Treatment type and line of treatment

Among studies of curative intent treatment for CRC, two studies examined neoadjuvant treatment alone for rectal carcinoma (De la Torre 2008; Allegra 2015), and one study explored use of both neoadjuvant and adjuvant treatment for rectal carcinoma (Hofheinz 2012). Six studies examined adjuvant treatment alone, including four studies for colon carcinoma (De Gramont 2012; Lembersky 2006; Shimada 2014; Twelves 2012), one study for rectal carcinoma (Kim 2001a), and one study for carcinoma of the colon or rectum (Pectasides 2015) (Table 1). Among studies that included patients with rectal carcinoma, two studies required the distal border of the tumour to be < 12 cm from the anal verge (Allegra 2015; Kim 2001a), one study required the distal border of the tumour to be < 16 cm from the anal verge (Hofheinz 2012), and two studies did not describe anatomical criteria (De la Torre 2008; Pectasides 2015).

Among studies of palliative intent treatment with chemotherapy for inoperable advanced or metastatic CRC, 31 were performed exclusively in the first‐line setting. One study had exclusion criteria that specified “no past history of chemotherapy or chemotherapy ceased for over six months” and included patients in the report who had been given first‐ and second‐line treatment (Yu 2005). Kato 2012 included patients given first‐ or second‐line treatment; if treatment was second‐line, first‐line therapy with FOLFOX was mandated. Two studies were conducted in the second‐line setting ‐ one in combination with oxaliplatin (Rothenberg 2008) and one in combination with irinotecan (Yasui 2015) (Table 2; Table 3).

Location

Among studies of curative intent treatment with neoadjuvant and/or adjuvant chemotherapy for CRC, capecitabine trials were performed in Greece (Pectasides 2015), in Europe (Hofheinz 2012), and in the USA, Europe, Asia, Australia, and other countries (Twelves 2012; De Gramont 2012). The Allegra 2015 study was predominantly performed in North America. UFT studies were conducted at sites in Asia (Shimada 2014), Europe (De la Torre 2008), and North America (Lembersky 2006). The single doxifluridine study was performed in Asia only (Kim 2001a).

Among studies of palliative intent treatment with palliative chemotherapy for inoperable advanced or metastatic CRC, all four S‐1 trials were performed in Asia only (Kato 2012; Yamada 2013; Yamazaki 2015; Yasui 2015). One Asia‐only study used capecitabine (Yu 2005); nine studies were conducted in Europe or included both European and non‐USA sites. Additionally, three European Intergroup studies were carried out ‐ Gruppo Oncologico Aree Metropolitane ‐ GOAM (Martoni 2006); Gruppo Oncologico dell'Italia Meriodionale ‐ GOIM (Silvestris 2010); and European Organisation for Research and Treatment of Cancer ‐ EORTC (Kohne 2008). Two capecitabine studies were conducted in the USA (Hochster TREE‐1 2008; Hochster TREE‐2 2008), and four studies had sites in the USA and in other countries. UFT trials were conducted in Europe and in non‐USA countries (Carmichael 2002; Douillard 2014), and in the USA and in other countries (Douillard 2002). Non‐USA sites in the European UFT trials included Canada, Australia, New Zealand, and Israel (Carmichael 2002); and Asia, South America, Australia, and Israel (Douillard 2014); the Douillard 2002 study also included non‐USA sites in Europe, Canada, and Puerto Rico. One UFT study was based in Japan (Shigeta 2016). Tegafur was used in two European studies (Andersen 1987; Nogue 2005). Eniluracil was used in one USA study (ECOG E5296 2012); one study was performed in the USA and Canada (Schilsky 2002a), and one was an international study (Van Cutsem 2001a). Doxifluridine was used in Europe (Bajetta 1996), and in South Korea (Ahn 2003)(Characteristics of included studies).

Performance status

Although most studies included patients with ECOG PS 2 or less (or the equivalent Karnofsky PS (KPS)), the study population for Seymour 2011 comprised elderly and frail patients who were considered by the treating oncologist to be unsuitable for upfront full‐dose chemotherapy. One study (Andersen 1987) included patients with ECOG PS 3, although the proportion of patients with ECOG PS 3 was not clear (Characteristics of included studies).

Monitoring of compliance and adherence to oral treatment

Among studies of curative intent treatment with neoadjuvant and/or adjuvant chemotherapy for CRC, only one study (Lembersky 2006) reported monitoring of compliance and adherence to oral treatment.

Among studies of palliative intent treatment with chemotherapy for inoperable advanced or metastatic CRC, 10 studies (Ahn 2003; Bajetta 1996; Douillard 2002; Douillard 2014; Ducreux 2011; Martoni 2006; Rothenberg 2008; Schilsky 2002a; Seymour 2011; Shigeta 2016) described oral chemotherapy pill monitoring or use of a patient diary.

Patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy

Of the nine studies, all but one study assessed DFS. Kim 2001a examined rates of local and systemic recurrence but did not report DFS. Although De la Torre 2008 examined DFS, we did not include this study in the DFS meta‐analysis owing to insufficient median follow‐up time (22 months) (Table 1).

All of the studies apart from Kim 2001a reported the OS outcome. We excluded De la Torre 2008 from the meta‐analysis of OS owing to insufficient follow‐up time (Appendix 5).

All of the studies reported outcome data for at least one specific grade ≥ 3 AE of interest for this review, and all provided data that were suitable for meta‐analysis. Included studies reported information for specific grade ≥ 3 AEs: diarrhoea (n = 9), HFS (n = 7), neutropenia/granulocytopenia (n = 7), febrile neutropenia (n = 4), vomiting (n = 8), nausea (n = 7), stomatitis (n = 5), mucositis (n = 4), and hyperbilirubinaemia (n = 4). Two studies of adjuvant treatment (Hofheinz 2012; Kim 2001a) described 'lowered leucocytes' or 'leukopenia' only and were excluded from the meta‐analysis (Appendix 6). One study (Twelves 2012) reported combined data for grade ≥ 3 vomiting and nausea, and one study (De la Torre 2008) reported combined data for grade ≥ 3 stomatitis and mucositis. Table 4 shows the relationships between reported AEs and treatment for the included studies. Included studies used the following AE assessment criteria: ECOG CTC (n = 1), NCI CTCAE version 4.0 (n = 1), NCI CTCAE version 3.0 (n = 1), NCI CTCAE version 2.0 (n = 3), NCIC‐CTG CTC 1991 version (n = 1), NCI CTC 1958 (n = 1), and WHO, version not specified (n = 1).

