Development and validation of a prognostic model to predict death in patients with traumatic bleeding, and evaluation of the effect of tranexamic acid on mortality according to baseline risk: a secondary analysis of a randomised controlled trial

Article history paragraph text

The research reported in this issue of the journal was funded by the HTA programme as project number 09/22/165. The contractual start date was in October 2010. The draft report began editorial review in July 2012 and was accepted for publication in November 2012. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors' report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

© Queen's Printer and Controller of HMSO 2013. This work was produced by Perel et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Model development

For the development of the prognostic model, we involved potential users from three settings: pre-hospital, battlefield and emergency departments. We held separate meetings with paramedics, military doctors and emergency medicine consultants to identify variables and interactions that they considered important and convenient for their settings and to obtain information on how to present the prognostic model in a user-friendly format. In each of these meetings we conducted a semistructured interview following a prespecified questionnaire (see Appendix 1).

Simple prognostic model

For ease of use at point of care, we developed a simple prognostic model. For this model we included the strongest predictors with the same quadratic and cubic terms as used in the full model, adjusting for TXA. The prognostic model was presented as a chart that cross-tabulates these predictors, but each of them recoded in a number of categories. The categories were made considering clinical and statistical criteria. In each cell of the chart we estimated the risk for an individual with values of each predictor at the mid-point of the predictor's range for that cell. We then coloured the chart cells in four groups according to ranges in the probability of death: < 6%, 6–20%, 21–50% and > 50%. We decided these cut-offs based on the feedback from the potential simple prognostic model users and by looking at previous publications. 17,18

Ethical approval

Ethical approval for this study and the use of the CRASH-2 trial data was obtained from the London School of Hygiene and Tropical Medicine, London, UK. TARN already has ethical approval (PIAG section 60; PIAG3-4(e)/2006) for research on the anonymised data that are stored securely on the University of Manchester server.

General characteristics

Tables 1 and 2 show the characteristics of the patients in the CRASH-2 trial and TARN data sets respectively.

In the CRASH-2 trial, the majority of patients were male and the median age was 30 years. Patients who died were on average older, had lower SBP, higher HR and lower GCS score. There were few missing data for all of the variables. There were 3076 deaths (15%) among the 20,127 CRASH-2 trial patients, and 1765 deaths (12%) in the 14,220 TARN patients. In comparison, the patients from TARN were older (median age 39 years), they had fewer penetrating injuries and they had a higher SBP.

Relationship between predictors and overall mortality

Age showed a positive and increasing association with risk of death. SBP, HR and RR showed a U-shape relationship, and GCS score showed a negative association with risk of death (Figure 1).

FIGURE 1.

Association between continuous predictors and death among CRASH-2 trial patients. The error bars represent the 95% CI of that observed unadjusted risk in that level of the biomarkers.

Table 3 shows that in the CRASH-2 trial, age of patient was positively associated with mortality for each of the described causes of death, with the highest relative increase for vascular occlusive death.

Prognostic model

We included quadratic or cubic transformations in the prediction model to accommodate the departures from linearity. In the multivariable analysis, GCS score, SBP and age were the three strongest predictors. HR, RR and hours since injury (HSI) were associated with mortality and also included in the final model. All of these variables were considered important by users. Patients in LICs and MICs were more likely to die than patients in HICs. Although capillary refill time was weakly associated with mortality, it was not included in the prognostic model because in situations with poor visibility, such as on the battlefield, it is difficult to measure. In addition, capillary refill time was missing in > 80% of the TARN patients. There was some evidence for statistical interaction between GCS score and type of injury. Low GCS scores were associated with worse prognosis for blunt injuries (Table 4).

Validation

The model showed a good internal validity, with a C-statistic of 0.84 (Figure 2) and good calibration except in very high-risk patients for whom the model over-predicted risk (Figure 3).

FIGURE 2.

Internal and external discrimination displayed by the receiving operating characteristic. AUC, area under the curve; PV, predictive value. (a) CRASH-2 trial data (internal discrimination); (b) TARN data (external discrimination). Curves showing point where the addition of sensitivity and specificity is maximum and corresponding cut-points of risk.

