Cancer Blog #22
By Brian Zimmerman
Begun on July 31, 2021
Email: dyingman1@yahoo.com
My Dying Words
Entry #22
January 27, 2022
A Reluctant Attempt
I decided with reluctance to give this discussion a try, but am not sure how well it will go. I want to give a talk about lab screening tests, though the principles I give can apply to any kind of medical screening tests, laboratory or not. So, what is a (laboratory) screening test? It is any lab test used to determine if you have a condition or disease. For instance, PAP smears for women to screen for cervical cancer, or PSA (prostate specific antigen) for men to screen for prostate cancer.
Let me say at the outset that I do have some background on these things. I didn’t just get the information off of the internet. I created lectures on these principles for hospital clinical laboratory students that I taught clinical chemistry (many years ago). In fact, I noticed that the pathologists would sneak in the back of the classroom because 40 years ago, these were cutting edge ideas. Now I’m sure every lab student (as well as every MD) are taught them.
The basic concept to grasp is that no screening test is perfect. Now, but you think, “Oh, brother, very profound”, let me have the chance to explain how this uncertainty has been more carefully defined than that general statement (Warning: this explanation involves math, but don’t quit now. We’re talking only about fractions here).
The Basics
So, here’s the basic principles and terms:
1. To have a screening test that’s cheap and useful, you first need another test that’s better and certain. In the prehistoric days, that better test was called the gold standard. Nowadays it is more modestly termed the reference standard.
2. There are two types of test parameters that most screening tests may have measured (and should be!). One type or set of parameters is the test’s sensitivity and specificity. These have very specific meanings in the test world, so let’s define them.
First, sensitivity is the percentage of people who you know for certain have a disease or condition that will produce a positive screening test result. For instance, if we knew for sure whether or not a group had colorectal cancer (which would be the disease we would be screening for), we could then see how many of those people who had the disease would give a positive screening test using the fecal occult blood test (this is a popular screening test that I’ve seen ads for you to collect at home. It’ a test in which a sample of your poop is collected (fecal) (I know, gross), and then is checked for blood that you can’t see (occult blood). So, if you had a hundred patients with that kind of cancer, and 80 of them had a positive screening test (and, obviously, 20 negative), the sensitivity would be: 80/100 = .8 x 100 = 80% (see, that wasn’t so bad).
Specificity is a related concept. Suppose you had another, different 100 patients who you knew for sure didn’t have colorectal cancer, and they had a fecal occult blood test results of 70 negative (and so 30 positive). Then the screening lab test specificity would be: 70/100 = .7 x 100 = 70%.
Now, how could we be sure these patients did or didn’t have colorectal cancer? The gold standard (or reference standard) used to be (and I guess still is) a colonoscopy in which they would actually go in and look to see if you had the cancer growing inside you.
Let’s Try a Diagram
We can diagram these concepts (and two more that we’ll discuss in a moment) using a 2×2 diagram (just a box divided into 4 smaller squares):
Disease Status (using gold standard)
Has disease No disease
| True positive screening test = 80 a | False positive screening test = 30 b |
| False negative screening test = 20 c | True negative screening test = 70 d |
But, second, there’s another pair of parameters that often are not reported and are far more important for you for a screening test: viz, the predictive value of a positive test (PPV, positive predictive value), and the predictive value of a negative test (NPV, negative predictive value). What do these mean? If someone has a positive screening test result, what are the odds that they have the disease? Using our diagram, it is defined as the % of positive screening test results that true, i.e., going across instead of up and down on the diagram we get PPV = a/a+b, so PPV = 80/80 + 30 = .727 x 100 = 72.7%.
On the other hand, what’s the predictive value of a negative screening test result? Well, again using our diagram, it is the % of negative screening test results that are true negatives (they don’t have the disease), so NPV = d/c+d = 70/70 + 20 = .777 x 100 = 77.7%.
Why Spend Time on this Exercise?
