– AI false positives pose serious danger to patients; cutoff changes recommended

For adrenal insufficiency (AI), reducing false positives means more than reducing resource utilization. Treatments like glucocorticoid replacement therapy can cause serious harm in people who do not actually have AI.

Research published in the Journal of the Endocrine Society makes multiple findings that report authors say could help bring down false positive rates for AI. This retrospective study ultimately analyzed 6,531 medical records from the Imperial College Healthcare NHS Trust in the United Kingdom.

Sirazum Choudhury, MBBS, an endocrinologist-researcher with the trust, served as a co-author of the report. He discussed the study with MedPage Today. The exchange has been edited for length and clarity.

This study ultimately addressed two related but distinct questions. What was the first?

Choudhury: Initially the path we were following had to do with when cortisol levels are tested.

Cortisol levels follow a diurnal pattern; levels are highest in the morning and then decline to almost nothing overnight. This means we ought to be measuring the level in the morning. But there are logistical issues to doing so. In many hospitals, we end up taking measurements of cortisol in the afternoon. That creates a dilemma, because if it comes back low, there’s an issue as to what we ought to do with the result.

Here at Imperial, we call out results of <100 nmol/L among those taken in the afternoon. Patients and doctors then have to deal with these abnormal results, when in fact they may not actually be abnormal. We may be investigating individuals who should really not be investigated.

So the first aim of our study was to try and ascertain whether we could bring that down to a lower level and in doing so stop erroneously capturing people who are actually fine.

What was the second aim of the study?

Choudhury: As we went through tens of thousands of data sets, we realized we could answer more than that one simple question. So the next part of the study became: if an individual is identified as suspicious for AI, what’s the best way to prove this diagnosis?

We do this with different tests like short Synacthen Tests (SST), all with different cutoff points. Obviously, we want to get the testing right, because if you falsely label a person as having AI, the upshot is that treatments will interfere with their cortisol access and they will not do well. Simply put, we would be shortening their life.

So, our second goal was to look at all the SSTs we’ve done at the center and track them to see whether we could do better with the benchmarks.

What did you find?

Choudhury: When you look at the data, you see that you can bring those benchmarks down and potentially create a more accurate test.

First, we can be quite sure that a patient who is tested in the afternoon and whose cortisol level is >234 does not have AI. If their level is <53.5 then further investigation is needed

There were similar findings for SSTs, which in our case were processed using a platform made by Abbott. For this platform, we concluded that the existing cut-offs should be dropped down to 367 at 30 minutes or 419 at about 60 minutes.

Did anything surprise you about the study or its findings?

Choudhury: If you look at the literature, the number of individuals who fail at 30 minutes but pass at 60 minutes is around 5%. But I was very surprised to see that our number at Imperial was about 20%.

This is a key issue because, as I mentioned, if individuals are wrongly labelled adrenally insufficient, you’re shortening their life. It’s scary to think about the number of people who might have been given steroids and treated for AI when they didn’t have the condition.

What do you see as the next steps?

Choudhury: I see centers unifying their cutoffs for SST results and making sure we’re all consistent in the way we treat these results.

From a research perspective, on the testing we’re obviously talking about one specific platform with Abbott, so research needs to be done on SST analyzers from other manufacturers to work out what their specific cutoffs should be.

Read the study here and expert commentary on the clinical implications here.