FIGURE 3.

Goodness of fit and external calibration of the prognostic model by levels of predicted risk. (a) CRASH-2 trial data (goodness of fit); (b) TARN data (external validation).

Internal validation using bootstrapping showed a minimal decrease of the C-statistic from 0.836 to 0.835. This indicates that there is very low over-optimism in the model development.

For the external validation we used the same variables as included in the derivation model except HSI, as this variable had a very large number of patients with missing data. Discrimination was good (C-statistic 0.88) and calibration satisfactory (see Figures 2 and 3).

Model presentation

The prognostic model is available online (www.crash2.lshtm.ac.uk), so risk of death can be obtained for individual patients. By entering the values of the predictors, the expected risk of death at 28 days is displayed. For example, a 70-year-old patient from a LIC, with a GCS score of 14, SBP of 100 mmHg, HR of 110 beats per minute and a RR of 35 breaths per minute, has a 32% probability of death at 28 days (Figure 4).

FIGURE 4.

Snapshot of the web-based prognostic model.

Users also highlighted the importance of a simple prognostic model that could be used at the bedside. The simple prognostic model included the three strongest prognostic variables (GCS score, SBP and age). We developed different prognostic models for patients within 3 hours of injury in LICs, MICs and HICs, and presented them as charts (Figure 5).

FIGURE 5.

Chart to predict death in trauma bleeding patients.

These simple charts also showed good internal and external calibration (Figure 6). The area under the curve (AUC) for the simpler model was 0.82 for CRASH-2 trial data and 0.86 for TARN data.

FIGURE 6.

Internal and external calibration of the simple chart.

Effect of tranexamic acid according to baseline risk

Table 5 shows the characteristics of CRASH-2 trial patients who were treated within 3 hours of injury according to baseline risk. Patients with higher baseline risk of death were older, had lower SBP and GCS scores, higher HRs and RRs, and were more likely to have blunt trauma than patients with a lower risk. Most patients (97%) were from MICs and this frequency distribution was similar for all of the risk strata.

Figure 7 shows the effect of TXA on all-cause mortality by baseline risk. TXA reduced all-cause mortality in each stratum of baseline risk. There was no evidence of heterogeneity in the effect of TXA on all-cause mortality (p-value for interaction = 0.96).

FIGURE 7.

Death from all causes.

Figures 8 and 9 show the effects of TXA on death due to bleeding, and on fatal and non-fatal thrombotic events by baseline risk of death.

FIGURE 8.

Death due to bleeding.

FIGURE 9.

Fatal and non-fatal thrombotic events.

There was no evidence of heterogeneity in the effect of TXA on death due to bleeding (p = 0.98) or on thrombotic events (p = 0.74). There was a reduction in non-bleeding deaths that was not statistically significant (OR = 0.97; 95% CI 0.86 to 1.09), with no evidence of heterogeneity according to baseline risk (p = 0.99).

Table 6 shows the effect of TXA on fatal and non-fatal thrombotic events in patients treated within 3 hours of injury. There was a significant reduction in the risk of fatal and non-fatal thrombotic events with TXA (OR = 0.69; 95% CI 0.53 to 0.89; p = 0.005).

There was a statistically significant reduction in arterial thrombotic events with TXA (OR = 0.58; 95% CI 0.40 to 0.83; p = 0.003). There was no evidence of heterogeneity in the effect of TXA on arterial thrombotic events (p = 0.91). There was a reduction in the risk of venous thrombotic events with TXA that was not statistically significant (OR = 0.83; 95% CI 0.59 to 1.17; p = 0.295). There was no evidence of heterogeneity in the effect of TXA on venous thrombotic events by baseline risk (p = 0.85).

Estimation of lives saved if tranexamic acid given to different risk strata

Assuming that the effect of TXA is the same in all risk strata (constant OR), the percentage of premature deaths that could be averted in TARN by administering TXA within 3 hours of injury in each risk stratum (< 6%, 6–20%, 21–50% and > 50%) is 17%, 36%, 30% and 17% respectively. Figure 10 shows the distribution of patients according to baseline risk in TARN, and Figure 11 shows the absolute number of deaths that could potentially be averted through the use of TXA in patients in the UK TARN by baseline risk of death.

