Early Cost-effectiveness Analysis of Risk-Based Selection Strategies for Adjuvant Treatment in Stage II Colon Cancer: The Potential Value of Prognostic Molecular Markers

After curative surgery, stage II patients with colon cancer have a 15% to 20% risk to develop a recurrence (1–3). This risk may be reduced by treatment with adjuvant chemotherapy. However, to prevent unnecessary exposure to potentially toxic treatment in patients who have already been cured by surgery alone, only stage II patients with a high risk of recurrence should be eligible for chemotherapy. The Dutch guidelines currently recommend to prescribe adjuvant chemotherapy only for stage II patients with microsatellite stable (MSS) pT4 tumors (4). However, insight into the molecular heterogeneity of tumors is increasing (5–7). Additional molecular markers, with an impact on either prognosis or response to adjuvant treatment, could potentially improve the currently used high-risk classification of patients with stage II colon cancer (5, 8).

In a recent model-based cost-effectiveness study, we showed that adding BRAF and/or KRAS mutation status as a selection criterion for adjuvant chemotherapy in stage II colon cancer can potentially improve patients' survival (9). In addition to BRAF and KRAS mutation status, the prognostic value of a large number of other molecular markers in the stage II colon cancer population has been demonstrated (5, 8). To take the interconnectivity between these markers into account, the consensus molecular subtypes (CMS) for colorectal cancer were introduced (7). The CMS classification distinguishes four subtypes of specific molecular markers: CMS1 (MSI immune), CMS2 (canonical), CMS3 (metabolic), and CMS4 (mesenchymal; ref. 7). A pooled analysis of 1,785 patients with stage I to IV colorectal cancer showed that patients with a CMS4 classification have a worse prognosis in terms of disease-free survival compared with patients with CMS1 [HR: 1.77; 95% confidence interval (CI), 1.34–2.34], CMS2 (HR: 1.70; 95% CI, 1.39–2.06), and CMS3 (HR: 1.74; 95% CI, 1.29–2.33). Given the prognostic character, the CMS classification was discussed by several studies as a promising feature to inform adjuvant treatment decisions in stage II colon cancer (7, 10, 11). However, limited survival data for CMS is available in the stage II population, since the classification system is only introduced in 2015.

In this study, we aimed to explore the potential value of using the prognostic information of CMS in stage II colon cancer treatment selection and to provide insight into the value of additional prospective research for the Dutch stage II colon cancer population by means of an early cost-effectiveness assessment. Using the Personalized Adjuvant TreaTment in EaRly stage coloN cancer (PATTERN) model (12), we compared health effects and costs of three selection strategies to allocate chemotherapy in the stage II colon cancer population: (i) the current Dutch guideline strategy, (ii) a mutation-based selection strategy, and (iii) a CMS-based strategy. We performed a number of sensitivity analyses to examine the robustness of the results. Furthermore, we evaluated decision uncertainty in our cost-effectiveness assessment using a value of information (VOI) analysis.

The PATTERN model has been comprehensively described elsewhere (12). The flowchart of the model is shown in Supplementary Fig. S1 and the model parameters are reported in Supplementary Table S1. In short, the PATTERN model is a Markov cohort model with five health states: diagnosis, recurrence, 90-day mortality, death of other causes, and death due to colon cancer. We used the Netherlands Cancer Registry (NCR) data for model parametrization (n = 2,271; ref. 13). Only patients without adjuvant treatment were selected for model quantification. As an additional step, an HR for treatment effect was included in the transition from diagnosis to recurrence, based on pooled trial data (Supplementary Table S1; ref. 14). The PATTERN model distinguishes 72 clinical and/or pathologic subgroups for pT stage, tumor sidedness, number of evaluated lymph nodes, and age, which were weighted such that the distribution of clinical and pathologic features reflects the Dutch stage II colon cancer population (Supplementary Methods and Materials). The PATTERN model was extensively validated, both internally and externally (12).