Overall, five studies presented data for 'any grade ≥ 3 AEs' (Allegra 2015; De Gramont 2012; Hofheinz 2012; Lembersky 2006; Twelves 2012).

Patients treated with palliative intent for inoperable advanced or metastatic CRC with palliative chemotherapy

Of the 35 studies, all but one study contributed to pooled effect estimates for an efficacy outcome and/or at least one grade ≥ 3 AE outcome (Silvestris 2010).

A total of 25 studies assessed PFS, and eight studies assessed the TTP outcome. Andersen 1987 did not assess either outcome. Hochster TREE‐1 2008, Hochster TREE‐2 2008, Hoff 2001, and Van Cutsem 2001b described TTP as the outcome examined but provided a definition compatible with the definition for PFS provided in this review. Bajetta 1996 stated that time to treatment failure was the examined outcome but provided a definition compatible with the definition for PFS provided in this review.  Ahn 2003 described PFS as the examined outcome but provided a definition compatible with the classification for TTP provided in this review. We excluded Hochster TREE‐1 2008 and Hochster TREE‐2 2008 (for the PFS endpoint) and Silvestris 2010 and Yu 2005 (for the TTP endpoint) from our meta‐analyses because we could not estimate the HRs either directly or indirectly from the information provided (Appendix 7). Douillard 2002 presented only median TTP times with a stratified log‐rank P value.

Thirty‐one studies reported the OS outcome. Kato 2012, Martoni 2006, Mei 2014, and Silvestris 2010 did not report the OS outcome, and we excluded Andersen 1987 and Yu 2005 from our quantitative synthesis because we could not estimate the HR either directly or indirectly from the report (Appendix 7).

All 35 studies assessed ORR using the following criteria: WHO 1979 (n = 3), WHO 1981 (n = 4), modified WHO (n = 2), RECIST, version 1.0 (n = 21), RECIST, version not specified (n = 1), ECOG (n = 1), and SWOG (n = 1). Two studies did not specify this information (Van Cutsem 2001a; Yu 2005). We excluded Mei 2014 and Seymour 2011 from meta‐analysis because investigators reported ORR only after two cycles of chemotherapy and at 12 to 14 weeks after the start of treatment, respectively. We excluded Silvestris 2010 because investigators assessed an unclear number of participants for ORR in both arms (Appendix 7). Of the 32 studies included in the meta‐analysis, 22 studies provided information on the number of participants assessable or evaluable for response. One study did not specify a separate ORR analysis population denominator (Van Cutsem 2001a).

All but one included study (Andersen 1987) reported outcome data on grade ≥ 3 AEs of interest for this review. Table 4 shows the relationship between reported AEs and treatment in the included studies. AE assessment criteria included NCI CTCAE, version 3.0 (n = 14); NCI CTCAE, version 2.0 (n = 9); NCI CTCAE, 1994 version (n = 2); NCI CTCAE, 1981 version (n = 1); NCI CTCAE, version not specified (n = 3); WHO, 1981 (n = 1); WHO, version not specified (n = 1); and an adaptation of SWOG, 1992 (n = 1). Two studies did not specify this information. Included studies provided information for specific grade ≥ 3 AEs as follows: diarrhoea (n = 30), HFS (n = 23), neutropenia/granulocytopenia (n = 29), febrile neutropenia (n = 19), vomiting (n = 23), nausea (n = 25), stomatitis (n = 21), mucositis (n = 12), and hyperbilirubinaemia (n = 12). Five trials provided combined grade ≥ 3 stomatitis and mucositis data (Carmichael 2002; Douillard 2002; Shigeta 2016; Yamada 2013; Yasui 2015), and eight studies reported combined grade ≥ 3 nausea and vomiting data (Ahn 2003; Carmichael 2002; Cassidy 2011a; Douillard 2002; Hochster TREE‐1 2008; Hochster TREE‐2 2008; Mei 2014; Nogue 2005). Fourteen studies presented data for 'any grade ≥ 3 AEs'. Three studies did not specify a separate safety analysis denominator (Comella 2009; Porschen 2007; Van Cutsem 2001a). Kato 2012 reported grade ≥ 3 AEs up to 12 weeks.

Four studies had no grade ≥ 3 AE outcomes that were suitable for meta‐analysis. Of these, one study included an unclear number of participants in the safety analysis population and unclear units of analysis (Ahn 2003). Andersen 1987 did not report any grade ≥ 3 AE outcomes. Silvestris 2010 included an unclear number of participants in the safety analysis population for each arm. For Yu 2005, it is unclear whether investigators reported AEs for the entire study population or only for a subset owing to discrepancies between the table title and participant numbers provided in the table (Appendix 8). Two additional included studies reported only 'leukopenia' or lowered 'white blood cells' (Bajetta 1996; Kohne 2008). One study reported grade 2 and 3 HFS (Porschen 2007), and one trial reported toxicities affecting skin/appendages which included but were not confined to HFS (Carmichael 2002). We did not include these studies in the meta‐analysis for neutropenia/granulocytopenia and HFS outcomes, respectively (Appendix 6).

Early stopping

Among studies of curative intent treatment for CRC, one study of neoadjuvant treatment (De la Torre 2008) was stopped early owing to slow accrual after 63% of the number of participants planned for accrual had been randomised. For similar reasons, one study of adjuvant treatment (Pectasides 2015) was prematurely closed after 55% of participants were enrolled.

Among studies of palliative intent treatment with chemotherapy for inoperable advanced or metastatic CRC, four studies were stopped early: Nogue 2005 owing to slow accrual, when 85% of the planned number of participants for accrual to the study had been randomised; ECOG E5296 2012 after 125 of the 950 planned participants had been accrued, owing to negative results from two other studies of eniluracil with oral 5‐FU (Schilsky 2002a; Van Cutsem 2001a); Kohne 2008 after enrolment of only 85 participants as a consequence of seven deaths that were assessed as unrelated to disease progression; and Fuchs 2007 after 547 of the 900 participants for Periods 1 and 2 combined had been enrolled. Accrual to this trial had slowed after reports described cardiovascular concerns with celecoxib, although celecoxib/placebo administration was permanently discontinued for patients in January 2005.

