840 patients were required, so it was always going to be international (there are 36 neurosurgical centres in the UK). The trial was stopped early because only six patients were recruited from the UK. Seven UK centres took part; Dundee recruited 2, Newcastle 3 and the other was from Southampton. The authors describe in the paper the strategies used to increase recruitment, but there was clearly a problem. Many other countries were similar; USA – 3 centres no patients, Germany 3 centres 3 patients, Spain 2 centres 3 patients but there are clear exceptions. One Indian centre recruited 29 patients, one Chinese centre recruited 30. In fact, of the 170 patients recruited, 115 were from India or China.
So from the start we have an underpowered study with most patients coming from a different healthcare system to the UK. Also, 111 of the patients were recruited from centres where screening records were not kept.
In general terms, why might a trial struggle to recruit?
Do we exaggerate the decreased validity of international studies?
Why are screening logs important?
One interesting aspect of the method was in relation to the pragmatic timeframes and the consequences of intention to treat analysis. A patient could be randomised up to 48hrs post injury, to either surgery within the first 12 hours, or conservative mx unless deterioration. It is entirely possible therefore that a patient in the conservative mx group could be randomised at 1 hour post injury and have surgery in hour 5, and a patient in the intervention arm could be randomised 45 hours post injury, and have surgery in hour 55. Indeed, those patients in the conservative mx group who underwent surgery had a median time to operation of 45 hours, with an interquartile range overlapping that of the intervention arm by 10 hours.
Can you think of a way whereby intention to treat could still be used, but the analysis could reflect this issue – or would you not use intention to treat at all?
There was no difference in the primary outcome (6 month GOS) as expected – the study was not adequately powered.
Do you think papers should continue to quote odds ratios, p values and confidence intervals when the result is non-significant?
With regard 6 month mortality (a secondary outcome) there was a difference between the groups, with the expected wide confidence intervals given the small numbers. This appears to be largely due to the initial injury, with the difference largely occurring within the first 15 days. As I’ve previously written, death as an outcome measure after brain injury is something I struggle to fully come to terms with. A number of people dying of a head injury do so because active treatment is withdrawn, a decision that is multifactorial to say the least.
Are there any reasons why treatment would be more likely to be withdrawn in the conservative management group?
There are a couple of conclusions I’m not sure I agree with. The authors say there was a “non-significant benefit on GOS [associated with early surgery]”. Does that make sense? Is there such a thing as a non-significant benefit? They also report a result as almost statistically significant, and say that if the 840 patients had been recruited and the same trend had transpired, the result would have been statistically significant. I don’t dispute that, but to say after matching the first ball on the lottery that if the next 6 match I’ll be rich doesn’t make it so. My final piece of pedantry is the emphasis placed on patients with a GCS of 9-12. Figure 4 shows that for this group there is no demonstrable benefit of either strategy. Can you say that this no difference is less of a no difference than with the GCS 3-8 group or the 13-15 group?
As the authors point out, what is needed for these patients is a crystal ball – knowing who to operate on is far from straightforward, and is one of the reasons I couldn’t do that job.
Unsurprisingly the authors call for a larger trial. What I’d be really interested to read however is a study into why recruitment for this trial was so unsuccessful in the UK despite their best efforts and the implications for other healthcare RCTs – to understand that may well give benefit to far more patients in the end.
I’d be really interested to read your comments about anything to do with this paper, but particularly in answer to the questions I’ve suggested…..