The Flawed Nature of the Calibration Factor in Breath-Alcohol Analysis

In the ClassroomThe Flawed Nature of the Calibration Factor in Breath-AlcoholAnalysisDominick A. LabiancaDepartment of Chemistry, Brooklyn College of The City University of New York, Brooklyn, NY 11210In recent years, a number of articles have appeared inthis Journal addressing various aspects of breath-alcohol analysis,including the role of Henry’s law in this type of analysisand the variability of the blood-alcohol to breath-alcohol partitionratio (blood:breath ratio, or BBR) central to the conversionof any measured breath-alcohol concentration(BrAC Meas. ) into a corresponding estimated blood-alcoholconcentration (BAC Est. ) (1–4). More specifically, within thecontext of breath-alcohol analysis, Henry’s law describes theequilibrium distribution of ethanol vapor between alveolarair and circulating pulmonary blood at 34 °C, the averagetemperature of such air. When breath analysis is used to determinea subject’s BAC Est. from his or her BrAC Meas. , eq 1,derived from Henry’s law, is used (2). In this equation, the2100:1 BBR, reflects the assumption that for all subjects undergoinga breath-alcohol analysis and having an average alveolarair temperature of 34 °C, 2100 mL of this expired aircontains the same mass of ethanol as 1 mL of blood.no. g ethanol 2100 mL breath no. g ethanol= ×100 mL blood 1 mL blood 210 L breath(1)BAC Est. BBR BrAC Meas.The Role of Simulator SolutionsA breath-alcohol simulator is an ideal Henry’s law systemconsisting of a dilute solution of ethanol in water maintainedat 34 °C. Simulator solutions are intended to simulatehuman test subjects and are routinely used to check theaccuracy of breath-alcohol analyzers that rely on a specificBBR, defined in the United States of America as 2100:1.These analyzers are utilized by law enforcement agencies todetermine the alcohol concentrations in samples of alveolarair provided by suspected DWI (driving-while-intoxicated)arrestees. Measured BrACs are reported in concentration unitsof g ethanol/210 L breath in jurisdictions where legal limitsare defined in terms of specific BrACs, for example, 0.08 gethanol/210 L breath in California. In other jurisdictions,measured BrACs are converted into equivalent estimatedBACs via eq 1. So in New York, for example, where the legallimit is 0.10 g ethanol/100 mL blood, or 0.10%, a BrAC Meas.of 0.10 g ethanol/210 L breath is equivalent to a BAC Est.of 0.10%.For a simulator solution maintained at 34 °C, the partitionratio corresponding to the BBR would be the water-alcoholto air-alcohol partition ratio (k w/a ). This ratio has beenestimated by Dubowski to be 2573:1, derived from his leastsquares,best-fit regression analysis of data documented in anumber of relevant studies (5). An equivalent form of theequation stemming from Dubowski’s analysis is eq 2, whichyields the value of 2573:1 cited above for k w/a when x, thetemperature in °C, is 34 °C.1 × 10 3k w/a=(2)0.04145e 0.06583 xLater investigations by Dubowski and Essary (6, 7) andSpeck et al. (8) reveal ranges of k w/a whose upper and lowerlimits do not differ significantly from Dubowski’s empiricallyderived ratio of 2573:1.Simulator-Based CalibrationsGiven that breath-alcohol analyzers rely on a fixed BBR,all test subjects, including simulator solutions, are assumedto be characterized by the same partition ratio. So if a breathalcoholanalysis is conducted on a DWI arrestee in New York,for example, the breath-alcohol analyzer is calibrated with asimulator solution that is expected to generate a result of0.10% if the analyzer is functioning properly at a fixed partitionratio of 2100:1. This result can be calculated prior toa given calibration using eq 3: R denotes the partition ratioof the test subject, which in the case of a simulator solutionis estimated to be 2573:1 at 34 °C; AC Actual is the true alcoholconcentration of the test subject, which for a simulatorsolution intended to produce a 0.10% result on a properlyfunctioning breath-alcohol analyzer is 0.1226% (5); andAC Reported is the alcohol concentration reported by a breathalcoholanalyzer relying on a 2100:1 partition ratio. 