Predictive Indices of Morbidity and Mortality After Liver Resection

Abstract

Objective:

To determine if use of Model for End-Stage Liver Disease (MELD) scores to elective resections accurately predicts short-term morbidity or mortality.

Summary Background Data:

MELD scores have been validated in the setting of end-stage liver disease for patients awaiting transplantation or undergoing transvenous intrahepatic portosystemic shunt procedures. Its use in predicting outcomes after elective hepatic resection has not been evaluated.

Methods:

Records of 587 patients who underwent elective hepatic resection and were included in the National Surgical Quality Improvement Program Database were reviewed. MELD score, CTP score, Charlson Index of Comorbidity, American Society of Anesthesiology classification, and age were evaluated for their ability to predict short-term morbidity and mortality. Morbidity was defined as the development of one or more of the following complications: pulmonary edema or embolism, myocardial infarction, stroke, renal failure or insufficiency, pneumonia, deep venous thrombosis, bleeding, deep wound infection, reoperation, or hyperbilirubinemia. The analysis was repeated with patients divided according to their procedure and their primary diagnosis. Parametric or nonparametric analyses were performed as appropriate. Also, a new index was developed by dividing the patients into a development and a validation cohort, to predict morbidity and mortality in patients undergoing elective hepatic resection. ROC curves were also constructed for each of the primary indices.

Results:

CTP and ASA scores were superior in predicting outcome. Also, patients undergoing resection of primary malignancies had a higher rate of mortality but no difference in morbidity.

Conclusion:

MELD scores should not be used to predict outcomes in the setting of elective hepatic resection.

The MELD score, developed to predict outcome after TIPS, has been used to predict outcome after liver resection and transplantation. A statistical analysis of the NSQIP database (587 liver resections) showed that ASA classification and CTP score were predictive of short-term morbidity and mortality but MELD was not. MELD should not be used to predict outcome following liver resection.

Many attempts have been made over the past few decades to classify patients with liver disease to determine their prognosis following hepatic surgery. These have included the well-known and widely used Child-Turcotte-Pugh score (CTP) as well as more obscure methods such as indocyanine green clearance. Most recently, the Model for End-Stage Liver Disease (MELD) score has gained widespread acceptance because of its use in prioritizing candidates for liver transplant. With the exception of measures of hepatic function, these scoring systems are all mathematical computations involving various laboratory values and clinical assessments and, as such, are reflections of different facets of pathology and physiology, and yield different degrees of reliability in different patient populations.

The MELD score was originally developed to predict short-term survival in patients undergoing transcutaneous intrahepatic portosystemic shunt procedures (TIPS), and has been shown to be reliable and predictive in this setting.^1–3 It is useful in assessing prognosis in patients with alcoholic hepatitis, although its superiority to CTP scores is not universally established.⁴ Furthermore, it has replaced the traditional CTP score in stratifying patients with end-stage liver disease awaiting transplantation according to guidelines adopted by the United Network for Organ Sharing in 2002.⁵ Interestingly, MELD seems to predict patient and graft survival in cadaveric liver transplantation but not in living donor liver transplantation.^6–8

Extrapolating scoring schema from the originally described population is a common practice. However, we practice in an era of healthcare resource scarcity coupled with increasingly sophisticated surgical technique and critical and anesthetic care. More and more, we are in need of accurate and reliable methods for allocating scarce resources, as well as educating patients in making informed choices concerning their care. Use of MELD in patients undergoing elective liver resection is becoming more common but has not been validated. We proposed to examine the appropriateness of using MELD scoring for these patients by measuring its power in predicting morbidity and mortality.

MATERIALS AND METHODS

The Department of Veterans Affairs National Surgical Quality Improvement Program (NSQIP) has collected information on surgical patients since 1991. In this program, a surgical nurse collects demographic, preoperative, surgical, and postoperative data on veterans undergoing surgery in 123 Veteran's Affairs Medical Centers across the country. Adverse events as well as general laboratory values are recorded for up to 30 days after surgery. There are standard definitions for all recorded variables. This program has been thoroughly described in other publications.⁹

Records on all patients who had undergone liver resections based on CPT codes were obtained from the NSQIP master database. Data were obtained on 587 patients. Distribution of cases between wedge resection, complete right or left lobectomy, and trisegmentectomy are listed in Tables 1 and 2. MELD scores were calculated for all patients with available data according to the formula:

TABLE 1. Morbidity and Mortality by Procedure

TABLE 2. Morbidity and Mortality by Diagnosis

In cases when only a prothrombin time was reported, the international normalized ratio was calculated in the manner described by van den Besselaar.¹⁰ CTP score was calculated according to the method described by Pugh et al in 1973 and reported as the letter category A, B, or C.¹¹

