The Problem With Blood Tests in DUI/DWAI Cases in Colorado and California
This is a long blog post but incredibly important. Unfortunately, most defense attorneys simply do not have the scientific training to understand the issues surrounding blood tests. Crime labs can certainly detect alcohol (or marijuana/THC) in our systems but it is a whole different ball game when trying to determine an actual BAC or THC concentration. The reality is most crime labs cannot actually say what a person's specific BAC or THC level is - they are only able to give approximations yet they report a specific level of concentration and never report their error rates and other statistical shortcomings associated with their procedures - bottom line is crime labs are reporting BAC and THC levels in an incredibly unethical manner and prosecutors/judges are more than happy to ignore these issues and take away your freedom. This post should provide some basic information if you took a blood test and are subsequently being prosecuted for driving under the influence (DUI) or driving with ability impaired (DWAI) in Orange County/Los Angeles, CA or Denver and the surrounding areas. Under the right circumstances, blood tests can actually be excluded from evidence because of these shortcomings - you need an attorney that understands the science and can work to get the blood tests kicked out of evidence.
1st Issue: Co-Elution
The first major issue is co-elution. Co-elution exists when all of the different elements found in our blood are not separated out during the gas-chromatography process. If co-elution occurs then there is a very good chance you get a higher BAC reading because ethanol is being mixed with other volatiles such as methane.
Crime labs often run a ‘standard mix’ in their machines prior to calibration where they detect the following 5 elements: Methanol, Ethanol, Isoproponal, Acetone and n-proponal. The standard mix is an actual blood sample from a person that has been verified as accurate. These same crime labs will run a sample called a ‘Matrix Blank.’ The matrix blank does not have any volatiles in it and when it is reported they only show n-proponal but on their calibration reports they do scan for ethanol and the number reported is: ‘--‘which indicates they can scan for elements in the blood and if they do not exist a reported number can be generated.
The issue here is that in our blood we are full of multiple elements like methanol and because of the fracking issues in CO we likely have levels of butane and other similar elements in our blood. These are likely to be small amounts but if not accounted for they can actually lead to a higher BAC reading. To prove co-elution does not exist, the reports generated must demonstrate that during the gas chromatography process these elements were scanned for and even trace amounts were separated out. By not reporting the separation of gasses like methane (e.g. methanol in gas form) they cannot prove co-elution did not occur. More often than not, crime labs do not get complete separation of elements.
The first reason co-elution likely occurs more often than not the ‘standard mix’ sample is only meant to find 5 primary elements in our blood. In their article testing new methods designed to prevent co-elution Jalali-Heravi & Parastar (2010) actually separate out 8 elements and point out that co-elution may inevitably occur because the retention times used by crime labs are too fast.
Second, crime labs are often calibrating their machines to only detect ethanol in distilled water. Remember, this calibration occurs after the ‘standard mix’ is run. The gas chromatographers need to be programmed to find all sorts of different elements so if they calibrate only for ethanol there is a good chance other volatiles are not being separated out and co-elution is occurring. These machines cannot think for themselves – we have to tell them what to look for and most crime labs only calibrate for ethanol because of the high volume of samples they must run each day.
2nd Issue: Statistical Analysis & Uncertainty
Crime labs using gas chromatography can only detect alcohol in your system – the machines must use statistical algorythems to actually determine what your BAC is. Most crime labs use linear regression as a way to measure BAC in blood. Because BAC is unknown in the blood what all crime labs are forced to do is create their own data set. One crime lab, according to their SOP, determines ethanol by regressing it against retention time to create a regression line. Thus the dependent variable is ethanol and the primary independent variable is retention time. I also think they use peak height and area scanned as independent variables but their inclusions in the models I will describe below don’t actually change anything.
This regression print out is using a prominent crime lab’s control points of 0.01, 0.08, 0.2 and 0.5 – they run their samples twice to get a total of 8 observations. What is important here is the F-Statistic of 0.81 and the probability > F that is 0.4036. The F-Statistic determines the goodness of fit for the regression model. A regression model is appropriate if the Prob > F score is below 0.05 or 0.01. A score below 0.01 means the model will be a good predictor of a dependent variable 99% of the time. Thus it appears this particular crime lab is not using an appropriate model.