Studies of curative intent treatment with neoadjuvant and/or adjuvant chemotherapy for CRC

We assessed seven curative intent studies as 'Low' risk. Two curative intent studies had 'Unclear' risk of bias owing to unspecified methods of randomisation, and we did not assess any studies as having 'High' risk of bias.

Studies of palliative intent treatment with chemotherapy for inoperable advanced or metastatic CRC

We assessed 20 palliative intent studies as 'Low' risk. Fifteen palliative intent studies had ‘Unclear’ risk of bias owing to unspecified methods of randomisation, and we did not assess any studies as having 'High' risk of bias.

Studies of curative intent treatment with neoadjuvant and/or adjuvant chemotherapy for CRC

We assessed seven curative intent studies as 'Low' risk. Two curative intent studies had 'Unclear' risk of bias owing to lack of information about allocation concealment, and we did not assess any studies as having 'High' risk of bias.

Studies of palliative intent treatment with chemotherapy for inoperable advanced or metastatic CRC

We assessed 21 palliative intent studies as 'Low' risk. Fourteen palliative intent studies had 'Unclear' risk of bias owing to lack of information about allocation concealment, and we did not assess any studies as having 'High' risk of bias.

DFS/PFS/TTP/ORR

We judged that lack of blinding of participants and/or personnel would not lead to ‘High’ risk of bias for these outcomes.

OS

We judged that lack of blinding of participants and/or personnel would not lead to ‘High’ risk of bias for these outcomes.

Grade ≥ 3 AEs

We judged that lack of blinding of participants and personnel would lead to 'High' risk of bias for this outcome.

DFS/PFS/TTP/ORR

We judged that lack of blinding of outcome assessors would lead to 'High' risk of bias for this outcome.

OS

We judged that lack of blinding of outcome assessors would not lead to ‘High’ risk of bias for these outcomes.

Grade ≥ 3 AEs

We judged that lack of blinding of outcome assessors would lead to 'High' risk of bias for this outcome, in particular for assessment of subjective grade ≥ 3 AEs such as HFS, diarrhoea, vomiting, nausea, stomatitis, and mucositis. We did not judge that lack of blinding of outcome assessors would affect assessment of grade ≥ 3 neutropenia/granulocytopenia, febrile neutropenia, or hyperbilirubinaemia, as these rely upon objective laboratory assessments.

ORR in studies of palliative intent treatment with chemotherapy for inoperable advanced or metastatic CRC

Of the 32 palliative intent studies that contributed to the ORR analysis, 23 studies had 'Low' risk of bias, four had 'Unclear’ risk, and five had ‘High’ risk. For Andersen 1987, ORR data were non‐evaluable for 20% of participants in the IV 5‐FU arm, and for Ahn 2003, ORR data were non‐evaluable for 29% of participants in the 5‐dFUR/LV arm. Twenty‐three per cent of participants in the CapeOx arm had missing confirmed tumour response data in Hochster TREE‐1 2008. Twenty‐four per cent (FT/LV) and 20% (5‐FU/LV) of participants in Nogue 2005 had non‐evaluable data for response owing to protocol deviations in response evaluation methods. In Pectasides 2012, 30.1% of participants in the XELIRI‐BEV arm and 19.7% of those in the FOLFIRI‐BEV arm had non‐evaluable response data owing to treatment discontinuation, early death, missing data, and non‐evaluable disease.

Time‐to‐event outcomes (DFS/PFS/OS/TTP)

Grade ≥ 3 AEs

Efficacy analysis

We judged studies to be at 'Low' risk of bias if an ITT analysis was performed as per the definition in our review, or if < 5% of randomised participants were excluded from the analysis.

Safety analysis

Most studies performed a safety analysis in the as‐treated population, which included participants who had received at least one dose of chemotherapy.

Disease recurrence/response (influences DFS/PFS/TTP/ORR)

This pertains to the detection of disease recurrence, response, and progression events.

Survival (influences DFS/PFS/OS)

This pertains to detection of death events.

Grade ≥ 3 AEs

This pertains to the detection of grade ≥ 3 AEs.

Studies of curative intent treatment with neoadjuvant and/or adjuvant chemotherapy for CRC

Of nine curative intent studies that contributed to meta‐analyses for any of the outcomes in this review, six studies were 'Low' risk, as participants in all treatment arms had similar baseline characteristics with regards to PS, median or mean age, and disease stage. Two studies had 'Unclear' risk, and one study had 'High' risk owing to a difference in mean age of 7.2 years (Kim 2001a).

Studies of palliative intent treatment with chemotherapy for inoperable advanced or metastatic CRC

Of the 34 palliative intent studies that contributed to meta‐analyses for any of the outcomes in this review, 12 studies had 'Low' risk of bias, as participants in all treatment arms had similar baseline characteristics with regards to PS, median or mean age, and number of organs involved with metastases, or KRAS mutation status in the case of EGFR inhibitor treatment. Eighteen studies had ’Unclear’ risk of bias. Four studies had ‘High’ risk of bias owing to differences between comparison arms. Hochster TREE‐2 2008 and Shigeta 2016 reported a five‐year difference in median age between oral and IV arms. Martoni 2006 described a 16.5% difference between arms with regards to the percentage with one versus more than one metastatic site at baseline. Douillard 2014 performed a post hoc analysis of participants evaluable for KRAS mutation status and found that a greater proportion of those in the UFOX + cetuximab arm (47/87; 54%) were KRAS mutant than in the FOLFOX4 + cetuximab arm (37/93; 40%). Whilst only 60% of the population was evaluable for KRAS mutation status, we considered that a 14% difference between oral and IV arms would lead to 'High' risk of bias.

Studies of curative intent treatment with neoadjuvant and/or adjuvant chemotherapy for CRC

Two curative intent studies provided information about subsequent treatment with adjuvant chemotherapy or chemotherapy following a recurrence or a new occurrence of CRC (Allegra 2015; Twelves 2012); both had 'Low' risk of bias. The remaining seven curative intent studies did not provide this information and had 'Unclear' risk.

