Meta-analysis: evaluation of adjuvant therapy after curative liver resection for hepatocellular carcinoma

Authors


Dr P. Mathurin, Service d'Hépatogastroentérologie, Hôpital Claude Huriez 2ème étage Est, Avenue Michel Polonovoski, CHRU Lille 59037, France.
E-mail: p-mathurin@chru-lille.fr

Summary

Aim : To evaluate adjuvant modalities after curative resection for hepatocellular carcinoma using a meta-analysis of randomized and non-randomized controlled trials.

Methods : In a first step, a meta-analysis of randomized controlled trials was carried out. Sensitivity analyses after inclusion of non-randomized controlled trials were performed. Four therapeutic modalities were evaluated: pre-operative transarterial chemotherapy, post-operative transarterial chemotherapy, systemic chemotherapy and a combination of systemic and transarterial chemotherapy.

Results : Only post-operative transarterial chemotherapy improved survival significantly at 2 years [difference, 22.8%; confidence interval (CI), 8.6–36.9%; P = 0.002] and 3 years (difference, 27.6%; CI, 8.2–47.1%; P = 0.005), and decreased the probability of no recurrence at 1 year (difference, 28.8%; CI, 16.7–40.8%; P < 0.001), 2 years (difference, 27.6%; CI, 8.2–47.1%; P = 0.005) and 3 years (difference, 28%; CI, 8.2–47.9%; P = 0.006). In a sensitivity analysis after inclusion of non-randomized controlled trials, post-operative transarterial chemotherapy still improved survival at 1 year (difference, 9.6%; CI, 0.8–18.3%; P = 0.03), 2 years (difference, 13.5%; CI, 0.9–26%, P = 0.04) and 3 years (difference, 18%; CI, 7–28.9%; P < 0.001), and decreased the probability of no recurrence at 1 year (difference, 20.3%; CI, 7.7–33%; P = 0.002), 2 years (difference, 35%; CI, 21.4–46.3%; P < 0.001) and 3 years (difference, 34.5%; CI, 18.7–50.3%; P < 0.001).

Conclusion : Post-operative transarterial chemotherapy improved survival and decreased the cumulative probability of no recurrence. New randomized controlled trials evaluating this modality are required.

Introduction

Hepatocellular carcinoma (HCC) is one of the most common malignancies in the world. In most cases, HCC develops in cirrhotic patients, with an annual incidence rate of approximately 3%.1 Despite immunization strategies and anti-viral therapy for hepatitis B and C, the incidence of HCC will increase dramatically over the next few decades. The treatment of HCC remains controversial and constitutes one of the main challenges in the near future. In patients with advanced tumours, two meta-analyses observed that no treatment has clearly proven efficacy in survival.2, 3 A recent systematic review, including several new randomized controlled trials (RCTs) published since these two meta-analyses were reported, showed a significant benefit of chemo-embolization with cisplatin or doxorubicin [odds ratio (OR), 0.42; 95% confidence interval (CI), 0.32–0.89; P = 0.017], but none with embolization (OR, 0.59; 95% CI, 0.29–1.2).4 The authors proposed chemo-embolization as the standard intervention in patients with unresectable HCC.4

Surgical resection, percutaneous ethanol injection and orthotopic liver transplantation are considered as first-line options to obtain disease-free survival in early HCC.5–10 Percutaneous ethanol injection has a similar efficacy to surgical resection and is highly effective for tumours of ≤ 3 cm.5, 11 These curative treatments have been shown to improve the natural history of patients with a single tumour smaller than 5 cm.12

Controversies continue to persist concerning the best option between surgical resection and orthotopic liver transplantation.10 When considering the high rate of tumour recurrence after resection, orthotopic liver transplantation was believed to be superior in a selective sub-group of patients.6 However, a study comparing these two modalities using an intention-to-treat analysis did not observe any significant difference in patient survival.13 The authors observed a 2.5-fold increase in waiting time during the follow-up period that considerably affected survival. The scarcity of donors largely explains why liver resection remains the first-line option for surgical procedures world-wide.

