I want to start this essay with what seems like a simple question: What are the chances that a potential employee who tests positive for an illegal substance during a pre-employment drug-screening actually uses that illegal substance?
Let me start with some background information:
The “accuracy” of drug testing
The first thing to note is that the term “drug-test” is a misnomer. Drug tests don’t actually test for the presence of drugs, but test for the presence of metabolites that the human body produces to process a drug when it enters the body. However, the human body can produce those same metabolites in response to non-illegal substances. For example, taking Ibuprofen can cause your body to produce the metabolite that will cause you to test positive for marijuana; certain antihistamines can result in metabolites for LSD; and yes, poppy seeds can cause your body to produce metabolites that test positive for opiates.
The second important thing to note is what exactly we mean when we talk about the accuracy. Generally, any test can fail in one of two ways: (1) either the test can produce a positive result falsely (a false-positive) or (2) the test can fail to produce a positive result when it is warranted (a false-negative).
In the technology world, there is a common joke: “Fast, cheap, good: you only get to choose two.” The same rule seems to apply in the world of drug testing. Generally, the cheaper a test is, the less accurate it is likely to be. Generally, two main types of drug tests exist for urinalysis: Immunoassay tests and chromatography tests. The first is cheaper and less accurate, the latter thought to be more accurate and is correspondingly more expensive.
Though several kinds of Immunoassay tests exist, one of the most common is the Enzyme Multiple Immunoassay test (EMIT). While under ideal conditions with properly trained personnel the tests claims to have a 96% accuracy rate, independent studies have shown that under typical conditions the EMIT test yields false positives up to 37% of the time. Because of how cheap the EMIT test is, it is the most widely used for pre-employment screening. A brochure from Siemens claims that this test is used 87-90% of the time.
The gas chromatography/mass spectrometry (GC/MS) test, the most common gas chromatography drug test, is largely believed to be much more accurate, although clearly the accuracy of this test is reliant on proper collection and laboratory techniques. This test is also considerably more expensive, costing roughly $80/test.
This measure of accuracy is not the point of my essay, and we can leave it to the medical professionals to fight about how accurate or not accurate these tests are. Lest my eyes glaze over from searching through and trying to understand articles from Pubmed, let’s move on.
Even if a test is accurate, is it worth it?
Well, that depends. We first have to start this discussion with a discussion of Bayesian statistics. Before you stop reading, don’t worry, this is not going to be as bad as you think: just hold on.
As background information on Bayesian statistics, I will quote myself from an earlier post.
I recently read Nate Silver’s book The Signal and the Noise, and the primary topic of that book is Bayesian statistics. For those that haven’t read the book, shame on you, but let me summarize briefly. Bayesian statistics is a mathematical model that helps us determine truth by looking at both past data, our assumptions, and new data. Silver argues convincingly that we tend to over sample new data and ignore old data, and that doing so skews our perception of what current reality is. We, as a modern society obsessed with the new, tend to throw out the old for the new when the old provides important context and relevant data for interpreting the present. This book changed the way I think about the world significantly.
As an example that is particularly clarifying, I will take one directly out of his book. Let me set up the background data for you: a woman in her forties has a 1.4% chance of getting breast cancer. Mammograms will incorrectly claim that a woman has breast cancer 10% of the time (false positives), and will fail to detect she has breast cancer 25% of the time (false negatives). Given this data, what is the chance that a woman in her forties that tests positive for breast cancer from a mammogram has breast cancer? Take a second to actually come up with a guess before you read on.
What did you come up with? Seriously, try to come up with something. If you are like me, you went with more than 90%. My reasoning was that if the test only has a 10% of being a false positive (so like 90% of the time a positive test is actually positive) and that not every cancer is detected, my brain immediately concluded that it had to be more than 90%. Had to be.
So what is the answer? 10 percent. 10 PERCENT, that a woman in her forties that tests positive for breast cancer actually has breast cancer. Yes really. If your reasoning was like mine, you completely threw out the old data (the 1.4% chance of a woman in her forties getting cancer) for the new data. Luckily for us (perhaps unluckily for society), we are not alone. Only 3% of people actually came up with the correct result when presented this scenario, and there’s no guarantee that all of that three percent didn’t arrive there purely by chance.
