Article Text
PDF
Abstract
Objectives Congestion is associated with worse outcomes in critically ill surgical patients but can be difficult to quantify noninvasively. We hypothesised that plasma volume status (PVS), estimated preoperatively using a validated formula that enumerates percentage change from ideal plasma volume (PV), would provide incremental prognostic utility after coronary artery bypass graft (CABG) surgery.
Methods In this retrospective cohort study, patients who underwent CABG surgery (1999–2010) were identified from a prospectively collected database. Actual ([1-haematocrit] x [a+(b x weight [kg])]) and ideal (c x weight [kg]) PV were obtained from equations where a, b and c are sex-dependent constants. Calculated PVS was then derived (100% x [(actual−ideal)/ideal]).
Results In 1887 patients (mean age 67±10 years; 79% male; median European System for Cardiac Operative Risk Evaluation [EuroSCORE] 4), mean PVS was −8.2±9%. While 8% of subjects had clinical evidence of congestion, a relatively increased PV (PVS >0%) was estimated in 17% and correlated with lower serum sodium, higher EuroSCORE and a diagnosis of diabetes mellitus. A PVS≥5.6% was optimally prognostic and associated with greater mortality (HR: 2.31, p=0.009), independently of, and incremental to, EuroSCORE, New York Heart Association class and serum sodium. A PVS≥5.6% also independently predicted longer intensive care (β: 0.65, p=0.007) and hospital (β: 2.01, p=0.006) stays, and greater postoperative renal (OR: 1.61, p=0.008) and arrhythmic (OR: 1.29, p=0.03) complications.
Conclusions Higher PVS values, calculated simply from weight and haematocrit, are associated with worse inpatient outcomes after CABG. PVS could help refine risk stratification and further investigations are warranted to evaluate the potential clinical utility of PVS-guided management in patients undergoing CABG.
- coronary artery disease surgery
- heart failure
Statistics from Altmetric.com
Introduction
Coronary artery bypass grafting (CABG) is the most common cardiac surgical procedure1 that remains the optimal route of revascularisation for patients with severe coronary artery disease.2 However, despite refinements in surgical, perfusion and anaesthetic techniques, patients undergoing CABG still endure a 16-fold higher 30-day mortality rate than the general population,3 and their postoperative course is frequently complicated by de novo arrhythmias, kidney injury and prolonged hospitalisation.2 Because congestion identifies those at a greater risk of harm,4–8 the relief of congestion is a fundamental goal of preoperative optimisation. This is particularly relevant to patients undergoing cardiac surgery whose haemodynamics are exquisitely sensitive to cardiopulmonary bypass and fluctuations in myocardial function. Despite appearing clinically euvolaemic, many patients with cardiac disease remain haemodynamically overloaded (covert congestion), and the prognosis for these individuals is nearly as bad as for those with overt oedema.9–11 Novel markers of congestion that provide prognostic information incremental to clinical indices might better stratify risk and guide diuretic therapy prior to CABG.
Plasma volume (PV) expansion underlies systemic congestion in patients with cardiac disease but is notoriously difficult to quantify noninvasively. Tracer-dilution techniques optimally measure PV but are expensive and clinically impractical.12 Echocardiography and physical findings can inform on volume status but have several limitations.10 13 14 Because PV is closely associated with weight and haematocrit, validated equations based on these simple indices can be utilised to objectively estimate PV levels. In a prior analysis, calculated PV values were correlated to those measured using tracer-dilution assays.15 Moreover, in acute and chronic heart failure cohorts, relative PV status (PVS), a measure of the degree to which patients have deviated from their ideal PV, strongly predicted mortality even after extensive covariate adjustment.15 16 The aims of this study were to assess the distribution, correlates and prognostic utility of calculated PVS in patients undergoing CABG.
Methods
Subjects
Patients who underwent CABG surgery at The Heart Hospital, London between 1999 and 2010 were identified from a prospectively and consecutively collected database that is validated and stored by a dedicated independent team as part of the UK’s National Adult Cardiac Surgery Registry. This registry is approved and access to it was granted by the clinical audit committee of The Heart Hospital. The need for individual patient informed consent was waived by the Ethics Committee as this was a retrospective analysis using a de-identified administrative database. Of the 1910 patients identified, 23 were excluded because of missing weights. Thus, 1887 patients were included in this study. Baseline, intraoperative and postoperative data collection and recording of health outcomes were done to the standards of the Registry. Clinical congestion was assessed prospectively and was defined by the presence of signs (eg, jugular venous distension, S3 gallop, laterally displaced apical impulse, pulmonary rales or peripheral oedema) of fluid overload on physical examination. In-hospital weights and haematocrits documented just before CABG were utilised to calculate PVS. Follow-up information was collected from medical records according to the standards of the Registry.
