Abstract
Background: Rectal cancer patients who achieve a good response to chemoradiotherapy (CRT), may be offered less invasive surgery or even no surgery at all. Implementation of such a policy, however, requires precise patient selection. This study identifies pretreatment clinical factors that are associated with pathological complete response (pCR) and ypT0-1N0 and evaluates their performance as a selection tool for organ-preserving strategies.
Material and methods: Patients with rectal cancer treated with CRT and total mesorectal excision between January 2000 and December 2014 were retrospectively included. Following clinical characteristics were extracted from the medical files: age, gender, body mass index, ASA score, cT-stage, cN-stage, distance from the anal verge, pretreatment carcinoembryonic antigen (CEA), pretreatment hemoglobin and distance from the mesorectal fascia. Univariable and multivariable binary logistic regression models were used to predict pCR and ypT0-1N0. The discriminative ability of the prediction models was evaluated by receiver operating characteristic analysis.
Results: A total of 620 patients were included of whom 120 experienced a pCR (19%) and 170 patients achieved ypT0-1N0 response (27%). A low pretreatment CEA, a high pretreatment hemoglobin and a high cN-stage were associated with pCR in multivariable analysis. A low pretreatment CEA, a low cT-stage and a high cN-stage were associated with ypT0-1N0. After cross validation, the area under the curve for the pCR and ypT0-1N0 prediction model equaled 0.609 and 0.632, respectively.
Conclusion: Despite their statistical significance, the value of pretreatment clinical variables in the prediction of pCR and ypT0-1N0 is very limited. To safely select patients for organ preservation, other strategies need to be explored.
Preoperative chemoradiotherapy (CRT) followed by total mesorectal excision (TME) is currently the standard treatment for locally advanced rectal carcinoma [Citation1,Citation2]. The tumoral response to this preoperative treatment is highly heterogeneous. Most patients’ experience tumor downsizing, and approximately 10–30% of the patients achieve a pathological complete response (pCR) in which no tumor cells can be found in the resection specimen [Citation3]. Patients who achieve a pCR have a favorable long-term outcome with excellent local control and disease-free survival regardless of their initial T- and N-stages [Citation3,Citation4]. Retrospective studies from Brazil have highlighted the ‘watch-and-wait’ policy in patients with a clinical complete response to CRT [Citation5]. More recent series support the feasibility of this approach [Citation6]. For patients with a near complete response, local excision has shown to be an alternative [Citation7]. Adopting a non-operative or minimally invasive strategy for well responding patients avoids the risks of surgical morbidity (e.g. anastomotic leakage, re-laparatomy, wound and pelvic infection…) and might spare patients the need for a permanent stoma.
With the advent of screening programs, it is expected that a higher percentage of rectal tumors will be diagnosed at an early stage. Patients with early rectal cancer usually undergo upfront surgery. However, as they are likely to respond well to CRT, administration of CRT with the aim of organ preservation is a valuable option for patients with early rectal tumors. To solve the dilemma of treating rectal cancer either with upfront surgery or with CRT with the aim to omit invasive surgery, accurate patient selection is important. It would be practical if pretreatment clinical variables predicted which patients respond well to the preoperative CRT and might be candidates for organ preservation. The aim of this study is to investigate whether pretreatment clinical factors can be identified that are associated with pCR and ypT0-1N0 response.
Material and methods
Patients
In total 620 patients with primary histologically confirmed rectal adenocarcinoma who were treated at the University Hospitals Leuven with preoperative CRT and TME between January 2000 and December 2014 were retrospectively included. This study was approved by the ethical committee of our institution.
The initial clinical stage was based on complete blood count, liver function tests and serum carcinoembryonic antigen (CEA) level, digital rectal examination (DRE), endoscopy, endorectal ultrasound (ERUS) (n = 486), chest/abdominopelvic computed tomography (CT) (n = 600), pelvic magnetic resonance imaging (MRI) (n = 359) or 18F-FDG positron emission tomography (PET)/CT (n = 165). Nodal stage was based on CT or MRI as these imaging modalities enable evaluation of the entire mesorectum. Tumor location, defined as the distance between the anal verge and the caudal tumor border, was assessed by DRE, endoscopy or MRI.
