https://orcid.org/0000-0002-5380-3144
Abstract
KRAS and BRAF mutations are well-established predictive and prognostic factors in metastatic colorectal cancer; however, their impact in the adjuvant setting has not yet been established.
We performed a meta-analysis of adjuvant phase III trials in patients with stage II and III colon cancer with available data on the impact of KRAS or BRAF mutations on both disease-free survival (DFS) and overall survival (OS). Trials were subgrouped based on whether adjustment for microsatellite instability (MSI) was performed and the subgroup effect was analyzed through a meta-regression. To increase the precision of the estimates, a joint DFS–OS (so-called “multivariate”) meta-analysis was performed. All statistical tests were 2-sided.
Nine trials were selected (QUASAR 2, PETACC-8, N0147, CALGB-89803, NSABP-C07, NSABP-C08, PETACC-3, QUASAR, MOSAIC) including a total of 10 893 patients. In the primary meta-analysis, KRAS mutation was associated with poor DFS (pooled hazard ratio [HR] = 1.36, 95% confidence interval [CI] = 1.15 to 1.61, P < .001) and OS (pooled HR = 1.27, 95% CI = 1.03 to 1.55, P = .03) and BRAF mutation was also associated with poor DFS (pooled HR = 1.33, 95% CI = 1.00 to 1.78, P = .05) and OS (pooled HR = 1.49, 95% CI = 1.31 to 1.70, P < .001). The effect of the mutations on outcome was enhanced in the MSI-adjusted subgroup for both the KRAS mutation (pooled HR for DFS = 1.43, 95% CI = 1.15 to 1.79, P = .001; and pooled HR for OS = 1.33, 95% CI = 1.03 to 1.71, P = .03) and the BRAF mutation (pooled HR for DFS = 1.59, 95% CI = 1.22 to 2.07, P = .001; and pooled HR for OS = 1.67, 95% CI = 1.37 to 2.04, P < .001). The interaction between BRAF and MSI adjustment was statistically significant for DFS (Pinteraction = .02). This interaction was even more pronounced in the DFS–OS multivariate meta-analysis.
Both KRAS and BRAF mutations were statistically significantly associated with both DFS and OS, with the mutation effect being enhanced by MSI adjustment. Effective adjuvant treatment for microsatellite-stable BRAF or KRAS-mutated colon cancer represents an unmet clinical need, and exploring the use of recently available BRAF and KRAS inhibitors in this setting would be highly desirable.
Colon cancer afflicts over 2 million people worldwide every year. It represents the second leading cause of cancer deaths and ranks third in terms of incidence (1). Approximately 75% of colon cancers are diagnosed as local disease suitable for radical surgery (2). In case of nodal involvement (stage III disease) or in case of node-negative disease but presence of adverse features (“high risk” stage II disease), such as serosal or adjacent organ invasion (T4), histologic grade 3, inadequate lymph node sampling, or obstruction or perforation, adjuvant fluoropyrimidine- or oxaliplatin-based chemotherapy is taken into consideration because it has demonstrated a statistically significant reduction in the risk of relapse and mortality (3). However, the absolute gain owing to adjuvant chemotherapy seems to not exceed 5% and 20% in stage II and III tumors, respectively (4).
Recent research has focused on how to optimize regimen and duration of adjuvant chemotherapy, especially looking at the reduction of oxaliplatin toxicity by shortening its exposure from 6 to 3 months while preserving its efficacy (5,6). The identification of patients at higher risk of relapse is fundamental to decide for whom a long course of treatment (6 months) and/or a doublet chemotherapy is necessary at the cost of increased toxicity (7) and could also provide insights for development of novel treatment strategies (8).
KRAS and BRAF genes have been extensively studied in the metastatic colorectal cancer setting. KRAS mutations are well-established negative predictive factors for the efficacy of a specific class of drugs (anti-EGFR drug resistance markers) (9,10). Similarly, the BRAF mutation seems to confer lack of benefit from anti-EGFR agents (11). It has been widely documented that especially BRAF mutation, but also KRAS mutations, are associated with poor prognosis in advanced disease (12-14).
BRAF mutations are frequently associated with the so-called “immune” subtype, which is typically characterized by the microsatellite instability-high (MSI-H) or deficient mismatch repair status (15). Approximately three-quarters of MSI-H cases are sporadic and dependent on epigenetic silencing of MMR (mismatch repair) genes and, in 35%-45% of cases , are associated with BRAF mutations (16,17). On the other hand, prevalence of MSI-H increases from approximately 5% in BRAF wild-type tumors to approximately 15% in BRAF-mutated tumors (18).
In localized disease, BRAF mutations in the context of MSI-H vs microsatellite-stable (MSS) tumors are thought to be distinct biological entities with different prognoses and treatment sensitivities. Resected MSI-H tumors seem to have a more favorable prognosis than MSS tumors regardless of the BRAF status (19). Moreover, some post hoc analyses of adjuvant trials have demonstrated in MSI-H tumors no benefit of 5-fluouracil monotherapy compared with surgery alone (20), whereas the adjuvant benefit of the fluoropimidine–oxaliplatin doublet seems to be maintained (21).
A number of adjuvant trials have been retrospectively analyzed for the possible impact of tissue KRAS and/or BRAF mutations on disease-free survival (DFS) and overall survival (OS) of stage II and III colon cancer, and for some of them adjustment for MSI status has been included in the analysis. A large joined analysis of 2 trials, PETACC-8 and N0147, demonstrated a consistent adverse prognostic effect of both KRAS and BRAF mutations in terms of time to recurrence, survival after recurrence, and OS in MSS tumors, with hazard ratios (HR) ranging from 1.2 to 3.0 (22). In another large cohort analyzed by The Adjuvant Colon Cancer End Points Collaborative Group (7 adjuvant studies, 2630 patients), BRAF mutations were found to be associated with shorter survival after recurrence in both the MSI-H and MSS context, with hazard ratios greater than 2.0 (23). Recently, a clear negative impact of KRAS and BRAF mutations on DFS was also demonstrated in a retrospective analysis of the QUick And Simple And Reliable 2 (QUASAR 2) study (24). However, less consistent results were provided in other trials, especially in terms of OS (25,26).
