Recent federal rules and traffic law changes impose breath-alcohol thresholds of 0.02 and 0.04 ~/210 L upon some classes of motor vehicle operators, such as ...
Journal of Analytical Toxicology, Vol. 23, October 1999
Measurement of Low Breath-AlcoholConcentrations: Laboratory Studiesand Field Experience Kurt M. Dubowski and Natalie A. Essary
The University of Oklahoma Health Sciences Center, Departmentof Medicine and ForensicScience Laboratories,P.O. Box 26901, Oklahoma City, Oklahoma 73190-3000 and Boardof Testsfor Alcohol and Drug Influence, Stateof Oklahoma
I A rad I Recent federal rules and traffic law changes impose breath-alcohol thresholds of 0.02 and 0.04 ~/210 L upon some classesof motor vehicle operators, such as juveniles and commercial vehicle operators. In federally regulated alcohol testing in the workplace, removal of covered workers from safety-sensitive duties, and other adverse actions, also occur at breath-alcohol concentrations (BrACs) of 0.02 and 0.04 ~/210 L. We therefore studied performance of vapor-alcohol and breath-alcohol measurement at low alcohol concentrations in the laboratory and in the field, with currentgeneration evidential analyzers. We report here chiefly our field experience with evidential breath-alcohol testing of drinking drivers on paired breath samples using 62 Intoxilyzer 5000-D analyzers, for BrACs of 0-0.059 g/210 L. The data from 62 law enforcement breath-alcohol testing sites were collected and pooled, with BrACs recorded to three decimal places, and otherwise carried out under the standard Oklahoma evidential breath-alcohol testing protocol. For 2105 pooled simulator control tests at 0.06-0.13 g/210 L the mean + SD of the differences between target and result were -0.001 + 0.0035 g/210 L and 0.003 _+0.0023 g/210 L for signed and absolute differences,respectively (spans-0.016-0.010, 0.000-0.016). For 2078 paired duplicate breath-alcohol measurements with the Intoxilyzer 5000-D, the mean + SD difference (BrAC 1 - BrAC 2) were 0.002:1:0.0026 (span 0-0.020 g/210 t). Variability of breathalcohol measurements was related inversely to the alcohol concentration. Ninety-nine percent prediction limits for paired BrAC measurements correspond to a 0.020 g/210 L maximum absolute difference, meeting the NSC/CAODrecommendation that paired breath-alcohol analysis results within 0.02 g/210 L shall be deemed to be in acceptable agreement. We conclude that the field system for breath-alcohol analysis studied by us can and does perform reliably and accurately at low BrACs.
Introduction
Breath-alcohol 1 analysis is a mature technology. Breath-alcohol analyzers have been commercially available and successfully used for about 60 years for applications in traffic law enforcement, the clinical diagnosis of alcoholic intoxication, and biomedical I The unmodified term "alcohol" in this article refersto ethanol.
386
research. The technology and the scientific basis of breath-alcohol analysis have been extensively studied and are well understood and established (1). Breath-alcohol testing was initially focused on the relatively high breath-alcohol concentrations (BrAEs) often found in both clinical and forensic settings. Hence, the great majority of the many extant published evaluations and studies of breathalcohol analyzers and their performance were carried out at such typical BrAEs as 0.08, 0.10, 0.15 g/210 L and greater. Three developments in recent times have involved numerous BrAE measurements at lower alcohol concentrations: (1) Increasing use of breath-alcohol analysis for research on the behavioral effects of alcohol and alcohol-related impairments after low or moderate intake of alcohol; (2) recent nationwide trends in traffic law legislation and enforcement practices to lower permissible alcohol concentrations (e.g., "zero tolerance") for motor vehicle operators; and (3) large-scale, federally regulated breath-alcohol testing in the workplace--especially the transportation workplace--pursuant to the mandate of P.L. 102-143, 1991 (2). Current traffic laws typically define alcohol-elements of traffic offenses and the legal consequences at BrAEs of 0.02, 0.04, 0.05 g/210 L or "any detectable alcohol concentration" for some classes of motor vehicle operator, such as commercial motor vehicle drivers and persons under 21 years of age. The pertinent federal (DOT) regulations for alcohol testing in the workplace also mandate removal of covered workers from safety-sensitive duties and entail other adverse personnel actions at BrAEs of 0.02 and 0.04 g/210 L for about 8 million workers in transportation industries. Field performance of evidential breath-alcohol measurement is also pertinent to the workplace setting because some federal rules make the results of breathalcohol tests conducted by federal, state or local law enforcement agencies and conforming to the applicable local requirements acceptable in lieu of employer-performed testing (3). We, therefore, examined the characteristics and reliability of breath-alcohol measurements at the relevant low concentrations between 0 and 0.059 g/210 L, and evaluated the performance of several current-generation quantitative evidential breath-alcohol analyzers in the laboratory and of such analyzers in the field. The studies of field performance involved the devices and procedures used for forensic breath-alcohol analysis in Oklahoma for traffic law enforcement and related purposes. This report summarizes our findings in the laboratory evaluations of three current-
Reproduction (photocopying) of editorial content of this journal is prohibited without publisher's permission.
Journal of Analytical Toxicology, Vol. 23, October 1999
generation analyzers and in the field testing of suspected drinking drivers in Oklahoma, both focused on low alcohol concentrations. For clarity, we have stated our findings and discussion in 3 categories: (1) In-vitro laboratory studies, (2) field experience in the in-vitro measurement of targeted vapor-alcohol concentrations, and (3) field experience in the in-vivo measurement of subject breath-alcohol concentrations. Throughout this article, VAC means in-vitro vapor-alcohol concentration (simulator test), g/210 L and BrAC means breath-alcohol concentration (human subject test), g/210 L.