Although resection is complete and is carried out at an early stage of the disease, there is still a high rate of early tumour recurrence that represents the main drawback of tumour resection. The cumulative recurrence rate at 3 years is approximately 80%.14 With regard to effective percutaneous treatment, recurrence is as frequent as that after surgical resection.15, 16 Even in carefully selected patients, at 3 and 5 years, tumour recurrence is still observed in 50% and 70% of cases, respectively.13, 17, 18

To our knowledge, the usefulness of adjuvant therapy after curative liver resection is still unknown. Several RCTs and non-randomized controlled trials (NRCTs) have evaluated the efficacy of adjuvant therapies. From these studies, no clear recommendations have been proposed.10 Most often, the small sample size of the RCTs evaluating adjuvant therapy has resulted in a lack of power to detect clinically meaningful differences. Meta-analysis, a quantitative technique for therapeutic evaluation, may be used when controversy persists after several trials.19 It is particularly useful when several trials have insufficient statistical power, as the pooling of trials decreases the random error. Although meta-analysis has traditionally been applied and is best confined to RCTs, meta-analytical techniques using NRCTs20 might be a valid method in some clinical settings in which either the number or the sample size of RCTs is insufficient.21

In the present study, we sought to identify adjuvant therapeutic modalities that improved the overall survival and cumulative probability of no recurrence in patients who had undergone curative liver resection for HCC. For this purpose, we used an evidence-based approach consisting of a meta-analysis of RCTs and NRCTs.

Methods

Literature research

Medline, Cancerlit, Embase and manual searches were combined, as Medline search alone was not sufficiently sensitive.22 General reviews and references from published RCTs and NRCTs were also used.7, 10, 13, 14, 23–34

Criteria for inclusion and exclusion

In order to be included, a study had to fulfil the following criteria: (i) comparison of an adjuvant therapy with the absence of treatment or with another therapeutic modality in patients who had undergone curative liver resection for HCC; (ii) full length papers reporting RCTs or NRCTs; (iii) use of overall survival or cumulative probability of no recurrence at 1 and 2 years (at least) as clinical end-points.

The following studies were excluded: (i) those evaluating treatment in patients with unresectable HCC2 or evaluating adjuvant therapies after non-curative resection; (ii) those with no information on survival or the cumulative probability of no recurrence; (iii) those evaluating patients with cholangiocellular carcinomas or liver metastases.

Criteria for determining whether the studies could be combined

Studies were combined when they investigated a similar modality of adjuvant therapy. The wide heterogeneity between the protocols for each modality led us to combine studies with different protocols as long as they evaluated the same route of administration. To assess the combinable nature of the studies for each modality evaluated, the overall survival and the cumulative probability of no recurrence at 1, 2 and 3 years of control groups were compared using the chi-squared test.

Review of the studies

All studies considered for inclusion were analysed independently by three observers (PM, BR, SK). For RCTs, methodological quality assessment was performed independently by two observers using a validated questionnaire.2, 35

Statistical methods

The intention-to-treat method was used. When actuarial survival was given, the number of deaths or recurrences was extrapolated from the curves. For each meta-analysis, the following methods were used: (i) assessment of the heterogeneity of results between control groups; (ii) assessment of efficacy by the methods of Peto et al.36 and Der Simonian and Laird.37

The Peto et al. method is a variant of the classical Mantel–Haenszel method. In each study, quantification of the size of the difference between the two groups was performed using the OR with an estimate of the 95% CI. An estimation of the overall size of the difference between the groups was performed using CI, a graphical representation, a test of statistical significance of the overall difference and a test of the homogeneity of the results obtained in individual studies.

In the Der Simonian and Laird method, the studies were considered as a random sample from a population of studies. The random effects model incorporates the heterogeneity of the studies. The overall treatment effect was estimated by a weighted average of the individual effects, with weights inversely proportional to the variance of the observed effects. The estimate was expressed by the difference in percentages (risk of treated − risk of controls). A significance level of 5% was taken as the alpha risk. Comparison of the percentages between states was performed using their 95% CI. Detailed results were given only for the Der Simonian and Laird analyses.

Only meta-analyses of therapeutic modalities that included two or more studies (RCTs or NRCTs) were performed. The core analysis group was defined as the group of studies that included the ‘pure’ control group using no treatment.

The meta-analysis of RCTs from the core group was performed in a first step. According to pre-established criteria, the following sensitivity analyses were performed: (i) after the inclusion of NRCTs in order to perform a meta-analysis restricted to RCTs and NRCTs which included a ‘pure’ control group; (ii) after the inclusion of RCTs and NRCTs in which the control group received treatment.