Data of three cohorts (n = 334) were used to distinguish between three biomarker subgroups in the PATTERN model: (i) microsatellite instable tumors (MSI), independent of BRAF and KRAS mutation status, (ii) MSS tumors without a mutation for BRAF or KRAS (MSSdwt), and (iii) MSS in combination with a mutation in BRAF and/or KRAS (MSSmut; refs. 15, 16). The prognostic impact of biomarker status was incorporated in the model by means of an HR affecting the transition from diagnosis to recurrence. These HRs were estimated to be 0.25, 0.88, and 1.53 for the MSI, MSSdwt, and MSSmut subgroups compared with the general stage II colon cancer population, respectively (Supplementary Table S1; ref. 12). A detailed description of how we estimated and incorporated these HRs in the PATTERN model is described elsewhere (12).

To include estimates for the prognostic impact of CMS subgroups in the PATTERN model, we used patient-level data of 428 patients with stage II colorectal cancer that were included in the pooled analysis of Guinney and colleagues (2015), which focused on the optimal classification of CMS (7, 17–21). Baseline characteristics for included patients are shown in Supplementary Table S2. The average age was 67.1 years. The majority of the patients had a CMS2 classification (45.6%), followed by CMS4 (23.1%), CMS1 (18.2%), and CMS3 (13.1%). Eighty-four recurrences (19.6%) were observed within an average follow-up time of 51 months (range: 1–201 months). As the disease-free-survival in patients with CMS1, CMS2, and CMS3 was comparable (Supplementary Fig. S2), we distinguished two CMS subgroups in the PATTERN model: CMS1–3 and CMS4 (Supplementary Methods and Materials). To incorporate CMS classification in the PATTERN model, the time-to-event model reflecting the transition from diagnosis to recurrence was adjusted by including HRs of 0.80 and 1.70 for CMS1to CMS3 and CMS4, respectively compared with the general stage II population (Supplementary Table S1; ref. 12).

Three strategies were evaluated. First, we evaluated a strategy that represented the current Dutch guideline recommendations. In this strategy, patients with a MSS status AND a pT4 stage receive adjuvant chemotherapy (Dutch guideline strategy). Second, a strategy was evaluated in which patients with a MSS status combined with a mutation in BRAF and/or KRAS OR a MSS status combined with a pT4 stage receive adjuvant chemotherapy (mutation strategy; ref. 22). In the third strategy, patients with a CMS4 classification or a pT4 stage receive adjuvant chemotherapy (CMS strategy). Note that these strategies were solely based on the prognostic value of the MMR, BRAF, KRAS, and CMS parameters (Supplementary Table S3). For all strategies we assumed adherence as observed in 2018 to 2019 NCR data. That is, 44% of the high-risk patients ages <75 received chemotherapy and 11% of the high-risk patients ages ≥75 received chemotherapy. In accordance with the Dutch guidelines, patients were treated with 3 months of capecitabine plus oxaliplatin (CAPOX).

We present an overview of resource use, costs, and utilities in Table 1. Costs were determined from a societal perspective and included costs for initial surgery, mutation testing, gene expression profiling, drug costs, costs of managing adverse events, absenteeism from work, patient's travel to the hospital, surveillance, and treatment for recurrence of disease (4, 22–30). Surveillance was based on the recommendations in the Dutch guideline (4), that is, a carcino-embryonal antigen (CEA) determination every 6 months and an ultrasound scan or CT scan of the liver once a year during the first 5 years after surgery. In addition, patients undergo colonoscopy every 3 years, with the first colonoscopy one year after surgery.

Health utilities were estimated using data from the Prospective Dutch ColoRectal Cancer cohort (PLCRC; ref. 31). For the present study, we selected 859 participants who were diagnosed with stage II or III CC between 2011 and 2019 and completed the EQ-5D-5L (32, 33). The patients’ scores on this quality-of-life questionnaire were summarized into a health utility score using the Dutch tariff (34).

