Multivariate logistic regression for in-hospital mortality after gastric resection for neoplasm. Data are represented as adjusted odds ratios (triangles) with surrounding 95% confidence intervals.
Smith JK, McPhee JT, Hill JS, Whalen GF, Sullivan ME, Litwin DE, Anderson FA, Tseng JF. National Outcomes After Gastric Resection for Neoplasm. Arch Surg. 2007;142(4):387-393. doi:10.1001/archsurg.142.4.387
That factors affecting outcomes of surgical resection in the treatment of gastric cancer can be identified using a large US database.
Retrospective observational study.
The Nationwide Inpatient Sample from January 1, 1998, through December 31, 2003.
We included 13 354 patient discharges (approximately 66 096 nationally by weighted analysis) who underwent gastric resection for neoplasm.
Main Outcome Measure
In-hospital mortality. Univariate analyses were performed by means of χ2 tests. A multivariate logistic regression was performed to determine which variables were independently predictive of in-hospital mortality.
During the study period, 50 738 patients (approximately 250 420 nationally) were discharged with the diagnosis of gastric neoplasm. Of those, 13 354 (26.3%) underwent gastric resection during their hospitalization. In-hospital mortality for patients undergoing surgery was 6.0%, without significant change from 1998 through 2003. Factors predictive of significantly increased in-hospital mortality included low annual hospital surgical volume (lowest [≤4 gastrectomies per year] vs highest [≥11 gastrectomies per year], 6.8% vs 4.9%; adjusted odds ratio [OR], 1.5; 95% confidence interval [CI], 1.2-1.8]), older patient age (50-69 vs <50 years, 4.0% vs 2.1%; adjusted OR, 1.5; 95% CI, 1.1-2.2) (≥70 vs <50 years, 8.6% vs 2.1%; adjusted OR, 2.9; 95% CI, 2.0-4.3), male sex (male vs female, 6.7% vs 5.0%; adjusted OR, 1.3; 95% CI, 1.1-1.5), and procedure type (total gastrectomy vs all other resections, 8.0% vs 5.3%; adjusted OR, 1.4; 95% CI, 1.2-1.7).
Higher annual surgical volume is predictive of lower in-hospital mortality for patients undergoing gastric resection for neoplasm. Other factors significantly associated with superior outcomes after gastric resection included diagnosis type, procedure type, younger age, female sex, and fewer comorbid conditions.
Gastric cancer is the second leading cause of cancer-specific mortality worldwide and the eighth leading cause in the United States. The American Cancer Society estimated that stomach cancer would have an incidence of 22 280 cases and would cause 11 430 deaths in the United States in 2006.1
Surgical resection is currently the only treatment option for gastric cancer with curative potential. Options for surgical treatment may depend on factors including the anatomic location of the tumor, histologic type, stage of disease, and patient's comorbid conditions. Gastrectomy confers the greatest likelihood for survival in the treatment of invasive gastric cancer.2 The type of gastrectomy remains an area of some debate; recent investigations have suggested that total gastrectomy offers no survival advantage over distal or proximal subtotal resections based on the anatomic location of the tumor.3 A regional study noted an overall decrease in operative mortality rates for gastric cancer resections, including total gastrectomy, extended total gastrectomy, distal gastrectomy, and proximal esophagogastrectomy, during the 20 years from 1976 through 1995.4 In contrast, a more recent national analysis of operative mortality trends from 1994 through 1999 found that the operative mortality rates of gastrectomy did not decline during the period studied.5
Hospital volume has been identified as having a significant influence on surgical outcome; several studies have demonstrated an inverse relationship between operative mortality and hospital surgical volume for high-risk procedures, including major cancer resections.6- 9 In light of this relationship, it has been proposed that implementation of volume standards could prevent many surgical deaths.10 Others have argued that the volume-outcome evidence is still incomplete, and a better understanding is needed of the factors that contribute to the apparent relationship between high volume and better outcome.11 Other predictive factors of operative mortality that have been identified include age and type of resection.12
Many studies examining surgical outcomes for gastric cancer are limited by the number of subjects or by a single-institution design. The goal of this study was to use the Nationwide Inpatient Sample (NIS), a national administrative database, to analyze contemporary in-hospital mortality rates for gastric resections for neoplasm. We examined the relative effects of predictors of in-hospital mortality, including hospital volume, patient age and sex, diagnosis, procedure, and comorbid conditions. We further examined the contribution of these factors to possible outcome differences among high-, medium-, and low-volume hospitals.
