Short‐term outcomes of open liver resection and laparoscopic liver resection: Secondary analysis of data from a multicenter prospective study (CSGO‐HBP‐004)

Abstract The aim of the present study was to compare short‐term outcomes of laparoscopic and open liver resection (LLR and OLR, respectively), and we first analyzed a preoperatively enrolled and prospectively collected database. We carried out a secondary analysis using a preoperative enrolled database that included the details of 786 patients who had been enrolled in a previously carried out randomized controlled trial to assess short‐term outcomes, including morbidities. Statistical analyses included logistic regression, propensity score matching (PSM) with replacement, and inverse probability of treatment weighting (IPTW) analyses. Among 780 liver resections, OLR was carried out in 543 patients and LLR was carried out in 237 patients. LLR was selected in patients with a worse liver function and was related to a smaller resected liver weight and/or partial resection. Logistic regression, PSM, and IPTW analyses revealed that LLR was associated with less blood loss and a lower incidence of morbidities, but a longer operating time. LLR was found to be a preferred factor in biliary leakage by IPTW only. LLR was a preferred factor for blood loss, morbidities and hospital stay, but was associated with a longer operating time. UMIN‐CTR, UMIN000003324.


| INTRODUCTION
Laparoscopic surgical techniques have recently been applied to liver resection, [1][2][3][4][5] despite the fact that their feasibility remains controversial. Although several randomized controlled trials (RCT) have been carried out to investigate the usefulness of laparoscopic techniques in gastric and colorectal surgery, [6][7][8] laparoscopic liver resection (LLR) has a relatively short history and the surgical techniques are still under development. As a consequence, it remains difficult to control quality in RCT. Thus, most studies on LLR and open liver resection (OLR) have been retrospective in nature and have analyzed a relatively small number of patients at a single hospital, [9][10][11] whereas other studies have used a nationwide database of postoperatively enrolled patients. [12][13][14] Thus, the potential for selection bias, enrolment bias, and missing values could not be avoided.
We previously carried out a multicenter RCT in which the endpoint was short-term surgical outcome. 15 Patients in the database were classified according to the method that was used to seal the liver cut surface. Briefly, over 700 patients from 11 institutes were enrolled in the RCT. Results showed that incidence of postoperative bile leakage and bleeding among the methods of sealing did not differ to a statistically significant extent. LLR accounted for approximately one-third of the procedures that were included in the database. All of the patients were enrolled preoperatively. Perioperative factors that were recorded in the database included: liver function, hepatitis, type of resection, operating time, blood loss, resected liver weight, detailed morbidities, and hospital stay.
We are of the opinion that the database would be useful for analysis of short-term outcomes of patients undergoing LLR and OLR for several reasons: patients were enrolled preoperatively, short-term surgical outcome and all morbidities were collected prospectively, and data were obtained from a multi-institutional study and therefore showed universality. Although an RCT is necessary to make a precise comparison, the results of the analyses in the present study would provide a useful rationale for the carrying out an RCT.
The present study shows how patients were selected for LLR and compares short-term outcomes between OLR and LLR. Propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) analyses were used to reduce selection bias.
Results showed that LLR was associated with less blood loss and a lower incidence of morbidities, but a prolonged operating time.
These results provide information that can be used until completion of an RCT. The results are useful for determining the indications for LLR and for obtaining informed consent from patients in whom LLR is indicated.

| Study design
In the present study, we aimed to carry out a secondary analysis of the data obtained in a previous open, multicenter RCT that was carried out to explore the efficacy of fibrin sealant (FS) with polyglycolic acid (PGA) versus fibrinogen-based collagen fleece (CF) in preventing postoperative biliary leakage and/or hemorrhage at the liver cut surface. 15  As a result of the lack of an international definition of biliary leakage 16 when planning this RCT, we defined biliary leakage as a drain bilirubin to serum bilirubin ratio of ≥5. When the ratio was 3 to <5, we re-measured the drain and serum bilirubin levels after 2 or 3 days. Postoperative hemorrhage was defined by the need for relaparotomy or transfusion to achieve hemostasis.
Patient selection was carried out as follows. Patients in whom hepatectomy was planned and who were ≥20 years of age were enrolled in the RCT and preoperatively assigned. Type of resection

| Statistical analysis
Statistical analyses were carried out according to the flow diagram in Figure 1. Methods of the analyses are described below.
The data were expressed as the mean. Differences between groups were tested using Student's t test or the chi-squared test, as appropriate. P values of <.05 were considered to indicate statistical significance. A multivariate logistic regression analysis was performed. PSM analysis was performed with replacement to increase the average quality of matching and to decrease bias. 18,19 IPTW was used to adjust for differences and reduce the impact of any treatment selection bias. 20 With this method, the weights for patients who were treated with LLR were the inverse of the propensity score (determined by logistic regression); the weights for patients who underwent OLR were the inverse of the 1-propensity score. Logistic regression was used to estimate the propensity scores. The following variables were included in the model: type of resection, type of resection, age, gender, platelet, bilirubin, albumin, prothrombin time, HBs-antigen positivity and HCV-antibody positivity. To visualize hazard ratio of surgical outcomes and similarity between analyses, we used a forest plot. All of the statistical analyses were conducted using the R software program (version 2.15.2, Foundation for Statistical Computing, Vienna, Austria, http://www.rproject.org).

