Omission of Adjuvant Therapy After Gastric Cancer Resection: Development of a Validated Risk Model

View More View Less
  • a From the Department of Surgery; Division of Hematology/Oncology, Department of Medicine; and Department of Radiation Oncology, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania.

NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines) for Gastric Cancer recommend adjuvant chemotherapy with or without radiotherapy following after resection of gastric adenocarcinoma (GA) for patients who have not received neoadjuvant therapy. Despite frequent noncompliance with NCCN Guidelines nationally, risk factors underlying adjuvant therapy omission (ATom) have not been well characterized. We developed an internally validated preoperative instrument stratifying patients by incremental risk of ATom. The National Cancer Data Base was queried for patients with stage IB–III GA undergoing gastrectomy; those receiving neoadjuvant therapy were excluded. Multivariable models identified factors associated with ATom between 2006 and 2011. Internal validation was performed using bootstrap analysis; model discrimination and calibration were assessed using k-fold cross-validation and Hosmer-Lemeshow procedures, respectively. Using weighted β-coefficients, a simplified Omission Risk Score (ORS) was created to stratify ATom risk. The impact of ATom on overall survival (OS) was examined in ORS risk-stratified cohorts. In 4,728 patients (median age, 70 years; 64.8% male), 53.7% had ATom. The bootstrap-validated model identified advancing age, comorbidity, underinsured/uninsured status, proximal tumor location, and clinical T1/2 and N0 tumors as independent ATom predictors, demonstrating good discrimination. The simplified ORS, stratifying patients into low-, moderate-, and high-risk categories, predicted incremental risk of ATom (30% vs 53% vs 80%, respectively) and progressive delay to adjuvant therapy initiation (median time, 51 vs 55 vs 61 days, respectively). Patients at moderate/high-risk of ATom demonstrated worsening risk-adjusted mortality compared with low-risk patients (median OS, 26.4 vs 29.2 months). This ORS may aid in rational selection of multimodality treatment sequence in GA.

Gastric adenocarcinoma (GA) is the second leading cause of cancer-related mortality worldwide.1 In the United States, an estimated 10,990 deaths were attributed to GA in 2014.2 Margin-negative surgical resection (R0) is the only potentially curative treatment for GA. Even after curative-intent gastrectomy, however, the 5-year survival rate remains approximately 28%.2,3

Despite these dismal statistics, surgery alone remained the mainstay of therapy in the United States until publication of results from the Intergroup 0116 (INT-0116) trial in May 2000. This phase III randomized controlled trial evaluated the impact of postoperative chemoradiotherapy (CRT) for patients with resected stomach/gastroesophageal junction adenocarcinoma. Patients receiving adjuvant CRT demonstrated significantly improved median disease-free survival (DFS; 30 vs 19 months; P<.001) and overall survival (OS; 36 vs 27 months; P=.005) compared with those undergoing surgery alone.4 This survival benefit persisted on longitudinal (ie, >10-year) follow-up.5 Despite criticism that CRT may have compensated for inadequate surgery6 (only 10% of patients underwent D2 lymphadenectomy, whereas 54% received D0 resections4), use of adjuvant multimodality therapy has since been incorporated into practice guidelines for GA management in the United States. Subsequently, several retrospective population-based studies have reaffirmed the benefit of adjuvant therapy in patients undergoing curative-intent gastrectomy.79

Based on these data, current NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines) for Gastric Cancer10 recommend postoperative CRT for patients with localized GA4 or postoperative chemotherapy after D2 (to view the most recent version of these guidelines, visit NCCN.org).11 The latter recommendation reflects findings from the CLASSIC trial; specifically, improved DFS and OS in patients receiving adjuvant chemotherapy (capecitabine/oxaliplatin) compared with those undergoing surgery alone.11 Alternatively, patients may receive neoadjuvant chemotherapy, drawing on results from the MAGIC trial, which demonstrated a survival advantage with perioperative chemotherapy versus surgery alone comparable with that achieved from adjuvant CRT in INT-0116.12 Importantly, in patients selected to undergo a surgery-first approach, clinical adoption of these NCCN Guidelines has remained inadequate. A recent National Cancer Data Base (NCDB) study revealed that use of adjuvant therapy in eligible patients by INT-0116 criteria plateaued at approximately 40% per year between 2003 and 2007. Alarmingly, a substantial proportion of INT-0116–eligible patients (nearly 40% of cases per year) underwent surgery alone in 2007.13 Risk factors for the omission of adjuvant multimodality therapy in eligible patients have not been well characterized.

Using a contemporary cohort from the NCDB, we (1) assessed temporal trends in use of adjuvant multimodality therapy after surgical resection; (2) identified demographic and clinical factors predicting omission of guideline-appropriate adjuvant therapy; and (3) developed an internally validated preoperative risk stratification model discriminating patients at low, moderate, and high risk for adjuvant therapy omission (ATom) after gastrectomy. This tool identifies vulnerable populations in whom delivery of cancer care could be improved, allowing for a more rational selection of treatment strategy.

Methods

Data Source

After Institutional Review Board approval, data from 1998 through 2011 were acquired from the esophagogastric participant use file of the NCDB, a collaborative effort between the American Cancer Society and the American College of Surgeons’ Commission on Cancer (CoC). Established in 1989, the NCDB is a comprehensive oncology surveillance program that captures approximately 70% of new cancer diagnoses from more than 1500 CoC-approved centers.14 Data available in the NCDB include site-specific operative codes, AJCC clinical/pathologic TNM staging (5th–7th editions), and multimodality treatment sequence.

