (2.55– 6.33). Non-significant ORs were found for γGT [1.02 (0.56–1.83)], ALT [1.36 (0.67–2.64)] and MCV [1.34 (0.80–2.26)]. Kaplan–Meier curves for drunk-driving recidivism In Fig. 1a, the probability of drunk-driving recidivism is plotted over time and compared between tertiles of %CDT (P < 0.001). There is a significant difference in the probability of drunk-driving recidivism between drivers with a %CDT ≥ 1.6 (third tertile) and drivers with a %CDT < 1.6 (first and second tertiles) (P < 0.001). Figure 1b illustrates the probability of drunk-driving recidivism compared between tertiles of the ROAD index (P < 0.001). Based on the ROAD index, drivers with a high risk of recidivating (ROAD index ≥ 25%; third tertile) can be distinguished from drivers with an intermediate risk (16% ≤ ROAD index < 25%; second tertile; P < 0.001) and a low recidivism risk (ROAD index < 16%; first tertile) (P < 0.05).
Figure 2 Receiver operating characteristic (ROC) curves for predicting recidivism within a 3-year period for percentage carbohydrate-deficient transferrin (%CDT), mean corpuscular volume (MCV), gamma-glutamyltransferase (GGT), alanine amino transferase (ALT), aspartate amino transferase (AST) and the Recidivism Of Alcohol-impaired Driving (ROAD) index based on 503 observations
Re-classification improvement of the ROAD index In Fig. 2, the ROC curves are shown for %CDT, γGT, AST, ALT, MCV and the ROAD index for distinguishing drunkdriving recidivists from the control group. The area under the curve (AUC) of the ROC curve of %CDT is 0.68, which is significantly higher (P < 0.001) than the AUC of the serum transaminase activities (AST: 0.50; ALT: 0.46), γGT (0.55) and MCV (0.55). When calculating the ROAD index for each driver, the AUC under the ROC was 0.71. The NRI of the ROAD index compared to %CDT is 0.28 © 2013 Society for the Study of Addiction
(P < 0.01), indicating a significant risk classification improvement of the ROAD index for classifying drivers as being at risk of recidivism of drunk-driving in comparison to classification based solely on %CDT. ROC comparison indicates that, at a fixed specificity of 75%, %CDT has a sensitivity of 60%, while the sensitivity of the ROAD index is 65%. A sensitivity evaluation revealed a significant classification improvement in drivers aged 40 or less (AUCROAD: 0.74 versus AUC%CDT: 0.69; P < 0.001). Addiction, 109, 71–78
76
Thomas M. Maenhout et al.
DISCUSSION Our results have shown that the determination of %CDT, γGT activity and ALT activity in serum (combined in the ROAD index) of drivers at their first admittance in a driver’s licence re-granting programme can assist in the prediction of drunk-driving recidivism. The most important predicting indirect alcohol biomarker was %CDT. Indirect biochemical evidence of an AUD, when starting a driver’s licence re-granting programme, seems to be an important predictor of drunk-driving recidivism. It has been reported previously that %CDT is a very specific biomarker for enhancing detection of heavy drinking in drivers admitted to a driver’s licence programme [7,9,13]. This may explain why %CDT was found to be the key predicting indirect alcohol biomarker of recidivism in the studied population. However, there are certain pitfalls in the interpretation of a %CDT results [7,9]. Rehabilitation researchers should be aware of the grey zone of %CDT between the ULN of 1.6% and the forensic cut-off (2.3%), which takes into account measurement uncertainty [9]. The use of a forensic cut-off could significantly limit the sensitivity of %CDT in certain subgroups to identify possible drunk-driving recidivists. Also, %CDT is known to have a lower sensitivity in young drivers (aged 30 or less), subjects with a high body mass index (BMI), drivers showing an intermittent (binge) drinking pattern and females [7,9,19]. The logistic regression analysis has also indicated a small positive contribution of γGT. This implies that γGT, in addition to %CDT, has a little, yet significant, contribution in predicting drunk-driving recidivism. This confirms findings from previous reports where increased γGT activities were found in populations of drunk-drivers, concluding that a considerable proportion of problem drinkers exists in this population group [2,20,21]. However, in the general population, elevated γGT activities show up in only 30–50% of excessive drinkers [22]. Additionally, it is not considered a specific marker for chronic heavy alcohol use, because intake of certain drugs and digestive diseases, such as pancreatitis and prostate disease, can also increase