Experimental Breath-alcohol analyzers Production models of the following instruments were employed without modification: BAC DataMaster (National Patent Analytical Systems, Mansfield, OH), Intoxilyzer 1400, and Intoxilyzer 5000-0 with the Oklahoma Software Program (CMI,Inc., Owensboro, t9/). All three devices are quantitative evidential breath-alcohol analyzers appearing in the National Highway Traffic Safety Administration's Conforming Products List for evidential analyzers (4), and each employs infrared spectrometry for measuring alcohol in breath. Every individual instrument in the laboratory or field was coupled with an alcoholic-breath simulator appearing on the relevant NHTSAConforming Products List for calibrating devices (5), for control tests and instrument maintenance. Sixty-two production models of the Intoxilyzer 5000-D and 62 production models of Guth models 34 or 10-4 simulators were used in the field. All had been laboratory-tested and validated prior to field deployment. Simulators Vapor-alcohol samples were generated with model 34 or model 10-4 Alcoholic Breath Simulators (Guth Laboratories, Harrisburg, PA) or TOXITEST II Alcohol Breath Simulators (CMI, Inc., Owensboro, KY), used without modification. All simulators were used in the effluent recirculating mode except with the Intoxilyzer 1400. Simulatorsolutions Simulators were charged with 0.5 L of aqueous alcohol soluTable I. Steps of the Oklahoma Forensic Breath-Alcohol Analysis Step 1 2 3 4 5 6 7 8 9 10
Element Room air blank Breath test 1 Room air blank Two-minute interval Room air blank Breath test 2 Room air blank Vapor-alcohol control test Room air blank Printout of results
tions prepared and verified to yield the various target vaporalcohol concentrations (VACs) by equilibration of room air with alcohol solution at 34~ The alcohol concentration of all simulator solutions used in the laboratory was verified in our laboratory. All simulator solutions for forensic breath-alcohol testing in Oklahoma were centrally prepared by the laboratory staff of the Oklahoma State Bureau of Investigation, and were validated independently by the OSBI Central Laboratory and by our laboratory. Both laboratories used headspace gas chromatography, as previously described (6), and target VACvalues were established as previously described (7). The accuracy of simulator-generated VACs was also confirmed for all individual lots of simulator solution by replicate analyses with a reference Intoxilyzer 5000-D and reference simulators maintained in our laboratory. Five-hundred milliliter aliquots of individual lots at any given target VAC are distributed to varying portions of the 220 field testing locations.
Breath-alcohol and vapor-alcohol measurements In-vitro vapor-alcohol measurements were carried out in the laboratory as previously described (7), using alcoholic breathsimulators with verified factory settings. The accuracy of low VACsthus generated in our laboratory was previously established by direct comparison with vapor-alcohol dry gas ethanol standards in the form of NIST-certified Research Gas Mixtures (RGMs) at VACs corresponding to 0.0375 and 0.0760 g/210 L, respectively (8). In the laboratory, VACswere measured by 50 within-run replicate measurements for each of the 3 instruments, at each tested VAC (0, 0.01, 0.02, 0.03, 0.04, 0.05, and 0.06 g/210 L). Each of the 50 replicates was paired with a control test at VAC = 0.10 g/210 L. The number of replicate tests at positive VACs, therefore, was 600 measurements per analyzer, for a grand total of 1800. All field breath-alcohol measurements were carried out in full compliance with State of Oklahoma regulations (9) and in accordance with official Breath-Alcohol Analysis Operating Procedures promulgated by the State Director of Tests for Alcohol and Drug Influence (KMD), which have the force of law in Oklahoma. Laboratory and field BrAC and VAC measurements and other experimental work were carried out in accordance with all pertinent safety considerations and in compliance with recognized standards of good laboratory practice, to the extent applicable. Results of VACand BrAC measurements are reported in this study in g/210 L to three decimal places (e.g., 0.050 g/210 L). For forensic purposes, VAC and BrAC measurements in the field are reported to two decimal places, truncated, in Oklahoma (9). The Oklahoma scheme for BrAC and control test measurements with the Intoxilyzer 5000-D in the field follows the ten step sequence shown in Table I. This is preceded by a preanalytical phase for subject and device preparation and a pretest deprivation-observation period of at least 15 min for human subjects (10). VAC measurements in the field (control tests) were performed with 62 sets of instruments. The number of VAC measurements at any given target VAC varied among analyzers. Breath-alcohol analysis in the field was performed on two paired separate consecutive breath samples collected within 3-4 min of each other. All VAC and BrAC measurements were performed in the field with identical analysis protocols and equipment and by identically trained operators and supervisors. 387
Journalof AnalyticalToxicology,Vol. 23, October 1999
Data treatment and statistical analysis The data obtained in the laboratory study of the three evidential infrared breath-alcohol analyzers were collected and examined statistically for each analyzer separately, and all results (n = 1800) obtained in this in-vitro study are included in the data base. They were also pooled to determine the correlation of the grand means of VAC measurements with the seven VACtarget values. The field data base includes 2105 VAC measurements (i.e., control tests) at 8 target VACsshown in Table II and 2078 paired BrAC measurements shown in Table III. The latter consists of all pairs of BrAC measurements, from 62 separate Intoxilyzer 5000-D instruments, which had a BrAC 1 of 0--0.059 g/210 L and an absolute difference BrAC 1 - BrAC 2 not greater than 0.02 g/210 L and which were accompanied by a control test result within • 0.01 g/210 L of the established VACtarget value. Both the VACand BrACfield data have peer status, and all were obtained under substantially identical conditions, warranting use of pooled data treatment. The control test results for a given VAC target value from all field instruments were combined for statistical analysis, as were the 2078 paired BrAC measurements from all field instruments. Statistical data treatment was performed by standard statistical methods (11-14), using STATGRAPHICS| Plus for Windows, Version 2 (Manugistics, Inc., Rockville, MD), with a microcomputer for both descriptive statistics and tests of significance.
