Prognostic Nomogram and Patterns of Use of FOLFIRI‐Aflibercept in Advanced Colorectal Cancer: A Real‐World Data Analysis

This article reports on a prognostic model for use in patients with colorectal cancer initiating therapy with FOLFIRI/aflibercept to predict overall survival and aid in decision making, with the goal of optimizing treatment results in actual patient populations.

Abstract

Introduction.

The VELOUR study evaluated the efficacy and safety of adding aflibercept to FOLFIRI (fluorouracil, leucovorin, irinotecan) in second‐line therapy for metastatic colorectal cancer (mCRC). However, a nomogram that can stratify patients according to prognosis is unavailable, and the frequency and effect of the pragmatic use of modified schedules in actual practice remains unknown.

Method.

The sample consists of 250 patients with mCRC treated with aflibercept and irinotecan‐based chemotherapy at nine Spanish academic centers between January 2013 and September 2015. The result of a Cox proportional hazards model regression for overall survival (OS), adjusted for covariates available in daily practice, was represented as a nomogram and web‐based calculator. Harrell's c‐index was used to assess discrimination.

Results.

The prognostic nomogram for OS includes six variables: Eastern Cooperative Oncology Group performance status, tumor location, number of metastatic sites, mutational status, better response to previous treatment(s), and carcinoembryonic antigen. The model is well calibrated and has acceptable discriminatory capacity (optimism‐corrected c‐index, 0.723; 95% confidence interval [CI], 0.666–0.778). Median OS was 6.1 months (95% CI, 5.1–8.8), 12.4 months (95% CI, 9.36–14.8), and 22.9 months (95% CI, 16.6–not reached) for high‐, intermediate‐, and low‐risk groups, respectively. Age, comorbidity, or use of modified FOLFIRI regimens did not affect prognosis in this series. Grade 3–4 adverse events were less common following modified schedules. The admission rate because of toxicity was higher in ≥65 years (9.7% vs. 19.6%; odds ratio, 2.26; p = .029).

Conclusion.

We have developed and internally validated a prognostic model for use in individuals with colorectal cancer initiating therapy with FOLFIRI‐aflibercept to predict both OS and the effect of pragmatic modifications of the classic regime on efficacy and safety. This can aid in decision making and in designing future trials.

Implications for Practice.

In this study, the authors developed and conducted the internal validation of a prognostic nomogram that makes it possible to stratify patients who are eligible for second‐line FOLFIRI‐aflibercept based on their probability of survival. This model was developed in a multicenter sample from nine Spanish hospitals. Furthermore, to increase the study's validity, the practical use of aflibercept in this setting was investigated, including doses or pragmatic modifications. The results suggest that the modified schedules often used in this daily clinical practice‐based patient population are associated with less severe toxicity without apparent detriment to survival endpoints. It is believed that these data complement the information provided by the VELOUR trial and are relevant for the oncologist in treating colon cancer in the second‐line setting.

Introduction

Metastatic colorectal cancer (mCRC) is a common neoplasm, with a median survival of 30 months [1]. Standard initial therapy for fit patients consists of polychemotherapy, typically FOLFOX (fluorouracil, leucovorin, oxaliplatin) or FOLFIRI (fluorouracil, leucovorin, irinotecan) schedules associated with targeted molecular vascular endothelial growth factor (VEGF) or epidermal growth factor receptor (EGFR) therapies. The choice is conditioned by the tumor's molecular profile, location, and treatment aim [2]. For individuals who have received first‐line FOLFOX and bevacizumab, second‐line FOLFIRI with antiangiogenic agents [3], [4], [5] or with anti‐EGFR therapy for tumors with wild‐type KRAS, NRAS, or BRAF could be used [6], [7].

VEGF is a core regulator of angiogenesis in mCRC, and high VEGF expression is associated with poor prognosis. Blocking VEGF increases the efficacy of chemotherapy in first and successive lines [8]. Aflibercept (VEGF‐Trap; ziv‐aflibercept; Zaltrap) is a human recombinant fusion protein that acts as a “decoy receptor,” blocking angiogenesis by targeting VEGF‐A, VEGF‐B, and placental growth factor [9].