FIGURE 10.

Trauma Audit and Research Network: distribution of patients by baseline risk.

FIGURE 11.

Trauma Audit and Research Network: distribution of lives saved if TXA was given to everyone (with constant OR −0.85).

Main findings

We have developed and validated a prognostic model for trauma patients with clinical parameters that are easy to measure. The model is available as a web calculator and can be used at the point of care in its simplified form. Separate models are available for patients from LICs, MICs and HICs. This simple prognostic model could inform doctors about the risk of death and guide them in the early assessment and management of trauma patients.

The ORs for all-cause mortality and death due to bleeding with TXA do not appear to vary by baseline risk. TXA reduced the odds of death due to bleeding by approximately 30% in each of the risk strata. TXA also appeared to reduce the odds of thrombotic events by 30%. Once again, this reduction does not appear to vary by baseline risk. Taken together, these data suggest that TXA can be administered safely to a wide spectrum of trauma patients and that its use should not be restricted to the most severely injured. However, although we did not find evidence of heterogeneity for the effect of TXA according to baseline risk, in the lowest-risk group the precision of the estimated effect is low.

Strengths and limitations

Our study has several strengths. Our models were based on a prospective cohort of bleeding trauma patients, with standardised collection of data on prognostic factors, very few missing data and low loss to follow-up. Unlike previous prognostic models, we explored more complex relationships between continuous predictors and mortality, and captured non–linear relationships. All of these factors provide reassurance about the internal validity of our models. The large sample size in relation to the number of prognostic variables is also an important strength. As most previous models were derived from single-centre studies in HICs, we developed separate models for LICs, MICs and HICs. Unlike most previous models, we conducted an external validation in a large cohort of trauma patients. This confirmed the discriminatory ability of the model in patients from HICs and showed good calibration. Another methodological strength was our use of imputation to address missing data, which has rarely been done in previous model validation studies. To the best of our knowledge this is the only prognostic model for this population that is available in a web-based calculator and a simplified chart that can be used at point of care. Importantly, we obtained advice from the potential users of the model throughout its development.

There are also some limitations. The data from which the models were developed come from a clinical trial and this could therefore limit external validity. For example, patients were recruited within 8 hours of injury, and we cannot estimate the accuracy of the models for patients evaluated beyond this time. Nevertheless, the CRASH-2 trial was a pragmatic trial that did not require any additional tests and, therefore, included a diversity of real-life patients. In addition, the relationship between predictors and outcome could be different in patients included in a clinical trial and in routine practice. The TARN data included a larger proportion of patients with blunt injuries than in the CRASH-2 trial; however, that the model's performance was very good in a trauma registry population provides reassurance that any potential bias (if present) was small. Another limitation was that for the validation we used a cohort of trauma patients that were not equally defined and we included them using an estimation of the blood loss. In any case, this weakness would have underestimated the accuracy of the model. Other potentially important variables such as pre-existing medical conditions, previous medications and laboratory measurement were not collected in the CRASH-2 trial and, therefore, were not available for inclusion into the model. However, these are variables that are usually unavailable in the acute care trauma setting in which the model is intended to be used. The prognostic model predicts overall death rather than death due to bleeding, as death due bleeding was not available in the TARN data set. However, it is expected that bleeding contributes to the other main causes of death in trauma patients. In addition, some deaths classified as non-bleeding could in fact be due to bleeding.

Regarding the effect of TXA according to baseline risk, it is important to note that the data examined here are a subset of the entire CRASH-2 trial. As most trauma protocols restrict TXA use to patients within 3 hours after injury, we restricted our analyses to these patients. Given that our results are based on a subgroup analysis, they should be interpreted cautiously. On the other hand, there was also a reduction in the risk of fatal and non-fatal myocardial infarction (relative risk = 0.64; 95% CI 0.42 to 0.97; p = 0.035) in the main trial analysis of patients treated up to 8 hours from injury, and there were fewer thrombotic events in TXA-treated patients. Although there may be grounds for scepticism about the reduction in arterial thrombotic events with TXA, the data provide reassurance that there is no increased risk.