The combined analysis of KRAS and BRAF mutations and MSI status could therefore be of particular value also in the adjuvant setting. In localized MSS tumors, KRAS and BRAF mutations might have a detrimental effect that would reinforce the need of genotype-oriented trial design, especially in light of the recent availability of specific KRAS/BRAF inhibitors (eg, sotorasib and encorafenib) and of other drugs targeting the EGFR-MAP -kinase pathway axis (27,28).
We performed a meta-analysis of available data to overcome the uncertainty and solve discordant reports on the role of these mutations in localized disease setting.
Methods
Research Question
The PICOS (Population, Intervention, Comparison, Outcomes and Study) method was used to define the research question as follows: Population, stage II and III colon cancer patients; Intervention (investigational group), presence of KRAS or BRAF mutation; Control, KRAS/BRAF wild-type status; Outcomes, DFS and OS; and Study Design, post hoc analysis of phase III randomized trials (29). The research project was registered in the Open Science Framework website (https://osf.io/67b5w/), and the Preferred Reporting Items for Systematic Reviews and Meta-analyses guidelines were followed (see Supplementary Methods, available online) (30).
Literature Search
The electronic PubMed database was systematically searched from January 1, 2010, to February 14, 2021, for relevant literature. The following syntax was used for the search: ([braf] OR [kras] OR [ras]) AND ([stage II] OR [stage III]) AND ([colon cancer] OR [colorectal cancer]). The search was performed independently by 2 researchers (V.F. and C.M.). In addition, manual examination of the reference list of included articles was performed.
Eligibility Criteria
Inclusion criteria were: 1) full English report on phase III randomized trials of adjuvant treatment in resected colon cancer, 2) stage II or III colon cancer, and 3) reports on the effect of KRAS and/or BRAF mutation on DFS and/or OS. We decided to select post hoc analyses of phase III adjuvant trials and exclude observational cohort studies to guarantee the highest possible quality of evidence and the highest technical standards for mutational analysis and minimize patient selection bias and adjuvant treatment bias.
The main exclusion criterion was absence of hazard ratio of mutated vs wild-type tumors for DFS or OS. We set 2 coprimary outcome measures, DFS and OS, to allow for a joint mixed-effects multivariate (bivariate) meta-analysis, which would also compensate for possible missingness of either outcome (31). When more than 1 article referring to the same trial was found, the most up-to-date and complete report was selected.
Study Selection
Records retrieved via PubMed search were imported in a tabular format in a CSV (comma-separated values) file (inclusive of titles and abstracts) and screened for potential inclusion by 3 investigators (V.F., M.C., S.R.) independently, in parallel. Full texts of articles deemed candidate for the meta-analysis were obtained and assessed for the final inclusion by the 3 investigators (V.F., M.C., S.R.) independently, in parallel. Discrepancies were resolved on discussion or by a fourth supervisor investigator (M.R.).
Data Extraction
Hazard ratio and confidence interval (CI) data of RAS/BRAF mutated vs wild-type subgroup for OS and DFS were extracted by 2 investigators (V.F and C.M.) independently, in duplicate. In included trials, OS was defined as the time from treatment allocation to death due to any cause or last follow-up. DFS was defined as the time from treatment allocation to cancer relapse or death or last follow-up. If time to treatment failure (that also includes treatment failure because of toxicity or other causes as an endpoint) was estimated instead of DFS, the former would be considered equivalent to DFS for the sake of simplicity. Relapse-free survival was considered as a synonym of DFS.
The following data were also collected: first author, trial name, year of publication, publishing journal, treatment options of the randomization design, number of mutated and wild-type cases, relative percentage of stage II and III tumors, type of Cox-regression model for hazard ratio estimation (univariate vs multivariate), and covariates included and/or adjusted for in case of multivariable models. Discrepancies in data collection were resolved appropriately. All data were extracted by means of a shared preestablished form and then transferred into an Excel spreadsheet.
The Newcastle-Ottawa Scale for cohort studies was used to evaluate risk of bias of included studies (32). A scale for cohort studies was used because the mutational status was not a matter of randomization, and the effect of mutations on survival was always a post hoc analysis with the whole study population analyzed independently of the treatment received.
Statistical Analysis
Hazard ratios were log-back transformed and respective standard errors were calculated from confidence intervals with the following formula: (ln [upper CI] – ln [lower CI]) /3.919928. The random-effects model with the restricted maximum likelihood method was used to obtain the pooled hazard ratio estimates and respective confidence intervals. Results were graphically presented with forest plots. The random-effects model was preferred over the fixed-effects model because heterogeneity was anticipated mainly because of the variety in the adjuvant treatment administered and in the proportion of stage II vs III tumors across the studies (33). Heterogeneity was assessed using the I2 and Cochran’s Q statistics. I2 greater than 50% was considered as substantial heterogeneity. Statistical significance of the summary estimates was based on Wald test P value.
Given the possible interaction between mutations and MSI in terms of survival effect, a meta-regression was performed by stratifying the studies according to whether the study hazard ratio was estimated by adjusting for MSI status (34). Within the meta-regression analysis, the “interaction hazard ratio” (ie, the exponent of the regression coefficient) and the interaction P value (Pinteraction) were calculated for the change of the pooled estimate across the 2 strata. A study was labeled as “adjusting for MSI” when either the mutation effect was assessed in the subgroup of MSS patients or when MSI status was included in a multivariable Cox-regression model. As mentioned above, the pooled hazard ratios for DFS and OS, and their interaction with MSI adjustment, were also jointly assessed for each mutation by means of a so-called “multivariate random-effects meta-analysis” and “multivariate random-effects meta-regression.” All multivariate estimates were obtained with the restricted maximum likelihood method (35). Bias of small study effect was assessed with the use of funnel plots.
