Article Text
Abstract
Objective It is difficult to predict the presence of histological risk factors for lymph node metastasis (LNM) before endoscopic treatment of T1 colorectal cancer (CRC). Therefore, endoscopic therapy is propagated to obtain adequate histological staging. We examined whether secondary surgery following endoscopic resection of high-risk T1 CRC does not have a negative effect on patients' outcomes compared with primary surgery.
Design Patients with T1 CRC with one or more histological risk factors for LNM (high risk) and treated with primary or secondary surgery between 2000 and 2014 in 13 hospitals were identified in the Netherlands Cancer Registry. Additional data were collected from hospital records, endoscopy, radiology and pathology reports. A propensity score analysis was performed using inverse probability weighting (IPW) to correct for confounding by indication.
Results 602 patients were eligible for analysis (263 primary; 339 secondary surgery). Overall, 34 recurrences were observed (5.6%). After adjusting with IPW, no differences were observed between primary and secondary surgery for the presence of LNM (OR 0.97; 95% CI 0.49 to 1.93; p=0.940) and recurrence during follow-up (HR 0.97; 95% CI 0.41 to 2.34; p=0.954). Further adjusting for lymphovascular invasion, depth of invasion and number of retrieved lymph nodes did not alter this outcome.
Conclusions Our data do not support an increased risk of LNM or recurrence after secondary surgery compared with primary surgery. Therefore, an attempt for an en-bloc resection of a possible T1 CRC without evident signs of deep invasion seems justified in order to prevent surgery of low-risk T1 CRC in a significant proportion of patients.
- COLORECTAL CARCINOMA
- COLORECTAL SURGERY
- THERAPEUTIC ENDOSCOPY
- COLONOSCOPY
- COLONIC POLYPS
Statistics from Altmetric.com
Significance of this study
What is already known on this subject?
Endoscopic therapy is adequate for T1 colorectal carcinoma (CRC) in the absence of histological predictors for lymph node metastasis (LNM).
It is difficult to predict the presence of histological predictors before endoscopic resection of a T1 CRC.
High-quality population based data on safety of endoscopic resection of T1 CRC in Western countries is absent or conflicting.
What are the new findings?
Endoscopic resection before surgical resection of a high-risk T1 CRC has no adverse effect on the percentage of patients with LNM at resection, and local and distant recurrence rate during follow-up.
Of all T1 CRCs treated with surgical resection, still 5% develops a local or distant recurrence.
Incomplete resections were observed in 15.3% of T1 CRCs, of which 75% was already judged incomplete during endoscopy.
How might it impact on clinical practice in the foreseeable future?
Endoscopic resection has the potential to be used as the first-line treatment for suspected malignant polyps as an ultimate staging procedure.
There is an increasing need for identification of malignant polyps before endoscopic resection to guide resection technique and optimise specimen handling.
Introduction
The gold standard for treatment of T1 colorectal carcinoma (CRC) used to be an oncological surgical resection including resection of draining lymph nodes. Colorectal surgery is however associated with an overall mortality of 1–5% and morbidity of about 30%, especially in the elderly population.1–4 A recent study from Japan showed that endoscopic resection is adequate for T1 CRC with a low risk for lymph node metastasis (LNM), which concerns approximately 19–29% of all T1 CRCs.5 ,6 Additional surgical resection is required for high-risk T1 CRC.5
Currently, histological evaluation of the resected specimen estimates the risk of LNM and guides therapy after endoscopic resection of T1 CRC. Negative resection margins, well or moderate differentiation, absence of vascular or lymphatic invasion, depth of submucosal invasion of <1000 μm and low-grade tumour budding are histological parameters associated with a low risk of LNM and local recurrence.7 Ideally, endoscopic resection should be used selectively for patients with a low risk of LNM.8–12 However, during colonoscopy it is difficult to reliably discriminate low-risk from high-risk T1 CRC. Since this differentiation can only be made after histological evaluation of the endoscopic resected specimen, it is important to know whether endoscopic resection of high-risk T1 CRC prior to surgical resection has adverse effects. Endoscopic resection through a tumour with sequential positive resection margins and residual tumour could cause tumour spill and raise the risk of LNM and recurrence. The current literature has shown conflicting evidence on this risk. Some studies suggest that incomplete endoscopic resection of T1 CRC might accelerate growth of the remaining tumour and promote metastasis.8 ,13 ,14 Even with complete endoscopic resection, tumour cell dissemination may occur when a piecemeal resection technique is used or when perforation occurs.15 ,16 In contrast, a recent study concluded that endoscopic resection before secondary surgery had no negative effect on short-term outcomes, morbidity, mortality and oncologic surrogate quality markers.1
Long-term outcome after surgical resection of T1 CRC with and without prior endoscopic resection has not been evaluated in Western countries. The aim of this study was to determine whether secondary surgical resection after endoscopic resection of high-risk T1 CRC has an adverse effect on patient outcome compared with primary surgical resection; regarding LNM and recurrence rate.
