Cardiovascular Anesthesiology: Research Report

The Relationship Between Fibrinogen Levels After Cardiopulmonary Bypass and Large Volume Red Cell Transfusion in Cardiac Surgery

An Observational Study

Karkouti, Keyvan MD^*†; Callum, Jeannie MD^‡; Crowther, Mark A. MD^§; McCluskey, Stuart A. MD^*; Pendergrast, Jacob MD^‖; Tait, Gordon PhD^*; Yau, Terrence M. MD^¶; Beattie, W. Scott MD^*

Author Information

From the ^*Department of Anesthesia and Pain Management, Toronto General Hospital, University Health Network and ^†Toronto General Research Institute and the Institute of Health Policy, Management, and Evaluation, University of Toronto; ^‡Department of Clinical Pathology, Sunnybrook Health Sciences Centre, Toronto; ^§Departments of Medicine and Pathology and Molecular Medicine, McMaster University, Hamilton; and ^‖Laboratory Medicine Program and ^¶Division of Cardiac Surgery, Toronto General Hospital, University Health Network, University of Toronto, Toronto, Ontario, Canada.

Accepted for publication March 1, 2013

Published ahead of print May 17, 2013

Funding: No external sources of funding were used for this study.

Conflict of Interest: See Disclosures at the end of the article.

Reprints will not be available from the authors.

Address correspondence to Keyvan Karkouti, MD, FRCPC, MSc, Department of Anesthesia, Toronto General Hospital, 3EN, Toronto, ON, M5G 2C4, Canada. Address e-mail to [email protected].

Anesthesia & Analgesia 117(1):p 14-22, July 2013. | DOI: 10.1213/ANE.0b013e318292efa4

Free
SDC

Despite major improvements in the conduct of cardiac surgery and cardiopulmonary bypass (CPB), coagulopathy leading to excessive blood loss that necessitates large volume allogeneic red cell transfusion continues to occur in approximately 10% of patients.¹^,² While the causes of coagulopathy in cardiac surgery are multifactorial and include excessive fibrinolysis and platelet dysfunction, a major contributing factor is thought to be reduced coagulation factor levels after CPB.¹^,³^,⁴ The conduct of cardiac surgery and CPB results in a substantial loss, consumption, and dilution of coagulation factors, and studies have shown that most coagulation factor levels, including fibrinogen, are typically reduced by 30% to 50% after CPB.¹^,^3–5

Since fibrinogen is a critical component of the coagulation process, as it is both a precursor to fibrin and a cofactor that enhances platelet aggregation, low fibrinogen levels may be a particularly important cause of coagulopathy after cardiac surgery with CPB.^6–8 The critical fibrinogen level that is needed for proper clot formation in bleeding cardiac surgical patients, however, has not been adequately investigated.⁹ According to most transfusion guidelines, the critical fibrinogen level is 0.8 to 1.0 g/L, and the recommendations are that therapy not be instituted in bleeding patients unless fibrinogen levels decrease below these levels.^10–14 These recommendations, however, are based primarily on expert opinion and are supported by low grade studies that may not be applicable to current practice.⁹^,¹⁵^,¹⁶

Clarifying the relationship between fibrinogen levels and bleeding in cardiac surgical patients has important clinical implications, because it can help improve the management of coagulopathy by guiding fibrinogen replacement therapy (with transfusion of plasma, cryoprecipitate, or fibrinogen concentrate),⁹^,¹⁷^,¹⁸ which could potentially reduce blood loss, transfusion of red cells and other coagulation products, and their associated adverse events. To that end, we explored the relationship between post-CPB fibrinogen levels and large volume red cell transfusion in a large cohort of cardiac surgical patients, using multivariable logistic regression and propensity score analyses to adjust for potential confounding variables.

METHODS

This was a retrospective analysis of data collected on a cohort of patients who underwent cardiac surgery with CPB from January 2005 to December 2011 at the Toronto General Hospital, which is a quaternary care teaching hospital of the University Health Network that is affiliated with the University of Toronto at Toronto, Ontario, Canada. A full range of adult cardiac surgical procedures are performed at this hospital. After approval from the institutional research ethics board, which waived the requirement for informed consent, study data were obtained from institutional databases. Full-time research personnel blinded to the details of this study adjudicated all patient outcomes from patients’ records. Quality assurance checks of the databases have consistently revealed a missing data rate of <2% and an error rate of <2%.

