Thursday, October 3, 2013

Coarctation Prediction in Neonates With PDA

Introducing a JavaScript interpretation of the recently published "coarctation prediction model" from Vanderbilt University Medical Center.

I particularly enjoy little challenges such as this one, issued by the authors of a recent JASE article:


... can be easily integrated into daily work flow by entering the CPM equation into a digital picture archiving and communication system.

(Actually, I find it much easier to put things like this on the web, rather than in our PACS—time for which is often dubbed “non-productive” by administrators. Sigh.)

The article is:

A Clinical Prediction Model to Estimate the Risk for Coarctation of the Aorta in the Presence of a Patent Ductus Arteriosus
Soslow JH, Kavanaugh-McHugh A, Wang L, Saurers DL, Kaushik N, Killen SA, Parra DA.
J Am Soc Echocardiogr. 2013 Sep 23. pii: S0894-7317(13)00658-5

The authors developed a model (a “logistic regression model”) to estimate the probability of developing a coarctation of the aorta, in neonates with a patent ductus arteriosus. If you were trained as I was, the admonishing rule was always that “you can’t rule out a coarctation in the presence of  a large PDA”. And now, it looks like maybe we can.

Using 5 simple linear measurements obtained (easily?) from the suprasternal notch view, several ratios are calculated and plugged into their formula. The resulting calculation is the likelihood (probability) of developing a coarctation. According to the authors, patients below the cutoff of 15%, “no longer require observation” and for patients above the 60% cutoff, we should “continue inpatient observation”.

I find it interesting that a model such as this performs as well as it does without using a z-score. Maybe that’s because z-scores are really only good for telling us if a particular measurement is normal— and not so good at telling how abnormal the measure is. Or, maybe it is because the existing z-scores are not particularly well suited for neonates. Or, maybe still, it’s because the arch anatomy changes dramatically when the ductus closes such that previously normal measurements become abnormal.

Perhaps it is a little of all of the above.

Anyhow, the inputs and calculations are fairly straightforward and easily adapted to a web calculator:

http://parameterz.github.io/vanderbilt-cpm

I thought this was worthy of it’s own project and so, it is now on GitHub (which is another little experiment). The live page is served directly from version control.
Bam.

I like it like that.