Vitreous alcohol is of limited value in predicting blood alcohol

ELSEVIER SCIENCE IRELAND

Forensic Science International 65 (1994) 73-80

ForeRsic Science

Internatiiill

Vitreous alcohol is of limited value in predicting blood alcohol

Derrick J. Pounder*“, Naohito Kurodab

“Department of Forensic Medicine, University of Dundee, Dundee. Scotland bDepartment of Legal Medicine, Keio University, Tokyo, Japan

(Received 14 April 1993; revision received 11 August 1993; accepted 16 August 1993)

Abstract

Vitreous humour alcohol concentration (VHAC) and blood alcohol concentration (BAC) measured by gas chromatography were available from 345 medico-legal autopsies. Simple linear regression with BAC as outcome variable and VHAC as predictor variable (range l-705 mg%) gave the regression equation BAC = 3.03 + 0.852 VHAC with 95% prediction interval * 0.019 J[7157272 + (VHAC - 189.7)*] and 99% prediction interval f 0.025 d[7157272 + (VHAC - 189.7)*]. The residual standard deviation of VHAC was 26 mg%, the standard error of the slope 0.0098 and the 95% confidence interval for the slope 0.833-0.871. In practice a BAC of 80 mg% is predicted with 95% certainty by a VHAC of ‘150 mg% and similarly a BAC of 150 mg% by a VHAC of 232 mg%. The prediction interval is too wide to be of real practical use, Previous authors have provided various formulae, including a simple conversion factor, to predict BAC from VHAC without taking into account the uncertainty of the prediction for an individual subject. A re-analysis of the raw data from previous publications gave in most instances regression equations significantly different from the authors’ own.

Key words: Alcohol; Vitreous humour; Blood; Autopsy

1. Introduction

The postmortem determination of ethanol is the most frequently performed test in a forensic toxicology laboratory. Blood is the usual specimen analysed when relating ethanol concentration to the degree of intoxication. However, it is

??Corresponding author, Department of Forensic Medicine, The Royal Infirmary, Dundee DDI 9ND, UK.

0379-0738/94/$07.00 0 1994 Elsevier Science Ireland Ltd. All rights reserved. SSDI 0379-0738(93)01451-V

14 D.J. Pounder, N. Kuroda / Forensic Sri. ht. 65 (1994) 73-80

sometimes necessary to resort to other body fluids and tissues if a satisfactory postmortem blood sample is unavailable or contaminated. The vitreous humour has been the subject of investigations of forensic interest since 1959 [l] and comparison of vitreous and blood alcohol concentrations were first undertaken in 1966 by Stumer and Coumbis [2]. Despite repeated studies [3-131, the predictive value of vitreous humour alcohol concentration (VHAC) in estimating blood alcohol concentration (BAC) in an individual case remains contentious. The authors undertook a retrospective analysis of 349 cases and compared their own results with previous publications to clarify this issue.

2. Materials and methods

Three hundred and forty nine cases with known VHAC and BAC were retrieved from medico-legal records from July 1988 to December 1992. Blood had been taken typically, but not universally, from the femoral vein. Vitreous humour had been obtained by direct needle puncture using the standard technique. Blood was collected into tubes containing potassium fluoride and potassium oxalate but this was not so for vitreous humour. Samples were stored at 4’C until analysis within 1 week.

Ethanol was determined in blood and vitreous humour by a gas chromatography (GC) method. A Perkin-Elmer model 8500 gas chromatograph with a flame ionisa- tion detector interfaced with a Perkin-Elmer HS-101 headspace sampler was used for analysis. The GC column was a 6 ft x l/8 in stainless steel column packed with 0.2% Carbowax 1500 on Graphpac GC (Alltech). The operating temperatures were vial heater, 70°C; needle, 100°C; transfer line, 120°C; oven, 75°C; injector, 150°C; and detector, 250°C. The carrier gas was nitrogen flowing at 20 ml/min. Analyses were performed in duplicate and the mean values used for the statistical analysis.

The relationship between vitreous humour alcohol concentration VHAC (x- variate) and blood alcohol concentration b-variate) was determined by simple least squares regression analysis. The prediction interval for the estimated BAC from a measurement of VHAC was calculated as described by Altman [ 141. All the calcula- tions were performed with the aid of MINITAB (Minitab Inc.) statistical software.

