Kukar M, Alnaji RM, Jabi F, Platz TA, Attwood K, Nava H, Ben-David K, Mattson D, Salerno K, Malhotra U, Kanehira K, Gannon J, Hochwald SN. Role of Repeat 18F-Fluorodeoxyglucose Positron Emission Tomography Examination in Predicting Pathologic Response Following Neoadjuvant Chemoradiotherapy for Esophageal Adenocarcinoma. JAMA Surg. 2015;150(6):555-562. doi:10.1001/jamasurg.2014.3867
Predicting complete pathologic response (CPR) preoperatively can significantly affect surgical decision making. There are conflicting data regarding positron emission tomography computed tomography (PET CT) characteristics and the ability of PET CT to predict pathologic response following neoadjuvant chemoradiotherapy in esophageal adenocarcinoma because most existing studies that include squamous histology have limited numbers and use nonstandardized PET CT imaging.
To determine if PET CT characteristics are associated with CPR in patients undergoing trimodality treatment for esophageal adenocarcinoma.
Design, Setting, and Participants
A retrospective medical record review was conducted at a large tertiary cancer center from a prospectively maintained database from January 1, 2005, to December 31, 2012. Inclusion criteria were patients undergoing esophagectomy for locally advanced esophageal adenocarcinoma post–neoadjuvant chemoradiotherapy with 2 standardized PET CT studies done at our institution (pre–neoadjuvant chemoradiotherapy and post–neoadjuvant chemoradiotherapy) for review. Data collected included clinical, pathologic, imaging, and treatment characteristics.
Main Outcome and Measure
The primary study outcome was the association of PET CT characteristics with histologic confirmed pathologic response.
Of the total participants, 77 patients met the inclusion criteria. Twenty-two patients (28.6%) had CPR vs 55 patients (71.4%) who had incomplete pathologic response. The 2 groups were similar in age, sex, race/ethnicity, comorbid conditions, Eastern Cooperative Oncology Group status, tumor grade, chemotherapy, and radiation regimen and days between the 2 PET CTs. The mean prestandardized uptake variable (SUV; 14.5 vs 11.2; P = .05), δ SUV (10.3 vs 5.4; P = .02), and relative δ SUV (0.6 vs 0.4; P = .02) were significantly higher in those with CPR vs incomplete pathologic response. Using the Youden Index, a δ SUV value less than 45% was predictive of residual disease with a positive predictive value of 91.7% (95% CI, 73-99; P < .05).
Conclusions and Relevance
To our knowledge, this is the largest study examining the role of PET CT characteristics in esophageal adenocarcinoma for patients undergoing neoadjuvant chemoradiotherapy that demonstrates that δ SUV of less than 45% is associated with patients with residual disease but not CPR. Based on the findings from our study, the current recommendation is still surgical resection regardless of the posttherapy PET SUV in the primary tumor. However, our study highlights the ability to detect patients with residual disease and the need to critically evaluate these patients for consideration of additional therapies.
For locally advanced esophageal cancer, neoadjuvant chemoradiotherapy (nCRT) followed by surgical resection has been well established as the standard treatment modality in the United States.1,2 Although a diagnostic modality that could help assess treatment response and further guide treatment decisions would be ideal, limited options currently exist. Most treatment centers use positron emission computed tomography (PET CT) before and after neoadjuvant therapy primarily to exclude metastatic disease but also to assess the disease response post-nCRT. While PET CT may help to exclude metastatic disease, its role in accurately predicting pathologic response is debatable. Multiple studies have aimed to address this issue with conflicting results.
Some studies have concluded that post-nCRT PET CT fails to accurately predict pathologic response, has been shown to have low predictive ability, and, therefore, does not justify altering current treatment.3,4 Alternatively, other studies conclude that post-nCRT PET CT demonstrates appropriate correlation between metabolic vs pathologic response and is useful in predicting pathologic response.5,6 Analyzed PET CT parameters vary from one study to another but most use the standardized uptake value (SUV). Others have looked at values such as metabolic tumor volume and tumor glycolytic activity3,7- 11 and concluded that such parameters can be predictive of pathologic response, potentially guiding treatment post-CRT.9
Existing studies have several limitations that make it difficult to draw useful conclusions. In this current study, we addressed several limitations and aimed to evaluate the accuracy of PET CT in assessing pathologic response in patients with esophageal adenocarcinoma undergoing nCRT followed up by surgical resection at a large tertiary cancer center.