A second indicator that the model is not a good fit is that the constant coefficient (_cons) of 46.50608 is not statistically significant. The probability value of the constant is 0.402 which is not below 0.05 or 0.01. The constant in a model should always be statistically significant. The standard errors are also very high. A standard error effectively means that is the distance a possible observation will be from the plotted regression line. In the above print out the standard error is 46.90149 and we have confidence intervals that literally take us below 0 which is impossible given we cannot have negative amounts of ethanol or any other volatile in our system. The standard error is important here because the retention time coefficient is negative which means that as retention time increases the level of ethanol decreases but the standard error will continually increase. This effectively means the longer a retention time occurs the further away a predicted value from a regression line will be. Hence, as retention time increases the farther away a predicted value from the regression line will be which equals higher likelihood of inaccuracy. Of course, all of this is happening because the model itself is not significant which means inappropriate construction.
This is a histogram representing the information presented above. This graph demonstrates that one of the primary assumptions of using linear regression is violated: data points must have a linear relationship – here we just don’t have any real indication that a linear relationship exists and hence the non-significant F-statistic. What is more the confidence intervals are way far off the regression line which is not correct and even dip into the negatives which is impossible in this context. A properly fitted regression model should have a confidence interval that is tight with the regression line:
Bottom line is that if crime labs want to use a regression analysis they need more data points like the above scatter plot which regresses miles against weight of a vehicle (this is data I use when teaching my introduction stats class). What the properly specified regression model allows us to do is calculate predictions for unknown data points with pretty good accuracy using the following formula: y = mx + b. (y = slope * x variable + y-intercept).
Here, I'm going to use some data from a case I worked on to show you the inaccuracy of the calculation. The crime lab claimed at a retention time of 1.1 my client's BAC was 0.11. Based on the data from the crime lab this is what the formula from the aforementioned crime lab’s data looks like: y = -42.13 * 1.1 + 0.3. Running the calculation we get y = 0.007 or rounded up it is 0.01. Y should equal 0.11 which is what they claimed my client’s BAC was in a case I previously litigated.
It is possible that the crime lab is not regressing concentration of ethanol against retention time but if this is true then their standard operating procedures (which I got through a discovery request) does not reflect this and my calculations are all wrong. However any other type of regression would make no sense because ethanol detection/concentration depends on retention time.
Nevertheless, I ran the models just so you can see how it works out and again they seem inappropriate because of the insignificant F-statistic but we do get a better standard error, a significant constant and a bit tighter confidence intervals that encapsulates all of the known data points which in a situation like this is what we would want. However, just remember the only thing a confidence interval tells us is that a mean is most likely going to fall on the regression line (retention time mean is 1.099 and ethanol mean is .197). FYI – a mean in a regression analysis is your y-intercept.
When I use the y = m*x + b equation I get the following: Y = -0.002 * 0.11 + 1.099 (the 0.11 is the BAC level of my client). My y value then becomes 1.098 or rounded up it is 1.1 which would seemingly make this crime lab’s approach acceptable and accurate (1.1 is the retention time used for my client). However this is with a known x value and thus this equation is a numeric representation of a tautology – or more simply put it is a self fulfilling conclusion which is not allowed in scientific research. We should actually be solving for x rather than y so the equation is 1.1 = -0.002 * x + 1.099 and we get for x a value of -5 which makes absolutely no sense but that is because the logic of this model is incorrect and there are just not enough known data points in existence to make any real calculations.
That said, the National Institute of Science and Technology issued Standard Operating Procedure 29 which is a way for labs to create uncertainty values to be included in their regression analysis and allow for a more accurate model. Moreover, other articles I have read argue that linear regression may not be appropriate in BAC analysis – this issue is most widely seen in the Widmark formula that is used to back calculate BAC levels. Finally, the only way a regression analysis makes sense is for crime labs to create a large data set of known BAC levels to compare their blood samples with unknown BAC levels to as a way to check and confirm their analysis is correct.
Ethically, it becomes important to report error rates and uncertainty measures in this type of analysis because of the problems I’ve identified. However, most crime labs do none of this which means their quality control is highly suspect.