Reply to 'Incorrect reanalysis of breath/blood alcohol data'.

The recent letter to the editor, "Incorrect Reanalysis of Breath/Blood Alcohol Data," by Wigmore and Robinson (1), is a restatement of the obvious. Of course it is important to know the shape of the distribution of in vivo blood/breath ratios in order to determine the uncertainty in BACs that occurs when breath analysis is used. It was not I, but Dubowski who stated the blood/breath ratio follows a normal distribution. On p. 102 of Reference 2 he said, "[T]hese experimentally determined [blood/breath] ratios have a Gaussian distribution. Hence a postabsorptive blood alcohol:breath alcohol concentration range of 1797:1 to 2763:1 can be estimated for 95% and 1555:1 to 3005:1 of 99.7% of such a population. These statistical projections agree closely with the experimentally found range of values." (emphasis added) In a more recent article, Dubowski applied Jones' mean and SD to the data in a 1978 article (3) to predict the range of blood/breath ratios expected in 99.7% of the population (see Table III of Reference 4).1 Using the same approach, I estimated the uncertainty in BAC results determined from breath analysis and the percentage of over and underestimates of actual BAC based on Dubowski's data and a normal distribution. It would have certainly been desirable to use Dubowski's actual data in order to independently check the shape of the distribution he described, the value of the SD, and the actual number of over and underestimates. Because these data appear not to have been published, this was not possible. Therefore, only Dubowski can tell us how close the estimates I made are to his actual results. I should also point out that Jones (3) used a detailed normal error analysis of his experimental blood/breath ratios to report the results from his 1978 study. He too did not publish his data in sufficient detail to allow comparison of the estimates I made (7) with actual values, but he probably can inform us how closely they agree with the actual values. He also appears to have recently adopted a normal error curve to recommend that 0.015 g/210 L be deducted from each breath test result to allow for 99.7% confidence limits (see p. 20 of Reference 8). Wigmore and Robinson also quoted an ostensibly authoritative source describing normal distributions. It is clear, however, that this quote confuses desk size in a medical school, which does not follow a normal distribution, with the random …