The pivotal, phase III VELOUR study evaluated the efficacy and safety of adding aflibercept to FOLFIRI as second‐line therapy in mCRC, following progression to an oxaliplatin‐based regimen [4]. Compared with placebo, aflibercept demonstrated significantly increased overall survival (OS) rates (hazard ratio [HR], 0.81; 95% confidence interval [CI], 0.71–0.93), as well as progression‐free survival (PFS; HR, 0.75; 95% CI, 0.66–0.86). Nevertheless, aflibercept was also associated with VEGF‐related toxicities, as well as increased grade 3–4 toxicity. Although the benefit was maintained across all strata [10], a post hoc analysis indicated that a profile of good and poor responders could be clinically defined (e.g., based on the Eastern Cooperative Oncology Group performance status [ECOG PS], time to relapse after adjuvant treatment, and number of metastases) [27]. Furthermore, another subanalysis confirmed that, despite efficacy data (PFS and OS) being consistent across age groups, grade 3–4 adverse events (AEs) were more common in the elderly [11]. Likewise, a registry of real‐world data suggested that clinical course during treatment with aflibercept (risk of progression or death) was heterogeneous and possibly affected by simple clinical variables [12].

All these data indicate that in the real world, where patients are often older, have comorbidities, and worse functional status, the risk‐benefit balance could differ from that initially expected [13]. Our group posed the working hypothesis that actual efficacy and safety outcomes might differ from the VELOUR study results, particularly bearing in mind that modified FOLFIRI regimens are commonly used with the aim of decreasing toxicity in vulnerable patients [14], [15], [16]. Therefore, the design of prognostic instruments in registries of real‐world data cannot be detached from the differential use of therapies in patients with unequal fitness. The correlation between the pragmatic variability of real‐world practice and prognostic factors must be fully analyzed before it can be generalized. Based on these premises, we have developed and internally validated a prognostic nomogram that can aid in optimizing treatment results in actual patient populations.

Materials and Methods

Participants and Study Design

A retrospective, observational, postmarketing study was conducted at nine Spanish university hospitals. Clinical and analytical data were obtained from patient histories and recorded by the same physicians caring for them; data were monitored in detail (A.F.M.) to filter inconsistencies, errors, or unjustified missing data.

Eligibility criteria included being adult (≥18 years) with histologically diagnosed, resistant or progressing mCRC, at least one oxaliplatin‐based first‐line chemotherapy or with tumor relapse in the first 6 months following adjuvant treatment (“rapid progressors”). Participants were to have received aflibercept with irinotecan‐based chemotherapy; recruitment was consecutive at each hospital. No intervention was performed, and treatment choice, clinical evaluations, and imaging studies were in line with each center's routine care practices.

The protocol was approved by a reference Clinical Research Ethics Committee (Galicia's CREC). All procedures carried out during the study period were performed in keeping with the Declaration of Helsinki of 1964. Those patients who were included and still alive when clinical information was collected provided written informed consent.

Definition of Variables and Endpoints

The primary aim of the study was the development and internal validation of a prognostic nomogram to predict OS after describing management patterns (modifications of standard treatment) and to evaluate survival and safety endpoints. PFS was defined as the time between initiating aflibercept and tumor progression or demise, censoring event‐free cases at last follow‐up. OS was established as the same time point until death from any cause.

Prior to devising the model, a descriptive analysis was made of possible treatment variability to verify that the sample had a comparable baseline hazard. Dose intensity (DI) was defined as the amount of drug administered per unit of time and expressed as mg/m² per week. Cumulative dose was specified as total dose and reported as total mg/m² administered. Relative dose intensity (RDI) was considered to be the DI administered with respect to the dose intensity scheduled for each regimen. Thus, three treatment groups were specified based on dose modifications: (a) FOLFIRI at the standard dose, the same dose and schedule as the VELOUR study [4]; (b) low‐dose FOLFIRI, for which the initial dosage included a reduction of at least 15% of one of the components; and (c) a subgroup labeled “other regimens” that included the suppression of the bolus and/or infusion of 5‐fluorouracil. Treatment groups were based on the regimen and dosage of the first cycle, regardless of subsequent dose reductions or discontinuations.

Potential predictors (see supplemental online Table 1) were chosen according to the criteria of being common and easily accessible at the beginning of aflibercept therapy and having demonstrated prognostic impact in at least one previous manuscript [17]. KRAS, NRAS, and BRAF mutational status was analyzed as per each center's clinical practice. The local researchers assessed tumor response using computerized tomographic images taken approximately every 3 months, in line with RECIST version 1.1 criteria [18]. This analysis refers to the maximum response achieved during treatment. Toxicities were categorized and recorded according to the grading scale in place during the treatment period (National Cancer Institute Common Terminology Criteria for Adverse Events version 4.0) every 15 days [19]. The number of hospital admissions because of toxicities was likewise recorded.