PV equations
Calculated PVS was obtained by determining actual and ideal PV levels. Actual PV was calculated with the following equation derived by curve-fitting techniques using the participants' haematocrit and weight compared with PV values measured with the gold standard radiolabelled albumin assay17:
actual PV = ([1−haematocrit] x [a + (b x weight [kg])]), where haematocrit is a fraction, and a=1530 in males and 864 in females, and b=41 in males and 47.9 in females.
Ideal PV was calculated from the following well-established formula18:
ideal PV=c x weight (kg) where c=39 in males and 40 in females
Relative PVS, an index of the degree to which patients have deviated from their ideal PV, was subsequently calculated from the following equation:
PVS = (actual PV – ideal PV) x 100% / Ideal PV
Statistical analysis
Data are presented as mean ±SD, proportions (%) or medians (IQR). Intergroup comparisons were made using a Student’s t-test, Mann-Whitney U test, Pearson χ2 test or Fisher’s exact test as appropriate.
Associations between covariates and PVS, length of hospital/intensive care unit (ICU) stay and postoperative complications were evaluated using linear or logistic regression. To minimise collinearity, regression models only considered the following covariates that were not directly related to PVS or to the European System for Cardiac Operative Risk Evaluation (EuroSCORE): current smoker, New York Heart Association (NYHA) class, Canadian Cardiovascular Society (CCS) class, extent of coronary disease, prior myocardial infarctions (MIs), heart rhythm, diabetes, hypertension, sodium and prior surgical interventions. For correlates of PVS>0%, all significant univariable predictors were entered into stepwise multiple regression models. For the association of PVS to lengths of stay and postoperative complications, multivariable models only included a priori determined clinically relevant covariates (EuroSCORE, NYHA class and serum sodium). Certain variables were dichotomised by their median values to facilitate regressions. A p value <0.20 was used to enter and retain covariates in multivariable models. The validity of linear and logistic regressions was verified by analysis of model residuals, linearity condition, testing for heteroscedasticity and the absence of interaction and multicollinearity. As lengths of stay were not normally distributed, bootstrap linear regression with 1000 bootstrap samples was performed to model these endpoints.
The association between PVS and all-cause death in hospital and at 1 year were determined using Cox proportional hazards analyses. Predictors of late (1 year) mortality were assessed only in patients who survived to hospital discharge (landmark survival analysis). Multivariable Cox models included a priori determined clinically relevant covariates (EuroSCORE, NYHA class and serum sodium) that were not directly related to PVS or to the EuroSCORE (to minimise collinearity). The significant levels for χ2 (likelihood ratio test) were calculated. The validity of Cox models was verified by assessing the proportionality of hazards, log-linearity and absence of interaction and multicollinearity. Kaplan-Meier cumulative survival plots were constructed for visualisation and assessed using the log-rank test.
Calculated PVS was assessed as both a continuous and a categorical variable in survival analyses. The shape of the association between PVS and survival was investigated by using a restricted cubic spline function on the PVS variable. The PVS cut-off that best discriminated mortality was determined using receiver operating characteristic (ROC) analysis. In this, sensitivity and specificity were of equal importance; therefore, the optimal PVS cut-off was the one giving the highest product of sensitivity and specificity for mortality prediction. Relative PVS was dichotomised by this cut-off in subsequent survival analyses, and in the analyses of postprocedural complications. The incremental prognostic utility of PVS was evaluated by calculating the continuous net reclassification improvement and the integrated discrimination index.
Missing data (3.3% of total data) were handled by imputing mean/median/mode values. Sensitivity analyses with a post-hoc Markov chain-Monte Carlo imputation with five iterations revealed similar results. No adjustments for multiple statistical comparisons were made.
Data were analysed using SPSS (V. 24.0, SPSS, IBM), STATA V. 12 and R (V. 3.4.3, R Development Core Team, California, USA). A two-tailed p value <0.05 was considered statistically significant.
Results
Baseline characteristics
The distributions of the baseline demographic and clinical features of the 1887 patients undergoing CABG are shown in table 1. Calculated PVS appeared normally distributed and ranged from −34.6% to 34.3% (figure 1).
Distribution of calculated PVS in the entire cohort (A), males (B) and females (C). PVS, plasma volume status.