Patient characteristics that were extracted from the medical files included gender, age, body mass index (BMI), ASA score, cT-stage and distance to the mesorectal fascia (cT-stage/MRF), cN-stage, tumor distance from the anal verge, pretreatment CEA and pretreatment hemoglobin (Hb). Histological evaluation of the resection specimen was performed according to the method described by Quirke et al. [Citation8]. Tumor response was defined as pCR and ypT0-1N0 and was based on the pathological reports.
Statistical analysis
Categorical and continuous variables were compared by Fisher’s exact and Mann-Whitney U-tests, respectively. Univariable and multivariable binary logistic regression models were applied to predict pCR and ypT0-1N0, using the variables gender, age, BMI, ASA score, cT-stage/MRF, cN-stage, distance from anal verge, pretreatment CEA and pretreatment Hb. A log transform was used for CEA to meet the linearity assumption in the logistic model (note that since base 2 is used, the odds ratio expresses the effect of doubling the CEA level). A multiple imputation approach was used to handle missing values in the predictor variables. Hundred imputed datasets were created applying the MCMC algorithm, which is based on the multivariable normal model. Following recommendations, imputation without rounding was used for the categorical predictors [Citation9]. The imputation model contained all predictor variables and both outcome variables. A multivariable prediction model was obtained by combining all predictors and by applying a backward selection procedure with 0.157 as critical level for the p-value. This value corresponds to the use of the Aikake Information Criterion (AIC) for model selection. The discriminative ability of the prediction model was evaluated by analyzing the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. To avoid that the same data were used to develop and to validate the model, the AUC was based on a leave-one out cross validation. In the multivariable models, pairwise interactions between predictors were verified. All analyses were performed using SAS software, version 9.2.
Results
Patients
Of the 620 included patients, 120 patients experienced pCR (19%) and 170 patients achieved ypT0-1N0 response (27%). Patient characteristics are depicted in . Radiotherapy dose ranged from 36 Gy to 55.8 Gy in fractions of 1.8 Gy and was delivered with a three-dimensional (3D) conventional radiotherapy technique. Concurrent chemotherapy usually consisted of a protracted continuous infusion of 5-FU (225 mg/m2/day, n = 572). Other patients received a continuous infusion of 5-FU with a weekly administration of oxaliplatin (50 mg/m2) (n = 4), capecitabine (825 mg/m2 twice daily) with (n = 14) or without (n = 11) weekly oxaliplatin (50 mg/m2), capecitabine (825 mg/m2 twice daily) with bevacizumab (5 mg/m2, every two weeks) with (n = 6) or without (n = 6) weekly oxaliplatin (50 mg/m2) or capecitabine (825 mg/m2 twice daily) with cetuximab (400 mg/m2 load dose, 250 mg/m2 weekly) (n = 7). All patients underwent TME after a mean interval of 7.3 weeks (SD: 2.2) after completion of CRT. Surgery consisted of sphincter saving operations (n = 523, 84.4%) or abdominoperineal resections (n = 97, 15.6%). Data were missing for cT-stage/MRF (15.3%), cN-stage (11.8%), Hb (7.6%), ASA (5.7%), CEA (4.0%) and BMI (1.8%), resulting in 65.3% complete cases. For none of the predictors, there was a relation between the presence of a missing value and the outcome.
Table 1. Patient characteristics according to their response to CRT (n = 620).
Prediction of pCR
shows the association between pretreatment clinical factors and pCR outcome. Univariable logistic regression demonstrated a significant relation with cN-stage, initial CEA and initial Hb (). Patients with cN =1 or 2, with lower initial CEA or with higher initial Hb, were significantly more likely to achieve pCR in univariable analysis. The multivariable model identified the same variables as independent predictors. Halving the initial CEA level increased the odds for pCR with 16.1% (1/0.861 = 1.161; p = 0.049). An initial Hb value being 1 g/dl higher increased the odds for pCR with 29.6% (p = 0.002). Patients with cN-stage 1 or 2 were more likely to achieve a pCR compared to patients with cN-stage 0, OR =3.42 (p = 0.04) and OR =4.52 (p = 0.02), respectively. In each of the 100 imputed datasets a full multivariable model was fitted and the AUC was calculated on the cross-validated probabilities. The resulting AUC equaled 0.609 (95% CI 0.548–0.669), indicating a low discriminative ability. To verify if the cross-validated AUC was not lowered due to inclusion of many noise predictors, a backward selection strategy was applied in each of the imputed datasets. The resulting mean cross-validated AUC equaled 0.596 (range 0.560–0.618). After backward model reduction, although being retained in the final model, cN was never significant.
