Viscoelastic point-of-care testing to assist with the diagnosis, management and monitoring of haemostasis: a systematic review and cost-effectiveness analysis

ROTEM delta point-of-care analyser

The ROTEM^® Delta (TEM International GmbH, Munich, Germany; www.rotem.de) is a point-of-care (POC) analyser, which uses thromboelastometry, a VE method, to test for haemostasis in whole blood. It was previously known as rotational TEG or ROTEG. 6 It is performed near the patient during surgery or when admitted following trauma. It is used to assist with the diagnosis, management and monitoring of haemostasis disorders, during and after surgery, which are associated with high blood loss. It is an integrated all-in-one system and analyses the coagulation status of a blood sample to differentiate between surgical bleeding and a haemostasis disorder. 17 It uses a combination of five assays to characterise the coagulation profile of a citrated whole blood sample (Table 2). Initial screening is performed using the INTEM and EXTEM assays; if these are normal then it is an indication that surgical bleeding rather than coagulopathy is present. The use of different assays allows for rapid differential diagnosis between different haemostasis defects and anticoagulant drug effects. 17 Training in how to use the technology is required but specialist laboratory staff are not needed.

Figure 2 shows the ROTEM system. A 340-µl blood sample, which has been anticoagulated with citrate, is placed into the disposable cuvette (sample cup) (7), using an electronic pipette. A disposable sensor pin (6) is attached to the shaft, which is connected with a thin spring (2), and slowly oscillates back and forth (1), suspended in the blood sample. The signal from the pin is transmitted via an optical detector system (3–5). The test is started by adding the reagents described above. Although the typical test temperature is 37 °C, different temperatures can be selected, for example for patients with hypothermia. Although the blood remains liquid the movement is unrestricted, as the blood starts clotting, the clot restricts the rotation of the pin with increasing resistance as the firmness of the clot increases. This is measured by the ROTEM system and translated to the output, which consist of graphical displays and numerical parameters.

FIGURE 2.

ROTEM system. 18 1, Oscillating axis; 2, counterforce spring; 3, light beam from LED; 4, mirror; 5, detector (electronic camera); 6, sensor pin; 7, cuvette with blood sample; 8, fibrin strands and platelet aggregates; 9, heated cuvette holder; 10, ball bearing; and 11, data processing unit. Reproduced with permission from ROTEM^®.

The graphical output of results produced by the ROTEM system is shown in Figure 3. A separate graphical display is produced for each reagent by an integrated computer. Numerical values for each of the following are also calculated and presented below the graph. Initial results are available within 5–10 minutes and full qualitative results are available in 20 minutes:

Clotting time (CT) – time from adding the start reagent until the blood starts to clot. A prolonged CT indicates abnormal clot formation.

Clot formation time (CFT) – time from CT until a clot firmness of 20-mm point has been reached and an α-angle, which is the angle of tangent between 2 and the curve. These measures indicate the speed at which the clot is forming and are mainly influenced by platelet function but are also affected by FIB and coagulation factors.

Amplitude 10 minutes after CT (A10) – used to predict maximum clot firmness (MCF) at an earlier stage and so allows earlier therapeutic decisions.

Maximum clot firmness – the greatest vertical amplitude of the trace. A low MCF value suggests decreased platelet numbers or function, decreased FIB levels of fibrin polymerisation disorders or low factor XIII activity.

Maximum lysis (ML) – fibrinolysis is detected by ML of > 15% or by better clot formation in APTEM compared with EXTEM.

FIGURE 3.

ROTEM: analysis and interpretation of results. 19 Reproduced with permission from ROTEM^®.

Thromboelastography

The ROTEM system is a variant of the traditional thromboelastography (TEG) method developed by Hartert in 1948. 20 The two techniques are very similar, and other recent reviews have evaluated them as a single intervention class. 12,21,22 Like ROTEM, TEG is a VE method and provides a graphical representation of the clotting process. TEG is used in the TEG^® 5000 analyser (Haemonetics Corporation, Niles, IL, USA; www.haemonetics.com). The rate of fibrin polymerisation and the overall clot strength is assessed. 1 Like ROTEM, TEG is able to provide an analysis of platelet function, coagulation proteases and inhibitors, and the fibrinolytic system within 30 minutes, or within 15 minutes if the rapid assay is used (Figure 4). The nomenclature used in TEG differs from that used in ROTEM; differences are summarised in Table 3. The practical differences between TEG and ROTEM are that TEG uses a torsion wire rather than the optical detector used in ROTEM to measure the clot formation, and, although the movement in ROTEM is initiated with the pin, with TEG it is initiated from the cuvette. 1 The assays used in TEG also differ (see Table 3). 24,25 The platelet mapping function means that TEG is able to measure platelet function, which cannot be assessed using ROTEM. Sample size requirements do not differ substantially between TEG and ROTEM; TEG uses a 360-µl blood sample compared with the 340-µl sample used in ROTEM. 25

FIGURE 4.

Thromboelastography: analysis and interpretation of results. 23 ACT, activated clotting time; EPL, estimated per cent lysis; LY30, lysis at 30 minutes; MA, maximum altitude; R, clotting time. Reproduced with permission from TEG^®.

Sonoclot coagulation and platelet function analyser

Another method that uses viscoelastometry to measure coagulation is the Sonoclot^® coagulation and platelet function analyser (Sienco Inc., Arvada, CO, USA). This analyser was first introduced in 1975 by von Kualla et al. 26 It provides information on the haemostasis process, including coagulation, fibrin gel formation, fibrinolysis, and, like TEG, is also able to assess platelet function. The Sonoclot process is similar to ROTEM and TEG; although Sonoclot is able to use either a whole blood or plasma sample, citrated blood samples can be used but are not required. 27 A hollow, open-ended disposable plastic probe is mounted on the transducer head. The test sample (blood or plasma) is added to the cuvette containing the reagents. A similar volume to ROTEM and TEG is used: 330–360 µl. As with ROTEM, it is the probe that moves within the sample; however, rather than moving horizontally the probe moves up and down along the vertical axis. As the sample starts to clot, changes in impedance to movement are measured. Like TEG and ROTEM, Sonoclot produces a qualitative graphical display of the clotting process and also produces quantitative results of activated clotting time (ACT), the clot rate and the platelet function (Figure 5 and Table 4). 24 However, the measure of ACT produced by Sonoclot reflects initial fibrin formation whereas the equivalent measures produced by TEG and ROTEM reflects a more developed and later stage of initial clot formation. 24 Most information on clot formation is available after 15 minutes. If details on platelet function are required this may take up to 20–30 minutes. 27

FIGURE 5.

Sonoclot analysis and interpretation of results.

Comparison of viscoelastic testing devices

This report refers to the three technologies – ROTEM, TEG and Sonoclot – as a class as ‘viscoelastic testing POC coagulation testing devices’ or ‘VE devices’; however, data from each device are analysed separately. Table 5 provides an overview of the different terms used by each device to refer to the different test outputs. This table also summarises the factors affecting clot formation at each stage and the different therapeutic options.

Current care pathway

The exact care pathway and use of SLTs before, during and after surgery will vary according to the specific type of surgery. Some centres routinely screen all patients pre operatively for coagulation disorders using SLTs, such as the PT and aPTT tests. 33 However, UK guidelines published in 200834 do not recommend routine coagulation tests to predict perioperative bleeding risk in unselected patients before surgery. Instead, pre-operative testing should only be considered in patients at risk of a bleeding disorder, for example those with liver disease, family history of inherited bleeding disorder, sepsis, diffuse intravascular coagulation, pre-eclampsia, cholestasis and those at risk of vitamin K deficiency. 33

It is generally recommended that patients stop taking anticoagulant medications (clopidogrel warfarin and aspirin) a number of days before surgery to reduce the risk of bleeding during surgery. 10,35 In the event of emergency surgery, this may not be possible, in which case coagulation testing should be performed. 33 If the surgery involves CPB then heparin may be administered prophylactically to reduce the risk of clotting while on CPB. 35 It is essential to monitor heparin anticoagulation if this has been administered. An initial ACT test should be performed after the first surgical incision and be repeated at regular intervals during surgery. 36 Standard coagulation tests (PLT, FIB concentration, PT, aPTT) are most commonly used to assess the coagulation status of patients who are experiencing high blood loss during surgery. However, these generally take too long to give a result that can inform treatment decisions. Instead, decisions on how to treat the bleed have to be based largely on clinical judgement. The same tests are used after surgery to monitor coagulation status.

If bleeding occurs, surgical intervention may be needed or packed RBCs are transfused if required. This is generally to maintain a haemoglobin concentration of > 6 g/dl during CPB and 8 g/dl after CPB or according to other requirements as indicated by national guidelines. Other therapeutic options depending on laboratory test results include FIB concentrate (bleeding patients with abnormal FIB), fresh frozen plasma (FFP) (if after transfusion of packed erythrocytes new laboratory results were not available and/or bleeding did not stop after FIB administration), prothrombin complex concentrate (PCC) (abnormal aPTT), antithrombin concentrate (when ACT analyses not controlled by heparin alone), desmopressin (suspected platelet dysfunction), platelet concentrates (low PLT),35 cryoprecipitate, antifibrinolytic drugs and tranexamic acid. If bleeding continues despite these treatments then additional treatment options include factor XIII concentrate and activated recombinant factor VII or factor VIIa, although not licensed for use in the UK. 10,35 Heparin dose adjustments may be made to try and control the bleeding.

Role of viscoelastic testing in the care pathway

Viscoelastic testing can be repeatedly performed during and after surgery and so can provide a dynamic picture of the coagulation process during these periods. The role of VE testing in the care pathway is unclear. It could be used either as an add-on test, in which case it would be performed as well as SLTs, or it could be as replacement test in which case SLTs would no longer be needed.

If VE testing does not prevent the need for SLTs and provides complementary findings then it should be performed in addition to any laboratory coagulation tests already recommended for specific populations. However, if the SLTs do not offer any supplementary information to that provided by VE testing then there should no longer be a need for standard tests and VE testing should replace some or all of the SLTs. VE tests offer two key potential benefits over SLTs: the shorter timescale in which they are able to provide results and the additional information on the clotting process that they offer compared with standard tests. It is hypothesised that by providing additional information and quicker results requirements for blood components/products could be targeted and so the patient is not subjected to risks associated with unnecessary transfusion. Time in theatre, resource use, length of stay (LoS) in a critical care unit, length of hospital stay, blood component/product usage and the associated costs may therefore be reduced.

Search strategy

Search strategies were based on index test (ROTEM delta, TEG and Sonoclot), as recommended in the CRD guidance for undertaking reviews in health care37 and the Cochrane Handbook for Diagnostic Test Accuracy Reviews. 39

Candidate search terms were identified from target references, browsing database thesauri [e.g. MEDLINE (MeSH) and EMBASE (EMTREE)], existing reviews identified during the rapid appraisal process and initial scoping searches. These scoping searches were used to generate test sets of target references, which informed text mining analysis of high-frequency subject-indexing terms using EndNote X4 reference management software (Thomson Reuters, CA, USA). Strategy development involved an iterative approach testing candidate text and indexing terms across a sample of bibliographic databases and aimed to reach a satisfactory balance of sensitivity and specificity.

Search strategies were developed specifically for each database, and the keywords associated with ROTEM, TEG, thromboelastometry and Sonoclot were adapted according to the configuration of each database.

Inclusion and exclusion criteria

Inclusion criteria for each of the three clinical review questions are summarised in Table 6. Studies that fulfilled these criteria were eligible for inclusion in the review.

Inclusion screening and data extraction

Two reviewers (MW and PW) independently screened the titles and abstracts of all reports identified by searches, and any discrepancies were discussed and resolved by consensus. Full copies of all studies that were deemed to be potentially relevant were obtained and the same two reviewers independently assessed these for inclusion; any disagreements were resolved by consensus. Details of studies excluded at the full-paper screening stage are presented in Appendix 4.

Studies cited in materials provided by the manufacturers of ROTEM, TEG or Sonoclot were first checked against the project reference database, in EndNote X4; any studies not already identified by our searches were screened for inclusion following the process described above.

Data were extracted on the following: participant characteristics; study design; inclusion and exclusion criteria; details of VE test and/or test parameters evaluated; details of SLTs, where applicable; details of outcomes assessed [main outcomes were bleeding outcomes, transfusion outcomes, hospital/intensive care unit (ICU) stay, re-operation and mortality]; results. Data were extracted by one reviewer, using a piloted, standard data extraction form and checked by a second (MW and PW); any disagreements were resolved by consensus. Full data extraction tables are provided in Appendix 2.

Quality assessment

The methodological quality of included RCTs was assessed using the Cochrane Risk of Bias Tool. 44 Prediction studies were assessed for methodological quality using QUADAS-2. 45 Risk of bias assessments were undertaken by one reviewer and checked by a second reviewer (MW and PW), and any disagreements were resolved by consensus.

The results of the risk of bias assessments are summarised and presented in tables and graphs in the results of the systematic review, and are presented in full, by study, in Appendix 3.

Methods of analysis/synthesis

We provided a narrative synthesis involving the use of text and tables to summarise data to show differences in study designs, population, VE device and potential sources of bias for each of the studies being reviewed. Studies were organised by research question addressed (study population), outcome and VE device. Where possible, meta-analysis was used to derive summary effect estimates. All meta-analyses were performed using the MetaExcel (Epigear International) add on for Microsoft Excel version 1997–2003 (Microsoft Corporation, Redmond, WA, USA).

How do clinical outcomes differ among patients who are tested with viscoelastic devices during or after cardiac surgery compared with those who are not tested?

We included 11 RCTs (n = 1089, range 22–228) (13 publications)35,46–57 for the assessment of VE devices in patients undergoing cardiac surgery; six assessed TEG,46–51 four assessed ROTEM35,53–55 and one52 assessed ROTEG. ROTEG was an early name for ROTEM and so the study assessing ROTEG52 was grouped with the ROTEM studies in the analyses. Two RCTs53,55 were available only as abstracts.

Risk of bias assessment

There were a number of methodological issues with the RCTs included in this assessment. Only three35,51,54 of the 11 RCTs35,46–55 were rated as ‘low’ risk of bias with respect to their randomisation procedures. The trials were generally poorly reported; all were rated as ‘unclear’ or ‘high’ risk of bias on at least 50% of the assessed criteria. Allocation concealment and blinding were particularly poorly reported. Only one study50 reported sufficient information to assess risk of bias in relation to allocation concealment and this study was considered to have a ‘high’ risk of bias on this criterion. This study50 moved four patients initially randomised to the algorithm group to the control group, and so allocation was not concealed for these patients.