Patients were excluded if they did not have a fibrinogen measure during surgery after termination of CPB, or if they underwent heart transplantation or insertion of ventricular assist devices (less common procedures that are often associated with severe coagulopathy at our institution). Patients with missing values for variables included in the multivariable analyses were excluded from the multivariable analyses. For patients who were readmitted for additional operations requiring CPB during the study period, only data from their first admission were used.

Clinical Practice

Laboratory Testing

Before surgery, complete blood count, international normalized ratio (INR) of prothrombin time, partial thromboplastin time, and creatinine values were routinely measured. Fibrinogen was not measured before surgery. During surgery, point-of-care arterial blood gases, hemoglobin (Hb), and activated clotting time (ACT) were routinely measured before CPB, every 15 to 30 minutes during CPB, and post-CPB after protamine administration. To help guide transfusion therapy post-CPB, complete blood count, INR, partial thromboplastin time, and fibrinogen (by the Clauss method) were frequently measured. Some of the anesthesiologists conducted these measures in every patient after they administered protamine, at the same time as the post-protamine ACT measure, while others performed them selectively only if patients were noted to have inadequate hemostasis after protamine reversal (i.e., if patients were subjectively deemed to be “wet” or “oozy”). All blood samples were drawn from indwelling arterial catheters (patency maintained with pressurized heparin-free flush) and were analyzed at the hospital laboratory. Turnaround time for coagulation tests were 30 to 60 minutes. Only patients who had their fibrinogen measured during surgery post-CPB were included in the analyses.

CPB Related

Anticoagulation during CPB was achieved with IV heparin (400 U/kg bolus with additional increments as necessary) to maintain an ACT above 480 seconds. The CPB circuit was primed with approximately 1.5 to 1.8 L Ringer’s lactate solution, 25 g mannitol, 2000 to 5000 units heparin, and 50 mEQ sodium bicarbonate. Albumin (25%) and synthetic colloids (up to 500 mL) were added to the circuit at the discretion of the clinical team. Management of CPB included retrograde autologous priming of the circuit where possible, systemic temperature drift to 32°C to 34°C, alpha-stat pH management, targeted mean perfusion pressure between 50 and 70 mm Hg, and pump flow rates of 2.0 to 2.5 L/min/m². Myocardial protection was achieved with intermittent antegrade and, occasionally, retrograde blood cardioplegia. When necessary, moderate or deep hypothermic circulatory arrest was achieved by cooling to 20°C to 28°C with or without antegrade or retrograde cerebral perfusion. CPB circuits were not heparin-coated. During CPB, shed pericardial blood was salvaged into the cardiotomy suction reservoir and reinfused via the CPB circuit for as long as patients were anticoagulated. After separation from CPB, heparin was neutralized with protamine sulphate to a target ACT of within 10% of baseline (the initial dose was calculated based on the initial heparin dose and the protamine’s neutralizing factor). All patients received full dose antifibrinolytic drugs as we have previously described in detail (aprotinin for high-risk cases until 2009, tranexamic acid for all others).¹⁹^,²⁰ Cell savers (which included processing of blood before reinfusion) were used at the discretion of the clinical team.

Blood Product Transfusion Practice

Antiplatelet drugs (other than acetylsalicylic acid) and warfarin were routinely stopped 5 to 7 days before surgery in nonemergent cases. Perioperative transfusion practice was in accordance with standard transfusion guidelines.¹¹^,²¹ Indications for red cell transfusion included hematocrit concentration <20% during CPB, <21% to 24% in stable patients, and <27% in unstable or bleeding patients post-CPB. Indications for platelet transfusion included a platelet count of <50 × 10⁹/L or ongoing bleeding after reversal of heparin with a platelet count of <80 × 10⁹/L. In cases where the risk of a qualitative platelet disorder after CPB was deemed to be high (e.g., recent use of antiplatelet drugs or prolonged CPB), platelets were often transfused irrespective of platelet counts in the setting of ongoing bleeding. All platelets (random donor or single donor) were prestorage leukoreduced by Canadian Blood Services. In patients who continued to bleed after full reversal of heparin, plasma (2–4 units) was transfused if the INR was >1.5 and cryoprecipitate (8–10 units) was transfused if the fibrinogen value was <1.0 g/L. Fibrinogen concentrates were not available for use during the study period. Given the long turnaround time for coagulation test results, plasma and platelets (but not cryoprecipitate) were at times administered empirically in patients with severe bleeding.