To evaluate trends in gastric resection for neoplastic disease at the national level, we used the NIS for January 1, 1998, through December 31, 2003. The NIS is the largest national all-payer hospital inpatient care database in the United States. Data exist for approximately 7 million hospital discharges per year from a stratified sample of 20% of nonfederal US community hospitals from participating states (n = 37 in 2003), including academic and specialty hospitals.13
The NIS is supported by the Healthcare Cost and Utilization Project and contains all-payer discharge information for 100% of patient discharges from participating hospitals. A weighting strategy is implemented by the NIS to allow population-based estimates to be drawn at the national level. Sampled hospitals are given appropriate weights based on the number of hospitals they represent in the database for a given year.13
To identify all patient discharges with the principal diagnosis of gastric neoplasm, we used selected codes from the ().14 We limited the initial cohort to patients undergoing gastric resections based on procedure codes. The diagnosis and procedure codes used in the analysis are listed in Table 1.
To evaluate annual surgical volume, we divided the participating hospitals into 3 groups. Hospitals that performed a mean of 4 or fewer gastric resections per year were classified as low-volume centers; more than 4 to less than 11 gastric resections per year, as medium-volume centers; and 11 or more gastric resections per year, as high-volume centers. The cutoff values of 4 and 11 resections were chosen because they evenly divided the national distribution of annual hospital volume for gastrectomy into thirds, a technique previously described.15,16
The primary measured outcome for this study was in-hospital mortality. Mortality was defined as death from any cause prior to discharge, regardless of the time from the operation.
We analyzed all data using the advanced survey procedures in SAS statistical software (version 9.1; SAS Institute Inc, Cary, NC). Univariate analyses of categorical variables were performed using Rao-Scott χ2 tests, with <.05 considered statistically significant. Trend analyses were performed using the Mantel-Haenszel χ2 test. A multivariate logistic regression was constructed with in-hospital mortality as the dependent variable. Independent variables included patient age group (<50, 50-69, and ≥70 years), sex, year of resection, type of resection, neoplasm diagnosis, and various comorbid medical conditions (congestive heart failure [CHF], renal failure, diabetes mellitus, and liver disease).17 Although a total of 30 comorbid conditions are listed for use with the administrative data sets,17 many are not readily applicable or important to surgical patients or were only recorded in a few patients. Therefore, CHF, renal failure, diabetes mellitus, and liver disease were chosen to be analyzed as a representative sample with clinical applicability to surgical procedures. In addition, certain hospital-level characteristics were included, such as hospital surgical volume (low, medium, or high), hospital type (teaching or nonteaching), and geographic region. Adjusted odds ratios (ORs) were calculated to determine the effects of the identified covariates on the outcome measure of in-hospital mortality.
From 1998 through 2003, 50 738 patient discharges (representing 250 420 patients nationally) occurred with the principal diagnosis of gastric neoplasm. All further data references list weighted numbers parenthetically to also highlight the data representative of the national trends and differences. Of this initial cohort of patients, 13 354 (26.3%) (representing 66 096 patients nationally) underwent gastric resection during their hospitalization. Of the patients undergoing surgery, 7996 (59.9%) (representing 39 605 patients nationally) were male. The mean age for patients undergoing surgery was 67 (range, 18-105) years; 11.6% were younger than 50 years; 39.4% were aged 50 to 69 years; and 49.1% were 70 years or older. Patient demographic characteristics are summarized in Table 2.