Five of the 11 RCTs46–48,52,55 reported details of blinding of study participants and personnel; only three46,47,52 of these were rated as ‘low’ risk of bias. In one of these studies46 the anaesthesiologist who performed the transfusion was blinded to the patient’s group assignments, in one47 the surgeons were blinded to the method of haemostasis assessment, and in the third52 the physician in charge of ROTEG and ICU physician were blinded. The other two studies48,55 explicitly stated that they were unblinded. Only three RCTs48,50,55 reported details on blinding of outcome assessors. Two RCTs48,50 were rated as ‘low’ risk of bias: one48 reported that outcomes were recorded by staff in the recovery unit who were unaware of group allocation; the other50 stated that surgeons and anaesthesiologists were not aware of group allocation at the time the decision on whether or not to transfuse was made. The third RCT55 reported that it was unblinded.

Inclusion of all study participants in analyses was the only notable area of methodological strength, with all but three trials rated as ‘low’ risk of bias for the completeness of outcome data criteria. 35,46,48,50–54 The results of risk of bias assessments are summarised in Table 8 and Figure 7; full risk of bias assessments for each study are provided in Appendix 3.

TABLE 8 - Risk of bias assessments for RCTs evaluating VE devices in patients undergoing cardiac surgery

☺, low risk; ☹, high risk; ?, unclear risk.

Study	Risk of bias
	Randomisation	Allocation concealment	Participant and personnel blinding	Outcome assessor blinding	Incomplete outcome data	Selective outcome reporting
Ak (2009)46	?	?	☺	?	☺	☺
Avidan (2004)48	?	?	☹	☺	☺	?
Girdauskas (2010)54	☺	?	?	?	☺	☺
Kultufan Turan (2006)52	?	?	☺	?	☺	☺
Nuttall (2001)50	☹	☹	?	☺	☺	?
Paniagua (2011)53	?	?	?	?	☺	☹
Rauter (2007)55	?	?	☹	☹	☹	☹
Royston (2001)49	☹	?	?	?	☹	?
Shore-Lesserson (1999)51	☺	?	?	?	☺	?
Weber (2012)35	☺	?	?	?	☺	☺
Westbrook (2009)47	?	?	☺	?	☹	?

FIGURE 7.

Proportion of studies fulfilling each risk of bias criteria for RCTs evaluating VE devices in patients undergoing cardiac surgery.

Results

Red blood cell transfusion

All but one49 of the included RCTs evaluated RBC transfusion as either a continuous or dichotomous outcome. Eight RCTs35,46,47,50,51,53–55 evaluated RBC transfusion within 24–48 hours as a continuous outcome (Table 9). All RCTs35,46–55 reported less volume of RBC transfusion in the VE algorithm group than in the control group but this was statistically significant in only three35,50,55 (two of ROTEM35,55 and one of TEG50); one RCT53 did not report on the statistical significance of the difference.

Six RCTs35,46,48,51,52,54 provided dichotomous data on the number of patients who received an RBC transfusion in each intervention group. The summary RR was 0.88 (95% CI 0.80 to 0.96), suggesting a significant beneficial effect of the VE testing algorithm in reducing the number of patients who received an RBC transfusion (Figure 8). There was no evidence of heterogeneity across studies (I² = 0%). Summary estimates were similar when stratified according to VE device: RR 0.86 (95% CI 0.72 to 1.02) for the three RCTs that evaluated TEG46,48,51 and 0.88 (95% CI 0.78 to 1.00) for the three RCTs that evaluated ROTEM35,54 and ROTEG. 52

TABLE 9 - Results from RCTs evaluating VE devices in patients undergoing cardiac surgery, which reported continuous data for blood component/product use

NR, not reported.

Comparisons that showed a significant difference (p < 0.05) between groups are highlighted in bold text.

Study	Data available	Intervention results	Control results	p-value for difference between groupsa
RBC transfusion (units unless otherwise stated) within 24–48 hours
Ak (2009);46 TEG	Median (IQR)	1 (0–1)	1 (1–2)	0.599
Nuttall (2001);50 TEG	Median (range)	2 (0–9)	3 (0–70)	0.039
Shore-Lesserson (1999);51 TEG	Mean (SD)	354 (487) ml	475 (593) ml	0.12
Westbrook (2009);47 TEG	Total	14	33	0.12
Girdauskas (2010);54 ROTEM	Median (IQR)	6 (2–13)	9 (4–14)	0.20
Paniagua (2011);53 ROTEM	Mean	3.8	6.4	NR
Rauter (2007);55 ROTEM	Mean	0.8	1.3	< 0.05
Weber (2012);35 ROTEM	Median (IQR)	3 (2–6)	5 (4–9)	< 0.001
Any blood component transfusion (units)
Ak (2009);46 TEG	Median (IQR)	2 (1–3)	3 (2–4)	0.001
Westbrook (2009);47 TEG	Total	37 (NR)	90 (NR)	NR
Girdauskas (2010);54 ROTEM	Median (IQR)	9 (2–30)	16 (9–23)	0.02
FFP transfusion (units, unless stated) at 12–48 hours
Ak (2009);46 TEG	Median (IQR)	1 (1–1)	1 (1–2)	0.001
Nuttall (2001);50 TEG	Median (range)	2 (0–10)	4 (0–75)	0.005
Royston (2001);49 TEG	Total	5	16	< 0.05
Shore-Lesserson (1999);51 TEG	Mean	36 (142) ml	217 (463) ml	< 0.04
Westbrook (2009);47 TEG	Total	22	18	NR
Kultufan Turan (2006);52 ROTEG	Mean (SD)	2.80 (0.95)	2.70 (1.46)	0.403
Girdauskas (2010);54 ROTEM	Median (IQR)	3 (0–12)	8 (4–18)	0.01
Paniagua (2011);53 ROTEM	Total	3.1	3.4	NR
Rauter (2007);55 ROTEM	Total	0	4	NR
Weber (2012);35 ROTEM	Median (IQR)	0 (0–3)	5 (3–8)	< 0.001
FIB (g) transfusion at 24–48 hours
Girdauskas (2010);54 ROTEM	Median (IQR)	2 (2–3)	2 (2–3)	0.70
Rauter (2007);55 ROTEM	Total	31	30	NR
Weber (2012);35 ROTEM	Median (IQR)	2 (0–4)	2 (0–6)	0.481
Platelet transfusion (units, unless otherwise stated) transfusion at 12–48 hours
Ak (2009);46 TEG	Median (IQR)	1 (1–1)	1 (1–2)	0.001
Nuttall (2001);50 TEG	Median (range)	6 (0–18)	6 (0–144)	0.0001
Royston (2001);49 TEG	Total	1	9	< 0.05
Shore-Lesserson (1999);51 TEG	Mean (SD)	34 (94) ml	83 (160) ml	0.16
Westbrook (2009);47 TEG	Total	5	15	NR
Girdauskas (2010);54 ROTEM	Median (IQR)	1 (0–4)	2 (1–3)	0.70
Paniagua (2011);53 ROTEM	Total	0.50	1.57	< 0.05
Rauter (2007);55 ROTEM	Total	0	0	NR
Weber (2012);35 ROTEM	Median (IQR)	2 (0–2)	2 (0–5)	0.010
PCC (international units) transfusion at 24–48 hours
Girdauskas (2010);54 ROTEM	Median (IQR)	0 (0–2000)	3000 (2000–3000)	< 0.001
Rauter (2007);55 ROTEM	Total	3000	13600	NR
Weber (2012);35 ROTEM	Median (IQR)	0 (0–1800)	1200 (0–1800)	0.155

FIGURE 8.

Forest plot showing RRs (95% CI) for number of patients receiving RBC transfusion in VE groups compared with control groups in cardiac patients.

Any blood component transfusion

Three RCTs46,47,54 evaluated any blood component transfusion as a continuous outcome (see Table 9). All three46,47,54 reported less volume of any blood component transfusion in the VE algorithm group than in the control group. This was statistically significant in two (one ROTEM54 and one TEG46); the third RCT54 did not report on the statistical significance of the difference.

Two RCTs51,54 provided dichotomous data on the number of patients who received any blood component (defined as any blood component in one and allogeneic blood component in the other) transfusion in each intervention group. One54 assessed ROTEM (RR 0.89, 95% CI 0.78 to 1.02) and the other51 assessed TEG (RR 0.63, 95% CI 0.44 to 0.92). The summary RR was 0.79 (95% CI 0.57 to 1.08), suggesting a beneficial effect of the VE testing algorithm in reducing the number of patients who received any blood component transfusion, although this did not reach statistical significance (Figure 9). There was some evidence of heterogeneity across studies (I² = 64%).

FIGURE 9.

Forest plot showing RRs (95% CI) for number of patients receiving any blood component transfusion in VE groups compared with control groups in cardiac patients.

Factor VIIa transfusion

Two RCTs35,54 that assessed ROTEM provided dichotomous data on the number of patients who received a factor VIIa transfusion in each intervention group. The summary RR was 0.19 (95% CI 0.03 to 1.17), suggesting a beneficial effect of the ROTEM testing algorithm, although this difference did not reach statistical significance (p > 0.05) (Figure 10). There was little evidence of heterogeneity across studies (I² = 29%).

FIGURE 10.

Forest plot showing RRs (95% CI) for number of patients receiving any factor VIIa transfusion in VE groups compared with control groups in cardiac patients.

Fresh frozen plasmas transfusion

All of the included RCTs35,46–55 evaluated FFP transfusion as either a continuous or dichotomous outcome. Ten RCTs35,46,47,49–55 evaluated RBC transfusion within 24–48 hours as a continuous outcome (see Table 9). All but two RCTs47,52 reported less volume of FFP transfusion in the VE algorithm group than in the control group; this was statistically significant in six35,46,49–51,54 (two of ROTEM35,54 and four of TEG46,49–51); three RCTs47,53,55 did not report on the statistical significance of the difference.

Five RCTs35,46,48,51,54 provided dichotomous data on the number of patients who received an FFP transfusion in each intervention group, all but one48 of which also reported continuous data. The summary RR was 0.47 (95% CI 0.35 to 0.65), suggesting a significant beneficial effect of the VE testing algorithm in reducing the number of patients who received an FFP transfusion (Figure 11). There was no evidence of heterogeneity across studies (I² = 0%). However, the study by Avidan et al. 48 appeared to be an outlying result, with a RR of 5.0; however, the CI was very wide (95% CI 0.25 to 101.61). This was due to the very small number of events (two in the intervention arm and zero in the control arm). Removal of this study48 from the meta-analysis had very little impact on the summary estimate (RR 0.47, 95% CI 0.37 to 0.61). Summary estimates were similar when stratified according to VE device: RR 0.52 (95% CI 0.20 to 1.35) for the three RCTs that evaluated TEG46,48,51 and 0.46 (95% CI 0.34 to 0.63) for the two RCTs35,54 that evaluated ROTEM.

FIGURE 11.

Forest plot showing RRs (95% CI) for number of patients receiving FFP transfusion in VE groups compared with control groups in cardiac patients.

Fibrinogen transfusion

Three RCTs35,54,55 evaluated any FIB transfusion as a continuous outcome (see Table 9). All three35,54,55 reported no difference between the VE algorithm group compared with the control group in the volume of FIB transfused. Two of these RCTs35,54 also provided dichotomous data on the number of patients who received a FIB transfusion in each intervention group. The summary RR was 0.94 (95% CI 0.77 to 1.14), suggesting no difference between the treatment groups (Figure 12).

FIGURE 12.

Forest plot showing RRs (95% CI) for number of patients receiving FIB transfusion in VE groups compared with control groups in cardiac patients.

Platelet transfusion

All of the included RCTs35,46–55 evaluated platelet transfusion as either a continuous or dichotomous outcome. Nine RCTs35,46,47,49–51,53–55 evaluated platelet transfusion within 24–48 hours as a continuous outcome (see Table 9). All RCTs reported less volume of platelet transfusion in the VE algorithm group compared with the control group but this was statistically significant in only five RCTs35,46,49,50,54 (two of ROTEM35,54 and three of TEG46,49,50); two RCTs47,55 did not report on the statistical significance of the difference.

Six RCTs35,46,48,51,52,54 provided dichotomous data on the number of patients who received a platelet transfusion in each intervention group. The summary RR was 0.72 (95% CI 0.58 to 0.89), suggesting a significant beneficial effect of the VE testing algorithm in reducing the number of patients who received a platelet transfusion (Figure 13). There was no evidence of heterogeneity across studies (I² = 0%). Summary estimates were similar when stratified according to VE device: RR 0.56 (95% CI 0.36 to 0.86) for the three RCTs46,48,51 that evaluated TEG and 0.78 (95% CI 0.60 to 1.00) for the three RCTs35,52,54 that evaluated ROTEM and ROTEG.

FIGURE 13.

Forest plot showing RRs (95% CI) for number of patients receiving platelet transfusion in VE groups compared with control groups in cardiac patients.

Prothrombin complex concentrate transfusion

Three RCTs35,54,55 evaluated any PCC transfusion as a continuous outcome (see Table 9). All three RCTs35,54,55 reported less volume of PCC transfusion in the VE algorithm group than in the control group but this was statistically significant in only one RCT (p < 0.001);54 one RCT55 did not report on the statistical significance of the difference.

Two of these RCTs35,54 also provided dichotomous data on the number of patients who received a PCC transfusion in each intervention group. The summary RR was 0.39 (95% CI 0.08 to 1.95), suggesting no difference between the treatment groups (Figure 14).

FIGURE 14.

Forest plot showing RRs (95% CI) for number of patients receiving PCC transfusion in VE groups compared with control groups in cardiac patients.

Bleeding

Nine RCTs35,46–52,54 evaluated bleeding, generally measured as mediastinal tube drainage, as a continuous outcome (Table 10). The majority reported less bleeding in the VE intervention group; however, only two studies35,50 reported a statistically significant difference in bleeding between the two groups.

TABLE 10 - Results from RCTs evaluating VE devices in patients undergoing cardiac surgery, which reported continuous data for clinical outcomes

NR, not reported.

Bold indicates statistically significant (p < 0.05) results.

