Definitive chemoradiotherapy versus neoadjuvant chemoradiotherapy followed by surgery for stage II-III esophageal squamous cell carcinoma

Abstract

Objective

Definitive chemoradiotherapy (CRT) remains the most commonly used treatment for locally advanced esophageal squamous cell carcinoma (SCC), owing to perceptions that esophagectomy offers an unclear survival advantage. We compare recurrence, overall survival (OS), and disease-free survival (DFS) in patients treated with definitive CRT or neoadjuvant chemoradiotherapy followed by surgery (trimodality).

Methods

This was a retrospective cohort study of patients with stage II and III SCC of the middle and distal esophagus in patients who completed CRT. Treatment groups were matched (1:1) on covariates using a propensity-score matching approach. The effect of trimodality treatment, compared with definitive CRT, on OS, DFS, and site-specific recurrence was evaluated as a time-dependent variable and analyzed using Cox regression with a gamma frailty term for matched units.

Results

We included 232 patients treated between 2000 and 2016: 124 (53%) with definitive CRT and 108 (47%) with trimodality. Trimodality was used less frequently over time (61% before 2009 and 29% after 2009, P<.0001). After matching, each group contained 56 patients. Median OS and DFS were 3.1 and 1.8 years for trimodality versus 2.3 and 1.0 years for CRT. Surgery was independently associated with improved OS (HR 0.57, 95% CI 0.34–0.97, P=0.039) and DFS (HR 0.51, 95% CI 0.32–0.83, P=0.007).

Conclusions

CRT followed by surgery may decrease local recurrence and increase DFS and OS in patients with esophageal SCC. Until better tools to select patients with pathological complete response are available, surgery should remain an integral component of the treatment of locally advanced esophageal SCC.

Introduction

Following the enthusiastic results of the CROSS trial, which proved the superiority of multimodality therapy over surgery alone for esophageal cancer,^1–2 neoadjuvant chemoradiotherapy (CRT) has become the standard of care for locally advanced disease in the United States. However, despite the recommendations of the National Comprehensive Cancer Network,³ the role of surgery after CRT as an indispensable component of curative therapy has been challenged in the last few years. In a recent analysis of the Surveillance Epidemiology and End Results–Medicare Linked Database, we found that only 5% of patients with locally advanced esophageal cancer received neoadjuvant therapy followed by surgery; most patients (49%) were treated with definitive CRT. The most common reason for exclusion of surgery was that surgery was not recommended (73%).⁴ This observed trend is likely attributable to several important misconceptions: (1) a nihilistic approach toward esophageal cancer in general, which is often considered a systemic disease at diagnosis without much hope for local therapy; (2) an unfavorable reputation of esophagectomy as a surgical procedure with high postoperative complication rates and impaired long-term quality of life; and (3) a perception that CRT leads to pathological complete response (pCR) in a high percentage of patients. The latter statement is particularly true for squamous cell carcinoma (SCC) of the esophagus, which is infrequently seen in the United States today but has been shown to have a higher sensitivity to CRT, with pCR rates up to 46%–49%.^1,5

Two major randomized trials compared definitive CRT with neoadjuvant CRT followed by surgery (trimodality) for esophageal SCC. Although surgery was associated with improved local control, both trials failed to show a survival benefit when surgery was added to the treatment regimen.^6–7 Conversely, several retrospective studies have found improved survival outcomes with trimodality treatment, supporting the view that definitive CRT should be reserved for patients who are unsuitable for surgery.^8–12 Although it seems reasonable to assume that patients with pCR would not benefit from surgical resection, there are no reliable tools to preoperatively identify these patients. We previously demonstrated that, after completion of CRT, a reduction of PET maximum standardized uptake value (SUV_max) >70%, a normal- appearing endoscopic examination, and no residual disease on biopsy are predictive of pCR. However, approximately 30% of patients with these characteristics still harbor disease. Nevertheless, we have observed an institutional shift in the approach to esophageal SCC, wherein selective referral for surgery is chosen for patients with suspected residual disease and close observation is chosen for patients with a high likelihood of pCR.⁵ The effect of this approach on outcomes is unclear.

The purpose of this study was to assess outcomes—overall survival (OS), disease-free survival (DFS), and site-specific recurrence—between definitive CRT and trimodality treatment for patients who completed CRT for locally advanced esophageal SCC.

