In 2013, researchers put a short arithmetic question to 61 people at a Boston-area hospital. Twenty-four were attending physicians. Twenty-six were interns, residents and fellows. Ten were medical students.
The question went roughly like this. A disease shows up in 1 of every 1,000 people. A test for it returns a false positive 5% of the time. Someone tests positive, and nothing else is known about them. What are the odds they really have the disease?
The right answer is about 2%. The most common answer was 95%. Twenty-seven of the 61 gave it. Arjun Manrai and colleagues published the results in JAMA Internal Medicine in 2014, reporting that “approximately three-quarters of respondents answered the question incorrectly.” The guesses ranged from a twentieth of a percent all the way up to 96%.
That study isn’t really about bad arithmetic. It’s about how a test result feels. Positive feels like a verdict. It isn’t. A result is evidence that moves the odds, and how far it moves them depends on something the test itself knows nothing about: how likely your condition was before anyone drew your blood.
Tests do three different jobs, and the job changes the meaning
A test can be doing one of three things, and the same test can do all three depending on who is taking it.
Screening hunts for a hidden problem before symptoms appear. Diagnosis tries to explain a symptom or an abnormal finding that already exists. Monitoring tracks a condition you already know about.
The National Library of Medicine’s MedlinePlus service describes the range of reasons blood gets drawn: to help diagnose diseases, to check how well the liver, kidneys, heart or thyroid are working, to see whether a treatment is working, and to follow long-term conditions like diabetes or high cholesterol over time.
Take a blood glucose test. In a 45-year-old with no complaints, it’s screening. In someone who’s lost 15 pounds without trying and can’t stop drinking water, it’s diagnostic. In someone already treated for diabetes, it’s monitoring. Same tube of blood, three different meanings.
The National Cancer Institute draws the line between the first two bluntly in its patient screening overview: “Screening tests usually do not diagnose cancer.” An abnormal mammogram or stool test doesn’t establish anything. It routes you to a diagnostic test, usually a biopsy, which is the one that answers the question.
That distinction matters for a reason most people never hear.
Why screening guidelines don’t apply once you have symptoms
The US Preventive Services Task Force is the body most US screening schedules trace back to. Volunteer experts in prevention and primary care have staffed it since 1984. The federal Agency for Healthcare Research and Quality has pulled the panel together since 1998, under authority Congress granted.
The Task Force grades each recommendation A, B, C or D, or issues an I statement when the evidence isn’t good enough to judge. An A grade means high certainty that the benefit outweighs the harm by a substantial margin, and the suggested practice is to offer the service. A D grade means the Task Force is reasonably sure the service has no net benefit or does more harm than good, and the suggested practice is to discourage it. The panel states that it doesn’t factor in cost when assigning a grade.
This is the part people miss. The panel writes for one audience only: people showing no signs and no symptoms of the condition in question.
So if a guideline says routine screening for something starts at 45, and you’re 38 with a symptom that points at that exact condition, the guideline isn’t about you. You’re not a candidate for screening anymore. You’re a candidate for diagnosis, and the rules about when and how often are different. People talk themselves out of getting checked using a number that was never written for their situation.
Sensitivity and specificity, without the jargon
Two numbers describe how a test performs, and they measure opposite failures.
Sensitivity is how good a test is at catching the people who have the condition. A test that’s 99% sensitive finds 99 out of every 100 real cases and misses one. High sensitivity means the test rarely lets something slip by.
Specificity is how good a test is at correctly clearing the people who don’t have the condition. A test that’s 99% specific correctly clears 99 out of every 100 healthy people and wrongly flags one. High specificity means the test rarely cries wolf.
Every test has a cutoff somewhere, and moving it trades one failure for the other. Lower the bar and you catch more real disease and also flag more healthy people. Raise it and you spare healthy people the scare and miss more real cases. There’s no setting that avoids both. Anyone who tells you a test is simply “accurate” is collapsing two different numbers into one word.
Clinicians struggle with this too. A 2015 systematic review in BMJ Open by Penny Whiting and colleagues pulled together 24 studies of how health professionals handle test accuracy information. Most of those studies looked at how well people could update the probability of disease after a positive result. The reviewers found those estimates were generally poor, and that the errors ran in a particular direction: people tended to overestimate how likely disease was. The review also found something useful. When the same information was presented as counts of actual people instead of percentages, estimates got better and faster.
So let’s count people.
The arithmetic that surprises everyone
Picture 100,000 people. A condition affects 1 in 1,000 of them, so 100 have it and 99,900 don’t. Everybody gets a test that is 99% sensitive and 99% specific. By any ordinary standard, that’s an excellent test.