Results For clarity, the results of the laboratory in-vitro studies and of the field control and subject tests are reported separately. Laboratory in-vitro measurementsof vapor-alcohol concentrations Descriptive statistics of representative data for accuracy and repeatability of within-run measurements of simulator-generated vapor-alcohol samples, in the laboratory setting, are shown in Tables IV-VI for the BAC DataMaster, Intoxilyzer 1400 and Intoxilyzer 5000-D, respectively. All "blank" analyses of room air between alcohol-containing samples yielded 0.000 g/210 L results with all three instruments. Results for the VACs omitted from Tables IV-VI for brevity (0.01, 0.03, and 0.05 g/210 L) were statistically comparable to those shown, and all of the results were normally distributed within a given VAC interval. The derived limits Table II. Accuracy and Reproducibility of Between-Run Vapor-Alcohol Measurements with the Intoxilyzer 5000-D--Pooled Field Data (n : 2105) n 21 13 260 367 568 296 536 44
388
Vapor-Alcohol Concentration, ~/210 t Target Mean SD CV% Median Mode 0.06 0.07 0.08 0.09 0.10 0.11 0.12 0.13
0.063 0.071 0.082 0.091 0.101 0.111 0.121 0.130
0.0019 0.0018 0.0034 0.0024 0.0036 0.0046 0.0036 0.0027
2.9 2.6 4.1 2.7 3.5 4.1 3.0 2.1
0.064 0.072 0.083 0.091 0.101 0.111 0.122 0.131
0.064 0.071 0.085 0.090 0.100 0.110 0.123 0.131
Span 0.061-0.069 0.068-0.074 0.071-0.091 0.082-0.105 0.090-0.111 0.101-0.126 0.110-0.130 0.125-0.135
of detection (LOD) and of quantitation (LOQ) for each of the three tested analyzers appear in Table VII, calculated as 3 x SD0 and 10 x SD0 at VAC = 0 g/210 L in keeping with usual analytical chemistry practice (15). The SD0 was obtained separately for each analyzer as the y-intercept value for the least squares linear regression of SDx upon each of the several target VACs. For the combined data (n = 1800) for all three analyzers, grand means of the results at each target VAC were calculated. Their correlation with the respective target VACsis shown in Figure 1, reflecting the accuracy and repeatability of the pooled laboratory results for the three tested analyzers as a class. The pooled data linear regression equation for grand mean measured VAC upon target VACwas y = 1.013x - 0.0009 g/210 L, with SEE = 0.0006 and Pearson correlation coefficient r = 0.99 and coefficient of determination r ~ = 99.95%. ANOVAyielded a P-value less than 0.01 indicating a statistically significant relationship between Table III. Distribution of Measured Breath-Alcohol Concentrations Between 0 and 0.059 g/210 L--Pooled Field Data (n = 2078) Breath-Alcohol
concentration span, g/210 L
Percent of low BrAC tests
n
0-0.009 0.010-0.019 0.020-0.029 0.030-0.039 0.040-0.049 0.050-0.059
74 288 341 394 484 497
0-0.059
2078
3.5 13.9 16.4 19.0 23.3 23.9 100
Table IV. Accuracy and Repeatability of Within-Run Vapor-Alcohol Measurements with the BAC DataMaster--Laboratory Data Vapor-Alcohol Concentration, g/210 L n Target Mean 50 50 50 300
0.02 0.04 0.06 0.10 Control
0.020 0.040 0.060 0.102
SD
CV%
0.0004 0.0005 0.0005 0.0020
2.0 1.3 0.8 2.0
SEM
Systematic Span
error,%
0.0001 0.019-0.021 0.0001 0.040-0.041 0.0001 0.059-0.061 0.0001 0.097-0.107
0 0 0 +2.0
Table V. Accuracy and Repeatability of Within-Run Vapor-Alcohol Measurements with the Intoxilyzer 1400--Laboratory Data Vapor-AlcoholConcentration,g/210L n Target Mean 50 50 50 300
0.02 0.04 0.06 0.I0 Control
0.019 0.039 0.058 0.I00
SD
CV%
0.0005 0.0006 0.0007 0.0010
2.5 1.5 1.2 1.0
SEM
Systematic Span
error,%
0.0001 0.018-0.019 0.0001 0.037-0.040 0.0001 0.057-0.060 0.0001 0.097-0.103
-5.0 -2.5 -3.3 0
Journalof Analytical Toxicology, Vol. 23, October 1999
target and measured VACsat the 99% confidence level. Field in-vitro measurements of vapor-alcohol concentrations
The descriptive statistics for the 2105 vapor-alcohol measurements (= control tests) in the field at eight VACtarget concentra0.12
.
. . . . . . . . . . . . .
0.1
y = 1.013X- 0.0009 /210 L
,.ooo
r = t.99
/
0.08
/
o 0.0.6 ~
o
0.04
E
0.02
/
/
tions are shown in Table II, summarizing accuracy and a betweenrun reproducibility of field control tests. Linear regression of these pooled control test results upon the corresponding target VACsyielded the equationy = 0.984x + 0.0030 g/210 L, with SEE = 0.0035, r = 0.97 and r 2 = 94.3%. Figure 2 is a scatterplot of the correlation of 210.5 pooled control test results with the respective VAC target values. Since the P-value obtained by ANOVAis less than 0.01, there is a statistically significant relationship between target and measured VACsin the field at the 99% confidence level. All "blank" analyses of room air between successive alcohol-containing samples yielded 0.000 g/210 L. Limits of detection and of quantitation for the Intoxilyzer 5000D instruments in the field calculated from the SD0 for this system, derived as described above, appear in Table VIII. The correlation of the mean results of the pooled control tests at each of the eight target VACswith the corresponding VACtarget value is shown in Figure 3, reflecting the accuracy of these field VACmeasurements as a class. The linear regression equation for this latter system wasy = 0.998x + 0.0012 g/210 L, with SEE = 0.0009, r = 0.99 and r 2 = 99.9%. For 2105 field VAC measurements, the signed and absolute differences between Target VACand Control Test Result were determined. Descriptive statistics for those differences appear in Table IX. The mean + SD for the absolute difference were 0.0030 +
0 i
i
i
i
i
i
0.02
0
i
i
f
0.04
i
i
i
0.06
r
i
0.08
Table VI. Accuracy and Repeatability of Within-Run
[
0.1
Vapor-alcohol target concentration (g/210 L) Figure1. Correlationof three-instrumentsgrand meansof vapor-alcohol measurementswith VAC target values--laboratory data.
0.15 y = ].984x~!0.0030 /210 L S E E = 0.0035 0.12
, - ;.;7
50 50 50 300
0102 0.04 0.06 0.10 Control
0.019 0.039 0.059 0.100
0.0007 0.0006 0.0006 0.0009
3.5 1.5 1.0 0.9
0.0001 0.0001 0.0001 0.0001
0.017--0.020 --5.0 0.039-0.041 --2.5 0.058--0.061 --1.7 0.099-0.103 0
Table VII. Derived Standard Deviation of Vapor-Alcohol Measurements at VAC = 0 g/210 L, and Calculated Limits of Detection and Quantitation for the Listed Alcohol
0.09
Analyzers--Laboratory Data Breath alcoholanalyzer
0.06
BAC DataMaster Intoxilyzer ] 400 Intoxilyzer 5000-D
e
8 o.0a
Systematic Vapor-AlcoholConcentration,~/210 L n Target Mean SD CV% SEM Span error,%
/
n = Z105
o
Vapor-Alcohol Measurements with the Intoxilyzer 5000-DELaboratory Data
/ i
Vapor-AlcoholConcenlrationg/210 L SD0atVAC=0 LOD=SD0x3 LOQ=SD0xl0 -0.00003 0.00032 0.00062
0.0001 0.0009 0.0018
0.0003 0.0032 0.0062
Table V l l l . Derived Standard Deviation of Vapor-Alcohol Measurements at VAC = 0 g/210 L, and Calculated Limits i
i
i
i
i
i
i
0.03 0.06 0.09 0.12 0.15 Vapor-alcohol target concentration (g/210 L) 0
of Detection and QuantitationmPooled Intoxilyzer 5000-DEField Data (n -- 2105) Vapor-AlcoholConcentrationg/210 L SD0 at VAC = 0
LOD = SDo x 3
LOQ = SD0 x 10
0.000663
0.0020
0.0066
Figure2. Correlationof control test resultswith vapor-alcohol target concentrations with the Intoxilyzer 5000-D--pooled field data (n = 2105).