Study	Data available	Intervention results	Control results	p-value for difference between groups
Bleeding/mediastinal tub drainage (ml) at 12-/24-hour follow-up
Ak (2009);46 TEG	Mean (SD)	480.5 (351.0)	591.4 (339.2)	0.087
Avidan (2004);48 TEG	Median (IQR)	755 (606–975)	850 (688–1095)	> 0.05
Nuttall (2001);50 TEG	Median (range)	590 (240–2335)	850 (290–10,190)	0.019
Royston (2001);49 TEG	Median (IQR)	470 (295–820)	390 (240–820)	NR
Shore-Lesserson (1999);51 TEG	Mean (SD)	702 (500)	901 (847)	0.27
Westbrook (2009);47 TEG	Median (IQR)	875 (755–1130)	960 (820–1200)	0.437
Kultufan Turan (2006);52 ROTEG	Mean (SD)	837.5 (494.1)	711.10 (489.2)	0.581
Girdauskas (2010);54 ROTEM	Median (IQR)	890 (600–1250)	950 (650–1400)	0.50
Weber (2012);35 ROTEM	Median (IQR)	600 (263–875)	900 (600–1288)	0.021
Length of ICU stay (hours)
Ak (2009);46 TEG	Mean (SD)	23.3 (5.7)	25.3 (11.2)	0.099
Westbrook (2009);47 TEG	Median (IQR)	29.4 (14.3–56.4)	32.5 (22.0–74.5)	0.369
Girdauskas (2010);54 ROTEM	Mean (SD)	175.2 (218.4)	194.4 (201.6)	0.6
Weber (2012);35 ROTEM	Median (IQR)	21 (18–31)	24 (20–87)	0.019
Length of hospital stay (days)
Ak (2009);46 TEG	Mean (SD)	6.2 (1.1)	6.3 (1.4)	0.552
Westbrook (2009);47 TEG	Median (IQR)	9 (7–13)	8 (7–12)	> 0.05
Girdauskas (2010);54 ROTEM	Mean (SD)	16.6 (16.4)	17.0 (14.8)	0.80
Weber (2012);35 ROTEM	Median (IQR)	12 (9–22)	12 (9–23)	0.718

Re-operation

Seven RCTs35,46,48–51,54 provided dichotomous data on the number of patients who required re-operation to investigate bleeding in each intervention group. The summary RR was 0.72 (95% CI 0.41 to 1.26), suggesting a beneficial effect of the VE testing algorithm in reducing the number of patients requiring re-operation, however, this difference was not statistically significant (Figure 15). There was no evidence of heterogeneity across studies (I² = 0%). Summary estimates were similar when stratified according to VE device: RR 0.75 (95% CI 0.31 to 1.83) for the five RCTs46,48–51 that evaluated TEG and 0.69 (95% CI 0.33 to 1.44) for the two RCTs35,54 that evaluated ROTEM.

FIGURE 15.

Forest plot showing RRs (95% CI) for number of patients requiring re-operation in VE groups compared with control groups in cardiac patients.

Surgical source of bleeding identified on re-operation

Four RCTs46,50,51,54 provided dichotomous data on the number of patients in whom a surgical source of bleeding was identified on re-operation in each intervention group. The summary RR was 1.04 (95% CI 0.42 to 2.57), suggesting no difference between the intervention groups (Figure 16). There was very little evidence of heterogeneity across studies (I² = 3%). One RCT assessed ROTEM54 and reported a RR of 0.86 (95% CI 0.26 to 2.87); the summary estimate for the three RCTs assessing TEG50,51,54 was similar at 0.99 (95% CI 0.18 to 5.36).

FIGURE 16.

Forest plot showing RRs (95% CI) for number of patients in whom a surgical source of bleeding was identified on re-operation in VE groups compared with control groups in cardiac patients.

Length of intensive care unit stay

Four RCTs35,46,47,54 evaluated the length of ICU stay as a continuous outcome (see Table 10). All studies35,46,47,54 reported shorter stays in the VE group than in the control group but this difference was statistically significant in only one study. 35

Length of hospital stay

Four RCTs35,46,47,54 evaluated the length of hospital stay as a continuous outcome (see Table 10). All studies35,46,47,54 reported that durations of stay were similar in the two treatment groups; none reported a statistically significant difference between groups.

Adverse events

One study35 reported on adverse events, including acute renal failure, sepsis and thrombotic complications. All were reduced in the ROTEM group compared with SLTs but differences were not statistically significant for individual outcomes. When the compound outcome of any adverse event was considered this was found to be significantly reduced in the ROTEM group compared with SLTs (RR 0.21, 95% CI 0.08 to 0.57).

Mortality

Four RCTs46,49,51,54 provided dichotomous data on the number of deaths (within 24 hours,51 48 hours,49 in hospital54 or ‘early mortality’46) in each intervention group. The summary RR was 0.87 (95% CI 0.35 to 2.18), suggesting no difference between the intervention groups (Figure 17). There was no evidence of heterogeneity across studies (I² = 0%). One RCT assessed ROTEM54 and reported a RR of 0.86 (95% CI 0.26 to 2.87); the summary estimate for the three RCTs46,49,51 assessing TEG was similar at 0.88 (95% CI 0.21 to 3.66). An additional RCT35 provided data on 6-month mortality. This study35 reported significantly reduced mortality in the VE testing group at 6 months compared with the SLT group (RR 0.20, 95% CI 0.05 to 0.87).

FIGURE 17.

Forest plot showing RRs (95% CI) for number of deaths in VE groups compared with control groups in cardiac patients.

Other reported outcomes

Data were also reported on the following outcomes but each were assessed in only one or two studies, and so are not discussed in detail here: cryoprecipitate use, desmopressin treatment, dialysis-dependent renal failure, duration of ventilation, factor VIIa, fresh blood transfusion, intubation time, need for additional protamine, non-RBC balance, post-operative confusion, reinfusion, reintubation, stroke, time to stop bleeding, total heparin dose, total protamine dose, total ventilation time, time to extubation and tranexamic acid use. Full results can be found in Appendix 2.

Summary

Pooled estimates from each of the meta-analyses are summarised in Table 11. Overall, there was a significant reduction in RBC transfusion, platelet transfusion and FFP transfusion in VE testing groups compared with control. There was no significant difference between groups in terms of any blood component transfusion, factor VIIa transfusion or PCC transfusion, although data suggested a beneficial effect of the VE testing algorithm but these outcomes were evaluated in only two studies. There was no difference between groups in terms of FIB transfusion. Continuous data on blood component/product use, although inconsistently reported across studies, supported these findings; the only blood component/product that was not associated with a reduced volume of use in the VE testing group was FIB. There was a suggestion that bleeding was reduced in the VE testing groups but this was statistically significant in only two of the nine RCTs that evaluated this outcome. Clinical outcomes (re-operation, surgical cause of bleed on re-operation and mortality) did not differ between groups. There was some evidence of reduced bleeding and ICU stay in the VE testing groups compared with control but this was not consistently reported across studies. There was no difference in length of hospital stay between groups. There was no apparent difference between ROTEM or TEG for any of the outcomes evaluated.

TABLE 11 - Pooled estimates for dichotomous outcomes from RCTs evaluating VE devices in patients undergoing cardiac surgery

Bold indicates statistically significant (p < 0.05) results.

Outcome	Summary RR (95% CI)	No. of studies	Heterogeneity
Blood component/product use
RBC transfusion	0.88 (0.80 to 0.96)	6	Q = 4.47, p = 0.48, I2 = 0%
Any blood component transfusion	0.79 (0.57 to 1.08)	2	Q = 2.79, p = 0.10, I2 = 64%
Platelet transfusion	0.72 (0.58 to 0.89)	6	Q = 4.47, p = 0.48, I2 = 0%
FFP transfusion	0.47 (0.35 to 0.65)	5	Q = 4.72, p = 0.45, I2 = 0%
Factor VIIa transfusion	0.19 (0.03 to 1.17)	2	Q = 1.40, p = 0.24, I2 = 29%
FIB transfusion	0.94 (0.77 to 1.14)	2	Q = 1.09, p = 0.30, I2 = 8%
PCC transfusion	0.39 (0.08 to 1.95)	2	Q = 10.23, p = 0.00, I2 = 90%
Clinical outcomes
Re-operation	0.72 (0.41 to 1.26)	7	Q = 3.49, p = 0.74, I2 = 0%
Surgical cause of bleed on re-operation	1.04 (0.42 to 2.57)	4	Q = 3.09, p = 0.38, I2 = 3%
Mortality	0.87 (0.35 to 2.18)	4	Q = 1.26, p = 0.74, I2 = 0%

How well do viscoelastic devices predict relevant clinical outcomes during or after cardiac surgery?

As none of the RCTs evaluated the Sonoclot VE test, we included lower levels of evidence for this device. Three prediction studies58–60 that evaluated Sonoclot were included in the review; two of these studies59,60 also evaluated TEG and so provided a direct comparison between these two devices. Baseline data from these studies are summarised in Table 12; full details of the studies are provided in Appendix 2.

Results

All three studies58–60 provided data that allowed calculation of ORs for the prediction of bleeding in patients who tested positive on a particular test or test parameter (Sonoclot, TEG or SLTs) compared with those who tested negative (Figure 18). Positive results on conventional tests, TEG and Sonoclot were all associated with an increased risk of bleeding with no clear differences according to test. Nuttall et al. 59 evaluated individual components of each of the tests separately, and found that all of the parameters investigated, with the exception of one TEG and one Sonoclot parameter, were associated with a significant (p < 0.05) increased risk of bleeding. Two of the SLTs (PT and aPTT) showed higher ORs than other parameters, but CIs overlapped with other SLTs and TEG and Sonoclot parameters. Bischof et al. 58 also evaluated individual test components but evaluated only the Sonoclot test; a direct comparison between Sonoclot and TEG or SLTs was therefore not possible from the results of this study. All three Sonoclot parameters showed a strong positive relationship with bleeding. Tuman et al. 60 was potentially the most informative study, as it evaluated each test class as a whole, that is, it evaluated a positive ‘TEG’ result rather looking at individual components of the TEG, similarly it evaluated SLTs as a class and Sonoclot as a whole. This study found that a positive TEG or Sonoclot result were both highly predictive of bleeding. However, the study60 was very small and CIs were wide. The limited data suggested that TEG results were more predictive than Sonoclot, but CIs overlapped. The SLTs performed less well and were not predictive of bleeding; this study60 was performed in 1989 and so may not be reflective of current practice.

FIGURE 18.

Forest plot showing ORs (95% CI) for prediction of bleeding by VE devices and SLTs in cardiac patients. A60, amplitude 60 minutes after clotting time; CR, clot rate; k, kinetics; MA, maximum amplitude; MPV, maximum platelet volume; NR, not reported; PF, platelet function; PTT, partial thromboplastin time; R, clotting time; R1, rate of fibrin monomer formation; R2, fibrinogenesis and platelet interaction; R3, rate of platelet mediated clot contraction.

How do clinical outcomes differ among patients with coagulopathy induced by trauma who are tested with viscoelastic devices compared with those who are not tested?

We identified one ongoing RCT62 that is comparing TEG (rapid assay) with conventional coagulation testing [INR, partial thromboplastin time (PTT), FIB, D-dimer] in adults with blunt or penetrating trauma who are likely to require a transfusion of RBCs within 6 hours from admission, as indicated by clinical assessment. 61,62 Additional information on this trial was provided by the study authors in the form of the study protocol. 62 The following outcomes are being evaluated in this study: quality and quantity of blood components transfused (packed RBCs, FFP, cryoprecipitate and apheresis platelets); patterns of transfusion ratios of RBC/FFP; haemorrhage-related deaths specified as very early mortality (< 2 hours post injury) and early mortality; late mortality; cessation of coagulopathic bleeding; multiple organ failure (MOF). Results from this study are not yet available. As no other RCTs were identified, we therefore considered lower levels of evidence for this objective. One CCT,63 reported only as an abstract, was included. This study63 compared a rapid-TEG-guided protocol with a standard transfusion protocol in adult trauma patients requiring massive transfusion (> 12 RBC units in 24 hours or > 4 units in 4 hours); both groups also included a near-patient haematocrit (HCT) assay. This study63 did not report numerical or statistical outcome data. It stated that there were no statistically significant differences between groups for death, acute respiratory distress syndrome (ARDS), systemic inflammatory response syndrome (SIRS), multisystem organ failure, sepsis, deep-vein thrombosis (DVT), stroke, acute coronary syndrome or days to discharge. There was a non-significant trend towards reduced pneumonia, days on the ventilator and ICU-days, and a trend towards increasing platelet use in the TEG-treated group. Baseline data from these studies are summarised in Table 14; full details of the studies are provided in Appendix 2. No other studies with a concurrent control group were identified for the trauma population.

How well do viscoelastic devices predict relevant clinical outcomes in patients with coagulopathy induced by trauma?

As there were insufficient data from studies that evaluated differences in clinical outcomes between VE tested and untested populations, we included lower levels of evidence for this objective. Fifteen prediction studies64–81 (18 publications; n = 4217) were included for this objective. Nine studies64,67,69,71–74,77,81 evaluated TEG, four of these71,74,77,79,80 also evaluated SLTs; the other six studies evaluated ROTEM,65,66,68,70,75,76 with four65,68,70,76 also evaluating SLTs. No studies of Sonoclot were identified. None of the studies evaluated both TEG and ROTEM in the same patients. Baseline data from these studies are summarised in Table 15; full details of the studies are provided in Appendix 2.

Risk of bias and applicability assessment

All of the studies64–77,80 that assessed the ability of VE testing devices to predict outcomes in trauma patients used a predictive accuracy or prediction modelling approach. The main areas of concern with regard to these studies were the process of participant selection and the applicability to the objectives of this assessment of the way in which both VE testing and the reference standard were applied. With two exceptions,73,74 all studies were rated as ‘high’ or ‘unclear’ risk of bias in the participant selection process, usually because of poor reporting or inappropriate exclusion of particular groups of patients. Ten of the 15 studies were rated as having ‘high’ applicability concerns for the index test because they assessed the predictive ability of selected individual components of VE testing rather than assessing the device as a whole or reporting data for all assays and parameters measured by the device;65,67,68,70,73–77,80 two further studies66,69 were rated as having ‘unclear’ applicability because, although the testing VE device was specified, no details of the assay(s) used or parameters measured were reported. Ten studies64,65,70,72–77,80 were rated as having ‘high’ applicability concerns with respect to the reference standard, where the reference standard was one or more measure(s) of transfusion requirements, because it was unclear whether or not the decision to transfuse was informed by VE testing results, this also resulted in an ‘unclear’ risk of bias rating with respect to the reference standard. In practice, the results of VE testing would inform the decision to transfuse, a situation that gives rise to the paradox that this type of study cannot have both ‘low’ risk of bias and ‘low’ applicability with respect to the reference standard; if the reference standard is applied as it would be in clinical practice, the study will necessarily be subject to incorporation bias. The remaining five studies66–69,71 were rated as ‘low’ applicability concerns because they reported objective reference standards (e.g. mortality). The results of QUADAS-2 assessments are summarised in Table 16 and Figure 19; full QUADAS-2 assessments for each study are provided in Appendix 3.