Patient Selection

Patients with invasive GA (defined by ICD-O-315) undergoing curative-intent resection for AJCC pathologic stage IB–III disease were identified. Patients receiving neoadjuvant chemotherapy and/or radiotherapy were excluded from the analysis. Patients were also excluded if they did not undergo at least a partial gastrectomy, received an indeterminate lymph node (LN) harvest, underwent palliative resection, or died within 30 days postoperatively.

Variables

The demographic/clinical NCDB variables used in this study (Table 1) have been defined previously.16 Non–privately insured (Medicare/Medicaid) patients were combined with uninsured patients to dichotomize the insurance variable for ease of incorporation into the risk model. Pre–INT-0116 (1998–2000) and post–INT-0116 (2001–2011) eras were defined to compare temporal differences in treatment utilization rates. The primary outcome of interest—ATom—was defined as nonreceipt of adjuvant chemotherapy or CRT (C±RT) in resected INT-0116–eligible patients selected to undergo a surgery-first approach. The remainder of patients received adjuvant C±RT. A variable representing duration to adjuvant therapy commencement was created by calculating the difference between duration to chemotherapy initiation and duration to gastrectomy (both from date of diagnosis). OS was defined as the interval between date of diagnosis and date of death/last contact.

Statistical Analysis

Descriptive statistics were performed. Simple linear regression was used to analyze temporal trends in adjuvant C±RT use. Bivariate analysis of independent variables by outcome of interest (ie, ATom) was performed. Pearson-χ2 or Fisher exact test and unpaired Student t tests were used to analyze categorical and continuous variables, respectively. Candidate predictive variables were entered via backward stepwise regression (P<.05 for entry); those variables demonstrating no independent association (P>.10) with ATom were removed. Covariates yielding a P value less than .10 were included in a bootstrap internal validation procedure, wherein 1000 random samples were generated with replacement.17 Each sample was

Table 1

Demographic and Clinical Characteristics of the Analytic Cohort (n=4,728) and Univariate Comparison Between ATom and Adjuvant C±RT Cohorts

Table 1Table 1
then subjected to multivariable regression. Predictors occurring in 50% or more of bootstrap models were retained in the final regression.

Model performance was evaluated using k-fold cross-validation, which split the dataset into 10 exclusive subsets of equal size, with 90% and 10% of subsets used for model training and testing, respectively.18 Results were used to calculate a bias-corrected C statistic comparable with the area under the receiver operating characteristic curve. A value of 0.5 demonstrates poor predictive capacity, whereas 1.0 demonstrates perfect model discrimination. Model calibration was performed via Hosmer-Lemeshow goodness-of-fit tests.19

A simplified risk stratification tool—the adjuvant therapy Omission Risk Score (ORS)—was created by assigning discrete numerical values to each risk factor based on β coefficients derived from the bootstrapped model. The referent for each variable was assigned a value of zero. For the remaining values, point assignments proportional to the lowest β coefficient were made. The aggregate of these values generated a composite risk score for each patient. Clinically applicable risk zones were developed using categorical regression of composite ORS.

The Kaplan-Meier method was used to estimate survival function.20 Univariate comparisons between treatment groups were performed by the log-rank test. Cox proportional hazards regression of covariates was performed using backward elimination. All tests were 2-sided. A P value of .05 or less was considered statistically significant. Analysis was conducted using SPSS V22.0 (Chicago, IL) and SAS 9.3 (Cary, NC).

Results

Descriptive Statistics and Temporal Trends in Treatment Utilization

The NCDB gastric cohort included 141,760 patients with invasive GA treated between 1998 and 2011. Structured queries allowed exclusion of patients with metastatic (n=19,732) or unknown stage (n=70,382) disease, as well as those who received preoperative C±RT (n=6,845), underwent indeterminate LN harvest (n=1,126), did not undergo gastrectomy (n=10,874), required palliative resection (n=108), died within 30 postoperative days (n=1,773), or had unknown clinical T-/N-classification data (n=22,666). Temporal treatment trends were analyzed in this cohort (n=8,254).

Between 1998 and 2011, the relative proportion of ATom patients decreased from 71.5% to 51.9%, whereas those receiving adjuvant C±RT increased by 68.8%, from 28.5% to 48.1% (both, P<.001). The largest annual change in undergoing ATom (67.9%–59.5%) or use of adjuvant C±RT (32.1%–40.5%) occurred between 1999 and 2000. Compared with the pre–INT-0116 era, the relative proportion of ATom patients decreased significantly in the post–INT-0116 era (54.0% vs 70.6%; P<.001). Nevertheless, as of 2011 in the United States, NCCN Guideline-appropriate10 adjuvant therapy was omitted in 51.9% of INT-0116–eligible patients selected for a surgery-first approach.