Studies of palliative intent treatment with chemotherapy for inoperable advanced or metastatic CRC

Twenty‐one palliative intent studies reported information about subsequent lines of treatment used for each treatment arm after disease progression. Study authors reported no major differences between treatment arms with regards to the percentage of participants who received subsequent therapy or the type of subsequent therapy used. None were at high risk of bias.

We identified no other reasons for high risk of bias in the included studies.

Patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy

Studies of palliative intent treatment with chemotherapy for inoperable advanced or metastatic CRC

1.1 DFS

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, DFS did not differ between participants treated with oral versus IV fluoropyrimidines. The pooled HR from seven studies with 8903 participants was 0.93 (95% CI 0.87 to 1.00) (Analysis 1.1; Table 5). Results show no heterogeneity (Chi² = 5.51, P = 0.48; I² = 0%) among effect estimates for these studies (Figure 4).

Figure 4

---

---

Forest plot of disease‐free survival.

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We assessed the quality of evidence as moderate (summary of findings Table for the main comparison).

1.2 DFS with subgroup analysis ‐ Treatment type

Chi² = 0.21, P = 0.64; I² = 0%.

1.3 DFS with subgroup analysis ‐ Infusional versus bolus intravenous fluoropyrimidine

Chi² = 0.06, P = 0.81; I² = 0%.

1.4 DFS with subgroup analysis ‐ Oral fluoropyrimidine backbone

Chi² = 1.70, P = 0.19; I² = 41.1%.

2.1 OS (curative intent)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, OS did not differ between participants treated with oral versus IV fluoropyrimidines. The pooled HR from seven studies with 8902 participants was 0.92 (95% CI 0.84 to 1.00) (Analysis 2.1; Table 5). Results show no heterogeneity (Chi² = 4.67, P = 0.59; I² = 0%) among effect estimates for these studies.

We did not identify any factors that reduced the quality of evidence for this outcome, and we assessed the quality of evidence as high (summary of findings Table for the main comparison).

2.2 OS with subgroup analysis ‐ Chemotherapy versus chemo‐radiotherapy

Chi² = 0.43, P = 0.51; I² = 0%.

2.3 OS with subgroup analysis ‐ Infusional versus bolus intravenous fluoropyrimidine

Chi² = 0.00, P = 0.96; I² = 0%.

2.4 OS with subgroup analysis ‐ Oral fluoropyrimidine backbone

Chi² = 2.20, P = 0.14; I² = 54.5%.

3.1 Grade ≥ 3 diarrhoea (curative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, grade ≥ 3 diarrhoea did not differ between participants treated with oral versus IV fluoropyrimidines. The pooled OR from nine studies with 9551 participants was 1.12 (95% CI 0.99 to 1.25) (Analysis 3.1; Table 5).

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We further downgraded quality by one level for inconsistency of results, as we noted substantial heterogeneity among included studies (Chi² = 23.79, P = 0.002; I² = 66%), and by one level for imprecision. We assessed the quality of evidence as very low (summary of findings Table for the main comparison).

3.2 Grade ≥ 3 diarrhoea (curative intent studies) ‐ Treatment type

Results show no subgroup differences by treatment type (chemotherapy versus chemo‐radiotherapy): Chi² = 1.24, P = 0.27; I² = 19.3% (Analysis 3.2; Appendix 10).

3.3 Grade ≥ 3 diarrhoea (curative intent studies) ‐ Infusional versus bolus intravenous fluoropyrimidine

Results show significant subgroup differences between ‘Infusional intravenous fluoropyrimidine’ (pooled OR 1.27, 95% CI 1.06 to 1.53 ‐ indicating more grade ≥ 3 diarrhoea with oral fluoropyrimidines) and ‘Bolus intravenous fluoropyrimidine’ (pooled OR 0.98, 95% CI 0.84 to 1.14 ‐ indicating that grade ≥ 3 diarrhoea did not differ between those treated with oral versus IV fluoropyrimidines) subgroups: Chi² = 4.52, P = 0.03; I² = 77.9% (Analysis 3.3; Appendix 10).

3.4 Grade ≥ 3 diarrhoea (curative intent studies) ‐ Oral fluoropyrimidine backbone

Results show significant differences between subgroups for the different oral fluoropyrimidine backbones.

Pooled effect estimates for the ‘Capecitabine’ and ‘UFT/Ftorafur’ subgroups indicate that grade ≥ 3 diarrhoea did not differ between participants treated with oral versus IV fluoropyrimidines, whilst the OR for the only study in the ‘Doxifluridine’ subgroup (Kim 2001a) indicated that grade ≥ 3 diarrhoea was increased with oral fluoropyrimidines (OR 32.14, 95% CI 1.89 to 545.41). Tests for subgroup differences yielded these results: Chi² = 6.73, P = 0.03; I² = 70.3%. Substantial or considerable heterogeneity remained between studies within the ‘Capecitabine’ subgroup (Chi² = 16.27, P = 0.003; I² = 75%) (Analysis 3.4; Appendix 10).

3.5 Grade ≥ 3 hand foot syndrome (curative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, odds of grade ≥ 3 HFS were higher with oral fluoropyrimidine treatment. The pooled OR from five studies with 5731 participants was 4.59 (95% CI 2.97 to 7.10) (Analysis 3.5; Table 5).

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We further downgraded quality by one level for inconsistency of results, as we noted substantial or considerable heterogeneity among included studies (Chi² = 16.34, P = 0.003; I² = 76%). We assessed the quality of evidence as low (summary of findings Table for the main comparison).

In four of the included studies, effect estimates favoured IV fluoropyrimidines, and in three of these, 95% CIs crossed the null value of 1.00 (Allegra 2015; Hofheinz 2012; Pectasides 2015). In one outlier study (Shimada 2014), the effect estimate favoured oral fluoropyrimidines and the upper limit of the 95% CI was 1.00. It is unclear whether this variation in effects was due to clinical diversity (this was the only study for this outcome that utilised UFT and enrolled patients only from Japan) and/or methodological diversity (AE assessments were less frequent in the oral than in the IV treatment arm). In a post hoc sensitivity analysis in which we incorporated heterogeneity into a random‐effects model meta‐analysis, the pooled OR was 2.36 and the 95% confidence interval crossed the null value of 1.00 (95% CI 0.52 to 10.74).