1 Clearly,when 2573 and 0.1226% are substituted, respectively, for Rand AC Actual in eq 3, the truncated value of AC Reported is0.10%.AC Reported=2100 × ACActualR(3)The NHTSA Conforming Products ListBreath-alcohol analyzers that appear on the ConformingProducts List (CPL) of Evidential Breath Measurement Devicesof the U.S. Department of Transportation’s National HighwayTraffic Safety Administration (NHTSA) are those devicesthat have been evaluated for precision and accuracy atcertain values of AC Reported . Specifically, instruments that meetthe model specifications established in 1993 (10) andamended in 1994 (11) are those instruments that have beenevaluated for precision and accuracy at 0.000 (blank reading),0.020, 0.040, 0.080, and 0.160% alcohol concentrations, andthat generate results within an acceptable systematic errorrange of ±0.002% (±2% at a BrAC of 0.10 g/210 L or a BACof 0.10%). For the blank reading, the regulations require that,“The tester shall use his or her own breath ... and he or shemay not consume alcohol for a period of 48 hours prior tothis test or smoke for a period of 20 minutes prior to thisJChemEd.chem.wisc.edu • Vol. 79 No. 10 October 2002 • Journal of Chemical Education 1237

In the Classroomtest” (11). For the remaining test concentrations, or AC Reportedvalues of 0.020, 0.040, 0.080, and 0.160%, simulator solutionshaving concentrations corresponding to AC Actual valuesof 0.0245, 0.0490, 0.0980, and 0.1960%, respectively, wouldbe used. These concentrations can be determined by usingeq 3 to solve for AC Actual in each case, in accord with eq 4,with 2573 substituted for R.AC = RActualAC2100 × Reported(4)The Flawed Nature of the Calibration FactorThe fundamental flaw of simulator-based calibrations isthat, while they produce values of AC Reported within an establishedmargin of error when a breath-alcohol analyzer functionsproperly at 2100:1 (or at another fixed BBR, such as2000:1 or 2300:1(12), both of which have been adopted outsidethe United States), they do not guarantee accurate resultswhen humans undergo breath-alcohol analysis. This flawoccurs because a properly maintained simulator solution isan ideal Henry’s law system, as noted above, while a humansubject is not. Therefore, calibrations of breath-alcohol analyzerswith such solutions deal only with instrument error,one of the three types of systematic error (13). Where humantest subjects are concerned, however, those calibrationsdo not address method error, the second type of systematicerror (13), which stems from the nonideal behavior of humansubjects and can be significant. 2 In this regard, Jones(14) reported that 70% of the uncertainty in breath test resultsis attributable to physiological variables, and Simpson(15) reported that 90% of the uncertainty in breath test resultsis “ascribable to variables involving the subject.” Theblanket claim, therefore, that the result of a breath-alcoholanalysis reflects the subject’s actual BAC within the same narrowmargin of error characterizing simulator-based calibrations,is misleading and untenable.This argument applies as well to calibrations of breathalcoholanalyzers involving dry gas ethanol standards that arealso used to simulate breath samples provided by human testsubjects. These standards are available as ethanol-in-nitrogencompressed gas mixtures having certified ethanol concentrationsthat can be expressed in units of g ethanol/210 L gas at34 °C and 760 torr (16). Dubowski and Essary (16) haveconfirmed that calibrations derived from such standards generateresults consistent with those based on aqueous simulatorsolutions. The effluent from any of the latter, in contrastto the dry gas standard, is essentially saturated with watervapor and is termed wet gas. Breath-alcohol analyzers capableof utilizing both wet gas and dry gas calibrations are equippedwith a barometric sensor to facilitate corrections for barometricpressure variability when the latter type of calibrationis used. 3The physiological variables, referred to above, that affectthe accuracy of a breath-alcohol analysis involving a humansubject are addressed below within the context of breathtest results reported in terms of BAC Est. . It must be emphasized,however, that these variables apply as well in those caseswhere breath test results are reported in terms of BrAC Meas. .This is a consequence of the fact that direct BrAC statutes inthe United States and elsewhere are based on a 2100:1 BBR(or on the other values of the BBR cited above), a point thatJones has stated explicitly (12) and that Labianca andSimpson (17) have demonstrated unequivocally via a simplemathematical proof.Variability of the BBRExtensive data demonstrating BBR variability have beendocumented in the scientific literature over the years, so onlykey conclusions stemming from these data and concerningthe postabsorptive and absorptive states of alcohol metabolismare emphasized here.For the postabsorptive state, Dubowski (18) reportednormally distributed BBR data characterized by a mean of2280:1 and a statistical range of 1555:1 to 3005:1 for 99.7%of the drinking population. Based on