self-reported drug-usage and crash-incidence in breathalyzed drivers

10

Click here to load reader

Upload: rd-macpherson

Post on 15-Jun-2016

220 views

Category:

Documents


6 download

TRANSCRIPT

Page 1: Self-reported drug-usage and crash-incidence in breathalyzed drivers

Accid. And. d Prm. Vol. 16, No. 2, pp. 139-148, 1984 lxm4575/84 $3.00 + Alo Pnnted in Great Bntain. G 1984 Pergamoo Press Ltd.

SELF-REPORTED DRUG-USAGE AND CRASH-INCIDENCE IN BREATHALYZED DRIVERS

R. D. MACPHERSON, J. F%RL and G. A. STARMER Department of Pharmacology, The University of Sydney, N.S.W. 2006, Australia

and

R. HOMEL School of Behavioural Sciences, Macquarie University, North Ryde, N.S.W. 2113, Australia

(Received 15 November 1982; in revised form 5 July 1983)

Abstract-Replies to a question on the medication usage of a large population of drivers subjected to evidential breath analysis were examined, and related to the age, sex and BAC of the driver, and to whether or not he was breath analysed after a crash. In an initial analysis, medications were classified into 13 major groups (including a drug negative, or control, group) and a log-linear analysis carried out on the cross-tabulation of age (five categories) by BAC (five categories) by drug (13 categories) by crash/no crash. (Analysis was restricted to males, since the number of females was very small). A reduced model was obtained, and the ratio of the odds of a crash in each drug group to the odds of a crash in the appropriate drug negative group computed. In a second stage of analysis, the analgesic and CNS depressant categories were expanded to individual agents. and odds ratios again computed. A number of individual drugs and drug groups were associated with an elevated crash risk. These included CNS depressants (diazepam, oxazepam, antidepressants), analgesics (d-propoxyphene) and drugs for the treatment of diabetes. In general, effects were most marked at low BAC’s.

INTRODUCTION

The full impact of drug use on driving behaviour and traffic crashes is at present unknown. Evidence that a problem exists stems from three different research approaches which demonstrate an apparent over-representation of certain drugs in the blood of traffic crash fatalities [Sterling-Smith, 1975; Cimbura et al., 19801, hospital admissions [Bonnichsen, 1975; Missen et al., 19781 and in the blood of drivers apprehended by the police [Alha et al., 19771. Alcohol had also been consumed by many of these drivers. For example, Lundberg et al. [1979] found that of 836 drivers apprehended by police, 765 were positive for drugs, alcohol or both. Drugs which were commonly detected included barbiturates, diazepam, methaqualone, chlordiazepoxide and meprobamate. Almost without exception these drugs and others detected in similar studies have been shown to be capable of impairing human perceptual, cognitive and motor performance either alone [Sharma, 1976; Linnoila, 1976; Clayton, 1976; Borland et al., 19751 or when combined with alcohol [Loomis and West, 1958; Mattila et al., 1977; Roden et al., 19771.

Moreover, Skegg, Richards and Doll [1979] reported a prospective study of 43,117 patients who were issued prescriptions by general practitioners. Crash related hospital admissions and deaths over a two year period were recorded. For 57 people injured or killed while driving cars, motor cycles or bicycles, the medicines dispensed in the previous 3 months were compared with those for 1425 matched controls. They found a highly significant association between the use of minor tranquillisers and the risk of a serious road accident.

The aim of this study was to extend the work of Skegg et al., [1979] in order to identify potentially hazardous drug-alcohol interactions. It was considered practicable to examine the replies to a drug usage question of a large population of drivers subjected to evidential breath analysis and to relate these to the age and sex of the drivers, the blood alcohol concentration found and whether or not a crash occurred. That is, our design involved comparing two major groups of drivers subjected to breath analysis: those who reported using medications (the drug positives) and those who did not (the drug negatives). The dependent variable was whether or not the driver was breath analysed after a crash, the aim of the analysis being to identify drugs which appear to lead to an elevated crash risk.

139

Page 2: Self-reported drug-usage and crash-incidence in breathalyzed drivers

140 R. D. MACPHERSON et al.

METHOD

Methodological problems There are three major methodological problems inherent in the approach adopted for

the present study. Firstly, in order to determine medication usage it is necessary to rely on information supplied by drivers subjected to breath analysis by the police. Secondly, the crash-medication correlations which emerge from the data apply to the population of breath analysed drivers, which is a non-random sample of the total population of motorists. Thirdly, the crash-medication correlations,are affected by the operation of many other variables (e.g. age, sex, blood alcohol level) which must be controlled in some way in order to establish the true influence of various forms of medications on the probability of a crash. We will deal with each of these problems in turn.