Baseline characteristics in the entire population and stratified by calculated PVS
While only 8% of patients were documented to have had current or prior signs of congestion, a relatively increased PV, as defined by a PVS >0%, was evident in 17% of subjects. Compared with those with a PVS ≤0%, patients with a PVS >0% were more likely to be older and female with a higher EuroSCORE, longer aortic cross-clamp time, and worse left ventricular ejection fraction, renal function and symptoms as reflected by higher NYHA and CCS class designations. Furthermore, patients with a PVS >0% were more likely to have cardiogenic shock, an urgent or emergent operative priority, a shorter interval between MI and CABG, a lower serum sodium, and a previous history of MI, atrial arrhythmias, diabetes mellitus or cerebrovascular accidents. On multivariable logistic regression analysis, only a history of diabetes mellitus (OR: 1.43, p=0.04), a lower serum sodium (OR: 0.84, p<0.0001) and a higher EuroSCORE (>4 vs.≤4; OR: 2.34, p<0.0001) were independently associated with a PVS >0%.
PVS and in-hospital, 30-day and 1-year mortality
After a median hospital stay of 7 (IQR: 6–10) days, 46 (2%) patients died. Higher levels of PVS were associated with a greater risk of inpatient death (HR 1.04, 95% CI: 1.01 to 1.07, p=0.01) in unadjusted Cox analyses. The relation between PVS and log relative hazard of mortality was nonlinear and ’J-shaped' fashion (figure 2) with PVS values between −7.5% and −26% visually associated with the lowest risk. On multivariable Cox analysis however, the association between continuous PVS and death was lost (HR 1.03, 95% CI: 0.99 to 1.07, p=0.18). Consequently, PVS was dichotomised using a ≥5.6% cut-off as it gave the highest product of sensitivity and specificity (area under the curve 0.66, 95% CI: 0.58 to 0.74, p<0.001) on ROC analysis. A PVS ≥5.6% was present in 672 (36%) patients, was associated with a threefold unadjusted increase in inpatient mortality (table 2, figure 3) and with a twofold increase after adjustment for EuroSCORE >4, NYHA class ≥III and serum sodium.
Cubic spline plot of relation of PVS to risk of inpatient mortality. The model is fitted using restricted cubic splines with four knots in the proportional hazards regression model. The solid line represents the HR and shaded area the 95% CI. PVS, plasma volume status.
Kaplan-Meier inpatient survival curve for calculated PVS stratified by optimal cut-off. PVS, plasma volume status.
Univariable and multivariable predictors of inpatient mortality
Addition of PVS ≥5.6% to a baseline model incorporating EuroSCORE >4, NYHA class ≥III and serum sodium incremented model performance as it enabled 30.4% (p=0.04) of patients dying in hospital to be correctly reclassified as higher risk, and 29.6% (P<0.001) of patients surviving to discharge to be reclassified as lower risk. The overall net reclassification improvement, reflecting the increment in prediction accuracy, was 60% (p<0.001). The integrated discrimination index, which reflects the change in calculated risk for each patient was 0.6% (95% CI 0.1% to 1.2%) among patients who died and 0% (95% CI −0.1% to 0.1%) for patients who survived during follow-up. The discrimination slope was 0.6 percentage points higher than the original.
Higher levels of PVS were also associated with a twofold unadjusted increase in 30-day mortality (HR 2.50, 95% CI: 1.39 to 4.52, p=0.002), but the association was not statistically significant on multivariable analysis (HR 1.47, 95% CI: 0.79 to 2.72, p=0.22). A PVS ≥5.6% was not associated with long-term (1 year) mortality (unadjusted HR 2.01, 95% CI: 0.97 to 4.16, p=0.60) in the landmark survival analysis.
PVS and postoperative duration and complications
In the total population, median ICU and hospital stay were 1 (IQR: 1–2) and 7 (IQR: 6–10) days, respectively. Patients with a PVS ≥5.6% were in the ICU and hospital longer after CABG surgery, compared with patients with a PVS ≤5.6% (figure 4). On multivariable bootstrap linear regression analysis, a PVS ≥5.6% remained an independent predictor of number of days spent in ICU (β: 0.65, 95% CI: 0.22 to 1.12, p=0.007) and in hospital (β: 2.01, 95% CI: 0.74 to 3.49, p=0.006) after adjustment for EuroSCORE >4, NYHA class ≥III and serum sodium. On restricting the analysis to only patients who were discharged alive (so as to remove confounding from the competing risk of death), a PVS ≥5.6% remained an independent predictor of number of days spent in ICU (β: 0.51 95% CI: 0.05 to 0.97, p=0.03) and in hospital (β: 1.79, 95% CI: 0.63 to 3.36, p=0.02) after similar covariate adjustment.
