Results of the search

We identified 1826 records through the electronic searches of the Cochrane Library (279 records), MEDLINE (332 records), MEDLINE In‐process & Other Non‐indexed citations (548 records) and Embase (667 records). We excluded 791 duplicates and 968 clearly irrelevant records through reading titles and abstracts. We retrieved the remaining 67 records for further assessment. Upon full‐text reading, we excluded 41 studies for reasons listed in the Characteristics of excluded studies table. We identified eight additional studies through scanning reference lists of the identified RCTs.

In total, 34 RCTs fulfilled the inclusion criteria. The study flow diagram is shown in Figure 3.

Figure 3

---

---

Study flow diagram.

A list of abbreviations used in the text are reported in Table 1.

Included studies

The main features of each of the 34 included trials (overall enrolling 13,787 participants; range: 55 to 1298; median: 247) are reported in the Characteristics of included studies table. One trial compared four arms (Bendell 2013_folfiri), four trials compared three arms (Cohn 2013_conat; Élez 2015; O'Neil 2014; Rothenberg 2003_folfox), and the remaining 27 trials compared two arms. Only one trial compared a second‐line chemotherapy regimen with best supportive care (Cunningham 1998). The other included studies compared two or more different second‐line treatments.

All trials were comparable in terms of baseline characteristics (all participants enrolled had metastatic CRC and had not responded to a first‐line treatment), except for the first‐line regimens that could be different. As regards outcomes, OS, PFS, ORR and SAE data were reported in 31, 31, 30 and 34 trials, respectively.

The drug regimens tested were heterogeneous. In particular, 17 different anticancer agents (alone or in combination) were tested: seven chemotherapy agents (5FU, oxaliplatin, irinotecan, mitomycin‐C, raltitrexed, capecitabine, S‐1) and 10 targeted agents (seven monoclonal antibodies: bevacizumab, cetuximab, panituzumab, abituzumab, conatumumab, ganitumab, ramucirumab; two small molecule inhibitors: vatalanib and gefitinib; and two recombinant blocking proteins: aflibercept and trebananib). The main features of these compounds are reported in Table 2. Two compounds (leucovorin and hyaluronan) utilized in some RCTs have no anticancer activity per se, but increase (leucovorin) or are supposed to increase (hyaluronan) the anticancer activity of chemotherapy agents.

The RCTs tested the following 25 drug combinations: 1. FOLFOX (5FU plus leucovorin plus oxaliplatin), 2. FOLFIRI (5FU plus leucovorin plus irinotecan), 3. IRIS (S‐1 plus irinotecan), 4. XELOX (capecitabine plus oxaliplatin), 5. bevacizumab plus FOLFIRI, 6. bevacizumab plus FOLFOX, 7. cetuximab plus irinotecan, 8. panitumumab plus FOLFIRI, 9. panitumumab plus irinotecan, 10. panitumumab plus bevacizumab plus FOLFIRI, 11. axitinib plus FOLFOX, 12. axitinib plus FOLFIRI, 13. conatumumab plus FOLFIRI, 14. ganitumab plus FOLFIRI, 15. hyaluronan plus irinotecan, 16. oxaliplatin plus irinotecan, 17. linifanib plus FOLFOX, 18. ramucirumab plus FOLFIRI, 19. vatalanib plus FOLFOX, 20. aflibercept plus FOLFIRI, 21. gefitinib plus raltitrexed, 22. mitomycin plus oxaliplatin, 23. mitomycin plus irinotecan, 24) trebananib plus FOLFIRI and 25. abituzumab plus cetuximab plus irinotecan.

Only one trial used no anticancer treatment (best supportive care) as the control arm (the experimental arm was treatment with irinotecan) (Cunningham 1998).

Excluded studies

After reading the full text of 65 articles, we excluded 39 studies. Main reason for exclusion was that the trial did not focus on second‐line therapy, rather the enrolment of people for first‐line treatment or for third‐ or higher‐line treatment.

Details regarding the reasons for exclusion of each study are reported in the Characteristics of excluded studies table.