It must be admitted at the outset that self-report data suffer from the major disadvantage of unknown reliability. To the best of our knowledge, no studies have correlated self-report and analytical data for a large sample, although Honkanen et al. [1980] did find discrepancies between self-report and analytical data on the use of psychotropic drugs by a small number of crash victims and a control sample. Due to recent changes in legislation in New South Wales, it is possible to compare the roadside drug data with the results of analysis of the blood of drivers who attend hospital as the result of a crash. A large scale study has therefore been planned to compare self-report and analytical data in about one thousand of these drivers.

While in the present study absolute veracity cannot be guaranteed, steps were taken to limit misreporting. Questioning was carried out by trained non-uniformed police breathalyser operators who were never the arresting officers. It is standard practice for these operators to ask the motorist to produce the medicine container or prescription (if available) for verification. Moreover, it should be remembered that although it is illegal to drive under the influence of certain medications in New South Wales, the public is almost totally unaware of this fact and in practice enough rapport is established for information to be offered readily. Nevertheless, the method does select against the non-medical use of prescription drugs, and against illicit drugs.

No doubt both under-reporting and over-reporting occurred in our study. To the extent that such misrepre~nting occurs in a random or haphazard fashion, the statistical power of the comparisons of crash risks in drug positive and drug negative groups is reduced. In other words, the findings will be conservative and where significant results are obtained they will probably reflect real effects. In any case, the findings are presented to aid in the ordering of priorities for pha~acological research. Such expe~mentation on the basis of the present study has already begun [Per1 et al.].

The second and third problems-the restriction of analysis to the population of breathalysed drivers and the confounding effects of variables other than medication usage-turn out on closer examination to be intimately connected. A mathematical expression of the problem will clarify the nature of the connection.

Let B be the event that a driver is breathalysed, C the event that a crash occurs (whether the driver is subsequently breathalysed or not), and 0 the event that a driver has some characteristic or performs some action other than having a crash which, if observed, would lead to being breathalysed. The events C and 0 together constitute the universe of reasons for being breathalysed in N.S.W. during the period 1974-78. Given that a driver has been breathalysed, his probability of being involved in a crash is known and is denoted by

JYCIB). Using elementary probability theory, one can show that if two groups (say drug

positive (+) and drug negative (- )) are being compared, then:

The quantity on the left hand side of this equation relates to the breathalysed population only, since each probability is conditional on the event of being breathalysed (B). Within this population, it is the odds of having a crash in the drug positive group

Page 3: Self-reported drug-usage and crash-incidence in breathalyzed drivers

Self-reported drug-usage and crash-incidence in breathalyzed drivers 141

[P+(C/B)P+(O/B)] divided by the odds of having a crash in the drug negative group [P_(C/B)P_(O/B)]. Another name for this quantity is the odds ratio, a fundamental tool in the interpretation of cross-tabulations using log-linear models [Fienberg, 19801. Using the above formulation, an odds ratio greater than unity indicates a higher proportion of crashes in the drug positive than in the drug negative group.

The odds ratios for each of the sub-groups of the breathalysed population considered in the present study may be computed from the data. However, the quantities on the right hand side of the equation are not able to be computed directly. P, (C)/P_ (C) is in fact the quantity of primary interest, since it is the ratio of probabilities (or relative risk) of a crash in the drug positive and drug negative groups in the complete population of motorists. In studies of the effects of differing levels of alcohol on crash probability, the relative risk has been estimated by comparing matched groups of crash involved and non-crash involved drivers [OECD Road Research Group, 19781. The equation above suggests another approach utilizing police data, provided the final quantity on the right hand side can be estimated. This quantity may be termed a “police enforcement factor”, and reflects two things: the relative risk of committing some offence (0), being observed by the police and then breathalysed [(P(B and O)], as well as the relative risk of being breathalysed given that a crash has occurred (P(B/C)). Unfortunately at the present time, the values of this police enforcement factor for the drug positive and negative groups are not known, and could probably only be determined through close observation of police activity in conjunction with some kind of roadside survey.

It follows, therefore, that in order to minimize the dangers involved in inferring a high relative risk in the general motoring population from a high odds ratio in the breathalysed population, it is necessary to control as many factors as possible which affect police behaviour. The aim is, essentially, to create subgroups in which the police enforcement factor on the right hand side of the equation is unity. A beginning has been made in the present study by controlling age, sex and blood alcohol level. Future research will incorporate many more variables, including time of day, day of week, occupation of driver, vehicle type and vehicle age. Moreover, new procedures in N.S.W. require police to Alcotest all persons involved in a crash, theoretically eliminating P(B/C) as a factor.