A contemporary subset of patients diagnosed between 2006 and 2011 was used to derive the ORS model (n=4,728). In this cohort, ATom was observed in 2,535 (53.6%) of patients; 2,193 (46.4%) received adjuvant C±RT (CRT, 1,546 [70.5%]; chemotherapy alone, 647 [29.5%]). The median age was 70 years (interquartile range [IQR], 60–78 years); a majority of patients were male (64.8%) and white (73.1%). Most patients held nonprivate (Medicare/Medicaid) insurance (66.7%), and a minority (1.9%) were uninsured. Facility type and location with the highest case representation were nonacademic centers (58.6%) and southern region (34.1%), respectively. Most tumors were larger than 2 cm (85.7%) and had poorly differentiated/undifferentiated histology (61.0%). Proximal tumor location accounted for 34.0% of cases; most (73.4%) required total gastrectomy. Microscopically negative margins (R0) and adequate LN staging (≥15 LNs examined10) were achieved in 86.0% and only 48.5% of cases, respectively. Median duration of postoperative stay was 9 days (IQR, 6–13 days), and the 30-day readmission rate was less than 10% (Table 1).

Factors Predicting ATom

Univariate comparison between adjuvant C±RT and ATom cohorts is presented in Table 1. Significantly different preoperative demographic/clinical variables (ie, age, sex, race, Hispanic ethnicity, Charlson/Deyo comorbidity index, insurance, tumor location, and clinical T/N classification) were entered in the final multivariable regression model. AJCC clinical stage, income, facility location or type, and urban/rural treatment status did not achieve statistical significance.

Predictors determined by backward stepwise regression to be significantly associated with ATom were advancing age (56–66 years: odds ratio [OR], 1.48, P<.001; 67–76 years: OR, 2.19, P<.001; >76 years: OR, 7.06, P<.001), Charlson/Deyo comorbidity index (score ≥1: OR, 1.35; P<.001), nonprivate insurance/uninsured (OR, 1.20; P=.02), proximal tumor location (OR, 1.63; P<.001), clinical T1/2 classification (OR, 1.20; P=.006), and clinical N0 classification (OR, 1.83; P<.001).

Risk Model Generation and Validation

Performance of the model was internally validated using a bootstrap procedure, an approach that minimizes the inherent bias (or optimism) toward an overestimated performance in the derivation dataset,17 yielding the following regression equation:

FD1
Regression β coefficients, adjusted ORs, and bias-corrected 95% CIs are detailed in Table 2. The C statistic
Table 2

Independent Variables in Bootstrap-Validated Logistic Regression Model Predicting ATom in Intergroup-0116–Eligible Patients

Table 2
of the original regression was 0.73 (Figure 1A), with a bias-corrected C statistic of 0.72, indicating good discrimination. The calibration of this model was acceptable by Hosmer-Lemeshow goodness-of-fit testing (8.28, P=0.41; Figure 1B).

ORS Derivation and Risk Stratification

A simplified ORS was derived, wherein an individual patient’s overall score could range from 0 to 12 (Table 3). Along a continuum, the frequency of ATom correlated significantly with progressively increasing ORS (P<.001; Figure 2A). The simplified model retained the ability to predict ATom as effectively as the original model (C statistic 0.72; Hosmer-Lemeshow 7.65, P=.37; Figure 1B).

Risk stratification using composite ORS is illustrated in Figure 2B. Patients in the low-risk (ORS, 0–3; n=325) and moderate-risk (ORS, 4–7; n=1,137) zones demonstrated a 30% and 53% risk of ATom, respectively; however, 80% of patients stratified as high risk (ORS, 8–12; n=1,042) did not receive adjuvant therapy.

Association of ORS With Duration to Adjuvant Therapy

We hypothesized that increasing ORS may also predict a delay in multimodality therapy initiation after gastrectomy. In 1,878 patients (85.6%) evaluable for this analysis, median duration to commencement of adjuvant chemotherapy rose incrementally from 51 days (IQR, 38–70 days) in low-risk patients, 55 days (IQR, 41–74 days) in moderate-risk patients, and 61 days (42.0–78.8 days) in high-risk patients (P=.001; Figure 2C). Moreover, a strong linear relationship was observed between mean duration to adjuvant C±RT and ORS, when plotted along a risk continuum (P=.001). Next, in accordance with the INT-0116 schedule, delay to adjuvant therapy was defined as treatment commencement greater than 6 weeks after gastrectomy. The ORS retained the ability to predict delay; the proportion of patients initiating adjuvant therapy more than 6 weeks after gastrectomy increased from 68.5% (low-risk) to 74.6% (moderate-risk) to 77.6% (high-risk; P=.005).

ORS Risk Stratification and Survival

The impact of ORS risk stratification on OS was examined. Survival analysis included 1,717 patients from 1998 through 2006 with minimum 5-year follow-up.14 Within this cohort, median survival was 35.7 months and 1- and 5-year survival rates were 79.6% and 38.3%, respectively. Each patient was assigned a composite ORS score and categorized into low- (ORS, 0–3) or

Figure 1
Figure 1

(A) Receiver operating characteristic (ROC) curve plotting capacity of the bootstrap-validated regression model to predict adjuvant therapy omission (ATom; blue curve). The C statistic (area under ROC curve) of the model is 0.73, indicating good model discrimination. (B) Hosmer-Lemeshow goodness-of-fit calibration plots depicting correlation between observed and predicted probabilities of ATom in bootstrap-validated (upper panel) and simplified (lower panel) models.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 13, 5; 10.6004/jnccn.2015.0073

moderate/high(ORS 4-12) risk categories. Risk-stratified univariate survival of ATom and adjuvant C±RT cohorts is illustrated in Figure 3. Although median OS between ATom and C±RT cohorts differed significantly (29.2 vs 43.1 months; P=.03) in low-risk patients, this survival disparity widened appreciably (26.4 vs 42.3 months; P<.001) in moderate/high-risk populations. Compared with adjuvant C±RT, ATom conferred an increased risk of death in low-risk (hazard ratio [HR], 1.80; 95% CI, 1.06–3.06; P=.029) and moderate/high-risk (HR, 1.49; 95% CI, 1.10–2.02; P=.011) groups on
Table 3

Simplified Adjuvant Therapy Omission Risk Score

Table 3
Cox proportional hazards modeling. Clinical LN positivity (N1–N3), tumor T classification 3/4, and margin positivity (R1/R2) were other strong predictors of risk-adjusted mortality in low- and moderate/high-risk populations (Table 4).