389
Journal of Analytical Toxicology,Vol. 23, October 1999
0.0023 g/210 L, and the absolute difference span was 0-0.016 g/210 L. When measured VACswere truncated to two decimal places, for example, 0.04 g/210 L, least squares linear regression of 2105 measured VACs upon target VhCs yielded y = 0.976x - 0.0001 g/210 L, with SEE = 0.0046, r = 0.95 and r 2 = 90.5%. The mean • SD of the absolute difference for the 2105 truncated Target VACControl Test Result were 0.0028 • 0.0045 g/210 L, with SEE = 0.00009, median = 0, mode = 0, and span of 0-0.01 g/210 L. Of the absolute differences of the truncated VACs, 72% were 0 g/210 L and 100% were 0.01 g/210 L or less. Field breath-alcohol measurements From all field BrACmeasurements over a 9-month period, 2078 paired low BrAC results constitute the data base for field breathalcohol measurements, with the distribution shown in Table III. All '~olank" analyses of room air between successive breath samples yielded 0.000 g/210 L. As expectable in the traffic law enforcement setting, the proportion of low-BrACs increases with increasing BrAC; BrACresults of 0.040-0.049 and 0.050-0.059 g/210 L constitute 47.2% of all low BrACs in this study. Descriptive statistics for the alcohol concentration of Breath 1 and Breath 2 appear in Table X. The total pooled 2078 paired BrACmeasurements are, of course, not normally distributed, and the mode greater than the mean reflects the proportion of higher values within the 0-0.059 g/210 L BrAC interval. The ratio of the variances for BrAC 1 and BrAC2 is 0.00022/0.00022 = 1.00, indicating that the null hypothesis applies at the 95% level to differencesbetween BrAC 1 and BrAC2. Figure 4 is a scatterplot for 2078 paired Breath 1 and Breath 2 alcohol measurement results with the 99% prediction interval shown. For these data, least squares linear regression of BrAC 2
upon BrAC 1 yielded .r = 0.963x + 0.0006 g/210 L, with SEE = 0.0037, r = 0.97, r 2 = 94.0%. ANOVA yielded a P-value less than 0.01, indicating a statistically significant relationship between the alcohol concentrations of paired breath specimens at the 99% confidence level. The dispersion of BrAC 1 and BrAC 2 values was
Y= C 998x+ q.0012, ~;/210 L ~"
i
0.09
i
0.06
0.0023 0.003
0.001
i
0
/
i
i
i
0.03
i
0
0.06
i
i
0.09
0.12
0.15
Vapor-alcohol target concentration (g/210 L) Figure 3. Correlation of field instrumentsmeans of vapor-alcohol measurements with VAC targetvalues--pooled field data.
006
Vapor-AlcoholConcentration,8/210 L Parameter Mean SD Median Mode Span -0.016-0.010
/
0
(n = 2105)
-0.003
/
/
0.03
Measurements--Pooled Intoxilyzer 5000-D Field Data
0.0035 -0.002
SEE = ( 0009 r - C ~,~
0.12
Table IX. Paired Sample VAC Differences (Target-Control VAC, g/210 L) for the Vapor-Alcohol Control
Signed -0.001 difference Absolute 0.003 difference
/
0.15
o
r .... l
.
.
.
.
.
.
Y [] 3.963x + 0.00 6
0.05
SEE = 0.0037 ;'= 3.~7 I n= 2078
fin4
I
/ /
.
.
.
.
.
.
8,~.~w~w~.~j
o'g~,:~l = io.o~r o ~
o T.~b~T';,,-
o-J
0.000-0.016 ,", .~
Table X. Descriptive Statistics for Measured Breath-
Alcohol Concentrations of Paired Breath Specimens 1 and 2--Pooled Field Data (n = 2078)
o,.." 98 /
002 "I m Im
'
."
"
..." o
9
~t"
0
0.01
1
o~--o8.,"'~" Q/' o
o,..
/
'/
Breath-AlcoholConcentration,g/210L Parameter n Mean S.D. Median Mode BrAC Span
390
Breath specimen 1 2078 0.036 0.0149 0.038 0.054 0-0.059
Breath specimen 2 2078 0.035 0.0148 0.037 0.045 0-0.059
/.~-9 % Pre ction imit 0
I,,,,-,',-,,,
/
....
0.02
/ ....
0.03
/ ....
0.04
/ ....
0.05
| 0.06
BrAC of breath 1 (g/210 L) ure 4., Corrclatior pai duplicate breath-alcohol measurements-Figure Correlation of paired pooled field data (n = 2078).