TABLE 16 - QUADAS-2 assessments for prediction studies evaluating VE devices in patients with coagulopathy induced by trauma

☺, low risk; ☹, high risk; ?, unclear risk.

Study	Risk of bias					Applicability concerns
	Patient selection	Index test	Reference standard		Flow and timing	Patient selection	Index test	Reference standard
Cotton (2011)73	☺	☹	?		☺	☺	☹	☹
Davenport (2011)70	☹	☹	?		☺	☺	☹	☹
Holcomb (2012)74	☺	?	?		☺	☺	☹	☹
Ives (2012)72	☹	☺	☺	?	☹	☺	☺	☺	☹
Jeger (2012)77	☹	☹	☺		☺	☺	☹	☹
Kaufman (1997)64	?	☺	☺		☺	☺	☺	☹
Korfage (2011)75	?	?	?		☺	?	☹	☹
Kunio (2012)67	?	☺	☺		☺	☹	☹	☺
Leeman (2010)65	☹	☺	?		☺	☹	☹	☹
Nystrup (2011)71	☹	?	☺		☺	☺	☺	☺
Pezold (2012)80	☹	☹	?		☺	☺	☹	☹
Schöchl (2011)76	☹	☹	?		☺	☺	☹	☹
Schöchl (2011)68	?	☹	☺		☺	☹	☹	☺
Tapia (2012)69	☹	?	☺		☹	?	?	☺
Tauber (2011)66	?	☺	☺		☺	☹	?	☺

FIGURE 19.

Proportion of studies fulfilling each QUADAS-2 criteria for prediction studies evaluating VE devices in patients with coagulopathy induced by trauma.

Results

Red blood cell transfusion

Three studies70,72,73 (two of TEG,72,73 one of ROTEM and SLTs70) evaluated the ability of VE devices to predict RBC transfusion (Figure 20). One used an end point of any RBC transfusion within 12 hours,70 one within 6 hours73 and one72 did not specify the time point. A positive result on each of the parameters assessed, with the exception of CT on ROTEM, was associated with an increased risk of RBC transfusion. There were no clear differences between ROTEM parameters or ROTEM and SLTs in the one study70 that reported multiple evaluations.

FIGURE 20.

Forest plot showing ORs (95% CI) for prediction of RBC transfusion by VE devices and SLTs in trauma patients. a, Adjusted OR based on multivariate analysis. EPL, estimated per cent lysis.

Any blood component transfusion

Three studies evaluated the ability of VE devices to predict any blood component transfusion (Figure 21). 64,75,77 Two evaluated TEG64,77 and one evaluated ROTEM;75 one of the studies of TEG77 also evaluated SLTs. The time frame for transfusion was within 24 hours in two studies64,77 and within 48 hours in the third. 75 A positive result on each of the parameters assessed was associated with an increased risk of any blood component transfusion; an overall TEG results suggesting the patient was hypercoagulable was associated with a decreased risk of transfusion (OR 0.14, 95% CI 0.03 to 0.76). One of the studies77 did not provide sufficient data to calculate CIs and so the significance of the ORs from this study could not be assessed. The other two studies64,75 both reported statistically significant (p < 0.05) associations for all parameters assessed. An overall TEG result indicating that the patient was hypocaoguable was found to be associated with the greatest increased risk of transfusion, but CIs were very wide (OR 180.00, 95% CI 14.15 to 2289.13). ORs for individual TEG, ROTEM or SLTs were much smaller, ranging from 2.50 to 15.26.

FIGURE 21.

Forest plot showing ORs (95% CI) for prediction of any blood component transfusion by VE devices and SLTs in trauma patients. a, Adjusted OR based on multivariate analysis. MA, maximum amplitude.

Massive transfusion

Six studies evaluated the ability of VE devices to predict massive RBC transfusion. 65,70,73,75,76,80 Three evaluated TEG73,74,80 and three evaluated ROTEM;65,70,76 all but one73 also evaluated SLTs. All used a threshold of ≥ 10 units of RBC transfused to define massive transfusion but the time frame within which this had to occur ranged from 6 hours to 24 hours. Three studies65,73,74 provided data as adjusted ORs for at least one of the VE test components; a further study70 provided data that permitted calculation of ORs (Figure 22). The other two studies76,80 provided data only on AUC for the ROC curve, together with 95% CIs (Figure 23). A positive result on each of the parameters assessed was associated with an increased risk of massive transfusion; however, this difference was not statistically significant for some of the ROTEM parameters and SLTs. There were no clear differences between ROTEM, TEG or SLTs, or individual test parameters in terms of ability to predict massive transfusion. AUCs, where reported, were between 0.70 and 0.92, with no clear differences between ROTEM, TEG or SLTs.

FIGURE 22.

Forest plot showing ORs (95% CI) for prediction of massive transfusion by VE devices and SLTs in trauma patients. a, Adjusted OR based on multivariate analysis. LY30, lysis at 30 minutes; MA, maximum amplitude.

FIGURE 23.

Forest plot showing AUCs (95% CI) of ROC curves for prediction of massive transfusion by VE devices and SLTs in trauma patients.

Mortality

Seven studies66–69,71,72,80 evaluated the association of the results of VE devices with mortality. Five studies67,69,71,72,80 evaluated TEG; two studies66,68 evaluated ROTEM; and three studies68,71,80 evaluated SLTs. Two studies66,72 defined mortality as death within 24 hours, one67 as death in hospital, two69,71 as death within 30 days, one68 did not provide a definition, and one80 restricted its definition of mortality to coagulation-related mortality [death after receiving a massive transfusion of ≥ 10 packed red blood cell (PRBC) units].

Two studies71,72 provided data as adjusted ORs; three further studies66,67,69 provided data that permitted calculation of ORs and associated CIs (Figure 24). The other two studies68,80 provided data on AUC only for the ROC curve, together with 95% CIs; these data were also reported in one of the studies71 that reported adjusted ORs (Figure 25). A positive result assessed was associated a statistically significant increased risk of death for most parameters assessed. The only exceptions were two parameters that were associated with a decreased risk of death, although this difference was not statistically significant: the presence of moderate hyperfibrinolysis (OR 0.76, 95% CI 0.09 to 6.20)66 and an overall TEG result suggesting that a patient was hypocoagulable (OR 0.23, 95% CI 0.03 to 1.91). 72 Three studies66,69,72 that evaluated a ROTEM or TEG result indicating the presence of hyperfibrinolysis showed the strongest association with death, with ORs ranging from 25 to 147, although CIs were wide. AUCs were between 0.63 and 0.93, with no clear differences between ROTEM, TEG or SLTs.

FIGURE 24.

Forest plot showing ORs (95% CI) for prediction of death by VE devices and SLTs in trauma patients. a, Adjusted OR based on multivariate analysis. EPL, estimated per cent lysis; MA, maximum amplitude; NR, not reported.

FIGURE 25.

Forest plot showing AUCs (95% CI) of ROC curves for prediction of death by VE devices and SLTs in trauma patients. MA, maximum amplitude; NR, not reported.

Other outcomes

Data were also reported on the following outcomes, but each outcome was assessed only in single studies and so are not discussed in detail here: FFP transfusion, massive transfusion of cryoprecipitate, massive transfusion of plasma, massive transfusion of platelets, plasma transfusion, platelet transfusion, substantial bleeding and neurosurgical intervention. Full results can be found in Appendix 2.

Summary

Fifteen studies provided data on the accuracy of TEG or ROTEM for the prediction of transfusion-related outcomes and death in trauma patients; eight studies also provided data on the accuracy of SLTs. The studies generally found that a positive result on each of the TEG or ROTEM parameters or on SLTs was associated with an increased risk of transfusion (RBC, any blood component and massive transfusion) and death. There was no clear difference between ROTEM, TEG or SLTs. However, none of the studies provided a direct comparison between TEG and ROTEM. An overall TEG result suggesting that a patient was hypocoagulable was the strongest predictor of any blood component transfusion. The presence of hyperfibrinolysis was the strongest predictor of mortality.

How do clinical outcomes differ among patients with post-partum haemorrhage who are tested with viscoelastic devices compared with those who are not tested?

No studies were identified that compared clinical outcomes among patients with PPH who were tested with VE devices compared with those who were not tested.

How well do viscoelastic devices predict relevant clinical outcomes in patients with post-partum haemorrhage?

As no studies evaluated differences in clinical outcomes between VE-tested and untested populations, we included lower levels of evidence for this objective. Two prediction studies82,83 were included in the review (n = 245). Both studies were available only as abstracts. Baseline data from these studies are summarised in Table 17; full details of the studies are provided in Appendix 2.

Risk-of-bias and applicability assessment

As with the trauma studies, the main areas of concern with regard to the two prediction studies82,83 conducted in patients with PPH were the applicability to the objectives of this assessment of the way in which both VE testing and the reference standard were applied. One study83 was rated as having ‘high’ applicability concerns for the index test because it assessed the predictive ability of selected individual parameters of the FIBTEM assay on the ROTEM device, rather than assessing the device as a whole, or reporting data for all assays and parameters measured by the device; the other study82 was rated as having ‘unclear’ applicability because, although it assessed the ROTEM device, no details of the assay(s) used were reported. Both studies were rated as having ‘high’ applicability concerns with respect to the reference standard because it was unclear whether or not the decision to transfuse was informed by ROTEM results, this also resulted in an ‘unclear’ risk of bias rating with respect to the reference standard. 82,83 In practice, the results of ROTEM testing would inform the decision to transfuse, a situation that gives rise to the paradox that this type of study cannot have both ‘low’ risk of bias and ‘low’ applicability with respect to the reference standard; if the reference standard is applied as it would be in clinical practice, the study will necessarily be subject to incorporation bias. The results of QUADAS-2 assessments are summarised in Table 18; full QUADAS-2 assessments for each study are provided in Appendix 3.

TABLE 18 - QUADAS-2 assessments for prediction studies evaluating VE devices in patients with PPH

☺, low risk; ☹, high risk; ?, unclear risk.

Study	Risk of bias				Applicability concerns
	Patient selection	Index test	Reference standard	Flow and timing	Patient selection	Index test	Reference standard
Bolton (2011)82	?	?	?	☺	☹	?	☹
Lilley (2013)83	☺	?	?	☺	☺	☹	☹

Results

Both studies82,83 provided data that allowed calculation of ORs for the prediction of outcomes in patients who tested positive on ROTEM compared with those who tested negative (Figure 26). The study83 that evaluated ROTEM and SLTs reported data in a format that allowed calculation of ORs only for the ROTEM parameter (MCF based on FIBTEM analysis) for the prediction of any RBC transfusion. There was a strong positive relationship between this parameter and RBC transfusion (OR 41.54, 95% CI 9.01 to 191.59). Data for other outcomes and for the SLT (Clauss fibrinogen) were reported as AUC for the ROC curve; these were very similar for Clauss fibrinogen and for MCF based on ROTEM (FIBTEM). 83 CIs were not presented and so formal comparisons of AUCs was not possible.

FIGURE 26.

Forest plot showing ORs (95% CI) for prediction of specified outcomes by ROTEM in women with PPH. NR, not reported.

The other study82 reported that a positive ROTEM result was associated with coagulopathy requiring treatment (OR 168.0, 95% CI 15.6 to 1814. 7). This study82 also evaluated FFP transfusion and platelet transfusion; data were available to calculate ORs for these outcomes but not associated CIs. The ROTEM results were also predictive of both these outcomes but the significance of the association was unclear. The size of the OR was smaller than for the association with coagulopathy requiring treatment (OR 76 for FFP transfusion and 19 for platelet transfusion). 82

Summary

Only two studies82,83 were identified that evaluated VE devices in patients with PPH. Both provided data on the accuracy of ROTEM for the prediction of outcomes; one also evaluated a SLT (Clauss fibrinogen). Both studies82,83 showed that ROTEM results were associated with the outcomes evaluated (RBC transfusion, invasive procedures, coagulopathy requiring treatment, FFP transfusion and platelet transfusion). The study that evaluated both ROTEM and Clauss fibrinogen reported similar results for both tests but did provide CIs to accompany effect estimates.

Search methods

Searches were undertaken to identify cost-effectiveness studies of VE POC testing. As with the clinical effectiveness searching, the main EMBASE strategy for each set of searches was independently peer reviewed by a second information specialist, using the CADTH Peer Review Checklist. 40 Search strategies were developed specifically for each database and searches took into account generic and other product names for the intervention. All search strategies are reported in Appendix 1.

The following databases were searched for relevant studies from inception to November 2013:

• MEDLINE (OvidSP): 1946–October 4 week 2013

• MEDLINE In-Process & Other Non-Indexed Citations and Daily Update (OvidSP): up to 5 November 2013

• EMBASE (OvidSP): 1974–5 November 2013

• NHS Economic Evaluation Database (EED) (Wiley): Issue 4, October 2013

• EconLit (EBSCOhost): 1990–1 September 2013

• Health Economic Evaluation Database (HEED) (Wiley): up to 7 November 2013, http://onlinelibrary.wiley.com/book/10.1002/9780470510933

• IDEAS via Research Papers in Economics (REPEC) (Internet): up to 7 November 2013, http://repec.org/

Identified references were downloaded in EndNote X4 software for further assessment and handling. References in retrieved articles were checked for additional studies.

Inclusion criteria

Cost minimisation and cost-effectiveness studies that evaluated the use of TEG, ROTEM or Sonoclot compared with a control group (either concurrent or historical) consisting of no-testing, clinical judgement or SLTs were eligible for inclusion. Studies in children were excluded.

Quality assessment

Full cost-effectiveness studies were appraised using the Drummond checklist. 84

Results

The searches identified 331 records, of which five studies12,85–88 fulfilled the inclusion criteria; two85,86 were available only as conference abstracts (Figure 27). Three studies86–88 were conducted in cardiac patients, one85 in patients undergoing liver transplant, and one12 in both cardiac and liver transplant patients. One study12 was a formal cost-effectiveness analysis of VE devices in cardiac and liver transplant patients. The other four studies85–88 were cost-minimisation studies performed alongside a retrospective before-and-after study.