Discussion

The current study proposes a novel preoperative tool to stratify patients according to risk of ATom following curative-intent gastrectomy for GA. Using

Figure 2
Figure 2

(A) Relationship between proportion of patients with adjuvant therapy omission (ATom) and simplified omission risk score (ORS) along a continuum. This linear trend was highly significant (P<.001). (B) Stratification into low-, moderate-, and high-risk zones using composite ORS scores. The incidence of ATom increases significantly in higher risk categories. (C) Ability of ORS to predict delay to adjuvant therapy initiation after gastrectomy. The box-and-whisker plot illustrates median (± interquartile range) duration to commencement of adjuvant therapy in low-, moderate-, and high-risk patient groups. Abbreviation: C±RT, chemotherapy with or without radiotherapy.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 13, 5; 10.6004/jnccn.2015.0073

bootstrap-validated risk modeling, advancing age, comorbidity, non–privately insured/uninsured status, proximal tumor location, and clinically “early” tumor-related characteristics were independently associated
Figure 3
Figure 3

Impact of adjuvant therapy omission (ATom) on overall survival (OS) in gastric adenocarcinoma from 1998 to 2006 (n=1,717), stratified by omission risk score category. OS is significantly worse in patients with ATom compared with those receiving adjuvant chemotherapy with or without radiotherapy in (A) low-risk, and (B) moderate/high-risk categories. The survival disadvantage conferred by ATom is exacerbated in moderate/high-risk versus low-risk patients.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 13, 5; 10.6004/jnccn.2015.0073

with ATom. Good model discrimination and calibration were verified by a bias-corrected bootstrap,18 k-fold cross validation, and Hosmer-Lemeshow procedures.19 A simplified ORS instrument,
Table 4

Cox Proportional Hazards Model for Overall Survival in Patients Stratified Into Low- and Moderate/High-Risk ORS Cohorts (n=1,717)

Table 4
demonstrating equivalent predictive accuracy, was created to facilitate clinical application. Stratification of patients into low-, moderate-, and high-risk zones predicted incremental risk of ATom and progressive delay to adjuvant therapy commencement. Moreover, moderate-/high-risk–stratified ATom patients demonstrated worsening risk-adjusted mortality compared with low-risk counterparts.

Compliance with NCCN Guidlines recommendations for adjuvant C±RT in resected GA has been poor in the United States. The proportion of eligible patients that had ATom (ie, undergoing surgery alone) approached 60% in 2006, and 50% in 2009 in recent SEER and Oregon State Cancer Registry analyses, respectively.8,21 The present assessment in a contemporary NCDB cohort suggests that ATom continues to be a pervasive problem nationally. As data from prospective randomized controlled trials suggest, these high ATom rates may be explained partly by attrition from prescribed adjuvant therapy regimens because of treatment-related adverse effects. Treatment cessation was observed in 25% of CRT-randomized INT-0116 patients because of toxicity or noncompliance.4 Similarly, only 42% of patients completed the designated postoperative component of their perioperative epirubicin/cisplatin/fluorouracil regimen in the MAGIC trial.12 Concerns about tolerability of adjuvant therapy may be addressed in part by contemporary efforts to apply chemotherapy-only regimens following standardized surgery.

Toxicity-related attrition may also be more pronounced in subsets of patients with high-risk features identified by the ORS model. These include both hostand tumor-related factors—advanced age/comorbidity22 and proximal tumors necessitating total gastrectomy23—which impart substantial risk of perioperative morbidity, suggesting that nonfatal surgical complications may be a major determinant of adjuvant therapy omission/delay in resected GA. Indeed, in other gastrointestinal tumors, perioperative morbidity not only delays and/or precludes initiation of adjuvant therapy,24,25 but also negatively impacts DFS and OS.26,27 It is possible, therefore, that the surrogates for worsening operative morbidity in this study may contribute to the deteriorating risk-adjusted survival observed in moderate/high-risk ATom patients.

It should be emphasized that ATom may be appropriate in a subset of patients deemed eligible for adjuvant therapy by clinical staging, barring other ORS-defined high-risk features. Discordance between clinical and pathologic staging is well recognized; discrimination between early T classifications (ie, T1 vs T2) is particularly challenging and may account for appropriate ATom in some cases. Nonetheless, given the reasonable concordance between clinical and pathologic staging in retrospective analyses28,29 when tumors are grouped as T1/2 versus T3/4 categories, the strong association of clinical T1/2 and N0 tumors with ATom may reflect assumptions regarding the dispensability of adjuvant therapy in “early” disease. Such assumptions have not been supported in the literature; conversely, suboptimal outcomes in US patients with GA, including those with stage IB and II disease,3 indicate that a multimodality approach should be applied more consistently. In addition, the increased risk of ATom in older/comorbid patients may reflect quality-of-life considerations or risk aversion. Although these risks may outweigh the putative benefits of adjuvant therapy in such populations, they should be balanced against the well-documented risk of morbidity and mortality from gastric resection, and total gastrectomy in particular.23 Finally, the disproportionate risk of ATom in non–privately insured or uninsured patients point to disparities in access to comprehensive cancer care, which merit further investigation.