the standard 2100:1ratio, it can be shown that Dubowski’s data are consistentwith a relative error range of 26% to +43%. Moreover, thosedata reflect a ratio of BAC underestimates to overestimatesof 77:23 (15, 19). In contrast, Jones’s postabsorptive data (20)indicate a lower, although still significant, ratio of underestimatesto overestimates of 69:31 (21).A noteworthy point in this regard is that the typical wetgas simulator calibration of a breath-alcohol analyzer is anexample of an analysis that produces an underestimated resultand, furthermore, definitive proof of the inability of suchanalyzers to adjust for test subjects having partition ratios thatdeviate from 2100:1. As noted above, a simulator solutionwhose AC Actual is 0.1226% is expected to generate a truncatedAC Reported of 0.10% on a properly functioning breath-alcoholanalyzer operating at a BBR of 2100:1. This is obviously afalse-low result for the simulator solution in question, but,nevertheless, the anticipated, correct false-low result that wouldbe generated by a breath-alcohol analyzer relying on a partitionratio lower than that of the 2573:1 defining the simulatorsolution. Moreover, and even more significant because ofthe adverse societal impact that would ensue, is the hypotheticalscenario in which the simulator solution would be aDWI arrestee in a jurisdiction having a legal limit of 0.12%.Although there are no such jurisdictions in the United States,the point is that, such an arrestee, although clearly guilty ofDWI, would not be so charged because his or her AC Reportedwould fall below the legal limit.For the absorptive state, the effect of BBR variability issubstantially more pronounced (22). Mason and Dubowski(23) have emphasized that, “... when blood and breath testsare available to a subject, the breath test can be discriminatoryin yielding a higher result than a blood test duringabsorption.” More recently, Dubowski (18) added,“significant variations from [the postabsorptive mean BBRof 2280:1] exist during active alcohol absorption and [asindicated above] in some individuals even in thepostabsorptive state.” In fact, Labianca and Simpson (24)determined a mean absorptive state BBR of 1836:1, derivedfrom lognormal statistical analysis of the data of Giguiere andSimpson (25). The logarithm-transformed data are normallydistributed and involve a statistical range of 1128:1 to 2989:1for 99% of the drinking population in the absorptive state.This reflects a relative error range of 46% to +42% and aratio of underestimates to overestimates of 24:76. The dataof Jones (20) for the absorptive state indicate a less1238 Journal of Chemical Education • Vol. 79 No. 10 October 2002 • JChemEd.chem.wisc.edu

In the Classroompronounced, but certainly significant, ratio of underestimatesto overestimates of 35:65 (26).The Effect of TemperatureConsistent with Henry’s law is the fact that temperaturemust be controlled in any given application, and this is certainlythe case for simulator solutions maintained at 34 °C,as indicated above. Yet oral temperature measurements ofDWI arrestees are not part of the protocol used by law enforcementagencies, despite the fact that such measurementsand the use of appropriate corrections where necessary havebeen endorsed (27–29). In this regard, studies have shownthat the BBR changes with temperature by factors rangingfrom 6.5%/°C (30) to 8.6%/°C (29).Breathing PatternThe length of time involved in breath sample delivery isalso a critical variable in breath-alcohol analysis. Hlastala (31)has found that errors in BAC Est. of as much as ±50%, or more,can occur by altering the breathing pattern. He attributes thisvariation to changes in alcohol concentration during thebreathing process, stemming from cooling and heating of thebreath and airways. These dynamics of airway alcohol exchangeeffect a positively sloped alveolar plateau (32). Thusthe longer a test subject exhales into a breath-alcohol analyzerafter breathing the cooler air of the surroundings, thegreater the positive deviation from the true BAC. Hlastala(31) offers the example of a BAC Est. of 0.14% stemming froma lengthy breath sample delivery that would be consistent witha significantly lower BAC of 0.09%, where the former concentrationis nearly 50% more than the latter. Ohlsson et al.