Material and Methods

Study Population

Patients with locally advanced esophageal SCC (clinical stage II-III) treated with either definitive CRT or trimodality treatment between 2000 and 2016 were included in the study. Patients were identified from a Memorial Sloan Kettering Cancer Center institutional database or a prospectively maintained thoracic surgical database. Patients with other cancers were included only if those cancers were considered cured or if that prognosis was better than their esophageal cancer prognosis. Patients with cancer of the cervical or upper thoracic esophagus (endoscopically measured up to 25 cm from the incisors) were excluded. Other exclusion criteria included clinical stage I or IV disease (according to the 7^th edition of the American Joint Committee on Cancer Staging Manual); treatment consisting of radiation, chemotherapy, or surgery alone; a radiation dose of <45 Gy; chemotherapy alone used for induction; a history of esophageal cancer. The final analysis cohort includes patients who completed CRT, defined as being alive with no progression of disease during treatment.. This study was reviewed and approved by our Institutional Review Board.

Variables

Patient demographic characteristics, comorbidities, tumor histological features, stage of disease, treatment details, surgical complications, and recurrence and survival status were retrieved from a prospectively maintained surgical database or the patient’s electronic medical record. Assessment of clinical stage was performed with EGD and biopsy, EUS, CT, and PET. PET SUV values were retrieved for both pre- and posttreatment evaluation. Clinical stage was retrospectively classified according to the 7^th edition of the American Joint Committee on Cancer Staging Manual. At the end of CRT, clinical response was evaluated using physical examination, clinical assessment of dysphagia, and PET/CT. Endoscopy with biopsy was selectively performed in patients with complete clinical response.

Recurrence was classified as local if disease was identified within the esophagus or stomach; as regional if disease was observed within the mediastinal, periesophageal, or gastric, celiac, and/or supraclavicular lymph nodes; and distant if disease was observed in other organs and/or retroperitoneal nodes.

Patients were grouped according to the treatment they received. Patients who were treated with definitive CRT and later underwent surgery for recurrence of disease were classified as patients undergoing salvage esophagectomy. For patients who underwent esophagectomy, pathological stage and response to treatment were determined by the amount of residual viable tumor in the surgical specimen, as previously described.¹³ A pCR was defined as the absence of histological evidence of cancer in the surgical specimen.

Statistical Analysis

The definitive CRT and trimodality treatment groups were compared in terms of demographic, tumor, and clinical characteristics using Fisher’s exact test for categorical variables or the Wilcoxon rank sum test for continuous variables. Descriptive data were summarized as frequencies and percentages for categorical variables and medians and interquartile ranges (IQRs) for continuous variables. Owing to the retrospective nature of this cohort, two issues were addressed: (1) potential selection bias for surgery and (2) guaranteed survival time for the trimodality group. The first issue was addressed by propensity-score matching between definitive CRT and trimodality; the second issue was addressed by considering surgery as a time-dependent variable (described below).

Propensity-score matching was conducted to overcome potential selection bias for the choice of treatment, by balancing treatment groups on patient and pretreatment characteristics. Treatment groups were defined as receiving surgery following CRT (trimodality) or not receiving surgery following CRT (definitive CRT). Patients who received surgery as part of a salvage regimen following recurrence were considered to be definitive CRT patients and were used as potential controls for matching. The propensity score—defined as the probability of receiving trimodality treatment, given the covariates—was estimated using logistic regression. The covariates included age, sex, comorbidities (pulmonary, cardiac, diabetes, previous cancer), previous chest radiation, tumor location, histological grade (poor vs. well/moderate), clinical stage (II vs. III), induction chemotherapy, chemotherapy duration, and radiation dose (≥5040 vs. <5040 cGy). Year of treatment (before or after 2009) was considered a proxy for potential changes in clinical practices. It was included as an adjustment factor (or control variable) in all time-to-event analyses. A one-to-one “nearest neighbor” matching without replacement was performed using the MatchIt package in R. The logit of the propensity score (PS) was used as the matching scale, and 0.2 times the standard deviation of the logit (PS) was used as the initial matching caliper.¹⁴ The caliper width was reduced until a matched sample with optimal balance in covariates between groups was formed. Balance diagnostics after matching were evaluated by estimating the standardized mean difference (SMD).¹⁵ An SMD <0.1 was considered a negligible difference and was indicative of success of the matching procedure.¹⁶ Analyses account for matching either through paired tests (McNemar’s or Wilcoxon signed-rank test) or with a gamma frailty term on the matched pairs in Cox regression models.