Of the 100 people who have the condition, the test catches 99 and misses 1.
Of the 99,900 who don’t, the test correctly clears 98,901 and wrongly flags 999.
Add up the positives: 99 real ones and 999 false ones, for 1,098 total. Divide 99 by 1,098 and you get about 9%.
Read that again. A 99% accurate test, run across a whole population for a condition that affects 1 in 1,000, produces roughly nine real cases for every 100 positive results. About 91 of every 100 people who got the bad news don’t have the thing.
Now change exactly one thing and keep the test identical.
Instead of testing everyone, test only people whose symptoms already make the condition reasonably likely, say 1 in 10. Picture 10,000 of them. A thousand have it, 9,000 don’t.
The test catches 990 of the 1,000 and misses 10. It wrongly flags 90 of the 9,000. That’s 1,080 positives, of which 990 are real. Divide and you get about 92%.
Same test. Same sensitivity. Same specificity. Nine percent versus 92 percent. The only variable that changed was who walked in the door.
This is the whole ballgame, and it’s why a doctor who orders a test because of something you said will treat the result very differently from an identical number that turned up on a panel nobody had a reason to run.
One more thing the numbers show, and it cuts the other way. In that first scenario, 98,902 people got a negative result and exactly one of them had the condition. When a condition is rare and the test is good, a negative result is close to conclusive. Rare conditions make positives weak and negatives strong.
A caveat worth stating plainly: real tests rarely post clean 99/99 numbers, real populations aren’t uniform, and no clinician consults a table to set your pre-test probability. They estimate it from your history, your exam and their experience, and that estimate carries real uncertainty. The underlying math, though, is settled and uncontroversial. It’s been taught in medical statistics for decades. What’s hard isn’t the formula. It’s holding onto it while a scary number sits on the screen in front of you.
What a false positive costs
The cost is the chain of events a flagged result sets off, not just a bad afternoon.
Ishani Ganguli and colleagues surveyed practicing US internists about this in 2019 and published the results in JAMA Network Open. They got 376 completed responses from a research panel built to represent American College of Physicians membership nationally, then weighted the answers. Nearly every respondent, 99.4%, said they’d experienced a cascade of follow-up testing after an incidental finding. And 94.4% said they’d experienced cascades that produced no clinically meaningful outcome at all.
Of those physicians, 68.4% pointed to psychological harm in the patient, 57.5% to a financial hit, and 15.6% to physical harm. On the most recent cascade they could recall, about a third of physicians said the test that turned up the finding may not have been appropriate to order in the first place. More than half said no follow-up guideline existed that they knew of.
False negatives have a cost too, and it’s a quieter one. The National Cancer Institute points out that a person told they’re clear may brush off a symptom that shows up later. A clean result is not a reason to ignore your body.
Researchers are still working on better ways to describe this. A 2026 review in Biochemia Medica by Farrokh Habibzadeh argued that sensitivity, specificity and predictive values don’t answer the two questions clinicians ask most, which are how often a test gets it wrong and whether its mistakes lean toward false positives or false negatives. The paper proposes measures for exactly that. Treat it as an active area of methods research, not settled practice.
Preparation is part of the test, not a formality
If you’ve ever wondered whether the fasting instruction really matters, it does, and not a little.
MedlinePlus explains why. Nutrients from your last meal are still circulating when the needle goes in, and they change certain results. Fasting usually means nothing but water for several hours or overnight, but the required window varies by test, so ask instead of guessing. For some blood draws you may be told to drink extra water so there’s more fluid in your veins. For some urine tests you may be told to drink 15 to 20 minutes beforehand. Some tests ask you to skip hard exercise first.
Two rules worth separating. Tell whoever orders the test about every prescription drug, over-the-counter medicine, vitamin and supplement you take, because some of them shift results. But do not stop taking anything on your own. MedlinePlus is explicit that medication changes come from your provider, not from you.
A small deviation from the instructions can produce a result that looks meaningful and isn’t. If you didn’t follow the prep, say so before the draw, not after the result.
Six questions to ask when a result comes back positive
These are ordered by how much they’ll change what happens next, not by how natural they feel to ask.
-
“Before you ran this, how likely did you think it was?” This is the single most useful question in the article, because the answer tells you how much weight the result deserves. A clinician who says “honestly, pretty unlikely, I was covering a base” is telling you something important.
-
“For me, is this a screening test or a diagnostic one?” The answer determines whether published screening guidance applies to your situation at all.