Materials and methods
Patients
The study design is a multicentre retrospective observational cohort study. All patients from 13 hospitals diagnosed with T1 CRC between 2000 and 2014 and treated with primary or secondary surgical resection were identified in the database of the Netherlands Cancer Registry. Patients were included in the study when they had a high risk for LNM based on the presence of one or more of the following histological criteria: poorly differentiated histology, positive resection margins, deep submucosal invasion depth (defined as submucosal invasion depth >1 mm, Sm2/Sm3 for sessile T1 CRC and Haggitt 3–4 for pedunculated T1 CRC) or presence of vascular or lymphatic invasion. These histological prognostic parameters were evaluated in the original pathology reports. Patients were subdivided into two groups: primary surgical resection and secondary surgical resection (endoscopic resection with additional surgical resection). Patients were excluded in case of (1) hereditary predisposition for CRC (Lynch syndrome, FAP or MAP associated polyposis); (2) IBD; (3) synchronous CRC (T1–T4); (4) death within 1 year due to another cause than T1 CRC; (5) non-adenocarcinoma (eg, carcinoid); (6) missing hospital records; (7) neoadjuvant radiotherapy and (8) metastasis of other origin than CRC (ie, pulmonary cancer).
Methods and definitions
Data on patient characteristics, recurrences, polyp characteristics, treatment and follow-up were collected from hospital records, endoscopy, surgery, radiology and pathology reports. Primary outcomes were LNM and recurrence rates.
LNM was defined as positive lymph nodes in the surgical resected specimen. Recurrence was defined as the detection of local recurrence or distant metastasis during follow-up. Local recurrence was defined as histologically confirmed recurrence at the anastomosis or surrounding the colon after surgery. Distant recurrence was defined as any metastasis localised outside the bowel (ie, liver, lung, brain, peritoneum or bone) confirmed with imaging and/or histology. A second primary CRC diagnosed during follow-up in another segment of the colon or rectum was defined as a metachronous lesion, not as a recurrence.
Comorbidity of patients was extracted from medical records and noted as the Physical Status Classification Score according to the American Society of Anesthesiology (ASA) as mentioned in the anaesthesiology report and the Charlson Comorbidity Index Score.17 ,18
Lateral spreading tumours were defined as lesions with a low vertical axis that extend laterally along the luminal wall. In contrast, pedunculated lesions were defined as polyps attached to the luminal wall with a stalk.19 This definition was used since the endoscopy reports almost never mentioned the Paris classification of colorectal neoplastic lesions.20 The proximal, right-sided colon was defined as the caecum, ascending and transverse colon including the splenic flexure. The distal, left-sided colon was defined as the descending and sigmoid colon, and the rectum. Transanal endoscopic microsurgery (TEM) was classified as an endoscopic resection since no lymph node dissection is performed and therefore it is compatible with local treatment.
Treatment-related morbidity was defined as occurrence of a complication during or after treatment. Complications of endoscopic resection include post-polypectomy bleeding, post-polypectomy syndrome and perforation with emergency surgery. Complications of surgical resection include postoperative bleeding, reoperation, prolonged ileus, wound infection, cardiovascular or pulmonary complications, leakage or stenosis of the anastomosis, abscess and multiple-organ failure. CRC treatment-related mortality was defined as death due to CRC treatment within 30 days after treatment. Follow-up started at the date of diagnosis and ended at the date of detection of a recurrence, death or last follow-up.
Data analysis
We used inverse probability weighting (IPW) to account for baseline differences in prognostic characteristics inherently present in an observational intervention study when comparing T1 CRC patient outcome between the secondary versus primary surgery groups. We specifically used IPW to correct for confounding as it provides marginal treatment effect estimates—also estimated in randomised clinical trials—and as LNM and recurrences are rare events in T1 CRC, limiting the number of parameters that can be included in multivariable models despite our large cohort.