Journal of Analytical Toxicology,Vol. 23, October 1999
found to increase slightly with increasing BrAC. The 95% prediction intervals for the means of duplicate breath-alcohol measurements at low BrACs appear in Table XI (excluding data for 74 BrACs of 0--0.009 g/210 L or less, all of which would be reported as negative). The 95% prediction intervals for these mean BrACs were calculated from • 1.96x
b = the slope. The fitted model equation wasy = 0.269/x + 0.3672, with SEE = 0.237, r = 0.99, r 2 = 99.9%. For BrAC measurements in g/210 L, truncated to two decimal places, the summary descriptive statistics and linear regression equation for BrAC2 upon BrAC I approximate the corresponding 3-digit findings. The regression equation for truncated BrACs wasy = 0.964x- 0.0005 g/210 L, with SEE = 0.0046, r = 0.95, and
s
where s is the standard deviation for the respective class of BrACs whose means are given in Table XI. Absolute differences between BrAC I and BrAC 2 for 2004 paired measurements appear in Table XII. The data in this table exclude the 74 BrAC differences between paired BrACs which would be reported as negative. Figure 5 is a histogram of the relative frequency distribution of 2078 BrAC 1 - BrAC 2 absolute differences of 0-0.020 g/210 L. The mean + SD for the 2078 absolute differences were 0.00274 • 0.0026 g/210 L. A histogram of the cumulative frequency distribution of the 2078 absolute differences BrAC 1 -BrAC 2 is shown in Figure 6. Of the BrAC 1 -BrAC 2 absolute differences, 13.3% were 0, about 39% were 0.001 or less, 95% were 0.007 g/210 L or less, and 99% were 0.012 g/210 L or less. Figure 7 illustrates the relationship of imprecision, as coefficient of variation, to mean result of duplicate BrAC measurements. The best fitted model was a reciprocal-x regression of CV upon mean BrAC of the form y = b/x + a where y = the CV in ~ercent, x = the mean BrAC in g/210 L, a = the y-intercept, and Table Xl. 95% Prediction Intervals for Duplicate BreathAlcohol Measurements at Low BrACsmPooled Field Data (n = 2004)
i tilt IIIIt. IIIIIIt,,.. .... o
0.011-0.019 0.021-0.029 0.031-0.039 0.041-0.049 0.051-0.059
8
12
16
(xlo-~
Figure 5. Relativefrequency distribution of absolute BrAC differences BrAC 1 - BrAC 2--pooled field data (n = 2078).
Breath-AlcoholConcentration,g/210 L Mean BrAC 95% predictioninterval 0.015 0.025 0.035 0.045 0.055
4
Absolute BrAC 1 - B r A C 2 difference (g/210 L)
1
v I=
Table XII. Absolute Differences Between Results of Paired Duplicate Breath-Alcohol Measurements with the Intoxilyzer 5000-DmPooled Field Data (n = 2004)
@1
.5
3 E
Breath-AlcoholConcentration, ~/210 t n Mean Pairs BrACSpan BrAC
Mean
SD
288 341 394 484 497 2004
0.002 0.002 0.002 0.002 0.002 0.002
0.0024 0.0023 0.0025 0.0025 0.0025 0.0026
0.010-0.019 0.020-0.029 0.030-0.039 0.040-0.049 0.050-0.059 0--0.059
0.015 0.025 0.034 0.045 0.055 0.036
r=1
Absolute difference'
* Absolute difference = [BrAC 1 - BrAC 2] g/210 U
Median Mode
Span
0.002 0.002 0.002 0.002 0.002 0.002
0--0.014 0-0.020 0-0.017 0-0.016 0-0.016 0-0.020
0.001 0.001 0.001 0.001 0.002 0.001
o
4
8
12
Is
(xlo~)
Absolute BrAC 1 - BrAC 2 difference (g/210L) Figure 6. Cumulative frequency distribution of absolute BrAC differences BrAC 1 - BrAc 2--pooled field data (n = 2078).
391
Journal of Analytical Toxicology, Vol. 23, October 1999
'=
0.26~ Ix + 0.3(;72 U . I O ~
--
= 0.99
"
,
,
,
,
~ 5 ~ , , "-., Predi(:bionLimi
,
,
,
,
,
,
,
,
,
,
,
J
i
i
i
i
0.01 0.02 0.03 0.04 0.05 0.06 Mean breath-alcohol concentraUon (g/210 L) Figure 7. Variability of coefficient of variation as a function of mean breathalcohol concentration--field data.
r 2 = 90.4%. In tests of the truncated means for BrAC 1 - BrAC2, the null hypothesis was supported at the 95% prediction level,as indicated by the ratio of variances of truncated BrAC I and BrAC 2 which was 0.00022/0.00022 = 1.00 for n = 2078. Of all 2078 absolute truncated differences, 78.4% were 0 and 21.4% were 0.01 g/210 L. The most striking contrast was in zero absolute differences BrAC 1--BrAC 2 for 3-digit or 2-digit truncated values. The nontruncated zero differences were 13.3% of all differences. The third digits of BrAC measurements in g/210 L follow an approximately uniform relative distribution between 0 and 9, as shown in Figure 8 in histogram form. In view of the demonstrated similarity of the third digit distributions for BrAC 1 and BrAC 2 shown by paired summary statistics and the essential identity of conditions for the paired measurements, the two peer sets of third digits were merged for n = 4156 to derive the third digit relative distribution, because larger data sets better approximate the uniform distribution. Each of the 10 digits occurs with approximately equal frequency, as in random occurrence; in a perfect uniform distribution, each digit 0 through 9 would appear as 10% of the distribution. A derived index was the predsion of a single low-BrAC measurement The precision of a single BrACmeasurement was calculated from SDI = SDdiffA ~ , where SDdiff is the standard deviation or variability of the mean absolute difference between duplicate breath-alcohol measurements (16). For this system SD1 = 0.0018 g/210 L
Discussion
Z ,I=
ee
0
1
2
3
4
5
6
7
8
9
Third digit of BrAC (g/210 L) Figure 8. Relative frequency distribution of the third digits of BrAC I and BrAC 2, g1210 L--pooled field data (n = 4156).
392
The principal purpose of this study was to obtain information about the performance of field measurements of low BrACs. We also compared field performance in vapor-alcohol measurements (control tests) with the performance of three current-generation quantitative evidential breath-alcohol analyzers in the laboratory under near-ideal conditions. The use of field measurements was necessary to obtain large data bases; it was also appropriate because that is where most breath-alcohol testing occurs. This study design entails advantages and limitations. Because of the small proportion of low-BrACsamong total breath-alcohol measurements in the law enforcement setting, it was necessary to pool field data from 62 instruments at different sites. The 2078 combined field BrAC measurements constitute a low-BrAC subset of the much larger data base for all breath-alcohol test results from these 62 instruments over a period of 9 months. In Oklahoma in recent years, BrACs less than 0.06 g/210 L typically constitute only 6.5% of total BrACfield test results. Data pooling in this study is acceptable because all of the data sets so combined have peer status. Their respective precision was about equal, and all field measurements employed the same alcohol analyzer make and model, identical analysis protocols, and analysts with uniform qualifications. All field control (simulator) tests at any given target VACused the same simulator models and aliquots of centrally prepared, validated simulator solutions with a use limit of the lower of 25 tests or 30 days. In analyzing and interpreting the findings from this pooleddata study for some purposes, it should be recognized that a
Journal of Analytical Toxicology, Vol. 23, October 1999
source of variability, that arising from the separate measurements at different locations and times, is added'to the usual sources of analytical and human biological variability to yield total system variability. Thus: VT= VA+ VB+ Vp where VTis total system variability, VAis the analytical variability, V8 is the breath variability, and Vp is the variability arising from the data pooling. The several component variabilities can be determined in the usual fashion by substituting known data and solving the equation for the desired variable. This study used certain statistical data treatments also used in other published studies of laboratory and field performance of breath-alcohol analysis (17-22), to allow ready comparison of these findings to those of others, even though no other published studies were focused on low BrACs. There is also another statistical consideration relevant to this study. We chose to use BrAC units of g/210 L because the commercial analyzers use that notation for official forensic BrAC measurements. That choice, instead of the more convenient mg/210 L notation, results in the retention of decimal zeros--which entails rounding of some calculated or derived factors and indices. The remainder of this discussion deals sequentially with the invitro laboratory test findings, the in-vitro simulators VAC measurements in the field, and the in-vivo human breath-alcohol measurements in the field.