Selecting the optimal multimodality treatment sequence for patients with resectable GA remains a considerable challenge.30 The apparent equipoise between adjuvant and neoadjuvant/perioperative C±RT protocols in the published literature continues to obfuscate treatment priorities. The ORS instrument, which preoperatively assesses individualized risk of adjuvant therapy delay/omission, may aid in clinical decision-making. For instance, a subset of patients at moderate/high ORS risk may be better served by induction therapy before surgery. Advantages of neoadjuvant therapy include tumor downstaging and enhanced tolerability of multimodality therapy.30 This approach, however, is also associated with a rate of attrition; in intent-to-treat trials, up to 15% of patients who initiate neoadjuvant therapy fail to undergo gastrectomy.31,32 Importantly, factors that drive attrition during neoadjuvant therapy likely differ from those predicting ATom and may aid more rational application of surgical therapy. Specifically, disease progression during induction therapy identifies patients for whom initial surgery would confer little advantage. Notwithstanding, there are patients for whom surgery provides essential locoregional disease control, and neoadjuvant therapy represents an ineffective detour. Dissecting apart these patient subsets is a logical next step in efforts to optimize multimodality treatment sequencing.

Several study limitations, characteristic of retrospective database analyses, warrant emphasis: (1) bias from missing data, (2) patient misclassification, and (3) underreporting of adjuvant C±RT receipt. The last of which may have inflated the prevalence of ATom in this cohort and impacted study findings. Additionally, because determination of treatment intent is not possible with the NCDB data, it remains unclear whether adjuvant therapy was intended but never initiated (eg, related to postoperative complications) or excluded (eg, not considered part of treatment plan8). Finally, specifics regarding the C±RT schedules used are not available and the effect of adjuvant therapy may be underestimated or overestimated.

Conclusions

This is the first report of an internally validated risk assessment tool assimilating patient-, tumor-, and provider-associated factors to predict ATom in patients with resected GA. Although the ORS is not intended to supplant nuanced clinical judgment, it does underscore the need for a more personalized approach to multimodality treatment of GA. The not infrequent tensions between optimizing long-term cancer-related outcomes and locoregional control preclude an algorithmic approach to this disease. Above all, the current data highlight shortcomings of the current armamentarium for the treatment of GA. Development of better tolerated and more effective adjuvant approaches, increased utilization of neoadjuvant therapies when appropriate, and more standardized surgery may represent important elements in improving overall GA outcomes in the United States.

The authors have disclosed that they have no financial interests, arrangements, affiliations, or commercial interests with the manufacturers of any products discussed in this article or their competitors.

References

  • 1.

    Siegel R, Ma J, Zou Z, Jemal A. Cancer statistics, 2014. CA Cancer J Clin 2014;64:929. Erratum in CA Cancer J Clin 2014;64:364.

  • 2.

    National Cancer Institute. SEER Stat Fact Sheets: Stomach Cancer. Surveillance, Epidemiology, and End Results Program Web site. Available at; http://seer.cancer.gov/statfacts/html/stomach.html. Accessed April 2, 2015.

    • Search Google Scholar
    • Export Citation
  • 3.

    Hundahl SA, Phillips JL, Menck HR. The National Cancer Data Base report on poor survival of U.S. gastric carcinoma patients treated with gastrectomy: Fifth Edition American Joint Committee on Cancer staging, proximal disease, and the “different disease” hypothesis. Cancer 2000;88:921932.

    • Search Google Scholar
    • Export Citation
  • 4.

    Macdonald JS, Smalley SR, Benedetti J. Chemoradiotherapy after surgery compared with surgery alone for adenocarcinoma of the stomach or gastroesophageal junction. N Engl J Med 2001;345:725730.

    • Search Google Scholar
    • Export Citation
  • 5.

    Smalley SR, Benedetti JK, Haller DG. Updated analysis of SWOG-directed Intergroup study 0116: a phase III trial of adjuvant radiochemotherapy versus observation after curative gastric cancer resection. J Clin Oncol 2012;30:23272333.

    • Search Google Scholar
    • Export Citation
  • 6.

    Dikken JL, Van De Velde CJ, Coit DG. Treatment of resectable gastric cancer. Therap Adv Gastroenterol 2012;5:4969.

  • 7.

    Coburn NG, Guller U, Baxter NN. Adjuvant therapy for resected gastric cancer—rapid, yet incomplete adoption following results of Intergroup 0116 trial. Int J Radiat Oncol Biol Phys 2008;70:10731080.

    • Search Google Scholar
    • Export Citation
  • 8.

    Enestvedt CK, Diggs BS, Shipley DK. A population-based analysis of surgical and adjuvant therapy for resected gastric cancer: are patients receiving appropriate treatment following publication of the Intergroup 0116 results? Gastrointest Cancer Res 2009;3:233238.

    • Search Google Scholar
    • Export Citation
  • 9.

    Kozak KR, Moody JS. The survival impact of the Intergroup 0116 trial on patients with gastric cancer. Int J Radiat Oncol Biol Phys 2008;72:517521.

    • Search Google Scholar
    • Export Citation
  • 10.