Statistical Analysis

Categorical variables were analyzed by means of Pearson's chi‐squared (χ2) and linear‐by‐linear association tests, and continuous variables were assessed using Kruskal‐Wallis tests. Candidate predictors were initially chosen after a review of the literature and consulting with experts. These predictors were significant in univariate analyses. We applied a Cox proportional hazards (PH) regression model for OS, and internal validation was carried out by bootstrapping. Nonlinear effects were modeled by restricted cubic splines. Multiple imputation by fully conditional specification was used in multivariate analyses to deal with missing values [20]. The PH assumption for each predictor was verified using the Schoenfeld residuals test. For discriminatory capacity, 1,000 bootstrap replications served as internal validation subsets to estimate the bias‐corrected c‐index. The 1‐year calibration was gauged visually and by the Gronnesby‐Borgan goodness‐of‐fit test. Survival was estimated via the Kaplan‐Meier method. Stratified log‐rank tests were also applied. All statistical assessments were two‐sided, and p values <.05 were deemed statistically significant. Based on previous investigations, a 250‐patient sample with more than 100 events is considered sufficient to adjust a model of ten preset predictors [21]. This analysis was performed with SPSS version 23.0 software (SPSS Inc., Chicago, IL) and RStudio [22], including rms, mstate, and survival packages [23], [24].

Results

Patients

The sample includes 250 patients consecutively treated between January 2013 and September 2015. Table 1 illustrates participants’ characteristics. Subjects had a median age of 63 years (range 35–82); 44% (n = 111) were aged at least 65 years, and most exhibited good functional status (ECOG PS 0–1 in 96.4%). The most frequent first‐line regimen was CAPOX (capecitabine‐oxaliplatin) or FOLFOX (5‐fluorouracil‐oxaliplatin; n = 234, 93.6%; Table 1), combined with bevacizumab (n = 111, 44.4%) or anti‐EGFR (n = 50, 20%). The presence of KRAS, NRAS, and BRAF mutations was detected in 168 tumors (67.2%). The most common tumor localization was the left colon (39.6%), followed by the rectum (31.6%). The primary tumor was treated surgically in 66% (n = 164); metastases were synchronously diagnosed in 75%. In most patients, aflibercept was used with FOLFIRI in the second line (n = 226, 90.4%), although 7.2% received it in the third or fourth line, and 2.4% in rapid progressors after adjuvancy. Following aflibercept, patients received a median of one further line of treatment (range 0–3). Surgery was performed on metastases in 25 patients (10%), following aflibercept in 10 of them (4%). Median follow‐up in subjects still alive (after initiating aflibercept) was 8.8 months (95% CI, 7.0–10.6).

Table 1. Baseline characteristics of patients at initiation of treatment with aflibercept.

Abbreviations: 5‐FU, 5‐fluorouracil; CAPOX, capecitabine‐oxaliplatin; CEA, carcinoembryonic antigen; CPT‐11, camptothecin‐11; CR, complete response; ECOG PS, Eastern Cooperative Oncology Group performance status; EGFR, epidermal growth factor receptor; FOLFIRI, fluorouracil, leucovorin, irinotecan; FOLFOX, fluorouracil, leucovorin, oxaliplatin; NE, not evaluable; PD, progressive disease; PR, partial response; SD, stable disease.

Aflibercept Dosage and Treatment Schedules

The participants received a median of six cycles of aflibercept (range 1–43), over a median of 14 weeks, with a mean dose per cycle of 3.9 mg/kg. The backbone of chemotherapy initially used with aflibercept consisted of FOLFIRI, the standard regimen, in 46%, low‐dose in 25%, and “other modified schedules” in 28.6% (FOLFIRI without 5‐fluorouracil bolus in 27%, irinotecan in monotherapy 1.2%, and others 0.4%; Table 1; for distribution by centers, supplemental online Table 2). Dosages of 5‐fluorouracil and irinotecan for each stratum are presented in Table 2. Of note is the relatively low mean RDI (<85%) in all subgroups, especially in those who received low‐dose FOLFIRI from the start (64%–67%). In contrast, treatment duration was comparable in the three strata. Aflibercept was discontinued because of toxicity (18%), progression (65%), medical decision (11%), and demise (4%).