The primary objective was to evaluate the effect of trimodality treatment, compared with definitive CRT, on OS and DFS. OS was calculated from the end of CRT until death or last follow-up. Patients who received salvage surgery after recurrence were censored on the date of salvage surgery. The purpose of this approach was to evaluate only the time receiving CRT. DFS was calculated from the end of CRT until recurrence, death, or last follow-up. We also considered the issue of “guaranteed survival time”: patients who received trimodality had to complete treatment and survive without further progression in order to receive surgery, whereas patients who received definitive CRT did not have this survival requirement. The longer survival requirement for the surgery patients may result in the CRT group unfairly appearing to have worse survival prognosis.¹⁷ To address this potential bias, surgery was considered a time-dependent variable, where all patients begin in the CRT group and switch to the trimodality group on the date of surgery (if any). This approach appropriately credits survival time before surgery to the CRT group.

A visual representation of OS and DFS stratified by treatment group, where trimodality treatment was treated as a time-dependent variable, was plotted using a nonparametric method by Simon and Makuch.¹⁸ To account for the propensity-score matched pairs, univariable and multivariable analyses for associations with OS and DFS were performed using Cox regression with a gamma frailty term, fixing year of treatment as an adjustment factor in addition to the treatment group (variable of interest).¹⁹

As a secondary objective, the types of recurrence that were associated with treatment were examined. Time to any recurrence was calculated from the end of CRT until the date of recurrence or of the last follow-up without detection of recurrence. If multiple recurrence types were detected at the same time, the patient was classified as having the most-severe recurrence type (i.e., distant > regional > local). We defined failure from the recurrence type of interest as events and failure from other recurrence types as censored observations. The same was done for each recurrence type. The cumulative incidence curves were generated from 1 minus the Kaplan-Meier estimates. We used cause-specific hazards Cox regression and performed nested likelihood ratio tests to obtain the P value, evaluating surgery as a time-dependent variable, including year of treatment as an adjustment factor.

Two-sided P<0.05 was considered statistically significant. Statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC) and R 3.2.4 (R Development Core Team, Vienna, Austria).

Results

Demographic Characteristics

We identified 491 patients with esophageal SCC treated between 2000 and 2016, of which 232 met the inclusion criteria. A CONSORT diagram with details on included and excluded patients is included as Supplementary Figure 1. Definitive CRT was the primary treatment for 124 patients (53%), whereas 108 patients underwent trimodality treatment (47%). Patient characteristics are summarized in Table 1. Patients in the definitive CRT group were older and had a higher rate of comorbidities. The temporal distribution of treatment reflects our institutional change of practice: before 2009, most patients who were good surgical candidates were offered surgery (61% patients received trimodality treatment), whereas after 2009, only patients with suspected persistence of disease were encouraged to undergo surgery (29% patients received trimodality treatment) (P<0.0001). Most patients were evaluated with PET/CT before treatment and after completion of CRT. Patients treated at our institution received induction chemotherapy before CRT, as it has become our standard practice to change chemotherapy during concurrent radiation for patients assessed to be nonresponders to induction chemotherapy by PET scan.²⁰ Most patients treated at our institution received the same dose of radiation therapy (5040 cGy) independently of surgical plans. There were no major differences in posttreatment SUV_max between the two groups. Of the patients who received definitive CRT, 17 underwent salvage surgery (16 for local and 1 for regional recurrence). Salvage surgery was abandoned in 2 patients determined to have unresectable disease. A comparative analysis of patients who underwent salvage surgery and those who underwent trimodality treatment found a higher rate of grade ≥3 pulmonary complications (29% vs. 11%, P=0.06) and 30-day mortality (18% vs. 2%, P=0.02) among patients who underwent salvage surgery (Supplementary Table 1).

Table 1.