Results

Of the 46 references identified, 25 studies (13 RCTs and 12 NRCTs) were excluded for the following reasons: three RCTs evaluating nutritional support,38–40 seven therapeutic modalities that were evaluated in fewer than two studies,41–47 three RCTs or NRCTs without data on survival or the cumulative probability of no recurrence,48–50 one RCT without survival curves or approximate points of survival probabilities,51 four RCTs or NRCTs testing adjuvant therapy in addition to palliative resection or palliative treatments,52–55 five NRCTs and one RCT mixing palliative and curative resection56–61 and one NRCT comparing pre-operative transarterial chemotherapy without embolization with pre-operative transarterial chemotherapy with embolization.62 Two publications concerned the same study.60, 61

Twenty-one studies (10 RCTs and 11 NRCTs), published as full papers, were included. The wide variability in the overall survival and cumulative probability of no recurrence of the patients enrolled in the RCTs and NRCTs was demonstrated by the assessment of the untreated control arms (data not shown).

We identified four different strategies that were evaluated in two studies or more: pre-operative transarterial chemotherapy whatever the modality (transarterial chemotherapy with or without embolization, lipiodol transarterial chemotherapy with or without embolization, iodine-131-labelled lipiodol), post-operative transarterial chemotherapy whatever the modality, systemic administration of chemotherapy and a combination of systemic chemotherapy with transarterial chemotherapy. The characteristics of the studies are described in Tables 1–4.

Table 1.  Characteristics of the studies included, published as full papers and evaluating pre-operative transarterial chemotherapy
ReferenceRegimen in addition to curative resectionNo. of patientsDesignMultiple tumours (%) Mean tumour size (mean ± s.d.) (cm)Cirrhosis (%)Overall survival at 1/2/3 years (%)Cumulative probability of no recurrence at 1/2/3 years (%)
  • LTA, lipiodol transarterial chemotherapy without embolization; LTAE, lipiodol transarterial chemotherapy with embolization; Nd, not determined; NRCT, non-randomized controlled trial; RCT, randomized controlled trial; TA, transarterial chemotherapy without embolization; TAE, transarterial chemotherapy with embolization.

  • *

    Core group.

  •  Pre-operative transarterial chemotherapy combined with post-operative systemic chemotherapy was compared with post-operative systemic chemotherapy in some cases; in others, pre-operative transarterial chemotherapy was compared with no treatment.

  •  For more details, see text.

  • §

     Study including only patients with large resectable tumours.