3.6 Grade ≥ 3 neutropenia/granulocytopenia (curative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, the pooled OR for grade ≥ 3 neutropenia/granulocytopenia from seven studies with 8087 participants was 0.14 (95% CI 0.11 to 0.16), favouring oral fluoropyrimidines (Table 5).

We downgraded the quality of evidence by one level for inconsistency of results, as we noted substantial or considerable heterogeneity between the included studies (Chi² = 53.38, P < 0.00001, I² = 89%). We assessed the quality of evidence as moderate (summary of findings Table for the main comparison).

The 95% CIs for effect estimates either favoured the oral fluoropyrimidine group (four studies) or crossed the null value of 1.00 (two studies). Only one outlier study (Allegra 2015) reported that the effect estimate and the 95% CI indicated more grade ≥ 3 neutropenia/granulocytopenia with oral fluoropyrimidine treatment (the only study for this outcome that included combination chemotherapy with radiotherapy) (Analysis 3.6).

3.7 Grade ≥ 3 febrile neutropenia (curative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, grade ≥ 3 febrile neutropenia events were few in the four studies with 2925 participants that reported this outcome (Analysis 3.7; Table 5). The pooled OR was 0.59 (95% CI 0.18 to 1.90), and we observed no heterogeneity (Chi² = 2.65, P = 0.45; I² = 0%).

We downgraded the quality of evidence by two levels for imprecision (small number of events and 95% CI included appreciable benefit and harm) and assessed quality as low.

3.8 Grade ≥ 3 vomiting (curative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, grade ≥ 3 vomiting did not differ between participants treated with oral versus IV fluoropyrimidines. The pooled OR from eight studies with 9385 participants was 1.05 (95% CI 0.83 to 1.34) (Analysis 3.8; Table 5).

We downgraded the quality of the evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We downgraded quality by one further level for imprecision. The final assessment for quality of evidence was low.

Heterogeneity among effect estimates for these studies was moderate (Chi² = 10.75, P = 0.10; I² = 44%), albeit not statistically significant. However, most of the effect estimates with their 95% CIs crossed the null value of 1.00, with the exception of one outlier study (Lembersky 2006), for which the effect estimate favoured oral fluoropyrimidines.

3.9 Grade ≥ 3 nausea (curative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, grade ≥ 3 nausea did not differ between participants treated with oral versus IV fluoropyrimidines. The pooled OR from seven studies with 9233 participants was 1.21 (95% CI 0.97 to 1.51) (Analysis 3.9; Table 5). Heterogeneity among effect estimates for these studies was minimal (Chi² = 6.40, P = 0.38, I² = 6%).

We downgraded the quality of the evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We downgraded quality by one further level for imprecision. The final assessment for quality of evidence was low.

3.10 Grade ≥ 3 stomatitis (curative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, the pooled OR for grade ≥ 3 stomatitis from five studies with 4212 participants was 0.21 (95% CI 0.14 to 0.30), favouring oral fluoropyrimidines (Analysis 3.10; Table 5).

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We downgraded quality by one further level for inconsistency of results, as we noted substantial or considerable heterogeneity between the included studies (Chi² = 26.70, P < 0.00001; I² = 89%). We assessed the quality of evidence as low. However, in the included studies, 95% CIs for effect estimates either crossed the null value of 1.00 (three studies) or favoured oral fluoropyrimidines (one study, Twelves 2012).

3.11 Grade ≥ 3 mucositis (curative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, grade ≥ 3 mucositis did not differ between participants treated with oral versus IV fluoropyrimidines. The pooled OR from four studies with 2233 participants was 0.64 (95% CI 0.25 to 1.62) (Analysis 3.11; Table 5). We noted no heterogeneity among effect estimates for these studies (Chi² = 1.56, P = 0.67; I² = 0%).

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We downgraded quality by two further levels for imprecision (small number of events and 95% CI included appreciable benefit and harm). We assessed the quality of evidence as very low.

3.12 Grade ≥ 3 hyperbilirubinaemia (curative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, grade ≥ 3 hyperbilirubinaemia did not differ between participants treated with oral versus IV fluoropyrimidines. The OR of 1.67 (95% CI 0.52 to 5.38) was derived from three studies with 2757 participants (Analysis 3.12; Table 5). However, few events occurred in both arms. Heterogeneity between effect estimates was moderate for these studies (Chi² = 3.45, P = 0.18; I² = 42%).

We downgraded the quality of evidence by one level for risk of bias, as studies at high risk of bias for this outcome contributed 44.3% of the weight for the pooled effect estimate. We downgraded quality by two further levels owing to imprecision (small numbers of events and 95% CIs included appreciable benefit and harm). We assessed the quality of evidence as very low.

3.13 Any grade ≥ 3 AEs (curative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with curative intent for CRC with neoadjuvant and/or adjuvant chemotherapy, we found that odds of any grade ≥ 3 AEs were lower with oral fluoropyrimidines, with a pooled OR of 0.82 (95% CI 0.74 to 0.90) from five studies with 7741 participants (Table 5).

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias. We downgraded quality by one further level for inconsistency of results, as heterogeneity among the included studies was considerable (Chi² = 99.17, P < 0.00001; I² = 96%). We assessed the quality of evidence as low.

The effect estimate for De Gramont 2012 (weight 33.6%) strongly favoured the oral fluoropyrimidine group (HR 0.32, 95% CI 0.26 to 0.39), and in the remaining four studies, 95% CIs for the effect estimates crossed the null value of 1.00 (Analysis 3.13).

Excluding studies at 'High' risk of bias

As we assessed all studies contributing to the DFS outcome as having 'High' risk of bias owing to lack of blinding (Table 7), we could not perform a sensitivity analysis that excluded studies at 'High' risk of bias.

4.1 PFS

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, PFS was worse in the oral fluoropyrimidine group. The pooled HR from 23 studies with 9927 participants was 1.06 (95% CI 1.02 to 1.11) (Analysis 4.1; Table 6). Heterogeneity among effect estimates for these studies was minimal (Chi² = 27.08, P = 0.25; I² = 15%).