Outcomes

The primary outcome was large volume red cell transfusion, defined as ≥5 units of allogeneic red cells transfused on the day of surgery or the day after surgery combined, an outcome that has been shown to be prognostically important.² Other outcomes measured included perioperative transfusion of ≥1 red cells, plasma, platelets, and cryoprecipitate.

Statistical Analyses

SAS™ version 9.3 (SAS Institute, Inc., Cary, NC) was used for the statistical analyses. Categorical variables were summarized as frequencies and percentages and continuous variables as medians and 25th and 75th percentiles. The unadjusted relationship of post-CPB fibrinogen level with the primary outcome was explored with cubic spline function plot and receiver operating characteristic analyses,²²^,²³ and based on the spline function plot and maximum sum of sensitivity and specificity (see Results), patients were dichotomized at the 2.0 g/L threshold (<2.0 g/L and ≥2.0 g/L). Differences in measured outcomes between the 2 fibrinogen level categories were measured by the χ² test. The unadjusted relationship of fibrinogen level as a binomial variable with the primary outcome was assessed by logistic regression.

Multivariable logistic regression modeling was used to determine the independent relationship of post-CPB fibrinogen level categories with large volume red cell transfusion. Covariates that were included in the modeling in addition to post-CPB fibrinogen level were patients’ calculated risk for large volume red cell transfusion, as determined by a previously developed and validated bleeding risk score that includes patients’ age, body surface area, preoperative shock, preoperative anemia and platelet count, urgency and type (redo, complex) of surgery, surgeon (high or low transfusion surgeon), duration of CPB and deep hypothermic circulatory arrest, and nadir hematocrit during CPB;²⁴^,²⁵ difficult wean from CPB, defined as the requirement for inotropes or intraaortic balloon pump to separate from CPB, a perioperative confounder that was not included in the bleeding risk score; preoperative Hb, as this confounder will influence the number of red cell transfusions; and post-CPB Hb, platelet count, and INR.⁴ Collinearity was assessed by the Spearman correlation test. Continuous variables were categorized if they were not linearly related to the outcome (as determined by cubic spline function plots). Modeling was by stepwise variable selection, with P < 0.05 used as the criterion for variable inclusion. Model discrimination and calibration were assessed by the c-index and the Hosmer–Lemeshow statistic (larger P value means better calibration), respectively. Bootstrap resampling was used to assess the stability of the models as follows: 100 computer generated samples, each including 4500 patients, were derived from the study cohort by random selection with replacement, and the models were refitted for each sample. The retention rates of the variables in the bootstrap models as well as the bootstrap mean odds ratios for the variables were obtained.

For sensitivity analysis, modeling was repeated after excluding the following groups: (1) patients who underwent emergent surgery (n = 106), because this group may have received antiplatelet and anticoagulant drugs up to the time of surgery, potentially important confounders for which data were not available; (2) patients who received aprotinin (n = 130), because it was reserved for high-risk cases and was not used after 2009; (3) patients who died (n = 121); (4) patients who received cryoprecipitate (n = 89), because cryoprecipitate was administered in response to low fibrinogen levels and may therefore have biased observed relationships; (5) patients with fibrinogen value <1.0 or >5.0 g/L (n = 65), to eliminate the effect of extreme fibrinogen levels; and (6) patients who were anemic before surgery (n = 1060), because they may have received multiple red cell transfusions during CPB, before fibrinogen was measured. Modeling was also repeated using the components of the bleeding risk score as separate covariates.