Points Scored

Laboratory value 1; 2; 3

Encephalopathy none; grades 1/2; grades 3/4

Ascites none slight moderate

Serum bilirubin (mg/dL) 1–2; 2–3; >3

Serum albumin (g/dL) >3.5; 2.8–3.5; <2.8

Prothrombin time 1–4; 4–6; >6 (seconds prolonged)

Grade A: <7 points, grade B: 7–9 points, grade C: 10–15 points

The American Society of Anesthesiologists (ASA) physical status classification is a scoring system designed to stratify patients according to their perioperative risk. It is assigned either in the preoperative period or at the time of surgery by the anesthesiologist according to the following guidelines:

ASA 1: normal healthy patient

ASA 2: patient with mild systemic disease

ASA 3: patient with severe systemic disease

ASA 4: patient with severe systemic disease, a constant threat to life

ASA 5: moribund patient who is not expected to survive without the operation

ASA 6: declared brain-dead patient whose organs are being procured in anticipation of transplantation.¹²

The ASA classification score is included as a variable field in the NSQIP database. The Charlson Index of Comorbidity is a weighted index of the number and severity of comorbid conditions designed to estimate the risk of death from comorbid disease in longitudinal clinical studies. Diagnoses included in the calculation are liver disease, malignancies, diabetes, renal disease, myocardial infarction, congestive heart disease, peripheral or cerebrovascular disease, dementia, pulmonary disease, connective tissue disorders, hemiplegia, and ulcer disease. The index has been used in perioperative clinical studies as an overall measure of morbidity unrelated to the surgical procedure or relevant diagnosis.^13,14 The score was calculated according to the method described by Charlson and Pompei in 1987 from data included as preoperative variables in the NSQIP database.¹⁵

The 4 primary indices (MELD, CTP, ASA, Charlson) as well as age were assessed for their ability to predict short-term mortality and morbidity. Short-term mortality in this case is defined as death occurring within 30 days. Morbidity was defined as the development of one or more postoperative complications (pulmonary edema, pulmonary embolism, myocardial infarction, stroke, renal failure or increased creatinine > twice preoperative value, pneumonia, deep venous thrombosis, bleeding, deep wound infection, reoperation, or increased bilirubin >3 times preoperative value). In addition, the 4 primary indices were examined for their ability to predict postoperative length of hospital stay as a surrogate for resource utilization. Patients were also di-vided into groups based upon their primary diagnosis (primary hepatic malignancy, metastatic malignancy, benign and unspecified), and their CPT procedure code (wedge resection, partial lobectomy, complete right or left lobectomy, and trisegmentectomy). Statistical analysis was used to determine differences in morbidity and mortality among these groups. Pearson χ² analysis, logistic regression, or Spearman correla-tion was used for binary and ordinal variables, with ANOVA, Pearson correlation, or linear regression for continuous variables as appropriate. The ability of the primary indices as models to predict mortality was assessed by calculating the area under the receiver operating characteristic curve. Finally, simple and stepwise logistic regression was used to evaluate specific preoperative characteristics for their ability to predict 3 specific outcomes. Infectious complications included were pneumonia, deep and superficial wound infections, dehiscence, urinary tract infection, and sepsis. Pulmonary outcomes included reintubation, failure to wean from mechanical ventilation for >48 hours, pulmonary edema, pulmonary embolism, and pneumonia. Renal outcomes included increases in creatinine by at least 100%, and need for dialysis, analyzed separately. All data are presented as mean ± standard deviation.

Two final indices were derived from and tested on this data set: one to predict mortality and one for morbidity. The data set was divided into random groups, one for development of the risk indices, and the other for validation, in a ratio of 2:3. Initially, χ² or Student t test was used for categorical and continuous variables, respectively, to test the relationship of each preoperative predictive variable against both mortality (death within 30 days) and morbidity (the development of one or more of the complications listed above). Logistic regression analysis was then performed with mortality and morbidity as the dependent variables and the preoperative predictors as the independent variables in the index development cohort patient group. Continuous variables were coded into normal and abnormal ranges with the exception of white blood cell count, which was coded into low, normal, and high ranges. Using the method previously described by Arozullah et al,¹⁶ point values were assigned to each risk factor by rounding the β-coefficients from the logistic regression model to the nearest integer. The point total for each patient in the validation cohort group was then calculated and designated as the risk index for mortality or morbidity, respectively. The new morbidity and mortality indices were then tested in a manner similar to that performed for the MELD and other indices earlier in the study. As they are continuous scores, Mann-Whitney rank sum test and Spearman correlation test were calculated as appropriate to test for significance in ability to predict the outcomes of interest.

