Introduction

Gastric cancer is the third leading cause of cancer-related death worldwide, responsible for nearly 9% of all cancer deaths [1]. Each year, gastric cancer is diagnosed in 20,000 individuals in the USA alone [2]. Although surgical resection of early-stage disease is associated with survival rates of more than 90%, outcomes are less favorable for more advanced disease [3]. New adjuvant therapies are the focus of ongoing efforts to improve survival outcomes for patients in whom gastric adenocarcinoma has been diagnosed.

Randomized clinical trials, including the Southwest Oncology Group (SWOG) 9008/Intergroup 0116 (INT-0116) [4, 5] and Medical Research Council Adjuvant Gastric Infusional Chemotherapy (MAGIC) [6] studies, have led to the addition of postoperative chemoradiotherapy (CRT) or perioperative chemotherapy to gastrectomy as standard therapy in the USA [7,8,9]. However, the utilization rates for these regimens are low (less than 20% and less than 2% respectively) [10], with uncertain effectiveness among the general population.

Studies based on Surveillance, Epidemiology, and End Results (SEER) data report a survival benefit associated with adjuvant radiotherapy following gastrectomy [11,12,13,14,15]. However, because the SEER Program does not include detailed chemotherapy data, the benefit of CRT remains unclear. Debate further continues on whether the addition of adjuvant CRT provides an incremental treatment benefit or only compensates for inadequate surgical resection by decreasing locoregional recurrences [16,17,18]. As more than 60% of gastric cancer patients are 65 years or older when gastric cancer is diagnosed, the effectiveness of adjuvant therapy among older individuals is of particular clinical importance [2].

Providing patient information on chemotherapy treatment, linked SEER–Medicare data can provide more informative estimates of the survival benefit associated with CRT among older individuals in whom gastric cancer has been diagnosed. Previous studies using SEER–Medicare data collected before the adoption of CRT as standard care in 2002 report conflicting results on the survival benefit [19, 20]. As such, we sought to provide updated estimates using SEER–Medicare data on patients treated since adjuvant CRT became standard care in 2002.

Materials and methods

Data source

We used the SEER–Medicare data to estimate the survival benefit associated with adjuvant CRT among individuals in whom gastric adenocarcinoma had been diagnosed and treated with gastrectomy. The database results from the linkage of two large population-based data sources: SEER cancer registries data and Medicare enrollment and claims files [21]. The SEER Program collects and publishes cancer incidence, prevalence, and survival data from population-based cancer registries covering approximately 26% of the US population [22]. SEER includes data on patient demographics, cancer site, histologic type, stage, and grade, and dates of diagnosis and death [21, 23]. Medicare includes data on Part A and Part B claims for covered health care services, including hospital, physician, outpatient, home health, and hospice bills [21]. SEER data for patients with diagnoses between January 1, 2002 and December 31, 2009 were matched to Medicare claims data from January 1, 2001 (or July 1, 2000 for 6 months before diagnosis) through December 31, 2010.

Study sample

The retrospective study cohort included patients aged 65–79 years with nonmetastatic, pathologically confirmed, stage IB–III gastric adenocarcinoma as their primary diagnosis between January 1, 2002 and December 31, 2009 who underwent gastrectomy within 6 months of diagnosis (Fig. 1). We excluded patients who were (1) not continuously eligible for Medicare Part A and Part B for the 12-month period between 6 months before cancer diagnosis and 6 months after diagnosis, (2) enrolled in a health maintenance organization (HMO) at any time during the period, (3) received any neoadjuvant therapy (e.g. chemotherapy and/or radiation therapy before gastrectomy), or (4) died within 90 days of gastrectomy (based on the landmark analysis [24], which determined that more than 80% of patients received adjuvant CRT within 90 days of gastrectomy, to minimize immortal time bias). We excluded patients aged 80 years or older as there is no established benefit from adjuvant CRT in this age group, and therefore it is not considered standard care for this subpopulation. Staging information was based on the American Joint Committee on Cancer (AJCC) cancer staging manual (sixth edition) [25]. Patients in whom the AJCC stage could not be determined were excluded.

