Magnetic resonance imaging accuracy in assessing tumour down-staging following chemoradiation in rectal cancer

Authors


A. Suppiah, Academic Surgical Unit, Castle Hill Hospital, Castle Road, Cottingham, East Yorkshire HU16 5JQ, UK.
E-mail: aravindsuppiah@hotmail.com

Abstract

Objective  Magnetic resonance imaging (MRI) is increasingly accepted as the radiological modality of choice staging rectal cancer but is subject to error. Neoadjuvant therapy is increasingly used in rectal cancer and MRI is used to stage response and occasionally plan surgery. We aim to assess the staging accuracy of MRI following chemoradiotherapy in rectal cancer.

Method  Retrospective analysis of 86 patients with MRI stage pre- and postlong-course chemoradiotherapy and comparison with pathological assessment.

Results  Fourty-nine patients (34 men, 15 women) with median age 68 years (60–74) were analysed. The median time from completion of CRT to MRI was 32 days (16–37). Chemoradiotherapy led to significant down-staging (< 0.001). MRI-staging accuracy was 43% (21/49) with over- and under-staging in 43% (21/49) and 14% (7/49) respectively. T-stage accuracy was 45% (22/49) with over-staging in 33% (16/49) and under-staging in 22% (11/49). MRI stage correlated poorly with pathological assessment for International Union Against Cancer (κ = 0.255) and T stages (κ = 0.112). MRI nodal assessment was 71% (35/49) accurate, with 82% (9/11) sensitivity, 68% (26/38) specificity and positive predictive value (PPV) of 43% (9/21) and negative predictive value of 93% (26/28). There was a significant difference in node positivity between MRI and pathological staging (= 0.005, Fisher’s exact). Complete radiological response was observed in 4% (2/49). Complete pathological response was observed in 10% (5/49), which were staged 0(1), I(1), II(2) and III(1) postchemoradiotherapy by MRI.

Conclusion  MRI staging following chemoradiation is poor. Over-staging occurs three times more commonly than under-staging. Over-staging is due to poor PPV of nodal assessment.

Introduction

Preoperative chemoradiation (CRT) is used in advanced local and/or nodal disease where it can lead to down-staging or complete response in up to 20% [1–4]. Magnetic resonance imaging (MRI) is now considered by many as the imaging modality of choice in staging rectal cancer [5]. There are few data on MRI accuracy following CRT due to limited MRI availability. However, larger studies are required as preoperative chemoradiotherapy and MRI utility is increasing. The aim of this study is to investigate MRI accuracy following CRT in one of the larger single-centre studies.

Method

Data were collected on all rectal cancer patients with pre- and postchemoradiotherapy MRI and pathological stages between 2003 and 2006. MRI was performed at 1.5 T with a dedicated surface coil on GE SIGNA, Philips Intera and Philips Achieva scanners (Philips Medical Systems, Andover, the Netherlands). The scanned sequences performed were sag T2 (4–5 mm slice thickness), axial T1 and axial fat sat T2 (4–7 mm slice thickness), high resolution (small field of view) axial T2 (3 mm, no interslice gap, scanned axial to tumour containing rectum). MRI scans were reviewed by at least two dedicated colorectal radiologists at the multidisciplinary team (MDT) meeting. Decision regarding use of CRT was made on case-by-case basis through the MDT. In general, CRT was given for T3/T4 disease and/or N1/N2 disease and/or threatened circumferential resection margin (CRM) involvement. All patients received standardized long-course CRT of 45 Gy in 25 fractions over 5 weeks with concomitant 5-fluorouracil on days 1–5 and 29–33 or oral capecitabine. Statistical Analysis was performed using spss® v.14 (SPSS Inc., Chicago, Illinois, USA) using pathological staging as the ‘gold standard’. Tumour down-staging was analysed using sign test for nonparametric related samples. The correlation between post-CRT MRI and pathological staging was analysed using kappa (κ) measurement of agreement (UICC and T stages) and Fisher’s exact test (N stage).

Results

Data were available on 64 patients who received CRT. One patient declined post-CRT MRI and had CT instead after pre-CRT MRI and claustrophobia. Fourteen patients did not have surgical resection – two patients were deemed unresectable perioperatively (both radiologically staged T4N0M0 and deemed resectable by post-CRT MRI), two patients had loop colostomy for palliation, one patient had argon beam ablation, six patients were initially fit for surgery but deteriorated during CRT or in the interim period between CRT and surgery and later considered unfit for surgery, two died and one patient refused. This left 49 patients available for analysis of pre- and post-MRI and final pathological staging – 34 men and 15 women with median age 68 years (60–74). The median time from completion of CRT to MRI was 36 days [interquartile range (IQR): 29–49]. The median time from post-CRT MRI to surgery was 18 days (IQR: 24–24). The median time from completion of CRT to surgery was 55 days (IQR: 49–63). Chemoradiotherapy led to significant down-staging (< 0.001, sign test).