For additional verification, the risk-adjusted relationship between post-CPB fibrinogen levels and the primary outcome was measured using propensity score methods.²⁶^,²⁷ For this analysis, a multivariable logistic regression model that included all measured covariates as predictor variables and the fibrinogen level categories as the dependent variable was developed, and this model was used to calculate the propensity score for each patient of having a low fibrinogen level (<2 g/L) post-CPB. Patients with low fibrinogen levels were then matched to patients with high fibrinogen based on similar propensity scores (using a 5→1 computerized greedy-matching technique).²⁸ The matched pairs were then compared on all measured covariates to determine whether the groups were balanced. Balance was determined by the standardized mean difference (SMD), which is the absolute difference of the group means as a percentage of their pooled standard deviation, and groups were considered balanced for a particular variable if the SMD was <10%.²⁶ The primary outcomes between these matched pairs were then compared using conditional logistic regression.²⁶

RESULTS

During the study period, 10,185 patients underwent cardiac surgery with CPB, of whom 4696 had a fibrinogen measure during surgery post-CPB. Of these, 87 patients who underwent heart transplantation or ventricular assist device implantation and 3 patients who were missing transfusion data were excluded, leaving 4606 patients to be included in the analyses. The proportion of patients who had a post-CPB fibrinogen measure increased from 21% in 2005 to 60% in 2009, and stabilized thereafter (Table 1). Patients included in the analyses had similar characteristics to those who were excluded, except that they underwent more complex, redo, and circulatory arrest cases. Consistent with this, they had a higher overall bleeding risk score and received more transfusions (Table 1).

T1-4 — Table 1:
Patient Characteristics, Surgical Details, and Laboratory Values

The cubic spline function graphs suggested that the probability of large volume transfusion increased when fibrinogen levels decreased below approximately 2.0 g/L (Fig. 1, A and B). The receiver operating characteristic analysis showed that the sum of sensitivity and specificity was maximized with fibrinogen level cutoffs of 1.7 to 2.0 g/L. Based on these data, we categorized patients into low fibrinogen (<2.0 g/L) and normal fibrinogen (≥2.0 g/L) groups to assess the independent relationship between low fibrinogen levels and large volume red cell transfusion. At the 2.0 g/L level, sensitivity was 50% and specificity was 60%.

F1-4 — Figure 1:
Cubic spline function curves of the unadjusted relationship between fibrinogen level and probability of perioperative transfusion of ≥5 units of red blood cells (RBC) (dotted lines represent the 95% confidence intervals [CIs]). Panel A includes the entire range of fibrinogen values, and panel B includes fibrinogen values of 1.5 to 2.5 g/L. Using <2.0 g/L as the threshold for low fibrinogen, which was selected for the primary analyses, the odds ratio for the outcome was 1.8 (95% CI, 1.4–2.2) (Table 3 and text). Thresholds around this value yielded similar odds ratios (e.g., <1.8 g/L threshold: odds ratio, 1.8; 95% CI, 1.4–2.2; <2.2 g/L threshold: odds ratio, 1.7; 95% CI, 1.3–2.1).

Of the 4606 included patients, 1918 (42%) had low fibrinogen levels, and these patients differed from those with normal fibrinogen levels in several characteristics that may have influenced transfusion rates. On the one hand, the low fibrinogen group included more women, were smaller, had lower preoperative platelet counts, and underwent more complex, redo, or circulatory arrest procedures; while on the other hand, they were younger, had fewer incidences of atrial fibrillation, left ventricular dysfunction, dialysis, and endocarditis, and had higher preoperative Hb concentrations (Table 1). The time from start of surgery to the post-CPB fibrinogen measure was similar in the 2 groups: 183 (145–222) minutes in the <2.0 g/L group vs 180 (145–219) minutes in the ≥2.0 g/L group (P = 0.4).

Overall, the 2 groups had very similar bleeding risk scores (Table 1). Despite this, the low fibrinogen group received significantly more transfusions than the normal fibrinogen group (Table 2). The unadjusted odds ratio (95% confidence interval) for transfusion of ≥5 units of red cells was 1.5 (1.3–1.7), and the risk-adjusted odds ratio obtained by multivariable logistic regression analysis was 1.8 (1.4–2.2) (Table 3). Similar odds ratios were obtained in the sensitivity analyses (Table 4). Propensity score matching resulted in 2050 matched pairs that were balanced (SMD < 10%) on all measured covariates (Table 1). Among the matched pairs, 192 (18.7%) patients in the low fibrinogen group and 133 (13.0%) patients in the normal fibrinogen group were transfused ≥5 red cell units (Table 2), resulting in an odds ratio (95% confidence interval) of 1.5 (1.2–2.0). The proportion of patients who received cryoprecipitate was also higher in the low fibrinogen group, but plasma and platelet transfusions were comparable between the 2 groups (Table 2).