Demographic and clinical characteristics pre- and postmatching

	Prematching				Postmatching
Characteristic	CRT	Trimodality	P	SMD	CRT	Trimodality	P	SMD
N	124	108			56	56
Matched variables
Age, years, median (IQR)	70 (63–76)	60 (56–67)	<0.001	0.908	65 (57–71)	64.5 (59–69)	0.686	0.078
Chemotherapy duration, months, median (IQR)	2.3 (1.7–2.8)	2.1 (1.3–2.5)	0.006	0.385	2.2 (1.6–2.6)	2.3 (1.4–2.7)	0.87	0.108
Male	68 (55)	58 (54)	0.97	0.023	30 (54)	30 (54)	0.99	<0.001
Pulmonary	15 (12)	11 (10)	0.80	0.061	6 (11)	5 (9)	0.99	0.060
Cardiac	91 (73)	42 (39)	<0.001	0.741	32 (57)	31 (55)	0.99	0.036
Diabetes	18 (15)	7 (7)	0.079	0.264	8 (14)	6 (11)	0.79	0.108
Previous cancer	38 (31)	7 (7)	<0.001	0.654	7 (13)	6 (11)	0.99	0.056
Previous chest radiation	11 (9)	3 (3)	0.095	0.262	1 (2)	2 (4)	0.99	0.111
Middle location	64 (52)	46 (43)	0.22	0.181	22 (39)	24 (43)	0.85	0.073
Poor differentiationa	49 (42)	33 (33)	0.21	0.192	22 (39)	21 (38)	0.99	0.037
Clinical stage III	85 (69)	68 (63)	0.45	0.118	36 (64)	36 (64)	0.99	<0.001
Induction chemotherapy	87 (70)	57 (53)	0.010	0.363	37 (66)	36 (64)	0.99	0.037
Radiation dose ≥5040 cGy	115 (93)	96 (89)	0.43	0.134	53 (95)	53 (95)	0.99	<0.001
Unmatched variables
Treatment after 2009	73 (59)	30 (28)	<0.001	—	31 (55)	16 (29)	0.007	—
SUVmax posttreatment, median (IQR)	3.2 (0.0–5.2)	3.5 (0.8–5.2)	0.47	—	3.1 (0.0–5.2)	3.8 (0.0–7.0)	0.33	—

Unless otherwise noted, data are no. (%). CRT, chemoradiotherapy; IQR, interquartile range; SMD, standardized mean difference; SUV_max, maximum standardized uptake value; trimodality, neoadjuvant chemoradiotherapy followed by surgery.

^aUnknown for 16 patients prematching.

Among the patients who underwent trimodality treatment, the median time from end of CRT to surgery was 2 months (IQR 1.6–2.9 months). The most common operation was Ivor Lewis esophagectomy (64% of cases), followed by 3-hole (33%) and transhiatal (3%) esophagectomy. Esophagectomy was abandoned in 2 patients deemed to have unresectable disease; in 1 patient, only an R2 resection was possible. Postoperative 30-day mortality was 2%; 27% of patients experienced overall grade ≥3 complications. A pCR was achieved in 49 patients (47%).

The propensity-score matching procedure yielded a balanced cohort of 112 patients—56 in each treatment group (Table 1). After matching, all 13 characteristics were balanced between the two groups, with the exception of chemotherapy duration (SMD=0.108), diabetes (SMD=0.108), and previous chest radiation (SMD=0.111). The two groups did not differ in other variables not included in the propensity-score matching, except that there were higher proportions of definitive CRT treated recently.

Overall Survival

In the matched cohort of 112 patients, there were 66 deaths. Median follow-up among survivors after the end of CRT was 2.3 years (range 0.3–14.6 years). Median OS was 2.3 years (95% CI 1.5–3.5 years) for definitive CRT and 3.1 years (95% CI 2.1–8.3 years) for trimodality treatment. The 5-year OS was 29% (95% CI 18%–49%) for definitive CRT and 45% (95% CI 33%–62%) for trimodality treatment (Figure 1).

Figure 1.

Overall survival, stratified according to treatment

Univariable analysis results for association with OS are shown in Table 2. In multivariable analysis, trimodality treatment was significantly associated with improved OS (HR 0.57, 95% CI 0.34–0.97, P=0.039), after year of treatment was controlled for. Among patients with available data on SUV following treatment (n=69), the effect of trimodality treatment remained protective and was significant in a model controlling for year of treatment and SUV_max posttreatment (HR 0.40, P=0.05). For sensitivity analyses, we included each of the three variables with an ASMD >0.1 into the final multivariable model for OS individually: chemotherapy duration, diabetes, and previous chest radiation. None of the three variables was significantly associated with OS. In addition, the protective effect of trimodality treatment was retained after the addition of the imbalanced variables (Supplementary Table 2).

