METHODS

Between 1994 and 2000, a total of 1486 patients was operated on for primary gastric cancer in Chang Gung Memorial Hospital, Taipei. According to the UICC's classification of residual tumors,(13) surgery with no residual tumors (R0) or curative resection was achieved in 961 (64.7%) patients. Surgery with microscopic residual tumors (R1) was performed in 104 (7.0%), and surgery with macroscopic residual tumors (R2) in 421 (28.3%). For this study, we enrolled 522 (35.3%) patients who had received palliative surgery (R1 or R2). The reasons for R1 included positive resection margins in 85 (81.7%), and/or distant lymph node metastasis in 49 (47.1%). The R2 surgeries were further subgrouped into resection (R2-A) in 257 (17.3%) patients and non-resection (R2-B) in 164 (11.0%) patients. R2-A patients had liver metastasis in 49 (19.1%), peritoneal seeding in 181 (70.4%), distant lymph node metastasis in 25 (9.7%), a residual T4 tumor in 67 (26.5%), and positive resection margins in 65 (25.4%). Finally, the reasons for R2-B included liver metastasis in 28 (17.1%), peritoneal seeding in 87 (53.0%), and an unresectable T4 tumor in 125 (76.2%).

Surgery
Whenever feasible, gastric resection was our first choice of treatment for patients with gastric cancer. A curative (R0) resection was defined as the surgeon's intent to excise all macroscopic disease, leaving all histologic margins free of tumors. The resection included not only the cancer and surrounding normal stomach, but also the greater and lesser omenta and perigastric lymph nodes en bloc. An extended lymphadenectomy was required to remove lymph nodes around the stomach and those along the hepatic, splenic, and celiac arteries (the anatomic N2 level in the Japanese system).(14) The R1 resection operation was initially performed the same as for an R0 resection with a curative intent. However, because of technical difficulties, no further resection, such as a pancreaticoduodenectomy or subtotal esophagectomy, was conducted to gain a free margin. For palliative purposes, a systemic lymphadenectomy was not required in R2 resections. Non-resection operations consisted of gastrojejunostomies (N=64), gastrostomies (N=17), jejunostomies, (N=60), and laparotomies only (N=23). Gastrostomies and jejunostomies were studied as a group, which together were classified as an intubation operation.
The location of the tumor within the stomach determined the extent of gastric resection: tumors in the upper third were treated with a total gastrectomy; tumors in the body of the stomach were treated with either a total gastrectomy or distal gastrectomy; and tumors in the distal third of the stomach were treated with a distal gastrectomy. For a T4 tumor, combined resection of the invaded neighboring organs such as the transverse colon or the liver was performed. A distal pancreaticosplenectomy was not routinely performed, except when there was direct invasion into the body or tail of the pancreas or spleen.

Variables studied
Resected specimens were studied pathologically according to the criteria described in the Japanese General Rules for Gastric Cancer Study(13) and the UICC's pTNM classification.(14) The study items included age, gender, tumor location, tumor size, gross (Borrmann) type, wall invasion, resection margin, histologic type, lymph node metastasis, vascular invasion, lymphatic invasion, and perineural invasion. The histologic features were classified into 2 types: 1) differentiated or intestinal type, consisting of papillary and/or tubular adenocarcinomas, and 2) undifferentiated or diffuse type, consisting of poorly differentiated, signet-ring cell, and/or mucinous adenocarcinomas.
The postoperative events registered during hospitalization were used to determine morbidity and mortality. Operative mortality was defined as death during the same admission period. After discharge, all patients were receiving periodic follow-up study in the outpatient department until the time of this writing, or until their death. The palliative effect was expressed as 'the duration of palliation', and was defined as the period in which symptoms were relieved and during which parenteral fluid or nutritional therapy was not required.(11)

Postoperative chemotherapy
Chemotherapy was performed if the patient consented, and the performance status was less than or equal to 3 in the scoring system of the Eastern Cooperative Oncology Group. Therapy commenced within 1 month of the operation. Chemotherapy could be categorized into 5-fluorouracil (5-FU)-based and cisplatin-based regimens. The most frequently used regimen was a combination of 2600 mg/m2 5-FU and 150 mg leucovorin, which were infused simultaneously through a portable pump over a 24-h period once a week for 6 weeks with a 2-week break prior to repetition of treatment.(15) Chemotherapy was repeated every 8 weeks until disease progression, unacceptable toxicity, or patient refusal.