To inform the PATTERN model, average health utilities were calculated for different time periods in the disease process, separately for patients with and without adjuvant chemotherapy (33). A full overview of the estimated utilities for each time period is provided in Table 1.

Model predictions were conducted from a societal perspective, using a lifelong time horizon. The outcomes included the number of recurrences and deaths due to colon cancer per 1,000 patients, life years per patient (pp), quality-adjusted life years (QALYs) pp, total life-time costs pp, and net monetary benefit (NMB) associated with each strategy. The NMB was calculated for each strategy as: total QALYs × willingness-to-pay threshold – total cost, using a threshold of €50,000/QALY. Dutch discount rates of 1.5% and 4% were used for health effects and costs, respectively (26). We conducted an incremental cost-effectiveness analysis in which the evaluated strategies were ordered from lowest to highest costs. Subsequently, incremental cost-effectiveness ratios (ICER) were calculated between successive nondominated strategies. The strategy associated with the highest ICER below a willingness-to-pay threshold of €50,000 per QALY was considered the cost-effective strategy (35).

To evaluate the impact of uncertainty on our model predictions, we conducted two one-way sensitivity analyses. First, we increased and decreased the risk to develop a recurrence with 10% in the CMS4 group. Note that the risk of a recurrence in the CMS1 to CMS3 group increased/decreased as well to maintain the same average recurrence risk in the population. Second, we increased and decreased the risk to develop a recurrence with 10% in the MSSmut group. Note that the recurrence risk for the MSI and MSSdwt increased/decreased as well to maintain the same average recurrence risk in the population.

Because of rapid developments in the field of genetic testing, the price of a test to determine CMS may likely decrease in the coming years (30). Therefore, we conducted a threshold analysis in which we decreased the costs for determining CMS from €500 in the base-case analysis to €0 in steps of €100. Subsequently, we compared the NMBs for the CMS strategy to the NMB for the mutation strategy to evaluate if the costs for CMS classification would influence the cost-effectiveness ordering of strategies.

In the base-case analysis, we focused completely on the prognostic value of the MMR, BRAF, KRAS, and CMS parameters. However, some studies indicated that patients with a CMS4 classification may be resistant to adjuvant chemotherapy (36–38). To investigate the influence of this potential resistance on optimal treatment selection for patients with stage II colon cancer, we evaluated a scenario in which we assumed that treatment has no effect in patients with a CMS4 classification. Since prescribing adjuvant treatment to patients with CMS4 would be useless without any benefit of the treatment, the CMS strategy was not included in this scenario analysis. Instead, we defined two strategies in which we used the CMS4 classification as feature to exclude patients from adjuvant chemotherapy. The following four strategies were evaluated in this scenario analysis: (i) The Dutch guideline strategy, (ii) the mutation strategy, (iii) a strategy in which patients with MSS and pT4 receive chemotherapy, unless a patient is classified as CMS4 (Dutch guideline strategy without CMS4), (iv) a strategy in which patients with MSS status in combination with a mutation in BRAF and/or KRAS OR pT4 receive chemotherapy, unless the patient is classified as CMS4 (mutation strategy without CMS4).

In line with the base-case analysis, a patients’ CMS status was not determined in strategy 1 and 2. That is, part of the patients which were selected for adjuvant chemotherapy in these strategies have a CMS4 classification. We assumed no treatment effect for those patients with a CMS4 classification in the evaluation of strategies 1 and 2, which is the HR for treatment was set to 1 (Supplementary S2; Supplementary Table S6). Subsequently, we adjusted the treatment effect for the other subgroups to 0.65, so that the weighted mean was in line with the treatment effect of 0.73, which was the estimate for the general stage II population (14). In strategy 3 and 4, the CMS status of the patient was determined, so that we were able to exclude patients with a CMS4 classification from adjuvant chemotherapy. Note that for patients that were selected for adjuvant chemotherapy in strategy 3 and 4, also the adjusted treatment effect of 0.65 was applied.