For all patients undergoing gastric resection, in-hospital mortality was 6.0%. Male patients were found to have significantly increased in-hospital mortality. Increasing age was also associated with increased mortality. Significant differences in in-hospital mortality were also found across diagnoses. Table 3 summarizes the types of procedures patients underwent based on diagnosis codes. The highest rate of mortality was found among patients undergoing resection who had been diagnosed as having abdominal esophageal tumors (15.5% in-hospital mortality) compared with patients who had benign neoplasms (1.1% in-hospital mortality) (overall <.001) (Table 4).
Gastrectomy/Billroth II was the most frequently performed operation (36.4%). The highest rates of in-hospital mortality were found among patients undergoing total gastrectomy with or without intestinal transposition, including esophagogastrectomy (8.0%), gastrectomy/Billroth II (6.0%), and proximal gastrectomy with esophagostomy (5.9%), compared with patients undergoing distal gastrectomy/gastropylorectomy-Billroth I (3.2%) (overall <.001) (Table 4).
Certain comorbid conditions were also found to be predictive of increased in-hospital mortality, including CHF, renal failure, and liver disease.
Comparisons of patient characteristics across surgical volume levels demonstrated that patients treated at hospitals with higher surgical volumes were slightly younger compared with the 2 lower-volume groups (mean ages, 65.6 vs 68.5 and 67.2 years; <.001). Patients at high-volume hospitals were also found to have a lower percentage of certain comorbid conditions, specifically CHF. Compared with the 2 lower-volume groups, high-volume centers surgically treated a higher proportion of patients diagnosed as having gastroesophageal junction tumors (28.5% vs 15.5% and 21.4%) and a lower proportion of patients with gastric body tumors (43.5% vs 47.4% and 47.5%), antrum/pyloric/prepyloric neoplasms (16.0% vs 22.9% and 18.9%), and benign neoplastic disease (8.8% vs 11.4% and 9.2%) (overall <.001). High-volume hospitals performed a greater number of total gastrectomies (32.7% vs 23.1% and 28.0%) and proximal gastrectomies (14.5% vs 9.0% and 11.6%) and fewer gastrectomies/Billroth II (30.8% vs 42.5% and 36.0%) (overall <.001) (Table 5).
Using a multivariate logistic regression model, we evaluated independent predictors of in-hospital mortality (Figure). Lower annual surgical volume was found to be predictive of significantly increased in-hospital mortality for patients undergoing gastric resection (lowest vs highest volume, 6.8% vs 4.9%; adjusted OR, 1.5; 95% confidence interval [CI], 1.2-1.8). Male sex was associated with significantly increased in-hospital mortality (male vs female, 6.7% vs 5.0%; adjusted OR, 1.3; 95% CI, 1.1-1.5). Increasing age was also predictive of increased mortality (50-69 vs <50 years, 4.0% vs 2.1%; adjusted OR, 1.5; 95% CI, 1.1-2.2) (≥70 vs <50 years, 8.6% vs 2.1%; adjusted OR, 2.9; 95% CI, 2.0-4.3). Among procedures, total gastrectomy was found to be predictive of significantly increased in-hospital mortality (total gastrectomy vs all other resections, 8.0% vs 5.3%; adjusted OR, 1.4; 95% CI, 1.2-1.7). Comorbid conditions found to be associated with significantly increased in-hospital mortality included CHF (adjusted OR, 2.4; 95% CI, 2.0-2.9), renal failure (adjusted OR, 6.3; 95% CI, 4.5-8.8), and liver disease (adjusted OR, 2.5; 95% CI, 1.6-3.9) (Table 6).
