Metabolic tumour and nodal response to neoadjuvant chemotherapy on FDG PET-CT as a predictor of pathological response and survival in patients with oesophageal adenocarcinoma

Objectives
 2-deoxy-2[18F]Fluoro-d-glucose (FDG) PET-CT has an emerging role in assessing response to neoadjuvant therapy in oesophageal cancer. This study evaluated FDG PET-CT in predicting pathological tumour response (pTR), pathological nodal response (pNR) and survival. Methods Cohort study of 75 patients with oesophageal or oesophago-gastric junction (GOJ) adenocarcinoma treated with neoadjuvant chemotherapy then surgery at Guy’s and St Thomas’ NHS Foundation Trust, London (2017–2020). Standardised uptake value (SUV) metrics on pre- and post-treatment FDG PET-CT in the primary tumour (mTR) and loco-regional lymph nodes (mNR) were derived. Optimum SUVmax thresholds for predicting pathological response were identified using receiver operating characteristic analysis. Predictive accuracy was compared to PERCIST (30% SUVmax reduction) and MUNICON (35%) criteria. Survival was assessed using Cox regression. Results Optimum tumour SUVmax decrease for predicting pTR was 51.2%. A 50% cut-off predicted pTR with 73.5% sensitivity, 69.2% specificity and greater accuracy than PERCIST or MUNICON (area under the curve [AUC] 0.714, PERCIST 0.631, MUNICON 0.659). Using a 30% SUVmax threshold, mNR predicted pNR with high sensitivity but low specificity (AUC 0.749, sensitivity 92.6%, specificity 57.1%, p = 0.010). pTR, mTR, pNR and mNR were independent predictive factors for survival (pTR hazard ratio [HR] 0.10 95% confidence interval [CI] 0.03–0.34; mTR HR 0.17 95% CI 0.06–0.48; pNR HR 0.17 95% CI 0.06–0.54; mNR HR 0.13 95% CI 0.02–0.66). Conclusions Metabolic tumour and nodal response predicted pTR and pNR, respectively, in patients with oesophageal or GOJ adenocarcinoma. However, currently utilised response criteria may not be optimal. pTR, mTR, pNR and mNR were independent predictors of survival. Key Points • FDG PET-CT has an emerging role in evaluating response to neoadjuvant therapy in patients with oesophageal cancer. • Prospective cohort study demonstrated that metabolic response in the primary tumour and lymph nodes was predictive of pathological response in a cohort of patients with adenocarcinoma of the oesophagus or oesophago-gastric junction treated with neoadjuvant chemotherapy followed by surgical resection. • Patients who demonstrated a response to neoadjuvant chemotherapy in the primary tumour or lymph nodes on FDG PET-CT demonstrated better survival and reduced rates of tumour recurrence. Supplementary Information The online version contains supplementary material available at 10.1007/s00330-023-09482-7.


Introduction
2-Deoxy-2[ 18 F]fluoro-d-glucose positron emission tomography-computed tomography (FDG PET-CT) is routinely used for staging of oesophageal cancer and has an emerging role in evaluating response to neoadjuvant therapy [1,2]. A reduction in FDG avidity of the primary tumour and locoregional lymph nodes (LN) following neoadjuvant chemotherapy has been shown to predict pathological response and survival [1][2][3][4]. However, no studies have evaluated the prognostic significance of metabolic parameters derived on FDG PET-CT with respect to pathological response in LNs, but have instead used primary tumour response or pathological nodal stage as a surrogate [5]. This is despite emerging evidence of a discrepancy, in up to 25% of patients, between the response in the primary tumour compared with that in LNs [1,6,7]. One hypothesis is that this discrepancy may be due to variations in chemo-sensitivity of different clonal populations of tumour cells [6,8].
Change in primary tumour maximum standardised uptake value (SUV max ) is the most widely used FDG PET-CT parameter to define metabolic response to treatment with various thresholds described. The most commonly utilised are a 35% reduction in SUV max as adopted by the MUNI-CON trial [9] and PET Response Criteria in Solid Tumours (PERCIST) which recommends a 30% reduction to define response [10]. However, there is no consensus regarding the optimal classification; the MUNICON threshold was derived from only 40 patients [11] and PERCIST is not tumour specific.
The primary aim of this study was to evaluate the ability of FDG PET-CT to predict pathological response in the primary tumour and LNs for patients with adenocarcinoma of the oesophagus or oesophago-gastric junction undergoing neoadjuvant chemotherapy prior to surgical resection. Secondary aims were to assess the prognostic effect of metabolic and pathological response on survival and to evaluate the predictive ability of currently utilised response classifications.

