The Value of Negative Predictive Value

Let’s say you come to the emergency room with chest pain.  Pretend you are a 40-something-year-old male with a history of high blood pressure but no previous heart problems.  After a few tests the doctor comes into the room to tell you that your EKG and lab work are normal but that you’ll be staying in the hospital overnight anyway.

The next morning, with your repeat EKG and blood work showing no abnormality, you are ushered to the treadmill room for a test called a stress echocardiogram.  You walk, then run, on the treadmill until your pulse reaches about 85% of your maximum rate and the echocardiogram technician takes pictures of your rapidly beating heart.  The doctor supervising the test interprets the pictures and declares you free of significant blockage in the arteries of the heart.  A few hours later you’re discharged—about 24 hours after you arrive—without a firm diagnosis and with the hopeful anticipation that your hospital bill won’t be too big.

This scenario describes the way we go about evaluating most patients who come through the ER with chest pain.  I say “most” patients because there are often exceptions.  Someone with an abnormal EKG, elevated cardiac enzymes, or features to their chest pain that we consider “high risk” would likely bypass the above evaluation and proceed directly to the cath lab.  Other patients whose symptoms are so clearly noncardiac might get dismissed with instructions to seek out their primary physician’s guidance for further evaluation.

The vast majority are admitted and treated as described above.  In the business we refer to this method as a “rule-out,” as in “we’ll rule out a heart attack (with the enzyme tests) and then do a stress test.”  Doctors are quite comfortable with this for good reason.  For many years it has been a safe and effective way to figure out who is really having heart problems and who is just still trying to digest last night’s Mexican dinner.

You can turn to demographic studies to get a sense of what might be happening with the people who fall into this category (we call them “moderate risk”).  Out of every hundred people admitted for a “rule out” only a handful—five or ten—will end up having coronary disease as the cause for their symptoms.  If you were to take all 100 of them to the cath lab and look at their arteries you would find absolutely no narrowing in the majority.  Using the current method described above we admit an overwhelming preponderance of patients with normal coronary arteries just to find those few with blockage.  Most patients coming in with chest pain can plan on at least a few meals of hospital food before they get their answer.

What we really need is a test that can reliably exclude those patients with clearly normal coronary arteries and that can do it quickly with minimal risk and expense.  When you come in the ER with chest pain you need a test that can tell you with as much certainly as scientifically possible whether your coronary arteries are normal or not.  If your test is normal you can be discharged with the knowledge that your chest pain is not your heart.  If the test is abnormal you can then come into the hospital for further testing.

In statistical parlance what you need is a test with an extremely high negative predictive value.

A little primer on medical tests is in order.  The screening tests we use (such as stress tests, mammograms, blood PSA, etc) are not perfect.  Each test has various limitations that make it so we cannot conclude that an abnormal test definitively signifies the presence of disease (or that a normal test predicts that absence of disease).  We rate the accuracy of a test using complex principles called sensitivity and specificity.  These values (which are not particularly intuitive and continually vex new medical students and residents) can be translated into the more useful terms positive (PPV) and negative predictive value (NPV) by incorporating into the equation the prevalence of a particular disease in the population being tested.

PPV  is defined as the likelihood that a positive (or abnormal) test correctly identifies the presence of disease (if a stress test has a PPV of 80%, then you, as a patient, have an 80% chance of actually having significant coronary blockage with an abnormal stress test).  Just the opposite, the NPV is the likelihood that a normal test will prove the absence of disease (a NPV of 90% means that you have a 90% chance of being normal if you have a normal test).

This may alarm some people.  We are accustomed to believing that our medical tests are precise all the time, but most are not, and we doctors have to, in effect, decode the test results (interpret the interpretation, if you will) based on the individual patient’s clinical profile.

At least one test is changing that, and if you plan to head to an ER anytime soon with chest pain you should know about it.  The technology has now advanced so far that we are now able to use computed tomography (CT, or “CAT” scans) to obtain beautiful images of the heart and its arteries—images good enough to be able to immediately determine if a patient with chest pain has normal coronary arteries or not.

The coronary CT angiogram, or CCTA, is a study where the patient undergoes a chest CT scan with intravenous contrast injection.  The test is noninvasive (only an IV line is needed), low risk and quick, and it produces beautiful picture (see linked images).

Best of all CCTA has an astoundingly high sensitivity for the presence of coronary disease.  In the moderate-risk population referred to above, this translates to a near-perfect NPV.  Two recent papers (in Circulation and JACC) assessed CCTA in the ER and determined that the sensitivity and NPV for this test is 100%.

There are downsides to CCTA.  It involves radiation (about the same amount you’d get with a nuclear stress test) and so we obviously don’t want to do this in anyone who might be pregnant.  It also doesn’t do as good a job of proving the presence of significant coronary disease as it does the absence thereof, and you might need more testing if the study is not normal.

So, if you’re planning to come to the ER with chest pain you may find that we are able use CCTA to get you in and out a little more quickly than we used to.  That’s good news for all involved (unless you were hoping to enjoy a few more hospital meals).