FIGURE 27.

Flow of studies through the health-economic review process.

Cardiac surgery

We adopted the model structure used by the Health Technology Assessment (HTA) undertaken for NHS Scotland in 2008,12 which was largely based on a cost-effectiveness study of cell salvage and alternative methods of minimising perioperative allogeneic blood transfusion by Davies et al. 90 As these studies were undertaken in 2008 and 2006, respectively, more recent data sources were used to update the input parameters of the model wherever possible.

Our model is based on a decision tree that starts with the choice of strategy to be followed, that is, VE device (ROTEM, TEG or Sonoclot) or SLTs. Within each strategy, patients then either do or do not receive a transfusion.

Red blood cell transfusion, where it occurs, may be associated with adverse events or complications. The complications included in the model were those considered in Davies et al. 90 and the Scottish HTA. 12 Most complications are a consequence of RBC transfusion, although some were modelled as a consequence of any transfusion.

Complications were categorised as (1) complications related to surgery and/or transfusion or (2) transfusion-related complications. Complications related to surgery and/or transfusion included in the model were renal dysfunction, myocardial infarction, stroke, thrombosis, excessive bleeding requiring re-operation, wound complications and septicaemia. Transfusion-related complications included transfusion-associated graft-versus-host disease, complications related to the administration of an incorrect blood component, haemolytic transfusion reactions (acute or delayed), post-transfusion purpura (PTP), transfusion-related acute lung injury (TRALI) and febrile reaction. In addition, we assumed that patients may also experience transfusion-transmitted infections. Transfusion-transmitted infections include bacterial contamination, variant Creutzfeldt–Jakob disease (vCJD), hepatitis A virus (HAV), malaria, human T-cell lymphotropic virus (HTLV), human immunodeficiency virus (HIV), hepatitis B virus (HBV) and hepatitis C virus (HCV). The model structure is shown in Figure 28.

FIGURE 28.

Cost-effectiveness model structure.

The model’s time horizons were set to 1 month and 1 year because the benefits of a reduction in RBC transfusion were considered to have occurred within this time frame. At 1 month, the model reflects the period of hospitalisation and accordingly captures the impact of complications related to surgery and blood loss, transfusion-related complications and infection caused by bacterial contamination. It should be noted that, as in Davies et al. ’s study,90 bacterial contamination is the only transfusion-transmitted infection that was assumed to occur during the hospitalisation period. For other transfusion-transmitted infections included in the model, a time horizon of 1 year was considered more appropriate, as these infections do not usually manifest themselves immediately. Discounting was not necessary, as the longest time horizon was set at 1 year. Costs were estimated from the perspective of the NHS in England and Wales. Consequences were expressed in life-years (LYs) gained and quality-adjusted life-years (QALYs). QALY weights (utilities) were assigned to adverse events to express their consequences. Sensitivity analysis relating to extended time periods would have been undertaken had there been potential to impact on results and conclusions.

Patients with coagulopathy induced by trauma

The model for trauma patients has largely the same structure as the model in cardiac surgery patients. The only difference relates to the ‘surgery- and/or transfusion-related complications’, which were replaced with ‘trauma- and/or transfusion-related complications’ – ARDS and MOF.

Cardiac surgery

Probability of red blood cell transfusion

We estimated the baseline risk of having a transfusion based on the number of transfusions in the SLTs group in the cardiac surgery trials included in the effectiveness review (Figure 29). This analysis was based on the studies by Ak et al. ,46 Avidan et al. ,48 Shore-Lesserson et al. 51 and Kultufan Turan et al. 52 We excluded two studies,35,54 as we did not think that the patients included in these studies were representative of general cardiac surgery patients; one study54 enrolled only high-risk patients (aortic surgery requiring hypothermic circulatory arrest, including urgent and emergency surgery) and the other study35 was restricted to patients with excessive bleeding. We used a double arcsine transformation before pooling data using the MetaXL (Epigear International, Wilston, QLD, Australia) add-on for Excel. 91 The summary estimate for the probability of RBC transfusion from these four studies46,48,51,52 was 0.592 (95% CI 0.528 to 0.654). The RR of RBC transfusion in cardiac surgery patients whose blood was tested with VE devices compared with SLTs was reported in Chapter 3 (see Results).

FIGURE 29.

Forest plot showing the probability of RBC transfusion (95% CI) in control groups in cardiac surgery trials.

To estimate the probability of RBC transfusion for the three VE strategies a RR was applied to the baseline prevalence of RBC transfusion. The effectiveness review found no evidence of a difference in the RR of RBC transfusion between studies that assessed ROTEM and those that assessed TEG (see Chapter 3, Results). We therefore applied the summary RR for RBC transfusion estimated for all studies for the ROTEM and TEG models. Limited data suggested that the accuracy of Sonoclot in predicting clinical outcomes may be similar to that of TEG. We therefore also assumed that this summary RR could be applied in the Sonoclot model. The baseline prevalence of RBC transfusion in patients who received SLTs and the RR for the three VE devices can be seen in Table 19. A beta and a normal distribution, respectively, were assigned for the PSA.

TABLE 19 - Probability of RBC transfusion for patients undergoing cardiac surgery according to SLTs management and RR associated with VE technologies

The 95% CI reported for the mean probability of RBC transfusion suggests a normal distribution.

Although theoretically log-normal, the 95% CI reported for the RR suggests a normal distribution.

Technology	Mean value	Distribution	Distribution parameters	Source
Baseline risk of RBC transfusion in SLTs group	Base case: 0.592	Normala (probability of RBC transfusion)	µ = 0.592; σ = 0.03	See Chapter 3, Results
RR: ROTEM, TEG and Sonoclot	Base-case RR = 0.88	Normalb	µ = 0.88; σ = 0.04

Complications related to surgery and transfusion

Complications included in the model relating to surgery and/or transfusion were renal dysfunction, myocardial infarction, stroke, thrombosis (any type, such as DVT or peripheral vascular thrombosis), excessive bleeding requiring re-operation, wound complications and septicaemia. The only one of these complications evaluated by the RCTs included in the effectiveness review (see Chapter 3, Results) was re-operation to investigate bleeding. As with the probability of transfusion, we excluded two studies35,54 from this analysis, as the patients included in these studies were representative of general cardiac surgery patients. The summary estimate for the probability of re-operation from the remaining five studies46,48–51 was 0.053 (95% CI 0.029 to 0.084) (Figure 30). The summary RR for the difference in transfusion risk for patients who received VE testing compared with SLTs was also taken from the clinical effectiveness review (see Table 11).

FIGURE 30.

Forest plot showing the probability of re-operation for bleeding (95% CI) in control groups in cardiac surgery trials.

Data on the other complications were limited and we therefore assumed that there was no difference in the direct risk of having a complication between those tested with VE devices and those tested with SLTs as in Davies et al. 90 However, the risk of complications in each testing strategy was influenced indirectly by the different RBC transfusion rates associated with each strategy. The probabilities of experiencing complications related to surgery and/or transfusion, and the probability distributions for the PSA are shown in Table 20. The probability of experiencing septicaemia was sourced, as in the Scottish study,15,89 from Karkouti et al. 92 However, the population in this study92 was not representative of our population, as it included only patients who received four or more units of RBC within 1 day of surgery (i.e. patients with massive bleed). As this estimate was judged to be too high, our model used the estimate in Karkouti et al. 92 reduced by an arbitrary factor of 0.5.

TABLE 20 - Probability of experiencing a complication related to surgery and blood loss in transfused patients undergoing cardiac surgery

Davies et al. 90 report only mean values. SDs were derived assuming a 95% CI with limits deviating 20% from the mean.

Type of complication	Mean value	Distribution	Distribution parametersa	Source
Renal dysfunction	0.03	Normal	µ = 0.03; σ = 0.003	Davies (2006)90
Myocardial infarction	0.03		µ = 0.03; σ = 0.003
Stroke	0.01		µ = 0.01; σ = 0.001
Thrombosis	0.03		µ = 0.03; σ = 0.003
Excessive bleeding re-operation:
Baseline risk SLTs	0.053	Normal	µ = 0.053; σ = 0.019	See Chapter 3, Risk of bias assessment
RR VE devices	0.72	Log-normal	µ = 0.72; σ = 0.285
Wound complications	0.07	Normal	µ = 0.07; σ = 0.007	Davies (2006)90
Septicaemia	0.0207 [0.0414 from Karkouti (2006)92]	Beta	α = 38; β = 917	Karkouti (2006)92 and assumption

Patients with coagulopathy induced by trauma

The model in patients with coagulopathy induced by trauma was based on the model that we developed for patients undergoing cardiac surgery. The difference between the models relates to the surgery- and/or transfusion-related complications, which we have replaced with trauma- and/or transfusion-related complications, that is, ARDS and MOF. Where possible, we have used trauma-specific data as inputs to the model. Where these data were not available, including the impact of VE testing on the various parameters, we have used the same input values as for the cardiac surgery population.

Probability of red blood cell transfusion

We estimated the baseline risk of RBC transfusion for the SLTs group using data from the studies included in the effectiveness review (see Chapter 3, How well do viscoelastic devices predict relevant outcomes in patients with coagulopathy induced by trauma?) that reported data on the proportion of patients who received an RBC transfusion. We used a random-effects model to estimate the mean proportion of patients who received an RBC transfusion. This gave a summary estimate of 0.321 (95% CI 0.209 to 0.444) (Figure 31). As there were no data comparing the proportion of transfused patients in a trauma population who received VE testing compared with those who received SLTs, we applied the same RR as in the cardiac surgery population. These data are summarised in Table 31.

FIGURE 31.

Forest plot showing RBC transfusion rates (95% CI) in trauma patients.

TABLE 31 - Probability of transfusion for trauma patients according to SLTs management and RR associated with VE technologies

The 95% CI reported for the mean probability of RBC transfusion suggests a normal distribution.

Although theoretically log-normal, the 95% CI reported for the RR suggests a normal distribution.

This is 0.04 in the cardiac surgery model (see Table 19). We have doubled it in order to account in the PSA for the uncertainty about the assumption that the RR for the cardiac surgery population is also valid for trauma.

Technology	Mean value	Distribution	Distribution parameters	Source
Baseline risk: SLTs	0.321	Normala (probability of RBC transfusion)	µ = 0.321; σ = 0.056	Estimated
RR: ROTEM, TEG and Sonoclot	RR = 0.88	Normalb	µ = 0.88; σ = 0.08c	See Chapter 3 (Figure 8) and assumption

Mortality

At 1 month, we estimated the baseline risk of mortality for the SLTs group using the same method used to estimate the baseline risk of RBC transfusion. We identified studies included in the effectiveness review (see Chapter 3, How well do viscoelastic devices predict relevant clinical outcomes in patients with coagulopathy induced by trauma?) that reported data on 30-day or in-hospital mortality; if the time frame for mortality was not reported, it was assumed to be longer-term (i.e. up to 30 day) mortality. We then used a random-effects model to estimate the mean 1-month mortality in the SLTs group of 15.7% (95% CI 11.7% to 20.1%) (Figure 32).

FIGURE 32.

Forest plot showing overall 1-month mortality rates (95% CI) in trauma patients.

As for the cardiac patients, we aimed to assign 1-month mortality rates to transfused and non-transfused patients, such that the overall mortality rate would be equal to 15.7%. We were able to retrieve one study110 that reported mortality rates separately for transfused and non-transfused. This study included 1172 trauma patients who were admitted to an ICU, of whom 67% received a transfusion. In-hospital mortality was reported for patients who received a blood transfusion (21.4%) and those who did not (6.5%), showing that mortality was 3.3 times higher among patients who received a transfusion. It should be noted that the number of days in hospital for transfused patients was also higher (18.6 vs. 9), so the mortality rates are less easy to interpret than if mortality had been reported for a fixed time period (e.g. 30 days). The severity of the trauma was also more severe in these patients than might be seen in a general trauma population (mean ISS = 24); however, as similar data were not available in a general trauma population, this was the best estimate available.

Thus we assumed that the ratio of mortality for transfused (Mort_trans) to non-transfused (Mort_{not trans}) was 3.3. Therefore, the goal was to estimate mortality rates such that the weighted average of these yielded an overall mortality of 15.7% – the mean mortality in the SLTs group derived from the systematic review, that is, 32% Mort_trans + 68% Mort_{not trans}= 15.7%. From this, it follows that mortality was 9.1% in patients who did not receive a transfusion and 29.8% in those that did.

We then estimated mortality for the two trauma- and/or transfusion-related complications: ARDS and MOF. As none of the papers included in the systematic review reported on the incidence of ARDS or MOF and their associated mortality rates, we estimated these data from other sources. We estimated the probability of mortality in patients with ARDS from a trial in ARDS patients that reported a mortality rate of 83/385 = 21.6%. 111 We pooled data from two studies106,107 to estimate the mortality rate in patients with MOF (Table 32): a 12-year prospective study of 339 patients with post-injury MOF,106 and a prediction modelling study of 104 trauma patients of whom 21 developed MOF. 107 This yielded an overall MOF mortality rate of 26.2%.

TABLE 32 - Probability of death due to MOF

Study	No. dead	No. MOF	Mean
Dewar (2013)107	5	21	0.238
Ciesla (2005)106	90	342	0.263
Overall mean (inverse variance)			0.262
SE			0.023

One-month mortality rates for transfusion-related complications and transfusion-transmitted infections were derived when possible from the SHOT survey,93 and, as in the cardiac surgery population, it was assumed that all infections apart from bacterial contamination would only manifest themselves after 1 month, implying a zero mortality rate in the first month.

However, as we calculated earlier, the overall mortality in the transfused group had to be 29.8% in order to achieve an overall mortality of 15.7% after 1 month. The ARDS and MOF mortality that we estimated from published studies were lower than this, which would imply that having ARDS or MOF lowers the mortality rate. As this is clinically implausible, we were confronted with a consistency issue caused by using data from various studies all with slightly different populations and ways of reporting. Any way of dealing with this issue involves arbitrary choices. We made the decision that all complication mortality rates that were below the overall mortality rate for the transfused patients would become part of a calibration similar to that applied in the cardiac population. De facto this means that only the mortality rate for transfusion-associated graft-versus-host disease (which is ‘1’) was not included in the calibration. The calibration procedure itself meant that all other transfusion-related mortality parameters were set to x, and a value of x was sought such that the transfusion mortality was 29.8%.