Design
This was a cohort study based on an ethically approved, prospectively maintained database of consecutive oesophagogastric resections performed at Guy's and St Thomas' NHS Foundation Trust, London, UK. The present study included patients treated with neoadjuvant chemotherapy who underwent oesophago-gastrectomy with curative intent for histologically confirmed adenocarcinoma of the oesophagus or oesophagogastric junction between January 2017 and December 2020. All patients had locally advanced tumours, i.e. cT2 or higher, positive LNs or both. Tumours were staged using the American Joint Committee on Cancer TNM version 8 [12]. Tumours of the oesophago-gastric junction were categorised using the Siewert classification [13]. The primary outcome measure was tumour regression grade in the primary tumour and LNs. Secondary outcomes were overall and disease-free survival.

FDG PET-CT protocol
Patients were assessed with FDG PET-CT before and after completion of neoadjuvant chemotherapy, prior to surgery. The majority of patients had FDG PET-CT performed on one of two GE Discovery 710 PET-CT systems (GE Healthcare) at the host institution. Patients were fasted for a minimum of 6 h. A standard PET acquisition minimum coverage from skull base to upper thighs was acquired post-injection of a 350-MBq + / − 10% and an uptake time of 60 min at 3 min per bed position. The standard clinical attenuation-corrected PET reconstruction was a time-of-flight ordered subset expected maximisation reconstruction using 2 iterations, 24 subsets and a 6.4-mm Gaussian filter (VPFX, GE Healthcare) and thickness of 3.27 mm. The low-dose unenhanced CT scan was performed in shallow respiration, using a standardised protocol: 140 kV, pitch 1.375, 40 mm beam collimation (64 × 0.625 mm), 0.5-s rotation time and auto mA (15-100 mA, noise index 40). CT images were reconstructed with a thickness of 2.5 mm. A minority of patients had a combination of FDG PET-CT examinations performed on different scanners at the same institution, or different scanners at different institutions. Image analysis was undertaken using Hybrid Viewer (Hermes Medical Solutions) by a board-certified radiologist (M.S.) with 10 years of PET-CT reporting and who provides specialist input to the upper gastro-intestinal multi-disciplinary team meeting.

Tumour measurements
A freehand region of interest (ROI) was used to calculate the SUV max in the primary tumour. For oesophageal tumours, metabolic tumour length was calculated by multiplying the number of axial images the primary tumour was visible by the section thickness of the PET images. For junctional tumours with a non-vertical orientation, a freehand measurement was made in the most appropriate orthogonal imaging plane. Axial fused PET-CT imaging, with its ability to demonstrate residual mural thickening, was used to assist measurement of SUV max in the primary tumour on post-treatment imaging when tumoural uptake was visually indistinguishable from background physiological oesophago-gastric uptake.

Metabolic tumour response (mTR)
Receiver operating characteristic (ROC) curve analysis was used to determine optimum SUV max and metabolic tumour length thresholds for predicting pTR in the study cohort. Metabolic tumour response (mTR) was defined using PER-CIST (ΔSUV max 30%) and MUNICON (ΔSUV max 35%) and based on results of ROC curve analysis. Patients with a complete metabolic response (CMR), i.e. tumour indistinguishable from background physiological oesophago-gastric uptake, or partial metabolic response (PMR), i.e. reduction in tumour SUV max ≥ pre-defined thresholds detailed above, were classified as responders. Patients with stable metabolic disease (SMD), i.e. reduction or increase in tumour SUV max < predefined thresholds, or progressive metabolic disease (PMD), i.e. increase in tumour SUV max ≥ predefined thresholds, were classified as non-responders.