Allocation

There was no high risk of selection bias, although the risk was unclear for about 50% of the trials (Figure 1).

Blinding

There was high risk of bias in 22 open‐label trials due to the lack of blinding (performance bias) intrinsic to the study design.

Incomplete outcome data

There was a high risk of attrition bias (incomplete outcome data) in only one study due to the following reasons: 1. for survival outcome, neither the arm the two withdrawn participants belonged to, nor the reasons for withdrawal were reported; and 2. for tumour response, "39 patients were evaluable for response", therefore 16 participants were not evaluable (Graeven 2007).

Selective reporting

There was no high risk of reporting bias.

Other potential sources of bias

There was no high risk of other bias (e.g. baseline imbalances between arms).

Modern chemotherapy regimens versus 5‐fluorouracil

Two RCTs (726 participants) compared 5FU (the first chemotherapeutic agent to be associated with survival advantage in first‐line treatment of people with metastatic CRC) with three different chemotherapy regimens (including more modern drugs): FOLFOX (Rothenberg 2003_folfox), oxaliplatin (Rothenberg 2003_oxa), and irinotecan (Rougier 1998).

Only one trial (167 participants) addressed OS and showed a significant advantage for participants treated with irinotecan (HR 0.69, 95% CI 0.51 to 0.94) (Rougier 1998). Both trials reported PFS: considering the single comparisons, both FOLFOX and irinotecan were better than 5FU, whereas oxaliplatin was not. By pooling the PFS data of the three comparisons, regimens other than 5FU alone performed better than 5FU alone (HR 0.84, 95% CI 0.73 to 0.96) (Figure 4), a finding characterized by a high between‐study heterogeneity (I² = 91%) sustained by the poor activity of oxaliplatin alone. The 'leave‐one‐out' sensitivity analysis showed that heterogeneity reduced to 11% if the oxaliplatin versus 5FU comparison was excluded; in this case, FOLFOX and irinotecan showed a clinically relevant and statistically significant PFS advantage compared to 5FU in second‐line therapy of metastatic CRC (HR 0.59, 95% CI 0.49 to 0.73). Based on the GRADE system, we rated this evidence as high (no reason for downgrading) (summary of findings Table for the main comparison).

Figure 4

---

---

For tumour response data, at meta‐analysis, ORR was significantly higher for other chemotherapy regimens compared to 5FU (RR 2.96, 95% CI 1.66 to 5.27), although this finding was not supported by the oxaliplatin versus 5FU comparison. Finally, the SAE rate was higher in participants treated with modern chemotherapy regimens (RR 1.39, 95% CI 1.22 to 1.58).

Irinotecan plus other anticancer agents versus irinotecan

Six trials assessed the effect of adding other anticancer agents (5FU, oxaliplatin, hyaluronan acid (a drug carrier), panitumumab, cetuximab) to irinotecan (2615 participants) (Clarke 2011; Gibbs 2011; Graeven 2007; Haller 2008; Seymour 2013; Sobrero 2008).

Only one trial (627 participants) found a statistically significant advantage for both OS (HR 0.78, 95% CI 0.67 to 0.90) and PFS (HR 0.60, 95% CI 0.55 to 0.66), when irinotecan was combined with oxaliplatin (also known as IROX regimen) (Haller 2008). The grade of evidence was moderate (because there was only one study) (summary of findings Table 4).

When the results of the six studies were pooled together, there was no OS advantage (HR 0.91, 95% CI 0.79 to 1.04), with a non‐negligible between‐study heterogeneity (I² = 42%). In contrast, there was both a PFS advantage (HR 0.68, 95% CI 0.60 to 0.76) (Figure 5) and an ORR advantage (RR 2.87, 95% CI 2.10 to 3.93) for combination regimens, with no significant difference in terms of toxicity (RR 1.18, 95% CI 0.96 to 1.45); the grade of evidence was moderate (since the PFS findings were not confirmed by the OS data) (summary of findings Table 3).

