Comparison of score-based prediction of 90-day mortality after liver resection

Background Indications for liver surgery are expanding fast and complexity of procedures increases. Preoperative mortality risk assessment by scoring systems is debatable. A previously published externally validated Mortality Risk Score allowed easy applicable and precise prediction of postoperative mortality. Aim of the study was to compare the performance of the Mortality Risk Score with the standard scores MELD and P-POSSUM. Methods Data of 529 patients undergoing liver resection were analysed. Mortality Risk Score, the labMELD Score and the P-POSSUM Scores (PS, OS, P-POSSUM mortality %) were calculated. The ROC curves of the three scoring systems were computed and the areas under the curve (C-index) were calculated using logistic regression models. Comparisons between the ROC curves were performed using the corresponding Wald tests. Results Internal validation confirmed that the risk model was predictive for a 90-day mortality rate with a C-index of 0.8421. The labMELD Score had a C-index of 0.7352 and the P-POSSUM system 0.6795 (PS 0.6953, OS 0.5413). The 90-day mortality rate increased with increasing labMELD values (p < 0.0001). Categorized according to the Mortality Risk Score Groups the labMELD Score showed a linear increase while the POSSUM Scores showed variable results. Conclusions By accurately predicting the risk of postoperative mortality after liver surgery the Mortality Risk Score should be useful at the selection stage. Prediction can be adjusted by use of the well-established labMELD Score. In contrast, the performance of standard P-POSSUM Scores is limited.


Background
Indications for liver surgery are expanding fast and the complexity of procedures has generally increased over the years. The preoperative assessment of the safety of liver surgery and identification of high-risk patients remains challenging. Multiple patient-related factors and procedure-specific variables directly influence the perioperative outcome as well as the 90-day mortality after hepatectomy.
To provide appropriate counseling to the patient regarding the risks and outcome of the planned procedure is the crucial point within a shared decision-making process. Quality of life after surgery has increasingly become an important patient-related endpoint after surgical interventions. Nevertheless, the surgeons' intuition is not the best when it comes to predicting the outcome after liver resection and "experience-based" indications are no longer acceptable [1,2].
Various scoring systems such as MELD Score, E-PASS, mE-PASS, Portsmouth-Physiological and Operative Severity Score for the enUmeration of Mortality and morbidity (P-POSSUM), Child-Turcotte-Pugh score or ASA classification have been used in an attempt to accurately predict mortality in liver surgery. However, the results were disappointing. General surgical risk models (E-PASS, mE-PASS and P-POSSUM) tended to overestimate the mortality rate by more than twofold [3]. For prognostication of patients with known liver disease the utility of MELD Score has been well documented. However, heterogenic reports are available regarding its ability to predict morbidity or mortality after elective liver surgery [4,5]. The P-POSSUM Score is now the most widely validated predictive scoring system used in perioperative care. P-POSSUM aims to calculate the risk of a surgical procedure based on the patient's physiological condition and operative parameters. In hepatobiliary-pancreatic surgery the predictive value of P-POSSUM is under debate [3,6,7]. Previously, we suggested an easy applicable externally validated scoring system for 90-day mortality [8]. The aim of the current study is to compare its performance with already established scoring systems.

Methods
Data of 529 patients undergoing liver resection between 2014 and 2016 at the Department of General and Transplant Surgery, University Hospital Heidelberg were analysed. The institution's liver resection database contained anonymous information for about 3447 cases performed between 2001 and 2018, which represented more than 99% of all liver resections performed in the department. Patients who underwent liver resection for hepatobiliary trauma as well as after liver transplantation and patients who underwent resections in conjunction with other operations (such as pancreaticoduodenectomy, laparoscopic resections or the deroofing of simple or parasitic cysts, cystectomy and necrosectomy) were excluded from the analysis. Resections performed in patients younger than 18 years of age were also excluded. Written informed consent was obtained by every patient included in the analysis and the study was approved by the ethical committee of the Medical Faculty of Ruprecht-Karls-University Heidelberg S-557/2017. For all patients the previously suggested Mortality Risk Score, the labMELD Score and the P-POSSUM Score (PS, OS, P-POSSUM mortality %) were calculated.
As described previously the Mortality Risk Score was composed of 8 different parameters. One point was assigned for each of the following factors: patient's age ≥ 60 years, the performance of right trisectionectomy, preoperative INR ≥ 1.1 and preoperative GGT values ≥60 U/l. Absence of these factors resulted in an assignment of 0 points. Two points were assigned for a preoperative platelet count value < 120/nl (preoperative platelet count values ≥120/nl, 0 points). Three points were assigned for a preoperative creatinine value ≥2 mg/dl (preoperative creatinine values < 2 mg/dl, 0 points). The histological diagnosis was stratified according to a 2-point scale; intrahepatic cholangiocarcinoma (1 point), perihilar cholangiocarcinoma (2 points) and all other diagnoses (0 points). ASA classification was stratified according to: ASA IV (5 points), ASA III (1 point) and ASA I or II (0 points). Based on this the patients could be divided into four different Risk Score Groups. Risk Score Group 1 (very low-risk, 0-1 points), Risk Score Group 2 (low-risk, 2-3 points), Risk Score Group 3 (medium-risk, 4-5 points), and Risk Score Group 4 (high-risk, ≥ 6 points) [8]. The labMELD Score was calculated using serum bilirubin, serum creatinine and International Normalized Ratio (INR) according to the following formula: LabMELD = 10x (0.957 x ln (serum creatinine mg/dl) + 0.378 x ln (serum bilirubin mg/dl) + 1.120 x ln (INR) + 0.643). Serum creatinine was capped at 4 mg/dl. For patients receiving dialysis at least 2 times in the previous week the creatinine value was set to 4 mg/dl. The value was rounded to the nearest integer. In order to avoid negative scores, the lower limit for each component of the score was set to 1. LabMELD Score values between 6 and 40 were possible [9][10][11][12]. The P-POSSUM equation for in-patient mortality was according to Prytherch et al.: ln (R/(1-R)) = − 9.065 + (0.1692 x physiological score) + (0.1550 x operative severity score). R is the predicted risk of mortality [13]. The score points of the physiological score and the operative severity score were awarded on the basis of the original research study of Copeland et al. [14] For the study population the 12 physiological parameters (age, cardiac history, respiratory history, systolic blood pressure, pulse rate, Glasgow Coma Scale, hemoglobin level, white cell count, urea concentration, sodium level, potassium level and electrocardiography) and the six operative parameters (operative severity, multiple procedures, total blood loss, peritoneal soiling, presence of malignancy and mode of surgery) were analysed and the physiological score (PS), operative severity score (OS) and P-POSSUM mortality % were calculated. For all patients the operative severity was set to major.