All tests were performed with packages “meta” and “rvmeta,” “mixmeta” of R software, version 3.6.3 (R Foundation for Statistical Computing). A 2-sided P value less than .05 was considered statistically significant.
Results
Study Selection
The initial search retrieved 494 articles, which were thoroughly reviewed for entry criteria (Figure 1). Nine trials were finally identified that met the inclusion criteria—QUASAR 2 (24), PETACC-8 (36), MOSAIC (26), N0147 (37), PETACC-3 (38), NSABP C-07 and NSABP C-08 (39), CALGB 89803 (25), QUASAR (40)—for a total of 10 893 patients (Table 1; Supplementary Table 1, available online). Overall, 3539 KRAS mutated and 1226 BRAF mutated patients were analyzed.
Study selection flow diagram. DFS = disease-free survival; OS = overall survival; RCT = randomized controlled trial.
Features of selected studies
| Reference | Trial name | Adjuvant treatment randomization | % of Stage III:II | No. of KRAS mutated patients | No. of BRAF mutated patients | Total analyzed patients (% from original population) | Primary result of study | Type of HR | MSI adjustmenta |
|---|---|---|---|---|---|---|---|---|---|
| Domingo et al., 2018 (24) | QUASAR 2 | Capecitabine vs capecitabine+bevacizumab | 60:40 | 203 | 79 | 511 (26) | Statistically significant multigene prognostic model including KRAS/BRAF (P = .00004) | multivariate | Yes |
| Taieb et al., 2016 (36) | PETACC-8 | FOLFOX vs FOLFOX+cetuximab | 100:0 | 558 | 94 | 1600 (63) | Shorter DFS and OS for mKRAS (HR = 1.64 and 1.71, P < .001 and .002) and mBRAF (HR = 1.74 and 1.84, P = .01 and .04), respectively, in MSS tumors | multivariate | Yes |
| André et al., 2015 (26) | MOSAIC | LV5FU2 vs FOLFOX | 64:36 | Not available | 94 | 902 (40) | FOLFOX superiority confirmed at 10 y (HR for OS = 0.85, P = .04) | Univariate | No |
| Sinicrope et al., 2015 (37) | N0147 | FOLFOX vs FOLFOX+cetuximab | 100:0 | 758 | 332 | 2193 (73) | Poor OS for mKRAS in distal cancers (HR = 1.98, P < .0001) | multivariate | Yes |
| Roth et al., 2012 (38) | PETACC-3 | LV5FU2 vs FOLFIRI | 70:30 | 515 | 112 | 1404 (67) | Better OS for MSI-high (HR = 0.43, P = .001) | multivariate | Yes |
| Gavin et al., 2012 (39) |
|
| 72:28 | 793 | 316 | 2226 (43) | Poor SAR for mBRAF (HR = 2.31, P < .0001) | Univariate | No |
| Ogino et al., 2012 (25) | CALGB 89803 | 5FU/LV vs IFL | 100:0 | 176 | 75 | 506 (40) | Inferior OS for mBRAF (HR = 1.66, P = .01) | multivariate | Yes |
| Hutchins et al., 2011 (40) | QUASAR | Observation vs 5FU/LV | 10:90 | 536 | 124 | 1551 (48) | Lower recurrence for MSI (RR = 0.53, P < .001) | Univariate | No |
| Reference | Trial name | Adjuvant treatment randomization | % of Stage III:II | No. of KRAS mutated patients | No. of BRAF mutated patients | Total analyzed patients (% from original population) | Primary result of study | Type of HR | MSI adjustmenta |
|---|---|---|---|---|---|---|---|---|---|
| Domingo et al., 2018 (24) | QUASAR 2 | Capecitabine vs capecitabine+bevacizumab | 60:40 | 203 | 79 | 511 (26) | Statistically significant multigene prognostic model including KRAS/BRAF (P = .00004) | multivariate | Yes |
| Taieb et al., 2016 (36) | PETACC-8 | FOLFOX vs FOLFOX+cetuximab | 100:0 | 558 | 94 | 1600 (63) | Shorter DFS and OS for mKRAS (HR = 1.64 and 1.71, P < .001 and .002) and mBRAF (HR = 1.74 and 1.84, P = .01 and .04), respectively, in MSS tumors | multivariate | Yes |
| André et al., 2015 (26) | MOSAIC | LV5FU2 vs FOLFOX | 64:36 | Not available | 94 | 902 (40) | FOLFOX superiority confirmed at 10 y (HR for OS = 0.85, P = .04) | Univariate | No |
| Sinicrope et al., 2015 (37) | N0147 | FOLFOX vs FOLFOX+cetuximab | 100:0 | 758 | 332 | 2193 (73) | Poor OS for mKRAS in distal cancers (HR = 1.98, P < .0001) | multivariate | Yes |
| Roth et al., 2012 (38) | PETACC-3 | LV5FU2 vs FOLFIRI | 70:30 | 515 | 112 | 1404 (67) | Better OS for MSI-high (HR = 0.43, P = .001) | multivariate | Yes |
| Gavin et al., 2012 (39) |
|
| 72:28 | 793 | 316 | 2226 (43) | Poor SAR for mBRAF (HR = 2.31, P < .0001) | Univariate | No |
| Ogino et al., 2012 (25) | CALGB 89803 | 5FU/LV vs IFL | 100:0 | 176 | 75 | 506 (40) | Inferior OS for mBRAF (HR = 1.66, P = .01) | multivariate | Yes |
| Hutchins et al., 2011 (40) | QUASAR | Observation vs 5FU/LV | 10:90 | 536 | 124 | 1551 (48) | Lower recurrence for MSI (RR = 0.53, P < .001) | Univariate | No |
