Based on WHO mortality data and a systematic review of the literature we estimate that there are about 400,000 in-hospital deaths from bleeding each year worldwide. If all hospitalised bleeding trauma patients could be treated with TXA within an hour of injury then up to 128,000 of these premature deaths could be averted. If they could be treated within three hours of injury then up to 112,000 premature deaths could averted. Although there is considerable uncertainty in the estimates even the most conservative suggest that tens of thousands of deaths could be averted every year.
We found no compelling evidence that the effect of TXA on death due to bleeding varies by geographical region. Our conclusion is based on a statistical test of interaction which is considered to be the most appropriate way to evaluate subgroup effects . As recommended by methodologists, we pre-specified that unless there was strong evidence against the null hypothesis of homogeneity of effects (i.e. p < 0.001), that the overall risk ratio (RR) would be considered to be the most reliable guide to the approximate RRs in all regions. We found no statistical basis to reject the null hypothesis.
The data sources used to parameterise the model are subject to a number of limitations which may have affected our results. First, although the WHO database provides the best available country-level mortality data, poor coverage and coding of mortality registration systems may affect the accuracy of the number of trauma deaths for some countries. Second, our classification of trauma deaths into blunt or penetrating trauma based on the cause of death categories in the WHO data was somewhat arbitrary and would have resulted in some misclassification. However, in the absence of accurate country-specific data, we judged that this approach would provide the most reliable estimates. Third, due to the absence of country-specific data for the proportions of deaths occurring in hospital and the proportion of deaths caused by haemorrhage, we chose to apply average global estimates. We were therefore unable to incorporate between-country variations in these parameter estimates into our analysis. Nevertheless, our estimates were derived from a systematic review of the recent literature and data from the CRASH-2 trial, and thus represent the most accurate estimates available. We also performed sensitivity analyses to assess the impact of uncertainty around the parameter estimates.
Since many deaths from self-inflicted injuries are not usually associated with life-threatening haemorrhage (e.g. self-poisoning, hanging) we excluded this category to avoid over-estimating the number of deaths due to bleeding. However, this is likely to have led to the exclusion of some self-inflicted deaths that were associated with haemorrhage, in which case we may have underestimated the potential of TXA administration.
Our analysis was based on a number of assumptions. We have assumed that there was no use of TXA as a treatment for traumatic bleeding prior to publication of the CRASH-2 trial results. It is possible that a small proportion of the trauma deaths in our sample did receive TXA prior to their death, which may over-estimate the number of deaths averted. However, given that any such prior use of TXA would have been minimal it is unlikely to have greatly affected our overall estimates.
The objective of our analysis was to estimate the potential number of deaths that could be averted assuming TXA use under optimal conditions, that is, when administered appropriately and within three hours of injury, to all eligible bleeding trauma patients. It is unrealistic that such conditions are consistently and fully achieved in clinical practice. For example, the opportunity to treat some eligible patients will be missed and errors in the dose used or its administration may reduce the beneficial effect of TXA.
We assumed that the results of the CRASH-2 trial could be extrapolated to all hospitalised bleeding trauma patients. The CRASH-2 trial used clinical criteria to recruit a large number of patients from 274 hospitals in 40 countries, which helps the results to be generalised widely. Whilst we acknowledge that the underlying risk of death will vary in different settings, this does not necessarily imply that that the relative effect will vary. Indeed, relative effects are often remarkably homogeneous despite differences in underlying risk. This is supported by empirical evidence from a range of trials in which the relative effects are constant across variations in baseline risk . Furthermore, there is no reason to suppose that the mechanism of action of TXA would vary in different populations. However, we acknowledge that the appropriateness of such extrapolation is a matter of judgement.
A further assumption is that all trauma patients reached hospital in time to receive early treatment with TXA; that is either within one hour or within three hours of injury. Such a time frame is unlikely to be realistic in many settings where long distances and other logistical difficulties may delay arrival at hospital. For this reason we performed sensitivity analyses based on the relative effect of TXA from a more conservative estimate of time-to-treatment of within eight hours of injury, the results of which still suggest that up to 60,000 deaths could be averted. Besides, there is reason to predict that time between injury and treatment would be shorter in clinical practice than in the CRASH-2 trial as delays caused by consent procedures would be avoided .
In applying the RR of death due to bleeding in our primary analysis we assumed that all deaths in this group would be avoided. However, it is possible that whilst TXA may prevent death due to bleeding, some patients would die from other causes instead. If this is the case, then our primary analysis would over-estimate the number of death averted. To address this we performed a sensitivity analysis in which the effect of TXA on all-cause mortality was used. Even using this smaller relative reduction, up to 50,000 deaths could be averted.
We restricted our analysis to the potential benefit of in-hospital use of TXA. However, our parameter estimate of the proportion of in-hospital trauma deaths indicates that most trauma deaths occur before arrival at hospital. TXA is a practicable treatment suitable for use in a range of health-care settings, including pre-hospital. If TXA was used in the pre-hospital setting then many more premature deaths might be averted.