Laboratory in-vitro measurements of vapor-alcohol concentrations The within-run vapor-alcohol measurements in the laboratory with the BAC DataMaster, the Intoxilyzer 1400, and the Intoxilyzer 5000-D yielded results with accuracy and repeatability typical of properly calibrated current-generation evidential breath-alcohol analyzers. The contrast between the laboratory and field findings is interesting. In the near-perfect laboratory environment within-run multi-replicate measurements of VAC with the Intoxilyzer 5000-D yielded CVs of 1.0% and 0.9% at VAC = 0.06 and 0.10 g/210 L, respectively (Table VI). The corresponding pooled field results were CVs of 2.9% and 3.5% at 0.06 and 0.10 g/210 L, respectively (Table II). The differences reflect the impact of many different instruments contributing data, at different locations and times and always between-runs, and as single measurements. The analyzers evaluated in this study use infrared spectrometry, are microprocessor controlled, very stable, and user4riendly. Some such analyzers are capable of performance in the field matching or exceeding that in a well-controlled laboratory environment. One of the authors (KMD) recently evaluated the performance of three individual BAC DataMaster analyzers in different law enforcement settings. An example of instrument performance was the accuracy and repeatability of within-run (n = 15) VAC measurements, using one KMD laboratory-based reference simulator (Guth Model 34CNP) and NIST-traceable alcohol solutions from our laboratory. At a laboratory-verified target VACof 0.101 • 0.0009 g/210 L, the mean • SD were 0.102 • 0.0003, 0.102 • 0.0008, and 0.101 • 0.0007 g/210 L, respectively (23). The corresponding CVs were 0.3%, 0.8%, and 0.4%. The derived sensitivity data (LOD) and LOQ for the BAC DataMaster, Intoxilyzer 1400, and Intoxilyzer 5000-D (Table VII) were comparable but not identical. However, all are adequate and
acceptable for application to human breath-alcohol testing for forensic, clinical, research and workplace related breath-alcohol testing, especially so when BrAC measurements are stated in g/210 L truncated to two decimal places (e.g., 0.01 g/210 L), as is nationally accepted scientific practice for forensic purposes, initially established by the National Safety Council's Committee on Alcohol and Other Drugs (24).
Field in-vitro measurements of vapor-alcohol concentrations The pooled field data for vapor-alcohol measurement all reflect single between-run VAC results. Each VAC field measurement in this data base represents a control test performed in association with and at the time of a breath-alcohol analysis with a given set of instruments (analyzer + simulator), for BrACs less than 0.06 g/210 L. The data base, therefore, does not contain the much greater number of simulator test results from the 62 analyzers which accompanied the approximately 93.5% of field breath-alcohol analyses with results of 0.06 g/210 L or greater. The total VACmeasurements in this data base (2105) slightly exceed the total breathalcohol measurements (2078) because of additional simulator tests occasioned by periodic maintenance activities and required simulator solution changes. Table II illustrates that the variability or imprecision of the pooled field tests at 0.10 g/210 L expressed as CV is greater than that of a single laboratory device of the same make and model (3.5%/0.9% = 3.9). For VACtarget results at 0.06 g/210 L, the proportion was (2.9%/1.0% = 2.9). Again these field tests entail between run, different instruments (analyzer and simulator), different environment, and different analysis dates as the extant conditions, plus data pooling. In the light of that consideration, these field VACmeasurements are excellent. The derived LOD and LOQ indices for field VACmeasurements, given in Table VIII, compare very favorably with those for the Intoxilyzer 5000-D in the laboratory: LOD 0.0018 (lab) versus 0.0020 g/210 L (field) and LOQ 0.0062 (lab) versus 0.0066 g/210 L (field). They are expected to be alike because the same make and model analyzer was used. The field system is clearly adequate to accurately measure any meaningful alcohol concentrations. We and others have on many occasions statistically evaluated breath-alcohol analysis data by determining the signed and absolute differences between target VACs and VAC results. The descriptive statistics are summarized in Table IX. The 0.003 g/210 L mean absolute difference for 2105 pooled field VAC measurements versus the respective targets at eight separate concentrations with a difference range of 0.016 g/210 L reinforce the accuracy of single simulator tests in the field in the system we studied. The NSC Committee on Alcohol and Other Drugs has recommended that the truncated result of a vapor-alcohol control test (then and now performed mostly with simulators) "must agree with the reference sample value within the limits of • 0.01% W/~' (25)2. This field system meets that recommendation, even for the pooled data when results are truncated to two digits as required by law and agency rules in Oklahoma and most other jurisdictions. Figure 2 gives the within-specifications correlation of target VACs of 0.06 - 0.13 g/210 L with pooled control test results, for pooled single test results. The y-intercept is close to • 0.01% w/v (blood-alcohol equivalent) in the earlier notation in the referenceequals • 0.01 g/210 9 L in current concentration units.