We downgraded the quality of evidence by one level for risk of bias, as studies at high risk of bias for this outcome contributed 48.5% of the weight for the pooled effect estimate. We assessed the quality of evidence as moderate (summary of findings Table 2).

4.2 PFS with subgroup analysis ‐ Single agent versus combination therapy

We found no evidence of subgroup differences (Chi² = 2.16, P = 0.14; I² = 53.8%) (Analysis 4.2; Appendix 11).

4.3 PFS with subgroup analysis ‐ Infusional versus bolus intravenous fluoropyrimidine

We found no evidence of subgroup differences (Chi² = 1.33, P = 0.25; I² = 24.7%) (Analysis 4.3; Appendix 11).

4.4 PFS with subgroup analysis ‐ Oral fluoropyrimidine backbone

Results showed significant subgroup differences by oral fluoropyrimidine backbone (Chi² = 13.46, P = 0.009; I² =70.3%). Pooled effect estimates for the ‘Capecitabine’ and ‘S‐1’ subgroups and the effect estimate for the ‘Doxifluridine’ subgroup (one study, Bajetta 1996) indicated that PFS did not differ between participants treated with oral versus IV fluoropyrimidines. However, pooled effect estimates for the ‘UFT/Ftorafur’ and ‘Eniluracil + oral 5‐FU’ subgroups indicated worse PFS in the oral fluoropyrimidine group (Figure 5; Analysis 4.4; Appendix 11).

Figure 5

---

---

Forest plot of comparison: 4 Progression‐free survival with, outcome: 4.4 Progression‐free survival with subgroup analysis ‐ oral fluoropyrimidine backbone.

4.5 PFS for combination therapy with subgroup analysis ‐ Oxaliplatin‐based versus irinotecan based

We found no evidence of subgroup differences (Chi² = 0.13, P = 0.72; I² = 0%) (Analysis 4.5; Appendix 11).

4.6 PFS for combination therapy with subgroup analysis ‐ with bevacizumab versus no bevacizumab

The post hoc subgroup analysis comparing studies of combination chemotherapy that included BEV versus those that did not include BEV found no subgroup differences (Chi² = 1.12, P = 0.29; I² = 11.0%) (Analysis 4.6; Appendix 11).

Assessment of publication bias for PFS

Visual inspection of a funnel plot of SE(lnHR)s against HRs for the 23 studies quantitatively synthesised for the PFS outcome revealed no asymmetry (Figure 6).

Figure 6

---

---

Funnel plot of progression‐free survival.

5.1 OS (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, OS did not differ between participants treated with oral versus IV fluoropyrimidines. The pooled HR from 29 studies with 12,079 participants was 1.02 (95% CI 0.99 to 1.05) (Table 6). Heterogeneity among effect estimates for these studies was minimal (Chi² = 33.69, P = 0.29; I² = 11%) (Analysis 5.1).

We did not identify any factors that reduced the quality of evidence for this outcome, and we assessed the quality of evidence as high (summary of findings Table 2).

5.2 OS with subgroup analysis ‐ Single‐agent versus combination therapy

Chi² = 0.40, P = 0.53; I² = 0%.

5.3 OS with subgroup analysis ‐ Infusional versus bolus intravenous fluoropyrimidine

Chi² = 0.10, P = 0.75; I² = 0%.

5.4 OS with subgroup analysis ‐ Oral fluoropyrimidine backbone

Chi² = 9.30, P = 0.05; I² = 57.0%.

However, the pooled effect estimate for the 'Capecitabine', 'UFT/Ftorafur', 'Doxifluridine', and 'S‐1' subgroups indicated that OS did not differ between participants treated with oral versus IV fluoropyrimidines, whereas the pooled effect estimate for the 'Eniluracil + oral 5‐FU' subgroup indicated a worse OS in the oral fluoropyrimidine group (Analysis 5.4).

5.5 OS for combination therapy with subgroup analysis‐ Oxaliplatin‐based versus irinotecan‐based

Chi² = 0.01, P = 0.90; I² = 0%.

6.1 TTP

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, TTP was worse in the oral fluoropyrimidine group. The pooled HR from six studies with 1970 participants was 1.07 (95% CI 1.01 to 1.14) (Analysis 6.1; Table 6). We noted no heterogeneity among effect estimates for these studies (Chi² = 4.95, P = 0.42; I² = 0%).

We downgraded the quality of evidence by one level for risk of bias, as studies at high risk of bias for this outcome contributed 93.4% of the weight for the pooled effect estimate. We assessed the quality of evidence as moderate.

7.1 ORR

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, ORR did not differ between participants treated with oral versus IV fluoropyrimidines. The pooled OR from 32 studies with 11,115 participants was 0.98 (95% CI 0.90 to 1.06) (Analysis 7.1; Table 6). Heterogeneity between the included studies was moderate (Chi² = 59.03, P = 0.005; I² = 42%).

We downgraded the quality of evidence by one level for risk of bias, as studies at high risk of bias for this outcome contributed 59.3% of the weight for the pooled effect estimate. We assessed the quality of evidence as moderate.

8.1 Grade ≥ 3 diarrhoea (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, odds of grade ≥ 3 diarrhoea were higher in the oral fluoropyrimidine arm. The pooled OR from 30 studies with 11,997 participants was 1.66 (95% CI 1.50 to 1.84) (Analysis 8.1; Table 6).

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We further downgraded quality by one level for inconsistency, as heterogeneity between the included studies was substantial (Chi² = 101.41, P < 0.00001; I² = 67%). The final assessment for quality of evidence was low (summary of findings Table 2).

We observed that for the included studies, 95% CIs for the effect estimates either crossed the null value of 1.00 or indicated more grade ≥ 3 diarrhoea with oral fluoropyrimidine treatment. One outlier study, which was an exception to this, favoured oral fluoropyrimidines (Diaz‐Rubio 2007). In this study, AE assessments were less frequent in the oral than in the IV treatment arm; however, many other studies included in the meta‐analysis for grade ≥ 3 diarrhoea also had high risk of bias as a result of this methodological issue (Characteristics of included studies).

8.2 Grade ≥ 3 diarrhoea (palliative intent studies) with subgroup analysis ‐ Single‐agent versus combination therapy

The pooled OR for the ‘Combination therapy’ subgroup favoured IV fluoropyrimidines more than the pooled OR for the ‘Single agent’ subgroup (Chi² = 21.70, P < 0.00001; I² = 95.4%) (Analysis 8.2).