(33) also reported an increase of over 50% in the BrAC Meas.of a subject with a large vital capacity (6.3 L) who exhaledinto the infrared-based breath-alcohol analyzer used in theanalysis well beyond the required minimum breath deliverytime of 4 s at a pressure exceeding 11 torr.The effect of breathing technique on the results ofbreath-alcohol analyses has also been explored by Jones (30).He found that, with breath-holding for 30 s prior to exhalation,BrAC Meas. and breath temperature increased by as muchas 18% and 0.7 °C, respectively. On the other hand, hyperventilationfor 20 s prior to exhalation produced decreasesin BrAC Meas. and breath temperature of up to 12% and 1.2°C, respectively.Other VariablesThe health of a subject undergoing breath-alcohol analysisis another important consideration that cannot be ascertainedby a breath-alcohol analyzer deemed to be accuratebased on a simulator calibration. A specific example in thisregard is a case described by Labianca and Simpson (17),which involved a severely asthmatic defendant charged withDWI in California. The charge stemmed from a BAC Est. of0.09% at a time when the statutory legal limit in Californiabased on breath tests was 0.08% (as indicated above, Californiacurrently has a direct breath statute in place that specifiesan equivalent legal limit of 0.08 g ethanol/210 L breath).The defendant presented evidence at his trial—derived fromcontrolled experiments conducted after he was charged andinvolving the analysis of samples of his blood- and breathalcoholthat were taken essentially simultaneously—thatshowed a postabsorptive BBR of 1233:1. With 1233 substitutedfor R, eq 4 revealed that the defendant’s AC Actual was0.05%, and this led to the ultimate dismissal of the DWIcharge. The above case is certainly not unique. As emphasizedelsewhere (17), Russell and Jones (34), in their studyof subjects with chronic obstructive pulmonary disease—which includes conditions such as asthma and emphysema(35)—concluded that, “quantitative measurement [involvingbreath-alcohol analysis] must be approached with caution”when test subjects lack effective pulmonary function.Additional variables that contribute to the uncertaintyin results obtained from a properly calibrated breath-alcoholanalyzer include mouth-alcohol contamination and contaminationof breath samples with compounds that can be mistakenlyidentified as alcohol. In the first instance, studies onan infrared-based breath-alcohol analyzer, which apparentlyhas the capability of identifying mouth-alcohol from sourcessuch as bleeding gums or regurgitation of stomach contents,produced false-high results and demonstrated that the identifyingmechanism is not foolproof (36). In the second case,the same type of infrared-based breath-alcohol analyzer 4 alsogenerated false-high results. This occurred because of the inabilityof the instrument to distinguish between the methylgroup of ethanol and the methyl group(s) of the solvents tolueneand the o-, m-, and p-xylenes to which test subjects hadbeen exposed (37).ConclusionObviously, method error in breath-alcohol analysis is asignificant issue, and if the impact of a relevant variable cannotbe ruled out in a particular DWI case, then that variablemust be taken into account. Clearly, the claim by users ofbreath-alcohol analyzers that such instruments are accuratebecause they produce accurate results within a specified marginof error when calibrated with simulator solutions (or drygas standards) is a very limited claim. The only acceptablepoint of accuracy in this situation is that the breath-alcoholanalyzer that functions properly at a fixed BBR is capable ofaccurately analyzing such a solution (or dry gas standard).However, when the same breath-alcohol analyzer is used totest a human subject, the result cannot automatically bedeemed accurate: it must be evaluated within the context ofits uncertainty. We teach our students in various chemistrycourses to abide by this protocol when evaluating experimentaldata, and we can certainly expect agencies that rely onwet or dry gas calibrations of breath-alcohol analyzers to abideby it as well.Moreover, it is important to place the message of thisarticle into a broader, general context. That is, a calibrationstandard should, ideally, mimic the composition of the sampleto be analyzed (38). Thus, for example, if a particular sampleis to be analyzed for a metal ion via absorption analysis of acolored complex ion derived from the metal ion, the presenceof sulfate and phosphate ions in the sample matrix caninterfere with the analysis. Sulfate and phosphate ions havea tendency to form colorless complexes with metal ions; as aconsequence a significant reduction in the concentration ofthe desired colored complex ion and its associated absorbancecan occur for the sample in question. To offset this type ofmatrix effect, calibration standards can be used that containJChemEd.chem.wisc.edu • Vol. 79 No. 10 October 2002 • Journal of Chemical Education 1239