Schaffer Online Library of Drug Policy Sign the Resolution
Contents | Feedback | Search
DRCNet Home
| Join DRCNet
DRCNet Library | Schaffer Library | Drugs and Driving 


Performance Comparison of the Intoxilyser 5000 with Blood Tests - A South African Study

TC Gilfillan

CSIR Information Services, PO Box 395, 0001 Pretoria, South Africa


An experiment is in progress to establish the relationship between breath samples analyzed by the Intoxilyser 5000 and standard blood tests, over blood alcohol levels in excess of the South African statutory limit of 0.08%. The paper describes the data collection design adopted to ensure the statistical validity of the relationship for at least three groups of alcohol consumers, namely, in BAC units: (0.08%, 0.12%), (>0.12%, 0.16%) and (>0.16%). The influence that various personal characteristics such as age, gender, race, and available physical attributes, may have on the relationship will also be established. The reasons for the adoption of this particular design will also be given.


Early in 1994 the Department of Justice of South Africa requested a thorough scientific study to justify the proposed use of breath samples to measure Breath Alcohol Concentration (BrAC) in place of the Blood Alcohol Concentration (BAC) measure, for driving under the influence (DUI) prosecutions. A study was proposed to establish whether a common relationship (conversion factor) exists between standard BAC results and the Intoxilyser 5000 BrAC readings, over BAC levels in excess of the statutory limit of 0,08%. To ensure the statistical validity of the relationship, the study was planned to collect data for three groups of subjects, with BAC levels within (0,08%, 0,12%), (>0,12%, 0,16%) and (>0,16%) respectively. The study was also designed to establish the influence that various personal characteristics such as age, gender, race, and available physical attributes, may have on the relationship.

Various European countries have opted for a range of blood/breath relationships: Great Britain and Holland (2300:1), Austria (2000:1), Norway and Sweden (2100:1). In the USA, Canada and Australia, where breath-analyzers have been in use for many years the 2100:1 factor is used (see Jones [1990] for further details).

In South Africa the introduction of new procedures to obtain evidential information is vigorously challenged in the courts, usually on technical grounds. It is for this reason that a search of the literature was undertaken to identify factors that could influence the blood/breath relationship. Factors found in the literature (Jones [1990] and Taylor [1981]) were built into the study design in order to anticipate grounds for challenging the introduction of BrAC for evidential purposes.

The objective of the study is to recommend a conversion of Intoxilyzer 5000 readings to corresponding blood test values, with adjustments, where necessary to take account of personal factors.


A study was conducted of the available literature, primarily to identify factors that could influence the relationship between the BrAC and BAC measurements. Where possible these would be incorporated in the study, so that, should they be challenged in court their impact, if any, on the relationship adopted from this study, can be explained.

Three key factors were identified from the paper of Jones [1990]:


A study comparing individuals with and without chronic pulmonary disease showed that the average blood/breath relationship for the age-matched control group (average age 65 years) was 3051:1. Another study of a group of younger healthy subjects (average 33 years) gave an average ratio of 2283. "There was a positive association between blood/breath ratio of alcohol and increasing age of the subject tested, although the data presented were not very convincing" (p 18-19).;


"the first pass effect ", where "a part of the dose of alcohol a person drinks is removed from the body before it reaches the general circulation" .... "is less evident in women compared to men" (p 8), and

the average blood/air partition ratios for dilute solutions of ethanol in specimens of whole blood obtained from healthy men and women suggest that female blood "will have a lower concentration of alcohol in the available water fraction and therefore a lower concentration in an air or vapor phase equilibrated with it", possibly due to lower Hematocrit values in female blood (p 12 and Table 5 on p 13).;


"The rate of disappearance of alcohol from blood is normally calculated from the slope of the rectilinear elimination phase of the BAC time-profile. By tradition, this is denoted as Widmark's beta factor. Beta factors vary among individuals, depending in part on genetic and racial differences,...." (p 7).

The height and mass of subjects were identified as measures of the physical attributes of subjects. These are easy to record and will reflect the physical variation between subjects.

As is shown later, these five factors are used in the design of the study as, so-called, control factors.

Further factors that cannot be controlled, but nevertheless can be easily monitored, were identified from the literature. These are:

the arm from which blood is drawn:

"Differences in alcohol concentration may also exist between left-right arm veins.."(p 7);

the health of the subject:

"Factors that elevate body temperature, such as a fever, might be expected to cause a rise in breath temperature and therefore in the expired breath-alcohol concentration" (p 17);

ambient temperature:

"The temperature and humidity of the ambient air a person breathes is an important consideration when quantitative evidential breath-alcohol measurements are made" (p 18). Humidity does not vary sufficiently in the Highveld where the study is being done, to justify its monitoring.