Obtaining IPW-adjusted treatment estimates is a two-step process. We first derived a propensity score by fitting a logistic regression model with treatment status as the dependent variable and potential confounders as predictors (using dummies for categorical and restricted cubic splines for continuous predictors; no interaction terms). As a second step, the inverse of the (propensity score-derived) predicted probability for actual received treatment was used to weigh patients in a regression model relating treatment status as a sole determinant of the outcome of interest (thus minimising the number of parameters to be estimated in the model). In this second step, we used logistic regression for the primary outcome LNM and Cox proportional hazard regression for the primary outcome recurrence.
As several clinicopathological variables had missing data (table 1), and as simply excluding patients with missing data is inefficient and increases the risk of selection bias, we used multiple imputation before data analysis (using multivariate imputation by chained equations21 with 32 variables including patient outcomes—10 imputation data sets, 25 iterations, healthy convergence). Rubin's rules were used to pool results across imputation data sets.22
For our primary analyses, we adjusted for age, sex, polyp location, size, morphology and differentiation grade, extended with the number of retrieved lymph nodes (10+ vs <10), and LNM when evaluating recurrence risk, as we considered these the most important prognostic variables and thus potential confounders. Invasion depth and lymphovascular invasion are also important prognostic variables but were missing too frequently in our data to be included in our primary analysis. Nevertheless, we additionally adjusted for these variables in a secondary supporting analysis. Other patient characteristics such as body mass index or comorbidity have likely influenced the decision for primary or secondary surgery, but are probably not causally related to LNM or recurrence, and we therefore did not consider these to be potential confounders.
Several checks were performed to evaluate the appropriateness and success of our analysis approach. First, we evaluated the improvement in baseline comparability between the two surgical approaches before and after IPW adjustment for confounders. Next, we assessed the c-index of the propensity models from step 1, after weighting patients by their propensity score-derived IPW (a c-index of 0.5 means that the resulting IPW model cannot discriminate which patient received which treatment, indicating successfully obtained balance; a c-index of 1.0 on the other hand indicates extreme remaining imbalance).23 As extreme weights for some individuals can lead to bias and loss of precision in treatment effect estimates,24 we inspected the distribution of weights and conducted sensitivity analyses truncating weights at the 97.5th percentile. With regard to the Cox proportional hazard regression models, we did not find any violation of the proportionality of the hazard assumption following inspection of the scaled Schoenfeld residuals. Finally, we did not evaluate the linearity assumption for continuous variables as these were fitted in the models by restricted cubic spline functions. We used 2000-fold bootstrapping to obtain p values and 95% CIs for IPW-derived estimates, repeating all analysis steps in each bootstrap sample. IBM SPSS Statistics V.20 and R V.3.2.1 for Mac were used for statistical analyses. All statistical tests were two-tailed, and statistical significance was defined as p<0.05.
Results
Baseline characteristics
In total, 602 patients with a high-risk T1 CRC were treated with either primary or secondary surgery and included for analysis (figure 1). Also, 263 patients were treated with primary surgical resection and 339 patients with secondary surgical resection after endoscopic resection. Of the 339 patients treated with secondary surgical resection, 162 had an endoscopic snare resection, 60 piecemeal snaring, 39 en-bloc endoscopic mucosal resection (EMR), 65 piecemeal EMR, 8 TEM, 1 endoscopic submucosal dissection (ESD), and 4 resection types were unknown. Residual tumour was found in 52 patients with endoscopic resection (52/339; 15.3%). An incomplete resection was already suspected in 39 (75.0%) endoscopy reports, while in the remaining 13 cases the endoscopic resection was thought to be radical, although pathology reported positive resection margins. In total, seven perforations occurred due to endoscopic resection (2.1%). Two patients with a perforation developed recurrence. Secondary surgical resection was performed at a median of 49 days after endoscopic resection (p25–p75 39–69). An overview of the baseline characteristics is provided in table 1. Before adjustment by IPW, patients treated with primary surgery were older and had more comorbidity than patients treated with secondary surgery (p<0.001). Polyps treated with primary surgery were larger in size (p<0.001), more often located in the right-sided colon (p<0.001) and had a sessile morphology almost twice as often (p<0.001). Following adjustment by IPW based on age, sex, polyp location, size, morphology, differentiation grade and number of retrieved and positive lymph nodes (ie, the propensity score used in the primary recurrence analysis), baseline characteristics between treatment groups were largely comparable, except for invasion depth. The frequency of lymphovascular invasion was similar for both treatment groups, before and after IPW adjustment.
