Near‐patient coagulation testing to predict bleeding after cardiac surgery: a cohort study

Abstract Essentials Near‐patient testing improves coagulopathy diagnosis in cardiac surgery patients with severe bleeding. We investigated how well pre‐emptive near‐patient testing predicted severe bleeding. Severe bleeding could be predicted using both near‐patient tests and patient clinical characteristics. Near‐patient test results gave little additional predictive value over clinical characteristics alone. Background Coagulopathic bleeding is common after cardiac surgery and is associated with increased morbidity, mortality and healthcare costs. Implementation of blood management algorithms in which patients with severe bleeding undergo near‐patient coagulation testing results in less overall bleeding and transfusion. However, it is unknown whether there is additional value from pre‐emptive near‐patient testing to predict whether severe bleeding will occur. Objectives To evaluate how well a comprehensive panel of 28 near‐patient platelet and viscoelastometry tests predict bleeding after cardiac surgery, compared to prediction using baseline clinical characteristics alone. Methods Single‐center, prospective cohort study in adults undergoing a range of cardiac surgery procedures. The primary outcome was clinical concern about bleeding (CCB), a composite of high blood loss (chest drain volume >600 mL within 6 hours), re‐operation for bleeding or administration of a pro‐haemostatic treatment directed by clinician judgement. Results In 1833 patients recruited between March 2010 and August 2012, the median number of abnormal near‐patient test results was 5/28 per patient (range 0‐18). CCB occurred in 449/1833 patients (24.5%). The c‐statistic for a predictive model for CCB using only baseline clinical characteristics (baseline‐only model) was 0.72 (95% CI 0.69‐0.75). Addition of near‐patient test results to this model (baseline‐plus‐test model) improved the prediction of CCB (c‐statistic 0.75 [0.72‐0.77]), but increased the number of correctly classified patients by only 18 (0.98%). Conclusions Near‐patient coagulation testing predicts bleeding in cardiac surgery patients, but offers little improvement in prediction compared to baseline clinical characteristics alone. trial registration: ISRNCTN 20778544 (http://www.isrctn.com/).


| INTRODUCTION
Severe bleeding caused by coagulopathy is common after cardiac surgery and frequently requires large volume red blood cell (RBC) transfusion (>4 units) or emergency re-operation. 1,2 Severe bleeding, RBC transfusion and re-operation are independently associated with organ failure, sepsis and death. [3][4][5] The provision of RBC and other blood components for cardiac surgery patients also has significant health care costs, and accounts for 10-15% of the UK blood supply. 6,7 Near-patient coagulation testing using viscoelastometry or rapid platelet function analysers detects the common sub-types of coagulopathy associated with cardiac surgery in a clinically useful timescale. 8,9 Several small, single-center, randomized controlled trials (RCTs) in cardiac surgery have shown that when compared to conventional laboratory testing or clinician judgement alone, near-patient testing reduced transfusion of RBC 10,11 or non-RBC blood components. [10][11][12][13][14][15] In a multicenter RCT of more than 7400 patients, implementation of a blood management algorithm incorporating near-patient tests, resulted in reduced RBC transfusion, platelet transfusion and overall bleeding. 16 In this trial, near-patient test results were used to direct targeted treatments for coagulopathy only in patients who had already developed severe bleeding. 16 Previous studies have also shown that some near-patient test results from before the start, or immediately after the end of cardiac surgery also enable prediction of postoperative bleeding. [17][18][19][20][21][22][23] This suggests an alternative blood management strategy in which nearpatient testing is performed pre-emptively before the highest risk period for severe bleeding in the immediate post-operative period. This is an attractive clinical strategy since identification of patients at the highest risk of bleeding could potentially enable selective targeted treatments to prevent severe bleeding starting. Predictive near-patient testing has been incorporated into several blood management algorithms evaluated in several previous RCTs, usually as a single step for selection of preventative treatments 10,12,13,15 or in combination with later diagnostic near-patient testing in patients who develop bleeding despite preventative treatments. 11 Blood management algorithms incorporating near-patient tests are recommended in US and European practice guidelines [24][25][26] , and are used widely. 27 However, there is poor consensus about the best algorithm design, particularly whether near-patient tests should be performed in response to severe bleeding, or whether there is additional value in pre-emptive testing to help prevention of bleeding.
We have performed a large prospective observational cohort study (Coagulation and Platelet Laboratory Testing in Cardiac Surgery [COPTIC] study; ISRCTN 20778544) to evaluate coagulation testing in a range of cardiac surgery procedures. In order to clarify the role of near-patient testing, we now report an analysis of how well nearpatient tests predict bleeding, compared with prediction using patient clinical and procedural characteristics alone.

