Article Text
Statistics from Altmetric.com
In his editorial on the Brain Trauma Foundation guidelines for the management of severe head injury, Kirkpatrick argues that standardisation of care is a prerequisite for the conduct of multicentre randomised trials.1 A similar concern seems to have motivated the European Brain Injury Consortium to develop its “expert opinion” based guidelines.2 This is not the case. Providing that a trial is large enough, randomisation will ensure that the intervention and control groups are identical with regard to known and unknown confounders. It is conceivable that the size of the intervention effect may vary a little depending on the other aspects of care given, but not the direction of the effect. Patients in the future will almost certainly receive different forms of care than they do today, and treatments shown to be effective today may be more or less effective in the future, but the direction of the effect will be the same. Rather than standardise care, it would be more useful to make sure that clinical trials were large enough to detect reliably moderate but clinically important treatment effects. Even though thousands of patients each year are treated with hyperventilation, barbiturates, mannitol, and steroids, clinical trials of these interventions, even in aggregate, have involved less than a few thousand patients, and for hyperventilation, mannitol, and barbiturates, existing trials comprise less than a few hundred patients. It is not surprising that the Brain Trauma Foundation was unable to define evidence based standards of care.
Kirkpatrick and Pickard reply:
We are grateful for the letter from Roberts which highlights the concept that randomised controlled trials can detect small treatment benefits provided such trials are large enough. Although correct in theory, this concept still assumes that the pathology treated shows some degree of homogeneity. It is now becoming clear that the number of known and unknown confounding variables with acute brain injury dramatically increases the number of patients required to prove efficacy. At some point this number exceeds that which is practicable and affordable, especially as the regulatory authorities demand that pharmaceutical phase III trials collect vast and expensive data sets. The randomised controlled trials which have shown treatment efficacy (for example, streptokinase for acute mycocardial infarction, aspirin and dipyridamole for the prevention of stroke and myocardial infarction, nimodipine for subarachnoid haemorrhage, and carotid endarterectomy for stroke prevention) have considered conditions with well defined end points sharing common pathological mechanisms. Head injury does not fall into this stereotype in view of the multiple interacting and opposing pathological mechanisms and the range of possible outcomes which are difficult to quantify.
Confounding variables influence outcome in particular conditions to different degrees. Attention to blood pressure, fluid balance, and systemic oxygenation may have only a modest influence on outcome after myocardial infarction, but are known to have a profound influence on the victim of brain injury. A specific treatment may have variable effects on different pathophysiological events operating within the same condition. Thus hyperventilating a patient with head injury with associated cerebral oligaemia is likely to be harmful, whereas this approach for raised intracranial pressure due to hyperaemia is more appropriate. By contrast the use of mannitol (which increases cerebral blood flow), may be beneficial for the first, but potentially harmful for the second. Hence the effect of a given intervention is not a simple matter of magnitude as suggested by Roberts, as the direction of the effect may differ in different subgroups.
Retrospective analysis of subgroups in phase III randomised control trials to correct for such imbalances have been received rightly with scepticism, but such an exercise can be used to direct the design of future trials to reduce the “noise” in the system. The suggestion that head injury subgroups be identified and a more stable physiological background be provided, is not dissimilar to that adopted for the randomised control trials of carotid endarterectomy, in which subpopulations were defined according to the grade of stenosis and the presenting symptom stereotype. If this had not been done, the efficacy would have been diluted to extinction.
The purpose of publishing standards of care in the head injury environment is largely educational, highlighting the lack of definitive evidence and the difficulty in obtaining it. None the less, ironing out key important variables will help to generate a substrate for testing intervention on defined subpopulations which share common pathology. Even if the standard proves to be harmful, delivery to the test population will provide a more stable pathophysiological baseline and a greater chance of showing individual therapeutic efficacy.