    Ajani JA, Bentrem DJ, Besh S. Gastric Cancer, Version 2.2013: featured updates to the NCCN Guidelines. J Natl Compr Canc Netw 2013;11:531546.

    • Search Google Scholar
    • Export Citation
  • 11.

    Bang YJ, Kim YW, Yang HK. Adjuvant capecitabine and oxaliplatin for gastric cancer after D2 gastrectomy (classic): a phase 3 open-label, randomised controlled trial. Lancet 2012;379:315321.

    • Search Google Scholar
    • Export Citation
  • 12.

    Cunningham D, Allum WH, Stenning SP. Perioperative chemotherapy versus surgery alone for resectable gastroesophageal cancer. N Engl J Med 2006;355:1120.

    • Search Google Scholar
    • Export Citation
  • 13.

    Sherman KL, Merkow RP, Bilimoria KY. Treatment trends and predictors of adjuvant and neoadjuvant therapy for gastric adenocarcinoma in the United States. Ann Surg Oncol 2013;20:362370.

    • Search Google Scholar
    • Export Citation
  • 14.

    Bilimoria KY, Stewart AK, Winchester DP, Ko CY. The National Cancer Data Base: a powerful initiative to improve cancer care in the United States. Ann Surg Oncol 2008;15:683690.

    • Search Google Scholar
    • Export Citation
  • 15.

    Fritz A, Percy C, Jack A, eds International Classification of Diseases for Oncology. 3rd ed. Geneva: World Health Organization; 2000. Available at: http://whqlibdoc.who.int/publications/2000/9241545348_eng.pdf. Accessed April 2, 2015.

    • Search Google Scholar
    • Export Citation
  • 16.

    Datta J, Lewis RS Jr, Mamtani R. Implications of inadequate lymph node staging in resectable gastric cancer: a contemporary analysis using The National Cancer Data Base. Cancer. 2014;120:28552865.

    • Search Google Scholar
    • Export Citation
  • 17.

    Chen CH, George SL. The bootstrap and identification of prognostic factors via Cox’s proportional hazards regression model. Stat Med 1985;4:3946.

    • Search Google Scholar
    • Export Citation
  • 18.

    Kohave R. A study of cross-validation and bootstrap for accuracy estimation and model selection. In Proceedings of the 14th International Joint Conference on Artificial Intelligence, Volume 2 (IJCAI’95). San Francisco, CA: Morgan Kaufmann Publishers Inc.; 1995;2:11371143.

    • Search Google Scholar
    • Export Citation
  • 19.

    Hosmer DW Jr, Wang CY, Lin IC, Lemeshow S. A computer program for stepwise logistic regression using maximum likelihood estimation. Comput Programs Biomed 1978;8:121134.

    • Search Google Scholar
    • Export Citation
  • 20.

    Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958;53:457481.

  • 21.

    Snyder RA, Penson DF, Ni S. Trends in the use of evidence-based therapy for resectable gastric cancer. J Surg Oncol 2014;110:285290.

  • 22.

    Grossmann EM, Longo WE, Virgo KS. Morbidity and mortality of gastrectomy for cancer in department of Veterans Affairs Medical Centers. Surgery 2002;131:484490.

    • Search Google Scholar
    • Export Citation
  • 23.

    Bartlett EK, Roses RE, Kelz RR. Morbidity and mortality after total gastrectomy for gastric malignancy using the American College of Surgeons National Surgical Quality Improvement Program database. Surgery 2014;156:298304.

    • Search Google Scholar
    • Export Citation
  • 24.

    Hendren S, Birkmeyer JD, Yin H. Surgical complications are associated with omission of chemotherapy for stage III colorectal cancer. Dis Colon Rectum 2010;53:15871593.

    • Search Google Scholar
    • Export Citation
  • 25.

    Merkow RP, Bilimoria KY, Tomlinson JS. Postoperative complications reduce adjuvant chemotherapy use in resectable pancreatic cancer. Ann Surg 2014;260:372377.

    • Search Google Scholar
    • Export Citation
  • 26.

    Howard TJ, Krug JE, Yu J. A margin-negative R0 resection accomplished with minimal postoperative complications is the surgeon’s contribution to long-term survival in pancreatic cancer. J Gastrointest Surg 2006;10:13381345; discussion 1345-1346.

    • Search Google Scholar
    • Export Citation
  • 27.

    Law WL, Choi HK, Lee YM, Ho JW. The impact of postoperative complications on long-term outcomes following curative resection for colorectal cancer. Ann Surg Oncol 2007;14:25592566.

    • Search Google Scholar
    • Export Citation
  • 28.

    Barbour AP, Rizk NP, Gerdes H. Endoscopic ultrasound predicts outcomes for patients with adenocarcinoma of the gastroesophageal junction. J Am Coll Surg 2007;205:593601.

    • Search Google Scholar
    • Export Citation
  • 29.

    Bentrem D, Gerdes H, Tang L. Clinical correlation of endoscopic ultrasonography with pathologic stage and outcome in patients undergoing curative resection for gastric cancer. Ann Surg Oncol 2007;14:18531859.

    • Search Google Scholar
    • Export Citation
  • 30.

    Mezhir JJ, Tang LH, Coit DG. Neoadjuvant therapy of locally advanced gastric cancer. J Surg Oncol 2010;101:305314.

  • 31.

    Ajani JA, Mansfield PF, Crane CH. Paclitaxel-based chemoradiotherapy in localized gastric carcinoma: degree of pathologic response and not clinical parameters dictated patient outcome. J Clin Oncol 2005;23:12371244.