Table 2. Univariable and multivariable logistic regression models to predict pCR.
Prediction of ypT0-1N0
The variables being significantly related to ypT0-1N0 were similar to those related to pCR (). Based on univariable logistic regression analysis, there was a significant relation with cT-stage/MRF, initial Hb and initial CEA (). Patients with a cT1-T2 tumor, with higher initial Hb or with lower initial CEA were more likely to achieve ypT0-1N0. The multivariable model identified initial CEA, cN-stage and cT-stage/MRF as independent predictors. Halving the initial CEA increased the odds for ypT0-1N0 with 27.4% (1/0.785 = 1.274; p = 0.0006). Patients with a cT1-T2 tumor and with cN =1 or 2 were more likely to achieve ypT0-1N0. However, when a backward selection strategy was applied, in none of the 100 imputed datasets cN-stage remained significant.
Table 3. Univariable and multivariable logistic regression models to predict ypT0-1N0.
The AUC based on the cross-validated probabilities from the 100 imputed datasets equaled 0.632 (95% CI 0.580–0.685), indicating again a low discriminative ability. After applying a backward selection strategy in each of the imputed datasets, a mean cross-validated AUC of 0.631 (range 0.621–0.640) resulted.
Discussion
Response to CRT is highly heterogeneous and up to 30% of the patients with locally advanced rectal cancer achieve a pCR. This percentage might be even higher in patient with early rectal tumors. This paper investigates the pretreatment clinical factors that are associated with pCR and ypT0-1N0 and evaluates their performance as a tool to select rectal cancer patients for organ-preserving treatment strategies.
Multivariable analysis demonstrated that a low pretreatment CEA, a high pretreatment Hb and a high cN-stage were associated with pCR, while a low pretreatment CEA, a low cT-stage and a high cN-stage were associated with ypT0-1N0.
A low pretreatment CEA has been correlated with pathological response to preoperative CRT before. Das et al. found that pretreatment CEA levels >2.5 ng/ml were significantly associated with lower pCR rates in univariate analysis [Citation10]. In contrast to these authors, we were not able to gather data on the percentage of the lumen that is involved by tumor, which was the only factor that was predictive for pCR in their multivariable analysis. In a study with 352 patients with locally advanced rectal cancer, Park et al. demonstrated that the rates of good response significantly decreased when the pre-CRT CEA levels increased and identified CEA as an independent prognostic factor [Citation11].
Patients with a low initial Hb experienced less response to CRT compared to patients with a high pretreatment Hb. Although the association of pretreatment Hb and response to CRT is especially established for tumors arising in the head and neck region, the cervix, the anal canal and the esophagus, data on the role of anemia in rectal cancer are emerging [Citation12,Citation13]. Anemia is present in many cancer patients at the time of diagnosis and has been hypothesized to lead to tumor hypoxia, angiogenesis, and resistance to chemotherapy and radiotherapy.
Surprisingly, multivariable analysis identified a higher cN-stage as an independent predictor for pCR and ypT0-1N0. This is in contrast to other studies which demonstrated that cN0-stage was associated with downstaging and tumor regression or with pCR [Citation12,Citation14]. Other studies found no association between cN-stage and response to CRT [Citation15]. These conflicting results can be explained by the fact that the identification of nodal disease remains a diagnostic problem for radiologists as nodal disease is frequently overstaged. Enlarged inflammatory nodes may appear as pathologically involved. As size on its own is not reliable, nodal staging is a subjective assessment based on the combination of size, shape, internal heterogeneity, border and presence of internal necrosis. A meta-analysis showed that the sensitivity and specificity of ERUS, CT and MRI for the assessment of nodal metastases in rectal cancer was 67% and 78% for ERUS, 55% and 74% for CT and 66% and 76% for MRI, respectively [Citation16]. Lymph node specific MRI contrast agents might improve the accuracy of nodal staging and restaging. Ultrasmall superparamagnetic iron oxide (USPIO) and the high molecular weight contrast agent gadofosveset have shown promising results, but are not widely available yet.
A longer interval between the end of CRT and surgery has been associated with a better response to CRT [Citation17–19]. We did not include this parameter as a pretreatment clinical factor because the interval between the end of CRT and surgery is not known at the start of CRT and we aimed to focus on pretreatment factors without addition of factors related to the treatment.