MRI accuracy in tumour stage after CRT

Postchemoradiotherapy radiological UICC, T and N stages are compared with pathological stage (Table 1–3; Fig. 1). The correlation between radiological down-staging and pathological stage is shown in Table 4. MRI-staging accuracy was 43% (21/49) with over- and under-staging in 43% (21/49) and 14% (7/49) respectively. Accuracy for T stage was 45% (22/49) with over-staging in 33% (16/49) and under-staging in 22% (11/49). MRI stage showed poor correlation with pathological stage for UICC (κ = 0.255) and T stages (κ = 0.112). MRI nodal assessment was 71% (35/49) accurate, with 82% (9/11) sensitivity, 68% (26/38) specificity and positive predictive value (PPV) of 43% (9/21) and negative predictive value (NPV) of 93% (26/28). There was significant difference in node positivity between MRI and pathology (= 0.005).

Table 1.   Comparison of postchemoradiation MRI and pathological TNM stage (κ = 0.255).
 Pathological stage
 0IIIIIIIVTotal
  1. MRI, magnetic resonance imaging.

MRI stage0100102
I134109
II2480014
III1294117
IV001157
Total59227649
Table 2.   Comparison of postchemoradiotherapy MRI vs pathological T-staging (κ = 0.112).
 Pathological T stage
 T0T1T2T3T4Total
  1. MRI, magnetic resonance imaging.

MRI stageT0100304
T1000202
T22146013
T320516023
T4001517
Total511032149
Table 3.   Post-CRT MRI vs pathological staging.
 pN (−)pN (+)Total
  1. MRIyN, Post-CRT MRI stage; pN, pathological staging (−) negative, (+) positive.

MRIyN (−)26228
MRIyN (+)12921
Total381149
Figure 1.

 TNM Stage distribution following chemoradiation (CRT) and final pathological stage.

Table 4.   Correlation between radiological down-staging and final pathological stage.
Radiological changesp0pIpIIpIIIpIVTotal
Down-stage47154232
No change1262011
Progression001146
Total59227649

Correlation between complete radiological and pathological response

Complete radiological response was observed in 4% (2/49) (Table 5). Only one of these two patients had complete response on pathological examination. This patient was radiologically staged T4N2 at diagnosis. This patient developed pulmonary metastases 6 months postoperatively. The other patient had radiological T3N0 tumour at diagnosis, which showed complete radiological down-staging following CRT but pathological assessment demonstrated residual T3N1 cancer.

Table 5.   Patients with radiological and/or pathological complete response and outcome.
Pre-CRT MRIPost-CRT MRIPathologicalOutcome
  1. CRT, chemoradiation; MRI, magnetic resonance imaging.

III (T4N2)00Alive at 9 months. Developed pulmonary metastases at 6 months
III (T2N1)I (T2N0)0Disease-free at 18 months
IV (T4N1M1)II (T3N0)0Alive at 24 months. Local recurrence at 22 months
II (T3N0)II (T3N0)0Alive at 48 months. Local recurrence at 40 months
III (T3N2)III (T2N1)0Disease-free at 30 months
II (T3N0)0III (T3N1)Disease-free at 7 months

Complete pathological response was observed in 10% (5/49) (Table 5). The complete response was only detected radiologically in one patient staged T4N2 pre-CRT (who was discussed above). Prechemoradiotherapy stage in the other patients was stages II(1), III(2) and IV(1). Postchemoradiotherapy MRI showed complete response in one patient and the remaining patients were staged as I(1), II(2) and III(1) (Table 4). Four patients developed new metastases (all pre-CRT stage III) on MRI and underwent local resection.

Discussion

This study represents one of the larger cohorts of patients with pre- and postchemoradiotherapy MRI. The stage distribution with peaks in stages II and III are typical of that seen in other CRT studies [6,7]. MRI accuracy has been reported at 70–85% in rectal cancer without chemoradiotherapy [3,8]. We were unable to measure the pretreatment-staging accuracy as all patients had chemoradiotherapy. Our postchemoradiotherapy MRI accuracy was consistent with other reports of 37%, 47% and 52% following long-course chemoradiotherapy [3,4,7]. There was poor agreement between postchemoradiotherapy radiological and pathological stages, which is less than some other studies reporting κ = 0.4 [7]. Thus, chemoradiotherapy significantly impairs MRI staging. In our study, over-staging occurred three times more frequently than under-staging. It follows that the reason for this is either overestimation of T and/or N stage.