Hospitalisation duration and postoperative complications in patients above or below a PVS of −5.6%. p Values were calculated from the Mann-Whitney U test (*), χ2 test (χ) or Fisher’s exact test (γ). ICU, intensive care unit.
Multiorgan failure, a new stroke, new arrhythmia, kidney injury and infections developed in 0.7%, 1.4%, 7.2%, 8.3%, 5.1% and 5.6% of the total population postoperatively. Patients with a PVS ≥5.6% were more likely to develop complications after CABG than patients with a PVS ≤5.6% (figure 4). On univariable logistic regression analyses, a PVS ≥5.6% was linked to greater renal complications (OR: 2.31, 95% CI: 1.66 to 3.20, p<0.0001) and new arrhythmias (OR: 1.74, 95% CI: 1.41 to 2.14, p<0.0001) but not multiorgan failure (OR: 2.95, 95% CI: 0.96 to 9.05, p=0.06), new stroke (OR: 1.71, 95% CI: 0.80 to 3.65, p=0.17) or infections (OR: 1.50, 95% CI: 0.99 to 2.30, p=0.06). In multivariable logistic regression analysis, a PVS ≥5.6% remained an independent predictor of renal complications (OR: 1.61, 95% CI: 1.14 to 2.29, p=0.008) and new arrhythmias (OR: 1.29, 95% CI: 1.03 to 1.62, p=0.03) after adjustment for EuroSCORE >4, NYHA class ≥III and serum sodium.
In patients who were discharged alive, multiorgan failure, a new stroke, new arrhythmia, kidney injury and infections developed in 0.1%, 1.3%, 7.0%, 7.0% and 4.5% of subjects postoperatively. Again, a PVS ≥5.6% was associated with a greater risk of kidney injury (OR: 1.48, 95% CI: 1.01 to 2.17, p=0.045) after adjustment for EuroSCORE >4, NYHA class ≥III and serum sodium.
Discussion
Preoperative congestion can worsen cardiac function, can be substantial despite clinical euvolaemia, and can be objectively estimated using validated equations based on weight and haematocrit. Using one such formula, we calculated PVS in patients awaiting CABG and made several observations. First, while only 8% of subjects had clinical evidence of congestion, a relative increase in PV (as defined by a PVS >0%) was estimated in 17% of individuals. Second, a lower serum sodium, history of diabetes mellitus and a higher EuroSCORE were independent correlates of an increased PV. Third, PVS was associated with inpatient mortality in a ’J-shaped' fashion with a PVS ≥5.6% linked to a twofold greater chance of death independently of, and incremental to, other conventional risk markers. Fourth, a PVS ≥5.6% identified patients at a heightened risk for postoperative complications and prolonged hospitalisation.
The disparity in the proportion of patients with clinically defined congestion (8%) compared with those with a PVS >0% (17%) potentially highlights the limitations of physical examinations in that they often underestimate PV status.10 This is consistent with data showing that many clinically euvolaemic heart failure patients are discharged with persistent haemodynamic (’covert') overload, as evidenced by objectively measured elevations in right atrial pressures and natriuretic peptides.9 11 Moreover, increases in these measures were associated with higher postdischarge event rates. Consequently, attention is now focused on identifying simple noninvasive metrics that can reflect both clinical and haemodynamic congestion.
Calculated PV levels have been validated as surrogates of intravascular filling using radiolabelled albumin,15 17 and the correlates of estimated PV expansion elucidated here further support this notion. That lower levels of serum sodium were linked to PV expansion accords with dilutional hyponatremia being a biochemical hallmark of volume overload. That prior diabetes mellitus and a higher EuroSCORE were also correlates possibly reflect their association with greater left ventricular systolic dysfunction, a major risk factor for systemic congestion.14
We found that patients in negative fluid balance had better clinical outcomes. Our observation that the best PVS cut-off for predicting adverse outcomes was a negative one (−5.6%) is consistent with previous analyses. Specifically, studies have shown that restrictive fluid administration and a consequent early negative fluid balance are associated with lower perioperative mortality and lower perioperative morbidity including the occurrence of acute renal failure.19–24
Our findings of an association between higher preoperative PVS and poorer outcomes are consistent with prior studies. In 5001 patients with chronic heart failure, a PVS ≥4% was linked to a 26% increased risk of death after covariate adjustment.15 Similarly, in 1115 patients hospitalised for acute heart failure, each 1% increment in admission PVS was associated with a 21% enhanced risk of death. Moreover, higher PVS values predicted increased morbidity.16 As previously observed,15 the association between PVS and mortality was ’J-shaped' fashion in this study, consistent with the notion that excessive dehydration or congestion can be detrimental. That a PVS ≤5.6% was linked to better survival is concordant with data showing that, in patients with cardiac disease, mild-to-moderate PV contraction, even to the point of uraemia, is associated with better outcomes.25 However, in contrast to short-term fatality, PVS did not predict long-term (1 year) mortality in our cohort likely due to the highly variable nature of PV levels over longer time intervals.
