Neurophysiological monitoring

Every other operating-room technology shows anatomy: where the tumour is, where the tracts are, what was left behind. Neurophysiological monitoring shows something else: function. It does not say where the motor pathway is; it says whether the motor pathway still works, now, this second.

The principle is simple. The motor cortex is stimulated — through the skull or directly on the surface of the brain — and the resulting muscle contraction is recorded. As long as the response holds, the pathway is intact. If it falls, something is happening. And unlike an image, the response does not go out of date: what is being measured is the brain on the table.

There is also a more precise variant: instead of stimulating from above, one stimulates from inside the resection cavity and measures how much current is needed to obtain a response. The less current, the closer the tract. A study of 69 patients with tumours less than a centimetre from the corticospinal tract showed that the relationship is close enough to work as a rangefinder: individual thresholds ran from more than 20 milliamps in some patients to 1 to 3 in others, and at three months only two patients had persistent motor deficit, both from vascular causes and none from mechanical injury to the tract³.

It is the operating-room tool with the strongest conceptual argument. And it is also the one that holds up worst under scrutiny of the evidence, for reasons worth understanding.

The figure that holds up the field

The most cited result is a meta-analysis of 90 reports and 8,091 patients with glioma¹. Those operated with stimulation mapping had severe late neurological deficit in 3.4 per cent of cases, against 8.2 per cent of those operated without mapping. Complete resection was 75 against 58 per cent. The authors conclude that mapping should be universally implemented as standard of care.

Now the small print, which they themselves record. These are observational studies, not randomised ones. And the groups are not comparable in the direction one would expect: among those operated with mapping, 99.9 per cent had the tumour in an eloquent area; among those without, 95.8 per cent. The comparison is between different patients operated by different teams in different eras.

The most influential individual series has the same problem, in its purest form. It compared 100 patients operated between 1985 and 1996 without stimulation with 122 operated between 1996 and 2003 with it, at the same institution². Severe permanent deficit fell from 17 to 6.5 per cent, and complete resection rose from 6 to 25.4. These are spectacular figures. But eleven years passed between the two series, during which MRI, navigation and the team's experience all changed. No adjustment separates the effect of stimulation from the effect of the calendar.

Why there are no randomised trials

The honest answer is published and signed. The argument is one of equipoise: randomising requires genuine uncertainty about which arm is better, and here there is none, because the belief that monitoring protects is already established. Randomising to "no monitoring" would mean deliberately exposing patients to avoidable harm⁹.

The reasoning is circular and its own defenders acknowledge it: we do not randomise because we assume the benefit, and we assume the benefit because we did not randomise. This is not bad faith; it is a real and difficult situation, common to many safety interventions. But it is worth naming, because it bears on how much can be claimed.

The level A that does not say what it seems to

The US neurology academy's guideline gives spinal monitoring a level A recommendation, the highest⁶. It is the argument that appears in every presentation on the subject, and it is usually misread.

What that recommendation says is that the surgical team should be alerted to the increased risk of adverse neurological outcome in patients with important changes in the potentials. The question the guideline set itself, verbatim, was whether monitoring predicts adverse outcomes. And the answer is yes: between 16 and 40 per cent of patients with changes in the potentials developed postoperative paraparesis, paraplegia or quadriplegia in the class I studies.

Predicting is not preventing. The guideline does not assess — because there was no evidence to assess — whether intervening after the alarm changes the outcome. And that is exactly the question that justifies the expense: stopping, raising blood pressure, releasing a retractor, removing instrumentation. The entire practical edifice of monitoring rests on a link no guideline has graded.

It detects poorly and rules out well

Monitoring has a very definite and rather counterintuitive diagnostic profile. A meta-analysis of 19 studies and 4,608 patients in cervical spine surgery found a pooled sensitivity of 56 per cent and specificity of 94⁷. Restricted to irreversible changes, sensitivity falls to 49 per cent and specificity rises to 98.

Translated: roughly four in ten postoperative deficits occur without the potentials having sounded the alarm. Conversely, when the potentials stay intact, the probability that the patient wakes up well is very high. It is a good instrument for reassurance and a mediocre one for warning.