Wu et al.63*Arm 1: pre-operative LTAE 18 weeks before surgery (20–30 mg doxorubicin mixed with 20–30 mL lipiodol and spongostan particles); mean 2.5 sessions carried out every 4–6 weeks 24RCT17.714.3 ± 4.2§58.367/46/3358/38/33
Arm 2: curative liver resection alone 28 21.414.5 ± 3.342.975/68/4357/43/39
Yamasaki et al.64*Arm 1: pre-operative LTAE (20 mg doxorubicin mixed with 5 mL lipiodol and 1–3 mm gelatin sponges); 1 session; exact timing was not given 50RCT103.1 ± 0.8Nd90/86/8076/56/46
Arm 2: curative liver resection alone 47 14.93.3 ± 0.9Nd87/81/7776.6/42.6/34
Harada et al.67*Arm 1: pre-operative TAE (10 mg mitomycin and 10 mg doxorubicin mixed with absorbable gelatin sponge particles); 1 session was performed an average of 79 days before hepatectomy 98NRCT39Nd in mean  ≤ 5 cm in 66% of cases Nd in mean65.790.8/84/7869.4/48/37.8
Arm 2: curative liver resection alone 33 37.1≤ 5 cm in 74% of cases68.697/88/6769.7/39.4/33.3
Uchida et al.70Arm 1: pre-operative LTA (5–35 mg mitomycin and/or 10–40 mg doxorubicin and/or 8–40 mg epirubicin mixed with 3.8 mL lipiodol in 70% of cases); 1 session was performed an average of 42 days before hepatectomy 60NRCT18.33.7 ± 3.17088.3/75/61Nd
Arm 2: curative liver resection alone 68 26.54.4 ± 3.47282/72/72Nd
Majno et al.66*Arm 1: pre-operative LTAE (50 mg doxorubicin or 50 mg cisplatin mixed with 10 mL lipiodol and gelatin sponge powder or gelatin pellets); 1, 2 and ≥ 3 sessions in 37%, 39% and 24% of cases; exact timing was not given 49NRCT20.55.05 ± 2.53100Nd63.3/42.9/32.7
Arm 2: curative liver resection alone 27 29.73.95 ± 1.89100Nd63/37/22.2
Adachi et al.65*Arm 1: pre-operative LTA, TA or LTAE (10–20 mg mitomycin and/or 10–40 mg doxorubicin mixed with 2–10 mL lipiodol and gelatin sponge cubes); 1 session was performed an average of 33 days before hepatectomy 46NRCT02.6 ± 0.963Nd89.1/76.1/52.2
Arm 2: curative liver resection alone 26 02.1 ± 0.984.6Nd88.5/65.4/50
Nagasue et al.68Arm 1: pre-operative LTA, TA or LTAE (6–20 mg mitomycin and/or 10–60 mg doxorubicin mixed with lipiodol (dosage: Nd) and gelatin sponges cubes or powder); 1 session was performed an average of 130 days before surgery 31NRCT29Nd83.974/61/32Nd
Arm 2: curative liver resection alone107 37.4Nd82.277/57/45Nd
Paye et al.69*Arm 1: pre-operative LTAE (40–60 mg doxorubicin mixed with 10–15 mL lipiodol and 1–3 mm gelatin sponges); number of sessions varied from 1 to 4; exact timing was not given 24NRCT37.57.8 ± 1Nd87.5/67/6362.5/41.7/29.2
Arm 2: curative liver resection alone 24 257.3 ± 1Nd70.8/71/6733.3/29.2/16.7
Imaoka et al.71*Arm 1: pre-operative LTAE (2 mg/kg cisplatin mixed with 10–20 mL lipiodol and gelatin sponges) in 37 patients and pre-operative LTAE (40–60 mg adriamycin mixed with 10–20 mL lipiodol and gelatin sponges) in 13 patients; 1 session; exact timing was not given 51NRCT Nd in mean  ≤ 10 cm in all casesNdNd84.3/64.7/Nd
Arm 2: curative liver resection alone 52  Nd in mean  ≤ 10 cm in all casesNdNd69.2/53.9/Nd
Table 2.  Characteristics of the studies included, published as full papers and evaluating post-operative transarterial chemotherapy
ReferenceRegimen in addition to curative resectionNo. of patientsDesignMultiple tumours (%)Mean tumour size (mean ± s.d.) (cm)Cirrhosis (%) Overall survival at 1/2/3 years (%)Cumulative probability of no recurrence at 1/2/3 years (%)
  • LTA, lipiodol transarterial chemotherapy without embolization; Nd, not determined; NRCT, non-randomized controlled trial; RCT, randomized controlled trial; TA, transarterial chemotherapy without embolization.

  • *

    Core group.