Both the second and third problems are solved, or at least the difficulties of interpretation somewhat mitigated through the introduction of these statistical controls. The assumption is that factors which differentiate self-reported medication users from non-users (age, sex, time of day and so on) also influence the way police enforce drink-drive law. There is considerable evidence for this overlap in the two sets of factors [Homel, 19841. The real problem, of course, is that not all possible covariates can be measured and incorporated into the analysis. This problem is inherent in all non- randomized designs, and ultimately requires a judgement as to whether the most obvious and important covariates have been included before the computed correlations (or odds ratios) can be accepted as probably reflecting causal connections between drug usage and crashes.

Procedure Drivers who fail the roadside breath test in New South Wales are arrested and

subjected to evidential breath analysis (Breathalyzer 900). Prior to the determination of the blood alcohol concentration, drivers are questioned regarding their state of health and whether they have taken ‘tablets, drugs, insulin or medicine’. During the period 1974-78, the replies of a large population of drivers (n = 32,000) to the ‘drugs’ question were recorded together with details of age and sex, the breath analysis result and whether or not a crash occurred.

Since it was found that fewer than one per cent of the drivers were female they were excluded from the study. A total of 32 cases were identified where drug usage was denied and no alcohol was found on breath analysis. These cases were also excluded from the analysis.

The data were classified according to: (1) Drug-positive and drug-negative; (2) Age range (17-25, 26-35, 36-45, 46-55, > 55); (3) Blood alcohol concentration range

AAP Vol. 16. No 2-F

Page 4: Self-reported drug-usage and crash-incidence in breathalyzed drivers

142 R. D. MACPHERWN et al.

Table I(a). Major categories of drugs reported by motorists

All analqesxs 502

Chemotherapeutx agents 510

Antlhlstamine preparations 218

Anti arthritic drugs 165

Cardiovascular drugs 513

All CNS deoressant druas 1457

Antidlabetlc drugs 118

Drugs for the treatment of gout 82

Respiratory drugs 612

Drugs for the treatment of gastrx ulcers 156

Vitamin preparations 74

All other drugs 411

TOTAL DRUGS REPORTED 4816

NO DRUG REPORTED 10006

Percentaqe Of al1

drugs reported

10.4

10.6

4.5

3.4

10.6

30.2

2.4

1.7

12.7

3.2

1.5

8.5

100.0

Table l(b). Subcategories of analgesic and CNS depressant drugs

Drug percentage of all

drugs reported

Analgesics

Codeine

d-propoxyphene

other

CNS depressant drugs All Tranqurllzers

DlaZepam

0*?Xzepam

Chlordiazepoxide

Other

All Sedatives

Barbiturates

Methaqualone/Diphenhydramlne

Nitrazepam

Phe"obarbitone/Codelne

Other

Anti-depressants

107 2.2

86 1.8

309 6.4

1073 22.3

759 15.8

66 1.4

58 1.2

190 3.9

300 6.2

94 2.0

20 0.4

38 0.8

25 0.5

123 2.6

84 1.7

(0.005-0.075, 0.080-0.115, 0.120-0.155, 0.160-0.195, >0.200 g/100 ml); (4) Drug category (12 broad groups-see Table 1).

The drugs were further subdivided where appropriate into individual agents or types making a total of 29 sub-groups.

The data which were analysed include 48 18 drug-positive cases and 10,006 cases where drug usage was denied. These “control” cases were a random sample from the full drug-negative population and were considered to be sufficient to form a “control” group.

Method of analysis The data form a four way cross-classification: age (5 categories) by blood alcohol

concentration (5) by crash/no crash (2) by drug (29). These factors will be referred to as A, B, C and D respectively. The primary aim of the research was to investigate factors related to crash involvement (factor C), with particular emphasis on the role of drugs. Thus a logit analysis suggests itself, with C as dependent variable and A, B and D together with all their interactions as predictors. However, there is some interest in knowing which drugs are reported by people of different ages and blood alcohol levels, and this involves an investigation of the ABD interaction if the four way interaction ABCD is not significant. This can only be done by log-linear analysis of the data, treating it as a full four way cross-classification. The two methods of analysis are very closely related, and the results of both analyses will be reported where appropriate.

A factor with 29 levels in a four way table poses severe computational problems,

Page 5: Self-reported drug-usage and crash-incidence in breathalyzed drivers

Self-reported drug-usage and crash-incidence in breathalyzed drivers 143

particularly since many categories consisted of very few cases. To simplify analyses and interpretation, the drug factor was reduced initially to 13 categories (including the drug negative category). The three analgesic drugs were treated as one category, as were the 10 central nervous system (CNS) depressant drugs. Other drugs which involved 50 or fewer cases in all were grouped with the residual category (“unknown name or nature”). The drugs grouped with the residual were anorectic drugs (50 cases), drugs used in the treatment of epilepsy (44), migraine headaches (22) steroid drugs (43) and illicit drugs (32). In a second stage of analysis, the effects of individual analgesic and CNS depressant drugs were investigated.