T3-4 — Table 3:
Multivariable Logistic Regression Model

T4-4 — Table 4:
Sensitivity Analyses Results

DISCUSSION

In this retrospective observational study, we explored the relationship between post-CPB fibrinogen levels and large volume red cell transfusion in a large cohort of patients who underwent cardiac surgery with CPB. Cubic spline function plots showed that the probability of large volume transfusion increased when fibrinogen levels decreased below approximately 2.0 g/L. Using multivariable logistic regression and propensity score methods to control for important confounders, we found that patients whose fibrinogen level was <2.0 g/L had an approximately 50% increase in the odds of large volume transfusion.

It is important to note that this study was not equipped to identify the critical fibrinogen level for instituting fibrinogen replacement therapy in bleeding patients. Nevertheless, its results are inconsistent with the commonly held position that clotting is not impaired, and hence replacement therapy should not be instituted until fibrinogen levels decrease below 0.8 to 1.0 g/L.^10–14^,^29–32 The literature in support of this position, however, consists of weak studies that may not be applicable to current practice.⁹^,¹²^,¹⁵^,¹⁶ In comparison, our study has several important strengths. We used a large cohort of patients who were managed in a homogenous manner at a single institution and underwent modern cardiac surgery, analyzed the relationship between post-CPB fibrinogen levels with an objective and prognostically important outcome, accounted for important potential confounders using multivariable logistic regression and propensity score analyses, and validated our findings with bootstrap and sensitivity analyses.

The findings of this current study are supported by several lines of recent evidence. First, in several studies that explored the effects of fibrinogen concentration on mechanical properties of whole blood clots either experimentally by serial dilution of collected blood or by studying blood samples from patient cohorts, it was found that clot formation and strength may be impaired even when fibrinogen levels are substantially >1.0 g/L.⁷^,^33–36 Second, several preliminary observational studies in various patient cohorts also found that low fibrinogen levels may be associated with increased blood loss or transfusions even at levels >1.0 g/L.⁴^,^37–40 Additional supporting evidence comes from several preliminary observational and randomized studies that administered fibrinogen to patients whose fibrinogen levels were low but >1.0 g/L and found the intervention to be associated with reduced blood loss or transfusion.^41–47 In light of these studies, some groups have increased their recommended treatment threshold from 1.0 g/L⁴⁸ to 1.5 to 2.0 g/L.⁴⁹

Our study has several limitations that must be considered when interpreting its findings. First, since it was a retrospective observational study, causality cannot be established and the influence of unmeasured confounding factors cannot be dismissed. Importantly, we did not measure specific coagulation factor levels other than fibrinogen and consequently could not assess their individual contributions to large volume transfusion. We did, however, account for deficits in coagulation factor levels as a group by including post-CPB INR in the multivariable analyses. We also did not measure platelet function and therefore could not assess the influence of platelet dysfunction on the observed associations. To minimize the potentially confounding effects of platelet dysfunction, we measured and controlled for important causes of platelet dysfunction in cardiac surgery such as renal dysfunction, duration of CPB, and type of surgery,⁵⁰ as well as platelet count. We did not have data on antiplatelet drug use, which is another important cause of platelet dysfunction, but since these drugs were routinely discontinued before nonemergent surgery, our sensitivity analysis that excluded patients who underwent emergent surgery likely minimized the potential confounding effects of these drugs. For these reasons, we believe that it is unlikely that platelet dysfunction was a major confounding variable in this study. Regarding other confounders, we measured and accounted for known important risk factors for large volume red cell transfusion after cardiac surgery²⁴^,²⁵^,⁵¹ and controlled for these potential confounders by both multivariable logistic regression and propensity score analyses. The results of the propensity score analysis are particularly illuminating, because they confirmed the results of the multivariable logistic regression analysis in a group of patients who were very well matched in terms of not only all measured covariates, but also plasma and platelet transfusions. This further suggests that the observed association between low fibrinogen levels and large volume red cell transfusion was not due to unmeasured confounders such as deficits in other coagulation factor levels or platelet dysfunction.