Fig. 1
figure 1

Study cohort. Asterisk exclusions were sequential, dagger based on landmark analysis that determined more than 80% of individuals received adjuvant chemoradiotherapy within 90 days of gastrectomy, double dagger no adjuvant therapy within 3 months after gastrectomy, AJCC American Joint Committee on Cancer, HMO health maintenance organization

Full size image

Adjuvant CRT

Variables were created to indicate gastrectomy within 6 months of diagnosis and receipt of chemotherapy and radiation therapy within 3 months after gastrectomy. Detailed information on the codes used to capture Medicare claims for gastrectomy, radiation therapy, and chemotherapy can be found in Table S1. Patients who had at least one Medicare claim with a gastrectomy code were coded as having had gastrectomy. Among these patients, we identified those who received chemotherapy or radiation therapy (or both) on the basis of Medicare claims. If the patient had at least one claim with a chemotherapy code from any Medicare file [Medicare Provider Analysis and Review (MEDPAR) (inpatient), Carrier Claims, Outpatient Claims, Home Health Agency, Hospice, and Durable Medical Equipment] and this claim was within 3 months of gastrectomy, the patient was coded as having had chemotherapy. Because Medicare Part D was not available for our entire study period, we included only capecitabine for oral chemotherapies. Patients receiving radiation therapy were similarly identified by radiation-specific codes. To ensure complete information about radiation therapy, both SEER and Medicare variables were used to code for radiation therapy. Patients who had at least one radiation therapy claim in Medicare within 6 months of diagnosis were coded as having had adjuvant radiation therapy.

Survival outcomes

The primary outcome was all-cause death, defined as the number of survival months from 90 days after gastrectomy until death or December 31, 2010, whichever came first. The secondary outcome was cause-specific death. The date of death and the date of gastrectomy were identified from Medicare records. Patients treated with gastrectomy only and gastrectomy plus adjuvant CRT were defined as the control and treatment groups respectively. We censored observations for patients who were alive at the end of the follow-up. For the secondary analysis, cancer-specific death was coded by the SEER cause-specific death classification in the Patient Entitlement and Diagnosis Summary File (PEDSF).

Baseline characteristics

Demographic and clinical data included age, sex, race/ethnicity, median income (census tract quintile), college education (census tract quintile), SEER geographic region, birth place, comorbidity, year of diagnosis, AJCC stage, tumor location, tumor histologic type, and number of lymph nodes resected (Table 1). Race/ethnicity was categorized as non-Hispanic white, non-Hispanic black, non-Hispanic Asian, and Hispanic. Race in SEER is based on patients’ medical records and registration information, while Hispanic ethnicity is determined through a Hispanic-surname algorithm. Tumor histologic type was categorized as intestinal, diffuse, and other tumors defined on criteria proposed by Lauren [26] and used by Henson et al. [27] and Wu et al. [28]. The number of lymph nodes resected was used as a proxy for determination of extent of lymphadenectomy (less than 15, 15–25, and more than 25). The burden associated with comorbidities was estimated by application of the modified Deyo adaptation [29,30,31] of the Charlson comorbidity index [32] to Medicare inpatient, outpatient, and physician claims during the 6-month period before cancer diagnosis. We classified the comorbidity into three groups with scores of 0, 1, and 2+.

Table 1 Characteristics of older patients with gastric adenocarcinoma treated with gastrectomy for two treatment groups
Full size table

Statistical analysis

We used a chi square test to evaluate differences in the distribution of baseline characteristics between the treatment group and the control group. To estimate median survival, we used the Kaplan–Meier survival method and tests for crude differences among the groups using a log-rank test.

To examine whether the adjuvant therapy improved overall survival in patients with gastric adenocarcinoma who underwent gastrectomy, we fitted two sets of Cox proportional hazards models for all-cause death. The first set of models estimated the overall relative hazard ratio between patients who underwent gastrectomy and received adjuvant CRT versus those treated with gastrectomy only. The second set of models estimated the stage-specific relative hazard ratios from the interaction between treatment effect and cancer stage at diagnosis (stages IB, II, and III).