A study was labeled as “adjusting for microsatellite instability (MSI)” when either the mutation effect was assessed in the subgroup of microsatellite-stable (MSS) patients or when MSI status was included in a multivariable Cox-regression model. 5FU/LV = bolus 5-fluorouracil + leucovorin; FOLFOX = 5FU/LV + oxaliplatin; DFS = disease-free survival; HR = hazard ratio; IFL = irinotecan + bolus 5-fluorouracil + leucovorin; LV5FU2 = the so-called de Gramont regimen; mBRAF = mutant BRAF; mKRAS = mutant KRAS; OS = overall survival; RR = relative risk; SAR = survival after recurrence. QUASAR trial = QUick And Simple And Reliable trial; PETACC = Pan-European Trials in Alimentary traCt Cancer; MOSAIC = Multicenter International Study of Oxaliplatin/Fluorouracil/Leucovorin in the Adjuvant Treatment of Colon Cancer; N0147 = North Central Cancer Treatment Group 0147 trial; NSABP = National Surgical Adjuvant Breast and Bowel Project; CALGB = Cancer and Leukemia Group B
Features of selected studies
| Reference | Trial name | Adjuvant treatment randomization | % of Stage III:II | No. of KRAS mutated patients | No. of BRAF mutated patients | Total analyzed patients (% from original population) | Primary result of study | Type of HR | MSI adjustmenta |
|---|---|---|---|---|---|---|---|---|---|
| Domingo et al., 2018 (24) | QUASAR 2 | Capecitabine vs capecitabine+bevacizumab | 60:40 | 203 | 79 | 511 (26) | Statistically significant multigene prognostic model including KRAS/BRAF (P = .00004) | multivariate | Yes |
| Taieb et al., 2016 (36) | PETACC-8 | FOLFOX vs FOLFOX+cetuximab | 100:0 | 558 | 94 | 1600 (63) | Shorter DFS and OS for mKRAS (HR = 1.64 and 1.71, P < .001 and .002) and mBRAF (HR = 1.74 and 1.84, P = .01 and .04), respectively, in MSS tumors | multivariate | Yes |
| André et al., 2015 (26) | MOSAIC | LV5FU2 vs FOLFOX | 64:36 | Not available | 94 | 902 (40) | FOLFOX superiority confirmed at 10 y (HR for OS = 0.85, P = .04) | Univariate | No |
| Sinicrope et al., 2015 (37) | N0147 | FOLFOX vs FOLFOX+cetuximab | 100:0 | 758 | 332 | 2193 (73) | Poor OS for mKRAS in distal cancers (HR = 1.98, P < .0001) | multivariate | Yes |
| Roth et al., 2012 (38) | PETACC-3 | LV5FU2 vs FOLFIRI | 70:30 | 515 | 112 | 1404 (67) | Better OS for MSI-high (HR = 0.43, P = .001) | multivariate | Yes |
| Gavin et al., 2012 (39) |
|
| 72:28 | 793 | 316 | 2226 (43) | Poor SAR for mBRAF (HR = 2.31, P < .0001) | Univariate | No |
| Ogino et al., 2012 (25) | CALGB 89803 | 5FU/LV vs IFL | 100:0 | 176 | 75 | 506 (40) | Inferior OS for mBRAF (HR = 1.66, P = .01) | multivariate | Yes |
| Hutchins et al., 2011 (40) | QUASAR | Observation vs 5FU/LV | 10:90 | 536 | 124 | 1551 (48) | Lower recurrence for MSI (RR = 0.53, P < .001) | Univariate | No |
| Reference | Trial name | Adjuvant treatment randomization | % of Stage III:II | No. of KRAS mutated patients | No. of BRAF mutated patients | Total analyzed patients (% from original population) | Primary result of study | Type of HR | MSI adjustmenta |
|---|---|---|---|---|---|---|---|---|---|
| Domingo et al., 2018 (24) | QUASAR 2 | Capecitabine vs capecitabine+bevacizumab | 60:40 | 203 | 79 | 511 (26) | Statistically significant multigene prognostic model including KRAS/BRAF (P = .00004) | multivariate | Yes |
| Taieb et al., 2016 (36) | PETACC-8 | FOLFOX vs FOLFOX+cetuximab | 100:0 | 558 | 94 | 1600 (63) | Shorter DFS and OS for mKRAS (HR = 1.64 and 1.71, P < .001 and .002) and mBRAF (HR = 1.74 and 1.84, P = .01 and .04), respectively, in MSS tumors | multivariate | Yes |
| André et al., 2015 (26) | MOSAIC | LV5FU2 vs FOLFOX | 64:36 | Not available | 94 | 902 (40) | FOLFOX superiority confirmed at 10 y (HR for OS = 0.85, P = .04) | Univariate | No |
| Sinicrope et al., 2015 (37) | N0147 | FOLFOX vs FOLFOX+cetuximab | 100:0 | 758 | 332 | 2193 (73) | Poor OS for mKRAS in distal cancers (HR = 1.98, P < .0001) | multivariate | Yes |
| Roth et al., 2012 (38) | PETACC-3 | LV5FU2 vs FOLFIRI | 70:30 | 515 | 112 | 1404 (67) | Better OS for MSI-high (HR = 0.43, P = .001) | multivariate | Yes |
| Gavin et al., 2012 (39) |
|
| 72:28 | 793 | 316 | 2226 (43) | Poor SAR for mBRAF (HR = 2.31, P < .0001) | Univariate | No |
| Ogino et al., 2012 (25) | CALGB 89803 | 5FU/LV vs IFL | 100:0 | 176 | 75 | 506 (40) | Inferior OS for mBRAF (HR = 1.66, P = .01) | multivariate | Yes |
| Hutchins et al., 2011 (40) | QUASAR | Observation vs 5FU/LV | 10:90 | 536 | 124 | 1551 (48) | Lower recurrence for MSI (RR = 0.53, P < .001) | Univariate | No |
A study was labeled as “adjusting for microsatellite instability (MSI)” when either the mutation effect was assessed in the subgroup of microsatellite-stable (MSS) patients or when MSI status was included in a multivariable Cox-regression model. 5FU/LV = bolus 5-fluorouracil + leucovorin; FOLFOX = 5FU/LV + oxaliplatin; DFS = disease-free survival; HR = hazard ratio; IFL = irinotecan + bolus 5-fluorouracil + leucovorin; LV5FU2 = the so-called de Gramont regimen; mBRAF = mutant BRAF; mKRAS = mutant KRAS; OS = overall survival; RR = relative risk; SAR = survival after recurrence. QUASAR trial = QUick And Simple And Reliable trial; PETACC = Pan-European Trials in Alimentary traCt Cancer; MOSAIC = Multicenter International Study of Oxaliplatin/Fluorouracil/Leucovorin in the Adjuvant Treatment of Colon Cancer; N0147 = North Central Cancer Treatment Group 0147 trial; NSABP = National Surgical Adjuvant Breast and Bowel Project; CALGB = Cancer and Leukemia Group B