Lau et al.31*Arm 1: 1 session of post-operative LTA 131I (1850 Mbq131I mixed with 2 mL lipiodol); 1 session within 2 weeks of randomization21RCT33.34.4 (range, 1.5–10)Nd90.5/85.7/85.785.2/76.2/74.5
Arm 2: curative liver resection alone22 18.23.8 (range, 1.4–11)Nd86.4/63.6/45.559.1/36/36
Li et al.73*Arm 1: post-operative LTA (40 mg/m2 doxorubicin and 6 mg/m2 mitomycin mixed with 4–10 mL lipiodol); 1–3 sessions (first session 3–4 weeks after hepatectomy; interval between courses was 4–6 weeks)47RCTNdNdNd97.9/85.5/69.585.1/Nd/Nd
Arm 2: curative liver resection alone47 NdNdNd72.3/52.7/35.153.2/Nd/Nd
Izumi et al.72*Arm 1: 1 session of post-operative LTA (in 7 cases: 20 mg/m2 doxorubicin and 10 mg/m2 mitomycin C mixed with 3 mL/m2 lipiodol; in 16 cases: 20 mg/m2 doxorubicin and 10 mg/m2 mitomycin S mixed with 2 mL/m2 lipiodol); 1 session 38 days after resection (range, 21–84 days)23RCT73.9Nd82.687/74/56.664.5/54.9/32
Arm 2: curative liver resection alone27 40.7Nd81.581/66.6/53.443/22/11.7
Ueno et al.74*Arm 1: post-operative TA (50–80 mg/kg cisplatin and 10 mg/kg mitomycin C); 1–3 sessions (first session 1 month after hepatectomy; interval between courses was 1 month)10RCTNdNdNdNd80/70/Nd
Arm 2: curative liver resection alone11 NdNdNdNd45.5/27.3/Nd
Takenaka et al.75*Arm 1: 1–2 sessions of post-operative LTA (30–50 mg/kg epirubicin mixed with 2–4 mL lipiodol); exact timing was not given17NRCT17.63.8 ± 482.4100/100/10094.1/82/82
Arm 2: curative liver resection alone19 26.31.9 ± 0.88494.7/94.7/89.594.7/31.6/15
Asahara et al.77*Arm 1: 4–6 sessions of post-operative LTA (30–50 mg doxorubicin mixed with 3–5 mL lipiodol); first session 2–4 weeks after hepatectomy, then every 2–3 months for 1 year68NRCTNdNd in mean ≤ 10 cm (96% of cases)75Nd/Nd/79.1Nd/Nd/50.8
Arm 2: curative liver resection alone67 NdNd in mean ≤ 10 cm (97% of cases)72.7Nd/Nd/69.2Nd/Nd/25.7
Tanaka et al.76*Arm 1: post-operative LTA (10 mg adriamycin and 5 mg mitomycin mixed with 4 mL lipiodol); four sessions were performed every 6 months for 2 years after hepatectomy24NRCT25NdNd100/79.2/79.295.7/67/67
Arm 2: curative liver resection alone41 19.5NdNd89.7/74.2/62.861.6/38.5/30.3
Table 3.  Characteristics of the studies included, published as full papers and evaluating systemic chemotherapy
ReferenceRegimen in addition to curative resectionNo. of patientsDesignMultiple tumours (%) Mean tumour size (mean ± s.d.) (cm)Cirrhosis (%) Overall survival at 1/2/3 years (%)Cumulative probability of no recurrence at 1/2/3 years (%)
  • 5-FU, 5-fluorouracil; Nd, not determined; NRCT, non-randomized controlled trial; RCT, randomized controlled trial.

  • *

     Core group.

Yamamoto et al.78*Arm 1: oral 5-FU 400 mg/day35RCTNdNd in mean  ≤ 2 cm in all casesNd91.4/80/74.382.6/62.9/48.6
Arm 2: curative liver resection alone32 NdNd in mean  ≤ 2 cm in all casesNd81.3/71.9/59.468.8/37.5/25
Takenaka et al.75*Arm 1: oral 5-FU 300–400 mg/day12NRCT33.31.9 ± 0.866.7100/100/10083.3/58.3/50
Arm 2: curative liver resection alone19 26.31.9 ± 0.88494.7/94.7/89.594.7/31.6/15
Table 4.  Characteristics of the studies included, published as full papers and evaluating post-operative systemic chemotherapy and transarterial chemotherapy
ReferenceRegimen in addition to curative resectionNo. of patientsDesignMultiple tumours (%)Mean tumour size (mean ± s.d.) (cm)Cirrhosis (%)Overall survival at 1/2/3 years (%)Cumulative probability of no recurrence at 1/2/3 years (%)
  • 5-FU, 5-fluorouracil; Nd, not determined; RCT, randomized controlled trial; TA, transarterial chemotherapy without embolization.

  • *

     Core group.

Ono et al.80*Arm 1: intravenous epirubicin  40 mg/m2/3 months associated  with oral 5-FU (300 mg/day) for 2 years  in addition to 1 session of post-operative  TA 1 month after hepatectomy  (40 mg/m2 epirubicin)29RCTNd in %;  No. of tumours  in mean ± s.d.  1.4 ± 0.74.2 ± 3.965.593.1/Nd/72.267.8/Nd/32
Arm 2: curative liver resection alone27 Nd in %;  No. of tumours  in mean ± s.d.  1.2 ± 0.53.7 ± 1.670.496.3/Nd/57.188.5/Nd/42
Lai et al.79*Arm 1: Intravenous epirubicin 40 mg/m2/6  weeks (maximum 8 cycles) in addition to  3 sessions of post-operative TA every  2 months (10 mg cisplatin)30RCT36.78.5 (95% CI,  6.8–10)56.676.7/66.7/66.750/38/18
Arm 2: curative liver resection alone36 41.710.4 (95% CI,  5.2–15.6)52.894.4/89/6469/53/48
Khono et al.81Arm 1: 1 session of post-operative TA  on day 28 after hepatectomy  (40 mg/m2 epirubicin) and oral 5-FU  (300 mg/day for 1 year)48RCT20.8Nd in mean  ≤ 10 cm  (95% of cases)83.389.6/68.8/5070.8/50/37.5
Arm 2: curative liver resection  and oral 5-FU (300 mg/day for 1 year)40 17.5Nd in mean  ≤ 10 cm  (94% of cases)67.582.5/57.5/5065/42.5/32.5