RESULTS

Tables 1 and 2 list procedure types and principal diagnoses along with the respective morbidity and mortality rates. Not surprisingly, complete lobectomies and trisegmentectomies are associated with higher rates of death and complications. Also, resection of primary neoplasms results in higher mortality rates and less impressive but still statistically significantly higher rates of morbidity. Table 3 lists the relevant complications and their frequencies. Overall morbidity was 32.0% with the most common adverse outcomes being pulmonary (failure to wean from mechanical ventilation >48 hours, reintubation, pulmonary edema, and pneumonia), renal insufficiency (increases in serum creatinine greater than double baseline), hyperbilirubinemia, deep wound infection, sepsis, bleeding, and return to the operating room. There were 50 deaths, for a mortality rate of 8.5%.

TABLE 3. Postoperative Complications

Tables 4 to 6 list the pertinent scoring systems and patient age, and their ability to predict morbidity and mortality. Morbidity was defined as the occurrence of one or more of the complications listed in Table 3. Total postoperative hospital length of stay was included as a surrogate marker of resource utilization in lieu of hospital charges or other cost variables, which are not part of the database. The ASA and CTP scoring systems were able to identify patients at risk for short-term morbidity (P = 0.023 and P = 0.0006, respectively) and mortality (P < 0.0001 and P < 0.001, respectively). In contrast, MELD and CI were not significantly related to either morbidity (P = 0.5 and P = 0.8, respectively) or mortality (P = 0.3 and P = 0.4, respectively). Interestingly, patient age was associated with development of complications (P = 0.003) but not with short-term mortality (P = 0.2). None of the predictive indices was significantly associated with increased postoperative hospital length of stay. Only CTP and ASA were able to identify patients at high risk of short-term mortality by analysis of the ROC, while only ASA identified patients who suffered significant complications.

TABLE 4. Summary of Predictive Indices

TABLE 5. Summary of Predictive Indices for 30-Day Mortality (Receiver Operator Curve)

TABLE 6. Summary of Predictive Indices for Morbidity (Receiver Operator Curve)

Table 7 lists the preoperative risk factors that were included in calculation of the new predictive risk scores for mortality and morbidity after the logistic regression procedure. Validation of the resulting scores for prediction of mortality and morbidity showed that both were highly successful. Statistical significance as well as ROC analysis results are listed in Table 8. If is of interest to note the complete lack of cardiovascular and renal predictors in both lists. Despite the fact that the average age of this patient set is 62 years, the rate of cardiovascular incidents is relatively low. This is probably due to a variety of factors, including the fact that surgeons may be selecting out those patients deemed “too sick” to survive hepatic resection. However, from this analysis, it seems that preoperative infectious complications, but particularly pulmonary factors, are far more predictive of poor outcome. It is well known that current alcohol use is a harbinger of poor prognosis in the setting of hepatic surgery.

TABLE 7. Postoperative Morbidity and Mortality Indices/ROC Curves

TABLE 8. Validation Patient Cohort (n = 336)

Table 9 describes the age, ASA, MELD, CTP, and CI for those patients who died compared with those who survived. The 2 groups were not significantly different in age, CI, or MELD. However, those who died had significantly higher CTP point scores (6.3 ± 1.9 versus 5.5 ± 0.9, P = 0.005) and higher ASA classification class (3.2 ± 0.6 versus 2.9 ± 0.6, P = 0.001). In addition, patients who died had lower hematocrit levels, as well as higher bilirubin counts and creatinine values (P < 0.0001 for all 3). Incidences of the most common complications are also listed for those patients who died compared with those who survived.

TABLE 9. Postoperative Complications and Predictive Indices in Deaths Versus Survivors

Three models of logistic regression were constructed to attempt prediction of pulmonary complications, renal insufficiency (creatinine increase >200% baseline), and infectious complications. The results are listed in Table 10. On univariate analysis, a history of chronic obstructive lung disease and alcohol use were associated with pulmonary complications, while diabetes, low preoperative albumin levels, prolonged operative time, and current tobacco use were associated with development of infection. Prolonged operative time, low preoperative albumin levels, and diabetes were associated with renal dysfunction. On multivariable regression, all remained significant except smoking for infectious complications and diabetes for renal dysfunction.

TABLE 10. Regression Analyses

DISCUSSION

Given the shortcomings of currently used scoring systems, there is a constant effort underway to create innovative, accurate, and clinically relevant assessment tools. However, the enthusiasm that accompanies any new system can lead to its application to situations for which it was not designed. MELD scores have been shown useful in stratifying patients with end-stage liver disease awaiting transplantation. By allocating organs to those with the highest scores rather than waiting time combined with subjective clinical assessment, improvements have been realized in short-term graft and patient survival.⁷ However, generalization to other types of hepatobiliary surgery does not seem warranted. In this analysis, ASA and CTP were both superior to MELD in predicting poor postoperative outcome. Interestingly, both CTP and ASA classification schemes involve intuitive, bedside assessment. This is especially true of the ASA classification, given its lack of explicit, defining criteria. The anesthesiologist assigning the score does so by the most general of guidelines with no specific laboratory values or mathematical calculations whatsoever. There does seem to be a role for clinical judgment and experience in predicting which patients will do well and which will not.