Effect of KRAS and BRAF Mutations on DFS and OS in the Adjuvant Setting
In the primary meta-analysis, 8 studies were included for the DFS analysis of both KRAS and BRAF mutations. Seven studies were included for the OS analysis of BRAF mutation. Six studies were included for the OS analysis of KRAS mutation (Figure 1). KRAS mutations were associated with a statistically significant deterioration of both DFS and OS (pooled HR for DFS = 1.36, 95% CI = 1.15 to 1.61, P < .001; and pooled HR for OS = 1.27, 95% CI = 1.03 to 1.55, P = .03) (Figure 2). However, study heterogeneity of approximately 70% (I2 score) was observed in both analyses.
Primary meta-analysis on the impact of KRAS and BRAF mutation on disease-free survival (DFS) and overall survival (OS). A) effect of KRAS on DFS; B) effect of KRAS on OS; C) effect of BRAF on DFS; D) effect of BRAF on OS. Random-effects model with the restricted maximum likelihood method was used to obtain the pooled hazard ratio (HR) estimates. Statistical significance of the pooled estimates was based on 2-sided Wald test P value. Heterogeneity was quantified by the estimated between-study variance τ2 and I2 statistics and tested using Cochran’s Q statistic. Percentage weight given to individual studies is also presented. Error bars indicate the 95% confidence intervals (CIs). mut = mutation; TE = treatment/analyzed variable effect; seTE = standard error of TE; wt = wild type.
BRAF mutation was statistically significantly associated with deterioration of OS with a pooled hazard ratio of 1.49 (95% CI = 1.31 to 1.70, P < .001) and no statistically significant study heterogeneity (I2 = 10.2%, Q P value = .35) (Figure 2). BRAF mutation was also associated with inferior DFS, but this was not statistically significant with a pooled hazard ratio of 1.33 (95% CI = 1.00 to 1.78, P = .05). A high degree of study heterogeneity was observed for this analysis (I2 = 72.8%, P < .001).
The DFS–OS joint multivariate (bivariate) meta-analysis confirmed the results of the primary meta-analysis with a statistically significant impact of KRAS mutations on DFS (HR = 1.36, 95% CI = 1.14 to 1.62, P < .001) and OS (HR = 1.34, 95% CI = 1.09 to 1.64, P = .005) and of BRAF mutation on OS (HR = 1.55, 95% CI = 1.27 to 1.88, P < .001) (Table 2). In the multivariate meta-analysis, the apparent absence of association between BRAF and DFS was more evident (P = .14) (Table 2). Study heterogeneity was confirmed to be high in both DFS and OS subanalyses (I2 = 69.3% and 61.7%, respectively).
Multivariate DFS–OS analysis of the impact of KRAS and BRAF mutationa
| Mutation and outcome | HR (95% CI) | Wald test P value | I2, % | Q test P value |
|---|---|---|---|---|
| KRAS | ||||
| DFS | 1.36 (1.14 to 1.62) | <.001 | 69.3 | <.001 |
| OS | 1.34 (1.09 to 1.64) | .005 | ||
| BRAF | ||||
| DFS | 1.26 (0.93 to 1.70) | .14 | 61.7 | .002 |
| OS | 1.55 (1.27 to 1.88) | <.001 |
| Mutation and outcome | HR (95% CI) | Wald test P value | I2, % | Q test P value |
|---|---|---|---|---|
| KRAS | ||||
| DFS | 1.36 (1.14 to 1.62) | <.001 | 69.3 | <.001 |
| OS | 1.34 (1.09 to 1.64) | .005 | ||
| BRAF | ||||
| DFS | 1.26 (0.93 to 1.70) | .14 | 61.7 | .002 |
| OS | 1.55 (1.27 to 1.88) | <.001 |
All statistical tests were 2-sided. CI = confidence interval; DFS = disease-free survival; HR = hazard ratio; OS = overall survival.
Multivariate DFS–OS analysis of the impact of KRAS and BRAF mutationa
| Mutation and outcome | HR (95% CI) | Wald test P value | I2, % | Q test P value |
|---|---|---|---|---|
| KRAS | ||||
| DFS | 1.36 (1.14 to 1.62) | <.001 | 69.3 | <.001 |
| OS | 1.34 (1.09 to 1.64) | .005 | ||
| BRAF | ||||
| DFS | 1.26 (0.93 to 1.70) | .14 | 61.7 | .002 |
| OS | 1.55 (1.27 to 1.88) | <.001 |
| Mutation and outcome | HR (95% CI) | Wald test P value | I2, % | Q test P value |
|---|---|---|---|---|
| KRAS | ||||
| DFS | 1.36 (1.14 to 1.62) | <.001 | 69.3 | <.001 |
| OS | 1.34 (1.09 to 1.64) | .005 | ||
| BRAF | ||||
| DFS | 1.26 (0.93 to 1.70) | .14 | 61.7 | .002 |
| OS | 1.55 (1.27 to 1.88) | <.001 |
All statistical tests were 2-sided. CI = confidence interval; DFS = disease-free survival; HR = hazard ratio; OS = overall survival.