To estimate the joint impact of uncertainty in all model parameters on model predictions for the three selection strategies evaluated in the base-case analysis, a probabilistic analysis with 1,000 iterations was conducted. The parameters in the PATTERN model were varied simultaneously according to their most appropriate distribution (Table 1; Supplementary Table S1). More detailed information about the probabilistic analysis is given in Supplementary Methods and Materials. To visually illustrate the results of the probabilistic analysis, a cost-effectiveness acceptability curve (CEAC) was constructed for a range of willingness-to-pay thresholds (€0-€100,000) per QALY. A CEAC depicts the probability for each strategy to result in the highest NMB at a specific willingness-to-pay threshold. Furthermore, the results of the probabilistic analysis were depicted in cost-effectiveness planes.

A VOI analysis provides insight in the decision uncertainty in the evaluation based on currently available evidence. The expected value of perfect information (EVPI) quantifies the value of eliminating uncertainty from all parameters included in the model, and reflects the maximum value that decision makers should be willing to pay for future research (39, 40). The expected value of perfect parameter information (EVPPI) quantifies the value of reducing decision uncertainty by eliminating uncertainty from one parameter or a group of parameters (39). The EVPPI helps to identify drivers of decision uncertainty and to select the potential outcomes that should be targeted in future research. To inform policy makers on the value of future research focusing specifically on an assessment of the prognostic value of genetic features in the stage II CC population, we examined the EVPPI for three sets of parameters: (i) CMS parameters; (ii) MMR, BRAF, and KRAS parameters; and (iii) CMS, MMR, BRAF, and KRAS parameters. As recommended by Rothery and colleagues, we extrapolated our VOI results to a population level using an annual incidence of 3,225 newly diagnosed patients with stage II colon cancer in the Netherlands (1, 39). Furthermore, we assumed a 10-year decision horizon in our VOI analyses. We estimated the EVPI and EVPPI using the Sheffield Accelerated Value of Information (SAVI) tool, which uses a non-parametric regression approach to estimate the EV(P)PI (41).

For the current Dutch guideline strategy, the model predicted 162 CC recurrences and 137 CC deaths per 1,000 patients (Table 2). The CMS strategy and mutation strategy were more effective compared with the Dutch guideline strategy with a decrease of 2.5% and 3.1% in the number of recurrences and 2.2% and 2.9% in CC mortality, respectively. The Dutch guideline strategy resulted in the lowest predicted QALYs of 8.10 pp, followed by the CMS strategy with 8.12 QALYs pp, and the mutation strategy with the highest number of predicted QALYs of 8.13 pp.

The predicted costs were with €23,660 pp lowest in the current Dutch guideline strategy (Table 2). The costs associated with the CMS strategy and with the mutation strategy were higher with €24,643 and €24,542 pp, respectively. The main cost driver in all three strategies was the cost for the initial surgery (Supplementary Table S4). The highest NMB at a willingness-to-pay threshold of 50,000 €/QALY was found for the mutation strategy (€381,827), indicating that this strategy was the preferred strategy at this threshold.

In the incremental cost-effectiveness analysis, the current Dutch guideline strategy served as the first comparator, due to the lowest predicted cost for this strategy (Table 2). The mutation strategy was associated with an ICER of 28,893 €/QALY, and represented a cost-effective strategy at a willingness-to-pay threshold of €50,000/QALY (Fig. 1). The CMS strategy was dominated by the mutation strategy as the mutation strategy predicted higher QALYs and lower costs compared with the CMS strategy.

Results of the one-way sensitivity analyses are shown in Supplementary Table S5. Increasing/decreasing the recurrence risk in the MSSmut group and decreasing the recurrence risk in the CMS4 group led to comparable ICERs and the same optimal strategy as in the base-case analysis.