Pre-operative transarterial chemotherapy

Two RCTs63, 64 and seven NRCTs65–71 evaluated pre-operative transarterial chemotherapy (Table 1).

Five studies used lipiodol and embolization,63, 64, 66, 69, 71 one embolization without lipiodol67 and one lipiodol without embolization.70 In two studies, transarterial chemotherapy was performed with embolization without lipiodol, with lipiodol without embolization or with lipiodol and embolization.65, 68

In two studies, pre-operative transarterial chemotherapy combined with post-operative systemic chemotherapy was compared with post-operative systemic chemotherapy in some cases, whilst, in other cases, pre-operative transarterial chemotherapy was compared with no treatment.68, 70 The study of Wu et al. included only patients with resectable large cancer (tumour diameter of 10 cm or more).63 The core group included seven studies.63–67, 69, 71

In the core analysis restricted to RCTs,63, 64 there was no significant difference in survival between the control groups at 1 year and 2 years, whereas the difference was significant at 3 years (χ2 = 8.7, P = 0.003). There was no significant heterogeneity in the cumulative probability of no recurrence between the control groups at 1 year (χ2 = 3.2, N.S.), 2 years (χ2 = 0.001, N.S.) or 3 years (χ2 = 0.21, N.S.). There was no significant effect of pre-operative transarterial chemotherapy on survival at 1 year (mean difference, 0.5%; CI, − 11% to 11.8%), 2 years (mean difference, − 6%; CI, − 32.4% to 20.1%) or 3 years (mean difference, − 0.2%; CI, − 14.1% to 13.7%). The benefit of pre-operative transarterial chemotherapy on the cumulative probability of no recurrence was not significant at 1 year (mean difference, − 0.1%; CI, − 14.4% to 14.3%), 2 years (mean difference, 6%; CI, − 11.6% to 24.2%) or 3 years (mean difference, 5.3%; CI, − 11.7% to 22.2%).

We performed two sensitivity analyses according to pre-established criteria (for more details, see ‘Statistical methods’ section).

  •  (a) In the core analysis including RCTs and NRCTs,63–67, 69, 71 the assessment of survival between the control groups showed significant heterogeneity at 1 year (χ2 = 9.2, P = 0.03) and 3 years (χ2 = 8.9, P = 0.03), but not at 2 years (χ2 = 4.5, N.S.). The assessment of the cumulative probability of no recurrence between the control groups showed significant heterogeneity at 1 year (χ2 = 21.4, P = 0.002), but not at 2 years (χ2 = 9.7, N.S.) or 3 years (χ2 = 8.2, N.S.). There was no significant effect of pre-operative transarterial chemotherapy on the survival at 1 year (mean difference, − 0.8%; CI, − 9.6% to 8.1%), 2 years (mean difference, − 3%; CI, − 12.2% to 6.1%) or 3 years (mean difference, 2.8%; CI, − 7.5% to 13%). The benefit of pre-operative transarterial chemotherapy on the cumulative probability of no recurrence was significant at 2 years (mean difference, 8.8%; CI, 0.6–17.1%; P = 0.03), but not at 1 year (mean difference, 5.2%; CI, − 2% to 12.4%) or 3 years (mean difference, 6.8%; CI, − 2% to 15.6%).