Meta-Regression Analysis
To find out possible determinants of the heterogeneity observed in the primary meta-analysis, a meta-regression was performed taking into account the factor more likely to influence the impact of mutations on survival, that is, the MSI status. With this aim, we categorized the hazard ratios from individual studies based on whether adjustment for the presence of MSI-high status was applied in the original report. DFS and OS hazard ratios were found to be adjusted for MSI in the QUASAR 2, PETACC-3, PETACC-8, CALGB-89803, and N0147 trials. Hazard ratios were not adjusted for MSI in the MOSAIC, NSABP C-07, NSABP C-08, and QUASAR trials.
The meta-regression demonstrated that the negative effect of mutations on survival was enhanced when adjusting for MSI. In the subgroup of trials adjusting for MSI status, the pooled hazard ratios increased to 1.43 (95% CI = 1.15 to 1.79, P = .001), 1.33 (95% CI = 1.03 to 1.71, P = .03), 1.59 (95% CI = 1.22 to 2.07, P = .001), and 1.67 (95% CI = 1.37 to 2.04, P < .001) for the effect of KRAS on DFS and OS and BRAF on DFS and OS, respectively (Figures 3 and 4). No statistically significant impact of KRAS on DFS and OS (Figure 3) and of BRAF on DFS (Figure 4) could be demonstrated if no adjustment for MSI was made in the trials: pooled hazard ratios = 1.25 (95% CI = 0.93 to 1.67, P = .13), 1.09 (95% CI = 0.70 to 1.69, P = .70), and 0.94 (95% CI = 0.66 to 1.33, P = .71), respectively. A negative impact of BRAF mutation on OS was also maintained in the subgroup of trials not adjusting for MSI (pooled HR = 1.34, 95% CI = 1.11 to 1.62, P < .001); however, only 3 trials (MOSAIC, NSABP C-07, NSABP C-08) were included in this subgroup (Figure 4).
Meta-regression by microsatellite instability (MSI) adjustment of the effect of KRAS mutation on (A) disease-free survival (DFS) and (B) overall survival (OS). Random-effects model with the restricted maximum likelihood method was used to obtain the pooled hazard ratio (HR) estimates. Statistical significance of the pooled estimates was based on 2-sided Wald test P value. Heterogeneity was quantified by the estimated between-study variance τ2 and I2 statistics and tested using Cochran’s Q statistic. Percentage weight given to individual studies is also presented. Error bars indicate the 95% confidence intervals (CIs). mut = mutation; TE = treatment/analyzed variable effect; seTE = standard error of TE ; wt = wild type. In light gray the test for heterogeneity is reported.
Meta-regression by microsatellite instability (MSI) adjustment of the effect of BRAF mutation on (A) disease-free survival (DFS) and (B) overall survival (OS). Random-effects model with the restricted maximum likelihood method was used to obtain the pooled hazard ratio (HR) estimates. Statistical significance of the pooled estimates was based on 2-sided Wald test P value. Heterogeneity was quantified by the estimated between-study variance τ2 and I2 statistics and tested using Cochran’s Q statistic. Percentage weight given to individual studies is also presented. Error bars indicate the 95% confidence intervals (CIs). mut = mutation; TE = treatment/analyzed variable effect; seTE = standard error of the TE; wt = wild type. In light gray the test for heterogeneity is reported.
The interaction between MSI adjustment and effect of the mutations on survival was particularly relevant for the DFS analysis of BRAF, with an interaction hazard ratio of 1.70 (95% CI = 1.09 to 2.64) and a statistically significant Pinteraction of .02 (Figure 4). The multivariate DFS–OS meta-regression confirmed the results of the primary meta-regression, yielding similar hazard ratios and P values (Table 3). Moreover, it demonstrated that stratification by MSI adjustment could explain most of the study heterogeneity observed in the primary meta-analysis for the effect of BRAF on DFS and OS. The residual heterogeneity after MSI adjustment stratification was in fact not statistically significant in the multivariate meta-regression (I2 = 37.7%, Q test P value = .10) (Table 3). Moreover, the interaction between BRAF mutation and MSI adjustment for the effect on survival was more pronounced in the multivariate meta-regression (interaction HR for DFS = 1.81, 95% CI = 1.17 to 2.81, Pinteraction = .008; and interaction HR for OS = 1.37, 95% CI = 0.98 to 1.91, Pinteraction = .07).