Table 2. Fluorouracil and irinotecan dosages used together with aflibercept.

Abbreviations: 5‐FU, 5‐fluorouracil; CI, continuous infusion; FOLFIRI, fluorouracil, leucovorin, irinotecan.

Supplemental online Table 3 displays patients’ baseline characteristics by the most common backbone of chemotherapy. The low‐dose FOLFIRI stratum had the highest median age (66 vs. 63 years in the remaining strata; p = .040). The percentage of ECOG PS 0 was highest in the standard FOLFIRI schedule versus low‐dose or without bolus (54.3%, 31.7%, 31.4%; p = .001). The remaining characteristics examined revealed no significant differences across treatment groups.

Safety

The most common AEs in the entire sample are shown in Table 3. The most prevalent toxicities were asthenia (73%), followed by mucositis (52%), neutropenia (38%), anemia (35%), and hypertension (34%). Most toxicities were mild, although 13.9% of the participants were admitted for grade 3–4 toxicity. Other frequent grade 3–4 toxicities were neutropenia (13%), hypertension (6.8%), asthenia (6.8%), diarrhea (5.4%), and thromboembolic disease (4.2%). Treatment‐related mortality was 1.6% (n = 4): two cases of severe hemorrhage, one gastrointestinal complication, and another febrile neutropenia. AEs stratified by type of FOLFIRI can be seen in Figure 1. Grade 3–4 AEs occurred more frequently with the standard schedule versus low‐dose FOLFIRI or without bolus (40%, 38%, 23%, respectively; linear‐by‐linear test, p = .033). Thus, grade 3–4 neutropenia developed more often in participants on standard FOLFIRI than those receiving low‐dose or without bolus (19.2%, 13.3%, and 4.6%, respectively; p = .018). Differences in grade 3–4 asthenia were likewise observed (36.6%, 24.1%, and 16.9%, respectively; p = .014); any grade bleeding (21.1%, 21.6%, and 4.6%; p = .009); dysphonia (36.6%, 24.1%, and 16.9%; p = .014), and hospitalizations for toxicity (11.9, 24.1%, and 8.6%; p = .039). No significant differences between treatment groups were identified in the remaining variables analyzed. Insofar as age is concerned, individuals aged at least 65 years exhibited more diarrhea, asthenia, and anemia. Moreover, the admission rate because of toxicity was also higher in these subjects (9.7% vs. 19.6%; odds ratio, 2.26; p = .029; supplemental online Fig. 1).

Table 3. Adverse events in the whole cohort (n = 250).

^aTwo cases of severe hemorrhage, one case of gastrointestinal toxicity, and one case of febrile neutropenia.

Figure 1.

Adverse events stratified by FOLFIRI regimen. (A): Standard FOLFIRI. (B): Low‐dose FOLFIRI. (C): FOLFIRI without 5‐FU bolus.

Abbreviations: 5‐FU, 5‐fluorouracil; FOLFIRI, fluorouracil, leucovorin, irinotecan.

Efficacy

A total of 220 patients had measurable and response‐evaluable disease at approximately 3 months of treatment initiation. In these cases, the RECIST 1.1 evaluation at 3 months revealed 0 complete responses, 53 partial responses (24%), 84 stable disease (38%), and 83 tumor progression (38%). No statistical evidence was found, indicating that the dosages and treatment regimens modified the likelihood of response (χ² = 2.055; degrees of freedom [d.f.] = 4; p = .720; supplemental online Fig. 2). At the time of analysis (March 2018), 189 progression events (78%) and 143 deaths (58%) were detected. Patients had a median PFS of 6.05 months (95% CI, 5.16–6.97) and OS of 12.4 months (95% CI, 10.2–14.4). Supplemental online Figure 3 is a survival graph. Figure 2A and B shows Kaplan‐Meier curves stratified by chemotherapy regimen. There was no statistical evidence of variations in PFS based on chemotherapy schedule: median PFS, 6.0 months (95% CI, 4.99–7.20) for the standard regimen; 6.97 months (95% CI, 4.6–8.7) for low‐dose; and 4.6 months (95% CI, 3.29–7.59) for “other regimens” (log‐rank test, χ² = 0.6, d.f. = 2, p = .736). Similarly, no statistically significant differences were found in OS based on these same schedules or treatment groups: 13.5 (95% CI, 10.4–15.8), 12.32 (95% CI, 9.3–32.6), and 8.6 months (95% CI, 5.7–15.8), respectively (log‐rank test, χ² = 4.9; d.f. = 2; p = .087). Subsequent systemic treatment was administered to 47% (n = 117), most often following standard FOLFIRI (53%) versus low‐dose FOLFIRI (33%), and FOLFIRI without bolus (49%; χ² = 6.32; d.f. = 2; p = .042).