  •  (b) After inclusion of the studies by Nagasue et al.68 and Uchida et al.,70 in which the control group received treatment in some cases, there was no significant effect of pre-operative transarterial chemotherapy at 1 year (mean difference, − 0.1%; CI, − 6.2% to 6%), 2 years (mean difference, − 0.7%; CI, − 7.8% to 6.4%) or 3 years (mean difference, − 3%; CI, − 10.8% to 4.9%). The benefit of pre-operative transarterial chemotherapy on the cumulative probability of no recurrence was significant at 2 years (mean difference, 8.8%; CI, 0.6–17.1%; P = 0.03), but not at 1 year (mean difference, 5.2%; CI, − 2% to 12.4%) or 3 years (mean difference, 6.8%; CI, − 2% to 15.6%).

Post-operative transarterial chemotherapy

Four RCTs and three NRCTs evaluated post-operative transarterial chemotherapy. Three RCTs tested post-operative transarterial chemotherapy,72–74 one RCT tested post-operative iodine-131-labelled lipiodol transarterial chemotherapy31 and three NRCTs tested post-operative transarterial chemotherapy75–77 (Table 2).

Two studies used lipiodol and embolization,72, 73 three lipiodol without embolization75–77 and one without embolization or lipiodol.74 In the study by Li et al., patients with palliative and curative resection were randomized.73 Only the data of patients with curative resection were assessed in this meta-analysis.73 The core group included seven studies.31, 72–77

In the meta-analysis restricted to RCTs from the core group,31, 72–74 the assessment of survival between the control groups did not reveal heterogeneity at 1 year (χ2 = 2.32, N.S.), 2 years (χ2 = 1.51, N.S.) or 3 years (χ2 = 1.8, N.S.). The assessment of the cumulative probability of no recurrence between control groups did not reveal heterogeneity at 1 year (χ2 = 1.3, N.S.) or 2 years (χ2 = 1.4, N.S.), but there was significant heterogeneity at 3 years (χ2 = 4.45, P = 0.03).

There was a trend towards a survival difference at 1 year (mean difference, 13.2%; CI, − 1.8% to 28.2%; P = 0.08). There was a significant improvement in survival at 2 years (mean difference, 22.8%; CI, 8.6–36.9%; P = 0.002) and 3 years (mean difference, 27.6%; CI, 8.2–47.1%; P = 0.005). There was a significant improvement in the cumulative probability of no recurrence in treated patients at 1 year (mean difference, 28.8%; CI, 16.7–40.8%; P < 0.001), 2 years (mean difference, 27.6%; CI, 8.2–47.1%; P = 0.005) and 3 years (mean difference, 28%; CI, 8.2–47.9%; P = 0.006).

We performed a sensitivity analysis according to pre-established criteria. After the inclusion of NRCTs in the core analysis, the assessment of survival between the control groups showed a trend towards heterogeneity at 1 year (χ2 = 8.2, P = 0.08) and significant heterogeneity at 2 years (χ2 = 11.5, P = 0.02) and 3 years (χ2 = 22.4, P < 0.001).31, 72–77 The assessment of the cumulative probability of no recurrence between the control groups showed significant heterogeneity at 1 year (χ2 = 14.2, P = 0.01), but not at 2 years (χ2 = 2.1, N.S.) or 3 years (χ2 = 5.5, N.S.). The survival effect of post-operative transarterial chemotherapy (Figures 1–3) was significant at 1 year (mean difference, 9.6%; CI, 0.8–18.3%; P = 0.03), 2 years (mean difference, 13.5%; CI, 0.9–26%; P = 0.04) and 3 years (mean difference, 18%; CI, 7–28.9%; P < 0.001). The benefit of post-operative transarterial chemotherapy on the cumulative probability of no recurrence (Figures 4–6) was significant at 1 year (mean difference, 20.3%; CI, 7.7–33%; P = 0.002), 2 years (mean difference, 35%; CI, 21.4–46.3%; P < 0.001) and 3 years (mean difference, 34.5%; CI, 18.7–50.3%; P < 0.001).

Figure 1.

Randomized (RCTs) and non-randomized (NRCTs) controlled trials evaluating the effect of post-operative transarterial chemotherapy on the overall survival at 1 year. Results are represented using the Der Simonian and Laird method.

Figure 2.

Randomized (RCTs) and non-randomized (NRCTs) controlled trials evaluating the effect of post-operative transarterial chemotherapy on overall survival at 2 years. Results are represented using the Der Simonian and Laird method.

Figure 3.

Randomized (RCTs) and non-randomized (NRCTs) controlled trials evaluating the effect of post-operative transarterial chemotherapy on overall survival at 3 years. Results are represented using the Der Simonian and Laird method.