| Study design and patients
The COPTIC study was a single center, observational cohort study in which patients undergoing cardiac surgery at the Bristol Heart Institute were recruited in accordance with a pre-specified protocol and in accordance with a UK NHS Research Ethics Committee approval (09/H0104/53). All patients aged over 18 years undergoing any non-emergency cardiac surgical procedure were eligible unless they were prisoners or were unable to consent due to mental incapacity.

| Care of patients and classification of prohaemostatic treatments
All participating patients gave written consent before surgery and were managed using standard anesthetic and surgical care pathways.
Protamine (1 mg per 100 units of heparin) was given to reverse heparin anticoagulation immediately at the end of surgery. For procedures other than off-pump coronary artery bypass grafting (CABG), patients received anti-fibrinolytic drugs and additional protamine after the return of heparinized blood from the cardiopulmonary bypass circuit.
These pro-haemostatic treatments were classified as directed by standard care because the decision to treat was made before the start of surgery. Some patients also received pro-haemostatic treatment with fresh frozen plasma, cryoprecipitate, platelets, recombinant Factor VIIa, fibrinogen concentrate, and additional anti-fibrinolytic drugs or protamine because severe bleeding was judged to have started. These treatments were classified as pro-haemostatic treatments by clinical judgement.

| Baseline characteristic and near-patient testing predictors
Patient clinical characteristics, surgical procedure and the results of conventional laboratory tests from pre-operative assessments were recorded from electronic hospital records. Near-patient tests were performed on a 'pre-operative sample' obtained at induction of Essentials • Near-patient testing improves coagulopathy diagnosis in cardiac surgery patients with severe bleeding.
• We investigated how well pre-emptive near-patient testing predicted severe bleeding.
• Severe bleeding could be predicted using both near-patient tests and patient clinical characteristics.
• Near-patient test results gave little additional predictive value over clinical characteristics alone. anesthesia and on a 'post-operative sample' obtained at the end of surgery, after protamine for heparin reversal but before chest closure and insertion of chest drains.

| Outcomes
The primary outcome was clinical concern about bleeding (CCB) after cardiac surgery, defined as a composite of any of the following: (i) a chest drain volume greater than 600 mL at 6 hours after admission to the cardiac intensive care unit (CICU); (ii) any re-operation for bleeding during the hospital stay in which a surgical cause of bleeding was not identified; or, (iii) any pro-haemostatic treatment by clinical judgement from the time of the post-operative blood sample until 12 hours after CICU admission. Pro-haemostatic treatments directed by clinical judgement were included in the primary outcome because these are the only reliable indicator of severe bleeding that occurs: (i) after the end of surgery but before chest drain insertion, or, (ii) after chest drain insertion but which is successfully reversed before the 600 mL chest drain volume threshold is reached.
The secondary outcomes were RBC transfusion, myocardial infarction (MI), stroke, acute kidney injury (AKI), sepsis and mortality (Table S1). The study observation period was the duration of hospital admission.

| Selection of predictors
The baseline characteristics age, sex, diabetes, type of procedure, antiplatelet drugs, surgical priority, estimated glomerular filtration rate, haematocrit, platelet count, and body mass index were selected as candidate predictors of CCB, before generation of predictive models (Table S2). Since a large number of different surgical procedures were performed in the study population, patients were classified into 12 categories depending on the type of procedure (CABG, CABG+valve, valve only, other high risk bleeding procedure) and pre-operative antiplatelet drugs (no anti-platelet drugs, aspirin alone and aspirin+P2Y 12 blocker sub-grouped according to duration of omission of P2Y 12 blocker before surgery: Table S2).
The candidate near-patient test predictors were a panel of 28 results from pre-operative or post-operative MEA platelet function analysis or from post-operative ROTEM or TEG viscoelastometry (Table S3) after cardiac surgery. 8,9 The MEA platelet function analyzer ADP-test and ASPI-test results measures platelet dysfunction associated with P2Y 12 blockers and aspirin, respectively. 28 The EXTEM maximum clot firmness -FIBTEM maximum clot firmness and TRAP-test results were selected to measure global platelet dysfunction function. 11,16 Test results were incorporated into predictive models for CCB as continuous variable. However, in order help describe the distributions for the test results, each result was also classified as above or below a 95% reference interval obtained by from 42 healthy volunteers (median age 48 years, 68% male), determined locally using the same analyzers as the main study.

