The neoadjuvant rectal score and a novel magnetic resonance imaging based neoadjuvant rectal score are stage independent predictors of long‐term outcome in locally advanced rectal cancer

Neoadjuvant rectal (NAR) score is an early surrogate for longer‐term outcomes in rectal cancer undergoing radiotherapy and resection. In an era of increasing organ preservation, resection specimens are not always available to calculate the NAR score. Post‐treatment magnetic resonance imaging (MRI) re‐staging of regression is subjective, limiting reproducibility. We explored the potential for a novel MRI‐based NAR score (mrNAR) adapted from the NAR formula.


INTRODUC TI ON
The incidence of rectal cancer reflects around 15-25 cases/100 000 population per year in Europe [1].Early rectal cancers can be managed by resection alone.A significant proportion present locally advanced (LARC), for which a standard management is neoadjuvant radiotherapy ± chemotherapy (nCRT) followed by total mesorectal excision (TME) [1].
The extent of radiotherapy response is an important prognosticator in LARC [2].There is a spectrum of response, however, varying considerably in patients of the same disease stage.Approximately 14.9% of patients can achieve pathological complete response (pCR) after nCRT, with no specimen evidence of tumour or positive nodes [3].Three-year disease-free survival (DFS) is 92.3%, 77.6% and 64.6% for complete, intermediate and poor regression (Dworak classification) respectively [4].
LARC trials historically rely on local recurrence (LR), overall survival (OS) and DFS outcomes.With clear links between nCRT response and long-term oncological outcomes, interest has grown to develop rapid and easily attainable surrogate end-points to increase trial efficiency and pace.They could also guide follow-up.
Magnetic resonance imaging (MRI) remains the standard for LARC local staging and post-nCRT re-staging [1,5].TNM downstaging alone, however, is not a consistent indicator of outcomes.
Although pCR (epitome of good response) has been used as a study end-point, a meta-analysis of 22 trials confirmed that pCR lacks validity as a 5-year OS surrogate [6].Pathological (pTRG) and MRI (mrTRG) tumour regression grading have shown promise as they associate with OS and DFS [2,[7][8][9].TRG assessments remain subjective, however, with limited inter-observer agreement, and thus reproducibility of prognostic information remains unclear [1,7,[10][11][12].mrTRG has therefore not been adopted as a preoperative surrogate end-point [13].
The neoadjuvant rectal (NAR) score was proposed by George et al. as a short-term trial end-point [14].It will henceforth be referred to as pathological NAR (pNAR).It incorporates weighted clinical (c) T stage and post-nCRT pathological (yp) T and N stages from resection specimens.It is a pseudo-continuous variable with 24 possible outcomes [14].pNAR is calculated as demonstrated in Figure 1A and is adapted from Valentini et al. nomogram data on rectal cancer recurrence and survival [15].It aims to serve as a DFS and OS surrogate with higher scores representing a poorer prognosis, for example 0 (ypCR from cT4) to 100 (progression from cT1 to ypT4N2) [14,16].The pNAR score has been validated in retrospective studies [17][18][19][20][21][22], phase II [16] and phase III trials [10]; however, some report no additional value as an OS end-point compared with ypTNM [23,24].
Pathologically derived surrogates have several issues in modern rectal cancer management.They are irrelevant with respect to organ preservation and cannot help inform preoperative conversations.
Preoperative surrogates may allow opportunity for nCRT optimization or provide a window of consideration for salvage preoperative management.They could also prove invaluable during informed consent for higher risk patients considering surgery associated with greater morbidity.
A specific description of MRI TN assessment after nCRT is a more accurate, less subjective assessment of re-staging.We hypothesize that an MRI NAR score (mrNAR), using the same parameters as the validated pNAR score, could satisfy the need for an early, preoperative surrogate of longer-term outcomes.
In this study we aimed to validate the prognostic value of George et al.'s [14] pNAR score within our cohort and explore the prognostic accuracy of a novel mrNAR score for future use as a preoperative marker of outcomes.