For each set, we fitted a series of models. The first model estimated the unadjusted relative hazard ratio comparing gastrectomy plus adjuvant CRT with gastrectomy only. The second model adjusted this relative hazard ratio for all demographic and clinical characteristics listed in Table 2. As the treatment effect estimates are likely confounded by factors related to treatment selection, the third model accounted for measurable confounders between the groups using propensity score analysis [33]. This entailed use of multivariable logistic regression to estimate the likelihood of adjuvant therapy among patients who underwent gastrectomy based on confounding covariates, including age, sex, race/ethnicity, median income, college education, geographic region, modified Charlson comorbidiy scores, year of diagnosis, AJCC stage, tumor location, tumor histologic type, and number of lymph nodes resected [34, 35]. Each patient received a propensity score based on his/her predicted probability of receiving adjuvant therapy. Patients who had scores lower than the higher of the two minima or higher than the lower of the two maxima were excluded to prevent unreasonable extrapolation. Using the estimated propensity scores, we adjusted the data for group differences in four ways: (1) propensity score stratification, in which the propensity scores were divided into five strata and stratum-specific hazard ratios calculated with Cox proportional hazards models were combined for an overall hazard ratio [33, 36, 37]; (2) propensity score 1:1 matching, using an algorithm with a caliper of 0.02 [38] to pair adjuvant CRT and gastrectomy-only patients similar in terms of their measurable characteristics; (3) regression adjustment (i.e. inclusion of propensity score as a linear predictor in the model); and (4) creation of stabilization weights defined as the inverse probability of treatment weighting [39, 40]. We present adjusted relative hazard ratios and 95% confidence intervals (CIs). An adjusted hazard ratio less than 1.00 indicated longer survival time among patients who underwent gastrectomy and received adjuvant CRT compared with those who underwent gastrectomy only, and an adjusted hazard ratio greater than 1.00 indicated shorter survival time.

Table 2 Crude median survival among patients in the two treatment groups and hazard ratios for overall survival after adjustment for patient characteristics
Full size table

To determine whether survival benefits differ by select patient and tumor characteristic, we performed subgroups analyses by cancer diagnosis age and tumor location. Sensitivity analyses evaluated the potential impact of defining adjuvant therapy on the basis of 2 or 4 months (versus 3 months in our base case analysis) from the date of gastrectomy and alternative definitions of adjuvant therapy that included any adjuvant chemotherapy (regardless of receipt of radiation therapy), only adjuvant chemotherapy, and only adjuvant radiation therapy.

All analyses were conducted with SAS, version 9.3 (SAS Institute, Cary, NC, USA). Statistical significance was set at p < 0.05, and all tests were two-tailed.

Results

Cohort description and patient characteristics

From an initial sample of 25,099 patients in whom gastric adenocarcinoma was diagnosed between 2002 and 2009, we identified 2936 patients who met the inclusion criteria (see Fig. 1). Among these patients, 1924 underwent gastrectomy within 6 months of diagnosis, did not receive any neoadjuvant therapy (e.g., preoperative chemotherapy and/or radiation therapy), and survived for more than 90 days after the procedure. A subset of 405 patients who received only adjuvant chemotherapy or radiation therapy within 3 months of gastrectomy were also excluded for the main analysis (but were included in sensitivity analysis), for a final cohort of 1519 patients. With adjuvant treatment defined as receipt of chemotherapy and radiation therapy within 3 months of gastrectomy, the final cohort included 886 patients (58.3%) who underwent gastrectomy only and 633 patients (41.7%) who underwent gastrectomy and received adjuvant CRT.

Compared with the gastrectomy-only group, adjuvant CRT treatment group had a higher proportion of patients younger than 75 years (75.5% vs 55.3%; p < 0.0001). They were also likelier to be in the higher median income quintile (28.4% vs 19.2%; p = 0.0002) and to have stage II or stage III cancer at diagnosis (84.0% vs 54.7%; p < 0.0001) (Table 1).

Patients who were treated with gastrectomy only were likelier to have stage IB cancer at diagnosis (45.3% vs 16.0%; p < 0.0001) or to have a Deyo modified Charlson comorbidity score of 2 or greater (23.6% vs 19.0%; p = 0.07). Differences in baseline characteristics of race/ethnicity, SEER region and birthplace were not statistically significant between the two groups (Table 1).

Survival outcomes

Overall survival

The median overall survival was 26.9 months [interquartile range (IQR) 40.9 months] for patients who also received adjuvant CRT within 3 months of gastrectomy versus 25.1 months (IQR 43.7 months) for patients who underwent gastrectomy only (Table 2). The Kaplan–Meier curves for overall survival are shown in Fig. 2. A greater proportion of patients who received adjuvant CRT were alive at 1 year after gastrectomy compared with those who underwent gastrectomy only (80.3% vs 73.5%). Cancer-specific survival was similar between the two groups (50% vs 47%; see Table 3).