When the recurrence risk in the CMS4 group was increased, the CMS strategy was no longer dominated by the mutation strategy and resulted in an ICER of 24,123 €/QALY compared with the Dutch guideline strategy. The mutation strategy was subject to extended dominance by the CMS strategy. Thus, in this analysis the CMS strategy was the preferred choice.

Results of the threshold analysis in which we decreased the test costs for a CMS classification are shown in Supplementary Fig. S3. The NMB for the CMS strategy was lower compared with the mutation strategy when the test costs were between €500 and €200. From test costs from €100 or lower, the NMB of the CMS strategy was higher compared with the mutation strategy, indicating that the CMS strategy was the optimal strategy when the test costs were €100 or lower.

Results of the scenario analysis are reported in Table 3. The Dutch guidelines strategy served as the first comparison in the incremental cost-effectiveness analysis, given the lowest costs pp (€23,656). The mutation strategy without CMS4 resulted in an ICER of 32,633 €/QALY compared with the Dutch guidelines strategy, which is considered as cost-effective assuming a willingness-to-pay threshold of 50,000 €/QALY. Furthermore, the Dutch guideline strategy without CMS4 was subject to extended dominance and the mutation strategy was dominated by the mutation strategy without CMS4.

The CEAC shows that the Dutch guideline strategy was most likely to give the highest NMB up to a willingness-to-pay threshold of 35,000 €/QALY (Fig. 2). From a willingness-to-pay of 35,000 €/QALY onwards, the mutation strategy had the highest probability to result in the highest NMB. The cost-effectiveness planes for the CMS strategy and mutation strategy compared with the Dutch guideline strategy are shown in Supplementary Fig. S4.

Figure 3 shows the population EVPI and EVPPI for a range of willingness-to-pay thresholds between €0 and €100,000 per QALY. Assuming a willingness-to-pay threshold of 50,000 €/QALY, the population EVPI was €19M, which is the expected value of eliminating all current decision uncertainty. Thus, this value represents the maximum value that decision makers should be willing to pay for future research to improve their decision-making on the use of CMS and gene mutation in stage II CC treatment selection. The prognostic parameters for MMR, BRAF, KRAS, and CMS status were the main drivers of the identified decision uncertainty as the population EVPPI for these parameters (€18M) was close to the population EVPI (€19M). We further evaluated the EVPPI for the CMS parameters and gene mutation (MMR, BRAF, and KRAS) parameters separately to examine whether the decision uncertainty was driven by one or both of these parameter subsets. Assuming a willingness-to-pay threshold of €50,000/QALY, we found a high EVPPI for both CMS parameters (€14M) and gene mutation parameters (€6M).

In this early cost-effectiveness study, we compared the (cost-)effectiveness of three selection strategies to allocate chemotherapy in the stage II CC population: (i) the current Dutch guidelines, (ii) a selection strategy in which patients with a MSS status combined with a mutation in BRAF and/or KRAS or a MSS status combined with a pT4 received chemotherapy, and (iii) a selection strategy in which patients with CMS4 OR pT4 received chemotherapy. These strategies were completely focused on the prognostic value of the MMR, BRAF, KRAS, and CMS parameters. The CMS strategy and the mutation strategy were both more effective and more costly compared with the current Dutch guideline strategy. In an incremental cost-effectiveness analysis, the mutation strategy was the optimal strategy at a willingness-to-pay threshold of €50,000/QALY with an ICER of €28,893/QALY compared with the current Dutch guidelines. In addition, we evaluated the population EVPI and EVPPI for the MMR, BRAF, KRAS, and CMS parameters. Assuming a willingness-to-pay threshold of €50,000/QALY, the expected value of reducing decision uncertainty by conducting further research on the MMR, BRAF, KRAS, and CMS parameters was €18M.