Patients were identified from a prospectively collected database from January 1, 2005, to December 31, 2012. Inclusion criteria included a diagnosis of esophageal adenocarcinoma and pre- and posttherapy PET CT scans obtained at our institute. Institutional review board approval was provided by the Roswell Park Cancer Institute. Participants did not provide consent owing to the retrospective nature of the study. Patients with a diagnosis of squamous cell carcinoma and/or outside PET CT imaging studies were excluded from the analysis. Data abstracted included demographic, clinical, radiologic, pathologic, treatment, and survival characteristics.
The 18F-fludeoxyglucose (FDG) PET CT imaging was performed on a Discovery ST (General Electric Medical Systems) with a postfilter resolution of 6.5-mm full-width-at-half maximum radiology. Computed tomography parameters ranged from 120 kV to 140 kV and 80 mA to 120 mA. The 18F-FDG PET CT images were reconstructed using standard ordered subset expectation maximization (2 iterations and 21 subsets). After a 4-hour fast, patients were given a dose of 15 mCi (to convert millicuries to millibecquerels, multiply by 3.7 × 1010; ±10%) and subsequently imaged 1 hour after injection from skull vertex to midthigh using 4-minute bed stop positions and correlative CT covering the same regions. Patients with a known diabetic history had blood glucose levels checked prior to imaging and all patients had a blood glucose level less than 200 mg/dL (to convert glucose to millimoles per liter, multiply by 0.0555).
Data were qualitatively and quantitatively analyzed using MedView postprocessing software (MedImage Inc) and retrospectively reviewed by an experienced nuclear medicine physician (J.G.) who was blinded to the clinical data at the time of image interpretation. The distribution of 18F-FDG in the distal esophageal mass on each scan was visually described as focal, diffuse, or heterogeneous. Uptake in a mass was reported as focal if it approximately assumed the shape of a sphere and was confined to a small segment shorter than 2 cm of the distal esophagus. A diffuse distribution involved the distal esophagus in a cylindrical pattern and a heterogeneous pattern of 18F-FDG uptake was reported as both focal and diffuse.
The SUV for body weight at maximum counts per pixel (SUVmax) was calculated automatically after manually drawing a region of interest around the esophageal lesion. Linear length was defined as the length of 18F-FDG avid tumor lesion linear extent (in centimeters) measured along the long axis of the distal esophagus and adjacent stomach (in approximately the craniocaudal direction) on the maximum projection image for each scan. Quantitative parameters analyzed included pre- and posttherapy SUVmax (SUV), δ SUV (post-SUV − pre-SUV), relative δ SUV (post-SUV − pre-SUV/post-SUV), mean linear length of uptake in centimeters pre- and posttherapy, and change in linear length. The interval between the pre-CRT and post-CRT PET CT scans was recorded in days.
All patients were treated with concurrent chemoradiation. The radiation dose was 50.4 Gy delivered in 1.8-Gy to 2-Gy fractions on Monday through Friday for a total of 28 fractions with concurrent chemotherapy. Three different chemotherapy regimens consisting of cisplatin with irinotecan, capecitabine with oxaliplatin, and carboplatin with paclitaxel were used in the patient cohort evaluated for this study. The cisplatin and irinotecan combination was administered intravenously at doses of 30 mg/m2 and 65 mg/m2, respectively, on days 1, 8, 22, and 29. Patients receiving the capecitabine/oxaliplatin combination received 3 doses of intravenous oxaliplatin at 85 mg/m2 on days 1, 15, and 29 along with oral capecitabine 625 mg/m2 twice a day on days of radiation. Carboplatin at the area under the curve 2 and paclitaxel at 50 mg/m2 were given intravenously weekly for a total of 6 doses in patients treated with this regimen.
Resected specimens underwent routine histopathologic examination and were also reviewed by a single pathologist (K.K.) for tumor viability to determine pathologic response. Residual esophageal cancer was assessed semiquantitatively based on the estimated percentage of viable cancer in relation to total cancer area, including the amount of radiation-induced tissue injury in mural histologic sections. The extent of viable cancer in the esophagectomy specimen was assigned to 1 of the following 3 categories: no viable tumor (complete response), 1% to 10% viable tumor (partial response), and more than 10% viable tumor (no response). Patients with partial response and no response were grouped together as an incomplete pathologic response (IPR) for the purpose of analysis.
Patient characteristics were reported by pathologic response as means, medians, and standard deviations for continuous variables and as frequencies and relative frequencies for categorical variables. Comparisons were made using the Wilcoxon rank sum test or Kruskal-Wallis and Fisher exact or χ2 tests for continuous and categorical variables, respectively. The ability of a variable to predict pathologic response was summarized using standard receiver operating characteristic methods. For continuous variables, the Youden Index was used to identify the optimal diagnostic threshold. The diagnostic measures of sensitivity, specificity, positive predictive value, and negative predictive value were reported as percentages with corresponding 95% confidence intervals (using the Clopper-Pearson method). The survival outcomes were summarized using standard Kaplan-Meier methods, with comparisons made using the log-rank test. All analyses were conducted in SAS version 9.3 (SAS Institute Inc) at a significance level of .05.