Fig. 2
figure 2

Kaplan–Meier survival curves illustrating the survival of Medicare beneficiaries with diagnosed gastric adenocarcinoma according to the treatment regimen. a All patients, b patients with stage IB disease, c patients with stage II disease, and d patients with stage III disease. Number of individuals at risk at 12, 36, and 60 months masked

Full size image
Table 3 Estimated cancer-specific survival outcomes by treatment regimen
Full size table

In an unadjusted Cox proportional hazards model, there was a significant difference in overall survival between the two treatment groups (hazard ratio 0.83; 95% CI 0.73–0.95) (Table 4). Controlling for baseline characteristics, the multivariable model also found a statistically significant overall survival benefit associated with gastrectomy plus adjuvant CRT (hazard ratio 0.58; 95% CI 0.50–0.67). All four propensity-score-adjusted models found a similar adjuvant CRT benefit. For example, the propensity-score-stratified model found that those who received adjuvant CRT in addition to gastrectomy had a 39% relative reduction in mortality (hazard ratio 0.61; 95% CI 0.41–0.79) compared with those who underwent only gastrectomy (Table 4).

Table 4 Effect of additional adjuvant chemoradiotherapy compared with gastrectomy only on hazard ratios for overall survival
Full size table

Stage-specific survival

For patients with stage IB tumors, the median overall survival from the time of gastrectomy was similar between gastrectomy plus adjuvant CRT and gastrectomy alone [39.3 months (IQR 47.0) and 44.2 months (IQR 51.1)] (Table 2). In contrast, the addition of adjuvant CRT was associated with greater survival for patients with stage II [33.2 months (IQR 49.1) vs 19.2 months (IQR 31.7)] and stage III [19.3 months (IQR 19.9) vs 12.3 months (IQR 16.7)] cancers. For all stages at diagnosis, adjuvant CRT had a greater proportion of patients alive at 1 year after gastrectomy (stage IB 90.1% vs 87.3%; stage II 84.0% vs 67.3%; stage III 71.1% vs 52.8%); the absolute difference was greater for more advanced tumors. Figure 2 depicts Kaplan–Meier survival curves by the stage at diagnosis. For cancer-specific survival at 1-year after gastrectomy, a greater proportion of stage II cancer (87% vs 72%) and stage III cancer (75% vs 61%) patients treated with the addition of adjuvant CRT were also alive compared with those who underwent gastrectomy only; outcomes among the two groups were similar however for stage IB cancer patients (93% vs 92%) (Table 3).

For stage IB tumors, adjusted multivariable models did not detect a survival difference between gastrectomy only and gastrectomy plus adjuvant therapy (hazard ratio 1.02; 95% CI 0.71–1.45; Table 4). However, adjuvant CRT was associated with a statistically significant survival benefit for stage II (hazard ratio 0.50; 95% CI 0.39–0.61) and stage III (hazard ratio 0.58; 95% CI 0.45–0.73) cancers. The results for propensity-score-adjusted models were similar (see Table 4).

Subgroup analysis

The survival benefit did not differ by tumor location or age at cancer diagnosis (Table 4). However, for all subgroup analyses, no significant survival benefit was detected for stage IB cancers.

Sensitivity analysis

The results were insensitive to assumptions on the adjuvant CRT treatment window [hazard ratio 0.63 (95% CI 0.54–0.74) for 2 months and hazard ratio 0.59 (95% CI 0.51–0.68) for 4 months]. Similarly, if adjuvant therapy was defined as any chemotherapy within 3 months (with or without radiation therapy) or only adjuvant chemotherapy, the treatment effects were largely unchanged [hazard ratio 0.63 (95% CI 0.55–0.72) and hazard ratio 0.61 (95% CI 0.52–0.70) respectively]. In contrast, if adjuvant therapy consisted of only radiation therapy (i.e., no chemotherapy) within 3 months of gastrectomy, no survival benefit was detected compared with gastrectomy only [hazard ratio 1.01; 95% CI 0.80–1.27). The results were consistent for overall and stage-specific survival (see Table 4).