As in the cardiac population, the 1-month mortality for each subgroup of patients in the VE group (Table 33) was assumed to be the same as in the SLTs group, implying that any mortality benefit in the VE group was due to fewer patients being transfused.

TABLE 33 - Probability of patient dying within 1 month per complication or infection (trauma population)

NA, not applicable.

SDs were derived assuming a 95% CI with limits deviating 20% from the mean.

Type of complication or infection		1 month (SLTs and VE)
		Mean value (SDa)	Source
No transfusion		0.091 (0.009)	Bochicchio (2008)110 and calibration
Transfusion and no complications		0.296 (0.030)	Calibration
MOF
ARDS
Transfusion-associated graft-versus-host disease		1	SHOT93
Incorrect blood component		0.296 (0.030)	Calibration
Haemolytic transfusion reactions	Acute
	Delayed
PTP
TRALI
Febrile reaction
Bacterial contamination
vCJD		NA	Assumption
HAV
Malaria
HTLV
HIV
HBV
HCV

For mortality of between 1 and 12 months after trauma, few data were available. One study112 was identified, which reported 3% mortality for this period. However, no information was identified on how this mortality is distributed over transfused and non-transfused patients. We therefore applied the same ratio as for 1-month mortality (3.3). Now we need to solve 32% Mort_trans + 68% Mort_{not trans} = 3.0%. This yielded a mortality in the non-transfused of 1.7% and mortality in the transfused of 5.7%. These values were assumed to apply to both the SLTs and VE group.

Probabilistic sensitivity analysis

The impact of statistical uncertainties regarding the model’s input parameters was explored through PSA. PSA results were presented in the cost-effectiveness plane for all the technologies compared. Cost-effectiveness acceptability curves (CEACs) were used to determine the probability of a strategy being considered cost-effective given a threshold incremental cost-effectiveness ratio (ICER). The probability distributions used in the PSA are listed in the tables presented above (see Model structure and methodology and Model input parameters).

Expected value of perfected information analysis

For the trauma model, we explored the value of information associated with the model uncertainty by estimating the expected value of perfect information (EVPI), which is the amount the decision-maker should be willing to pay to eliminate all uncertainty in the decision. The decision is made based on the expected net monetary benefit given current information, that is the technology with the highest expected net monetary benefit is chosen as optimal. The EVPI per patient was calculated as the average of the maximum net benefits across all PSA outcomes (expected net benefit of perfect information) minus the maximum average net benefit for the different technologies (expected net benefit given current information). Additional research might be justified when the expected net benefit for future patients, defined as the population EVPI, exceeds the expected costs of additional research. Therefore, the per-patient EVPI is multiplied by the population size to give the population EVPI. This is then summed over the lifetime for which the research recommendation is expected to be valid, discounted at 3.5% to give the net present value. 118 We selected a period of 5 years for this value. For the trauma model, a potential population of 16,825 adult patients in the UK was assumed, based on data from the National Audit Office. 119 This was calculated as follows:

• total number of major trauma (ISS ≥ 16) patients in England was approximately 20,000 per year

• number who die before they get to hospital is 2400

• proportion aged ≤ 15 years is 4.4%.

Note that this provides an upper limit of the potential population, as SLTs and VE testing will probably not be indicated for the whole trauma population. We distinguished two approaches to the population EVPI depending on whether or not the problem to be addressed was which of the four different strategies should be recommended, or whether or not to recommend VE testing (e.g. ROTEM) instead of SLTs. In the former case, all four technologies were included in the EVPI estimation. For the latter situation we compared only ROTEM, as it the most expensive strategy.

Scenario analyses

Scenario analyses were performed to investigate the influence of number of years of machine usage, number of tests performed per year, number of tests per patient, RR of be probability of transfusion, baseline prevalence of RBC transfusion, units of blood component transfused, 1-month mortality, and the probability of experiencing complications related to trauma and/or transfusion (trauma model only). We only performed these analyses for the most expensive VE device (ROTEM) as if the results were cost-effective for this device then they would also be cost-effective for the other devices (TEG and Sonoclot).

Base-case results for model in cardiac surgery patients

The base-case results from the analysis reported as LYs, QALYS and costs per technology for patients undergoing cardiac surgery are summarised in Table 37.

Under the assumptions made above (see Model structure and methodology), all of the VE technologies dominated SLTs. As the same treatment effects were assumed for each VE testing device, effectiveness (measured using LYs and QALYs) was the same for each device. The cost of Sonoclot was lower than that of ROTEM or TEG and so this device was associated with greater cost-savings (£132) than TEG (£79) or ROTEM (£43).

Note that, in general, when four strategies are compared, a full incremental analysis should be performed. As is clear from Table 38, this would result in the simple conclusion that Sonoclot dominates all other options. Such an approach would be helpful to address the question of which of the four different testing strategies should be recommended. However, if the actual question is whether or not to recommend VE testing instead of SLTs then the pairwise comparisons against SLT shown here are more informative.

The total cost of testing per patient undergoing cardiac surgery for the four technologies included in the base-case analysis was £139 for ROTEM, £103 for TEG, £78 for SLTs, and £50 for Sonoclot. Other outputs of interest from the base-case analysis were overall 1-month and 1-year mortality, the percentage of patients experiencing surgery and/or transfusion complications, the percentage of patients experiencing transfusion-related complications, the percentage of patients experiencing transfusion-transmitted infections, transfusion costs and hospitalisation costs. These are summarised in Table 38. Note that for these outputs there is no difference between the three VE devices. These results show that, compared with SLTs, the use of VE devices is associated with less mortality, a reduced probability of experiencing complications, less transfusion and fewer hospitalisation costs. The probability of experiencing transfusion-transmitted infections was very low (almost zero) in both groups but was lower in the VE group.

Results of the probabilistic sensitivity analyses in cardiac surgery patients

The impact of the statistical uncertainties in the model was investigated in the PSA. As the model assumed differences in technology costs between only the three VE technologies, the scatterplot of the PSA outcomes in the cost-effectiveness plane was not very informative (Figure 33).

FIGURE 33.

Cost-effectiveness plane with PSA outcomes for all technologies in cardiac surgery patients.

The CEACs for each technology are shown in Figure 34. PSA confirmed that SLTs is the strategy with the lowest probability of being cost-effective. This is to be expected, as the base-case scenario suggested that all three of the VE devices were both cheaper and more effective than SLTs. The CEACs for ROTEM, TEG and Sonoclot are very close together, especially at higher ceiling ratios, which would be expected, as the only difference between the three strategies assumed in the model was a difference in technology cost. At lower ceiling ratios, larger differences were observed, as Sonoclot was the cheapest technology in our model and so had the highest probability of being cost-effective.

FIGURE 34.

Cost-effectiveness acceptability curves for all technologies in cardiac surgery patients.

The information presented in Figure 34 is helpful to address the question of which of the four different testing strategies should be recommended. However, as mentioned earlier, if the actual question is whether or not to recommend VE testing instead of SLTs, then pairwise comparisons may be more informative. The deterministic pairwise results are presented in Table 37. The CEACs in Figures 35–37 illustrate the difference between ROTEM, TEG or Sonoclot and SLTs in terms of the probability of being cost-effective. At a cost-effectiveness threshold of £30,000 per QALY, the probability of cost-effectiveness for each of the three VE technologies was 0.79 for ROTEM (the most expensive device), 0.82 for TEG, and 0.87 for Sonoclot (the cheapest device). At higher thresholds, the cost-effectiveness probabilities converged to around 0.8 for all technologies.

FIGURE 35.

Cost-effectiveness acceptability curves: ROTEM vs. SLTs (cardiac surgery).

FIGURE 36.

Cost-effectiveness acceptability curves: TEG vs. SLTs (cardiac surgery).

FIGURE 37.

Cost-effectiveness acceptability curves: Sonoclot vs. SLTs (cardiac surgery).

Results of scenario analyses in cardiac surgery patients

All scenario analyses suggested that ROTEM remained cost-saving (Table 39). CEACs for all analyses (not shown) were similar to those in Figure 35. The only two exceptions were the number of tests run on each device per year, and using VE testing as an add-on to SLTs rather than a replacement. After reducing the number of tests run on each device from 500 to 200, ROTEM no longer dominated SLTs, and an ICER of £16,487 is found (see Table 39 and Figure 38). At a cost-effectiveness threshold of £30,000 per QALY, the probability of cost-effectiveness for ROTEM was 0.62. As the cost-effectiveness threshold increased, the probability of cost-effectiveness for ROTEM converged to around 0.70. We estimated, using iterative analysis, that if all other parameters in the model remain unchanged, the costs of ROTEM and SLTs would be equal if 326 tests were run on ROTEM each year. At this level the ICER would be £0. If number of tests per year is reduced to 152 then the ICER is around £30,000. When VE testing is performed in addition to SLTs rather than as a replacement, the ICER of ROTEM + SLTs compared with SLTs becomes £7487. Table 40 presents the results for the various assay combinations explored. Note that in these combinations TEG is now the more costly VE test. However, in all of the scenarios VE testing is still dominant compared with SLTs.

FIGURE 38.

Cost-effectiveness acceptability curves ROTEM vs. SLTs: scenario based on 200 tests per year.

Base-case results for model in patients with coagulopathy induced by trauma

The base-case results from the analysis reported as LYs, QALYs and costs per technology for patients with coagulopathy induced by trauma are summarised in Table 41.

All of the VE technologies dominated SLTs. As with the cardiac surgery model, the cost of Sonoclot was lower than that of ROTEM or TEG, and so this device was associated with greater cost-savings (£818) than TEG (£721) or ROTEM (£688). The total cost of testing per trauma patient for the four technologies was £203 for ROTEM, £170 for TEG, £130 for SLTs, and £73 for Sonoclot (£84). Other intermediate outcomes are summarised in Table 42.

Results of the probabilistic sensitivity analyses in patients with coagulopathy induced by trauma

The impact of statistical uncertainties in the model was investigated in the PSA. The scatterplot of the PSA outcomes in the cost-effectiveness plane (Figure 39) did not show clear preference for any one of the VE technologies.

FIGURE 39.

Cost-effectiveness plane with PSA outcomes for all technologies in trauma population.

The CEACs for each strategy are shown in Figure 40. The PSA confirmed that SLTs was the strategy with the lowest probability of being cost-effective (0.022 at most). This is to be expected, as the base-case scenario suggested that all three of the VE devices were both cheaper and more effective than SLTs. As with the cardiac surgery model, the CEACs for ROTEM, TEG and Sonoclot were very close together, which would be expected as the only difference between the three strategies assumed in the model was a difference in technology cost. At lower ceiling ratios, larger differences were observed as Sonoclot was the cheapest technology in our model.

FIGURE 40.

Cost-effectiveness acceptability curves for all technologies in trauma population.

A comparison of ROTEM with SLTs found a cost-effectiveness probability equal to 0.96 for ROTEM for a ceiling ratio equal to £0 (see CEAC in Figure 41). As the ceiling ratio increased, the CEAC for ROTEM converged to 0.87. A similar pattern was observed for TEG and Sonoclot (CEACs not shown).

FIGURE 41.

Cost-effectiveness acceptability curves: ROTEM vs. SLTs trauma population.

Results of the expected value of perfect information analysis

The population EVPI results are presented in Figure 42. This shows that, at a cost-effectiveness threshold of £30,000 per QALY, the population EVPI when all four technologies are considered was £25,017,471, whereas the population EVPI when only ROTEM and SLTs were compared was more than 22 times lower at £1,263,131. This huge difference in EVPI is to be expected, given that there is little uncertainty whether or not any one of the VE devices is superior to SLTs, but much uncertainty as to which of three devices is the optimal device. This is illustrated in the results of the PSA (see Figures 37–39).

FIGURE 42.

Population EVPI in trauma model (all technologies and ROTEM vs. SLTs only).

Clinical effectiveness

All completed RCTs identified by our systematic review were conducted in patients undergoing cardiac surgery. Pooled estimates, derived from meta-analyses of dichotomous data, indicated that VE testing (TEG or ROTEM) was associated with significant reductions in the numbers of patients receiving RBC transfusion, platelet transfusion and FFP transfusion, compared with a SLTs-based strategy. There were no significant differences between the VE testing and SLTs in terms of factor VIIa transfusion, any blood component transfusion or PCC transfusion; although data suggested a beneficial effect associated with VE testing, these outcomes were evaluated in only two studies. 35,54 There was no apparent difference in the rates of FIB transfusion between patients managed using VE testing and those managed using SLTs. Continuous data on blood component/product use, although inconsistently reported across studies, supported these findings; the only blood component/product that was not associated with a reduced volume of use in the VE testing group was FIB. There were no apparent differences in clinical outcomes (re-operation, surgical cause of bleed on re-operation and mortality) between patients managed using VE testing and those managed using SLTs. There was some evidence of reduced bleeding35,50 and ICU stay35 in the VE testing groups compared with SLTs groups, but this was not consistently reported across studies. There was no apparent difference in the length of hospital stay between groups. All meta-analyses, with the exception of factor VIIa transfusion, FIB transfusion and PCC transfusion, which included only studies of ROTEM, included both studies of TEG and studies of ROTEM; summary estimates were similar when stratified by VE device, thus, there was no evidence to indicate a difference in effectiveness between the two devices. However, it should be noted that none of the included RCTs reported a direct comparison between TEG and ROTEM.

As none of the RCTs described above evaluated the Sonoclot VE test, we included lower levels of evidence for this device. Three prediction studies that evaluated Sonoclot were included in the review,61,62,120 two of these also evaluated TEG and SLTs, enabling a direct comparison between the two devices and between VE devices and SLTs. 61,62 Data reported by the three studies in this group were not suitable for meta-analyses. All three studies61,62,120 used measures of bleeding as the reference standard or as the dependent variable in multivariable models. Positive results on conventional tests, TEG and Sonoclot were generally associated with an increased risk of bleeding with no clear differences according to test. The limited available data do not suggest a significant difference in the ability of Sonoclot and TEG to predict bleeding; however, there were insufficient data to rule out a difference in the overall clinical effectiveness of these two devices. No studies reported any data comparing Sonoclot and ROTEM.