Based on the propensity scores, the primary surgery group had a lower probability of being selected for secondary surgery than the secondary surgery group. None of the propensity scores yielded extreme weights, with a maximum of 18.1 (97.5th percentile 5.5) for the propensity score used in the primary recurrence analysis, which was 18.3 (97.5th percentile 5.4) for the one used in the primary LNM analysis (which included the same confounders as for recurrence except the LN variables). The post-IPW c-indexes (0.59 and 0.58, respectively) indicated appropriately obtained balance, agreeing with the adjusted baseline characteristics in table 1.
Treatment-related morbidity and mortality
In the primary surgery group, 67 (25%) treatment-related complications were observed (13 cardiovascular events, 13 leakages at the anastomosis, 11 wound infections, 8 bleedings, 6 prolonged ileus episodes, 6 pulmonary events, 4 strictures at the anastomosis, 3 multiorgan failures, 3 abscesses). In the secondary surgery group, 92 complications occurred (27%), of which 22 were associated with the primary endoscopic resection (14 post-polypectomy bleedings, 7 perforations, 1 post-polypectomy syndrome) and 70 associated with the following surgical resection (13 leakages at the anastomosis, 11 bleedings, 9 prolonged ileus episodes, 8 cardiovascular events, 8 wound infection, 7 strictures at the anastomosis, 5 reoperations, 5 pulmonary events, 3 abscesses).
A total of 16 treatment-related complications leading to death within 30 days after treatment were seen. One complication was a perforation during endoscopic resection followed by cardiovascular complications during emergency surgery. The other 15 complications were surgery related, that is, 9 leakages at the anastomosis, 5 cardiovascular complications and 1 abscess.
Patient outcome
Short-term outcome
The median number of retrieved lymph nodes was 6 (p25–p75 3–11) in the secondary surgery group and 7 (p25–p75 3–11) in the primary surgery group (p=0.192). Of all patients included, 55 (9.1%) had LNM. The secondary surgery group had LNM in 8.3% (28/339) of patients, which was 10.3% (27/263) in patients treated with primary surgery, yielding an unadjusted OR of 0.79 (95% CI 0.45 to 1.38; p=0.398). IPW analysis showed that after adjustment for primary confounders the secondary surgery group was at similar risk of LNM as the primary surgery group (OR 0.97; 95% CI 0.49 to 1.93; p=0.940). Further confounder adjustment for invasion depth and lymphovascular invasion did not change these results (table 2).
Long-term outcome
Overall median follow-up was 4.3 years (p25–p75 1.9–6.9). Total follow-up in the primary surgery group (n=263) was 1294.0 years, with a mean follow-up per patient of 4.92 years (SD 3.3). Total follow-up in the secondary surgery group (n=339) was 1548.9 years, with a mean follow-up per patient of 4.57 years (SD 3.3). During follow-up, 34 recurrences were observed, 19 in the primary surgery group and 15 in the secondary surgery group. Distant metastases in the primary surgery group were located in the lung (n=6), peritoneum (n=3), mediastinal lymph nodes (n=3) and liver (n=8), with some patients having more than one distant metastasis. Distant metastases in the secondary surgery group were located in the lung (n=5), liver (n=5), peritoneum (n=4) and brain (n=1). Additional resection of recurrence was performed in 15 patients (44%). The overall recurrence rate of T1 CRC treated with primary surgery was 14.7 per 1000 person-years and was not significantly different from the overall recurrence rate of T1 CRC treated with secondary surgery, that is, 9.7 per 1000 person-years (unadjusted; p=0.297). Univariate Cox proportional hazard regression analysis showed no factors associated with a higher risk of recurrence. IPW analysis using the propensity score showed no difference in risk of recurrence between primary and secondary surgery (HR 0.97; 95% CI 0.41 to 2.34; p=0.954). Further confounder adjustment for invasion depth and lymphovascular invasion did not change these results (table 2). Also adjusting for the number of lymph nodes retrieved, and for the number of patients with ≥10 lymph nodes, did not change long-term outcome.
Discussion
In this study, endoscopic resection before surgical resection of a pT1 CRC with risk factors for LNM (high-risk T1 CRC) had no adverse effect on patients' outcome, regarding LNM and recurrence rates.