The analysis was carried out using computer programs REG and ECTA [Francis, 19811. The theory of log-linear and logit models is set out in Bishop et al. [1975] and Nelder and Wedderburn [ 19721.

RESULTS

In the sample of 14,824 male motorists breathalyzed and found to have a positive BAC over the period 1974 to 1978, one third (33.8%) were tested after a crash. Slightly more than a third (36.4%) were aged 25 years or younger, compared with about 20% in this age

group in the motoring population. The modal blood alcohol level was 0.120-0.155g/100m1, with 17.4% below the prescribed level of 0.08 and 17.0% above 0.200 g/100 ml.

Two-thirds of the sample (67.5%) reported using no medications. Of these who did report using drugs at the time of testing, the category most frequently nominated was CNS depressants (particularly diazepam, anxiolytic agents and unspecified sedatives) followed by drugs for the treatment of respiratory disorders, cardiovascular diseases, antibiotics and analgesics (particularly unspecified pain killers). For details see Tables l(a) and l(b).

The complete four way table was inspected by computing, for each age and blood alcohol level combination, the ratio of the odds of a crash in each drug group to the odds of a crash in the drug negative group. Only a few of the odds ratios reached the five percent level of statistical significance, and it was difficult to discern clear patterns. However, three points deserve mention. Firstly there were no significant differences between any drug group and the control for blood alcohol levels exceeding 0.200 g/100 ml. Most meaningful differences occurred at the low blood alcohol levels. Secondly, tranquilizers were frequently associated with higher crash probabilities (at the lower blood alcohol levels). Thirdly, some drugs occasionally produced odds ratios significantly less than unity. These included antibiotics and sedatives at higher blood alcohol levels.

The preliminary analysis of the complete table suggested that a four way interaction may be present. However, its deviance was only 197.0 with 172 degrees of freedom (P = 0.09), so a reduced model was sought. The ABD interaction was highly significant when fitted in all positions (for example, when fitted first the deviance was 295.1 with 172 degrees of freedom, P < 0.001). This made the log-linear and logit reduced models equivalent, since the only problematic terms involved the factor C. The remaining three-way interactions (ABC, ADC, BDC) were fitted in all possible orders (adjusted for ABD), indicating that only ADC (the age by drug interaction in the logit model) could be omitted (deviance = 59.9 with 48 degrees of freedom, P = 0.12). The deviance for ABC, with 16 degrees of freedom, was 30.4 (P = 0.02) when adjusted for BDC, and 35.0 (P = 0.004) when not adjusted. This means that the effect of blood alcohol concentration on the odds of being tested after a crash depend on the age of the driver, a result consistent with the findings of research which compares matched groups of crash involved and non-crash involved drivers [OECD Road Research Group, 19781. The deviance for BDC, with 48 degrees of freedom, was 68.1 (P = 0.03) when adjusted for ABC, and 72.7 (P = 0.01) when not adjusted. This means that self-reported drug usage is associated with mode of apprehension, but that the effects vary with blood alcohol level.

Further simplification of the model was attempted by investigating the linear, quadratic, cubic and quartic components of age and blood alcohol level. The interaction of the quartic components of age and blood alcohol level yielded a deviance of 8.1 with one degree of freedom (P = 0.004), preventing simplification of the ABC interaction.

Page 6: Self-reported drug-usage and crash-incidence in breathalyzed drivers

144 R. D. MACPHERSON et al.

Fortunately, however, the BDC interaction reduced unambiguously to the linear com- ponent (deviance of linear component = 31.1 with 12 degrees of freedom, P = 0.001) which means that for a given drug, the relationship between blood alcohol level and odds of apprehension as a result of a crash was linear (in the log scale). Thus the reduced model consisted of the three-way interactions ABD, ABC and B (linear) DC and, of course, all main effects and two-way interactions. The residual deviance was 293.9 with 256 degrees of freedom, which corresponded to a just acceptable P value of 0.051.

Interpretation of blood alcohol by drug interaction Table 2 presents the interpretation of the BDC interaction in terms of odds ratios. For

a given drug and blood alcohol level, the odds of apprehension as a result of a crash are divided by the odds for the drug negative group at the same blood alcohol level. Since the ADC interaction was omitted from the model, these odds ratios are the same for each age group. However, they differ from the odds ratios which would be obtained by collapsing the full table over age). The statistical significances of the odds ratios were assessed at the 0.05,O.Ol and 0.00083 levels, this later figure being the Bonferroni level (0.05/60). The 0.05 or 0.01 levels seem appropriate since in an exploratory study, it seems desirable to commit Type 1 rather than Type 2 errors.