3. Results

An initial scattergram and regression line disclosed eight cases falling outside the 95% prediction interval and the autopsy reports and background information on these were reviewed. All eight cases had a BAC greater than the VHAC (Table 1). There were several factors which might account for an anomalous result: sampling of thoracic blood with possible ethanol diffusion from, or contamination by, gastric contents; multiple injuries with possible spillage of gastric contents; and suicide with probable alcohol consumption close to the time of death (death in the absorptive phase). The four trauma cases were eliminated on the grounds that the blood samples were probably contaminated. A regression line was established on the basis of the remaining 345 cases (Fig. 1). A plot of the residuals showed an apparently even scatter but a visual assessment of a normal plot of residuals indicated too many values in the tails of the distribution. The normality of y-variate residuals was for-

D.J. Pounder, N. Kuroah /Forensic Sci. Jnt. 65 (1994) 73-80 15

Table 1 Details of the cases where BAC fell outside the 95% prediction interval

Case no. VHAC (mg%) BAC (mg%) Circumstances

PM 366188 172 PM 272191 160 PM OOU92 355

PM 285192

T89298 380 404 T91619 430 500 T9258 I 54 169 PM 534192 34 128

62

308 530 400

I31

Multiple injuries; hit by train Multiple injuries; hit by train Chest cavity blood; homicidal stabbing;

postmortem burning Chest cavity blood; multiple injuries; fall from

height (bridge) Suicidal stabbing Suicidal salicylate overdose with vodka Suicidal propoxyphene overdose Pethidine overdose with whisky; urine ethanol

141 mg%

mally assessed in the Shapiro-Francis W’ test [14]. Shapiro-Francis W’ was 0.948 (P = 0). Elimination of the four deaths in the absorptive phase gave W’ = 0.986 (P = 0.003).

The regression equation (n = 345) with 95% prediction interval was BAC = 3.03 + 0.852 VHAC f 0.019 J[7157272 + (VHAC - 189.7)2] and with 99% prediction interval was BAC = 3.03 + 0.852 VHAC f 0.025 J[7157272 + (VHAC - 189.7)2]. The residual standard deviation of VHAC was 26 mg%, the standard error of the slope 0.0098 and the 95% confidence interval for the slope 0.833-0.871. The standard error of the intercept was 2.34 and the 95% confidence interval for the intercept - 1.57 to 7.63, being not significantly different from zero.

The practical application of these regression equations is shown in Table 2. Set

,’ ,’

600 ,I’ /

300

200 1

VHAC (mg%)

Fig. I. Scattergram of BAC (mg%) against VHAC (mg%) with regression line. The 95% prediction interval is shown with broken lines running parallel with the regression line.

76 D.J. Pounder, N. Kuroda /Forensic Sri. ht. 65 (1994) 73-80

Table 2 Prediction of critical values of BAC (as mean, or lower limit of 95% and 99% prediction intervals) from VHAC (mg%)

Observed VHAC Predicted BAC (mg%) (mg%)

Mean 95% PI 99% PIa

90 80 29-131 13-147 150 131 80-182 64-198 169 147 96-198 80-214 173 150 loo-201 83-217 232 201 150-251 134-268 251 217 166-268 150-284

aPI, prediction interval for the determination of a single BAC value. Thus for an observed VHAC of 90 mg% the best estimate of BAC is 80 mg%; there is a 95% probability that the true value of BAC is in the range 29-l 31 mg”/, and a 99% probability that the true value is in the range 13-147 mg%.

out are the VHAC values which predict key BAC values of 80 and 150 mg% as the mean, or the minimum value at either the 95% or 99% prediction interval for the determination of a single BAC value. A BAC of 80 mg% is predicted with 95% certainty by a VHAC of 150 mg% and similarly a BAC of 150 mg% by a VHAC of 232 mg%. A BAC of 80 mg% is predicted with 99% certainty by a VHAC of 169 mg% and similarly a BAC of 150 mg% by a VHAC of 251 mg%.