MRI and endorectal ultrasound (ERUS) display accuracy of 71–91% and 69–97% for T stage, which is superior to 52–74% for CT [5,9]. MRI is less operator-dependant with less interobserver variation but is associated with higher cost when compared with ERUS [10]. Accuracy decreases following CRT, which is reflected in this and several other studies (47–52%) [3,4,7]. This is probably due to inability of MRI to differentiate tumour tissue from radiation-induced fibrosis and scarring. In our study, errors were slightly higher for over- (33%) than under-staging (22%). Other studies report a more marked difference in staging errors. Two studies report T stage over- vs under-staging of 38%vs 10% and 47%vs 6% [3,4]. This suggests that the main reason for over-estimating UICC stage is poor nodal rather than T-stage assessment. However, it is also worth noting that the only two patients who were found to be unresectable perioperatively in this study were erroneously deemed resectable T4N0 tumours in post-CRT MRI.

Accuracy of nodal staging varies by modality – ERUS (62–83%), CT (22–73%) and MRI (39–95%) and decreases following CRT [5]. In our study, nodal assessment was categorized into node positive/negative as the differentiation between N1 and N2 diseases was clinically insignificant. The 70% postchemoradiotherapy accuracy of nodal assessment is comparable to other reports of 60–70% [2–4]. This accuracy of nodal assessment is superior to that of UICC and T stage accuracy and also results in a high NPV in post-CRT MRI nodal assessment [3,7]. Most of the errors are over-staging (24–28%) and reflected in a poor PPV of <50% [3,4]. This is probably due to the inability of MRI to identify viable cancer cells which in turn accounts for most of the UICC-staging errors.

Significant down-staging occurred with complete pathological response in 10% of cases, which is consistent with other studies reporting up to 5–20% [1,2,11,12]. MRI was unable to identify complete pathological response as only one of the five patients with a complete pathological response was identified radiologically. In view of this, complete radiological response should not be taken as a marker of complete tumour response. Some studies report complete pathological downstage or tumour regression in association with decreased recurrence [11,12]. The sample sizes, however, are small and may represent the selection of biologically favourable tumours. Two patients had complete pathological response in our series, one with initial T4N2 disease who developed lung metastases 6 months postsurgery. This may be due to biologically aggressive tumours or the presence of micrometastases, which was radiologically undetectable. Hence, the prechemoradiotherapy stage of the tumour may be indicative of potential future metastases and should be considered, despite complete response to CRT. However, the number of patients (= 5) with complete pathological response in our study is too small to make a conclusion and contrasts with findings from larger studies of pathological response rates. Habr Gama et al. demonstrated 122/361 (34%) complete pathological response following CRT. Furthermore, 99 of the 122 patients remained disease-free at 12 months, suggesting better disease control with CRT than demonstrated in our study. Also in contrast to our smaller study, failure was slightly higher due to systemic disease (7.1%) compared with local recurrence (5%) [13].

It was not possible to ascertain if the routine MRI postchemoradiation affected surgical treatment due to the retrospective nature of this study. The extent of surgical resection (e.g. sphincter sacrifice vs preservation) was decided prior to CRT or perioperatively. This is reasonable in view of the poor radiological staging accuracy following CRT. It is possible that the presence of new metastases on post-CRT MRI may alter surgical management. However, all four patients in this study who developed new metastases underwent local resection. In patients without metastatic disease, adjuvant chemotherapy was provided based on pathological not radiological stage. Hence, the routine use of post-CRT MRI did not appear to influence surgical management in this study.

There are a number of limitations due to the retrospective nature of this study. First, there was no objective assessment of change in tumour size or distance to CRM. It is unlikely that decrease in tumour size has a significant impact on stage as other studies report a > 30% shrinkage in 63% (19/30) patients but a decrease in T stage in only 17% (5/30) [7]. Other studies report decreased tumour volume, which does not correlate with down-staging following CRT [14,15]. Overall accuracy of CRM involvement prediction is estimated at 73%, although this varies between 22–83% depending on the height and location of the tumour [16]. Re-assessment of MRI-predicted tumour size and CRM measurements in this series is presently ongoing. More recent studies from the MERCURY Study Group suggest preoperative high-resolution MRI is able to predict CRM involvement with up to 94% accuracy [17]. Secondly, all MRI staging was reviewed at MDT by at least two experienced colorectal radiologists who were not-blinded and reached staging agreement by consensus. Studies on the interobserver variation in MRI staging are also underway.

Conclusion

MRI accuracy following neoadjuvant long-course CRT is poor compared with reported accuracy in nontreated rectal cancer. Over-staging occurs three times more frequently than under-staging. Errors in MRI stage are due to poor PPV of the nodal assessment. The poor accuracy and lack of predictive ability suggest post-CRT MRI is a waste of valuable resources. The cost-effectiveness and impact of routine MRI following chemoradiotherapy on deciding further surgical management requires further assessment.

Acknowledgements

Authors would like to thank Dr Victoria Allgar, Senior Lecturer in Medical Statistics, Hull-York Medical School for aid in statistical analysis and Mrs Jenny Ward for data collection.

Ancillary