Seventy-seven patients were included in the analysis using the inclusion criteria and divided into 2 groups based on pathologic response (complete pathologic response [CPR; n = 22] vs IPR [n = 55]). Sixty-three patients underwent open Ivor-Lewis resection with 2-field lymphadenectomy. Fourteen patients underwent a minimally invasive thoracoscopic and laparoscopic esophagectomy with 2-field lymphadenectomy (5 patients with intrathoracic anastomosis and 9 patients with cervical anastomosis).
Table 1 details the demographic and clinical characteristics where both groups were similar in age, sex, race/ethnicity, Eastern Cooperative Oncology Group status, smoking and alcohol history, presence of Barrett disease, and grade of disease. Comorbidities including cardiac disease, hypertension, pulmonary disease, and diabetes mellitus were similarly distributed in both groups as well. Moreover, the chemotherapeutic regimens used were similar in both groups. The median interval between the 2 PET CTs was 96.6 days (range, 42-324 days). The median interval between the end of radiation therapy to the second PET CT was 32 days.
Table 2 details the quantitative and qualitative PET CT data used to correlate with pathologic response. The pretherapy mean SUV was higher in the CPR group vs the IPR group (14.45 vs 11.23; P = .05). The mean δ SUV and relative δ SUV were statistically significantly higher in the CPR group. Mean pre- and posttherapy linear length and change in linear length were similar in both groups. In terms of the qualitative parameters (pattern of uptake), pre- and posttherapy patterns of uptake were similar in both groups.
Because relative δ SUV provides more reliability, it was chosen as a variable to identify an optimal cutoff point to predict pathologic response. The Youden Index identified a δ SUV cutoff value of 45% for reliably predicting pathologic response. Patients with a less than 45% decrease in SUV were more likely to have an IPR with a positive predictive value of 91.7% (95% CI, 73.0-99.0). In contrast, this cutoff yielded a positive predictive value of 38% (95% CI, 24.8-52.1) for predicting CPR.
A subgroup analysis was completed, dividing the patients in the IPR group into 2 subgroups of partial response (n = 29) vs no response (n = 26). The 3 groups were analyzed and Table 3 details the radiographic characteristics for each response group. The pretherapy mean SUV was significantly the highest in the CPR group followed by the no response group and was lowest in the partial response group (14.45 vs 12.72 vs 9.89; P = .02). A similar trend was also observed with respect to the mean δ SUV, which was also statistically significant and highest for the CPR group followed by the no response and partial response groups (10.25 vs 7.48 vs 5.83; P = .05). As indicated, the δ SUV was not predictive of the amount of residual disease.
Median follow-up was 29.5 months (range, 1.2 months-84.5 months) with a 5-year actuarial survival rate of 36.3%. Three-year overall survival was 82.5% vs 53.1%, disease-specific survival was 82.5% vs 58.1%, and recurrence-free survival was 71.8% vs 39.8% for patients with CPR and patients with IPR, respectively (Figure). Both quantitative and qualitative PET characteristics including the relative δ SUV of 45% were not predictive of overall, disease-specific, or recurrence-free survival.
Despite several publications, existing literature remains unclear regarding the role of PET CT in predicting pathologic response in patients with esophageal adenocarcinoma following nCRT. Existing data regarding the ability of PET CT to predict pathologic response have several limitations. Apart from small sample size,8- 10,12- 14 a few studies focus exclusively on squamous histology6,15- 17 while others focus on mixed histology.9,10,12- 14,18- 20 Because squamous histology has been demonstrated to be more chemoradiosensitive than adenocarcinomas from a histological standpoint20- 22 and are associated with a higher rate of complete histopathologic response,18,23 it is difficult to draw meaningful conclusions.
Furthermore, the function of standardizing PET CT cannot be understated because there is a considerable variability in absolute SUV measurements between institutions and even within a system using different software.4,24 A similar problem exists in defining pathologic response, where several criteria have been used based on institutional preferences.12,19,25- 28 With this heterogeneity in the literature, we decided to evaluate our experience by standardizing most of the variables.
There were several key findings from our study. The pretherapy SUV was higher in complete pathologic responders compared with incomplete responders and also higher in the subgroup analysis with statistical significance. This was demonstrated in a study by Rizk et al,24 who showed that although the initial PET SUVmax did not predict survival, patients with higher initial SUVmax (cutoff, 4.5) were more likely to respond to preoperative therapy. Others have demonstrated this finding in the setting of squamous cell esophageal cancer, which routinely demonstrates higher initial SUVmax compared with esophageal adenocarcinoma and is more responsive to nCRT.18,20 This finding may help in selecting patients who will benefit most with nCRT, although this needs to be evaluated further because we were not able to identify a reliable cutoff.