393
Journal of Analytical Toxicology, Vol. 23, October 1999
zero and the slope is close to unity. Figure 3 presents the data as target VACversus mean control test result for each of the eight target VACclasses. Excellent correlation is evident from the graph and the equation for the best-fit linear regression line. 1Yuncated result findings in the field were reported above. The absolute difference span for all 2105 target VAC - measured (Control) VAC was 0 - 0.01 g/210 L for truncated VAC units. A simple way to observe the effect of truncation on the data in the tables in this article, for example, Table II, is to block out momentarily the third decimal digits of VACsor BrACs with a ruler or paper edge. The • 0.01 g/210 L or closer agreement between of target and result VACsbecomes clearly evident. The system is in good control. A recent IACT/NHTSAstudy (26) provides a basis for comparison with some of our findings. Accuracy and repeatability were assessed for 12 Intoxilyzer 5000 instruments by 10 within-run replicate measurements with each instrument at 0.05 g/210 L target VAC, with data pooling. The mean • SD were 0.050 • 0.0018 g/210 L, CV = 3.6%, span 0.046-0.053. The closest comparison is with our pooled results at target VACof 0.06 g/210 L in Table II which had a CV = 2.9%. The IACT/NHTSAstudy report also contains an x-g plot of target VACsof 0.02, 0.05, 0.10, and 0.15 g/210 L (x) with pooled VACmeasurements (n = 140) with 12 Intoxilyzer 5000 analyzers. Their linear regression equation was y = 1.0010x + 0.0064, which corresponds to ours, .q = 0.998x + 0.0012 g/210 L with r = 0.99 for pooled single measurements of VACsof 0.06 - 0.13 g/210 L. Field measurements of breath-alcohol concentrations
All BrAC measurements consisted of duplicate measurements on two consecutive breath specimens. The NSC Committee on Alcohol and Other Drugs recommendation on duplicate breathalcohol testing includes the statement "Consecutive breathalcohol analysis results within 0.02 g/210 L, without regard to sign, shall be deemed to be in acceptable agreement" (27). As in several other published studies of duplicate breath-alcohol measurements in the field (e.g., reference 22), we combined into our data base for this study only BrAC measurements within 0.02 g/210 L of each other, without regard to sign, although Oklahoma rules for forensic breath-alcohol analysis stipulate "Results of duplicate breath alcohol analyses, on the same subject on the same occasion, which are within three-hundredths gram per two hundred and ten liters of breath (+ 0.03 g/210 L) shall be deemed to be in acceptable agreement and mutually confirmatory and substantiative" (28). The rank distribution of BrACs included in this study, given in Table III, confirms the known pattern for law enforcement-associated breath-alcohol analyses--low BrACs are a small portion of total results, and the actual number of lowBrACs increases with increasing BrAC. The 74 pairs of duplicate breath-alcohol measurements which yielded results less than 0.01 g/210 L were excluded from computations for Tables XI and XII, because those results are reported as negative in our system. The descriptive statistics for paired duplicate breath specimens in Table X show the expected close agreement and nearly identical precision of Breath I and 2 measurements, with Breath 2 yielding a result lower than that for Breath I by a mean absolute difference of 0.001 g/210 L. That statistical finding is to be expected given the established fact that most persons tested for alcohol in the 394
law-enforcement context are in the postabsorptive state. With duplicate breath sampling typically4 rain apart, and a population mean breath-alcohol clearance of 0.015 g/230 L per hour (29), the predicted BrACdecrease from Breath 1 to Breath 2 is 4/60 x 0.015 = 0.001 g/210 L. The scatterplot in Figure 4 presents the least squares linear regression of BrAC 2 upon BrAC 1 for all 2078 result pairs with the regression equation and 99% prediction limits shown. As expectable, the dispersion of the results around the best fit linear trend line is slightly greater at higher BrACvalues. The 95% prediction intervals for five BrACclasses shown in TableXI were constructed by separate statistical treatment of each paired BrAC cohort for mean BrAC0.015, 0.025...0.055 g/210 L rather than by use of a single mean SD for the entire breath sample population. All data summarized in Table XII are presented as relative and cumulative frequency distributions in Figures 5 and 6. Courtney et al. in 1992 reported on the BrAC differences between paired breath specimens obtained in their field tests with the Intoxilyzer 5000 analyzer (22). Their large data base includes tests of 7089 subjects with a median BrACof 0.16 g/210 L (span 0--0.40). 5% of all test results were invalidated, under Texas rules, because of lack of 0.020 g/210 L agreement. The mean difference BrAC 1 - BrAC 2 for their 7089 paired tests was 0.0073 g/210 L, with a span of 0.0022-0.0055. Retests of 78% of the subjects whose initial duplicate BrAC differences were greater than 0.020 g/210 L also yielded a mean difference of 0.0073 g/210 L. The histogram of the distribution of their BrAC differences is very similar to our Figure 5, except for a lower proportion of zero g/210 L differences, 5.9% of their differences versus 13.3% for our data. The cumulative frequency distribution for our absolute differences BrAC 1 - BrAC 2 is presented as a histogram in Figure 6. Differencesof 0.01 g/210 L or less constitute about 98% of all differences. The variability or imprecision of breath-alcohol measurements varies inversely with the BrAC. For these field results, that relationship is reflected graphically in Figure 7 as a reciprocal-x regression. The CVs also reflect the increased imprecision at lower BrACswhich necessarily arises from the analyzer output for these devices which is limited to a statement of the concentration result to two or three decimal places. For an assumed imprecision of 0.001 g/210 L at a mean BrACof 0.010 g/210 L, the CV = (0.001/0.010) x 100 = 10.0%, the same 0.001 g/210 L imprecision at a mean BrACof 0.050 yields a CV- 2.0% and for a mean BrAC of 0.100 yields a CV = 1.0%. Figure 8 illustrates the uniform distribution of the third digits of field BrACs in g/210 L. Gullberg (30) also reported an essentially uniform distribution of third digits in field forensic breathalcohol testing. In truncating to two digits from three digits, the omitted third digits followa uniform distribution. Third digits are discrete random variables with an equal probability of being 0, 1...9. An unknown third digit is as likely to be 9 as 0. Truncating BrAC measurements in g/210 L to two decimal places does not introduce bias other than the intended deletion of the third digit. A 1987 report compared our and others' duplicate breathalcohol analysis results for 11 different breath-alcohol analyzers (31). For pooled duplicate BrACpairs in our laboratory (n = 624), measured with a MKIVGC Intoximeter and an Intoxilyzer 4011, we found signed differences to be normally distributed, and found the absolute difference mean + 8D to be 0.0034 • 0.0031 g/210 L
Journal of Analytical Toxicology, Vol. 23, October 1999
with a span of 0-0.018 g/210 L. All BrAC differences were 0 - 0.018 g/210 L. These statistics are like the pooled field data findings in Table IX. Apparently not much has changed over the past decade in analytical performance of breath-alcohol analysis.