The most significant difference between patients undergoing TIPS procedures or awaiting liver transplantation and these patients undergoing elective hepatic resection is the variable rate of severe liver disease among the 2 groups. Presumably, all those in the first group already have severe liver dysfunction, although the exact pathogenesis does differ. The incidence of hepatic dysfunction in the second group is much lower. Although primary hepatocellular carcinoma does most often develop in the setting of cirrhosis, 91% of the patients in this series had minimal or completely no evidence of liver disease. It is true that clinical signs and symptoms of liver disease are often absent until liver disease is quite advanced. However, the clinical differences between the 2 groups are significant. This may account for the different degrees of success with which MELD is able to predict postoperative mortality and morbidity.

The most common postoperative complications in this series were pulmonary complications (pneumonia, failure to wean from the ventilator >48 hours, pulmonary embolism, pulmonary edema, and need for reintubation), infection (deep and open wound infection, wound dehiscence, sepsis), renal (acute renal failure or significant increases in creatinine without need for dialysis), need for reoperation, and excessive postoperative bleeding (>4 units packed red blood cells transfused). These are in agreement with other published reports, as is the overall complication rate of 32.0%. Mortality in this series is slightly higher than in other reports and may reflect a higher degree of comorbidity of patients seen in veteran's hospitals, or perhaps the surgical volume done at each center, a variable not addressed in this study. Furthermore, this series had a greater proportion of procedures for malignancies, as well as a greater number of larger resections than other published reports, both factors associated with higher morbidity and mortality.^17,18

Attempts to identify preoperative factors that are most closely associated with poor outcome were disappointing. The relationship between sepsis and diabetes is well known. Also, the predictive power of prolonged operative time is probably related to a variety of factors, including extent of dissection, degree and difficulty of resection, and surgical skill. It is not possible to say that a long operation is independently responsible for a poor outcome but is more likely a surrogate marker for other factors. Also, the strong association between elevated preoperative liver transaminases is most likely a testament to the dangers of operating on an acutely inflamed liver.

Looking at the differences between patients who died within 30 days of surgery and those who survived showed significantly different patterns in postoperative complications. It is not surprising that those who died had a greater incidence of cardiac arrest, myocardial infarction, and respiratory failure. However, the increased incidence in renal dysfunction is compelling support for the intimate physiologic link between the liver and the kidney. Almost every patient who died showed at least an increase in serum creatinine levels, and the rate of need for dialysis or ultrafiltration was more than 30 times greater than in those who survived. In addition, although infectious complications were among the most common, only frank sepsis was significantly more common in those who died. Even serious wound infections and pneumonia did not herald a poor outcome.

Our results are in agreement with prior series that showed greater mortality following resection of larger portions of the liver. It is also not surprising that, among those undergoing resection for malignant disease, patients with hepatocellular carcinoma have poorer outcome than those with metastatic disease given the propensity of hepatocellular carcinoma to develop in the cirrhotic liver.

It is important to note that the range of MELD scores in this series is wide, but they are clustered in the low range, undoubtedly due to selection bias. Patients with greatly advanced liver disease are refused elective resection due to prohibitively high risk of morbidity and mortality predicted by any measure. However, in this study group, the mean MELD score in the Childs A group was 5.7 ± 3.3 while in the Childs B/C group it was 9.5 ± 7.3. As a normal MELD is 6, it is obvious that a large proportion of this group did not have severe liver disease. The low score and the large standard deviation in the Childs B/C group supports the contention that use of MELD in prognostication for patients without severe intrinsic liver disease is inappropriate even if they are undergoing hepatic procedures.

CONCLUSION

While MELD may be appropriate and useful in stratifying patients awaiting liver transplantation or undergoing TIPS procedures, it does not predict morbidity or mortality after elective liver resection. Both CTP and ASA, despite their shortcomings as subjectively clinical in nature, were superior in assessing outcome in these patients.

The shortcomings of this study are those common to all database research. As a retrospective review of prospectively collected data, the results should be confirmed by a prospective trial. This group includes a disproportionate number of male patients with high levels of comorbidity, factors that may limit the generalisability of the conclusions to other populations. If it is unsuccessful in this group of relatively sick patients, it is unlikely to be useful in the broader population.

ACKNOWLEDGMENTS

The authors thank the Chiefs of Surgery and the NSQIP Surgical Clinical Nurse Reviewers for their dedication and hard work in assuring the integrity of the NSQIP data.