| Statistical analysis
In order to compare predictive models that incorporated near-patient test results with alternative models that included the baseline characteristics, the analysis population was defined as all patients with complete data for all predictors. Bias due to this constraint was investigated by calculating standardized mean differences (SMD) 29 to compare the analysis population with those excluded because of missing data.
Logistic regression was used to develop predictive models and to estimate associations between CCB or secondary outcomes and the baseline characteristics (baseline-only model) or alternative models that also included near-patient test results (baseline-plus-test models).
The near-patient tests were further evaluated post hoc, with a model that included only the best-fitting near-patient test results (test-only model). The best baseline-only, baseline-plus-test, and test-only models were selected as the models with the highest c-statistic. Predictive value was also expressed as the proportion of patients correctly classified as CCB or no CCB, where those with a predicted probability of CCB ≥0.5 were classified as CCB. For all models, multivariable fractional polynomial techniques were used to investigate the linearity of terms. Model fit was assessed with Hosmer-Lemeshow goodnessof-fit tests and individual contributions to the models were evaluated using likelihood ratio tests.
For the best baseline-plus-test model, three sensitivity analyses considered alternative formulations of the primary outcome (Table S4). In the first two sensitivity analyses, patients who were classified as CCB solely because of a pro-haemostatic treatment by clinical judgement were either (i) excluded from analysis (SA1); or (ii) reclassified as no CCB (SA2). In a third post hoc sensitivity analysis, patients classified as CCB solely because they were transfused with 1-2 units of plasma or with 1 unit of platelets, were reclassified as no CCB (SA3).
The c-statistics from the primary outcome models were internally validated by bootstrapping with 1000 replicates and cross-validated by removing one observation and then using the remaining analysis population to create models that generated the predicted probability of CCB for that one observation. After repeating the process for each observation, the predicted probabilities were used to build receiver operator characteristic curves and to calculate c-statistics. All analyses were performed in STATA (version 14.0; STATA Corp, College Station, TX, USA).  Table 1). Compared to the analysis population, the 708 patients excluded due to missing data had a smaller proportion of patients who underwent CABG with aspirin (31.7% vs 40.2%; SMD 0.18) and a higher proportion who underwent valve replacement with aspirin (10.3% vs 7.3%; SMD 0.11). For all the other baseline characteristics, the SMDs were less than 0.10, indicating that the groups were similar (Table S5). 30 The primary outcome of CCB occurred in 449 (24.5%) of the analysis population, with 182 (9.7%) patients having more than one qualifying component. Considering the components separately, 362 (80.6%) of patients with CCB received a pro-haemostatic treatment by clinical judgement, 244 (54.3%) had a chest drain volume greater than 600 mL and 57 (12.7%) had re-operation for bleeding (Fig. S1).

| Near-patient coagulation test results
When compared to a 95% reference interval (RI) from a group of healthy controls not receiving anti-platelet drugs, the analysis population had an overall median of 5 abnormal test results per patient (range 0/28-18/28

| Prediction of CCB using baseline characteristics
Baseline characteristics relating to sex, diabetes, procedure/antiplatelet drugs, haematocrit, platelet count, and body mass index were statistically significant independent predictors of CCB. Most of the variation in CCB was accounted for by procedure/anti-platelet drugs.
CABG with aspirin was the largest group and was the reference category. Compared with this group, CABG+valve and valve procedures in patients receiving dual anti-platelet treatment (aspirin plus P2Y 12 blocker within 7 days or less of surgery) conferred the highest odds of CCB (Fig. 2, Table 2). The predictive model for CCB  Table S7.

| Prediction of CCB using baseline characteristics and near-patient test results
Alternative predictive models incorporating near-patient test results altered the prediction of CCB compared to the baseline-only model (   (Table S8).

| Prediction of CCB using only near-patient test results
The best test-only model for CCB, which included only the nearpatient test results without baseline characteristics, had a c-statistic of 0.71 (0.69-0.74; Fig. 4, Table S7). CABG, coronary artery bypass grafting; Valve, valve replacement; ASP, aspirin or aspirin plus P2Y 12 blocker stopped more than 7 days before surgery; APT, any anti-platelet drugs; DAPT, aspirin plus P2Y 12 blocker 7 or less days before surgery shown with duration of omission of P2Y 12 blocker before surgery shown in brackets; eGFP, estimated glomerular filtration rate; SD, standard deviation.