Table 2.

Cox regression analysis for association between factors and overall survival (OS) using matched data^*

Factor(s)
	HR	(95% CI)	P
Treatment Trimodality vs. CRT	0.57	(0.34–0.97)	0.039
Age per year	1.02	(0.99–1.04)	0.263
Chemotherapy duration per month	0.9	(0.65–1.24)	0.514
Sex Male vs. Female	1.27	(0.78–2.08)	0.338
Pulmonary Yes vs. No	1.62	(0.69–3.8)	0.268
Cardiac Yes vs. No	1.27	(0.78–2.08)	0.343
Diabetes Yes vs. No	1.97	(0.95–4.08)	0.068
Previous Cancer Yes vs. No	0.81	(0.38–1.71)	0.58
Previous Chest Radiation Yes vs. No	1.93	(0.46–8.08)	0.367
Location Mid vs. Distal	1.19	(0.73–1.94)	0.476
Differentiation Poor vs. Well/Moderate	1.35	(0.83–2.21)	0.224
Clinical Stage 3 vs. 2	1	(0.61–1.66)	0.993
Induction Chemo Yes vs. No	0.82	(0.49–1.35)	0.43
Radiation dose >=5040 vs. <5040	0.82	(0.3–2.28)	0.706
SUV max post-treatment	1.06	(0.97–1.15)	0.18

CRT, chemoradiotherapy; SUV_max, maximum standardized uptake value.

^*Year of treatment (before vs. after 2009) and a gamma frailty term on the matched pairs were included in the analysis of each variable.

Disease-Free Survival

There were 51 recurrences and 80 recurrence or death events in the matched cohort. Median DFS was 1.0 years (95% CI 0.7–1.7 years) for definitive CRT and 1.8 years (95% CI 1.2–6.9 years) for trimodality treatment. The 5-year DFS was 23% (95% CI 14%–40%) for definitive CRT and 40% (95% CI 29%–56%) for trimodality treatment (Figure 2). Results of the univariable analysis for association with DFS are shown in Table 3. In multivariable analysis, trimodality treatment was significantly associated with improved DFS (HR 0.51, 95% CI 0.32–0.83, P=0.007), after year of treatment was controlled for. Among patients with available post- CRT SUV_max data (n=69), the effect of trimodality treatment remained protective after controlling for year of treatment and SUV_max posttreatment (HR 0.55, P=0.065). In sensitivity analyses, of the three variables with an ASMD >0.1, none was significantly associated with DFS. In addition, the protective effect of trimodality was retained after the addition of the imbalanced variables (Supplementary Table 3).

Figure 2.

Kaplan-Meier curve demonstrating disease-free survival, stratified by treatment group

Table 3.

Cox regression analysis of disease-free survival using matched data^*

Factor(s)
	HR	(95% CI)	P
Treatment Trimodality vs. CRT	0.51	(0.32–0.83)	0.007
Age per year	1.01	(0.99–1.04)	0.392
Chemotherapy duration per month	0.86	(0.65–1.15)	0.31
Sex Male vs. Female	1.38	(0.88–2.17)	0.157
Pulmonary Yes vs. No	1.78	(0.87–3.63)	0.113
Cardiac Yes vs. No	1.28	(0.81–2)	0.288
Diabetes Yes vs. No	1.5	(0.78–2.89)	0.223
Previous Cancer Yes vs. No	0.87	(0.43–1.76)	0.706
Previous Chest Radiation Yes vs. No	1.67	(0.4–6.94)	0.479
Location Mid vs. Distal	0.94	(0.6–1.47)	0.79
Differentiation Poor vs. Well/Moderate	1.37	(0.88–2.15)	0.163
Clinical Stage 3 vs. 2	1.08	(0.68–1.71)	0.749
Induction Chemo Yes vs. No	0.71	(0.45–1.13)	0.15
Radiation dose >=5040 vs. <5040	0.75	(0.3–1.86)	0.532
SUV max post-treatment	1.04	(0.96–1.12)	0.343

CRT, chemoradiotherapy; SUV_max, maximum standardized uptake value.