While PVS might be an epiphenomenon of disease severity, credible pathophysiological mechanisms exist to explain how PVS could causally influence outcomes in patients undergoing CABG. Neurohormonal activation is thought to partly underlie oedema formation in patients with cardiac disease.26 As PVS increases, progressive venous congestion could drive further renin–angiotensin and sympathetic system activation to perpetuate fluid retention and pathological myocardial remodelling. Progressive ventricular dilatation then leads to increases in end-diastolic pressures and wall stress which can exacerbate ischaemia and cardiomyocyte loss to precipitate terminal heart failure after CABG.27 Atrial dilatation then acts as a substrate for atrial arrhythmias. Furthermore, increases in central venous pressures are transmitted to pulmonary, renal and hepatic veins, compromising oxygenation and organ perfusion and likely giving rise to the propensity to renal injury seen after CABG in patients with a PVS ≥5.6%.28 29
Our findings have potentially important clinical implications. Plasma volume expansion underlies covert and overt congestion in patients awaiting CABG and could be estimated using PVS to facilitate patient optimisation. The equation could potentially be used to identify patients in need of decongestion with diuretic therapy titrated to keep PVS below −5.6% prior to anaesthetic induction. Because the calculation of PVS is simple and utilises readily available weights and haematocrits, we contend that it may have wide applicability in patients undergoing CABG. The current study may justify future prospective observational and interventional studies to determine whether PVS could be used to identify those who are intravascularly depleted despite clinical oedema who might benefit from additional intraoperative haemodynamic support, and whether the formula could be employed postoperatively to guide patient care prior to discharge.
Our study has limitations. First, it is observational so causality cannot be inferred. Second, the data are from a single centre but is still likely to be representative of contemporary patients undergoing CABG in other centres. Third, information on blood transfusion administration, B-type natriuretic peptides and worsening heart failure was unavailable. Fourth, few patients had postoperative multiorgan failure or stroke so we are likely underpowered for these endpoints. Similarly, the number of deaths in the study was relatively small, which limited the power to detect a statistically significant difference in 30-day mortality on multivariable analysis. Fifth, the PVS formula will likely need additional refinement in CABG cohorts prior to prospective assessment of its true clinical utility. Finally, there were few female patients in this study despite the consecutive nature of the database. This likely reflects referral bias and the later presentation of coronary artery disease in women with higher comorbidity profiles.30
In conclusion, a higher preoperative PVS, calculated simply from weight and haematocrit, is associated with an increased risk of inpatient death, complications and prolonged hospitalisation in patients undergoing CABG. Further refinement of the PVS equation in patients undergoing CABG and a prospective evaluation of its utility in this cohort are warranted.
Key questions
What is already known on this subject?
Congestion is associated with worse outcomes in critically ill surgical patients, and its relief is a fundamental goal of preoperative optimisation.
Many patients with cardiac disease remain haemodynamically overloaded despite clinical euvolaemia.
Plasma volume status (PVS) can be estimated using a validated formula, incorporating simple clinical indices (weight and haematocrit), which enumerates percentage change from ideal plasma volume.
What might this study add?
In this retrospective cohort study that included 1887 adults who underwent coronary artery bypass graft surgery (CABG), a higher preoperative PVS (>−5.6%) was independently associated with a twofold risk of death early after CABG and was also associated with complications and prolonged hospitalisation.
How might this impact on clinical practice?
Calculated PVS could help refine risk stratification.
Randomised clinical trials may be warranted to evaluate the clinical utility of PVS-guided fluid management in patients undergoing CABG.
References
Footnotes
Contributors AM wrote the first draft of the manuscript. AM, MFB, AK and DOO performed the statistical analyses. DOO conceived the study and is a guarantor. All authors were involved in data acquisition and revision of the manuscript for important intellectual content.
Funding Dr Maznyczka (FS/16/74/32573), Dr Okonko (FS/14/77/30913) and Dr Barakat (FS/14/77/30913) are supported by the British Heart Foundation. Drs Okonko and Barakat are supported by the NIHR Biomedical Research Center at Guy’s and St Thomas' NHS Foundation Trust and King’s College London.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement All data relevant to the study are included in the article.
Patient consent for publication Not required.