As smoking may influence lung performance the smoking habits of the subject, such as the number of cigarettes smoked daily, and whether smoking preceded apprehension, is a further factor that is monitored.

Other information routinely collected is:

All the monitored factors, and the routine information will be used to explain anomalies in the data. Introduced through indicator variables their influence on the blood/breath relationship will be determined.

A factor of interest is the altitude at which the study is being conducted. In Gauteng province the altitude ranges from a 1370 metres (4500 feet) above sea level to 1750 meters (5740 feet). According to Taylor [1981] "the presumed alveolar air-to-blood ratio of 1:2100 can be further in error if the test is administered at higher altitudes. In Denver, for example, it has been found that the 5,000 foot altitude significantly effects test results: the breath-alcohol reading will increase as altitude increases." If an average significantly higher than 2100:1 is found for this study it may be a consequence of altitude. It would require a comparative study to confirm such a result.


In implementing the study a control sheet was developed to include the five control factors. The objective is to ensure that the study will continue until all combinations of the five factors are represented in the data set. Each factor is considered at two levels:

There are 32 combinations (2^5) of five factors, each at two levels. To monitor variation within a category, at least three subjects are required. Therefore 96 subjects, falling into 32 distinct categories, are required for each of three groups of subjects.

The BAC statutory limit in South Africa is 0,08%. The three groupings are of subjects with BAC within (0.08%, 0.12%), (>0.12%, 0.16%) and (>0.16%). The ratio can then be determined for each group.

Therefore the first group will consist of subjects with BAC levels at or exceeding the statutory 0.08%, but not more than 0.12%. One category will have three or more subjects that are older than 30 years, male, taller than 1.7 metres, heavier than 75 kilograms, and all Caucasian. Each of the other 31 combinations will also contribute at least three subjects to this group.

A minimum of 288 subjects will therefore be in the study. For practical reasons not all the combinations will be forthcoming from DUI cases. We expect some categories to be heavily over-subscribed while others will simply not feature. Our experience is that few women commit DUI offences. Also we do not expect many subjects within the first group.

To make up the numbers we are enroling volunteers to consume alcohol under controlled conditions.


This is a highly structured data set, known to statisticians as 2^5 factorial design.

In addition to an estimate of the blood/breath ratio that has been balanced over all five factors, the structure of the data allows us to interrogate it for the effect of each factor on the ratio estimate, independent of the remaining factors.

For the study being conducted in South Africa, we will estimate, and test for statistical significance, the effect of each of the five factors on the ratio, and the effect of ten interactions between all pairs of factors.

For each group of subjects we will have a group average ratio, five factor effects and ten interaction effects.

A statistical analysis of the combined data can tell us whether the group averages differ significantly from the average for all the data, and whether the five factors and ten factor interactions are significantly different from zero.

The hypothesis we will be testing is that the three group averages do not differ from the grand average, and that the factor effects and the factor interactions are zero.


The effectiveness of the study depends on the spread of age, height and mass within each combination. Where categories have more than the minimum of three subjects, subjects will be selected for the analysis that demonstrate as wide a spread as possible over the three factors.

For example consider the category:

older than 30 years, male, taller than 1.7 metres, heavier than 75 kilograms, all Caucasian.

We would want the three ages to be about 10 years apart, the heights differing by .15 metres and the masses differing by 5 kilograms. The three subjects could, for example, be

35 yr, 1.8 m, 90 kg; 40 yr, 2.1 m, 85 kg; 45 yr, 1.95 m, 80 kg;

giving a mix over the three factors. Provided the data base is sufficiently large it should be possible to select combinations with a healthy spread in the factors.

In the selection of volunteers every effort will be made to ensure that the required spread is realised.


A more informative alternative to published studies, such as that conducted in the United Kingdom in 1984 (Cobb et al [1985]), is in progress in South Africa. For the UK study, involving over 40000 subjects, the vast majority of subjects were men [p 37]. In contrast the considerably smaller South African study will involve both genders in equal numbers. As a number of additional factors are included in the study, it will be possible to establish the influence of a wide range of factors on the blood to breath ratios. In addition the study will be conducted above sea level, at an altitude above 13 000 metres.


Cobb, P G W, Dabbs, M D G (1985). Report on the Performance of the Lion Intoximeter 3000 and the Camic Breath Analyser evidential breath alcohol measuring instruments during the period 16 April 1984 to 15 October 1984. His Majesty's Stationary Office, London.

Jones, A W (1990). Physiological Aspects of Breath-Alcohol Measurements. Alcohol Drugs Driving; 6:1-25.

Taylor, L E (1981). Drunk Driving Defense. Little, Brown and Company, Boston, Toronto.