Part of the problem is that "the alarm" is not standardised. A review of 68 studies in supratentorial surgery found that the amplitude drop considered significant ranges from 20 to 80 per cent depending on the paper, and that current thresholds in direct cortical stimulation vary between 3, 4 and 5 milliamps⁵. The position statement of the US monitoring society sets criteria — more than 50 per cent amplitude reduction in supratentorial, brainstem or facial nerve surgery, with disappearance of the response always a major criterion — but it is a position statement, not a guideline with evidence levels, and it supplies no sensitivity or specificity figures⁴.

There is an elegant example of how the moment of measurement decides the result. A study of intramedullary tumours compared 50 monitored patients with 50 matched historical controls⁸. At discharge, the difference in the functional scale did not reach significance. At follow-up of three months or more, it did. The authors attribute this to transient deficits in the monitored group that later recover. It is a real result; it is also a reminder that measuring at 48 hours and measuring at three months can give opposite answers.

Where the community itself withdrew

The most instructive case is a procedure where monitoring was routine and stopped being so. An analysis of 15,395 anterior cervical fusions found neurological injuries in 0.23 per cent of monitored cases and 0.27 per cent of unmonitored ones: no difference¹⁰. And use fell from 22.8 per cent in 2007 to 4.3 per cent in 2014.

It is an administrative database and the event rate is so low that statistical power is limited. But the direction of the finding matters: the surgical community abandoned the routine indication in a procedure where it did not see it paying off. That is exactly what should happen and almost never does.

Awake or asleep

The most eloquent form of monitoring is not putting the patient to sleep at all: operating awake and asking them to speak, count or move a hand while stimulation is applied. It is the technique that allows language mapping, which no electrode can measure in an anaesthetised patient.

Its failure rate has been measured: in a series of 424 awake craniotomies, 6.4 per cent failed, from lack of intraoperative communication in 4.2 and from seizures in 2.1 per cent¹². And failure has consequences: in those cases complete resection was 54 against 83 per cent, speech deterioration at three months 15.4 against 2.3, and major complications 14.8 against 4 per cent.

There is also a counterpoint worth knowing. A meta-analysis of 3,011 patients compared awake and asleep motor mapping¹¹. Extent of resection was practically identical: 92.2 against 92.5 per cent. Immediate and late deficits, no significant difference. The only advantage of the awake procedure was a lower rate of postoperative seizures. For motor mapping, awake craniotomy did not demonstrate superiority over general anaesthesia with monitoring. For language, the awake patient remains irreplaceable.

The arithmetic of money

The economic case for monitoring has a striking flaw. A simulation model estimated that monitoring saves 23,189 dollars per patient and remains cost-saving as long as the surgery's neurological complication rate exceeds 0.3 per cent¹³. It is an attractive result. But that model assumes a sensitivity of 94.3 per cent, nearly double the 56 per cent the empirical meta-analysis measured later. The whole economic argument depends on a parameter the evidence does not support.

The largest cost-effectiveness study, across more than 50,000 records, concludes that monitoring is cost-effective, with a ratio of 60,734 dollars per quality-adjusted life year¹⁴. And it states in its own abstract that the population was extracted from a multicentre database collected by a single national monitoring provider. The study concluding that the service is worth it was built with data supplied by the party selling the service.

What to take away

Neurophysiological monitoring is the only operating-room tool that measures function in real time, which is why it is the least replaceable. When the evoked potential disappears as a retractor is being placed, that information comes from no image, no navigation system and no algorithm.

At the same time, almost everything claimed about it rests on non-randomised comparisons between patients who were not comparable, and its real performance as a diagnostic test is modest: it rules out well and detects poorly.

Both things are true at once, and the honest synthesis is this: monitoring does not guarantee that the patient will wake up well, but it is the only thing that warns in time, while something can still be done. That is worth a great deal even if it is never demonstrated in a trial, and it should be defended with that argument rather than with figures the evidence does not support.