A reading under 300 on a common heart-failure blood test clears most emergency-room patients of a heart-failure crisis, missing only about 2 in 100 who turn out to have one. That rule-out number is the cleanest result in a new analysis of NT-proBNP, the peptide fragment the test measures, pooled from 31 studies and 26,948 patients and published online October 17 in the European Journal of Heart Failure ↗.
NT-proBNP is a piece of molecular bookkeeping. When the heart's walls are stretched by too much fluid or pressure, heart muscle cells release a hormone called B-type natriuretic peptide. That hormone tells the body to shed salt and water and relax blood vessels. It is cut from a longer precursor, and the inactive leftover, the N-terminal fragment, is what piles up in the blood and is easy to measure. A high NT-proBNP is a readout of a heart under strain. The heart-failure drug nesiritide ↗ is a lab-made copy of that active hormone. It is the same molecule used in reverse, given during a crisis to open vessels and clear fluid.
One cutoff rules out, a sliding scale rules in
The meta-analysis, led by Guillaume Baudry with the natriuretic-peptide researcher James Januzzi among the authors, tested two ways of reading the number. A single fixed floor of 300 ng/L is very good at ruling the crisis out. Its sensitivity was 97.9 percent, meaning almost everyone with real heart failure scores above it. Its negative predictive value was 97 percent, so a reading below the floor is rarely wrong when it says no crisis. But the same cutoff is a blunt instrument for ruling heart failure in. Its specificity was only 46 percent, so more than half of the people above 300 do not have an acute crisis. Overall accuracy, a score where 1.0 is perfect and 0.5 is a coin flip, was 0.73.
Switching to thresholds that climb with age did better. Using 450 ng/L under age 50, 900 between 50 and 75, and 1,800 above 75 lifted overall accuracy to 0.84 and balanced the two kinds of error, holding sensitivity at 86 percent while raising specificity to 82 percent. Older hearts run higher baselines, so a flat cutoff reads the old and the young differently. The sliding scale corrects for that.
The groups that need the test most are measured least
The catch is in who was studied. Two groups make NT-proBNP hardest to read, and both are thin in the data. In obesity, fat tissue clears natriuretic peptides faster, so levels run low and a real crisis can hide under the threshold. In chronic kidney disease the fragment clears slowly, so levels run high and a healthy heart can look alarming. These are exactly the readers this section serves most. Obesity is the shared starting point for the GLP-1 drugs the field is built around. The weight that patients lose on those drugs pushes their NT-proBNP back upward over months.
Yet the obesity subgroup rested on 3,562 patients and only the age-dependent thresholds. Its accuracy of 0.82 is one the authors call broadly comparable to the general population. Chronic kidney disease rested on 2,271 patients, where every fixed threshold scored below 0.80. The paper's own conclusion flags that the evidence in both groups is limited, in sample size and in the range of cutoffs tested. The reassuring top line, that the thresholds hold across subgroups, is carried by the smallest slices of the dataset.
Why the number is never just a number
This is the second time this year the peptide behind this test has turned out to be shakier than its routine use suggests. We reported in June that NT-proBNP drifts by about a third on its own ↗ between draws in the same stable person, a noise floor that swallows small changes. Put the two together and a single NT-proBNP value in the emergency room carries two kinds of slack at once. The thresholds are subgroup-sensitive, and the molecule wobbles on its own. The 300 floor is still a strong way to send someone home. Reading a borderline number as proof of a crisis, especially in a patient who is obese or on dialysis, asks more of one peptide than it can give.