Despite the significant association between pretreatment CEA, Hb and cN-stage and pCR and between pretreatment CEA, cN-stage and cT-stage and ypT0-1N0, ROC analysis showed that the predictive performance of pretreatment clinical characteristics is too poor to safely select rectal cancer patients for organ-sparing treatment strategies. In analogy with our results, Van Stiphout et al. demonstrated that a model based on clinical factors reached an AUC of 0.61 ± 0.03 for the prediction of pCR [Citation20]. Their model consisted of a limited number of clinical factors (age, gender, cT-stage, cN-stage, tumor location and tumor length) and did not include pretreatment CEA and Hb, which are frequently reported to be associated with response to CRT. Huh et al. developed a predictive model based on circumferentiality, macroscopic ulceration and pretreatment CEA and found an AUC of 0.706 (0.635–0.776 CI) for discriminating pCR from non-pCR [Citation21].
The poor predictive value of a model based on clinical parameters alone points the need to explore other strategies to select patients for a less invasive surgical approach. Various molecular markers have been associated with response to CRT [Citation22]. However, none of them has consistently proven its value in response prediction. Functional imaging techniques depict the microstructural and metabolic characteristics of the tumor, allowing assessment of treatment-induced changes before morphological changes become apparent. In this respect, imaging and re-imaging with diffusion-weighted MRI and 18F-FDG PET have emerged as promising tools in the response prediction and assessment in rectal cancer [Citation23]. Tumor phenotypes can also be characterized by extracting a large number of quantitative features from imaging modalities, a research field known as radiomics [Citation24,Citation25]. By correlating imaging features with treatment response, radiomics might be applied as a predictive tool to assess tumor response to CRT in rectal cancer. Future research should focus on the integration of clinical factors with molecular markers and imaging acquired at different time points throughout the treatment course.
This analysis is limited by its retrospective nature and by the associated drawbacks such as missing data and heterogeneity with regard to diagnostic work-up and the CRT regimen. We tried to address these issues by imputing missing data and by limiting this analysis to patients who were treated in a single center. We did not exclude patients who were treated with oxaliplatin, bevacizumab and cetuximab as phase II and III trials demonstrated that their addition does not increase the pCR rate. Another limitation is the fact that during the 14-year study period, some diagnostic and therapeutic aspects of rectal cancer management have evolved, such as the increased use of MRI for both staging and restaging and the introduction of laparoscopic TME surgery.
In conclusion, despite their statistical significance, the value of pretreatment clinical variables in pCR and ypT0-1N0 prediction is very limited. To safely select patients for organ-preserving treatment strategies, other strategies should be explored.
Disclosure statement
The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.
References
- Sauer R, Becker H, Hohenberger W, Rödel C, Wittekind C, Fietkau R, et al. Preoperative versus postoperative chemoradiotherapy for rectal cancer. N Engl J Med 2004;351:1731–40.
- Heald RJ, Ryall RD. Recurrence and survival after total mesorectal excision for rectal cancer. Lancet 1986;1:1479–82.
- Maas M, Nelemans PJ, Valentini V, Das P, Rödel C, Kuo LJ, et al. Long-term outcome in patients with a pathological complete response after chemoradiation for rectal cancer: a pooled analysis of individual patient data. Lancet Oncol 2010;11:835–44.
- Rödel C, Martus P, Papadoupolos T, Füzesi L, Klimpfinger M, Fietkau R, et al. Prognostic significance of tumor regression after preoperative chemoradiotherapy for rectal cancer. J Clin Oncol 2005;23:8688–96.
- Habr-Gama A, Perez RO, Nadalin W, et al. Operative versus nonoperative treatment for stage 0 distal rectal cancer following chemoradiation therapy: long-term results. Ann Surg 2004;240:711–17.
- Maas M, Beets-Tan RG, Lambregts DM, Lammering G, Nelemans PJ, Engelen SM, et al. Wait-and-see policy for clinical complete responders after chemoradiation for rectal cancer. J Clin Oncol 2011;29:4633–40.
- Lezoche G, Baldarelli M, Guerrieri M, et al. A prospective randomized study with a 5-year minimum follow-up evaluation of transanal endoscopic microsurgery versus laparoscopic total mesorectal excision after neoadjuvant therapy. Surg Endosc 2008;22:352–8.
