Meta-analysis data needs to be handled “with care”

Dierk Vorwerk

The level of evidence for data obtained through registries is usually regarded as low. This level is unlikely to be improved by combining data in a review which usually suffers from the weakness of its weakest links—its underlying source material. Also, the basic assumption of combining small sets of heterogeneous pieces of data for review needs to be handled “with care”, writes Dierk Vorwerk.


Paraskevas et al published a systematic review on stroke and death rates following carotid artery stenting and carotid endarterectomy in contemporary administrative dataset registries (Eur J Vasc Endovasc Surg 51 [2016] 3-120).

The authors were motivated by the 2011 AHA  guidelines that expanded the indications for carotid stenting as an alternative to surgery in  patients with symptomatic carotid artery stenosis in centres with a proven complication rate of less than 6%  and to selective patients with asymptomatic carotid artery stenosis with a proven complication rate <3%. These expansions are based on the results from randomised controlled data from CREST.

The authors found that for symptomatic patients, the 30 day combined complication rate (stroke and death) was higher than the threshold of 6% in 72% of the included registries and carotid artery stenosis but in only 11% for endarterectomy. For asymptomatic patients, they found a complication rate >3% in 43% of carotid artery stenting patients but in only 5% of endarterectomy patients.

The authors claim that data from registries are helpful because they report “real world data” instead of selective data as experimental and randomised studies do. This leads us to believe that combining many smaller “real-world” data in a review will help us to gain a global view.
Unfortunately this is not true.

This basic assumption, of combining many smaller “real-world” data in a review, needs to be handled with care. Data collections from registries have many weak points, are open to bias and many confounding factors – as the authors correctly stated. Their level of evidence is usually regarded as low. This level is unlikely to be improved by combining data in a review which usually suffers from the weakness of its weakest link—its underlying source material.

A real-world scenario might be better obtained if one huge, uniform registry is evaluated including many patients with a rigid conformation to the same rules of data collection, data selection and judgement. Unfortunately, this is not the case in a review that combines many different smaller registries with heterogeneous backgrounds.

When comparing treatment options that are so close to each other with regard to success and complication rates such as carotid artery stenting and endarterectomy, utmost care has exercised in handling data which relies on weak sources such as registries. Success and complications depend on very small details such as level of training of the performing physician, delay between a first ischaemic event and time of treatment, age, anatomy and also how an adverse event is evaluated. Has the neurological outcome of carotid artery stenting or endarterectomy been evaluated by the performing surgeon/interventionalist, or by an independent neurologist? Or, in an even worse scenario, was the evaluation performed differently between one cohort and the other? How were neurological sequelae qualified and monitored? Were transient ischaemic attacks counted as strokes? All these important details are usually unknown, or widely heterogeneous, particularly in data from registries.

This review compares registry data which are necessarily vague with a fixed cut-off of 3% or 6% complication rate which have been evaluated from randomised trial data. This is difficult to swallow.

The analysis may correctly state again that stroke is more frequent after carotid artery stenting than after endarterectomy. So what? It is pretty logical that ischaemic events are an intrinsic and typical complication of a treatment that uses a transluminal approach just as peripheral nerve damage is a typical complication of open surgery. That is not surprising and we know it well from randomised trials. The more illuminating question is: are these strokes lasting and how severe they are?

The authors still continue to count strokes and deaths without any differentiation. But this is not one complication but a variety of them of different severity. Stroke is an unhappy event but is not uniform. A stroke can have subtle or massive clinical signs and can sometimes be difficult to evaluate. A stroke that completely resolves is a terrifying experience as is a myocardial ischaemic event but cannot be regarded as equal to a stroke that is disabling, ie one that changes a person´s life drastically. 

Endarterectomy is an established technique and perhaps better than stenting with regard to neurological events – but it not better with regard to neurological outcome. Carotid artery stenting is still evolving and there is even some hope for improvement with the introduction of newer stent designs to better control its inherent problems.

We do know from two large, randomised, controlled trials (the long-term results of ICSS [International Carotid Stenting Study] and CREST [Carotid revascularization endarterectomy vs. stenting trial]) that carotid artery stenting is as good as endarterectomy concerning disabling strokes and neurological outcome, and that there is merely no difference between the two procedures.

This needs to be openly discussed with symptomatic patients and it is mandatory both for surgeons, neurologists and interventionalists to give them unbiased information. We should stop arguing about tiny differences and using unreliable data to blur the vision on carotid artery stenting. And, finally, it is time to re-evaluate our strategies on how to treat asymptomatic carotid artery stenosis, which are relying on ancient data that does not take into account up-to-date best medical treatment regimens.

Dierk Vorwerk is chairman of the Department of Diagnostic and Interventional Radiology, Klinikum Ingolstadt, Ingolstadt, Germany. He is also the editor-in-chief of CardioVascular and Interventional Radiology (CVIR)