8.3 Grade ≥ 3 diarrhoea (palliative intent studies) with subgroup analysis ‐ Infusional versus bolus IV fluoropyrimidine

The pooled OR for the ‘Infusional IV fluoropyrimidine’ subgroup favoured IV fluoropyrimidines more than the pooled effect estimate for the ‘Bolus IV fluoropyrimidine’ subgroup (Chi² = 15.57, P < 0.0001; I² = 93.6%) (Analysis 8.3).

8.4 Grade ≥ 3 diarrhoea (palliative intent studies) with subgroup analysis ‐ Oral fluoropyrimidine backbone

Results showed significant subgroup differences by oral fluoropyrimidine backbone (Chi² = 21.15, P = 0.0003; I² = 81.1%). The pooled OR for the ‘Capecitabine’, ‘UFT/Ftorafur’ and ‘S‐1’ subgroups indicated worse grade ≥ 3 diarrhoea with oral fluoropyrimidine treatment, and 95% CIs for pooled effect estimates for the ‘Eniluracil + oral 5‐FU’ and ‘Doxifluridine’ (one study, Bajetta 1996) subgroups crossed the null value of 1.00 (Analysis 8.4).

8.5 Grade ≥ 3 diarrhoea (palliative intent studies) with subgroup analysis for combination therapy ‐ Oxaliplatin‐based versus irinotecan‐based

The pooled OR for the ‘Irinotecan‐based’ subgroup favoured IV fluoropyrimidines more than the pooled effect estimate for the ‘Oxaliplatin‐based’ subgroup (Chi² = 12.72, P = 0.0004; I² = 92.1% ) (Analysis 8.5).

8.6 Grade ≥ 3 hand foot syndrome (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, results showed greater grade ≥ 3 HFS with oral fluoropyrimidine use. The pooled OR from 18 studies with 6481 participants was 3.92 (95% CI 2.84 to 5.43) (Table 6). Heterogeneity between the included studies was moderate (Chi² = 33.79, P = 0.03; I² = 41%).

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We assessed the quality of evidence as moderate (summary of findings Table 2).

We observed that for the included studies, effect estimates with their 95% CIs crossed the null value of 1.00 (10 studies, and one arm of Hochster TREE‐1 2008 and Hochster TREE‐2 2008) or indicated increased grade ≥ 3 HFS with oral fluoropyrimidine treatment (four studies, and one arm of Hochster TREE‐1 2008 and Hochster TREE‐2 2008). One outlier, which was an exception to this, favoured oral fluoropyrimidines (ECOG E5296 2012, the only study for this outcome using Eniluracil + oral 5‐FU). Another study (Shigeta 2016) reported no events in either arm (Analysis 8.6).

8.7 Grade ≥ 3 hand foot syndrome (palliative intent studies) subgroup analysis ‐ Single‐agent versus combination therapy

Results showed subgroup differences in the 'Single‐agent' and 'Combination therapy' subgroups. In the 'Single‐agent' subgroup, grade ≥ 3 HFS did not differ between participants treated with oral versus IV fluoropyrimidines. However, the 'Combination therapy' subgroup showed an increase in grade ≥ 3 HFS with oral fluoropyrimidine treatment (Chi² = 9.86, P = 0.002; I² = 89.9%). Only two studies were included in the 'Single‐agent' subgroup (one was ECOG E5296 2012, the outlier study), and heterogeneity between these two studies was considerable (Chi² = 9.56, P = 0.002; I² = 90%) (Analysis 8.7; Appendix 12).

8.8 Grade ≥ 3 hand foot syndrome (palliative intent studies) subgroup analysis ‐ Infusional versus bolus IV fluoropyrimidine

The pooled OR for the ‘Bolus IV fluoropyrimidine’ subgroup favoured IV fluoropyrimidines more than the pooled effect estimate for the ‘Infusional IV fluoropyrimidine’ subgroup (Chi² = 4.48, P = 0.03; I² = 77.7%) (Analysis 8.8; Appendix 12). However, heterogeneity between studies within the ‘Infusional IV fluoropyrimidines’ subgroup was moderate (Chi² = 30.02, P = 0.03; I² = 43%).

8.9 Grade ≥ 3 hand foot syndrome (palliative intent studies) subgroup analysis ‐ Oral fluoropyrimidine backbone

The effect estimate for the ‘Eniluracil + oral 5‐FU’ subgroup (one study, ECOG E5296 2012) favoured oral fluoropyrimidines, the 95% CI for pooled effect estimates for the ‘UFT/Ftorafur’ and ‘S‐1’ subgroups crossed the null value of 1.00, and the pooled OR for the ‘Capecitabine’ subgroup indicated increased grade ≥ 3 HFS with oral fluoropyrimidine treatment (Chi² = 19.58, P = 0.0002; I² = 84.7%) (Analysis 8.9; Appendix 12).

8.10 Grade ≥ 3 hand foot syndrome (palliative intent studies) subgroup analysis for combination therapy ‐ Oxaliplatin‐based versus irinotecan‐based

We found no evidence of subgroup differences (Chi² = 0.32, P = 0.57; I² = 0%) (Analysis 8.10; Appendix 12).

8.11 Grade ≥ 3 neutropenia/granulocytopenia (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, the pooled OR for grade ≥ 3 neutropenia/granulocytopenia from 29 studies with 11,794 participants (Table 6) was 0.17 (95% CI 0.15 to 0.18), favouring oral fluoropyrimidines.

We downgraded the quality of evidence by one level for risk of bias, as studies at high risk of bias for this outcome contributed 29.2% of the weight for the pooled effect estimate. We further downgraded quality by one level for inconsistency of results, as heterogeneity between included studies was substantial to considerable (Chi² = 295.88, P < 0.00001; I² = 90%). We assessed the quality of evidence as low (summary of findings Table 2).

We observed that for the included studies, effect estimates with their 95% CIs either favoured oral fluoropyrimidines (14 studies and infusional arms of Hochster TREE‐1 2008 and Hochster TREE‐2 2008 studies), or included the null value of 1.00 (13 studies and bolus arms of Hochster TREE‐1 2008 and Hochster TREE‐2 2008 studies) (Analysis 8.11).