In this study, we used representative US data from the NIS to demonstrate that the overall in-hospital mortality after gastric resection for neoplasm was 6.0% from 1998 through 2003. Our analysis shows that a number of factors affect the in-hospital mortality rate for gastrectomy. Factors predictive of significantly increased mortality included male sex, older patient age, diagnosis type, gastrectomy type, and low annual hospital surgical volume. On univariate analysis, men had an approximately one-third excess risk of in-hospital mortality compared with women (6.7% vs 5.0%). Patients in the oldest group (≥70 years) had a more than 4-fold risk of death compared with the youngest patients (8.6% vs 2.1% for age <50 years). Patients in the intermediate age group (50-69 years) had an intermediate mortality rate (4.0%). In our study, patients undergoing total gastrectomy had the highest in-hospital mortality (8.0%) of any gastric resection type, in contrast to patients who underwent distal gastrectomy/gastropylorectomy/Billroth I (3.2%). Patients with certain comorbid conditions had much higher in-hospital mortality, including individuals noted to have renal failure (29.0%) and CHF (16.2%). A hospital's annual gastrectomy volume correlated with decreasing mortality, with low-, medium-, and high-volume hospitals having crude mortality rates of 6.8%, 6.4%, and 4.9%, respectively (<.001). In the present study, we found that in-hospital mortality rates did not significantly change from 1998 through 2003, and that rates ranged from 5.3% to 6.5% during this era ( = .40).
To examine possible reasons for the mortality differences seen across hospital volume strata, we performed comparisons of selected covariates across low-, medium- and high-volume hospitals. These analyses demonstrated considerable heterogeneity in the patient populations undergoing gastric resection at different-volume hospitals, with significant differences in mean patient age, the presence of major comorbidities, diagnosis types, and types of surgical resection. Patients treated at hospitals with higher surgical volumes were slightly younger compared with the 2 lower-volume groups (mean age, 65.6 vs 68.5 and 67.2; <.001); these patients were also found to have a lower percentage of certain comorbid conditions, specifically CHF. This may imply that high-volume centers are treating younger, healthier patients. However, the data also show that high-volume centers are treating less benign neoplastic disease compared with the 2 lower-volume groups (8.8% vs 11.4% and 9.2%; <.001), are more often performing the higher-risk procedures, and have the highest rates for total gastrectomies (32.7% vs 23.1% and 28.0%; <.001), the procedure found to have the highest in-hospital mortality. These data suggest a complex relationship between the patient type and hospital volume. After adjustment for these and other known factors in our multivariate model, patients undergoing gastric resection at low-volume hospitals had odds for perioperative death 1.5-fold those of patients at higher-volume hospitals; patients undergoing resection at medium-volume hospitals had an intermediate OR for death at 1.3.
Our results, which showed an in-hospital mortality rate of 6.0% for gastric resections for neoplasm, compare favorably with those reported previously in the literature. Birkmeyer et al,9 who used the Medicare Claims database and the NIS for 1994 through 1999, found that the in-hospital or 30-day mortality for gastrectomy for cancer ranged from 8.7% to 13%. Finlayson et al7 used the NIS from 1995 through 1997 to examine mortality after 8 types of cancer resections and reported a mortality rate for gastrectomy ranging from 6.9% to 8.7%. Neither of these studies presented an overall mortality rate after gastrectomy, only ranges over volume strata. A national study from the Department of Veterans Affairs medical centers showed a 7.6% overall 30-day mortality rate after gastrectomy for cancer.18 Case series studies from single institutions in the United States have reported postoperative mortality rates as low as 4% after gastrectomy for adenocarcinoma of the stomach19; however, these rates are not directly comparable to our national level data. Although gastric neoplasms and their surgical treatment are known to vary widely by geographic region and by patient race and ethnicity,20- 23 international reports of interest include an operative mortality of 6.5% for curative resections for gastric cancer at a single institution in Chile from 1982 to 1990.24 Gastric cancer results differ most strikingly in Asian countries20; a recent study from Taiwan reported in-hospital mortality ranging from 1.4% to 5.4%.25
Our results showing that factors including male sex, older patient age, and procedure types including total gastrectomy increase perioperative mortality validate in part the previous work of others. In an analysis based on the Dutch Gastric Cancer Trial, investigators found that being older than 65 years and male were the most important risk factors for death, together with the extent of nodal dissection.26 Total gastrectomy has been found by some investigators to confer an additional risk of mortality,19 whereas the national Veterans Affairs study did not find a significance difference for morbidity or mortality by gastrectomy type.18
An increasing body of evidence has documented the inverse relationship between operative mortality and hospital surgical volume.6- 9 However, unlike the results for procedures such as pancreatic resection, for which the volume-outcome relationship has been consistently established,16,27- 32 the existing literature has painted an indeterminate picture as to volume effects on mortality after gastrectomy. Investigators using the Medicare Claims database and the NIS for 1994 through 1999 found that the mortality with gastrectomy for cancer ranged from 8.7% to 13% across volume strata.9 The same group,7 using the NIS for 1995 through 1997 to examine mortality with 8 types of cancer resections, found that differences for in-hospital mortality between low- and high-volume hospitals only were significant for pancreatectomy, esophagectomy, and pulmonary lobectomy. Mortality rates for gastrectomy ranged from 8.7% in low-volume hospitals (<9 procedures) to 6.9% in high-volume hospitals (>17 procedures), and after adjustment for confounders no difference was found in the adjusted OR for mortality.