^*Year of treatment (before vs. after 2009) and a gamma frailty term on the matched pairs were included in the analysis of each variable.

Recurrence

Twelve patients had local recurrence only, 12 had regional recurrence, and 27 had distant recurrence. The 5-year cumulative incidence of local recurrence was 38% (95% CI 15%–55%) for patients treated with definitive CRT, compared with 0% for patients treated with trimodality (P<0.001) (Figure 3). Regional and systemic recurrence rates did not significantly differ between the two groups (5-year cumulative incidence of regional recurrence was 19% versus 18% and distant recurrence was 38% versus 27% for CRT versus trimodality, respectively) (Figure 4A–B).

Figure 3.

Cumulative incidence of local recurrence

Figure 4.

Cumulative incidence of regional (a) and distant (b) recurrence

Discussion

In this study, we sought to evaluate the role of surgery for curative intent after CRT in patients with locally advanced esophageal SCC. We found a statistically significant decrease in risk of recurrence, particularly local recurrence, and increased DFS and OS among patients who received surgery after induction CRT. Moreover, trimodality treatment was an independent factor associated with prolonged DFS and OS. As definitive CRT has become the most common treatment for locally advanced esophageal SCC in the United States, despite national guidelines recommending trimodality treatment for patients suitable for surgery, our results are particularly relevant.⁴ The results of two randomized, controlled trials from the previous decade, showing no benefit in patients receiving surgery, have certainly contributed to the belief that surgery does not offer much advantage and potentially leads to increased morbidity and impaired quality of life.^6–7 However, these trials have been strongly criticized, and the quality of their evidence has not been considered appropriate to generate recommendations for best treatment practices.³ In the trial reported by Stahl et al. from Germany, mortality in the surgical group was 13% overall, with the majority of deaths occurring during the perioperative period (11%).⁷ This is significantly higher than the perioperative mortality reported by highly specialized and high-volume centers^21–25 and the 2% in this study. Moreover, patients in the surgical arm received only 40 Gy of radiation, whereas patients in the nonsurgical arm received 50 Gy. Only 66% of patients randomized to surgery actually went on to receive surgery, and their 3-year OS of 31% was much lower than the 58% reported in the CROSS trial and the 52% (95 % CI 40%–68%) in this study. In the French trial reported by Bedenne et al., only patients who responded to CRT were randomized, and the study included patients with SCC and adenocarcinoma. Again, there was a high 90-day mortality (9%) in the surgical arm, a low rate of complete resection (75%), and a 2-year survival of 34%, which is significantly lower than the 67% reported in the CROSS trial and the 63% (95% CI 51%–78%) in this study.^2,6 Conversely, several nonrandomized studies from the United States and the East, where this cancer is much more prevalent, have reported results in accordance with ours. Nomura et al. and the Japan Clinical Oncology Group (JCOG) pooled data from two previous prospective trials of patients with locally advanced esophageal SCC treated with either definitive CRT (JCOG9906)²⁶ or trimodality treatment (JCOG9907)²⁷ and reported significantly improved progression-free survival and OS for patients treated with esophagectomy (definitive CRT vs trimodality: adjusted HR 1.76 [95% CI 1.3–2.5] vs 1.72 [95% CI 1.2–2.5]).²⁸ Similarly, Liao et al. reported statistically significant improvements in OS, DFS, and locoregional control with esophagectomy in a matched-pair analysis of patients with stage II and III esophageal cancer.⁹

In this study, we found that local recurrence was the main reason for treatment failure in patients who received definitive CRT. The 5-year cumulative incidence of distant recurrence was, in fact, similar between groups (38% for definitive CRT vs 27% for trimodality), suggesting that local control may influence survival in these patients. This result is also supported by both the German and French randomized trials mentioned above, which showed higher local control rates and higher DFS with surgery. The difference in 5-year cumulative incidence of local recurrence between definitive CRT (38%) and trimodality treatment (0%) highlights the inability to accurately assess for tumor response and the inaccuracy of clinical complete response. The differences in DFS between the two groups were likely driven by worse local control in the definitive CRT group.