| DISCUSSION
In this large, prospective, observational cohort study we evaluated

| Comparison with other studies
The baseline-only model developed in the analysis population enabled the correct classification of 76.8% of patients as either CCB or no CCB. This is consistent with several previous studies in which demographic details, comorbidities, conventional pre-operative laboratory test results [31][32][33][34][35] , and procedural characteristics 31,35 contributed to prediction of RBC transfusion. The baseline-only model confirmed that exposure to anti-platelet drugs before cardiac surgery is also a strong predictor of post-operative bleeding. 17,20 We also showed that for patients having CABG, aspirin plus a P2Y 12 blocker stopped for less than 2 days increased the risk of severe bleeding fourfold compared to aspirin alone. However, there was no increased risk if the P2Y 12 blocker was stopped for 6 to seven days.
This confirms the empiric advice in current practice guidelines 36 , that stopping P2Y 12 blockers for at least 5 days achieves the greatest reduction in bleeding risk.
A unique feature of the COPTIC study was the comprehensive panel of near-patient tests used to detect coagulopathy. One striking finding was that the analysis population had a median of 5/28 test results per patient that were abnormal (range 0-18) when compared to healthy control reference intervals, which were similar in our study to previously published reference intervals. [37][38][39] By far the most common abnormalities were reduced platelet function with the ASPI-test and adrenaline reagents. Since both reagents are sensitive to dysfunction of the platelet cyclooxygenase pathway 28,40 , this finding is consistent with the high proportion of patients receiving aspirin (approximately 75%). Reduced platelet function with the ADP-test was less common, reflecting the lower number of patients with any exposure to P2Y 12 blockers before surgery (approximately 20%). 28 Compared to the platelet function tests, abnormal viscoelastometry tests were less common (median 0/28 abnormal tests per patient; range 0-12), but included reduced ROTEM INTEM α-angle, ROTEM F I G U R E 2 Baseline characteristics and clinical concern about bleeding. Data are odds ratios with 95% confidence intervals for clinical concern about bleeding. Odds ratios are for male vs female, presence of diabetes vs no diabetes, procedure type/anti-platelet drug group vs CABG with ASP, and urgent vs elective priority. Odds ratios are adjusted for other factors in the table and for whether the patient was in an interventional study at our center. CABG, coronary artery bypass grafting; Valve, valve replacement; ASP, aspirin alone or aspirin plus P2Y 12 blocker stopped for longer than 7 days before surgery; APT, any anti-platelet drugs; DAPT, aspirin plus recent P2Y 12 blocker stopped 7 days or less before surgery, with duration of omission of P2Y 12 blocker indicated in brackets. Odds ratios and 95% confidence intervals are also reported in Table S7 1

| Strengths and weaknesses
The main strength of the COPTIC study was the low risk of bias ena- cardiac surgery studies. 43 However, we also included pro-haemostatic treatments by clinical judgement administered in response to observed severe bleeding, to identify patients that would not be captured with the other endpoint definitions. It is a potential criticism that some pro-haemostatic treatments by clinicial judgement may have been given before severe bleeding was observed, resulting in misclassification of patients as CCB. However, in sensitivity analyses using more conservative definitions of severe bleeding for the primary outcome, the main findings remained consistent.
It is also a potential weakness of the study that some participants had one or more items of missing data, usually because of quality con- Best test-only model; c-statistic = 0.71 the baseline characteristics of the excluded population were similar to the analysis population, indicating that it is unlikely that this exclusion strategy introduced bias.

| Clinical impact of study findings
Current practice guidelines support incorporating near-patient coagulation testing into blood management algorithms for cardiac surgery. [24][25][26] However, these guidelines do not distinguish between the use of near-patient tests to inform treatment selection after patients develop severe bleeding, and the use of pre-emptive testing to predict bleeding and to direct preventative treatments.
The results of the COPTIC study provide general support for nearpatient testing to assist blood management in cardiac surgery because our findings confirm that currently available near-patient analyzers distinguish different coagulopathies in this setting. Identifying high bleeding risk patients before bleeding occurs so that targeted preventative treatments can be given is an attractive potential clinical strategy. We have shown that bleeding can be predicted to a modest extent by considering clinical baseline characteristics that are readily available before the start of surgery, and that there is unlikely to be any additional benefit from pre-emptive near-patient testing in unselected patients.

ACKNOWLEDGMENTS
AD Mumford co-designed the study, interpreted study data, and wrote the manuscript. J Harris and CA Rogers analyzed the study data. Z Plummer and V Verheyden coordinated the study. K Lee performed laboratory analyses. BC Reeves and GD Angelini codesigned the study and interpreted study data. GJ Murphy codesigned the study.

RELATIONSHIP DISCLOSURES
The study was funded by a Programme for Applied Health Research