Why go through all this palaver? It’s because our news sources are full of stories about getting and using the home COVID screening test. The question you are sure to ask is: if I have a positive test result, how accurate is that answer? What about if I get a negative test result? The problem is trying to find that information. If the screening test manufacturer (which you can google) will tell you that it is a great test because it has a sensitivity of 90% (meaning it’s catching 90% of people who have the disease), just realize that it is telling you virtually nothing about how accurate your positive screening test result is. Sensitivity and the predictive value of a positive test (PPV, positive predictive value) are not the same and to mix them up is to commit the logical error of confusion of the inverse. So, to say that “this animal is a dog and, therefore, is likely to have 4 legs,” is not the same as saying “this animal has 4 legs and, therefore, is likely to be a dog.” That’s like the difference between sensitivity and PPV.
Think of it this way: the Public Health people would want to know this fact: of all the people who truly have COVID (according to the reference standard), how many have a positive test, in other words, how many are we missing? That’s sensitivity.
You, at home, don’t really care so much about that except in some theoretical way. What you want to know is does my positive screening test result mean I have COVID or not, which is a very different question (the inverse, in fact). In other words, what is the PPV? If my test is negative, how likely is it that I really don’t have COVID? See if the test manufacturer can give that information if you google them and those parameters.
The media is, of course, oblivious to all these distinctions, and, in my humble opinion, really have no business reporting on this subject or any other scientific issue (such as global warming), as they probably are reporting on things they have no understanding of. I’m quite sure the Public Health people (as in the CDC or the local public health department MD) fully understand them (I hope), and most likely your PCP (Primary Care Physician, again, I hope), and likely the hospital epidemiologist and maybe the hospitalists. I guess if you’re symptomatic and go to your PCP or the ED, you could ask them, if, at that point, you still wanted to know. They may do the PCR test (the gold or reference standard, I believe, to test for COVID). I guess if it differs from your screening test, that will give you some clue about reliability (or accuracy, actually. Reliability in this context is something different than accuracy, but that’s another issue.)
Sorry, Two More Factors
Believe me, I hate to mention two more factors, but they are truly important. The first is the disease prevalence, which means how many people in the population have the disease? (the per cent, or %) Obviously, the more people who have a disease in your population, the easier it is to find them. Imagine if you had a population of 100 (to keep it simple), and 90 people had a condition or disease. It wouldn’t take much of screening test if positive to be accurate. On the other hand, if only one person out of the hundred had the disease or condition, you could be pretty certain that a negative screening test would be accurate. The significance of this factor for the COVID disease may not be obvious, but imagine at any one time one million people out of 300 million people had COVID. Now imagine you had 50 million people had COVID. The prevalence in the second case went way up and would make the job of the positive screening test much easier. Would the test manufacture tell you the prevalence of the test population? Doubtful, but it is useful information.
The second factor I want to be sure to mention is the test’s ability to become positive or negative in the presence of disease at different times of the disease’s progress. That ability can be adjusted somewhat by the manufacturer to make the test more or less able to detect the presence of the disease. So, for instance, if you test for COVID very early on in the disease process, the viral load may be so small that there is no good way to detect its presence. Second, the screening test manufacturer might, for various reasons, decide to adjust the test’s ability to become positive in the presence of the disease, which may affect the test’s sensitivity or the PPV. Maybe you want to catch as many diseased people as possible so the manufacturer raises the NPV (negative predictive value of the test, i.e., how accurate is a negative test result) to be as high as possible because they really don’t want someone who has the disease to give a negative screening test result. But, you can see from the diagram that it’s possible that might affect the PPV. There’s no perfect test, remember?
I hope this discussion wasn’t too confusing and was in some way helpful. I just felt with all the COVID concern, that I ought to try to help some with the most likely way you are to be tested. I again apologize for the length, but I felt I should put it all in one entry rather than spread it out. By the way, there are 2 articles (for free) that might be useful if you want more information. The first is: “Sensitivity, Specificity, and Predictive Values: Foundations, Pliability, and Pitfalls in Research and Practice,” by Robert Trevethan, submitted to the Epidemiology section of the journal “Frontiers in Public Health, published 20 November 2017. The second is an article cited by Trevethan. It is: “Understanding and Using Sensitivity, Specificity and Predictive Values,” by Rajul Parikh, et al, in the “Indian Journal of Ophthalmology,” 2008; 56:45-50 (some very useful practical examples using screening tests in ophthalmology for diseases such as types of glaucoma).
Next entry: Another discussion of Paul on how he recommends we prepare to face death.