Multivariate DFS–OS meta-regression by MSI adjustment of the impact of KRAS and BRAF mutation
| Mutation and outcome | MSI adjustment | HR (95% CI) | Interaction HR (95% CI) | Pinteractiona | I2, % | Q test P valueb |
|---|---|---|---|---|---|---|
| KRAS | ||||||
| DFS | No | 1.25 (0.92 to 1.69) | 1.14 (0.78 to 1.67) | .50 | 68.8 | <.001 |
| Yes | 1.42 (1.13 to 1.79) | |||||
| OS | No | 1.20 (0.83 to 1.74) | 1.18 (0.75 to 1.85) | .47 | ||
| Yes | 1.42 (1.09 to 1.84) | |||||
| BRAF | ||||||
| DFS | No | 0.88 (0.62 to 1.24) | 1.81 (1.17 to 2.81) | .008 | 37.7 | .10 |
| Yes | 1.59 (1.21 to 2.08) | |||||
| OS | No | 1.28 (1.00 to 1.64) | 1.37 (0.98 to 1.91) | .07 | ||
| Yes | 1.75 (1.39 to 2.21) |
| Mutation and outcome | MSI adjustment | HR (95% CI) | Interaction HR (95% CI) | Pinteractiona | I2, % | Q test P valueb |
|---|---|---|---|---|---|---|
| KRAS | ||||||
| DFS | No | 1.25 (0.92 to 1.69) | 1.14 (0.78 to 1.67) | .50 | 68.8 | <.001 |
| Yes | 1.42 (1.13 to 1.79) | |||||
| OS | No | 1.20 (0.83 to 1.74) | 1.18 (0.75 to 1.85) | .47 | ||
| Yes | 1.42 (1.09 to 1.84) | |||||
| BRAF | ||||||
| DFS | No | 0.88 (0.62 to 1.24) | 1.81 (1.17 to 2.81) | .008 | 37.7 | .10 |
| Yes | 1.59 (1.21 to 2.08) | |||||
| OS | No | 1.28 (1.00 to 1.64) | 1.37 (0.98 to 1.91) | .07 | ||
| Yes | 1.75 (1.39 to 2.21) |
2-sided Wald test P value. CI = confidence interval; DFS = disease-free survival; HR = hazard ratio; MSI = microsatellite instability; OS = overall survival.
All tests were 2-sided.
Multivariate DFS–OS meta-regression by MSI adjustment of the impact of KRAS and BRAF mutation
| Mutation and outcome | MSI adjustment | HR (95% CI) | Interaction HR (95% CI) | Pinteractiona | I2, % | Q test P valueb |
|---|---|---|---|---|---|---|
| KRAS | ||||||
| DFS | No | 1.25 (0.92 to 1.69) | 1.14 (0.78 to 1.67) | .50 | 68.8 | <.001 |
| Yes | 1.42 (1.13 to 1.79) | |||||
| OS | No | 1.20 (0.83 to 1.74) | 1.18 (0.75 to 1.85) | .47 | ||
| Yes | 1.42 (1.09 to 1.84) | |||||
| BRAF | ||||||
| DFS | No | 0.88 (0.62 to 1.24) | 1.81 (1.17 to 2.81) | .008 | 37.7 | .10 |
| Yes | 1.59 (1.21 to 2.08) | |||||
| OS | No | 1.28 (1.00 to 1.64) | 1.37 (0.98 to 1.91) | .07 | ||
| Yes | 1.75 (1.39 to 2.21) |
| Mutation and outcome | MSI adjustment | HR (95% CI) | Interaction HR (95% CI) | Pinteractiona | I2, % | Q test P valueb |
|---|---|---|---|---|---|---|
| KRAS | ||||||
| DFS | No | 1.25 (0.92 to 1.69) | 1.14 (0.78 to 1.67) | .50 | 68.8 | <.001 |
| Yes | 1.42 (1.13 to 1.79) | |||||
| OS | No | 1.20 (0.83 to 1.74) | 1.18 (0.75 to 1.85) | .47 | ||
| Yes | 1.42 (1.09 to 1.84) | |||||
| BRAF | ||||||
| DFS | No | 0.88 (0.62 to 1.24) | 1.81 (1.17 to 2.81) | .008 | 37.7 | .10 |
| Yes | 1.59 (1.21 to 2.08) | |||||
| OS | No | 1.28 (1.00 to 1.64) | 1.37 (0.98 to 1.91) | .07 | ||
| Yes | 1.75 (1.39 to 2.21) |
2-sided Wald test P value. CI = confidence interval; DFS = disease-free survival; HR = hazard ratio; MSI = microsatellite instability; OS = overall survival.
All tests were 2-sided.
Publication Bias Analysis
Funnel plots are presented to inspect for publication bias (Figure 5). The highest risk of publication bias was observed for OS analysis of BRAF. The Newcastle-Ottawa Scale for selected studies is presented in Supplementary Table 2 (available online).
Funnel plots to inspect for publication bias. DFS = disease-free survival; HR = hazard ratio; OS = overall survival.
Discussion
Single molecular markers have so far been proven fruitless to stratify prognosis in the adjuvant setting of colorectal cancer, with the only exception probably being the lower recurrence rate associated with MSI-high status in stage II tumors (41). Risk estimation in daily practice is currently based in most cases on clinical predictors (such as nodal involvement, histologic grade, clinical presentation). Identification of mechanistic molecular prognostic factors would therefore be highly desirable (42).
KRAS and BRAF mutations have repeatedly been shown to associate with poor survival in metastatic patients; however, their impact in the early stages has not been conclusively confirmed (43). To increase the chance of high-quality assessment of KRAS and BRAF mutations in this setting, we searched for phase III adjuvant trials with post hoc analyses and found 9 well-known adjuvant trials in which the effects of these mutations on DFS and/or OS were investigated (24–26,36–40).
Initial reports from retrospective cohorts found no prognostic value of KRAS/BRAF in stage II and III colon cancer (44). However, with quality selection of phase III trials, the detrimental effect of these mutations was evident, with a pooled 30%-50% higher risk of death (HR = 1.27 to 1.49, P = .03 to <.001) in the primary meta-analysis (Figure 2). KRAS mutation did also statistically significantly affect DFS in the primary meta-analysis (P < .001), and the effect of BRAF on DFS was not statistically significant but with a P value of .05. The impact of both mutations was enhanced when adjustment for MSI status was included in the survival analysis, thus suggesting that their prognostic value is especially relevant in MSS tumors.
In particular, BRAF effect clearly interacts with MSI, with the MSI-high status probably being the predominant molecular feature when co-occurring with BRAF mutation and conferring a more favorable prognosis regardless of the BRAF status. On the other hand, the adverse effect of mutant BRAF is clearly observable in MSS tumors (Figure 4).