Discussion

Using SEER–Medicare data since 2002 when adjuvant CRT became part of US standard care [4, 5], we found that Medicare enrollees in whom gastric adenocarcinoma was diagnosed between age 65 years and age 79 years who received adjuvant CRT in addition to gastrectomy had more favorable survival. These findings suggest that older patients, who account for the most of gastric cancer cases, benefit from adjuvant CRT. In particular, patients in whom a stage II or stage III nonmetastatic gastric adenocarcinoma had been diagnosed received an estimated 42–50% relative reduction in death risk. Patients in whom stage IB tumors had been diagnosed demonstrated a negligible survival benefit associated with adjuvant CRT.

Among our study cohort of older patients in whom gastric adenocarcinoma had been diagnosed who underwent gastrectomy, we found that only 42% received adjuvant CRT within 3 months of gastrectomy. The vast majority of these patients had stage II or stage III disease at diagnosis. The reasons for these low rates of postoperative adjuvant treatment are uncertain, and may include the emergence of neoadjuvant therapies. Of note, although our study cohort excluded patients who received neoadjuvant therapy, when we included these individuals in our cohort, we found that 11% of individuals who underwent gastrectomy (n = 2458) received some type of neoadjuvant therapy, with the proportion increasing over time from 5.8% in 2002–2004 to 20.5% in 2008–2009 (Table S2).

Prior studies used the SEER [11,12,13,14,15] and SEER–Medicare data [19, 20] to evaluate the association between adjuvant treatment and survival, primarily using data collected before CRT became part of standard care in 2002 or incomplete chemotherapy information. Using SEER–Medicare data collected since 2002 that include detailed data on chemotherapy use, our current study adds to the growing body of evidence that suggests that adjuvant CRT, compared with gastrectomy alone, is associated with a survival benefit, in particular for patients with more advanced stage II and stage III tumors (Table 5).

Table 5 Comparison of estimated effect of adjuvant therapy on overall survival for different American Joint Committee on Cancer stages
Full size table

With the ability to distinguish between different types of adjuvant therapy on the basis of chemotherapy information available in the SEER–Medicare database we were also able to assess the benefit associated with alternative regimens, including a statistically significant survival benefit for any adjuvant chemotherapy, regardless of receipt of radiation therapy, and a negligible benefit for adjuvant radiation therapy if it is administered without any chemotherapy. Although not directly comparable, our results are consistent with findings from the Adjuvant Chemoradiation Therapy in Stomach Cancer (ARTIST) trial, which found that the addition of radiation therapy to chemotherapy did not significantly reduce recurrence after curative resection and D2 lymph node dissection among patients with gastric cancer [41]. Similarly, the CRITICS study did not find any differences in 5-year survival rates among patients who received postgastrectomy chemotherapy and CRT (after neoadjuvant therapy) [42], suggesting little benefit from intensifying postoperative treatment with radiation therapy.

As with any observational study based on administrative data, our study is subject to limitations. The patient cohort was limited to those who were Medicare beneficiaries living in a SEER region and aged 65 years or older in 2002–2009. These patients are not representative of all US gastric cancer patients who underwent gastrectomy. Although the SEER–Medicare database is large, gastric cancer is uncommon in the USA. Therefore, our cohort size is too small to detect differences between certain subgroups. The database also does not have information on total radiation therapy dose, and it is possible that some patients in the adjuvant CRT group did not complete the course of treatment or received nonstandard doses. In addition, the SEER–Medicare database does not have complete information for patients who received care through an HMO. Some studies have suggested that patterns of care and comorbidities differ between Medicare patients and HMO patients [43, 44]. Although agreement between SEER and Medicare databases for surgical procedures was not completely concurrent, studies have shown high concordance between the two databases [45]. Since we anchored survival on the date of gastrectomy, approximately 300 patients listed as having surgery in the SEER database (with no information on the date of procedure) but not in the Medicare database were excluded. Lastly, although we used propensity score adjustment to account for imbalances between the two groups based on known factors, selection bias based on other unknown, unmeasured variables, such as tumor residual information, may have influenced treatment choice.

Conclusion

Our findings suggest that adjuvant CRT in conjunction with gastrectomy is associated with a survival benefit among older patients in whom stage II and stage III gastric adenocarcinoma has been diagnosed. Comparative effectiveness studies that capture both the morbidity risks and survival benefits associated with adjuvant CRT are needed to further identify opportunities to improve outcomes of gastric cancer treatment.