MRI assessment
A standardized MRI protocol was performed on a variety of 1.5 T MRI scanners.The protocol includes large field of view sagittal and axial images with 4-6 mm slice thickness, small field of view true axial and coronal tumour images with slice thickness of 3-3.5 mm, and axial diffusion-weighted sequences performed with b value ranges of 0-50, 300-800 and 800-1000 (s/mm 2 ).Post-nCRT scans were performed on the same scanner as initial staging for any given patient.No routine rectal cleansing or insufflation was performed.MRI staging was performed pre-nCRT in all patients and post-nCRT from 2016 as outlined in Results.Pre-and post-nCRT T-staging, N-staging and circumferential resection margin status was assessed using the MERCURY mrTRG proforma from 2016 to 2019 [9,25] and the ESGAR proforma from 2019 [26].As per proforma, pre-nCRT nodal assessment was based on size and morphological criteria (shape, signal intensity heterogeneity and margin regularity).For post-nCRT, either no remaining nodes or nodes <5 mm were considered N0 and the presence of any nodes with a short axis diameter ≥5 mm was considered N+.
All scans were performed within a single NHS Scotland Health Board, NHS Greater Glasgow and Clyde.The mean time between the end of CRT and re-staging MRI was 8.84 weeks.All MRI scans were reported by a consultant radiologist and re-reviewed and presented at the colorectal MDT by subspecialist gastrointestinal radiologists.Information from the final MDT reports were accessed retrospectively and categorized using the TNM classification (8th edition) [1].

Pathological evaluation
Pathology reports followed the Royal College of Pathologists (RCP) TNM classification guidelines [27].Reports were accessed retrospectively having been released at the time of resection.'Vascular invasion' was deemed present if there was evidence of intramural, extramural or both venous invasions.Specimens were considered to be margin positive/R1 if there was tumour encroachment (direct involvement or nodal) ≤1 mm from non-peritonealized 'circumferential' or longitudinal margins.pCR was defined in line with RCP guidelines as the absence of viable tumour locally (ypT0) and in lymph nodes (ypN0) [27,28].The reporting pathologist's impression of tumour regression was reviewed retrospectively.Tumour regression grading followed the recommended four-tier system by the American Joint Committee on Cancer/RCP guidelines.This is based on a modification of TRG described by Ryan et al. as follows: TRG0, no viable cancer cells (complete response); TRG1, single or rare small groups of cancer cells (near-complete response); TRG2, residual cancer with evident tumour regression but more than single or rare small groups of cancer cells (partial response); TRG3, extensive residual cancer with no evident tumour regression (poor or no response) [27,29].

Recurrence/survival
The following definitions were used: overall survival (OS), nCRT start date to any death; recurrence-free survival (RFS), nCRT start date to any recurrence; local recurrence (LR) refers to pelvic or intra-luminal recurrence only (isolated) or to both isolated local and distant recurrence together (occurring simultaneously or at different time points before death or end of follow-up); distant recurrence (DR) was defined as any recurrence outside the pelvis alone.

Clinicopathological characteristics
Between

Survival outcomes
From the nCRT start date, median follow-up time was 52 months for the whole cohort and 38 months for the 177 patients who also had   We analysed the prognostic significance of pNAR for survival in our cohort.pNAR was a marker for OS (HR 2.05, 95% CI 1.33-3.14,p = 0.001) on univariate analysis (Table 2).Despite lower event numbers, mrNAR was also significant for OS (HR 2.96, 95% CI 1.38-6.34,p = 0.005) (Table 2).Survival analysis revealed that the pNAR risk categories significantly stratified OS (p < 0.001) up to 60 months (Figure 2A).mrNAR categories also stratified for OS (p = 0.017) (Figure 2B).
Multivariate analysis was performed to compare ypTNM, R1 and pNAR score as pathological surrogate markers of OS (

Recurrence outcomes
Table 2 summarizes the relationship of clinicopathological factors with recurrence.Three-year recurrence rates were 19.6%.Median time to any recurrence was 18 months.

pNAR versus mrNAR
Explorative analysis of the 177 patients with pNAR and mrNAR scores was performed.As a pseudo-continuous variable, pNAR and mrNAR scores had a low correlation coefficient (r = 0.22, p = 0.002).