The absence of evidence of an interaction between the effect of TXA and baseline risk could be due to a lack of statistical power for the interaction tests; however, graphically it seems unlikely that the findings are due to lack of power and they suggest that the effect of TXA is not influenced by baseline risk. We used a multivariable approach that has been shown to improve the power of subgroup analysis in comparison with individual predictor subgroup analyses. 19 Simulation studies have shown that the power to detect an interaction is increased from 23% to 83% when a multivariable approach is used. Nevertheless, the absence of evidence of interaction should not be taken as evidence of absence of an interaction.

That we analysed data from a randomised controlled trial using a prognostic model derived from the same trial might be considered a weakness. On the other hand, the prognostic model we used has good external calibration and discrimination, and there are few models available that can be reliably applied to the trauma population included in the CRASH-2 trial. The revised trauma score, one of the most widely used scores, was developed many years ago, has several methodological limitations, and showed poor calibration in the CRASH-2 trial and TARN data sets (Figure 12). 20

FIGURE 12.

Calibration of the revised trauma score in CRASH-2 trial and TARN data. (a) CRASH-2 trial data (prediction with revised trauma score); (b) TARN data (prediction with revised trauma score).

Establishing cause of death in trauma patients can be difficult and any inaccuracy might have affected our estimate of the effect of TXA on fatal thrombotic events. There may also have been inaccuracy in the diagnosis of non-fatal events. The diagnosis of myocardial infarction is particularly challenging in trauma patients, many of whom are anaesthetised or sedated. Elevations in creatine kinase-MB isoenzyme are hard to interpret because of muscle injury and we did not collect data on troponin. However, diagnostic inaccuracy tends to obscure treatment effects and would not readily explain the observed reduction in thrombotic events with TXA.

Implications of the study

Many trauma protocols use blood pressure as the main criterion for determining which patients should receive urgent intervention. However, according to our model, a 75-year-old patient with blunt trauma and a SBP of 110 mmHg, HR of 80 beats per minute, RR of 15 breaths per minute and a GCS score of 15 has a similar risk of death as a 45-year-old patient with exactly the same parameters but with a SBP of 60 mmHg. These findings have important practical implications. According to many trauma protocols, only the younger patient would receive urgent interventions such as TXA, whereas the older patient would be denied this life-saving intervention. The effect of age is particularly important bearing in mind that in HICs the average age of trauma patients is increasing. Data from TARN show that one-quarter of trauma deaths in England and Wales occur in patients > 70 years. The effect of age is likely to reflect the increased incidence of co-existing diseases, in particular cardiovascular diseases. Elderly patients are more likely to have coronary heart disease and the decrease in oxygen supply associated with traumatic bleeding can increase the risk of myocardial ischaemia. 21 Another potential explanation for the increased risk of vascular occlusive death is related to the inflammation process trigger after trauma. It is known that after trauma there is a potent inflammatory response that involves increased serum levels of interleukin 1 (IL-1), IL-2, tumour necrosis factor alpha (TNF-α), IL-6, IL-12 and interferon-gamma. 22 In patients with traumatic bleeding there is activation of plasmin which plays a key role in the fibrinolytic response in the early hours after injury. Plasmin also has pro-inflammatory effects through the activation of cytokines, monocytes, neutrophils, platelets and endothelial cells. 23 It has been suggested that vascular risk rises in short time periods of inflammatory responses to exposures such as infections or major surgery. 24 It is possible that some of the observed prognostic role of age in trauma patients is due to the acute trauma inflammatory response, which might trigger acute vascular events, particularly in those older patients who have a more widespread atherosclerotic condition. Furthermore, the prognostic role of age could be explained partially by a self-fulfilling prophecy phenomenon, as age has been shown to be positively associated with do-not-resuscitate orders. 25

We found that trauma patients in LICs and MICs were at higher risk of death than patients from HICs. It is important to emphasise that the income classification refers to the country and not to individual patients. Some of the effect of income classification might be the consequence of the differences on health-care settings. Other studies have shown similar results, but to our knowledge this is the first one to include a large number of LICs and MICs. 26 Although we did not have enough information to explore the causes of these differences, it is likely that the rapid increase in the number of trauma patients and the lack of resources in poorer countries are among the most important reasons. Scaling up cost-effective interventions in these settings could save hundreds of thousands of lives every year.