Another limitation of the study was that the indication for and timing of post-CPB fibrinogen measure were not standardized. Thus, more than half of the patients who underwent surgery at the institution did not have a post-CPB fibrinogen measure, and therefore could not be analyzed. Our results, therefore, may not be generalizable to the general cardiac surgery population. Of note, however, excluded patients did not greatly differ from included patients (Table 1). As for the lack of standardized timing for measuring fibrinogen, we do not believe this to be a major limitation because, as outlined earlier, clinical practice was such that fibrinogen was typically measured immediately or very soon after reversal of heparin post-CPB before there could be substantial blood loss or blood product transfusion. In fact, during the study period, there was a gradual change in practice such that more anesthesiologists measured fibrinogen routinely at the same time as the first ACT measure post-CPB, as is illustrated by the annual increase in the proportion of patients who had a fibrinogen measure post-CPB. In addition, the timing of fibrinogen measure relative to the start of surgery was very similar in the low and high fibrinogen groups, indicating that the timing of fibrinogen measure was not a confounding variable.

Another limitation of our study is that while the date of transfusion was available, the exact timing was not available. Thus, we could not differentiate between transfusions that occurred during or after CPB. Owing to the high transfusion threshold used to define high volume transfusion and our inclusion of preoperative and post-CPB Hb as covariates in the multivariable analyses, however, it is unlikely that our results were unduly influenced by blood loss and transfusions before the post-CPB fibrinogen measure. The consistency of our findings when we excluded patients with preexisting anemia, who are most likely to have received some transfusions during CPB, supports this conclusion.

Finally, since preoperative fibrinogen levels were not routinely measured, we were not able to analyze the effects of the relative change in fibrinogen levels, which may be more predictive of large volume red cell transfusion than absolute post-CPB fibrinogen levels.⁴ There is, however, a strong correlation between preoperative and postoperative fibrinogen levels,³⁹ and in our experience clinicians base their transfusion decisions on the absolute post-CPB fibrinogen levels rather than the preoperative levels or the relative change in levels. We also could not assess the effects of plasma or platelet transfusions (which contain some fibrinogen) on the observed relationships because few patients who reached the red cell transfusion threshold did not receive plasma (only 35 patients who received ≥5 red cell units did not receive plasma and only 94 did not receive platelets).

Despite the limitations of our study, we postulate, based on our findings and other supportive evidence, that in bleeding cardiac surgical patients, the critical fibrinogen level for instituting replacement therapy may be higher than the commonly recommended 0.8 to 1.0 g/L, and that maintaining higher fibrinogen levels may lead to reduced blood loss, blood transfusion, and related adverse events and costs. We must emphasize, however, that while our study suggests that the critical fibrinogen level is >1.0 g/L, it was not equipped to identify the critical fibrinogen level. That is, one cannot infer from our results that bleeding patients should be treated to a fibrinogen level of 2.0 g/L. Adequately powered trials randomizing bleeding patients to different fibrinogen levels are needed to identify the critical fibrinogen level in bleeding patients. Moreover, since increased fibrinogen levels have been associated with thrombosis⁵² and fibrinogen replacement therapies have important risk profiles,⁹^,¹⁸ future studies must be adequately powered to clarify the risk–benefit balance of raising the critical fibrinogen threshold after cardiac surgery.

In summary, in this single-center retrospective cohort study that included 4606 patients who underwent cardiac surgery from 2005 to 2011 at a single institution, we found that the probability of large volume transfusion increased when fibrinogen levels decreased below approximately 2.0 g/L. The findings of this study suggest that current recommendations that fibrinogen replacement not be initiated in bleeding patients until fibrinogen levels decrease below 0.8 to 1.0 g/L may be too conservative. Randomized trials are needed to determine whether maintaining higher fibrinogen levels in bleeding patients reduces blood loss, red cell and other blood product transfusions, and by that means improves outcomes in cardiac surgery.

DISCLOSURES

Name: Keyvan Karkouti, MD.

Contribution: This author helped in study design, study conduct, data collection, data analysis, and manuscript preparation.

Attestation: This author has read and agrees to the manuscript as written. Keyvan Karkouti has reviewed the original study data and data analysis, and attests to the integrity of the data and the reported analyses. Keyvan Karkouti is the archival author.

Conflicts of Interest: Dr. Karkouti is involved in a research project being conducted by CSL Behring, but has received no direct funding from the company. Dr. Karkouti has received research support, honoraria, or consulted for Bayer, Novo Nordisk, and The Medicines Company.

Name: Jeannie Callum, MD.

Contribution: This author helped in study design and manuscript preparation.