DISCUSS ION
In this study we sought to validate George et al.'s NAR score (pNAR) and explore the prognostic utility of a novel, non-subjective mrNAR score for use preoperatively.

pNAR
For the first time in a UK cohort, our study contributes to and supports the view of the current literature that pNAR associates with survival outcomes [18,22].We confirmed pNAR risk categories consistently predicted survival and recurrence.This was apparent through regression analysis and stratification by pNAR risk category on survival analysis up to 60 months.
On multivariate analysis, pNAR achieved independent status for OS at 3 years above ypTNM and pTRG.For recurrence, ypTNM, R1 and vascular invasion were independent risk factors over pNAR.Our findings dispute the findings of van der Valk et al. who determined their combined model of cT, ypT and ypN variables a superior predictor over pNAR for OS [23].Imam et al. also concluded that the pNAR score gave no more prognostic information than ypT and ypN combined, but still believed pNAR to be of clinical value [18].We consider pNAR to be a useful tool, with readily available parameters, that can be expanded to 1 of 24 pseudocontinuous scores to provide extra prognostic information.Despite known limitations of post-nCRT MRI, particularly nodal re-staging [5,[33][34][35], we demonstrated that mrNAR was associated with ypTNM (p = 0.001) despite its formula not containing pathological parameters (Table 1).Dichotomous χ 2 analysis supported previous reports that yMRI was more accurate for yT, particularly yT3-4, than yN re-staging [35].In our cohort, MR was more sensitive (48.4%) for detecting pathological N+ disease compared to 42% sensitivity in a study of 2062 Swedish colorectal cancer patients [35].The poor ability of mrNAR to match the corresponding pNAR risk category could be explained by the cumulative inaccuracies that develop when combining T and N within the formula.This is supported as <8 risk category groups were least compatible (15% sensitivity, PPV 0.23) of all, highlighting MRI limitations to assess lower T stages and negative nodes.mrNAR, however, did demonstrate good agreement of outcome with pNAR on Bland-Altman analysis (Figure S1), supporting its witnessed utility as a prognosticator.

K
E Y W O R D S cCR, clinical complete response, LARC, MRI, mrNAR score, NAR score, neoadjuvant, organ preservation, pathological complete response, pCR, rectal cancer, surrogate end-point, watch and wait What does this paper add to the literature?We present an imaging-based neoadjuvant rectal (NAR) score (mrNAR) of response to therapy.Extent of response is an important prognosticator and this novel mrNAR score is a stage-independent predictor of recurrence-free and overall survival.It does not rely on resection specimens so has use in an era of organ preservation.F I G U R E 1 (A) Equation for George et al. neoadjuvant rectal score (NAR) (referred to as pNAR within this work).Clinical tumour staging was based on the initial diagnostic MRI T stage.(B) Equation for our novel proposed mrNAR score; cT, clinical tumour stage; ymrN, post-nCRT MRI nodal N stage; ymrT, post-nCRT MRI tumour T stage.between nCRT start dates 13 February 2008 to 28 December 2020.Clinicopathological characteristics, data and outcomes were collected retrospectively from electronic records by a single investigator (RKM).Patients were included if they underwent nCRT and proceeded to curative intent resection within a single National Health Service (NHS) health board.Selection for nCRT consideration was made following regional multidisciplinary team (MDT) meetings and regimens were delivered at the treating oncologist's discretion.Radiotherapy was most frequently delivered over 5 weeks (45-54 Gy in 25 fractions) usually with concomitant fluoropyrimidine-based chemotherapy regimens including oral capecitabine or intravenous 5-fluorouracil.Alternatively, a short-course schedule was delivered (25 Gy in 5 fractions) over a week, potentially followed by systemic chemotherapy.Patients were excluded if transanal excision, contact or brachytherapy was performed, if they received palliative or postoperative radiotherapy, if they did not progress to curative intent surgery for any reason, or no staging MRI performed.Distant metastases at diagnosis (TxNxM1) or preoperatively (yTxNxM1) were excluded.LARC was defined as MRI T3-T4 and/or locoregional nodes and/or circumferential resection margin threatening.Colonoscopy biopsy confirmed histological diagnosis of adenocarcinoma.MRI and computed tomography were used for local and distant staging respectively.Post-nCRT (y) MRI (mr), with documented inclusion of ymrTN re-staging, was not performed in all, as outlined in Results.TME was performed by open or laparoscopic techniques.The median time from nCRT start date to date of surgery was 18 weeks.George et al.'s pNAR was calculated as described in Figure 1A [14].Our post-nCRT mrNAR score was calculated by adjusting the pNAR calculation as follows: cT, clinical tumour T stage; ymrT, post-nCRT MRI T stage; and ymrN, post-nCRT MRI nodal stage (Figure 1B) [14].Clinical T-staging was based on the diagnostic MRI.As previously described by George et al., pNAR was categorized into three risk groups: low (NAR <8), intermediate (NAR Descriptive statistics were used for baseline clinicopathological characteristics.χ 2 was used to assess the associations between both pNAR TA B L E 1 Association with clinicopathological characteristics and pathological NAR (n = 381, median FU 52 months) and MRI NAR (n = 177, median FU 38 months).
and mrNAR with clinicopathological characteristics.A p value ≤0.05 was considered significant.Cox regression analysis was used to determine relationships at 3 years between OS, RFS and clinicopathological characteristics including pNAR and mrNAR.Postoperative pathological factors with p value ≤0.01 on univariate analysis were included in the multivariate model.Kaplan-Meier and log-rank tests were used to assess survival probability difference between pNAR, mrNAR categories and survival outcomes.Spearman's rank correlation coefficient assessed the relationship between pseudo-continuous mrNAR and pNAR scores, and the magnitude of the percentage error was calculated as [(mrNAR -pNAR)/pNAR] × 100.Dichotomous χ 2 analysis for trend and positive predictive values (PPVs) were used to describe the association of mrNAR to pNAR risk categories as well as ymrTNM to ypTNM.A Bland-Altman plot was used to analyse agreement between pNAR and mrNAR scores.Statistical analysis was performed using SPSS software (version 26.0; SPSS Inc.).