- Quirke P, Durdey P, Dixon MF, Williams NS. Local recurrence of rectal adenocarcinoma due to inadequate surgical resection. Histopathological study of lateral tumour spread and surgical excision. Lancer 1986;2:996–9.
- Allison PD. Imputation of Categorical Variables with PROC MI, presented at the 30th Meeting of SAS Users Group International, April 10–13 2005, Philadephia, PA.
- Das P, Skibber JM, Rodriguez-Bigas MA, Feig BW, Chang GJ, Wolff RA, et al. Predictors of tumor response and downstaging in patients who receive preoperative chemoradiation for rectal cancer. Cancer 2007;109:1750–5.
- Park JW, Lim SB, Kim DY, Jung KH, Hong YS, Chang HJ, et al. Carcinoembryonic antigen as a predictor of pathologic response and a prognostic factor in locally advanced rectal cancer patients treated with preoperative chemoradiotherapy and surgery. Int J Radiat Oncol Biol Phys 2009;74:810–17.
- Yoon SM, Kim DY, Kim TH, Jung KH, Chang HJ, Koom WS, et al. Clinical parameters predicting pathologic tumor response after preoperative chemoradiotherapy for rectal cancer. Int J Radiat Oncol Biol Phys 2007;69:1167–72.
- Khan AA, Klonizakis M, Shabaan A, Glynne-Jones R. Association between pretreatment haemoglobin levels and morphometric characteristics of the tumour, response to neoadjuvant treatment and long-term outcomes in patients with locally advanced rectal cancers. Colorectal Dis 2013;15:1232–7.
- Garland ML, Vather R, Bunkley N, Pearse M, Bissett IP. Clinical tumour size and nodal status predict pathologic complete response following neoadjuvant chemoradiotherapy for rectal cancer. Int J Colorectal Dis 2014;29:301–7.
- Garcia-Aguilar J, Hernandez de Anda E, Sirivongs P, Lee SH, Madoff RD, Rothenberger DA. A pathologic complete response to preoperative chemoradiation is associated with lower local recurrence and improved survival in rectal cancer patients treated by mesorectal excision. Dis Colon Rectum 2003;46:298–304.
- Bipat S, Glas AS, Slors FJ, Zwinderman AH, Bossuyt PM, Stoker J. Rectal cancer: local staging and assessment of lymph node involvement with endoluminal US, CT, and MR imaging-a meta-analysis. Radiology 2004;232:773–83.
- Wolthuis AM, Penninckx F, Haustermans K, De Hertogh G, Fieuws S, Van Cutsem E, et al. Impact of interval between neoadjuvant chemoradiotherapy and TME for locally advanced rectal cancer on pathologic response and oncologic outcome. Ann Surg Oncol 2012;19:2833–41.
- Kalady MF, de Campos-Lobato LF, Stocchi L, Geisler DP, Dietz D, Lavery IC, et al. Predictive factors of pathologic complete response after neoadjuvant chemoradiation for rectal cancer. Ann Surg 2009;250:582–9.
- Tulchinsky H, Shmueli E, Figer A, Klausner JM, Rabau M. An interval >7 weeks between neoadjuvant therapy and surgery improves pathologic complete response and disease-free survival in patients with locally advanced rectal cancer. Ann Surg Oncol 2008;15:2661–7.
- van Stiphout RG, Lammering G, Buijsen J, Janssen MH, Gambacorta MA, Slagmolen P, et al. Development and external validation of a predictive model for pathological complete response of rectal cancer patients including sequential PET-CT imaging. Radiother Oncol 2011;98:126–33.
- Huh JW, Kim HR, Kim YJ. Clinical prediction of pathological complete response after preoperative chemoradiotherapy for rectal cancer. Dis Colon Rectum 2013;56:698–703.
- Buijsen J, van Stiphout RG, Menheere PP, Lammering G, Lambin P. Blood biomarkers are helpful in the prediction of response to chemoradiation in rectal cancer: a prospective, hypothesis driven study on patients with locally advanced rectal cancer. Radiother Oncol 2014;111:237–42.
- Joye I, Deroose CM, Vandecaveye V, Haustermans K. The role of diffusion-weighted MRI and (18)F-FDG PET/CT in the prediction of pathologic complete response after radiochemotherapy for rectal cancer: a systematic review. Radiother Oncol 2014;113:158–65.
- Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology 2016;278:563–77.
- Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun 2014;5:4006.