The CNS depressant drugs yielded highly significant odds ratios at blood alcohol levels up to 0.16. At the lowest blood alcohol level (under 0.08/ g/lOOml) the odds of apprehension as a result of a crash were nearly double the odds in the drug negative group. However, although they did not reach the same level of significance because of smaller numbers, the odds ratios for the drugs for the treatment of diabetes mellitus (up to a blood alcohol level of 0.12 g/100 ml) were among the highest in the whole study. Under 0.08 g/100 ml, the odds of a crash were 2.6 times higher than in the control. The third drug category which is a cause for concern is the analgesics, which (again at the two lowest blood alcohol levels) yielded odds ratios significantly greater than unity.

The trend for most drug categories was for higher blood alcohol levels to yield odds ratios not significantly different from unity. The odds of a crash were generally highest at the low blood alcohol levels, suggesting a “swamping” effect as blood alcohol concen- tration increases. That is, at high blood alcohol levels the addition of a drug did not materially alter the odds of being caught as a result of a crash. The exceptions to this rule were antibiotics (at blood alcohol levels exceeding 0.12 g/100 ml) and the residual category drugs (at levels of 0.20 g/100 ml and above), which yielded odds ratios significantly less than unity. This means that, especially for antibiotics, behaviour other than a crash was more likely than in the control group to attract police attention. In fact, the same statement could be made for about half the drug categories at the highest blood alcohol level, but it was only for antibiotics and the residual category that the trend reached significance.

The only other odds ratio in the table which was significant at P = 0.05 was for

Table 2. Ratios of the odds of a crash (drug: control) for major drug categories (reduced model)

orug .005 - .075

Analgesics 1.65.' Chemotherapeutic agents .9E! Antihistamines 1.31 Anti-arthritic 1.52 Cardiovascular 1.28 CNS depressants 1.94". Diabetes 2.56" Gout 2.68 Respiratory disorders 1.37 Gastric ulcers .70 Vitamins .99 All others 1.34

Blood alcohol concentration

.080 - .115 .120 - .155 .160 - .195

1.39" 1.17 .98 .87 .78.' .70"

1.16 1.03 .92 1.45 1.38' 1.31 1.23 1.18 1.13 1.64"' 1.383*** 1.17' 1.89" 1.40 1.04 1.81 1.22 .I33 1.18 1.00 .88 .78 .88 .99 .t?B .78 .69

1.13 .95 .I31

.200+

.83

.62*

.81 1.25 1.08 .99 .77 .56 .76

1.11 .61 .68*

l P < .05 t. p < .Ol

.** p < .05/60 = .00083

Page 7: Self-reported drug-usage and crash-incidence in breathalyzed drivers

Self-reported drug-usage and crash-incidence in breathalyzed drivers 145

anti-arthritic drugs at the middle blood alcohol level, but in the absence of significance at higher or lower blood alcohol levels it is difficult to interpret this result. However, several further features of the table should be highlighted. Firstly, as noted above, odds ratios generally declined with blood alcohol level. The only exception to this was drugs for the treatment of gastric ulcers, which yielded an odds ratio greater than unity only at the top blood alcohol level. Secondly, two categories (antibiotics and vitamin preparations) yielded odds ratios which at all blood alcohol levels were less than unity. Thirdly, although not statistically significant, drugs for the treatment of gout yielded the highest odds ratio in the table (2.68 at blood alcohol levels up to 0.08 g/100 ml) and exhibited the sharpest decline in odds ratios as blood alcohol increased.

Analysis of drug sub-groups As noted above, the analysis of all 29 drug categories simultaneously did not

recommend itself as a sensible procedure computationally. However, the prominence of the analgesic and CNS depressant drugs in Table 2 underlined the importance of considering the effects of the sub-groups of drugs in these two broad categories (see Table lb). The strategy adopted was to refit the reduced model, with the analgesic and CNS depressant categories expanded to cover all the sub-groups. This resulted in a drug factor with 25 categories, and a model with residual deviance of 648.5 with 471 degrees of freedom (P < 0.001). The lack of fit of the model to the expanded table highlights the complexity of the data, and suggests a return to the odds ratios from the full table with which we commenced the analysis. Nevertheless, the reduced model does clarify patterns and trends suggested by the cell-by-cell analysis, and appears to be a useful summarizing tool (see Table 3).

The most striking feature of Table 3 is the very high odds ratios at low blood alcohol levels for oxazepam and, to a lesser extent, the antidepressants. Although based on relatively small numbers (nine cases of oxazepam usage for the two lowest blood alcohol levels), the evidence seems very strong that there is greatly heightened crash risk among users of these drugs at low blood alcohol levels. Although also highly significant, diazepam was not associated with such high odds ratios. Unlike the CNS depressants, the analgesic sub-groups exhibited a consistent pattern and can be grouped together without much loss of information.