Both relative and δ SUV were significantly higher in complete pathologic responders. Although we could not reliably identify a cutoff to predict CPR, we found that an δ SUV change of less than 45% reliably predicted residual disease. Although nonresponders have been shown to have a poor metabolic response, absolute cutoff values have not been reported. The 5-year overall and disease-specific survival for patients with IPR (data not shown) were similar to those undergoing upfront surgery in large randomized studies.1 This questions the function of neoadjuvant therapy in these patients. The findings from our study may form the basis of further investigations, including the need for additional or different neoadjuvant therapy regimens or the need for an extended/aggressive nodal dissection in these patients because we can reliably predict this subset of patients prior to surgical resection. Although the current recommendation is still surgical resection, our study results may have some effect on critically evaluating this subset of patients who have an IPR for consideration of additional therapies.
Although patients with CPR had a significantly higher recurrence-free survival, their overall and disease-specific survival were similar to patients with IPR. This may represent a type II error; however, the notion that CPR is a predictor of improved overall and disease-specific survival in esophageal cancer has been extrapolated from studies with either squamous histology or mixed histology. Our findings were also supported by a recent analysis by Kauppi et al,29 where 66 patients with esophageal adenocarcinoma were evaluated and histopathologic response was not predictive of overall or disease-specific survival in a multivariate analysis. This highlights another important point that local control using multimodality treatment in adenocarcinoma histology does not seem to affect overall outcome, although this may need to be validated with a large sample size. Another key finding from the subgroup analysis was that it was difficult to predict the amount of residual disease based on metabolic response because patients classified to have no pathologic response had more decreases in δ SUV and relative δ SUV. This could be explained by the fact that more aggressive tumors are more likely to respond to therapy but they may still have a substantial amount of residual tumor despite reduction in SUV.
In addition to the retrospective nature of the study, our study had some limitations. The data spanned from 2005 to 2012 and the chemotherapeutic regimen and radiation doses changed across time with evolving data, which may have potentially confounded pathologic responses. However, the distribution of chemotherapeutic regimen used in 3 groups were similar. Although the exact timing of second PET CT posttherapy has not been well-defined yet, we did not use strict criteria that may have potentially affected quantitative and qualitative PET data in the setting of radiation-induced inflammation. Qualitative data assessment is also less objective and subject to interobserver variability, although we did not find any statistical differences in the 2 groups. The SUV definition itself is subject to several sources of error, including patient size, measurement duration, plasma glucose concentration, recovery coefficients, partial volume, and region of interest. Additionally, our study was a single institution study and, although a strength in this case, it has some inherent flaws and limits generalizability.
Despite these limitations, this study was a single institution study containing a large number of patients with uniform histology. In addition, the PET CTs were standardized and imaging and pathologic data were reviewed by a single nuclear medicine physician and a single pathologist, respectively, to minimize interobserver variability. An δ SUV of less than 45% on PET CT was associated with residual disease but not CPR in patients with locally advanced esophageal adenocarcinoma treated with neoadjuvant chemoradiotherapy.
Corresponding Author: Steven N. Hochwald, MD, Department of Surgical Oncology, Roswell Park Cancer Institute, Elm Street and Carlton Street, Buffalo, NY 14263 (firstname.lastname@example.org).
Accepted for Publication: October 9, 2014.
Published Online: April 22, 2015. doi:10.1001/jamasurg.2014.3867.
Author Contributions: Dr Kukar had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Kukar, Nava, Ben-David, Salerno, Hochwald.
Acquisition, analysis, or interpretation of data: Kukar, Alnaji, Jabi, Platz, Attwood, Ben-David, Mattson, Salerno, Malhotra, Kanehira, Gannon, Hochwald.
Drafting of the manuscript: Kukar, Alnaji, Hochwald.
Critical revision of the manuscript for important intellectual content: Kukar, Jabi, Platz, Attwood, Nava, Ben-David, Mattson, Salerno, Malhotra, Kanehira, Gannon, Hochwald.
Statistical analysis: Kukar, Alnaji, Attwood.
Administrative, technical, or material support: Jabi, Platz, Ben-David, Mattson, Gannon.
Study supervision: Kukar, Nava, Salerno, Malhotra, Hochwald.
Conflict of Interest Disclosures: None reported.
Previous Presentation: This paper was presented in part as an oral presentation at the 67th Annual Cancer Symposium Meeting of the Society of Surgical Oncology; March 14, 2014; Phoenix, Arizona.