Although the CMS strategy was dominated by the mutation strategy in the base-case analysis, the differences between both strategies were very small. To illustrate, comparing both the CMS strategy and mutation strategy to the current Dutch guideline strategy led to ICERs of €40,747/QALY and €28,893/QALY, respectively. Thus, assuming a willingness-to-pay threshold of €50,000/QALY, both strategies would be considered cost-effective compared with the current Dutch guidelines. In addition, in the sensitivity analysis in which we increased the recurrence risk with 10% in the CMS4 group, the CMS strategy became the optimal strategy with an ICER of €24,123/QALY and the mutation strategy was no longer considered as cost-effective as this strategy was subject to extended dominance by the CMS strategy. The results of this sensitivity analysis show that the optimal treatment decision depends on the prognostic value of the MMR, BRAF, KRAS, and CMS parameters, which were the biggest drivers of decision uncertainty in our VOI analysis. Given the weak underlying evidence, small differences in predicted QALYs and costs, and the high identified decision uncertainty, the optimal strategy cannot be determined on the basis of the currently available evidence.

When interpreting our results, it should be noted that the dataset available from the study of Guinney and colleagues, which was used for parametrization of the prognostic value of CMS, also included rectal cancer patients (approximately 15%). As the distribution of patients with colon cancer and rectal cancer is almost equal across the CMS subtypes and the proportion of rectal cancer patients was low, we expect that this had only a minor influence on the estimated HRs (7). Furthermore, it should be noted that in this study the molecular-based strategies were compared with the current Dutch guideline, which recommends adjuvant chemotherapy in patients with a MSS or pT4tumor. The Dutch guideline deviates from other international guidelines such as the European Society for Medical Oncology (ESMO) and American Society for Clinical Oncology (ASCO), as these include, besides MSS and pT4, also other high-risk features, that is, <12 lymphnodes evaluated, primary tumor perforation, poor tumor differentiation, extramural vascular invasion (ASCO only), and obstructive tumors (ASCO only) in the decision to select patients for adjuvant chemotherapy (42, 43). We were not able to compare the molecular-based strategies to strategies based on the ESMO and ASCO guideline recommendations as the high-risk features included in ESMO and ASCO guidelines were not included in the PATTERN model. As the ESMO and ASCO guidelines include more clinical and pathological high-risk factors than the Dutch guidelines, the percentage of treated patients would have been higher in a potential ESMO or ASCO strategy compared with the Dutch guideline strategy. If these clinical and pathologic factors partly correlate with the molecular markers, the added value of the molecular markers might be more limited for the ESMO and ASCO guideline.

In addition, it should be kept in mind that we assumed in the base-case analysis an equal treatment effect of adjuvant chemotherapy in all subgroups in the PATTERN model. That is, patients were selected for adjuvant chemotherapy solely based on their expected prognosis. To our knowledge, there are currently no indications for heterogeneous treatment effects in the clinical, pathologic, and biomarker subgroups. However, there are cautious indications that CMS4 patients are less likely to benefit from adjuvant chemotherapy (36–38, 44). We did not take into account this potential resistance in the base-case analysis. The rationale for this choice is that currently available evidence is weak. However, it is of great importance that more data will be collected in the near future on the predictive value of CMS. When sufficient data is available, model analyses should be updated. Alternatively, we conducted a scenario analysis in which we assumed that adjuvant chemotherapy has no effect in patients with CMS4. The results of this scenario analysis showed that it would be beneficial to exclude patients with CMS4 from adjuvant chemotherapy in case of resistance, because it leads to higher QALYs (less toxicity from ineffective treatment) and lower costs. This emphasizes the importance of including the predictive value of molecular markers in the decision making process, in addition to the prognostic value of these markers.

Conducting additional research which focuses on the MMR, BRAF, KRAS, and CMS parameters has the potential to improve decision-making on the optimal use of CMS and gene mutation data in stage II colon cancer treatment selection and to reduce current decision uncertainty. A suitable design for additional data collection is a prospective observational cohort study. More specifically, data could be obtained as part of PLCRC, an observational nationwide cohort study in the Netherlands (31). Given that the expected costs of an observational study are substantially lower than the EVPPI estimate identified in our study (€18M; ref. 45), further research would be valuable to reduce the current decision uncertainty and improve decision-making on the value of MMR, BRAF, KRAS, and CMS for treatment selection (46).