It must be made clear that the findings of the present meta-analysis only highlight the prognostic value of KRAS and BRAF mutations. No assumption can be made at the moment with regard to their predictive role, especially in determining the duration of adjuvant chemotherapy. The DFS noninferiority of 3 months of the standard doublet chemotherapy fluoropyrimidine plus oxaliplatin compared with 6 months was the primary objective of the recent International Duration Evaluation of Adjuvant Chemotherapy project (45). The noninferiority could not be demonstrated in a population of nearly 13 hundred patients; however, 3 months of chemotherapy seemed to be adequate for most of patients. A post hoc analysis of the International Duration Evaluation of Adjuvant Chemotherapy population looking for a possible predictive role of KRAS/BRAF would be highly desirable because mutations might be confirmed as risk factors requiring longer adjuvant treatment duration.
A number of limitations of the present analysis need to be discussed, above all its retrospective nature. Statistically significant heterogeneity among trials is also evident for KRAS when adjusting for MSI status (Figure 3), and this may be explained by the different treatments adopted. In particular, the potential detrimental impact of adding cetuximab to oxaliplatin-containing regimens in KRAS-mutated tumors may have amplified the results in the PETACC-8 and N0147 trials (46). In the same trials, the differential efficacy of the anti-EGFR among KRAS wild-type right- vs left-sided tumors might be a confounding effect (47). On the other hand, KRAS has no apparent survival effect in trials using adjuvant irinotecan (CALGB 89803 and PETACC-3; Figure 3). Because survival probabilities were reported to be inferior when irinotecan was included in the adjuvant treatment (48), a detrimental effect of irinotecan-based regimens for both KRAS wild-type and KRAS mutated patients, attenuating the prognostic value of KRAS mutations, cannot be excluded. Probably for all these reasons, looking at individual trial hazard ratios, a seemingly bimodal distribution is observed especially in the KRAS analysis (Figure 3), with higher hazard ratios observed for the trials including the use of the anti-EGFR cetuximab and lower hazard ratios for the trials including irinotecan.
Most of the population included in the present meta-analysis was treated with adjuvant chemotherapy, with only a minority of patients assigned to observation in the QUASAR trial. This constitutes a bias for the estimation of the true prognostic effect of the mutations independently of adjuvant chemotherapy. The absence of data on expanded RAS mutations including NRAS, exclusion of some KRAS mutated cases after amendments of PETACC8 and N0147 trials (49,50), and the unavailability of adequate data stratified by stage are other drawbacks of the present research and possible sources of heterogeneity. Finally, we should acknowledge that different proportions of patients from the original study populations were assessed for KRAS and BRAF mutational status in available trials (from 26% to 73%; Table 1).
Our findings suggest that MSS BRAF/KRAS-mutated tumors have a more aggressive tumor biology, which is observed from the early stages of cancer growth and is maintained at the metastatic stage. The use of KRAS/BRAF genotyping for stage II and III colon cancer appears particularly appealing for many reasons: 1) testing platforms have long been used in the metastatic setting where the test is mandatory for the treatment choice. Most of the pathology laboratories are therefore already equipped to provide results of KRAS/BRAF profiles; 2) KRAS and BRAF assessment might be incorporated in stage II and III colon cancer work-up and prompt the use of longer or more intense adjuvant chemotherapy in cases of uncertain biological behavior (eg, T3N1 with histological grade 3, lymphovascular or perineural invasion, tumor budding); 3) KRAS/BRAF mutations and MSI status can be taken into consideration as stratification factors in future trials and also be used to include patients in adjuvant trials of novel anti-BRAF or anti-KRAS agents or immunotherapy (51–53); 4) having KRAS/BRAF status available from the early stages would speed up the treatment decision making in the metastatic setting in the case of the unfortunate occurrence of relapse; and 5) the early evaluation of tissue KRAS/BRAF mutational status would encourage the use of these mutations, when detected, for the assessment of postoperative minimal residual disease through the liquid biopsy circulating tumor DNA (ctDNA), thus helping identify patients at higher risk of recurrence (54,55). Trials are ongoing where patients with minimal residual disease positivity to BRAF V600E mutation or MSI-high ctDNA are treated with BRAF inhibitor or immunotherapy, respectively (ClinicalTrials.gov Identifier: NCT03803553).
In conclusion, in the present meta-analysis, we were able to confirm a statistically significant overall effect of both KRAS and BRAF mutations on prognosis in the nonmetastatic setting, with the effect being particularly evident when the estimates were adjusted for presence of MSI-high status. Prospective confirmation of these findings and assessment of their use to modulate intensity, duration, and type of adjuvant therapy are warranted.
Funding
No funding to declare related to the present project.
Notes
Role of the funder: Not applicable.
Disclosures: Vincenzo Formica reports personal fees from and consulting or advisory role with Amgen, Servier, Merck, Sanofi; Chiara Cremolini reports personal fees from Roche, Amgen, Bayer, and Servier; research funding from Merck Serono; and a consulting or advisory role with Roche, Bayer, Amgen. Hendrik-Tobias Arkenau is an investigator in studies sponsored by Taiho and reports an advisory role in Guardant, Roche, and Servier. Francesco Sera, Silvia Riondino, Cristina Morelli and Mario Roselli report no conflicts of interest.
Author contributions: Conceptualization: VF, CC, SR, CM, HTA, MR. Methodology: VF, FS, CC, CM. Software: VF, FS. Data curation: VF, FS, CM. Writing- Original draft preparation: VF, CC, SR, HTA. Visualization, Investigation: VF, FS, CC, SR, CM, HTA, MR. Supervision: HTA, MR. Software, Validation: FS. Writing- Reviewing and Editing: VF, FS, CC, SR, CM, HTA, MR.
Acknowledgements: Cristina Morelli has conducted the present study within the PhD Program on Experimental System and Medicine (XXXV cycle) at the University of Rome Tor Vergata.
Data Availability
The data underlying this article are available in the article itself and in its online Supplementary Material.