With the exception of one small, non-RCT,65 all studies conducted in trauma patients or women with PPH included in our systematic review were prediction studies. These studies either reported the predictive accuracy of different VE device parameters and/or SLTs with a reference standard consisting of clinical outcome or measure of transfusion requirements. These studies generally found that a positive result on each of the TEG or ROTEM parameters or on SLTs was associated with an increased risk of transfusion (RBC, any blood product and massive transfusion) and death. There was no clear difference between ROTEM, TEG or SLTs. However, none of the studies provided a direct comparison between TEG and ROTEM. An overall TEG result suggesting that a patient was hypocoagulable was the strongest predictor of any blood component transfusion. The presence of hyperfibrinolysis was the strongest predictor of mortality. No studies of the Sonoclot device were identified that fulfilled inclusion criteria for the either the trauma or PPH populations.

A previous Cochrane review,21 last updated in 2011, evaluated the effectiveness of transfusion strategies guided by VE devices in patients with severe bleeding. This review21 concluded that there was no evidence that TEG or ROTEM improved morbidity or mortality and that, although transfusion strategies guided by VE devices appeared to reduce the amount of bleeding, the clinical implications of this remained uncertain. Our systematic review differs from the Cochrane review21 on a number of key points. The Cochrane review21 was not restricted to any specific clinical groups – as a result, it included one study of patients undergoing liver surgery, as well as eight RCTs of patients undergoing cardiac surgery, all of which were also included in our review. Our review represents an advance on the Cochrane review21 in that it identified three further RCTs35,53,55 conducted in patients undergoing cardiac surgery. In addition, because the Cochrane review21 was restricted to RCTs, it did not include any studies assessing Sonoclot, whereas we were able to include some limited data on this device. A key difference in approach between our systematic review and the Cochrane review21 was in the handling of continuous data. The Cochrane review21 converted median values to means in order to allow pooled estimates to be generated, even though the Cochrane handbook includes a specific recommendation that this approach should not be used; the Cochrane Handbook (section 7.7.3.6) states that ‘Ranges are very unstable and, unlike other measures of variation, increase when the sample size increases. They describe the extremes of observed outcomes rather than the average variation. Ranges should not be used to estimate SDs. One common approach has been to make use of the fact that, with normally distributed data, 95% of values will lie within 2 × SD either side of the mean. The SD may therefore be estimated to be approximately one-quarter of the typical range of data values. This method is not robust and we recommend that it should not be used’. 121 We do not believe that this approach can be justified and have therefore reported individual study results in forest plots and summarised findings in a narrative synthesis. Finally, we noted two specific errors in data extraction in the Cochrane review. 21 First, the study by Westbrook et al. 47 was included in a meta-analysis of the proportion of patients undergoing surgical re-intervention for exploration of bleeding; data from this study47 had been erroneously extracted from the baseline characteristics table that reported the number of patients in each arm who were undergoing a repeat cardiac surgical intervention. Second, meta-analyses of the proportion of patients undergoing FFP transfusion and the proportion of patients undergoing platelet transfusion, which were reported in the Cochrane review,21 included data derived from a graph reported in Nuttall et al. 50 The graph recorded the numbers of patients, in each arm of the trial, who received FFP only, platelets only, or platelets and FFP and/or cryoprecipitate;50 this means that the graph cannot be used to derive either the total number of patients who received FFP or the total number who received platelets. The Cochrane review21 appeared to have extracted the numbers of patients receiving FFP and the numbers of patients receiving platelets as though these were the total numbers of patients receiving each blood component. As more patients in the control (SLTs) arm received multiple blood components,50 this error had the effect of producing a RR which favoured the control group, a result which was in the opposite direction to all three of the other studies included in the meta-analysis. 21 A systematic review conducted for a Health Technology Assessment (HTA) report, published in 2008, included studies of VE devices in cardiac surgery, but did not restrict inclusion by study design;91 the two RCTs included in this assessment, which met the inclusion criteria for our review,48,50 were also included in both our review and the Cochrane review. 21 The Health Technology Assessment report92 concluded that, assuming 200 tests per annum, the use of VE devices appeared to be clinically effective and cost-effective, reducing the need for inappropriate transfusions, decreasing blood component requirements and reducing the number of deaths, complications and infections. The results of our systematic review are consistent with previous reviews21,91 in that they suggest that the use of VE devices may be a clinically effective approach to the management haemostasis in patients undergoing cardiac surgery.

We are not aware of any previous systematic reviews assessing the effectiveness of VE devices for the management of patients with trauma-induced coagulopathy or PPH. A Cochrane Diagnostic Test Accuracy protocol has recently been published with the title ‘Thromboelastography (TEG) and thromboelastometry (ROTEM) for trauma-induced coagulopathy in adult trauma patients with bleeding’. 22

Cost-effectiveness

We assessed the cost-effectiveness of VE devices in two key populations: patients undergoing cardiac surgery and patients with trauma acquired coagulopathies. There were insufficient data from the clinical effectiveness review to construct a model to assess the cost-effectiveness of VE devices in women with PPH. There were no data on the clinical effectiveness of Sonoclot; we therefore assumed that the TEG- and ROTEM-based estimates used in the model would also be applicable to Sonoclot; thus the same health effect estimates were used for all three VE devices.

The cost-effectiveness model suggested that VE testing is cost-saving and more effective than standard laboratory testing in cardiac surgery patients. The per-patient cost-saving was slightly smaller for ROTEM (£43) than for TEG (£79) and Sonoclot (£132). This finding was entirely dependent on material costs, which are slightly higher for ROTEM in the base case. When other combinations of assays are assumed, TEG could be slightly less cost-saving than ROTEM. When all of the uncertainties included in the model were taken into account, at a cost-effectiveness threshold of £30,000 per QALY, the probability of cost-effectiveness for each of the three VE technologies was 0.79 for ROTEM (the most expensive device), 0.84 for TEG and 0.87 for Sonoclot (the cheapest device). At higher thresholds, probabilities converged to around 0.8 for all technologies. Scenario analyses were used to assess the potential impact of changing various input values for the model. In these scenarios the results remained largely unchanged. Only when the number of tests performed per machine per year was VE testing was no longer cost-saving when the number of tests performed per machine was fewer than 326. When this number was 152, the ICER was around £30,000.

For the trauma population, the per-patient cost-savings due to VE testing were more substantial, amounting to £688 for ROTEM compared with SLTs, £721 for TEG and £818 for Sonoclot. A comparison of the most expensive technology, ROTEM, with SLTs found a cost-effectiveness probability equal to 0.96 for ROTEM for a ceiling ratio of £0. As the ceiling ratio increased, this probability converged on 0.87. The increased cost savings observed for the trauma compared with the cardiac population were primarily due to the higher blood volumes that are transfused in the trauma patients. Scenario analyses constructed to assess the impact of various parameters showed similar results. Given the lack of effectiveness data in trauma patients, the current results should be regarded as indicative of the potential cost-effectiveness of VE testing only in trauma patients.

Clinical effectiveness

Extensive literature searches were conducted in an attempt to maximise retrieval of relevant studies. These included electronic searches of a variety of bibliographic databases, as well as screening of clinical trials registers and conference abstracts to identify unpublished studies. Because of the known difficulties in identifying test accuracy studies using study design-related search terms,122 and potential need to include non-RCTs and prediction modelling studies, search strategies were developed to maximise sensitivity at the expense of reduced specificity. Thus, large numbers of citations were identified and screened, many of which did not meet the inclusion criteria of the review.

The possibility of publication bias remains a potential problem for all systematic reviews. Publication bias was not formally assessed in this review because, for RCTs, the number of studies was too small for such an assessment to be meaningful and, for prediction studies, there is no reliable method of assessing publication bias. However, our search strategy included a variety of routes to identify unpublished studies and resulted in the inclusion of a number of conference abstracts and the identification of one ongoing RCT. 62 Considerations may differ for systematic reviews of test accuracy studies. It is relatively simple to define a positive result for studies of treatment, for example a significant difference between the treatment and control groups which favours treatment. This is not the case for test accuracy studies, which measure agreement between index test and reference standard, or prediction modelling studies, which measure the extent to which a particular test result is predictive of outcome(s) once other potentially predictive variables have been adjusted for. However, it would seem likely that studies finding greater agreement between the index test and reference standard (high estimates of sensitivity and specificity) or that the index test is a significant, independent predictor of outcome will be published more often.

Clear inclusion criteria were specified in the protocol for this review and the one protocol modification that occurred during the assessment has been documented in the methods section of this report (see Table 6). The eligibility of studies for inclusion is therefore transparent. In addition, we have provided specific reasons for excluding all of the studies considered potentially relevant at initial citation screening (see Appendix 4). The review process followed recommended methods to minimise the potential for error and/or bias;37 studies were independently screened for inclusion by two reviewers, and data extraction and quality assessment were undertaken by one reviewer and checked by a second (MW and PW). Any disagreements were resolved by consensus.

Studies included in this review were assessed for risk of bias using published tools appropriate to study design and/or the type of data extracted. Studies that provided data on the accuracy of VE testing to predict clinical outcomes and/or transfusion requirements were assessed using the QUADAS-2 tool. 45 QUADAS-2 is structured into four key domains covering participant selection, index test, reference standard and the flow of patients through the study (including timing of tests). Each domain is rated for risk of bias (low, high or unclear); the participant selection, index test and reference standard domain are also separately rated for concerns regarding the applicability of the study to the review question (low, high or unclear). Although designed specifically for this type of study, QUADAS-2 was also considered the best option for assessment of the prediction modelling studies. This was because the prediction modelling studies included in this assessment are unusual in that they generally present the results of several multivariable models for each outcome/dependent variable; a separate model is needed for each VE testing parameter or SLTs, as parameters and tests frequently measure the same or similar coagulation properties and cannot be considered independent. In addition, studies aimed to assess the ability of individual VE testing parameters or SLTs to predict the occurrence of very short-term outcomes. For these reasons, the studies were considered to have more in common with diagnostic accuracy studies than with classic prognostic/prediction modelling studies. RCTs were assessed using Cochrane’s tool for assessing risk of bias in randomised trials. 44 The results of the risk of bias and QUADAS-2 assessments are reported, in full, for all included studies (see Appendix 3) and in summary in the results [see Chapter 3, Risk of bias assessment, How do clinical outcomes differ among patients with coagulopathy induced by trauma who are tested with viscoelastic devices compared with those who are not tested? (Risk of bias and applicability assessment), How well do viscoelastic devices predict relevant clinical outcomes in patients with post-partum haemorrhage? (Risk of bias and applicability assessment); Tables 8, 16 and 18; and Figures 5 and 17].

Although we identified 11 RCTs35,46–57 that compared the effectiveness of VE testing with a SLTs-based approach for the management of haemostasis in patients undergoing cardiac surgery, the potential to produce summary effect estimates was limited by the wide variety of outcomes reported and a lack of standardisation of the way in which these were measured. The assessment of heterogeneity was limited by the relatively small number of studies that contributed to each meta-analysis. The Q-statistic has limited power to detect heterogeneity when the number of studies included in a meta-analysis is low. No summary estimates of continuous data (e.g. duration of hospital or ICU stay, or volume of blood component/product transfused) were possible as the majority of these data were appropriately reported as medians with range or IQR. Pooling only those studies that reported continuous outcomes as mean ± SD would be unrepresentative of the group as a whole, and would be likely to result in greater weight being given to studies that reported data as mean ± SD without consideration of whether or not these data were normally distributed.

At the start of this assessment the role of VE testing in the care pathway was considered to be unclear; it could be used either as an add-on to, or replacement for SLTs. Three of the RCTs included in our systematic review compared the effectiveness of VE testing combined with SLTs (two studies using TEG50,51 and one using ROTEM55) to SLTs alone, that is, these studies provided data on the add-on value of VE testing. For all outcomes assessed, the results of these studies were consistent with those of studies that compared VE testing alone with SLTs. These findings indicate that performing SLTs in addition to VE testing is unlikely to give further benefit over that provided by VE testing alone. VE testing can therefore be regarded as a replacement for SLTs.

All of the studies conducted in trauma patients or women with PPH included in this review have considerable limitations with respect to their ability to address the overall aim of assessing the clinical effectiveness of VE devices for assessment of haemostasis in these patient groups. With the exception of one small, non-RCT,63 all studies in these patient groups were prediction studies, which either reported the predictive accuracy of different SLTs and/or VE device parameters for which the reference standard was a clinical outcome or measure of transfusion requirements, or the results of prediction models, for which each test or parameter was modelled separately, as described above, with clinical outcome or transfusion requirement as the dependent variable. When the reference standard or dependent variable in the model was a measure of transfusion requirements, it is not possible for studies to be rated as both ‘low risk of bias’ and ‘low applicability concerns’ with respect to the reference standard. This is because, in order for such a study to reflect clinical practice and be rated ‘as low applicability concerns,’ the decision to transfuse would need to be made with knowledge of the test results; however, there is then an inevitable risk of incorporation bias leading to a rating of ‘high risk of bias.’ The need for a separate model for each VE testing parameter or SLTs, as described above, creates a further problem in that prediction studies cannot adequately assess the overall predictive performance of VE devices compared with SLTs as they would be used in practice. Finally, any type of prediction study is suboptimal, in that these studies can only ever provide an indication of the ability of VE testing or SLTs to predict clinical outcomes or transfusion requirements. These studies cannot provide information on how interventions and subsequent clinical outcomes may differ according to whether or not a POC VE testing or SLTs-based strategy is used; these data can be derived only from controlled trials.

Cost-effectiveness

Our study can be regarded as an important update of the cardiac surgery aspect of the evaluation undertaken for NHS Scotland,12 and is the first cost-effectiveness analysis of VE devices in trauma patients. It was informed by an up-to-date, high-quality systematic review that included a number of RCTs published since the NHS Scotland evaluation. 12 We also added a PSA to the model in order to assess the simultaneous impact of the various uncertainties. A further strength of our model is that we included longer-term mortality data than those that were included in previous evaluations, which included mortality up to only 1 month. Our cardiac surgery model used data based on a large study by Murphy et al. ,15 which showed that the effects of transfusion on mortality continued up to and beyond 1 year. Similar data were not available for the trauma population. We therefore had to make some assumptions for this population. We extrapolated the ratio of mortality in transfused to non-transfused patients found in a study that provided this information up to hospital discharge to 1-year follow-up and then applied these data to the overall mortality rate for this period from another study in trauma patients. It would be expected that a RR at hospital discharge is too high at 1 year; the study in cardiac patients showed that the difference in mortality between transfused and non-transfused patients decreased over time. Scenario analyses showed that changing the ratio of mortality in transfused patients compared with non-transfused patients did not affect results. We might reasonably assume that, given that mortality is low between 1 month and 1 year, this would also be the case if we had made similar changes to 1-year mortality.