Table 3. Ratios of the odds of a crash (drug: control) for analgesic and CNS depressant drug subgroups (based on reduced model)

orug

Ardges1cs Codexne d-propoxyphene Other

CNS depressants Tranquilizers

Diazepam oxazepam Chlordiazepoxide Other

Anti-depressants

Sedatives Barbiturates Methaqualone/ Diphenhydramine

Nitrazepm Phenobarbitone/ Codeine

Other

Blood alcohol concentration

.005 - .075 .080 - .115 .120 - .155 .160 - .195 .200+

1.48 1.39 1.31 1.23 1.16 2.00 1.07* 1.75' 1.64 1.54 1.59 1.26 1.00 .80 .63

1.59.' 1.4P.9 1.37." 1.27' 1.18 14.13*** 7.03"' 3.509" 1.74 .86 1.23 1.05 .90 .77 .65 1.64 1.37 1.14 .94 .79

4.63'** 3.00*** 3.13*** 2.57*' 2.11

1.67 1.24 .92 .68 .51

3.02 .2.24 1.66 1.23 .91 3.28 2.36 1.71 1.23 .89

1.13 1.27 1.42 1.60 1.79

1.60 1.21 .92 .70 .53

. P ( .05 .* p < .Ol

l *. p < .05/65 = .00077

Page 8: Self-reported drug-usage and crash-incidence in breathalyzed drivers

R. D. MACPHERSON er a/.

The further terms in the reduced model (ABC and ABD) require some interpretation. The ABC interaction contirrned previous research [Homel, 1980; Homel, 19841 which showed that older men were most likely to come to police attention as the result of a crash. This pattern held for all blood alcohol levels, with some minor variations in the ordering of age groups. Not surprisingly, the odds of being apprehended after a crash generally increase with blood alcohol concentration, for all age groups, ahhough the age group profiles were far from parallel.

The ABD interaction was extremely complex. Nevertheless an attempt will be made to summarize its main features, since it is important to document variations in self-reported drug usage across age groups and blood alcohol levels. Firstly, the younger the driver, the more likely he was to report not using any medications. This pattern is consistent with commonsense expectations, and encourages confidence in the general reliability of self-report data. Secondly, however, there was a slight decline in reported drug usage as blood alcohol level increased, a trend which was more pronounced with young drivers. This could indicate a certain degree of under-reporting at higher blood alcohol levels, especially among the young.

Remarks about particular drugs will be restricted to those which figured prominently in Table 2. CNS depressant drugs were more likely to be used by older drivers (36 years of age and older), and reported usage was consistent across blood alcohol levels. Similarly, diabetes sufferers were invariably older, but reported use greatly deciined as blood alcohol level increased. Analgesics were much less likely to be reported at high blood alcohol levels, but no pattern with respect to age could be discerned. Finally, contrary to the main trends, antibiotic users were clearly young, with little variation by blood alcohol level.

DISCUSSION

It has been repeatedly demonstrated that alcohol increases the likelihood of a driver having a crash [e.g. Fox and Borkenstein, 19751, the nature of alcohoi-indu~d impai~ent of human performance is well-documented [e.g. Moskowitz and Murray, 19751 and a causal relationship between these effects is considered to exist.

In this study, this concept has been extended to include drug-alcohol interactive effects. All the drugs identified as increasing the probability of a driver being apprehended as a result of a crash (compared with that in the drug-negative group within the same blood alcohol concentration range) are those for which at least additive interactive effects with alcohol on human performance can be demonstrated. It is interesting to note that increased crash probabilities where they occurred were most readily demonstrated for drugs in association with “social” blood alcohol concentrations and that as the blood alcohol concentration increased, the influence of the drug appeared to decline. This is considered to be of social significance for drivers receiving medication who do not modify their drinking habits. At grossly impairing blood alcohol concentrations, the apparent influence of the drugs was minimal.

Since alcohol is a central nervous system depressant, interactiveeffects with other central nervous system depressants, which are at least additive, would be expected. This has proved to be the case with the results presented here and the odds ratios for having a crash were increased for the class as a whole and for diazepam, oxazepam, and the antidepressants. There is epidemiological data to support these findings for diazepam [Missen et al., 1978; Gelbke ef al., I9783 and also a number of human performance studies [Linnoila, 19731. Similar considerations apply to oxazepam [Palva and Linnoila, 19781, although the num- bers were small in the present study, and to the antidepressants [Milner and Landauer, 19711 especially during the first few weeks of therapy.