| Patient flow
A total of 786 patients were enrolled in the present study. Among these patients, 780 patients underwent hepatectomy and were analyzed in the present study ( Figure 1). OLR was carried out in 543 patients and LLR was carried out in 237 patients. We first carried out a logistic regression analysis using all of the patient data. Next, we carried out a PSM analysis after confirming that the C-index was >.8 (C-index: .8093). PSM analysis was done with replacement to increase the average quality of matching and to decrease bias. 18,19 Two hundred and eighty-six patients from each group were included in the PSM analysis by replacement. Finally, we carried out an IPTW analysis to reduce selection bias and to analyze all of the data in the database.

| Demographic characteristics of the patients who underwent OLR and LLR
We summarized the demographic characteristics of the patients who underwent OLR and LLR (Table 1). In the trial database, platelet count and prothrombin time of the LLR group was lower, whereas the rate of HCV positivity was higher in comparison to the OLR group. Regarding type of liver resection, the rate of partial resection was higher in the LLR group. Liver resection weight, estimated blood loss, and duration of postoperative hospital stay in the LLR group were lower in comparison to the OLR group. Rates of biliary leakage and other postoperative adverse events were lower in the LLR group.
In summary, laparoscopic resection was selected in patients with a worse liver function (lower platelet count, lower prothrombin time, and HCV positivity). Laparoscopic operative procedures were related to a smaller liver resection weight and/or partial resection.

| Surgical outcomes of OLR and LLR (analyzed by logistic regression, PSM, and IPTW)
We analyzed estimated blood loss, operating time, biliary leakage, other adverse events and duration of postoperative hospital stay as the surgical outcomes of liver resection. Biliary leakage was defined according to the definition of the International Study Group of Liver Surgery (ISGLS). 16 With regard to blood loss, logistic regression analysis showed that LLR, albumin, and partial resection were associated with less estimated blood loss (Figure 2). After PSM, LLR remained a preferred factor ( Table 2). In the IPTW analysis, LLR, albumin, HCV positivity, and partial resection were preferred factors (Figure 3).
With regard to operative time, logistic regression analysis showed that age, albumin, HCV positivity, and partial resection were associated with a shorter operative time, whereas LLR was associated with a longer operative time (Figure 2). After PSM, LLR was still associated with a longer operative time (Table 2). In IPTW analysis, age, albumin, HBs-positivity, and partial resection were preferred factors, whereas LLR remained associated with a longer operative time (Figure 3).  (Table 2). In IPTW analysis, LLR, albumin, and partial resection were preferred factors (Figure 3). Regarding biliary leakage, in logistic regression analysis, male sex and partial resection were preferred factors (Figure 2). After PSM, LLR was not associated with biliary leakage ( Table 2). In IPTW analysis, LLR, male sex, and partial resection were preferred factors (Figure 3).
Regarding length of hospital stay, LLR and partial resection were preferred factors. After PSM, LLR remained a preferred factor (Table 2). In IPTW analysis, LLR and partial resection were preferred factors ( Figure 3).
In summary, LLR was a preferred factor for estimated blood loss, adverse events and hospital stay. However, the relationship between LLR and biliary leakage depended on the type of analysis. LLR was associated with a longer operative time.

| DISCUSSION
Numerous retrospective studies have investigated the short-term outcomes of LLR. 21,22 Some were case-control studies that analyzed more than 200 patients 23,24 ; others used PSM to analyze nationwide large-scale databases. [12][13][14] However, retrospective studies are associated with certain limitations: they might lack cases and/or information, and they involve selection and enrolment biases.
Thus, retrospective studies risk underestimating the incidence of morbidities.
In the present study, we used a preoperatively enrolled and prospectively collected database to investigate perioperative morbidities; which there was no biliary leakage according to our definition, the drainage tube could be removed without any complications. However, the incidence of biliary leakage according to this definition was too low to analyze in the multivariate analysis. We therefore used the definition of the ISGLS in the present study.
Next, we compared the short-term outcomes observed in the present study with those of previous reports (Table 3). 9,13,14,[26][27][28][29][30][31][32] Almost all of the reports showed less blood loss and a shorter hospital stay; however, there were discrepancies among the reports with regard to operating time and incidence of morbidities. As Table 3 shows, in most of the previous reports, less than 100 patients underwent LLR. One report analyzed over 300 LLR patients 13