Conclusions We concluded:(1) Quantitative evidentialbreath-alcohol analyzers of the current generation have demonstrated accuracy, precision (repeatability and reproducibility), and sensitivity (LOD and LOQ) adequate and appropriate for clinical, research, forensic, and workplace alcohol-testing applications. (2) Breath-alcohol analysis performance in the field with current-generation analyzers is likewise capable of meeting appropriate analytical requirements for both forensic and workplace alcohol testing under current laws and regulations. (3) field results of breath-alcohol measurements on paired breath specimens, in our system, agree closely and when combined with a vapor-alcohol control test afford a firm basis for establishing that the analytical system is in acceptable control, and that the field breath-alcohol measurements are reliable and valid. (4) The field system we studied can and does consistently provide correct measurements of low breath-alcohol concentrations, including 0.02, 0.04, and 0.05 g/210 L.
Acknowledgments We acknowledgeand appreciate the splendid services of the many Oklahoma law enforcement officers performing breathalcohol analyses in connection with traffic law enforcement, whose work provided much of the data base for this report. We also thank CMI,Inc. for making the Intoxilyzer1400 availableto us, and thank National Patent AnalyticalSystems for making the BAC DataMasteravailable to us.
References 1. K.M. Dubowski. The Technology of Breath-Alcohol Analysis. DHHS Publication No. (ADM) 92-1728. U.S. Department of Health and Human Services, National Institute on Alcohol Abuse and Alcoholism, Washington, D.C., 1992. 2. K.M. Dubowski and Y.H. Caplan. Alcohol testing in the workplace. In MedicolegalAspects ofAIcohol, 3rd ed., J.C. Garriott, Ed. Lawyers and Judges Publishing, Tucson, AZ, 1996, pp 439-475. 3. Department of Transportation, Federal Highway Administration: Controlled Substances & Alcohol Use and Testing. Fed. Regist. 59: 7508 (I 994). 4. Department of Transportation, National Highway Traffic Safety Administration: Highway Safety Programs; Model Specifications for Devices to Measure Breath Alcohol. Fed. Regist. 63:10066-10068 (1998). 5. Department of Transportation, National Highway Traffic Administration: Highway Safety Programs; Model Specifications for Calibrating Units for Breath-Alcohol Testers;Conforming Products List of Calibrating Units. Fed. Regist.62:43416-43425 (1997).
6. K.M. Dubowski. Manual for Analysis of Ethanol in Biological Liquids. (Report No. DOT-TSC-NHTSA-76-4). U.S. Department of Transportation, National Highway Traffic Safety Administration, Washington, D.C., January 1977. 7. K.M. Dubowski and N.A. Essary.Evaluation of commercial breathalcohol simulators. Further studies. J. Anal. Toxicol. 15:272-275 (1991). 8. K.M. Dubowski and N.A. Essary.Vapor-alcohol control tests with compressed ethanol-gas mixtures: scientific basis and actual performance. J. Anal. ToxicoL 20" 484-491 (1996). 9. State of Oklahoma, Board of Testsfor Alcohol and Drug Influence. Oklahoma Administrative Code. Title 40. Chapter 30. Analysis of Alcohol in Breath, 1996, sections 40:25-1-2-40:25-1-4, 40:30-11-40:30-1-3. 10. K.M. Dubowski. Quality assurance in breath-alcohol analysis. J. Anal. Toxicol. 18:306-311 (1994). 11. M.G. Natrella. Experimental Statistics. National Bureau of Standards Handbook 91, Washington, D.C., National Bureau of Standards, 1963. 12. J.K.Taylor. Statistical Techniques for Data Analysis. Lewis Publishers, Chelsea, MI, 1990. 13. J.K. Taylor. Quality Assurance of Chemical Measurements. Lewis Publishers, Chelsea, MI, 1987 pp 15-39, 75-93, 205-207. 14. J.C. Miller and J.N. Miller. Statistics for Analytical Chemistry, 2nd ed. John Wiley & Sons, New York, NY, 1988. 15. J.K. Taylor. Quality Assurance of Chemical Measurements. Lewis Publishers, Chelsea, MI, 1987, pp 78-79. 16. M. Thompson and R.J.Howarth. The rapid estimation and control of precision by duplicate determinations. Analyst 98:153-160 (1973). 17. R.G. Gullberg. Statistical evaluation of truncated breath-alcohol test measurements. J. Forensic Sci. 33:507-510 (1988). 18. National Highway Traffic Safety Administration. The Accuracy of Evidential Breath Testersat Low BrACs. (Report No. DOT HS 807 415). Washington, D.C., May 1989. 19. R.G. Gullberg. Determining total method level of detection and level of quantitation for breath alcohol analysis programs. J. Forensic Sci. 37:1208-1210 (1992). 20. R.G. Gullberg and A.J. McEIroy. Identifying components of variability in breath- alcohol analysis. J. Anal. ToxicoL 16:208-209 (1992). 21. R.G. Gullberg. Repeatability of replicate breath alcohol measurements collected in short time intervals. Sci. Justice 35:5-9 (1995). 22. M. Courtney, S. Kleypass, and J. Benton. Differences between first and second breath specimens. SWAFSJ. 14(2): 9-20 (1992). 23. K.M. Dubowski. Unpublished observations, 1998. 24. Committee on Alcohol and Other Drugs. Committee Handbook. Chicago, IL, National Safety Council, 1992, p 31. 25. Committee on Alcohol and Other Drugs. Committee Handbook. Chicago, IL, National Safety Council, 1992, p. 39. 26. S. Ezelle and P. Harding. IACT/NHTSA breath alcohol research project update and findings, Part I. IACTNewsletter. 9(3): 16-22 (1998). 27. Committee on Alcohol and Other Drugs. Committee Handbook. Chicago, IL, National SafetyCouncil, 1992, Recommendation J. 28. State of Oklahoma, Board of Testsfor Alcohol and Drug Influence. Oklahoma Administrative Code. Title 40. Chapter 30. Analysis of Alcohol in Breath, 1996, section 40:30-1-3(d). 29. K.M. Dubowski. Absorption, distribution and elimination of alcohol: h ighway safety aspects.J. Stud. Alcohol. Suppl. 10:98-108 (1985). 30. R.G. Gullberg. Distribution of the third digit in breath alcohol analysis. ]. Forensic Sci. 36:976-978 (1991}. 31. K.M. Dubowski and N.A. Essary.Breath-alcohol analysisof duplicate samples. In Alcohol, Drugs and Traffic Safety - T86, P.C. Noordzij and R. Roszbach, Eds. Excerpta Medica, Amsterdam, The Netherlands, 1987, pp 373-377.
Manuscript received April 8, 1999; revision received May 25, 1999.
395