Statistical analysis
When appropriate, the Mann-Whitney U test or Fisher's exact test was used for between-group comparisons. Logistic regression was used to identify which of the clinicopathological variables were associated with operative mortality after gastric resection. Follow-up was carried out until December 2001 or until the patient expired. The survival of patients was expressed by median survival times and 95% confidence intervals (CIs). The cancer-specific survival outcome was expressed by applying the Kaplan-Meier method for all patients excluding those who died immediately after surgery. The log-rank test was used to compare the prognostic significance of individual variables on survival and the duration of palliation. A p value of < 0.05 was considered statistically significant. In pairwise multiple comparisons, however, the significance level, a, was adjusted to avoid a type I error and to retain an overall significance level of 0.05 by using the Bonferroni correction, where the adjusted a level was equal to 0.05 divided by the number of tests or comparisons. The adjusted a level is noted in the text wherever necessary.

RESULTS

Demographics of patients
The median age of all 1486 patients was 63.4 years, and the gender ratio (male: female) was 1.9:1; both parameters were very similar among the R0, R1, and R2 groups (p=0.826 and p=0.654, respectively). Table 1 presents a comparison of the clinicopathological characteristics of the 3 groups which underwent gastric resection. R2 patients without gastric resection (R2-B) were not included, for there were no surgical specimens for histopathological study. Compared to the R0 group, both the R1 and R2 groups presented a bigger tumor size, a greater percentage of upper-third location, grossly infiltrative type (Borrmann type 3 and 4), histologically undifferentiated type, vascular invasion, lymphatic invasion, and perineural invasions (p<0.001 for all, except for the percentage of upper location, p=0.001). The percentage of serosal invasion and lymph node metastasis was greater in the R1 and R2 than in the R0 group (p<0.001). The proportion of total gastrectomies was 26.2% and 32.7% for the R0 and R1 groups, while it was 41.6% for the R2 group (p<0.001). Combined spleen and/or pancreas resection was performed more frequently in the R1 and R2 than in the R0 group (25.8% and 25.9% vs. 13.7%, p<0.001). It was a concomitant procedure in 53.4% of patients that underwent a total gastrectomy.

Postoperative complications
Table 2 lists postoperative complications of gastrectomies in the curative group and the 3 palliative groups. The morbidity rate was significantly greater for patients after palliative surgery than after curative surgery (23.3% vs. 16.0%, p=0.002). However there were no differences among the 3 groups (R1, R2-A, and R2-B) of patients after palliative surgery (p=0.675). Anastomotic leakage and an intra-abdominal abscess were the 2 major complications after gastric resection. The incidence of anastomotic leakage was significantly greater in patients after palliative resection, especially total gastrectomies. The incidence of postoperative gastric stasis was especially greater in patients after gastrojejunostomies (p<0.001). The non-resection group had a higher incidence of postoperative bowel obstruction (p=0.002) due to progressive peritoneal carcinomatosis.

Operative mortality
Table 3 shows the operative mortality following various surgical procedures stratified by the UICC's residual tumor classification. The operative mortalities for R1 resection, and R2 resection and non-resection were 3.8%, 6.2%, and 17.1%, respectively, with significant difference among them (p<0.0001). The operative mortality did not significantly differ among R1 distal gastrectomies (4.5%), R2 distal gastrectomies (3.3%), and R1 total gastrectomies (2.9%) (p=0.919). However, R2 total gastrectomies showed a particularly higher operative mortality (10.9%) than other resection procedures (p= 0.00382). Combined pancreaticosplenectomies resulted in 2 surgical deaths (12.5%) following R2 resection (N=16), however there were no deaths following R0 (N=22) or R1 (N=4) resections (p= 0.182).
Mortality rates of the non-resection group were 23.4%, 10.9%, and 10.5%, respectively, after intubation surgery, gastrojejunostomies, and laparotomies only. The major cause of hospital deaths for the non-resection group was downhill progression of the disease, in contrast to surgical complications, mainly anastomotic leakage, for the resection group.

Risk factors for operative mortality with gastric resection
The surgical mortality of our patients (R0, R1, and R2, N=1322) who underwent gastric resection was closely related to severe medical diseases (presence/absence) (p=0.019), residual tumor classification (R0, R1, or R2) (p=0.017), pathological stage (stage III, IV/stage I, II) (p=0.013), peritoneal seeding (presence/absence) (p=0.001), extent of resection (total gastrectomy/distal gastrectomy) (p=0.024), and very old age (M 80 years/< 80 years) (p<0.001). It was inversely correlated with lymph node dissection (yes/no) (p=0.046). It was not correlated to tumor location, tumor size, gross type, serosal invasion, lymph node metastasis, positive margin, or combined resection of adjacent organs. Logistic regression analysis revealed that the most important risk factors for operative mortality with gastric resection were peritoneal seeding, severe medical disease, and very old age (risk ratios of 3.39, 2.05, and 1.87, respectively) (Table 4).
The three most important risk factors were compared in the 4 groups of patients undergoing resection and included an R1 distal gastrectomy, R2 distal gastrectomy, R1 total gastrectomy, and R2 total gastrectomy. There were no significant differences in the proportions of very old patients or severe medical diseases in the 4 groups (p=0.939 and p=0.309, respectively). A significantly higher proportion of peritoneal seeding was noted in the group with R2 total gastrectomies than in the other 3 groups (p<0.001).