Figure 5

---

---

Since the between‐study heterogeneity was moderate to high for all outcomes, we searched for more homogeneous results by subgroup analysis. Pooling the data of two trials (144 participants) comparing the same regimens (irinotecan versus FOLFIRI) found no OS, PFS, ORR or SAE differences (Clarke 2011; Graeven 2007).

Heterogeneity was drastically reduced (I² less than 25%) when we meta‐analyzed the data of the two trials (1758 participants) testing the hypothesis that the addition of targeted agents (panitumumab, cetuximab) might enhance the therapeutic efficacy of irinotecan (Seymour 2013; Sobrero 2008): both PFS (HR 0.71, 95% CI 0.64 to 0.79) and ORR (RR 3.38, 95% CI 2.52 to 4.53) were significantly better in the combination regimen; however, there was no significant OS difference (HR 0.99, 95% CI 0.88 to 1.10), and the rate of SAEs was significantly higher with the combination regimen (RR 1.44, 95% CI 1.31 to 1.59); the grade of evidence was moderate because the PFS findings were not confirmed by the OS data (summary of findings Table 8).

Fractionated irinotecan versus single‐dose irinotecan

Three RCTs compared irinotecan administered as single dose (every three weeks) with that of irinotecan fractionated (with different schedules) (Fuchs 2003, 391 participants; Shoemaker 2004, 85 participants; Tsavaris 2003, 120 participants). There were no statistically significant differences in terms of OS (HR 0.86, 95% CI 0.71 to 1.05; 2 RCTs) (Fuchs 2003; Tsavaris 2003), PFS (HR 0.90, 95% CI 0.71 to 1.14; 2 RCTs) (Fuchs 2003; Tsavaris 2003), ORR (RR 0.83, 95% CI 0.45 to 1.53; 2 RCTs) (Shoemaker 2004;Tsavaris 2003), and SAE (RR 0.97, 95% CI 0.87 to 1.10; 3 RCTs) (Fuchs 2003; Shoemaker 2004; Tsavaris 2003).

Bevacizumab plus chemotherapy versus chemotherapy

Four RCTs (1723 participants) addressed the addition of bevacizumab to chemotherapy (oxaliplatin‐ or irinotecan‐based regimens) (Bennouna 2013; Cao 2015; Giantonio 2007; Masi 2015).

All four studies demonstrated a survival advantage for the combination arms, although this advantage was only statistically significant in two RCTs (Bennouna 2013; Giantonio 2007). Pooling the OS data from all four studies showed that bevacizumab can lead to a survival advantage (HR 0.79, 95% CI 0.70 to 0.88) (Figure 6), with no between‐study heterogeneity (I² = 0%). The four single trials reported even better results for PFS, which was confirmed by meta‐analysis (HR 0.67, 95% CI 0.60 to 0.75) (Figure 7). Based on the GRADE system, we rated the evidence as high (no reason for downgrading) (summary of findings Table 2).

Figure 6

---

---

Figure 7

---

---

ORR data were in line with survival data as the response rates were higher in the bevacizumab‐containing regimens (RR 1.72, 95% CI 1.23 to 2.43). Interestingly, the addition of bevacizumab did not significantly increase the risk of SAEs (RR 1.07, 95% CI 0.93 to 1.25).

FOLFIRI plus targeted agents versus FOLFIRI

Six trials (3335 participants) tested the hypothesis that adding other anticancer agents to the FOLFIRI regimen (conventional chemotherapy drugs only) might increase therapeutic efficacy investigating seven different targeted drugs: conatumumab (Cohn 2013_conat), ganitumab (Cohn 2013_ganit), panitumumab (Peeters 2010), bevacizumab (plus panitumumab) (Liu 2015), trebananib (Peeters 2013), ramucirumab (Tabernero 2015), and aflibercept (Van Cutsem 2012). Singularly taken, only three RCTs showed a significant improvement in OS (Liu 2015; Tabernero 2015; Van Cutsem 2012); pooling the data of all six trials, we found that the addition of targeted agents did increase life expectation (HR 0.84, 95% CI 0.77 to 0.91) (Figure 8). There were similar results when pooling PFS data (HR 0.78, 95% CI 0.71 to 0.87) (Figure 9) and ORR data (RR 2.07, 95% CI 1.31 to 3.28).