Approximately 19–29% of all T1 CRCs can be identified as having a low risk for LNM with current criteria,5 ,6 and this subgroup of patients can be treated with endoscopic treatment only.12 It is however difficult to recognise this specific subgroup of low-risk T1 CRCs before endoscopic resection as, besides depth of invasion, risk factors for LNM such as lymphatic or vascular invasion, tumour budding and tumour differentiation cannot be assessed during endoscopy. It is therefore propagated to perform an en-bloc resection as a staging procedure in cases of submucosal invasion with absence of identifiers of deep invasion (eg, Sano IIIB, Hiroshima C3, NICE 3 or Kudo Vn or Vi-high pit pattern with magnifying chromoendoscopy25–29) and to proceed to adjuvant surgery when histological risk features for LNM are present.30 Such a strategy would prevent unnecessary surgery for both low-risk T1 CRC and patients with high-grade dysplasia (intramucosal carcinomas), which often cannot be discriminated from superficial invasion with current optical classification systems. It is however important to know whether this strategy is safe and does not have a negative effect on the outcome of patients with a high-risk T1 CRC.
To our knowledge, the present study is the largest multicentre observational cohort study addressing this question, and the first correcting for confounding by indication as efficient as considered possible. The results suggest that the strategy of an endoscopic removal attempt before surgical resection does not increase the risk of LNM or recurrence during follow-up. Recently, Rickert and colleagues reported that endoscopic resection before surgical resection has no adverse effect on short-term surgical and oncological outcomes.1 However, their study had some limitations. First, the control group of patients treated with primary surgery consisted of pT2 CRC in the majority of cases. As pT2 CRCs have a known inferior oncological outcome due to deeper invasion and higher risk of LNM, this may have underestimated the adverse effect of endoscopic resection of a high-risk T1 CRC. Second, only short-term outcomes were reported, without data on distant metastases during follow-up. Finally, their study only included five cases with positive lymph nodes and did not correct for baseline differences between the primary and secondary surgery group. In a study by Nozawa and colleagues, comparing 145 patients with surgical resection following endoscopic treatment to 133 patients with primary surgery for a pT1 CRC, the risk of LNM between secondary and primary surgery was not different.31 However, this study also did not adjust for confounding by indication, and the number of cases with a recurrence was only 4.
The present study has some limitations. First, this is a retrospective observational cohort in which many factors may have contributed to the decision to perform primary surgery over endoscopic treatment. Besides measured factors such as size, location and polyp morphology, unmeasured factors such as the aspect of the polyp (excavation, depression, mucosal friability), difficulty to remove the polyp and local expertise may also have played a role. Most of these factors are related to depth of invasion. Even after adjusting for depth of invasion, no higher risk of LNM in the secondary surgery group was observed (table 2). The drawback of adjusting for depth of invasion however was that only 54% complete cases could be analysed, which necessitates imputation and makes analyses less robust. Adjusting for lymphovascular invasion, the strongest predictor of LNM did not show an increased risk of LNM in the secondary surgery group (table 2). Furthermore, when corrected for positive lymph nodes and the number of lymph nodes retrieved, we could not detect a higher risk of recurrence in the secondary surgery group. Although we have adjusted for confounding by indication as much as possible, it cannot be completely ruled out that differences in risk of both LNM and recurrence in both groups remained based on unmeasured factors. In addition, it can also not be ruled out that a group of patients with a poorer prognosis based on tumour biology was selected for primary surgery. As only seven cases of perforation were observed after endoscopic resection, it was impossible to calculate the risk of perforation on LNM and recurrence. We were also unable to evaluate the risk of LNM and recurrence for incomplete resections, with only 3/52 (18.6/1000 person-years) patients with an incomplete resection showing recurrence versus 10/283 (7.8/1000 person-years) with a complete resection. Although this study supports the strategy of an endoscopic resection before surgery, it should be noted that not all T1 CRCs are good endoscopic candidates. It is necessary to predict the presence of deep invasion with narrow band imaging (NBI) or chromoendoscopy in order to select cases suited for endoscopic resection, and to prevent an endoscopic treatment of deep invasive T1 CRC where endoscopic cure can often not be achieved and the risk of perforation is increased. Furthermore, endoscopic resection should be performed safe and en-bloc with proper margins and sample handling in order to increase success. In our opinion, endoscopic resection of T1 CRC should therefore be performed by experienced endoscopists who are sufficiently skilled in both EMR and ESD to minimise the risk of complications, to assure high rates of en-bloc resections for adequate staging and to lower the risk of residual tissue.32 ,33