Previous data are conflicting regarding improvement in operative mortality of gastric resections over time.4,5 Our data do not show a decrease in operative mortality for gastrectomy during the recent period from 1998 through 2003, although our overall mortality skews lower than those of earlier studies. However, given known differences in databases such as the NIS, Medicare Claims Database, Medicare Provider Analysis and Review, and Veterans Affairs National Surgical Quality Improvement Program, our results, while encouraging, are not directly comparable to those of the previous studies.
Our study is by definition limited by the confines of the NIS data. Because the NIS is an administrative database, the accurate assessment of case mix, including comorbidities, and the separation of preexisting comorbid factors from postoperative complications can be difficult.33,34 Stage information and preoperative patient performance status measurements are not included in the NIS data set, although other investigators have concluded that such patient-level characteristics are not likely to explain mortality differences across hospital volume levels.7,9,26 A further caveat is that the present study examines the outcome measure of in-hospital mortality. Other important outcomes, such as perioperative complications (eg, hemorrhage, infection, anastomotic leak, need for reoperation or intervention, and hospital readmission) and long-term survival, are also important in understanding the global risks after gastrectomy for neoplasm.
Our data demonstrate that, in a large national database, patients undergoing gastrectomy at high-volume hospitals have decreased perioperative mortality compared with patients undergoing these procedures at lower-volume hospitals. Although patients at low-volume hospitals have a number of risk factors predisposing to poorer outcomes, such as older age and significant medical comorbidities, these volume-mortality relationships maintain significance after a logistic regression to control for these and other covariates. Our overall in-hospital mortality rate of 6.0% for gastrectomy for neoplasm is comparable to the 5.9% rate of in-hospital mortality for pancreatic resection for neoplasm in the NIS during the same period, which we have recently reported.16 Because gastric neoplasms carry a relatively poor prognosis, patients being considered for gastric resection for neoplasm may be able to markedly improve their odds of long-term survival by seeking care at hospitals that perform relatively high volumes of such surgery.
Jennifer F. Tseng, MD, Department of Surgery, University of Massachusetts Medical School, UMass Memorial Medical Center, 119 Belmont St, Swift House, Worchester, MA 01605 (email@example.com).
December 4, 2006.
Dr Tseng had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. : Tseng. : Anderson and Tseng. : Smith, McPhee, Hill, Whalen, Litwin, and Tseng. : Smith and Hill. : McPhee, Hill, Whalen, Sullivan, Litwin, Anderson, and Tseng. : Smith, McPhee, Hill, and Tseng. : Anderson and Tseng. : Whalen, Sullivan, Anderson, and Tseng. : Litwin, Anderson, and Tseng.
This study was supported by the Pancreatic Cancer Alliance (Dr Tseng) and a PanCAN–Samuel Strom Young Investigator Award from the American Society of Clinical Oncology (Dr Tseng). Dr Tseng is also a Howard Hughes Medical Institute Early Career Awardee.
This study was presented at the 87th Annual Meeting of the New England Surgical Society; September 16, 2006; Groton, Conn, and is published after peer review and revision. The discussions that follow this article are based on the originally submitted manuscript and not the revised manuscript.
We thank Theodore P. McDade, MD, for his contributions to the UMass Surgical Outcomes Analysis and Research Group, including the careful review of the manuscript.