For previous English language publications [4-6,8-131 the numbers of cases studied, the mean, standard deviation and range of the ratio BAC:VHAC and the recommended conversion (regression) equation are set out in Table 3. Seven pre-

Table 3 Analyses and formulae for predicting BAC from measurements of VHAC provided in previous publications

Reference N BAC:VHAC Recommended conversion

Mean S.D. Range equation (mg%)

4a 25

5 174 6b 110

8C 43

9 44

10 43 II 86 12b 205 13 75

0.72 0.74 0.89 0.95

0.95

0.77

0.80

0.85 0.81

0.09 0.11

-

0.16

0.17

0.04

0.54-0.88 0.55-0.95

0.58-2.08

0.72- 1.37

0.57-1.75 0.63-2.42 0.52-4.0 0.55-1.38

BAC = 0.73 VHAC

BAC = 0.89 VHAC BAC = 0.96 VHAC +

29 zt 102 (BAC = 0.76 VHAC +

35.7) BAC = 0.81 VHAC BAC = (0.77 f 0.34)

VHAC BAC = 0.57 VHAC BAC = 0.76 VHAC + 4.7 - Unreliable

N, number of cases; SD., standard deviation. ‘Used two analytical techniques for ethanol. bGives VHAC:BAC in original publication. Qerives various regression equations but recommends simple conversion factor.

D. J. Pounder, N. Kuroda /Forensic Sci. ht. 65 (1994) 73-80 II

Table 4 Regression equations with 95% prediction intervals derived from the raw data of previous publications and Shapiro-Francis W’ test for normality of dataa

Ref Regression equation ??95% Prediction interval W’ P

2 BAC = -7.0 + 1.04 VHAC + 0.07 J[732894 + (VHAC - 170.5)2] 0.916 0.0083 3 BAC = 19.9 + 0.84 VHAC f 0.11 +0148 + (VHAC - 236.5)*] 0.984 0.9434 4 BAC = 13.8 + 0.68 VHAC f 0.13 J[249594 + (VHAC - 200.9)‘] 0.837 0.0013 8 BAC = 37.9 + 0.76 VHAC f 0.08 J[528190 + (VHAC - 225.2)*] 0.972 0.3049

10 BAC = 13.9 + 0.70 VHAC f 0.09 J[599696 + (VHAC - 227.3)*] 0.900 0.0024 11 BAC = 13.6 + 0.78 VHAC f 0.05 J[1369425 + (VHAC - 182.0)*] 0.948 0.0029 13 BAC = 7.7 + 0.77 VHAC f 0.07 J[862282 + (VHAC - 228.9)*] 0.991 0.0002

aP value is the probability that the value of W’ would be obtained in a sample if the population had a normal distribution of y-variate residuals.

vious publications [2-4,8,10,11,13] provide the raw data of VHAC and corresponding BAC. Simple linear regression was performed using MINITAB (Minitab Inc.) statistical software and the regression equation and 95% prediction interval together with the Shapiro-Francis W’ test for normality of y-variate residuals was derived for each of these data sets (Table 4).

4. Discussion

There are many advantages to the postmortem toxicological analysis of vitreous humour. In most cases, the specimen is easily obtained and can be sampled without a full autopsy. Vitreous humour is a clear, serous specimen which is easy to work with analytically. Its anatomically isolated position protects it from putrefaction and it demonstrates good chemical stability. Analysis of vitreous humour may be used to corroborate a postmortem blood alcohol level and assist in distinguishing intoxication from postmortem alcohol production.

The first study comparing vitreous and blood alcohol concentrations was undertaken by Stumer and Coumbis [2], who concluded that ‘there is a close correlation between levels of ethyl alcohol in the blood and the vitreous humour; a wider range of variation between the two exists at the higher values’. Another study [3] lead its authors to conclude similarly that ‘the concentration of ethanol in the vitreous humour accurately reflects that in the blood, particularly when blood values are less than 200 mg%. When blood alcohol concentrations are high, there is a slight upward deviation in the observed vitreous humour values’.

In another study [4] the question of predicting BAC from VHAC first arose. Its authors concluded that in cadavers the blood alcohol concentration can, within safe limits, be estimated from the vitreous alcohol concentration using the equation: blood alcohol concentration = 0.73 x vitreous alcohol concentration. This simple idea was reinforced by Coe and Sherman [5] who published a larger series of 174 cases and performed regression analysis. They concluded that the best fitting conversion factor for converting vitreous humour to blood alcohol concentration with 95% confidence was 0.89 f 0.023; i.e. the best estimate of blood alcohol concentra-

78 D. J. Pounder, N. Kuroda /Forensic Sci. Ini. 65 (1994) 73-80

tion = 0.89 (kO.023) x vitreous humour alcohol concentration. They promoted the idea of a simple ‘conversion factor’.