Before the CMS classification system could be used in daily clinical practice to guide adjuvant treatment decisions in the stage II colon cancer population, a number of challenges should be overcome. First, a more standardized procedure to determine a patients’ CMS classification is needed. Nowadays, different procedures are used between studies, which may lead to heterogeneous results (7, 47). Thus, future research should focus on optimizing the CMS test procedure. Second, additional data collection is not only needed to reduce the uncertainty around the prognostic and potential predictive value of molecular markers, but also to gain more insight in the interconnectivity between the CMS classification system and clinical and pathologic high-risk features, such as pT stage and number of evaluated lymphnodes. To illustrate, a recent cohort study of 30 patients with stage II CC with a CMS4 classification showed that the 5-year overall survival (OS) differs between low-risk patients (5-year OS: 41.7%) and high-risk patients (pT4 or < 10 lymphnodes evaluated; 5-year OS: 68.0%), which suggests that clinical an pathologic characteristics are related to tumor biology (48). The association between CMS and clinical and pathologic features should be further investigated in a larger sample size to optimize the identification of high-risk stage II patients.

In conclusion, this is the first study that evaluates the cost-effectiveness of a CMS-based selection strategy for adjuvant chemotherapy in the stage II colon cancer population. Both the CMS-based and mutation-based strategy were more effective and more costly compared with the current Dutch guideline. Given a threshold of €50,000/QALY, the preferred option would be to treat patients according to their BRAF and KRAS mutation status in addition to their MMR status and pT4. However, there was substantial decision uncertainty concerning the choice of optimal selection strategy, with the MMR, BRAF, KRAS, and CMS parameters as the main drivers of this decision uncertainty. Given this decision uncertainty, the choice for the optimal strategy cannot be based on the currently available evidence. Additional research on the prognostic and predictive value of the MMR, BRAF, KRAS, and CMS parameters has the potential to reduce this decision uncertainty and improve decision-making on the use of molecular markers to guide the treatment selection.

M.J.E. Greuter reports grants from ZonMw during the conduct of the study. M. Koopman reports other support from Nordic Farma, Merck-Serono, Pierre Fabre, and Servier, and grants from Bayer, Bristol Myers Squibb, Merck, Roche, and Servier outside the submitted work. L. Vermeulen reports personal fees from Genentech, and grants from Pierre Fabre, MSD, Bayer, and Servier outside the submitted work. G.R. Vink reports grants from Servier, Sirtex, BMS, Bayer, Merck, Pierre Fabre, and Personal Genome Diagnostics outside the submitted work. V.M.H. Coupé reports grants from ZonMw during the conduct of the study, as well as grants from KWF, ZonMw, Novartis, and GSK outside the submitted work. No disclosures were reported by the other authors.

G. Jongeneel: Conceptualization, formal analysis, methodology, writing–original draft. M.J.E. Greuter: Conceptualization, supervision, methodology, writing–review and editing. N. Kunst: Writing–review and editing. F.N. van Erning: Writing–review and editing. M. Koopman: Writing–review and editing. J.P. Medema: Data curation, writing–review and editing. L. Vermeulen: Data curation, writing–review and editing. J.N.M. Ijzermans: Writing–review and editing. G.R. Vink: Writing–review and editing. C.J.A. Punt: Writing–review and editing. V.M.H. Coupé: Conceptualization, supervision, funding acquisition, writing–review and editing.

Financial support for this study was provided by a grant from ZonMw (grant no.: 848015007). ZonMw had no role in designing the study, interpreting the data, writing the manuscript, and publishing the report. The authors thank the registration team of The Netherlands Comprehensive Cancer Organization (IKNL) for the collection of the data for the NCR.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.