TXA. Huh. Good god y’all. What is it good for? Absolutely everything.
This appears to be the alternative hypothesis for the TICH-2 study 1, published last week in the Lancet. Off the back of recent evidence suggesting early IV tranexamic acid (TXA) can improve outcome in trauma 2, post partum haemorrhage 3, possibly traumatic brain injury 4 and maybe epistaxis 5, the investigators here decided to evaluate use of TXA in primary intracerebral haemorrhage. This is FOAMed of the highest order – the paper is from a high impact journal and open access to all at present. Core reading for emergency physicians and critical care folk alike. Like always we would strongly suggest you download, read, appraise and reflect on your practice in light of this new evidence.
Before we start our appraisal journey, perhaps we should just remind ourselves a little about primary intracerebral haemorrhage (ICH)6 . This awful condition is responsible for about 20% overall stroke diagnoses, yet disproportionately owns 50% of the mortality. It describes bleeding from a diseased blood vessel within the parenchyma of the brain itself, usually in the basal ganglia, cerebellum, cortex or brainstem. It occurs most often as a consequence of hypertension, leading to vessel wall rupture and bleeding. As such, other causes of intracranial bleeding such as subarachnoid haemorrhage, arteriovenous malformation, infection and trauma are NOT included in this definition. Important to remember.
What was TICH-2 all about then.
Following on from TICH-1 7, TICH-2 1 was a phase 3 trial looking to evaluate whether the use of early TXA in primary ICH could reduce death and/or dependence. When we say ‘early’ we mean <8h from onset. When we say ‘TXA’ we mean a standard 1g loading dose over 10mins, then a further 1g over 8h. Phases of trials have been discussed before, but here is a reminder 8.
The abstract is below, but as we always say, please go and read the full paper. It’s open access and so there is no excuse and you should always make up your own mind about evidence and how it relates to your practice.
Tell me about the methodology
Well, it was pretty (very) good overall. This was an international, double blind, placebo controlled parallel group randomised controlled trial. The authors put a lot of effort into attempting to reduce bias through independently adjudicated and blinded outcomes. They also attempted to negate confounding through robust and centralised randomisation protocols, stratified by centre and then further minimised by key prognostic factors (to ensure an even distribution of ICH severity across centres and countries). Double blinded TXA administered as an investigative medicinal product also lends itself well to this – it gives no early clues regarding administration (unlike ketamine for example), so it is very difficult for clinicians to ascertain the trial drug and adjust conscious or subconscious behaviour.
There are some interesting points to note – dependence as an outcome was reported as the modified Rankin scale 9 assessed by telephone follow up or postal questionnaire. This assumes an accuracy of data collection from remote follow up that would leave me with some concerns. Haematoma expansion was of key interest to the trial group also; to assess this, they included the baseline routine CT which I think would be pretty reliably delivered, but then aimed to compare this to further imaging after 24h of treatment. For patients with multiple CT images, they used the scan closest to 24h. A slight lack of standardisation here raises concern that patients could receive interim management that may influence outcome. What if a 6h CT showed an increase in haematoma size necessitating clot evacuation by neurosurgery. Then a 24h scan showed a small residual haematoma only. How would this be dealt with?
There was a prepublished statistical analysis plan, including prespecfied subgroups for secondary and safety outcomes. The trial protocol was previously published 10. The trial design included independent blinded adjudication of all relevant outcomes, a trial steering committee, a trial management committee and an independent data monitoring committee. This is about as transparent as it gets to my knowledge and should all be applauded.
What was the primary outcome?
The authors chose to use the modified Rankin Scale for the primary outcome as a measure of death or dependence. This is a common outcome for stroke trials and recognised internationally. However, they chose to report this as an ordinal odds ratio (OR) – it’s worth considering for a second what this actually means. Usually trials will select a dichotomised outcome (death or no death for example) and then report a static odds ratio of effect – for example, that TXA might halve the odds of death. An ordinal OR gives the change in odds for a unit increase in a continuous predictor. In this situation, it provides a reflection of how the odds change across the whole of the mRS when receiving TXA, compared to not receiving it. This is a reasonable way of doing things – indeed it arguably gives a more accurate reflection of effect across the board. However it assumes that we want to know about all degrees of severity, that the gradations in the scale are equivocal and that the scale in use is reliable and proportionate. To fend off this concern, the authors have also dichotomised the mRS in a sensitivity analysis using mRS 0-3 (a good outcome) vs. 4-6 (a bad outcome). This is helpful and perhaps more familiar to most of us. Remember though, you could set this line of good and bad anywhere. Would that give you different results?
They powered to look for an ordinal OR of 0.79 with an appropriate alpha and beta. This is quite the improvement, suggesting that the use of TXA would improve the odds of a better mRS score across the board by a fifth. The power calculator therefore suggested 2000 patients would provide enough confidence for a definitive answer. That sounds like a lot. But is it? It’s worthwhile comparing this to the previous TXA trials, which put this figure in perspective – CRASH 2 and the WOMAN trial recruited >20,000 each for example. CRASH 3 is aiming for >13,000 4. So quite different numbers really.
Who did they recruit?
Over 4 and a half years, they recruited 2325 participants from 124 sites in 12 countries. Sounds great, but we should note that >80% of these patients were recruited in the UK. This is good news for UK clinicians as it enhances the validity of the findings, but significantly reduces the international generalisabililty.