Survival outcomes
Figure 1 illustrates the cumulative survival curves of the 4 groups of patients according to the residual tumor classification. For the R1, R2-A, and R2-B groups, the median survival times were 18.8 (95% CI, 13.9-23.7), 9.5 (95% CI, 7.7-11.4), and 5.3 months (95% CI, 3.7-6.9), respectively, whereas for the R0 group it was 59.7 months (95% CI, 51.7-67.7). Differences in survival rates between any 2 of the 4 groups were statistically significant (log-rank p<0.0001, R1 or R2-A or R2-B vs. R0; log-rank p<0.0001, R2-B vs. R1; log-rank p=0.0004, R2-A vs. R1; Bonferroni's adjusted a level=0.0167).
Figure 2 illustrates survival curves of patients after palliative surgery. The median survival times after distal gastrectomies and total gastrectomies were 11.3 (95% CI, 9.6-13.0) and 7.1 months (95% CI, 6.0-8.2), respectively. Median survival times after gastrojejunostomies, intubation, and laparotomies only were 6.2 (95% CI, 3.5-8.9), 3.8 (95% CI, 2.1-5.5), and 6.6 months (95% CI, 3.0-10.2), respectively. Patients who underwent distal gastrectomies survived significantly longer than those who had total gastrectomies, laparotomies only, or intubation procedures (log-rank p=0.0006, <0.0001, and <0.0001, respectively), and nonsignificantly longer than those who had gastrojejunostomies (log-rank p=0.014; Bonferroni's adjusted a level=0.0125). Survival times of patients did not differ among total gastrectomies, gastrojejunostomies, and laparotomies only (total gastrectomies vs. gastrojejunostomies, log-rank p=0.125; total gastrectomies vs. laparotomies only, log-rank p=0.722; gastrojejunostomies vs. laparotomies only, log-rank p=0.454). The survival time after intubation was significantly shorter than those after distal gastrectomies or total gastrectomies (log-rank p<0.0001 and=0.0001, respectively), but non-significantly shorter than those of gastrojejunostomies or laparotomies only (log-rank p=0.019 and =0.021, respectively; Bonferroni's adjusted a level=0.0125).
The median survival time of patients with liver metastasis (N=77) was 6.9 months (95% CI, 4.6-9.2). Among them, patients (N=49) who underwent gastric resection survived longer than those (N=28) who underwent non-resection operations (median survival time: 11.1 vs. 4.0 months, log-rank p<0.0001). Combined resection of liver metastasis (N=19) did not offer a longer survival time than leaving the liver metastasis alone (N=58) (median survival time: 12.6 vs. 11.1 months, log-rank p=0.577). The median survival time of patients with peritoneal seeding (N=258) was 7.5 months (95% CI, 6.3-8.7). Patients with peritoneal seeding survived longer after gastric resection (N=181) than did those after non-resection operations (N=77) (median survival time: 9.5 vs. 4.3 months, respectively, log-rank p<0.0001).

Survival benefits of chemotherapy
Fifty-six (53.8%) R1 patients, 102 (39.7%) R2-A patients, and 42 (25.6%) R2-B patients received postoperative chemotherapy. Figure 3 illustrates the cumulative survival curves of patients who underwent palliative surgery with or without postoperative chemotherapy. The median survival time of patients who received postoperative chemotherapy was 14.1 months (95% CI, 12.2-16.0), in contrast to 6.7 months (95% CI, 5.5-7.9) for those who did not (log-rank p<0.0001). If stratified into R1, R2A, and R2B, patients who received postoperative chemotherapy survived significantly longer than those who did not (log-rank p=0.010 for R1, p<0.0001 for R2-A, and p=0.002 for R2-B).

Duration of palliation
Regarding the duration of palliation, the median time was 3.6 months (95% CI, 0.4-6.8) in the gastrojejunostomy group and 1.2 months (95% CI, 0.3-3.4) in the intubation group (log-rank p=0.054). The median durations of palliation after distal and total gastrectomies were 8.9 (95% CI, 7.0-10.8) and 4.9 months (95% CI, 3.9-5.9), respectively (long-rank p=0.170). The median duration of palliation after distal gastrectomies was significantly longer than that after any non-resection procedures (log-rank p<0.0001). Total gastrectomies had no significant palliative benefits compared to gastrojejunostomies (log-rank p=0.054).