We do not, at present, have a reliable clinical tool to preoperatively identify pCR—assuming that pCR is meaningful of absence of locoregional disease. Even with a negative endoscopic examination, benign histological findings on biopsy, and significantly decreased SUV_max after treatment, up to 30% of patients harbor disease in their esophagus. Waiting longer before performing surgery may seem appealing in that it allows for better selection of patients with residual disease or recurrence; however, in our experience, salvage esophagectomy is associated with higher morbidity and worse overall outcomes and should not be taken lightly. Previous studies on this subject have shown contradictory results.^29–33 The delayed effects of radiation might not only eliminate dissection planes but might also significantly impair lung function and overall healing. The overall pCR rate in this study was 47%, which is similar to the 49% reported in the CROSS trial.¹ Overall, more than half of patients who underwent surgery had residual disease found in their specimen. This is important information that should be shared with patients and providers.

Among the limitations of this study is its retrospective design, which likely introduced selection bias—the most important being the direction of healthy patients to the trimodality group. Although this trend was probably true before 2009, the matching procedure takes into account clinically relevant differences that were observed. After 2009, most patients directed to surgery were those with findings suspicious for persistent disease; therefore, patients with the worst prognosis were potentially selected into the trimodality group. Nevertheless, even with a higher percentage of nonresponders in the trimodality group (66% vs 47%), these patients had higher OS and DFS. Bias due to the survival requirement among the surgery group was also a concern; to address this, we excluded patients from the CRT group who were unable to complete CRT due to early death or recurrence to ensure adequate follow-up after CRT and to have comparable groups. Another important limitation was the use of different chemotherapy agents during the study period and the use of induction chemotherapy before CRT and PET-directed chemotherapy only for patients treated at our institution. Moreover, our study cohort may not match the US population in terms of the distribution of stage of esophageal cancer patients, since our data are from a tertiary cancer treatment center and only patients with locally advanced disease were considered for our study.

Overall, this study has several strengths. Only patients with esophageal SCC who completed CRT were included, and, in the majority of cases, clinical staging included PET and EUS. There was no difference in radiation dose between the two groups, and all patients received a dose of at least 45 Gy. Even if the median posttreatment SUV_max was approximately 3 in both groups, most patients selected for observation after CRT had a higher PET-SUV reduction (the rate of SUV reduction >70% was 66% for the definitive CRT group and 47% for the trimodality group in the matched groups, respectively), perhaps highlighting a higher rate of clinical response to treatment in the definitive CRT group. Despite this fact, patients who underwent surgery still had better outcomes than patients treated with definitive CRT.

In conclusion, surgery should be considered after completion of CRT for treatment of esophageal SCC in patients who are fit for surgery. Surgery was associated with a decreased probability of local recurrence and increased DFS and OS. As there is not, at present, a reliable way to preoperatively identify patients with pCR, and as salvage surgery may be associated with increased morbidity and mortality, the potential advantages of surgical resection after CRT should be reviewed and discussed with patients undergoing treatment for esophageal SCC.

Supplementary Material

2

Supplementary Figure 1. CONSORT flow diagram demonstrating selection of patients included in the current study, based on inclusion and exclusion criteria.

3

Supplementary Table 1. Overview of surgical patients: definitive chemoradiotherapy followed by salvage surgery versus trimodality

Supplementary Table 2. Models of overall survival: treatment, year of treatment, and unbalanced covariates

Supplementary Table 3. Models of disease-free survival: treatment, year of treatment, and unbalanced covariates

Figure 5. Central picture.

Overall survival stratified according to treatment

Daniela Molena, MD, discussing the implications of the study

Perspective statement.

Since patients with esophageal SCC have a high rate of pCR after chemoradiotherapy, surgery is often considered optional and reserved for patients with poor response. In this study, we demonstrate that CRT followed by surgery is associated with a lower risk of local recurrence and improved DFS and OS. Patients with esophageal SCC should be consulted about the potential advantages of surgical resection.

Central message.

Surgery after chemoradiation in the treatment of esophageal SCC is associated with improved OS and DFS and decreased risk of local recurrence.

Glossary of abbreviations

CRT

chemoradiotherapy

DFS

disease-free survival

IQR

interquartile range

OS

overall survival

pCR

pathological complete response

SCC

squamous cell carcinoma

SMD

standardized mean difference

SUVmax

maximum standardized uptake value

Trimodality

neoadjuvant chemoradiotherapy followed by surgery