Median time to randomisation from stroke onset was 3.6h, although only just over a third were recruited within 3h. This is pertinent given recent developments with CRASH3 11. Mean and median haematoma volume were 24 and 14.1ml respectively, a decent difference implying some fairly big outliers and a nice example of why we need several methods of calculating the average.
Adherence was good as you would expect for a simple medication like TXA and loss to follow up was <1%.
And the results?
An intention to treat analysis was performed on valid data for 2307 patients. The first result of note is a powerful reminder of the severity of this condition; over half of each group, TXA or otherwise, were dead, significantly disabled or unable to walk/complete activities of daily living. This, within a trial context also where patients will arguably get the highest level of care available through standardised protocols, additional research expertise and specialised care.
And the primary outcome? The ordinal OR was 0.88 (95% CI 0.76 to 1.03, p=0.11) for a shift in the mRS. When dichotomised at sensitivity analysis, the adjusted OR for a ‘bad outcome’ was 0.82 (95% CI 0.65 to 1.03, p=0.08). So, no statistically significant difference in the odds of death or dependence at 3 months, whether you got TXA or placebo.
Now then. I don’t want to use the word ‘trend’ for a variety of reasons. And I also don’t want to imply a statistically significant result was found when it wasn’t. But let’s be clear about these results. The odds of a better outcome were improved by TXA in this patient group, but the margin of potential variation around this result means that we can’t be absolutely sure that TXA had an effect. The ordinal OR found was not as good as that proposed in the power calculation, and hence the trial lacked power to clarify significance for this result. Had they recruited 10,000 patients with similar findings, the authors may have been able to demonstrate a statistically significant improvement in outcome.
The margin of error around the OR is such that we can be pretty sure from this data that TXA does not worsen outcome. This is supported by the safety outcome results, which showed less predefined safety outcomes, predefined SAEs and thrombotic complications in the TXA group when compared to placebo.
The subgroups were interesting as always. The only reported significant interaction was in patients with a systolic BP <170mmHg at recruitment, who seemed to do better with TXA. Looking at the rest of the forest plot on figure 3 some familiar themes emerge though – trends towards improved outcome following TXA in younger patients, males, those receiving the intervention at <4.5h and patients with interventricular haemorrhage. None of these interactions achieved significance by odds ratio and should be taken as hypothesis generating predominately.
The most interesting prespecified analysis was of the effect on haematoma volume – a reduction in the proportion of patients with haematoma expansion was seen in the TXA group versus placebo at day 2 (25% vs 29% respectively). This was followed by quantitative analysis of haematoma volume (reported by a blinded independent radiologist) which suggested less of an increase from baseline to 24h in the TXA group (3.72ml [SD15.9]) compared to placebo (4.9ml [SD 16.0]). Does this difference float your boat? It’s a small absolute volume on average, and it doesn’t appear to correlate in this sample with improved neurological recovery or less dependence. However, it does provide some internal validity to the rationale that TXA may reduce haematoma expansion. Interesting stuff.
I see – no significant difference in the primary outcome. Is that the end of TXA in ICH then?
I don’t think so. Although this trial was powered appropriately, the effect size was overestimated and as such the conclusions are unfortunately not that clear. More research would address that, but will perhaps be challenging to perform given this result. I suspect that given the suggestion of potential improvement, champions for early TXA will question the need another huge and expensive trial? Given the lack of adverse events, the low costs and robust safety data of the drug I suspect many will say they now know enough about this drug and given the devastating nature of the condition, they would be happy to try it.
What about your practice?
Well that’s the kicker isn’t it. Which side do you fall down on? Would you champion the low costs and familiarity of this drug? This weak evidence of suggested benefit regarding haematoma volume and trend towards outcome improvement? Or would you highlight the opportunity costs 12 of giving, the lack of definitive evidence and the costs of any therapy (even if it’s cheap), for a lot of people.
Is there a middle ground? Looking at the subgroups there is not much to discriminate, other than a systolic BP <170 as mentioned previously.
I think on balance, if my mum had a primary ICH next week I would probably advocate for her to get TXA early. For the patients that I see clinically, I will carefully consider and discuss with my stroke service. I’d be conscious of the fact that for patients with other severe adverse prognostic features, there is likely to be little overall gain. But for those who are being actively and aggressively treated, a reduction in haematoma volume could be helpful provided the specialty team responsible for ongoing care have no objection. I certainly don’t think you can justify handing it out to everyone based on this evidence. But careful individualised assessment might be the way forward. And the downsides of administration appear to be clinically negligible, and mostly to do with cost, administration and time.
Any last points?
On the above note, it’s just worth remembering that evidence based medicine is of course supposed to be delivered in context. We are sometimes guilty of dichotomising evidence in the same way we like to dichotomise outcomes. Either it works or it doesn’t. This is rarely true.
It usually works only in certain cases, or it might work a bit but not all the time, or it doesn’t cause harm and it might work but we’re not 100% sure is a bit more like it.
Read the trial, think about EBM in context and ask yourself if you would be up for receiving TXA in this situation, if you were the patient. Then remember those thoughts when the next ICH comes through your door, and share them with your team, the patient where appropriate and their relatives.
Before you go please don’t forget to…
- Subscribe to the blog (look top right for the link)
- Subscribe to our PODCAST on iTunes
- Follow us on twitter @stemlyns
- See our best pics and photos on Instagram
- PLEASE Like us on Facebook
- Find out more about the St.Emlyn’s team
- Learn with us at the teachingCoOp course in Manchester 2018
- Come to our conference #StEmlynsLIVE 2018