A second limitation is that the number of lymph nodes retrieved in our observational cohort is below the advised number of 10 lymph nodes in 64–66%. In advanced CRC, sampling more lymph nodes has shown to be associated with a decreased risk of recurrence, suggesting understaging of the N-status when insufficient lymph nodes are retrieved.34 Therefore, a minimum of 10 or 12 lymph nodes are advocated in different international guidelines.35–39 The low number of retrieved lymph nodes could therefore potentially have influenced both end points of our study. Earlier studies observed that advanced T-stadium is associated with increased lymph node yield and reported yields in stage I CRC in line with our findings.40–42 A study that focused on T1 and T2 CRC reported that in case of T1 CRC examination of four lymph nodes may be adequate.43 However, the authors conclude that the lowest lymph node yield with the highest statistical significant association with survival is the minimum cut-off value. In a reply by Metze, this was shown to be a biased approach in a simulation study.35 It is therefore still unclear what the optimal cut-off value for lymph node yield in T1 CRCs should be in clinical practice. In our opinion, the influence of the observed lower number of lymph nodes retrieved on the outcome of our study is limited. First, we observed that lymph node yield was equally distributed over both groups. Second, we adjusted for both the presence of LNM and the number of lymph nodes retrieved. The latter was performed with retrieved lymph nodes as flexible continuous variable, dichotomised as ≥10 lymph nodes and ≥4 lymph nodes. All these additional analyses could not show an increased risk of LNM or recurrence after secondary surgery (data not shown).
Another limitation is that different endoscopic resection techniques were used. Due to this variation in techniques, it was impossible to assess the risk of LNM and recurrence for each endoscopic resection technique individually. Therefore, we cannot rule out that the risk of LNM or recurrence varies between the different resection techniques and that our case mix of techniques influences the risk observed in our secondary surgery group. Patients were however selected on a histological diagnosis of a T1 CRC in consecutive order in time. Selection of patients was therefore independent of the endoscopic resection technique and our cohort reflects current practice in the Netherlands.
Although our cohort is one of the largest cohorts at present with a long mean follow-up of 4.7 years, a high number of patients with positive lymph nodes (n=55), and the first study on this topic correcting for confounders as efficiently as possible, the number of recurrences after surgery (n=34) is still limited. The bootstrapped 95% CI surrounding the estimate are therefore wide (0.41 to 2.34). If a parallel-group randomised controlled trial (RCT) with 80% power would be conducted to conclude that secondary surgery is non-inferior to primary surgery, we would need to recruit >1000 patients in each group.44 It could be argued that such a study needs to be able to exclude an HR of even smaller than 1.5, further increasing the study size substantially. We believe that such an RCT will probably never be conducted, and the best available evidence regarding the safety of endoscopic resection followed by surgery for T1 CRC will therefore be provided by observational data.
In conclusion, our data do not show an increased risk of LNM and recurrence after secondary surgery compared with primary surgery. Therefore, an attempt for an en-bloc resection of a possible T1 CRC without evident signs of deep invasion seems justified in order to prevent surgery of low-risk T1 CRC in a significant proportion of patients. In case of high-risk histological features of the resected endoscopic specimen, an additional surgical resection should always be performed.
References
Footnotes
Contributors AO, AvdB, PS and LM conceived and designed the study. PS and LM are the principal investigators. AO, KK, AvdB, BS, TS, HP, WdV, JvB, MK, JMJG, JNG, NL, FB and FW are responsible for patient accrual and inclusion. ML and GO are the expert GE pathologists. SE is the study statistician. AO, KK and LM drafted the manuscript. AvdB, BS, TS, HP, WdV, JvB, MK, JMJG and PS all analysed and interpreted the data, and critically revised the manuscript. All authors approved the study protocol and read and approved the manuscript.
Competing interests None declared.
Ethics approval This study has been approved by the Medical Ethics Review Committee of the University Medical Center Utrecht (reference number WAG/om/15/031817).
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement No unpublished data relating to this study exist. The data for this study were obtained from the Dutch Comprehensive Cancer Organisation as well as from the electronic medical records of the participating centres.