Nodal measurements
Metabolic nodal (mN) stage was defined as nodes discrete from the primary tumour within a standard lymphadenectomy territory visible above background soft tissue uptake and categorised as mN0 (0 avid nodes), mN1 (1-2 avid nodes), mN2 (3-6 avid nodes) or mN3 (> 6 avid nodes). A freehand ROI was drawn around the most avid node to determine nodal SUV max . Post-treatment nodal SUV max measurement was not possible if nodal uptake was visually indiscernible from background soft tissue uptake and the size of the nodal residuum on axial fused PET-CT imaging was small thereby precluding reliable assessment.

Metabolic nodal response
Metabolic nodal response (mNR) was defined as CMR (no discernible uptake above background soft tissue uptake), PMR (reduction in mN category or SUV max ≥ 30%), SMD (stable mN category or reduction/progression in SUV max < 30%) or PMD (progression of mN category or SUV max ≥ 30%). ROC curve analysis was used to identify optimum SUV max nodal threshold for predicting pNR in the study cohort.

Pathological tumour and nodal response (pTR and pNR)
Pathological response was evaluated using the Mandard classification, a 5-point categorical scoring system based on the proportion of fibrosis and viable tumour cells in the surgical resection specimen [14]. Mandard scores were calculated for response in the primary tumour (pTR) and resected LN (pNR) as described previously [6]. Patients with Mandard scores of 1-3 were classified as responders and those with scores of 4-5 as non-responders.

Treatment
All patients were on a curative-intent treatment pathway and received neoadjuvant chemotherapy followed by surgical resection. Patients who received extended neoadjuvant therapy as part of an advanced disease protocol were excluded. Transthoracic or transhiatal oesophagectomy or extended total gastrectomy was performed by either open or minimally invasive approach.

Statistical analysis
Clinicopathological characteristics were compared using the chi-squared test. Overall survival was defined as time from surgery to death or date of last outpatient department visit. Disease-free survival was defined as time from surgery to cancer recurrence, death or date of last outpatient department visit. Continuous variables (SUV max and metabolic tumour length) were evaluated using ROC analysis to identify the threshold which provided the greatest area under the curve (AUC) for predicting pathological response in the study cohort. An AUC of 1.0 indicated perfect predictive ability and an AUC of 0.5, no ability. Crude and adjusted logistic regression analyses were performed adjusting for age (continuous), sex (male or female), chemotherapy regimen (epirubicin, cisplatin and capecitabine (ECX) or fluorouracil, leucovorin, oxaliplatin and docetaxel (FLOT)) and presence of signet ring cells (yes or no). Survival curves were created using the Kaplan-Meier method, with subgroups compared using the log-rank test. Crude and adjusted survival analyses were performed using Cox proportional hazards regression adjusted for age (continuous), sex (male or female), chemotherapy regimen (ECX or FLOT), cT stage (cT1-2 or cT3-4), cN stage (cN0 or cN +), tumour grade (well/moderately differentiated or poorly differentiated) and presence of signet ring cells (yes or no). Confounders for logistic regression and Cox proportional hazards analyses were defined based on directed acyclic graphs [15]. Harrell's C-index was calculated to compare the accuracy of different metabolic response parameters to predict survival. p values < 0.05 were considered to be statistically significant. Statistical analysis was performed using IBM SPSS statistics (IBM Corp, IBM SPSS statistics Version 27.0.) and GraphPad Prism 9 (GraphPad Software Inc.).