So while “accuracy” can be described as a function of false-positives and false-negatives, the true “accuracy” of a drug test will largely depend on an estimation of the initial condition, i.e. the percent of the population an employer is dealing with that uses a particular drug. Putting aside all of the controversy about how “accurate” a drug test is aside, let’s just for a moment assume that the test is fairly “accurate,” with only a 4% false positive rate, and a 10% false negative rate.
Let’s for a moment focus on adults aged 26 or older. Among this overall population, one study indicates that 0.6% of this population uses cocaine. Let’s say you are hiring some older than 26, and you want them to take a pre-employment drug-screen and the test comes back positive for cocaine. What is the probability that that person actually uses cocaine?
The Bayesian formula is pretty simple, but this is not a math essay, so I’m not going to necessarily explain how to get to this result, but you can fairly easily do this on your own. The answer: 5%. Again, given that the test only gives false positives 4% of the time, gives false negatives 10% of the time, a person over the age of 26 who tests positive for cocaine is only 5% likely to actually have done cocaine. Among adults over 26, the incidence of any illicit drug use is 6.3%, so a positive test for any drug would only indicate that that individual is 38% likely to be a drug user.
So what did the drug test actually tell you? Only that your potential employee is more likely to be doing cocaine than the general population. Perhaps testing positive warrants additional questions or another test, but on its own it is by no means conclusive. And I would argue that you are might be excluding a potentially good employee than to be preventing a drug-using employee from coming onto your work force.
(Note that multiple tests are likely to be more conclusive than one test alone. For example, the random over 26-year-old was only .6% likely to be using cocaine before the first test, but is 23% likely to be using cocaine after the first test. Thus 23% becomes our initial estimation for the second test. If the person tests positive again, (s)he becomes 74% likely to be using cocaine. A third positive test will mean that the potential employee is 96% likely to be using cocaine.)
Should an employer do pre-employment drug screening? I’m not here to answer that question. But the point I’m trying to make here is that if an employer does the drug screens, (s)he ought to put them in the proper prospective, and know exactly what information a positive test yields Sure, information deemed from drug-screening can add additional information, but the information gleaned from a single drug test is not completely conclusive. Employers who do drug testing might consider the drug test as part of a totality test when hiring an employee (with other factors such as prior experience, interview, etc), but it shouldn’t be completely disqualifying test. Employers who are serious about maintaining zero-tolerance drug policies should perform more than one test if they want accurate information.
Employers (and lawyers who advise employers) ought to carefully consider their estimation of the incidence of drug use among the type of employees they are trying to hire in deciding whether or not to use pre-employment drug screens—the lower the incidence in the population they are trying to target, the less information a positive drug test is likely to give. Given the costs of testing, certain employers may find that the cost of the test outweighs the information given from the test. Costs of drug-testing are not only the costs of the actual test, but costs of compliance with federal regulations on medical record-keeping.
Employer’s certainly don’t want to have employees that use illicit drugs—drug-using employees can be less productive and especially in safety-conscious fields can be a hazard in the work-place. But how should employers prevent this? I would argue that habitual illicit drug use will manifest itself in ways that are pretty easily picked up without the use of drug-testing: habitual illicit drug users are more likely to have problems with breaks in their resume etc. And I would say that if you can’t pick up whether someone is a habitual cocaine or heroin user during an interview, you should seriously re-consider how you do interviews.
Bottom line: I’m not trying to argue that drug-testing is worthless, but only that pre-employment drug tests only reveal so much information: that the information gleaned from a single positive test ought not be viewed as dispositive of whether that potential employee is using drugs. Employers should be fully informed and make careful decisions as to the procedures for drug-testing and in evaluating the information that drug tests give as part of an overall evaluation of whether to hire someone.
The information about the types of drug tests and the accuracy of the drug tests come from:
Scott S. Cairns & Carolyn V. Grady, Drug Testing in the Workplace: A Reasoned Approach for Private Employers, 12 Geo. Mason L. Rev. 491 (1990)
Kenneth G. Dau-Schmidt et al., Legal Protection for the Individual Employee (4th ed. 2011).