    • Search Google Scholar
    • Export Citation
  • 32.

    Ajani JA, Winter K, Okawara GS. Phase II trial of preoperative chemoradiation in patients with localized gastric adenocarcinoma (RTOG 9904): quality of combined modality therapy and pathologic response. J Clin Oncol 2006;24:39533958.

    • Search Google Scholar
    • Export Citation

If the inline PDF is not rendering correctly, you can download the PDF file here.

Correspondence: Robert E. Roses, MD, Department of Surgery, University of Pennsylvania Perelman School of Medicine, 3400 Spruce Street, Philadelphia, PA 19104. E-mail: robert.roses@uphs.upenn.edu
  • View in gallery

    (A) Receiver operating characteristic (ROC) curve plotting capacity of the bootstrap-validated regression model to predict adjuvant therapy omission (ATom; blue curve). The C statistic (area under ROC curve) of the model is 0.73, indicating good model discrimination. (B) Hosmer-Lemeshow goodness-of-fit calibration plots depicting correlation between observed and predicted probabilities of ATom in bootstrap-validated (upper panel) and simplified (lower panel) models.

  • View in gallery

    (A) Relationship between proportion of patients with adjuvant therapy omission (ATom) and simplified omission risk score (ORS) along a continuum. This linear trend was highly significant (P<.001). (B) Stratification into low-, moderate-, and high-risk zones using composite ORS scores. The incidence of ATom increases significantly in higher risk categories. (C) Ability of ORS to predict delay to adjuvant therapy initiation after gastrectomy. The box-and-whisker plot illustrates median (± interquartile range) duration to commencement of adjuvant therapy in low-, moderate-, and high-risk patient groups. Abbreviation: C±RT, chemotherapy with or without radiotherapy.

  • View in gallery

    Impact of adjuvant therapy omission (ATom) on overall survival (OS) in gastric adenocarcinoma from 1998 to 2006 (n=1,717), stratified by omission risk score category. OS is significantly worse in patients with ATom compared with those receiving adjuvant chemotherapy with or without radiotherapy in (A) low-risk, and (B) moderate/high-risk categories. The survival disadvantage conferred by ATom is exacerbated in moderate/high-risk versus low-risk patients.

  • 1.

    Siegel R, Ma J, Zou Z, Jemal A. Cancer statistics, 2014. CA Cancer J Clin 2014;64:929. Erratum in CA Cancer J Clin 2014;64:364.

  • 2.

    National Cancer Institute. SEER Stat Fact Sheets: Stomach Cancer. Surveillance, Epidemiology, and End Results Program Web site. Available at; http://seer.cancer.gov/statfacts/html/stomach.html. Accessed April 2, 2015.

    • Search Google Scholar
    • Export Citation
  • 3.

    Hundahl SA, Phillips JL, Menck HR. The National Cancer Data Base report on poor survival of U.S. gastric carcinoma patients treated with gastrectomy: Fifth Edition American Joint Committee on Cancer staging, proximal disease, and the “different disease” hypothesis. Cancer 2000;88:921932.

    • Search Google Scholar
    • Export Citation
  • 4.

    Macdonald JS, Smalley SR, Benedetti J. Chemoradiotherapy after surgery compared with surgery alone for adenocarcinoma of the stomach or gastroesophageal junction. N Engl J Med 2001;345:725730.

    • Search Google Scholar
    • Export Citation
  • 5.

    Smalley SR, Benedetti JK, Haller DG. Updated analysis of SWOG-directed Intergroup study 0116: a phase III trial of adjuvant radiochemotherapy versus observation after curative gastric cancer resection. J Clin Oncol 2012;30:23272333.

    • Search Google Scholar
    • Export Citation
  • 6.

    Dikken JL, Van De Velde CJ, Coit DG. Treatment of resectable gastric cancer. Therap Adv Gastroenterol 2012;5:4969.

  • 7.

    Coburn NG, Guller U, Baxter NN. Adjuvant therapy for resected gastric cancer—rapid, yet incomplete adoption following results of Intergroup 0116 trial. Int J Radiat Oncol Biol Phys 2008;70:10731080.

    • Search Google Scholar
    • Export Citation
  • 8.

    Enestvedt CK, Diggs BS, Shipley DK. A population-based analysis of surgical and adjuvant therapy for resected gastric cancer: are patients receiving appropriate treatment following publication of the Intergroup 0116 results? Gastrointest Cancer Res 2009;3:233238.

    • Search Google Scholar
    • Export Citation
  • 9.

    Kozak KR, Moody JS. The survival impact of the Intergroup 0116 trial on patients with gastric cancer. Int J Radiat Oncol Biol Phys 2008;72:517521.

    • Search Google Scholar
    • Export Citation
  • 10.

    Ajani JA, Bentrem DJ, Besh S. Gastric Cancer, Version 2.2013: featured updates to the NCCN Guidelines. J Natl Compr Canc Netw 2013;11:531546.

    • Search Google Scholar
    • Export Citation
  • 11.

    Bang YJ, Kim YW, Yang HK. Adjuvant capecitabine and oxaliplatin for gastric cancer after D2 gastrectomy (classic): a phase 3 open-label, randomised controlled trial. Lancet 2012;379:315321.

    • Search Google Scholar
    • Export Citation
  • 12.