8.12 Grade ≥ 3 febrile neutropenia (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, the pooled OR for grade ≥ 3 febrile neutropenia from 19 studies with 9407 participants was 0.27 (95% CI 0.21 to 0.36), indicating lower odds of grade ≥ 3 febrile neutropenia in the oral fluoropyrimidine arm (Table 6).

We downgraded the quality of evidence by one level owing to inconsistency of results, with substantial heterogeneity between the included studies (Chi² = 60.67, P < 0.00001; I² = 67%). We assessed the quality of evidence as moderate.

However, we observed that for the included studies, effect estimates with their 95% CIs either crossed the null value of 1.00 (11 studies) or favoured oral fluoropyrimidines (seven studies), with the exception of one outlier study (Yasui 2015) in which participants treated with oral fluoropyrimidines had greater grade ≥ 3 febrile neutropenia (Analysis 8.12).

8.13 Grade ≥ 3 vomiting (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, the pooled OR of 1.18 (95% CI 1.00 to 1.40) from 23 studies with 9528 participants indicated higher odds of vomiting with oral fluoropyrimidine treatment (Analysis 8.13; Table 6). Heterogeneity among the effect estimates for these studies was minimal (Chi² = 32.08, P = 0.19; I² = 19%). This pooled OR included data from seven studies that combined data for grade ≥ 3 vomiting and nausea (Carmichael 2002; Cassidy 2011a; Douillard 2002; Hochster TREE‐1 2008; Hochster TREE‐2 2008; Nogue 2005; Mei 2014).

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We downgraded quality by one further level as the result of imprecision, and the final assessment for quality of evidence was low.

8.14 Grade ≥ 3 nausea (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, grade ≥ 3 nausea did not differ between participants treated with oral versus IV fluoropyrimidines. The pooled OR from 25 studies with 9796 participants was 1.16 (95% CI 0.99 to 1.36) (Table 6). Heterogeneity between studies was moderate (Chi² = 48.04, P = 0.01; I² = 42%). However, for the included studies, effect estimates with their 95% CIs either crossed the null value of 1.00 or indicated higher odds of grade ≥ 3 nausea with oral fluoropyrimidine treatment, with the exception of two outlier studies (Mei 2014; Schilsky 2002a) (Analysis 8.14).

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We downgraded quality by one further level owing to imprecision. The final assessment for quality of evidence was low.

8.15 Grade ≥ 3 stomatitis (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, the pooled OR for grade ≥ 3 stomatitis from 21 studies with 8718 participants (Table 6) was 0.26 (95% CI 0.20 to 0.33), favouring oral fluoropyrimidines.

We downgraded the quality of evidence by one level owing to risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We downgraded quality by one further level for inconsistency of results, with substantial heterogeneity between the included studies (Chi² = 62.38, P < 0.00001; I² = 66%). We assessed the quality of evidence as low.

We observed that for the included studies, effect estimates with their 95% CIs either crossed the null value of 1.00 (15 studies) or favoured oral fluoropyrimidines (six studies) (Analysis 8.15).

8.16 Grade ≥ 3 mucositis (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, the pooled OR for grade ≥ 3 mucositis was 0.17 (95% CI 0.12 to 0.24) from 12 studies with 4962 participants (Table 6), favouring oral fluoropyrimidines.

We downgraded the quality of evidence by one level owing to risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We further downgraded quality by one level owing to inconsistency of results, as heterogeneity between the included studies was substantial or considerable (Chi² = 39.81, P < 0.0001; I² = 75%). We assessed the quality of evidence as low.

We observed that for the included studies, effect estimates with their 95% CIs either crossed the null value of 1.00 (seven studies) or favoured oral fluoropyrimidines (four studies) (Analysis 8.16).

8.17 Grade ≥ 3 hyperbilirubinaemia (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, grade ≥ 3 hyperbilirubinaemia did not differ between oral and IV fluoropyrimidine arms. The pooled OR from nine studies with 2699 participants was 1.62 (95% CI 0.99 to 2.64). We noted no heterogeneity between the included studies (Chi² = 3.86, P = 0.70; I² = 0%) (Analysis 8.17; Table 6).

We downgraded the quality of evidence by one level owing to risk of bias, as studies at high risk of bias for this outcome contributed 28.5% of the weight for the pooled effect estimate. We downgraded quality by one further level for imprecision, and we assessed the quality of evidence as low.

8.18 Any grade ≥ 3 AEs (palliative intent studies)

For the comparison of oral versus IV fluoropyrimidines in patients treated with palliative intent for CRC with chemotherapy, the pooled OR for any grade ≥ 3 AEs from 14 studies with 5436 participants was 0.83 (95% CI 0.74 to 0.94), favouring oral fluoropyrimidines (Table 6).

We downgraded the quality of evidence by one level for risk of bias, as all studies that contributed to the pooled effect estimate had high risk of bias owing to lack of blinding. We further downgraded quality by one level for inconsistency of results, as heterogeneity between the included studies was substantial or considerable (Chi² = 69.88, P < 0.00001; I² = 77%). We assessed the quality of evidence as low.

We observed that for the included studies, 95% CIs for the effect estimates crossed the null value of 1.00 or favoured oral fluoropyrimidines, with the exception of the bolus arm of Hochster TREE‐1 2008, Kohne 2008, and Seymour 2011 (Analysis 8.18).

Excluding studies with 'High' risk of bias

When we excluded studies with 'High' risk of bias from the meta‐analysis for the PFS outcome (Table 8), the pooled HR was 1.01 (95% CI 0.96 to 1.07). Whilst results showed no substantial change in the direction or magnitude of the effect estimate compared with the original analysis, which included studies at 'High' risk of bias (HR 1.06, 95% CI 1.02 to 1.11), the 95% CI included the null value of 1.00 (Table 9). We found no heterogeneity (I² = 0%, P = 0.91).

Other sensitivity analyses

Results showed no change in direction nor substantial change in magnitude of the pooled effect estimate for PFS when we performed sensitivity analyses excluding the Seymour 2011 study (with a frail and elderly study population) or excluding studies of second‐line chemotherapy (Table 9).