Since it is the writers’ contention that the key to this problem is correct use of regression analysis, we emphasise here a few points concerning the method [14,15]. We have plotted (Fig. 1) the outcome variable (BAC) against the predictor variable (VHAC) and then fitted a straight line to the data that in some sense gives the ‘best’ prediction of BAC for any value of VHAC. Importantly, the method assumes, amongst other things, that the values of the outcome variable, BAC, have a normal distribution for each value of the predictor variable, VI-UC. The regression line can be thought of as joining the mean values of BAC for each value of VHAC. The Shapiro-Francis W’ test applied to previous datasets (Table 4) indicated some non- normality in five of the seven previous studies. W’ for the authors’ own data also indicated non-normality but this was small; the P value reflected the large sample size and was heavily influenced by a few data points. In the authors’ judgement the degree of non-normality is unimportant.

The 95% confidence interval for the regression line includes the true relation with 95% probability. The slope is the parameter of main interest, as it indicates the strength of relationship between the two variables, and we have calculated the standard error of the slope, as several previous authors have done. However, a critical fact is that the 95% prediction interval for the determination of a single BAC value is much wider than the 95% confidence interval for the mean of BAC for a given value of VHAC. A greater uncertainty when trying to predict BAC for an individual must naturally be expected. For any value of VHAC it is expected that 95% of future subjects will have BAC values between the prediction interval. Consequently there is a 95% probability of an individual’s BAC value being within this interval although the best estimate is given by the value on the regression line corresponding to the value of VHAC. Obviously a narrow prediction interval is needed for a relationship to have any practical usefulness. From the data which we have provided (Table 2) it is clear that the prediction interval of BAC from VHAC is so wide as to make it practically useless from a quantitative medicolegal perspective.

From an evidential viewpoint it would be unreasonable to give an estimate of the mean BAC based upon the VHAC without also providing the 95% prediction interval. Yet some authors have done this and produced an average ‘conversion factor’. Providing a confidence interval for the slope does not redress the omission; in giving an estimate of BAC from VHAC it is essential to give the 95% prediction interval.

The literature does contain expressions of unease over the use of a simple ‘conversion factor’. Noting the discrepancies between the calculated ‘conversion factors’ obtained by previous workers, Coe and Sherman [5] remarked that ‘while such discrepancies would make it seem undesirable to utilise the vitreous alcohol concentration as absolutely predictive of blood alcohol concentration, it has been found ad- vantageous to do vitreous humour alcohol determinations routinely in conjunction with blood alcohols as a check on the blood values’. The authors agree with this broad conclusion. In a study of 3 1 cadavers [7], blood ethanol levels were calculated from the corresponding levels in vitreous humour using a conversion factor of 0.93. The comparison between the observed and the calculated blood ethanol levels gave an average difference of 12.2% (S.D. = 10.8; range = O-37.9%). Jollymore and col-

D.J. Pounder, N. Kuroda/Forensic Sci. Int. 65 (1994) 73-80 19

leagues [lo] recognised that using a simple conversion factor of 0.81 leads to an estimated blood alcohol level which is greater than the observed blood alcohol level in 66% of cases. Their solution to increase the reliability of a blood alcohol level determined from a vitreous humour level was to use the lowest blood to vitreous humour ratio observed in practice, namely a value for their study of 0.57. A better solution would be to apply regression analysis and the 95% or 99% prediction intervals.

It was first pointed out in 1969 [4] that in cases where the ratio of blood to vitreous humour alcohol concentrations exceed 1 .O, the most likely explanation is that death occurred before diffusion equilibrium had been attained and this observation may be of forensic significance. Later, in a study of 43 cases, Caughlin [8] suggested a bi-modal distribution of blood:vitreous alcohol ratios with the first mode in the range 0.72-0.90 and a positively skewed distribution in the range 0.94- 1.37. He pro- posed that the first mode of the distribution ratio represented the elimination phase of the blood alcohol curve and that the second mode represented the absorption phase prior to equilibrium being established. Yip and Shum 1111, in an analysis of 86 cases reproduced this b&modal distribution of blood:vitreous alcohol concentration ratios and suggested that a blood:vitreous ratio >0.95 indicates that death occurred before equilibrium had been achieved, i.e. in the early absorptive phase. The blood:vitreous ratio during this early phase had a mean of 1.09 (S.D. = 0.38) in contrast to the late absorptive and elimination phases where the mean was 0.80 (SD. = 0.09). Others [IO] failed to reproduce this bi-modal distribution.