γGT levels [22]. Nevertheless, a mathemathical combination of γGT and %CDT (γ-CDT) has already been proposed to discriminate social drinkers and alcohol abusers within the clinical context [23–25]. A recent study has indicated that γ-CDT emerged as a notable predictor of DUI recidivism in a group of random breath-tested drivers [2]. However, the CDT methodology (CDTect; Pharmacia & Upjohn Diagnostics, Kalamazoo, MI, USA) used for developing the combined γ-CDT marker is withdrawn from routine because it was prone to analytical interference, which could lead to false positive results in women and drivers with a high transferrin level [26]. © 2013 Society for the Study of Addiction
A third biomarker included in the ROAD index was serum ALT activity. Serum AST activity was excluded from the analysis due to collinearity with ALT (Pearson’s R2 = 0.75). The latter was selected as it is considered a more specific measure of liver injury than AST, because it is found predominantly in the liver, whereas AST is found in several organs, including the liver, heart, muscle, kidney and brain [27]. The negative β-value of ln(ALT) in the equation of the ROAD index can be interpreted as a correction term for enzyme leakage from non-hepatic tissue or non-alcoholic liver damage. When the logistic regression analysis is performed without %CDT the negative β-value for ALT remains in equation, next to a positive contribution of γGT. This could confirm that an isolated γGT increase could have a higher predictive value for drunk-driving recidivism compared to a simultaneous increase of γGT and ALT (or AST) [28]. No significant contribution to the prediction of drunk-driving recidivism was found for MCV. Although moderately specific to ethanol, MCV is far less sensitive for chronic alcohol abuse than other markers [28]. In a clinical context, increased MCV and serum ALT and AST activities may indicate alcoholic liver disease [27]; however, due to the low specificity and sensitivity, their application in re-licensing investigations should be questioned and the test results should be interpreted with caution. When applying routinely used clinical cut-off limits for these biomarkers in the dichotomized logistic regression analysis, their contribution to the prediction of recidivism disappears completely. The non-contribution of MCV and the minimal negative contribution of ALT (or AST, due to the high correlation) to the ROAD index indicate that, while they are still clinically useful, they serve no clear purpose in DUI studies. The Kaplan–Meier analysis indicated that both %CDT and the ROAD index were capable of distinguishing drunk-drivers with a low and a high risk of recidivating (Fig. 1). It seems that the ROAD index has little value above %CDT for the identification of high-risk drunk drivers. However, based on the ROAD index, which adds γGT, ALT and the sex of the driver to %CDT, an intermediate-risk group could be identified while, based on %CDT results alone, this was not possible. This might be explained by the fact that the sensitivity of the combination of CDT and γGT to detect heavy drinking exceeds the diagnostic sensitivity of both CDT and γGT without sacrificing specificity [2]. A sensitivity analysis has shown that this classification improvement was especially seen in drivers aged 40 years or less, where %CDT is known to be less sensitive [9]. A last predictor included in the ROAD index was the sex of the driver. Previous research by our group indicated that male drivers are the predominant population Addiction, 109, 71–78
Biomarkers and recidivism of drunk-driving
in the Belgian driver’s licence re-granting programme [9]. This study indicates that male drivers are also more likely to become a recidivist of drunk-driving than female drivers, with an OR of 2.26 for men for becoming a recidivist of drunk-driving. The latter is also by far the most consistent finding in the literature [29]. An important remark concerning the study is that [based on the Nagelkerke R square (0.13)] the predictors included in the ROAD index could predict only a small fraction of the drunk-driving recidivism. This indicates that other predictors, not included in the model (e.g. psychometric measures and clinical investigation), remain very important for determining a driver’s drinking pattern and predicting drunk-driving recidivism. Demographic, behavioural and psychological person-level factors have all been cited as important predictors of which DUI offenders are likely to recidivate [5]. There are significant associations between DUI offender status