Would you consider yourself normal?
Average?
Run-of-the mill? Middle of the road? Typical?
The chance that you are correct is vanishingly small.
If you're a clinician, like me, you get a LOT of medical questions. While we've all been asked to look at suspicious rashes, moles, or (gulp) fluids, many of us are also used to being handed a printout of "the labs" and asked to interpret.
Clinical laboratories, in their wisdom, often report the "normal" range of the test being administered. Helpfully, should your result fall out of the normal range, they will flag it somehow. Often with a * or maybe an "H." Or highlighted like this:
So ... who decides what's normal?
The short answer is -- each lab, in the U.S. at least. If you want to keep your lab accredited by the College of American Pathologists, you need to establish or validate your reference values at least annually. But it gets quite a bit more complicated. Because there is a "normal range," which is technically called the "reference interval" and then there are clinical decisions that get made when a test is out of the normal range (or sometimes even if a test is in the normal range), and these two issues are in fact, quite different.
But let's start simply. Here's how we get a reference interval for, say, hemoglobin:
A recipe to create a normal range for hemoglobin concentration
Step 1: Find 300 friends.
Step 2: Realize that your 300 friends are not representative of the general population.
Step 3: Recruit 300 random people.
Step 4: Draw their blood.
Step 5: Calibrate your analyzer using known standards.
Step 5: Measure the hemoglobin concentration in the samples. In the same way. Exactly the same way.
Step 6: Make a histogram. Here's one:
Right off the bat, you see some structure. There seems to be a peak around 13 or 13.5. And the range goes from around 10 to 17.
Now, if you can find 3,000 people (instead of 300), you get an even more refined picture:
Again, we get the sense that there is a peak somewhere below 14, and a range that goes from around 10 to 17.
For you data people, here are the numbers behind that picture:
Mean hemoglobin -- (this is the average as we usually think of it): 13.5.
Median hemoglobin -- (the value smack in the middle of the other values): 13.5
Step 8: Decide how many people you want to freak out.
We're about to set a reference range for this lab value. The more narrow we make it, the more people will fall outside of it, get a little * on their lab form, and run to their closest doctor friend for explanation. The standard practice is to freak out 5% of the people. In other words, we'll make cutpoints such that 95% of the population will fall inside. If you're with me on the math, that means 2.5% of people will have a value that's too high and 2.5% will have a value that's too low.
Here's the thing to keep in mind:
Those people in red aren't abnormal.
"Wait, but you just said they are!"
No, I said we'll mark them as abnormal. It's not the same thing. People in high school called me abnormal -- does that mean I really was?
OK, bad example.
The point is that these distinctions are arbitrary. We could have made the cut-point such that the top and bottom 5% are abnormal, or 10%, or 1%, or whatever. There will always be someone in that group, despite them being perfectly healthy.
Now here's where statistics gets fun.
Let's say you are in perfect health. But remember, through the vicissitudes of random chance, you may fall into the abnormal range on 5% of your blood tests. Assume you get 10 blood tests at a time (this might be a low number for a checkup). What are the chances you will have NO abnormal values. The binomial distribution comes to the rescue here.
Well, 60% of people will have no abnormal values, with around 30% having one abnormal value, 8% or so with 2 abnormal values, 1% with 3 abnormal values and so on.
Assuming you get your blood drawn once a year from ages 40 to 70, the chance that perfectly healthy, normal you would get by without a single abnormal value is about 1 in 5 million. Keep that in mind when you see that little * on your blood work.
Decisions, decisions
To a physician, the number associated with a lab value is less important than what we do about it. Does that number change our clinical decision making? To us, that's really a better definition of normal. And while there are no really reliable data to figure out what percentage of the time an abnormal lab value prompts additional tests or treatments, it is clearly only a fraction of the time a lab value is abnormal.
Deciding on a "cut-off" value of a lab to prompt a clinical decision requires tying the laboratory result to some outcome that really matters like death or political affiliation. For that reason, the medical literature is chock full of studies examining lab values, and looking for associations with various outcomes. Take vitamin D for example.
According to PubMed, there were 4,311 scientific articles about vitamin D published in 2015, up from 3,994 the year before, and 3,885 the year before that. Back in 1999, there were only 1,216 articles. Vitamin D is hot.
And studies have attempted to link low vitamin D levels to everything. Here is a completely non-exhaustive list (I'm not linking to fracture/fall studies -- there are simply too many):
- Anemia
- Asthma
- Cognitive decline
- Miscarriage
- Inflammatory bowel disease
- Erectile dysfunction
Actually, it's hard to find an outcome that someone hasn't linked vitamin D to. Let me think ... Brain tumors? Insomnia? Anxiety?
Whether you believe these studies or not, vitamin D levels may tell us something, and our patients want to know what to do about them. That means you need cutoffs. We need a cold, hard, number to say "this much is too low, give a supplement."
That number, as I pointed out above, is not necessarily the reference interval threshold. In the case of vitamin D, 42% of the U.S. population may be deficient, so using the bottom 2.5% of the typically measured range would barely scratch the surface of the problem.
So how do you decide what is "deficient?" Well, you look at a variety of, at first, observational studies and basically pick a number. A nice round number like <20 ng/mL or <30 ng/mL helps stick in the mind, so authors often go with that. You then analyze the data again, but instead of looking at the continuous relationship between vitamin D levels and outcomes, you dichotomize it (deficient/not deficient) and look at outcomes. A big difference in outcomes across those groups should make you feel like you picked a good cutpoint.
It gets more complicated
A brief aside here. Cutpoints are generally universal (low is low for everyone) but perhaps they shouldn't be. Vitamin D is shepherded around the circulation, in part, by vitamin D-binding protein. Only "free" vitamin D is biologically active, yet all the vitamin D (even the bound part) gets measured when you have your vitamin D level checked. In other words, if your vitamin D binding protein doesn't hold on as tight to vitamin D, your measured level will not be changed, but your physiologically important level may be significantly higher. It turns out that a significant proportion of black individuals have a "looser" phenotype of vitamin D-binding protein, which may explain why vitamin D deficiency appears to be so incredibly common in black populations, and yet clinical outcomes like fracture and osteoporosis are not.
Above, I said we would choose a cutpoint and reanalyze the data. A really rigorous researcher will repeat that deficient/not deficient analysis at multiple cutpoints. Perhaps at every possible cutpoint. As you can imagine, picking a higher cutpoint will classify more people as deficient, and a lower cutpoint will classify fewer people as deficient. The choices both have their tradeoffs. Classifying more people means fewer people who might benefit from treatment will fall through the cracks, but you'll also worry a lot of people needlessly.
In the case of vitamin D, we could argue that the current cutpoint of 20ng/mL might be a bit high (given that we are including roughly half of the U.S. population in the deficient category).
The real proof, though, is in the trial. Once you've decided who is deficient, you can then, you know, treat them. If you decided correctly, treatment should significantly improve the outcome rate that you saw in the observational phase of this whole endeavor.
I will not go into the slew of trials evaluating vitamin D supplementation, but I will say that they have overall been disappointing.
That's right. In the end, after all the rigmarole of observational studies, debates about cutpoints, and clinical trials, we still don't know what normal is for vitamin D. In fact, the U.S. Preventive Services Task Force states that there is insufficient evidence to recommend for or against screening of vitamin D in the general population. We can fight a lot about what is normal, but maybe we shouldn't even be looking in the first place.