Most deaths occur during the elimination phase and it is clear that the observation of a bi-modal distribution of blood:vitreous alcohol ratios in any study depends upon the inclusion of cases dying in the absorptive phase. Animal studies [16,17] indicate that, following intra-peritoneal or intravenous injections of ethanol, BAC:VHAC ratios may be >0.95 for 30 min or longer. We suggest that it is likely that the proportion of absorptive phase cases included in published series has varied considerably and that this accounts, at least in part, for the differences in published BAC:VHAC ratios (Table 3).

We conclude that although regression analysis allows the development of a predictive model for BAC given a specified value of VHAC, the prediction interval is so wide as to make it of little practical use from a quantitative perspective. Re-analysis of the data provided in previous studies confirms this view. The prediction interval cannot be improved appreciably by developing larger case series. However, vitreous humour alcohol measurement remains a valuable means of corroborating an already known blood alcohol value.

5. References

I H.N. Naumann, Postmortem chemistry of the vitreous body in man. Am. Med. Assoc. Arch. Ophthalmol., 62 (1959) 356-363.

2 W.Q. Sturner and R.J. Coumbis, The quantitation of ethyl alcohol in vitreous humor and blood by gas chromatography. Am. J. Clin. Pathol., 46 (1966) 349-351.

3 M.S. Leahy, E.R. Farber and T.R. Meadows, Quantitation of ethyl alcohol in the postmortem vitreous humor. J. Forensic Sci., 13 (1968) 498-502.

4 S. Felby and J. Olsen, Comparative studies of postmortem ethyl alcohol in vitreous humor, blood, and muscle. J. Forensic Sci., I4 (1969) 93-101.

80 D.J. Pounder. N. Kuroda / Forensic Sci. Int. 65 (1994) 73-80

5

6

8

9

10

11

12

13

14

15 16

17

J.I. Coe and R.E. Sherman, Comparative study of postmortem vitreous humor and blood alcohol. J. Forensic Sci., 15 (1970) 185-190. R.C. Backer, R.V. Pisano and I.M. Sopher, The comparison of alcohol concentrations in postmortem fluids and tissues. J. Forensic Sci., 25 (1980) 327-331. C.L. Winek and F.M. Esposito, Comparative study of ethanol levels in blood versus bone marrow, vitreous humor, bile and urine. Forensic Sci. Znt., 17 (1981) 27-36. J.D. Caughlin, Correlation of postmortem blood and vitreous humor alcohol concentration. Can. Sot. Forensic Sci. J., 16 (1983) 61-68. B.E. Stone and P.A. Rooney, A study using body fluids to determine blood alcohol. J. Anal. Tox- icol., 8 (1984) 95-96. B.D. Jollymore, A.D. Fraser, M.A. Moss and R.A. Perry, Comparative study of ethyl alcohol in blood and vitreous humor. Can. Sot. Forensic Sci. J., 17 (1984) 50-54. D.C. Yip and B.S. Shum, A study on the correlation of blood and vitreous humour alcohol levels in the late absorption and elimination phases. Med. Sci. Law, 30 (1990) 29-33. Y.H. Caplan and B. Levine, Vitreous humor in the evaluation of postmortem blood ethanol concentrations. J. Anal. Toxicol., 14 (1990) 305-307. P. Neil, A.J. Mills and V.M. Prabhakaran, Evaluation of vitreous humor and urine alcohol levels as indices of blood alcohol levels in 75 autopsy cases. Can. Sot. Forensic Sci. J., 18 (1985) 97-104. D.G. Altman, Practical Statistics for Medical Research, Chapman and Hall, London 1991, pp. 139, 299-309, 315. M.J. Gardner and D.G. Altman, Statistics with confidence. Br. Med. J., London 1989, Chap. 5. J.E. Olsen, Penetration rate of alcohol into the vitreous humor studied with a new in vivo technique. Acta Ophthalmol. Copenh., 49 (1971) 585-588. P. Fernandez, M. Lopez-Rivadulla, J.M. Linares, F. Tato and A.M. Bermejo, A comparative phar- macokinetic study of ethanol in the blood, vitreous humour and aqueous humour of rabbits. Foren- sic Sci. Znt., 41 (1989) 61-65.

Documents

Vitreous alcohol is of limited value in predicting blood alcohol