Failure to demonstrate an increased crash-risk for chlordiazepoxide is interesting in view of the conflicting evidence which exists concerning its interactive effects with alcohol. Some authors have even claimed that the drug is capabie of antagonising the effects of alcohol to some extent [Dundee and Isaac, 197 I]. Chlordiazepoxide, oxazepam and diazepam can be used interchangeably as minor tranquillisers and this lack of interactive potential with alcohol could be of importance to some patients.

Page 9: Self-reported drug-usage and crash-incidence in breathalyzed drivers

Self-reported drug-usage and crash-incidence in breathalyzed drivers 147

Apart from the central nervous system depressants, which were the largest group, other drug groups associated with a significantly increased crash probability where combined with alcohol included the antidiabetic agents and analgesic drugs, particularly dextropropoxyphene.

Oral antidiabetic drugs of the sulphonylurea type are known to interfere with alcohol and drug clearance [Truitt er al., 19601 and thus their identification as increasing crash- probability,is not surprising. The position of dextropropoxyphene is a little more obscure since although Kiplinger et al. [ 19741 found no interactive effects with alcohol on human performance, epidemiological findings indicate that the drug is apparently over-represented in the blood of traffic crash victims [Finkle er al., 19681.

Here, only the significantly increased odds ratios for crash involvement have been commented on but it should be noted that trends existed for the following drug groups and individual agents: anti-gout drugs, anti-arthritics, analgesics, codeine-containing preparations and methaqualone-diphenhydramine combinations. In all these cases there is evidence for interactive effects with alcohol. For other drug groups (such as vitamins) where no such laboratory evidence exists, no increased crash probability was identified.

It is fair to say that, from a pharmacological point of view, there are no surprises in the data. Nevertheless the methodological problems discussed earlier need to be kept in mind. Many variables related to drug usage and to police enforcement have been omitted from the present analysis. This may be particularly important for those drugs for which odds ratios less than unity were found. For example, as noted earlier, antibiotic users were younger than non-antibiotic users. This suggests that the low odds of a crash for antibiotic users in Table 2 could be partly a reflection of social factors correlated with age (given that age itself has been controlled in the model), such as vehicle type or time of day or night on the roads. In the comparison of subcategories within the same drug group the role of these confounding social factors is likely to be minimized (e.g. within the tranquilliser group) because the social characteristics of users (and hence patterns of police enforcement) are likely to be similar.

There is a further problem that the effects of drugs are confounded with the effects of the diseases or illnesses for which the medications were being taken. It is possible that the elevated crash rates reported in this paper reflect the effects of ill-health, and that the individuals concerned would have had even higher crash risks if they had not treated their illnesses. Pharmacological experiments are needed to shed light on this issue. However, it should be noted that in the real world drugs are consumed mainly by people with illnesses, and that from a traffic safety point of view the distinction between the effects of the illness and the effect of the drug may be of academic rather than practical interest.

Acknowledgemenrs-Thanks to Inspector A. S. Hammond of the New South Wales Police for providing access to police records and the Traffic Authority of N.S.W. for funding.

REFERENCES

Alha A. R.. Carlsson N., Linnoila M. and Lukkari I., Prevalence of drugs among drivers arrested for drinking and driving in Finland. 2. Rechrsmed. 79, 225-234, 1977.

Bishop Y. M. M.. Fienberg S. E. and Holland P. W., Discrete Mulfiuariate Analysis: Theory and Practice. M.I.T. Press, Cambridge, Mass.. 1975.

Bonnichsen R.. Aspects of drug analysis in relation to road traffic legislation and supervision. Proc. 6th Inf. Conf. Alcohol, Drugs and Traffic Safety. (Edited by S. Israelstam and S. Lambert) pp. 495-508. Addiction Research Foundation, Toronto, 1975.

Borland R. G., Nicholson A. M. and Wright C. M., Behavioural sequelae of methaqualone in man and in the monkey (Macaca Mulatta). Brit. J. Clin. Pharmac. 2, 131-141, 1975.

Cimbura G., Warren R. A., Bennett R. C., Lucas D. M. and Simpson H. M., Drugs detected in fatally injured drivers and pedestrians in the Province of Ontario, Ottawa. Trugfc Injury Research Foundation ofcanodu, 1980.

Clayton A. B., The effects of psychotropic drugs upon driving related skills. Human Factors 18, 241-252, 1976. Dundee J. W. and Isaac M., Interaction between intravenous alcohol, sedatives and tranquillisers. Med. Sci. Law

11,49-50. 1971. Fienberg S. E.. The Ana/.vsis of Cross-Classified Categorical Data. M.I.T. Press, Cambridge, Mass., 1980. Finkle B. F.. Biasotti A. and Bradlord L. W., The occurrence of some drugs and toxic agents encountered in

drink drivers investigated. J. Foren. Sci. 13, 236245, 1968. Fox B. H. and Borkenstein R. F., Patterns of blood alcohol concentrations among drivers, in Alcohol, Drugs

and Traffic Sqfery (Edited by S. Israelstam and S. Lambert) pp. 51-67. Addiction Research Foundation, Toronto. 1975.