Figure 2.

Kaplan‐Meier survival estimates based on modifications to chemotherapy. (A): Progression‐free survival. (B): Overall survival.

Abbreviation: FOLFIRI, fluorouracil, leucovorin, irinotecan.

Development and Internal Validation of a Prognostic Nomogram

After verifying the uniformity of survival endpoints with respect to the pragmatic use of the treatment regimen, a prognostic nomogram was then created. Univariate Cox PH regression models for PFS and OS, including the effect of surgery, are shown in supplemental online Tables 4 and 5. Predictive factors for OS were ECOG PS 1 versus 0 (hazard ratio [HR] 2.58; p < .001), rectal localization (HR 0.52; p = .007), number of metastatic sites (continuous variable; HR 1.28; p = .037), BRAF mutations (HR 6.54; p < .001), complete response to first line (HR 0.10; p = .003), and carcinoembryonic antigen (CEA) levels (135 vs. 5 ng/mL; HR 1.41; p = .038; supplemental online Table 7). Age had no prognostic effect (e.g., in subjects >70 years, HR 1.07; 95% CI, 0.70–1.64). The nomogram is represented in Figure 3, and a web‐based calculator has been designed and is available at http://www.iricom.es/prognostictools/vtrap. The model was well calibrated (Groennesby‐Borgan score test: χ2 = 5.067; p = .535; supplemental online Fig. 4) and has an acceptable discriminatory capacity, with optimism‐corrected c‐index of 0.723 (95% CI, 0.666–0.778). Patients considered high (>164 points), intermediate (128–164 points), and low risk (<128) had a median OS time of 6.1 months (95% CI, 5.1–8.8), 12.4 months (95% CI, 9.36–14.8), and 22.9 months (95% CI, 16.6–not reached), respectively (log‐rank, p < .001; supplemental online Fig. 5). A sensitivity analysis illustrates that the model continues to be robust after excluding subjects who have undergone surgery for metastases. Additionally, the Akaike information criterion (AIC) does not support the inclusion of resection of the primary tumor or of metastases (ΔAIC 2.9 and 6.6, respectively). The multivariate Cox PH regression for PFS is shown in supplemental online Table 6.

Figure 3.

Predictive overall survival nomogram for patients treated with FOLFIRI‐aflibercept.

Abbreviations: CEA, carcinoembryonic antigen; CR, complete response; ECOG PS, Eastern Cooperative Oncology Group performance status; NE, not evaluable; PD, progressive disease; PR, partial response; SD, stable disease.

Discussion

Although several observational studies have corroborated the efficacy and safety data for FOLFIRI plus aflibercept [12], [25], [26], there is evidence that the clinical course of patients treated with second‐line antiangiogenic agents is variable. In this study, we have integrated putative factors that lead to efficacy and safety heterogeneity, developing and internally validating a prognostic nomogram for OS in patients who initiated FOLFIRI and aflibercept in routine conditions of care. Basically, the resulting model exhibited adequate calibration and acceptable discriminatory capacity; therefore, if externally validated by other international groups, it could prove to be valuable to stratify these patients. Regarding the previous literature, this nomogram can integrate the biological and clinical characteristics of the disease (mutations, tumor localization, CEA, ECOG PS, and number of metastatic sites) and identifies the variable of chemosensitivity to the previous line of chemotherapy as having the greatest weight. Two prior analyses revealed ECOG PS and number of metastatic sites to be important prognostic factors in individuals treated with aflibercept [12], [27]. High CEA levels [28], right‐sided colon tumor location [29], and BRAF mutations [30] correlated with poor prognosis in mCRC. Nevertheless, the VELOUR study evinced similar benefit, regardless of whether the primary was left‐ or right‐sided colon, and subjects with mutated BRAF in the placebo group had worse OS, although aflibercept showed superior efficacy in these cases [31]. The resulting nomogram is capable of correctly stratifying patients based on their risk of death and, in particular, correctly identifies individuals with a poor prognosis, with median OS of less than 6 months, who might be eligible to participate in clinical trials to improve outcomes.