and age, ethnicity, education, employment status, income and marital status [29]. In the studied population, approximately 21% has relapsed within the 3-year follow-up period. Previous findings have indicated that when the follow-up period ranges between 2 and 30 years, a recidivism rate of roughly one-third among drunken drivers is seen [2]. Of course, data on frequency of recidivism are only preliminary, as many offenders are never caught and an unknown proportion of DUI offenders continue to drink and drive. Nevertheless, this could indicate that, irrespective of the follow-up period of drunk-drivers, they are a special population who seem to relapse continually. Therefore, it is important to have as many tools as possible available to identify the high-risk offenders and keep them out of traffic. We have shown that indirect alcohol biomarkers have a small, yet significant contribution to the prediction of drunk-driving recidivism Finally, in addition to indirect alcohol biomarkers, non-oxidative ethanol metabolites (direct alcohol biomarkers) should be taken into account in re-licensing decisions. Direct markers of alcohol intake [phosphathidyl ethanol (PEth) in whole blood, fatty acid ethyl esters (FAEE) in hair, ethyl glucuronide (EtG) and ethyl sulphate (EtS) in hair or urine] have been found to be strongly predictive of driver-related risks, but these laboratory measurements are not routinely available [10,30,31]. Due to their high sensitivity, high specificity and intermediate normalization rates, they could fill the gap between direct ethanol measurement and chronic alcohol biomarkers such as CDT and γGT [10]. Direct alcohol biomarkers could significantly increase specificity and sensitivity for detecting AUD in drivers and, thereby, probably assist in predicting the risk of drunk-driving recidivism. © 2013 Society for the Study of Addiction
77
Implications The ROAD study revealed that, of all routinely used indirect alcohol markers, %CDT is the major predictor of recidivism of drunk-driving. Additionally, we have indicated that non-specific indirect markers of alcohol intake, such as γGT, AST, ALT and MCV, have minimal or no added value for distinguishing drunk drivers with a low and a high risk of recidivating. The ROAD index reveals that taking into account the γGT activity, ALT activity and the sex of the driver next to the %CDT result could have an additional value over %CDT for distinguishing drunk drivers with a low and an intermediate risk of recidivating. Future research should focus on investigating the value of the ROAD index as an interpretative tool for indirect alcohol markers test results in addition to direct alcohol markers, questionnaires, patient history, social status and clinical examination to estimate the risk of a drunk driver to become a drunk-driving recidivist. Declaration of interests None. Acknowledgements None. References 1. Janitzek T. Drink driving: Young drivers and recidivist offenders. European Transport Safety Council (ETSC). 2008. Available at: http://www.etsc.eu/documents/ETS_004-08.pdf (accessed 10 July 2012). (Archived at http://www .webcitation.org/6KsWRPKaa on 4 November 2013). 2. Portman M., Pentillä A., Haukka J., Eriksson P., Alho H., Kuoppasalmi K. Predicting DUI recidivism of male drunken driving: a prospective study of the impact of alcohol markers and previous drunken driving. Drug Alcohol Depend 2010; 106: 186–92. 3. Christophersen A. S., Beylich K. M., Bjørneboe A., Skurtveit S., Mørland J. Recidivism among drunken and drugged drivers in Norway. Alcohol Alcohol 1996; 31: 609–11. 4. Caetano R., Raspberry K. Drinking and DSM-IV alcohol and drug dependence among white and Mexican-American DUI offenders. J Stud Alcohol 2000; 61: 420–6. 5. Schell T. L., Chan K. S., Morral A. R. Predicting DUI recidivism: personality, attitudinal, and behavioral risk factors. Drug Alcohol Depend 2006; 82: 33–40. 6. Bishop N. J. Predicting rapid DUI recidivism using the Driver Risk Inventory on a state-wide sample of Floridian DUI offenders. Drug Alcohol Depend 2011; 118: 423–9. 7. Delanghe J. R., De Buyzere M. L. Carbohydrate-deficient transferrin and forensic medicine. Clin Chim Acta 2009; 406: 1–7. 8. Bjerre B. Primary and secondary prevention of drinkdriving by the use of alcolock device and program. Swedish experiences. Accid Anal Prev 2005; 37: 1145–52. 9. Maenhout T. M., Baten G., De Buyzere M. L., Delanghe J. R. Carbohydrate deficient transferrin in a driver’s license regranting program. Alcohol Alcohol 2012; 47: 253–60. Addiction, 109, 71–78
78
Thomas M. Maenhout et al.