2
Kaplan-Meier plots for survival; statistical significance using log-rank test.(A) Survival analysis and prognostic significance of pNAR risk category score for OS up to 60 months with corresponding lifetables (n = 381).(B) Survival analysis and prognostic significance of mrNAR risk category score for OS up to 60 months with corresponding lifetables (n = 177).survival at 60 months (p = < 0.001)

TA B L E 3
Multivariate analysis of postoperative pathological factors for 3-year OS and RFS.

Further study is required
to truly understand pNAR use within an era of personalized treatment; for example, should patients with NAR >16 be preferentially considered for adjuvant chemotherapy?We did not deem the wide 2008-2020 interval range a limitation in our study of pNAR.Simple T-staging (TNM I-IV) and vascular invasion status has not differed significantly in MRI or pathological reporting in this period, nor did the distribution of pNAR scores across years in our cohort.mrNAR Habr-Gama et al. first described similar long-term survival rates when comparing pCR to those in the so-called watch-and-wait TA B L E 4 Multivariate analysis of postoperative pathological factors and mrNAR for 3-year OS and RFS.

F I G U R E 3
Kaplan-Meier plots for recurrence; statistical significance using log-rank test.(A) Recurrence analysis of pNAR risk categories for RFS, DR, LR up to 60 months with corresponding lifetables (n = 381).(B) Recurrence analysis of mrNAR risk categories for RFS, DR, LR up to 60 months with corresponding lifetables (n = 177).

Table 1 .
Most patients 2008 and 2020, following exclusions, 381 patients were available for analysis with pNAR scores.Post-nCRT MRI with TNM re-staging since 2016 was available in 177/381 (46.5%) with mrNAR scores calculated.Baseline demographics are summarized in

Multivariate Cox regression N (%) Overall survival at 3 years (OS) Any recurrence at 3 years (RFS) Events n = 48 % (SE) HR (95% CI) p Events n = 75 % (SE) HR (95% CI) p
mrNAR has shown early promise as a predictor of recurrence.Risk categories were independently associated with RFS on multivariate analysis.Similar to OS Kaplan-Meier plots, >16 appeared to separate from the other two closely linked categories which were 91% and 95% for intermediate and low categories respectively.We hypothesize separation of these categories with increased cohort size.