    Cunningham D, Allum WH, Stenning SP. Perioperative chemotherapy versus surgery alone for resectable gastroesophageal cancer. N Engl J Med 2006;355:1120.

    • Search Google Scholar
    • Export Citation
  • 13.

    Sherman KL, Merkow RP, Bilimoria KY. Treatment trends and predictors of adjuvant and neoadjuvant therapy for gastric adenocarcinoma in the United States. Ann Surg Oncol 2013;20:362370.

    • Search Google Scholar
    • Export Citation
  • 14.

    Bilimoria KY, Stewart AK, Winchester DP, Ko CY. The National Cancer Data Base: a powerful initiative to improve cancer care in the United States. Ann Surg Oncol 2008;15:683690.

    • Search Google Scholar
    • Export Citation
  • 15.

    Fritz A, Percy C, Jack A, eds International Classification of Diseases for Oncology. 3rd ed. Geneva: World Health Organization; 2000. Available at: http://whqlibdoc.who.int/publications/2000/9241545348_eng.pdf. Accessed April 2, 2015.

    • Search Google Scholar
    • Export Citation
  • 16.

    Datta J, Lewis RS Jr, Mamtani R. Implications of inadequate lymph node staging in resectable gastric cancer: a contemporary analysis using The National Cancer Data Base. Cancer. 2014;120:28552865.

    • Search Google Scholar
    • Export Citation
  • 17.

    Chen CH, George SL. The bootstrap and identification of prognostic factors via Cox’s proportional hazards regression model. Stat Med 1985;4:3946.

    • Search Google Scholar
    • Export Citation
  • 18.

    Kohave R. A study of cross-validation and bootstrap for accuracy estimation and model selection. In Proceedings of the 14th International Joint Conference on Artificial Intelligence, Volume 2 (IJCAI’95). San Francisco, CA: Morgan Kaufmann Publishers Inc.; 1995;2:11371143.

    • Search Google Scholar
    • Export Citation
  • 19.

    Hosmer DW Jr, Wang CY, Lin IC, Lemeshow S. A computer program for stepwise logistic regression using maximum likelihood estimation. Comput Programs Biomed 1978;8:121134.

    • Search Google Scholar
    • Export Citation
  • 20.

    Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958;53:457481.

  • 21.

    Snyder RA, Penson DF, Ni S. Trends in the use of evidence-based therapy for resectable gastric cancer. J Surg Oncol 2014;110:285290.

  • 22.

    Grossmann EM, Longo WE, Virgo KS. Morbidity and mortality of gastrectomy for cancer in department of Veterans Affairs Medical Centers. Surgery 2002;131:484490.

    • Search Google Scholar
    • Export Citation
  • 23.

    Bartlett EK, Roses RE, Kelz RR. Morbidity and mortality after total gastrectomy for gastric malignancy using the American College of Surgeons National Surgical Quality Improvement Program database. Surgery 2014;156:298304.

    • Search Google Scholar
    • Export Citation
  • 24.

    Hendren S, Birkmeyer JD, Yin H. Surgical complications are associated with omission of chemotherapy for stage III colorectal cancer. Dis Colon Rectum 2010;53:15871593.

    • Search Google Scholar
    • Export Citation
  • 25.

    Merkow RP, Bilimoria KY, Tomlinson JS. Postoperative complications reduce adjuvant chemotherapy use in resectable pancreatic cancer. Ann Surg 2014;260:372377.

    • Search Google Scholar
    • Export Citation
  • 26.

    Howard TJ, Krug JE, Yu J. A margin-negative R0 resection accomplished with minimal postoperative complications is the surgeon’s contribution to long-term survival in pancreatic cancer. J Gastrointest Surg 2006;10:13381345; discussion 1345-1346.

    • Search Google Scholar
    • Export Citation
  • 27.

    Law WL, Choi HK, Lee YM, Ho JW. The impact of postoperative complications on long-term outcomes following curative resection for colorectal cancer. Ann Surg Oncol 2007;14:25592566.

    • Search Google Scholar
    • Export Citation
  • 28.

    Barbour AP, Rizk NP, Gerdes H. Endoscopic ultrasound predicts outcomes for patients with adenocarcinoma of the gastroesophageal junction. J Am Coll Surg 2007;205:593601.

    • Search Google Scholar
    • Export Citation
  • 29.

    Bentrem D, Gerdes H, Tang L. Clinical correlation of endoscopic ultrasonography with pathologic stage and outcome in patients undergoing curative resection for gastric cancer. Ann Surg Oncol 2007;14:18531859.

    • Search Google Scholar
    • Export Citation
  • 30.

    Mezhir JJ, Tang LH, Coit DG. Neoadjuvant therapy of locally advanced gastric cancer. J Surg Oncol 2010;101:305314.

  • 31.

    Ajani JA, Mansfield PF, Crane CH. Paclitaxel-based chemoradiotherapy in localized gastric carcinoma: degree of pathologic response and not clinical parameters dictated patient outcome. J Clin Oncol 2005;23:12371244.

    • Search Google Scholar
    • Export Citation
  • 32.

    Ajani JA, Winter K, Okawara GS. Phase II trial of preoperative chemoradiation in patients with localized gastric adenocarcinoma (RTOG 9904): quality of combined modality therapy and pathologic response. J Clin Oncol 2006;24:39533958.

    • Search Google Scholar
    • Export Citation
All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 218 195 11
PDF Downloads 23 21 3
EPUB Downloads 0 0 0