Figure 8

---

---

Figure 9

---

---

Importantly, the addition of targeted agents also significantly increased the incidence of SAEs (RR 1.30, 95% CI 1.17 to 1.45). At subgroup analysis, considering the three trials (2440 participants) that investigated targeted agents blocking exclusively the VEGF pathway (bevacizumab, ramucirumab, aflibercept), findings were even more homogeneous in terms of survival (OS: HR 0.82, 95% CI 0.74 to 0.90; PFS: HR 0.77, 95% CI 0.70 to 0.84).

Based on the GRADE system, we rated the evidence as high.

FOLFOX plus targeted agents versus FOLFOX

Two RCTs (1432 participants) tested the hypothesis that adding targeted agents (bevacizumab and vatalanib) to the FOLFOX regimen (chemotherapy drugs only) might increase therapeutic efficacy (bevacizumab: Giantonio 2007; vatalanib: Van Cutsem 2011). The meta‐analysis of the two trials showed a significantly improved PFS when targeted agents were added to FOLFOX (HR 0.76, 95% CI 0.66 to 0.86) (Figure 10), although there was a high degree of heterogeneity (I² = 78%). Addition of targeted agents also improved OS, although this result did not reach statistical significance (HR 0.92, 95% CI 0.82 to 1.04).

Figure 10

---

---

Finally, the incidence of SAEs was significantly higher among participants receiving the combination regimen (RR 1.2, 95% CI 1.13 to 1.28). Only one study reported ORR, which showed an advantage for the regimen including bevacizumab (Giantonio 2007).

Based on the GRADE system, we rated the evidence as low (due to high heterogeneity and PFS data not confirmed by OS data).

FOLFIRI plus bevacizumab versus FOLFIRI plus other targeted agents

Two trials (282 participants) compared the FOLFIRI chemotherapy regimen in combination with either bevacizumab or other targeted agents (axitinib or panitumumab) (Bendell 2013_folfiri; Hecht 2015). Neither study reported a significant OS difference between the two approaches, a result confirmed by data meta‐analysis (HR 1.15, 95% CI 0.86 to 1.53). There were similar findings for PFS and ORR both in single studies and meta‐analysis (PFS: HR 1.10, 95% CI 0.81 to 1.50; ORR: RR 1.44, 95% CI 0.94 to 2.20).

Toxicity findings favoured the bevacizumab‐containing regimen (other agents versus bevacizumab comparison: RR 1.53, 95% CI 1.10 to 2.11). Due to the low number of participants randomized, any result should be interpreted with caution (due to low statistical power considerations).

Using the GRADE system, we rated the evidence as low (due to imprecision and low statistical power).

FOLFOX plus bevacizumab versus FOLFOX plus other targeted agents

Two trials (169 participants) compared the FOLFOX chemotherapy regimen in combination with either bevacizumab or other targeted agents (Axitinib or Linifanib) (Bendell 2013_folfox; O'Neil 2014). Neither study reported a significant OS difference between the two approaches, a result confirmed by data meta‐analysis (HR 1.14, 95% CI 0.81 to 1.61). There were similar findings for PFS and ORR both in single studies and meta‐analysis (PFS: HR 1.26, 95% CI 0.83 to 1.91; ORR: RR 0.77, 95% CI 0.46 to 1.29).

Toxicity findings favoured the bevacizumab‐containing regimen (other agents versus bevacizumab comparison: RR 1.27, 95% CI 1.03 to 1.55). Due to the low number of participants randomized, any result should be interpreted with caution (due to low statistical power considerations).

Using the GRADE system, we rated the evidence as low (due to imprecision and low statistical power).

Other comparisons based on single trials

Single RCTs compared the following regimens (for details see Characteristics of included studies and Table 3).