Moreover, this analysis of prognostic factors has been expanded with an enriched description of the study population's baseline characteristics and details regarding their treatment at each center (e.g., supplemental online Table 2). We believe that this description is always essential before extrapolating the data from one setting to another. In addition, in actual conditions, potentially confounding variables must be assessed, as well as investigating all the strata of patients who may have a different baseline hazard for one reason or another. Thus, specific risks can be conditioned both by the greater vulnerability in the real world, as well as by pragmatic modifications to the schedules commonly being made by local oncologists, in pursuit of greater safety in groups that are frailer than those treated in trials [10], [12], [16]. In and of itself, this is certainly a valid research objective that should interest the clinician. This series confirms that, indeed, elderly patients are a commonly seen group who are prone to more serious AEs, as previously reported by other authors [11], which appears to justify clinicians’ use of modified regimens to reduce toxicities when confronted with this situation [16]. Therefore, in addition to helping us describe the usual management patterns in these individuals with this treatment alternative and in this specific clinical context, this analysis of routine clinical practice has enabled us to develop a nomogram with the potential to optimize treatment outcomes, as well as being applicable in developing subsequent clinical trials.

As for conditions under which it is possible to generalize, we have found that the characteristics here have been reasonably comparable to those of the VELOUR study, with the exception of the use of more than one prior treatment in 7.2%, greater prior exposure to bevacizumab (44% vs. 30% in VELOUR), as well as a slight increase in KRAS mutations (47% vs. 40% in VELOUR). Nonetheless, the most substantial difference lies in that more than half of the participants in our study received modified FOLFIRI schedules or FOLFIRI at nonstandard dosages schedules from the very beginning. This is consistent with the relatively low RDI (<85%) in all the strata, which comprises one of the model's key conditions for extrapolation. Furthermore, our series includes patients with a slightly older median age compared with the VELOUR trial (63 years, range 35–82, vs. 61 years, range 21–82 years), in addition to 44% of the cases being aged more than 65 years versus 36% in the pivotal trial. In our cohort, 42% and 54% presented with ECOG PS 0 and 1, whereas in VELOUR, these proportions were 57% and 40%, respectively. These two circumstances a priori make our population more similar to what is typically found in the clinical care setting. All in all, survival‐related endpoints in our cohort (median OS and PFS of 13.5 and 6.05 months, respectively) concur with those of the VELOUR study, 13.5 and 6.9 months [4]. We have not observed any differences in these endpoints according to age or type of chemotherapy used. However, as for best overall response, the rate of progressive disease in our registry (38%) was higher than in the experimental arm in VELOUR (10.4%), although in our study, we found no statistical evidence to indicate that the modifications to the schedules were the cause for them being less active. These changes can evidently be attributed to the fact that the evaluation was not centralized. Nonetheless, the most notable differences with the VELOUR study were seen regarding the toxicity profile, both in serious general AEs (e.g., rate of grade 3–4 diarrhea, 5.4% vs. 19.3%; grade 3–4 asthenia, 6.8% vs. 16.8%, in our series vs. in VELOUR, respectively), as well as class‐specific toxicity (e.g., hypertension, 6.8% vs. 19.3%, or grade 3–4 thrombosis, 4.2% vs. 7.8%). Our data reveal fewer AEs with modified schedules, not attributable to different treatment durations. Although the tolerance profile appears to be slightly better here than in the pivotal trial, it must be taken into account that serious toxicity‐related hospitalizations were still greater among elderly patients (≥65 years). This result essentially coincides with those of Ruff et al. who found an increase in the toxicity profile of individuals aged more than 65 years in an age‐based analysis of the VELOUR trial [11]. This group continues to be more complex in second‐line treatment for mCRC [13]. Therefore, much remains to be done in this field.

Insofar as limitations are concerned, the reader must be aware of the relatively small sample size, which precludes the detection of slight differences across strata. The study includes patients who would not have been eligible for VELOUR because of prior therapies, line of treatment, previous combination, etc. Other limitations are inherent to analyzing retrospective data, with the imprecision that typically entails, although grade 3–4 toxicity, tumor progression, and demise are generally robust events that are reliably recorded. Although no differences were observed in the appreciable magnitude in survival‐related endpoints according to stratum after adjusting for confounding factors, the presence of residual bias that could affect the apparent result of the modifications cannot be entirely ruled out.