10. Maenhout T. M., De Buyzere M. L., Delanghe J. R. Nonoxidative ethanol metabolites as a measure of alcohol intake. Clin Chim Acta 2013; 415: 322–9. 11. Korzec A., Bar M., Korter M. W. J., de Kieviet W. Diagnosing alcoholism in high risk drinking drivers: comparing different diagnostic procedures with estimated prevalence of hazardous alcohol use. Alcohol Alcohol 2001; 36: 594– 602. 12. Stibler H. Carbohydrate-deficient transferrin in serum: a new marker of potentially harmful alcohol consumption reviewed. Clin Chem 1991; 37: 2029–37. 13. Helander A. Biological markers of alcohol abuse. J Neural Transm 2003; 66: 15–32. 14. Nochajski T. H., Stasiewicz P. R. Relationship of initial depression with 18-month follow-up substance related outcomes for court-mandated DUI offenders. Proceedings of the International Conference on Alcohol, Drugs and Traffic Safety. Seattle: ICADTS; 2007. 15. Schellenberg F., Girre C., Nalpas B., Plat A., Tome A., Pagès J. C. Analytical evaluation of a new capillary electrophoresis method for carbohydrate deficient transferrin measurement. Clin Chim Acta 2007; 382: 48–53. 16. Carrol R. J., Ruppert D. Transformation and Weighting in Regression. Monographs on statistics and applied probability. New York: Chapman and Hall; 1988, 241 pp. 17. Schellenberg F., Wielders J. P. M. Evaluation of capillary electrophoresis assay for CDT on SEBIA’s Capillarys system: intra and inter laboratory precision, reference interval and cut-off. Clin Chim Acta 2010; 411: 1888–93. 18. Thomas L. Reference ranges for adults and children. Labor und Diagnose (English translation: Laboratory and Diagnosis), 5th edn. Frankfurt, Germany: TH Books; 1998, p. 83. 19. Whitfield J. B., Fletcher L. M., Murphy T. L., Powell L. W., Halliday J., Heath A. C. et al. Smoking, obesity, and hypertension alter the dose–response curve and test sensitivity of carbohydrate-deficient transferrin as a marker of alcohol intake. Clin Chem 1998; 44: 2480–9. 20. Pikkarainen J., Penttilä A. Drinking, driving and road-side studies—the Finnish experience. In: Utzelmann H.-D., Kroj G., Berghaus G., editors. Alcohol, Drugs and Traffic Safety T92. Cologne: TUV-Akademie Rheinland GmbH; 1992, pp. 1582–602. 21. Dunbar J. A., Penttilä A., Pikkarainen J. Random breath testing: an effective means of reducing drinking and driving
© 2013 Society for the Study of Addiction
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
and identifying problem drinkers. In: Valverius M., Roszbach R., editors. Proceedings 10th International Conference on Alcohol, Drugs and Traffic Safety. Amsterdam: Elsevier; 1987, pp. 517–20. Conigrave K. M., Davies P., Haber P., Whitfield J. B. Traditional markers of excessive alcohol use. Addiction 2003; 98: 31–43. Sillanaukee P., Olsson U. Improved diagnostic classification of alcohol abusers by combining carbohydrate-deficient transferrin and γ-glutamyltransferase. Clin Chem 2001; 47: 681–5. Hietela J., Koivisto H., Antilla P., Niemela O. Comparison of the combined marker γGT–CDT and the conventional laboratory markers of alcohol abuse in heavy drinkers, moderate drinkers and abstainers. Alcohol Alcohol 2006; 41: 528–33. Marques P., Tippetts S., Allen J., Javors M., Alling C., Yegles M. Estimating driver risk using alcohol biomarkers, interlock BAC tests and psychometric assessments: initial descriptive. Addiction 2010; 105: 226–39. Anttila P., Järvi K., Latvala J., Blake J. E., Niemelä O. Diagnostic characteristics of different carbohydrate-deficient transferrin methods in the detection of problem drinking: effects of liver disease and alcohol consumption. Alcohol Alcohol 2003; 38: 415–20. Halvorson M. R., Campbell J. L., Sprague G., Slater K., Noffsinger J. K., Peterson C. M. Comparative evaluation of the clinical utility of three markers of ethanol intake: the effect of gender. Alcohol Clin Exp Res 1993; 17: 225–9. Mundle G., Munkes J., Ackermann K., Mann K. Sex differences of carbohydrate-deficient transferrin, gammaglutamyltransferase, and mean corpuscular volume in alcohol dependent patients. Alcohol Clin Exp Res 2000; 24: 1400–5. Nochajski T. H., Stasiewicz P. R. Relapse to driving under the influence (DUI): a review. Clin Psychol Rev 2006; 26: 179–95. Albermann M. E., Musshoff F., Madea B. Comparison of ethyl glucuronide (EtG) and fatty acid ethyl esters (FAEEs) concentrations in hair for testing abstinence. Anal Bioanal Chem 2011; 400: 175–81. Muskovich M., Haag-Dawoud M. Alcohol consumption among drivers subject to the Swiss license restriction of zero tolerance when driving. Traffic Inj Prev 2012; 13: 537–43.
Addiction, 109, 71–78