Potential biological mechanisms

The effect of TXA in arterial events was unexpected. Treatments to reduce haemorrhage often increase thrombosis risk and the use of anticoagulants tends to increase the risk of bleeding. However, TXA reduces the risk of haemorrhage death and appears to also reduce the risk of arterial thrombotic events. What mechanisms might account for this? Trauma marks the onset of a period of greatly increased myocardial oxygen demand as pain and hypovolaemia lead to a rapidly increased HR. At the same time, blood loss results in hypotension and anaemia, which reduces myocardial oxygen delivery. The imbalance between increased oxygen demand and reduced supply may be sufficient to cause myocardial infarction, particularly in patients with pre-existing coronary artery stenosis. Cohort studies have shown that even a modest reduction in haemoglobin is a strong risk factor for death in patients with stable angina. In this respect, the pathophysiology of myocardial ischaemia in trauma patients may be similar to stress-induced myocardial infarction in perioperative patients (type 2 perioperative myocardial infarction). 27 By reducing traumatic blood loss and anaemia, an early short course of TXA could reduce myocardial oxygen demand and increase oxygen supply.

Tranexamic acid might also play a role in the prevention of acute coronary thrombosis following trauma. 28 Traumatic injury triggers an acute inflammatory state with increases in TNF-α, IL-1 and IL-6. These inflammatory mediators have been implicated in coronary plaque fissuring and acute coronary thrombosis. TXA reduces bleeding through the inhibition of plasmin, the enzyme responsible for fibrinolysis. However, plasmin also has a wide spectrum of pro-inflammatory effects, raising the possibility that the observed reduction in the risk of arterial thrombotic events in trauma patients treated with TXA may be mediated via an anti-inflammatory effect. 29

Implications for clinical practice

The result that the effect of TXA does not vary by baseline risk has important implications for the care of trauma patients. First, it suggests that TXA can be used safely in all trauma patients with or at risk of significant bleeding, as per the inclusion criteria used in the CRASH-2 trial, and not just in those with massive haemorrhage. Second, the observed reduction in the risk of arterial events with TXA suggests that the absolute benefits from TXA administration are likely to be greatest in older trauma patients, who at any given level of injury severity have a higher baseline risk of haemorrhage death and thrombotic events. Clinical concern about an increased risk of ischaemic cardiac events may be a reason to give rather than to withhold TXA. It is worth noting that trials of TXA in cardiac surgery patients, a group at high risk of cardiac events, provide no evidence of any increased risk with TXA. 30 We acknowledge that estimating the risk of significant bleeding is a challenging ongoing process that uses not only physiological variables but also other variables such as laboratory measurements and response to treatments. Physicians will use all of this information and their clinical judgement when deciding whether or not to use TXA. However, in situations of uncertainty physicians can use the prognostic model to support the decision-making process and should certainly consider administering TXA to patients with a risk of death > 5%.

Future research

The relationship between age and mortality needs further exploration; a better understanding of the mechanism underlying the association between age and increasing mortality could lead to effective interventions to improve the outcome in this vulnerable population.

As we were able to validate the model only in patients from high-income regions, future studies should also explore its performance in LIC and MIC settings.

Finally, future research should evaluate whether or not the use of this prognostic model in clinical practice has an impact on the management and outcomes of trauma patients.

Contribution of authors

Pablo Perel (Senior Lecturer), Haleema Shakur (Senior Lecturer) and Ian Roberts (Professor) designed the study.

David Prieto-Merino (Lecturer) analysed the data.

All authors contributed to writing the paper.

Disclaimer

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.