With these limitations in mind, our data might be applicable in clinical practice for patients deemed candidates to receive FOLFIRI and aflibercept, who present a factor of vulnerability, such as old age, comorbidity, or impaired functional status, which suggests an increased risk of severe toxicity, raising the possibility of using an adapted or modified schedule. The model might be useful for patient counseling, patient stratification for clinical trials, or become an adjuvant tool for prognosis‐sensitive decision making. Nevertheless, the covariates of the model are prognostic factors, and their predictive ability has not been described before [31]; likewise, it cannot be ruled out that a subgroup of patients categorized as having a poor prognosis might benefit from chemotherapy. Furthermore, these data complement the efficacy and safety outcomes of the pivotal clinical trial by evaluating these parameters in a real‐world population and help to clarify the potential role of variables such as age and dose modifications in the final clinical benefit.

Conclusion

We have developed and internally validated a prognostic nomogram that is potentially useful in selecting patients to receive treatment with FOLFIRI and aflibercept within the context of routine clinical practice. Our data also confirm the efficacy and safety of this regimen even with modified schedules and introduce the concept of pragmatically individualized therapy as an attitude that can potentially mitigate serious risks, without substantially compromising efficacy.

See http://www.TheOncologist.com for supplemental material available online.

Author Contributions

Conception/design: Ana Fernández Montes, Carlos López López, Paula Jiménez Fonseca, Alberto Carmona‐Bayonas

Provision of study material or patients: Ana Fernández Montes, Carlos López López, Guillem Argilés Martínez, David Páez López, Ana María López Muñoz, Beatriz García Paredes, David Gutiérrez Abad, Carmen Castañón López, Paula Jiménez Fonseca, Javier Gallego Plazas, María Carmen López Doldán, Eva Martínez de Castro, Manuel Sánchez Cánovas, María Tobeña Puyal, Beatriz Llorente Ayala, Ignacio Juez Martel, Mariana López Flores, Alberto Carmona‐Bayonas

Collection and/or assembly of data: Ana Fernández Montes, Carlos López López, Guillem Argilés Martínez, David Páez López, Ana María López Muñoz, Beatriz García Paredes, David Gutiérrez Abad, Carmen Castañón López, Paula Jiménez Fonseca, Javier Gallego Plazas, María Carmen López Doldán, Eva Martínez de Castro, Manuel Sánchez Cánovas, María Tobeña Puyal, Beatriz Llorente Ayala, Ignacio Juez Martel, Mariana López Flores, Alberto Carmona‐Bayonas

Data analysis and interpretation: Ana Fernández Montes, Carlos López López, Guillem Argilés Martínez, David Páez López, Ana María López Muñoz, Beatriz García Paredes, David Gutiérrez Abad, Carmen Castañón López, Paula Jiménez Fonseca, Javier Gallego Plazas, María Carmen López Doldán, Eva Martínez de Castro, Manuel Sánchez Cánovas, María Tobeña Puyal, Beatriz Llorente Ayala, Ignacio Juez Martel, Mariana López Flores, Alberto Carmona‐Bayonas

Manuscript writing: Ana Fernández Montes, Carlos López López, Paula Jiménez Fonseca, Alberto Carmona‐Bayonas

Final approval of manuscript: Ana Fernández Montes, Carlos López López, Paula Jiménez Fonseca, Alberto Carmona‐Bayonas

Disclosures

Ana Fernández Montes: Sanofi, Celgene, Roche, Servier, Amgen (C/A), Sanofi (RF); Carlos López López: Sanifo, Merck, Roche, Amgen (C/A, RF, SAB); Guillem Argilés Martínez: Bayer, Servier, Sanofi, Bristol‐Myers Squibb, Amgen, Merck, Roche (C/A), Bayer (RF), Bayer, Merck, Sanofi, Roche, Amgen, Servier (SAB); David Páez López: Amgen, Sanofi, Merck Serono, Servier (C/A); Beatriz García Paredes: Sanofi (C/A); Manuel Sánchez Cánovas: Leo Pharma (RF), Kyowakirin (C/A). The other authors indicated no financial relationships.

(C/A) Consulting/advisory relationship; (RF) Research funding; (E) Employment; (ET) Expert testimony; (H) Honoraria received; (OI) Ownership interests; (IP) Intellectual property rights/inventor/patent holder; (SAB) Scientific advisory board