The main outcome used in the economic models was the proportion of patients at risk of RBC transfusion. From this, it was possible to impute other effects, such as units of blood transfused, adverse events, complications, changes to quality of life and overall survival. This is consistent with the only cost-effectiveness study in the field, the Scottish HTA report. 12,89 It is also consistent with the study by Davies et al. ,90 on which the Scottish HTA report12 was based, in which costs and effects of methods of minimising perioperative allogeneic RBC transfusion were assessed for cardiac patients as a subpopulation. In order to estimate the mortality for VE testing, we assumed that any mortality benefit from VE testing resulted only from fewer patients receiving an RBC transfusion. However, literature suggests that both whether or not a patient is transfused, and the number of units transfused, may be predictive of mortality. 123 Thus it is possible that differential mortality between VE and SLTs could result from reasons other than simple differential rates of transfusion, such as reduced volume transfused or differential transfusion of other blood components, for example FFP and platelets. This could potentially have resulted in an underestimation of the benefits of VE testing on mortality than currently accounted for in the model. Conversely, the mortality estimate was derived from the study by Murphy et al. ,15 which was based on observational data. It may therefore be suggested that differences in mortality between transfused and non-transfused could have arisen because of confounding, as those patients who were transfused may have been more ill than those who were not. This could have had the effect of biasing the model in favour of VE testing. However, Murphy et al. 15 argue against this explanation as prognostic factors were well balanced across transfused and non-transfused groups, mean nadir HCT was similar between groups, adjustment for confounding did not alter observed effects, and effects were observed within strata of patients with and without important risk factors. Thus we consider it reasonable to assume that any bias we introduced by using the observed mortality data for our model is limited. This decision is supported by the fact that the RR of mortality for ROTEM and TEG was 0.90 in our model, which was almost identical to the RR estimated in the systematic review of RCTs (0.87).

A strength of our study was the detailed consultation with manufacturers regarding the costs of each VE device. This was important as each device is available with different numbers of channels and runs different assays, which are not directly comparable between devices. We decided which assays and number of tests to model based on the combination of assays and numbers of tests used in the trials so that the costs included in the model correspond to the source of the effectiveness data. However, it is unclear whether or not the results found in the trials would also be applicable to different assay combinations and numbers of tests used in clinical practice. We found that varying the number of tests, which could also be a proxy for assay combinations, did not alter the conclusions in terms of cost-effectiveness. The length of time that a machine is used for and the average number of tests run per machine per year influences the material cost of a test. However, scenario analysis showed that the number of tests had to be very low before VE testing was no longer cost-effective.

A major limitation of both models was the lack of data on the effectiveness of the Sonoclot device. None of the RCTs included in our review assessed this device. As the only difference in the models was the costs of the devices, and Sonoclot was the cheapest device, Sonoclot was the most likely to be cost-effective. However, this should be interpreted with extreme caution as a result of the lack of evidence.

There were no data on the clinical effectiveness of any of the VE devices in trauma patients. We therefore assumed equivalent clinical effectiveness to the cardiac surgery population. Clinical experts were consulted regarding their views on the validity of this assumption. They indicated that patients undergoing (elective) cardiac surgery are likely to differ from trauma patients, which may affect the relative effectiveness of the VE devices. Specifically, it was noted that trauma patients are likely to have higher blood loss and therefore have greater blood transfusion requirements. We were able to estimate the baseline risk of RBC transfusion in trauma patients from the predictive accuracy studies included in the systematic review, but these studies could not inform the RR of transfusion in patients who were and were not tested with a VE device. There was general agreement that an assumption of equivalent clinical effectiveness in terms of the RR of RBC transfusion between the cardiac surgery and trauma populations was a reasonable assumption, given the lack of other reliable data. Although this assumption may be clinically problematic, scenario analysis indicated that if the RR of RBC transfusion was as high as 0.98, VE testing would still be cost-saving in this population. This compares with a value of 0.88 derived from the systematic review of cardiac surgery patients and used in the base-case analysis.

The 1-year time horizon used by our model could be regarded as a further limitation. However, we would argue that extrapolation over a longer time horizon is unnecessary. This is because at 1 year all VE devices were shown to be both more effective and cheaper than SLTs, and with little uncertainty (probabilities of at least 0.68 of being cost-effective); effectiveness would only increase and costs would be likely to decrease over a lifetime. The expected increase in effectiveness is based on the avoidance of transfusions supported by Murphy et al. ,15 who showed that transfusion continues to increase mortality beyond 1 year. In addition, long-term complications, such as stroke, which are likely to be avoided by fewer transfusions, would also imply lower cost.

Where possible, we used cardiac surgery and trauma-specific utility and cost estimates in our models. However, for some of the short-term utility parameters we were unable to find trauma-specific data. We made the conservative assumption that during the first month trauma patients would have the same utility as that of cardiac surgery patients. Given that many trauma patients spent quite some time on an ICU, often being ventilated, the true utility is likely to be lower. In addition, we had no good data on costs of a hospital stay once trauma patients leave the ICU. This is related to the fact that these patients may go to a wide variety of departments, depending on the type of trauma (e.g. brain trauma or mainly orthopaedic trauma). We therefore made the assumption that costs per day would also be the same as for cardiac patients; it was unclear whether or not this was likely to be an overestimation or underestimation. However, given that these utilities and costs apply to only a very short time period, they are unlikely to have influenced whether or not VE testing was cost-effective.

We conducted an EVPI for the trauma population, as we felt that there less evidence and therefore greater uncertainty for this population. This showed that it may be worth spending money on further primary research given that, when comparing all four technologies (ROTEM, TEG, Sonoclot and SLTs) the population EVPI was around £25M for an ICER of £30,000. However, the EVPI should be interpreted with caution, given that the value when comparing only a single VE device (ROTEM) with SLTs was 22 times lower at just over £1.25M. This would suggest that there is relatively little uncertainty as to whether or not ROTEM would be cost-effective in comparison with SLTs. This is inconsistent with the evidence, as the data to inform the trauma model was derived from trials conducted in cardiac surgery patients. The full uncertainty associated with this limitation, as well as other assumptions, may not have been captured by this analysis.

Clinical effectiveness

The results of our systematic review are consistent with previous reviews,21,89 in that they suggest that the use of VE devices may be a clinically effective approach to the management haemostasis in patients undergoing cardiac surgery. Our results indicate that the use of VE devices may be associated with a reduction in transfusion rates; however, whether or not this reduction represents a decrease in inappropriate transfusions and whether or not it translates into changes in important clinical outcomes (e.g. duration of ICU/hospital stay, morbidity and mortality) remains less clear. Studies included in our review provided some indication that the use of VE devices may be associated with a reduction in the duration of ICU stay. However, data were not considered suitable for meta-analyses and only one study35 showed a statistically significant decrease in the length of ICU stay for patients managed using an algorithm based on a VE device compared with those managed using an algorithm based on SLTs; this study35 restricted inclusion to patients who were bleeding from capillary beds or had blood loss > 250 ml/hour or 50 ml/10 minutes.

The existence of a link between the use of VE devices and clinical outcome is even more uncertain when these devices are used in the management of trauma patients or women with PPH. Studies in trauma patients or women with PPH included in our review consistently indicated a link between a positive test result (VE device of SLTs) and transfusion outcomes or mortality. However, we did not identify any completed RCTs in these patient groups, although we did identify one ongoing RCT62 (recruitment has reached 105 participants out of a target of 120) and additional information on this study was provided by the authors in the form of the study protocol. As described in the above (see Strengths and limitations of assessment), prediction studies cannot provide information on how interventions and subsequent clinical outcomes may differ according to whether or not a POC VE testing or SLTs-based strategy is used. Further, in contrast with the RCTs conducted in patients undergoing cardiac surgery, they cannot provide data on how transfusion rates may differ according to whether or not the decision to transfuse is based on the use of a VE device or on SLTs. Our systematic review included one small (n = 50) controlled clinical trial63 that compared the effectiveness of an ‘institutional massive transfusion protocol’ (details not reported) to a TEG-guided protocol (details not reported) for the management of trauma patients. This study63 was published only as a conference abstract and no numerical data were reported; however, the results section stated that there were no statistically significant differences, or trends towards differences, between groups in mortality, ARDS, SIRS, MOF, sepsis, cardiovascular events, or duration of hospitalisation; a trend towards reduced pneumonia, reduced days on ventilation and reduced duration of ICU stay in the TEG-guided group was reported. 63 We did not include studies of VE devices with a historical control group in our review, as it is not possible to attribute any observed differences between groups in these studies solely to the introduction of the VE device. One such study,124 from a German level I trauma centre, reported reductions in the annual use of transfusion products from 2002 to 2010 (PRBC –33%, FFP –79%, platelet concentrates –65%) following the introduction of an algorithm for coagulation management in trauma patients based on POC ROTEM combined with calculated goal-directed therapy with FIB and PCC; the number of study participants was unclear, but approximately 250 trauma patients per year were treated in the emergency room. The study protocol provided by the authors of the ongoing trial64 also reported the results of a before-and-after study, conducted in their institution. Although much smaller than the German study, this study64 had the advantage of assessing two immediately consecutive populations: before (n = 34) and after (n = 34) rapid TEG was added to the institution’s massive transfusion protocol; unlike the German study, this implies that rapid TEG was the only change to management strategy. Results from this study64 indicated that patients managed with a protocol that included rapid TEG had more effective resuscitation than those managed using the standard massive transfusion protocol; median improvement in lactate from presentation to 6 hours was 2% for the standard massive transfusion protocol and 44% for the standard massive transfusion protocol + rapid TEG, and median improvement in pH from presentation to 6 hours was 1% for the standard massive transfusion protocol and 2% for the standard massive transfusion protocol + rapid TEG. 125 Rates of transfusion of all blood components were consistently less after the introduction of rapid TEG, but differences did not reach statistical significance. 125 Finally, mortality fell from 65% to 29% (p = 0.04) after the introduction of rapid TEG. 125 Taken together, the results of these studies could be considered to indicate that further investigation of the clinical utility of VE devices in trauma patients and women with PPH is warranted.

There is currently a lack of adequate information on the potential role of VE devices in the early detection of hyperfibrinolysis and any consequent effects on clinical outcomes, and this is an area that may particularly warrant further investigation. Data from the CRASH-2 trial indicate that greatest survival benefit from antifibrinolytic therapy in trauma patients is seen with very early (< 1 hour after injury) intervention. 126 There are also some published data indicating that the risk of death from bleeding increases at levels of clot lysis < 7.5% (at 30 minutes post maximum clot strength) which is generally regarded as normal. 127,128 The ROTEM FIBTEM assay and the TEG functional FIB assay use a reagent that is specific for the fibrin polymerisation process, which declines more rapidly than FIB levels as measured in the laboratory. 129 This adds the potential to detect the pathology at an earlier stage in its evolution to the time gained from using POC testing compared with laboratory-based testing. 130,131 A small observational study, which did not meet the criteria for inclusion in our systematic review, reported that primary fibrinolysis, as diagnosed by TEG, occurred < 1 hour post injury in 18% of a series of severely injured patients requiring massive transfusion, and was associated with increased blood component requirements, coagulopathy and haemorrhage-related death. VE devices therefore have the potential to provide a sufficiently timely and sensitive method of detecting fibrinolysis to enable optimally effective intervention. FIB is also thought to play a major role in the evolution of PPH and can be an early predictor of severity;129 however, data in this population are even more sparse than for trauma. Neither of the two PPH studies included in review82,83 reported hyperfibrinolysis as an outcome, although one83 did evaluate the ROTEM FIBTEM assay.

The extent to which VE devices may be considered to be clinically equivalent remains uncertain. As outlined in the background section of this report (see Tables 2–4), the range of parameters measured differs between the three devices included in this assessment (TEG, ROTEM and Sonoclot). Despite these differences, the available data provide no strong evidence of a difference in clinical effectiveness between TEG and ROTEM; however, it should be noted that there is no strong evidence that the devices are equivalent, as there were no studies providing a direct comparison between the two devices. Data on Sonoclot were very sparse, limited to three studies59,60,120 in the cardiac surgery population; data from two of these studies,59,60 which provided a direct comparison with TEG, did not suggest a significant difference in the ability of the two devices to predict bleeding.

Issues of training requirements and implementation are outside the scope of this assessment; however, a 2010 published report132 of studies undertaken by the UK National External Quality Assessment Scheme (NEQAS) for Blood Coagulation on the use of TEG and ROTEM devices in operating theatres has indicated that there may be some areas of concern. The published article132 reported the results of a series of four quality assurance studies, with up to 18 TEG users and 10 ROTEM users involved in testing two samples per study. The samples were normal plasmas, factor VIII or XI deficient samples, or normal plasmas spiked with heparin. The precision of the tests varied greatly for both devices, with coefficients of variances ranging from 7.1% to 39.9% for TEG and 7.0% to 83.6% for ROTEM. 132 Some centres returned results that were judged to be sufficiently different from those obtained by other participants to predict alterations in patient management decisions. 132 Based on these findings, it would seem that staff training requirements are likely to be an important consideration for the implementation of these devices. A UK study,133 published in 2009, compared users’ experience of TEG and ROTEM over a 1-week period; the study133 included seven consultant anaesthetists, one consultant haematologist, one associate specialist anaesthetist and two senior trainee anaesthetists, all of whom were trained by the manufacturers. The summary of the opinions of study participants suggested that the TEG training programme was preferred, and that better service support was provided for this device. 133 However, this is a very small study133 and may not be reflective of current experience in the NHS.

Cost-effectiveness

Substantial uncertainties around the cost-effectiveness of VE devices for the identification and management of coagulopathies remain, particularly with respect to the trauma population. The main uncertainties in the cost-effectiveness analyses follow directly from those described for the review of clinical effectiveness. Uncertainties are caused by lack of clinical effectiveness data for Sonoclot in the cardiac surgery population, and by a lack of clinical effectiveness data for any of the VE devices in the trauma and PPH populations. Once the results of the ongoing RCT,62 and any future RCTS, in the trauma population become available, our trauma model can readily be updated.

Other uncertainties pertain particularly to the trauma patients. As well as a requirement for data on the clinical effectiveness of VE testing in this population, this also includes data on trauma-specific costs and utilities. The influence of RBC transfusion on longer-term mortality (beyond in hospital mortality) in trauma patients is also unclear.

Randomised controlled trial searches

Post-partum haemorrhage searches

Trauma searches