Patient and treatment characteristics
There were 76 eligible patients identified from the institutional database. One patient had a non-avid tumour on baseline FDG PET-CT and was excluded, leaving 75 patients for final analysis. Mean age was 63 years (range 26-79) with the majority male (65/75, 86.7%) ( Table 1). Almost two thirds received FLOT (48/75, 64.0%) with the remainder receiving ECX chemotherapy. Most patients (71/75, 94.7%) completed all prescribed cycles of neoadjuvant chemotherapy with a minority completing 2 cycles of ECX (n = 2) or 2 cycles of FLOT (n = 2). There was discordance between pathological response in the primary tumour and LN in several patients, including a significant proportion (6/26, 23.1%) who demonstrated a LN response in the absence of a response in the primary tumour. Median SUV max pre-and post-chemotherapy, respectively, was 11.5 (range 3.8-47.6)  (Supplementary table 2, Supplementary table 3) did not demonstrate any major differences to clinicopathological, response or survival results, so this cohort was included in the final analysis.
Logistic regression analysis (     (Table 4) demonstrated that percentage change of nodal SUV max was more accurate than mN stage for predicting pNR or pathological node negativity (pN −) (SUV max concordant n = 41, 56.9%; mN stage concordant n = 33, 45.8%). The combined category of mNR was equally as accurate as percentage change in nodal SUV max alone. ROC analysis (Fig. 1) excluding patients with metabolically negative nodes (mN −) at baseline (n = 38) demonstrated an optimum nodal SUV max decrease of 32.6% for predicting pNR or pathological node negativity. Using an SUV max threshold of 30% to define mNR (Table 2), this demonstrated high sensitivity and PPV but relatively low specificity and NPV for predicting pNR or pN − (AUC 0.749, sensitivity 92.6%, specificity 57.1%, PPV 86.2%, NPV 66.7%, p = 0.010). There was significantly higher concordance when excluding mN − patients compared with the entire cohort (85.3% vs 56.9%). Three patients had missing pNR data so were excluded from nodal analysis.

Prognostic effect of metabolic response on survival
Metabolic responders in the primary tumour demonstrated better overall and disease-free survival compared with non-responders using all SUV max thresholds of mTR. However, when survival models were compared using Harrell's C-index (Table 5), SUV 50% demonstrated the best predictive ability for overall and disease-free survival in both crude and adjusted analyses compared with PERCIST and MUNICON criteria. ROC analysis for metabolic and pathological tumour characteristics (Fig. 3) demonstrated pTR to be the best predictor of recurrence and survival, and SUV 50% was the second best predictor. Metabolic nodal responders demonstrated improved survival compared with non-responders on adjusted analysis (responder overall survival HR 0.13, 95% CI 0.02-0.66; disease-free survival HR 0.19, 95% CI 0.04-0.90; node negative overall survival HR 0.30, 95% CI 0.08-1.21; diseasefree survival 0.32, 95% CI 0.08-1.32). No other confounders in the adjusted model were prognostic for survival.