-
“What test would confirm or rule this out, and how good is that one?” Most positives are an invitation to a second, better test. Find out what it is before you lose a night to it.
-
“What happens if we just repeat this in a few weeks?” Sometimes waiting is the medically correct choice, not a stalling tactic. Sometimes it isn’t. Ask which this is.
-
“If this turns out to be a false alarm, what will finding that out have cost me?” Time, money, radiation, a biopsy. You’re entitled to know the price of the follow-up before you agree to it.
-
“What would change your mind in either direction?” This turns a verdict back into a working hypothesis, which is what it was all along.
When a result genuinely can’t wait
Most of this article argues for slowing down. This is where that advice stops.
If a person from the clinic or the lab calls you instead of posting the result to the portal, treat it as urgent and call back the same day. Someone decided it shouldn’t wait for your next visit.
If you’re told a screening test was abnormal and you need a diagnostic test to follow up, book it and keep the appointment. The delay between an abnormal screen and the test that resolves it is where real harm happens, and it’s entirely avoidable.
If your symptoms are getting worse while you wait on a result, call. A pending test is not a reason to ride something out, and neither is a negative one. A test that came back clean does not cancel a symptom that’s escalating.
And if a test for a treatable infection comes back positive, move on the confirmatory testing quickly, because timing affects both your treatment and whether you pass something to someone else.
One more category of test, and why it’s a separate problem
Tests you buy yourself sit in different regulatory territory. The FDA generally doesn’t review a test before it goes on sale when the seller aims it at general wellness, at low-risk medical uses, or at something that isn’t medical at all. When the FDA does review a test, it looks at analytical validity, meaning whether the test measures what it says it measures, and clinical validity, meaning whether that measurement predicts a state of health. The FDA’s guidance to consumers is direct: these tests aren’t a substitute for a regular health care evaluation, and results should be discussed with a provider. That page’s content is dated December 2019, so check the current version for anything specific. What at-home kits and consumer lab panels do well, and where they fall down, is a longer conversation that belongs in its own article.
The one thing to take away
A positive result is a change in the odds. That’s all it is.
Two numbers determine what it means: how well the test performs, and how likely the condition was before you took it. You can ask about both, and a good clinician will answer both without being annoyed.
Practically, do this. Write down the result, the date, the name of the lab, and the specific next step with a date attached. Bring that page to the follow-up visit. If nobody has told you what the next step is, that’s the question to ask before you leave, call or sign off the portal message.
Reading the range printed next to the number, and deciding what a single out-of-place value means, are related but separate skills with their own pitfalls. They’re worth learning next.
Keep exploring: home medical tests and lower medical bills.
Sources
- MedlinePlus, National Library of Medicine. “What You Need to Know About Blood Testing.” medlineplus.gov
- MedlinePlus, National Library of Medicine. “How to Understand Your Lab Results.” medlineplus.gov
- MedlinePlus, National Library of Medicine. “How to Prepare for a Lab Test.” medlineplus.gov
- National Cancer Institute. “Cancer Screening Overview (PDQ) – Patient Version.” cancer.gov
- NIH News in Health, National Institutes of Health. “To Screen or Not to Screen? The Benefits and Harms of Screening Tests.” newsinhealth.nih.gov
- U.S. Preventive Services Task Force. “About the USPSTF.” uspreventiveservicestaskforce.org
- U.S. Preventive Services Task Force. “Grade Definitions.” uspreventiveservicestaskforce.org
- Office of Disease Prevention and Health Promotion (MyHealthfinder). “Get Screened.” odphp.health.gov
- U.S. Food and Drug Administration. “Direct-to-Consumer Tests.” fda.gov
- JAMA Internal Medicine (PubMed). “Manrai AK et al. "Medicine's uncomfortable relationship with math: calculating positive predictive value." JAMA Internal Medicine, 2014. PubMed. DOI: 10.1001/jamainternmed.2014.1059.” doi.org
- BMJ Open (PubMed). “Whiting PF et al. "How well do health professionals interpret diagnostic information? A systematic review." BMJ Open, 2015. PubMed. DOI: 10.1136/bmjopen-2015-008155.” doi.org
- JAMA Network Open (PubMed). “Ganguli I et al. "Cascades of Care After Incidental Findings in a US National Survey of Physicians." JAMA Network Open, 2019. PubMed. DOI: 10.1001/jamanetworkopen.2019.13325.” doi.org
- Biochemia Medica (PubMed). “Habibzadeh F. "Beyond accuracy: how often and in which direction diagnostic tests err." Biochemia Medica, 2026. PubMed. DOI: 10.11613/BM.2026.030101.” doi.org