Statistical analysis
SAS software (Release 9.4, SAS Institute, Cary, NC, USA) was used for statistical data analysis. Categorical parameters were presented as absolute and relative frequencies. Differences between the 90-day mortality subgroup and the rest of the total cohort were analysed using Fisher's exact test. Continuous parameters were expressed as the median, the interquartile range (IQR), the mean and the standard deviation (SD). With respect to the continuous parameters, comparisons between the 90-day mortality subgroup and the rest of the total cohort were performed using the Mann-Whitney U test. Logistic regression analyses were performed in order to examine the categorized Risk Score and the categorized MELD Score with respect to the 90-day mortality. Odds ratios (OR) with their corresponding 95% confidence intervals (CI) are presented here. The receiver operating characteristic (ROC) curves of the three scoring systems Mortality Risk Score, MELD Score, and P-POSSUM Score were computed and the areas under the curve (known as the C-index) were calculated using logistic regression models. Comparisons between the ROC curves were performed using the corresponding Wald tests from the underlying logistic regression models. Twosided p-values of less than 0.05 were considered statistically significant.

Comparison with labMELD score and P-POSSUM scores
The median labMELD Score of the study population of the 529 patients was 7.0 (IQR 6-8) and the mean lab-MELD Score was 7.5 (SD 2.7). The 90-day mortality rates according to the labMELD Score were as following: labMELD Score 6 1.7% (n = 4 of 236, OR1), labMELD Score 7 3.4% (n = 5 of 148, OR 2.03, 95% CI 0.53-8.31,  Table 3. The labMELD Score was predictive for a 90-day mortality rate with a C-index of 0.7352. The P-POSSUM scoring system consists of two parts, the physiological score (PS) and the operative severity score (OS). The mortality rate is defined as in-patient mortality. The median PS was 16.0 (IQR 14-18.8) and the mean PS was 17.1 (SD 4.5). The median OS was 17.0 (IQR [13][14][15][16][17][18][19] and the mean OS was 16.2 (SD 4.2). The median mortality rate of the P-POSSUM scoring system amounted to 2.3% (IQR 1.4-4.4) and the mean mortality rate amounted to 4.3% (SD 7.3). The C-value for AUC for the prediction of in-patient mortality for the PS part of P-POSSUM scoring system was 0.6953, for the OS part 0.5413 and for the mortality rate of the P-POSSUM system 0.6795 Table 4 Fig. 1.

Discussions
Selection criteria for hepatectomy and prediction of the individual mortality risk are debatable. This study aimed at evaluating prognostic models of postoperative mortality in patients undergoing hepatectomy. By accurately predicting the risk of 90-day mortality the Mortality Risk Score should be useful at the selection stage. Using only preoperatively available parameters it is easy applicable and does not need subjective judgement of patients' condition.
The current study identifies a precise prediction of the postoperative mortality risk by the previously suggested Mortality Risk Score compared to the labMELD and P-POSSUM scoring systems. Depending on the individual risk profile an increasing risk for 90-day mortality could be identified with high predictive power. In contrast, the labMELD Score showed quite heterogenic results with medium predictive power. One has to recognize that despite increasing score points the mortality risk was evaluated to be not linear increasing as one might have suspected. The predictive power of the P-POSSUM Score was low with the OS clearly underperforming.
When the labMELD Scores, the PS and OS as well as the whole P-POSSUM system were categorized according to the suggested Mortality Risk Score groups it could be demonstrated that the median labMELD as well as the median PS/P-POSSUM Score values increase gradually within the different risk groups and were able to dynamically reflect low risk to high risk patients. However, further studies are needed; especially on eastern populations.
In summary, the trend of expanding the pool of patients undergoing liver resection towards high risk surgical patients continues. The 90-day mortality rate improved over the past decade but still varies across cohorts. Overall, risk scores will never replace comprehensive preoperative assessment and proper patient selection criteria. However, they might help to estimate the individual risk profile and support planning of a risk adapted operation strategy including pre-treatment to enhance the functional liver reserve, staged resections or combined surgical and interventional radiological concepts to reduce the postoperative morbidity and mortality rates. This will not only have significant implications on the patients' quality of life but also on healthcare-economics.

Conclusions
The Mortality Risk score shows a precise prediction of the risk of postoperative mortality after liver surgery and should be used at the selection stage. The wellestablished labMELD Score estimates the postoperative mortality risk with a medium predictive power, the standard P-POSSUM Scores fail in predicting the 90-day mortality after liver resection.