Page 10: Self-reported drug-usage and crash-incidence in breathalyzed drivers

148 R. D. MACPHER~ON et al.

Francis I., StatisricaI Soffware: A Comparatioe Review. North Holland, Amsterdam, 1981. Gelbke H. P.. Schlicht H. J. and Schmidt G., Haufigbeit positive. Diazepam befunde in Blutproben

Alkoholisertev, Verkehrsteilnehmer. 2. Rechtsmedizin 80, 319-328, 1978. Homel ROSS. Penalties and the Drink/Driver: A Stud?; of One Thousand Ofinders. N.S.W. Bureau of Crime

Statistics Research, Sydney, 1980. Homel Ross, Young men in the arms of the law: An Australian perspective on policing and punishing the

drinking driver. Accid. Anal. & Prev. in press. Honkanen R., Ertama L., Linnoila M., Alha A., Lukkari I., Karlsson M., Kiviluoto 0. and Puro M., Role of

drugs in traffic accidents. Brir. Med. J. UIl, 1309-1312, 1980. Kiplinger G. G., Sokol G. and Rodda B. E., Effect of combined alcohol and propoxyphone on human

performance. Arch. Inr. Pharmacodvn 212, 175-180, 1974. Linnoila M., Effects of diazepam, chlordiazepoxide, thioridazine, haloperidol and fluphenthixate on psychomotor

skills related to driving. An. Med. Exp. Eiol. Fenn. 51, 125-132, 1973. Linnoila M., Tranquilizers and driving. Accid. Anal. & Prev. 8, 15-19, 1976. Loomis T. A. and West T. C., Comparative sedative effects of a barbiturate and some tranquilliser drugs on

normal subject. J. Pharmac. Exp. Ther. 122, 525-531, 1958. Lundberg G. D., White J. M. and Hoffman K. I., Drugs (other than or in addition to ethyl alcohol) and driving

behavior: A collaborative study of the California Association of Toxicologists. J. Forens. Sci. 24. 207-215. 1979.

Mattila M. J., Palva E., Seppala T. and Saario J., Effects of trithiozine on psychomotor skills related to driving: A comparison with diazepam and interactions with alcohol. Curr. Ther. Res. 22, 875-884, 1977.

Moskowitz H. and Murray J., The effect of alcohol on human information processing rate, in Alcohol, Drugs and Trajic Safery (Edited by S. Israelstam and S. Lambert) pp. 399403. Addiction Research Foundation, Toronto, 1975.

Milner G. and Landauer A., The effects of doxepin alone and together with alcohol in relation to traffic safety. Med. J. Aust. 1, 837-841, 1971.

Missen A. W., Cleary W., Eng L. and McMillan S., Diazepam, alcohol and drivers. N.Z. Med. J. 87, 275-277, 1978.

Nelder J. A. and Wedderbum R. W. M., Generalized linear models. J. Roy. Statist. Sot. Ser. A. 135. 37&384, 1972.

O.E.C.D. Road Research Group, New Research on the Role of Alcohol and Drugs in Road Accidents, 1978. Palva E. S. and Linnoila M., Effect of active metabolites of chlordiazepoxide alone or in combination with

alcohol on psychomotor skills related to driving. Eur. J. Clin. Pharmac. 13, 345-350, 1978. Per1 J., Starmer G. A. and Bird K. D., The interactive effects on human psychomotor performance of ethanol

with dextropropoxyphene and/or diflurisal (in preparation). Roden S.. Harvev P. and Mitchard M.. The influence of alcohol on the persistent effects on human performance

of the hypnotics, mandrax and nitrazepam. Inf. J. Clin. Pharmac. iiopharmac. 15, 350-355, 1977. Sharma S., Barbiturates and driving. Accid. Anal. & Prev. 8, 27-31, 1976. Skegg D. C. G., Richards S. M. and Doll R., Minor tranquillizers and road accidents. Brir. Med. J. 1, 917-919.

1979. Sterling-Smith, R. S., Alcohol. marijuana and other drug patterns among operators involved in fatal motor

vehicle accidents, In Alcohol, Drugs and Traffic Safety (Edited by S. Israelstam and S. Lambert) pp. 93-105. Addiction Research Foundation, Toronto, 1975.

Truitt E. B. Jnr.. Puritz G. and Morgan A. M., Disulliram-like actions produced by hypoglycaemic sulphonylurea compounds. Q.J. Smd. Ale. 23. 197-207, 1960.