Discussion
In this study, metabolic response on FDG PET-CT after neoadjuvant chemotherapy was predictive of pathological response in both the primary tumour and LN metastases of patients with adenocarcinomas of the oesophagus or oesophago-gastric junction. Patients who had a favourable Fig. 2 Kaplan-Meier survival analysis for overall and diseasefree survival by pathological tumour response (pTR) (a + b), pathological nodal response (pNR) (c + d), metabolic tumour response (mTR) (e + f) and metabolic nodal response (mNR) (g + h) in 75 patients with adenocarcinoma of the oesophagus or oesophagogastric junction treated with neoadjuvant chemotherapy followed by surgical resection a b c d e f g h metabolic or pathological response in the primary tumour or LNs had improved survival compared to non-responders. Some methodological issues deserve attention. This study allowed for the long-term follow-up of a consecutive cohort of patients treated for oesophago-gastric adenocarcinoma and compared FDG PET-CT response with pathological LN response, making these results novel. Although most data were collected prospectively, the observational design made it difficult to exclude confounding. However, the results were adjusted for the key prognostic factors to counteract this. As a single-centre study, one advantage was that there was consistency of pathological and radiological reporting. This reduced heterogeneity; however, the singlecentre design reduced the generalisability of the findings compared to a population-based approach. Although most patients had FDG PET-CT performed on the same scanner, the inclusion of 14 patients who did not may have introduced bias due to potential differences in SUV max measurements and calibration between scanners. However, stratified analysis revealed no obvious differences in clinicopathological, response or survival results between these patients.
It has been suggested that commonly utilised thresholds of SUV max are not optimal for assessing response to neoadjuvant therapy in oesophageal cancer, with higher cut-offs proposed [5,16,17]. This is supported by the results of the present study which suggest a reduction in SUV max of 50% in the primary tumour was a better predictor of not only pathological response but also tumour recurrence and survival compared with PERCIST (ΔSUV max 30%) and MUNICON (ΔSUV max 35%). Although increasing the SUV max threshold to 50% slightly decreased the sensitivity for predicting pathological response compared with PERCIST and MUNICON, it significantly increased the specificity and, overall, correctly identified pathological response in more patients. In this study, change in SUV max was more accurate than change in metabolic Table 5 Hazard ratios (HR) with 95% confidence intervals (CI) of overall and disease-free survival for metabolic tumour response (mTR), pathological tumour response (pTR), metabolic nodal response (mNR) and pathological nodal response (pNR) in 75 patients with adenocarcinoma of the oesophagus or oesophago-gastric junction treated with neoadjuvant chemotherapy followed by surgical resection. The accuracy of each mTR classification to predict survival was compared using Harrell's C-index a Adjusted for: age, sex, chemotherapy regimen (FLOT/ECX), cT stage (cT0-2, cT3-4), cN stage (cN0, cN +), differentiation (well-mod, poor), signet ring cell (yes, no) b Stable/progressive metabolic disease c Complete/partial metabolic response d Mandard  tumour length for predicting pathological response and survival. However, advances in FDG PET-CT allowing calculation of three-dimensional spatial metrics such as metabolic tumour volume and total lesion glycolysis may provide even more reliable prognostic information [5]. Although such metrics were not calculated for the present study due to technical limitations, these results are still relevant since change in SUV max remains the most reproducible and widely utilised metabolic parameter for response assessment. The prognostic ability of FDG PET-CT to identify pathologic nodal disease was relatively poor. Approximately half of patients with metabolic negative nodes on baseline and restaging FDG PET-CT had pathological positive nodes, approximately a quarter of the patients in the entire cohort. This is not unexpected given that false-negative observations are well recognised, particularly with respect to small LNs due to a combination of limited tumour burden (insufficient to generate a detectable PET signal) and size below the spatial resolution of PET. When patients with metabolic negative nodes were excluded from response analysis to counteract the high false-negative rate, concordance, sensitivity and positive predictive value of mNR for predicting pNR or pathological node negativity were high, although specificity and negative predictive value remained low. Interestingly, there was also a false-positive rate of just over 10%, i.e. patients with metabolically positive but pathologically negative nodes, a finding which might be explained by post-treatment inflammation, also a well-recognised pitfall in FDG PET-CT interpretation.
The results of the present study support the findings of several previous studies demonstrating that a decrease in metabolic activity in the primary tumour following chemotherapy predicts pathological response [5,18] and survival [3,9,19]. However, to our knowledge, only two studies from the same institution have evaluated the association between mNR and these outcomes [3,5], with similar results obtained. Furthermore, no previous studies have evaluated the ability of FDG PET-CT to predict pathological nodal response. This is despite recent findings Fig. 3 Receiver operating characteristic (ROC) analysis evaluating the prognostic effect of metabolic tumour response (mTR) parameters and pathological tumour response (pTR) on survival (a) and recurrence (b) in 75 patients with adenocarcinoma of the oesophagus or oesophago-gastric junction treated with neoadjuvant chemotherapy followed by surgical resection a b suggesting it is an independent predictor of survival and evidence of a discrepancy, 23% in this study, between patients who demonstrate a nodal response in the absence of a response in the primary tumour [6,7]. The use of FDG PET-CT in assessing response to neoadjuvant chemotherapy in patients with oesophageal or junctional adenocarcinoma remains an under-researched area, and a large-scale study is needed to establish optimum response thresholds and its prognostic effect on pathological response and survival. Although this analysis focused on FDG PET-CT response after completion of neoadjuvant chemotherapy, at which point treatment alteration would not generally be considered, recent studies have evaluated early response assessment after induction chemotherapy in patients undergoing chemoradiation, with promising results [20]. Further research is warranted into whether FDG PET-CT performed earlier in the treatment pathway could be used to tailor individual treatment strategies in patients treated with neoadjuvant chemotherapy.
In conclusion, this study has demonstrated that metabolic response on FDG PET-CT was predictive of pathological response in both the primary tumour and LNs in a cohort of patients with locally advanced adenocarcinoma of the oesophagus or oesophago-gastric junction treated with neoadjuvant chemotherapy followed by surgical resection. However, a relatively high false-negative rate makes metabolic evaluation of pathological nodal response challenging. Metabolic tumour response, pathological tumour response, metabolic nodal response and pathological nodal response were all independent prognostic factors for survival.