Figure 4.

Randomized (RCTs) and non-randomized (NRCTs) controlled trials evaluating the effect of post-operative transarterial chemotherapy on the cumulative probability of no recurrence at 1 year. Results are represented using the Der Simonian and Laird method.

Figure 5.

Randomized (RCTs) and non-randomized (NRCTs) controlled trials evaluating the effect of post-operative transarterial chemotherapy on the cumulative probability of no recurrence at 2 years. Results are represented using the Der Simonian and Laird method.

Figure 6.

Randomized (RCTs) and non-randomized (NRCTs) controlled trials evaluating the effect of post-operative transarterial chemotherapy on the cumulative probability of no recurrence at 3 years. Results are represented using the Der Simonian and Laird method.

Systemic chemotherapy [5-fluorouracil (5-FU) effect]

One RCT78 and one NRCT75 were included. Takenaka et al.'s NRCT was also used in the meta-analysis of post-operative transarterial chemotherapy, as this study compared three groups: a control group, an oral 5-FU group and a post-operative transarterial chemotherapy group75 (Table 3). In the present analysis of 5-FU, the 5-FU and control arms were used. The core group included the studies of Yamamoto et al. and Takenaka et al., which evaluated a 5-FU derivative given orally vs. a control arm receiving no treatment.75, 78 Meta-analysis restricted to RCTs was not possible.

In the core analysis, the assessment of survival between the control groups showed significant heterogeneity at 1 year (χ2 = 4.6, P = 0.03), 2 years (χ2 = 7, P = 0.008) and 3 years (χ2 = 10.7, P < 0.001). The assessment of the cumulative probability of no recurrence between the control groups showed significant heterogeneity at 1 year (χ2 = 5.1, P = 0.02), but not at 2 years or 3 years. There was no significant survival effect of oral 5-FU at 1 year (mean difference, 3%; CI, − 7.2% to 13%), 2 years (mean difference, 1.2%; CI, − 9.6% to 12%) or 3 years (mean difference, 2.5%; CI, − 8.6% to 13.6%). There was no significant effect of 5-FU on the cumulative probability of no recurrence at 1 year (mean difference, 1.9%; CI, − 12.8% to 16.7%) or 3 years (mean difference, 17.4%; CI, − 2% to 37%).

Combination therapy including systemic chemotherapy with transarterial chemotherapy

Three RCTs evaluated combination therapy of post-operative systemic chemotherapy and transarterial chemotherapy after curative resection.79–81 The study by Khono et al. evaluated combination therapy of oral 5-FU and transarterial chemotherapy after curative resection vs. a control arm receiving oral 5-FU alone (Table 4). The core group included two studies.79, 80 The protocols of systemic chemotherapy differed: the study by Lai et al. tested intravenous epirubicin,79 whereas that by Ono et al. analysed a combination of intravenous epirubicin and oral 5-FU.80

In the core analysis,79, 80 only data at 1 and 3 years were analysed, as the study by Ono et al. did not provide any data at 2 years. At 1 and 3 years, there was no significant heterogeneity between the control groups in survival or the cumulative probability of no recurrence. Post-operative transarterial chemotherapy in addition to systemic chemotherapy had an effect on the survival at 1 year (mean difference, − 9.2%; CI, − 23.2% to 5%) and 3 years (mean difference, − 3.4%; CI, − 19.3% to 12.4%), and a deleterious effect on the cumulative probability of no recurrence at 1 year (mean difference, − 19.7%; CI, − 35.2% to − 4.3%; P = 0.01) and 3 years (mean difference, − 21.3%; CI, − 41.6% to − 1%; P = 0.04).

After inclusion of the study by Khono et al.,81 there was still no significant effect of this therapy combination on survival at 1 year (mean difference, − 4%; CI, − 16.7% to 8.7%), 2 years (mean difference, − 5.5%; CI, − 38.4% to 27.3%) or 3 years (mean difference, − 2.2%; CI, − 14.8% to 10.5%), or the cumulative probability of no recurrence at 1 year (mean difference, − 10.6%; CI, − 27.9% to 6.8%), 2 years (mean difference, − 3.6%; CI, − 26.7% to 19.5%) or 3 years (mean difference, − 11.6%; CI, − 33.2% to 9.9%).

Discussion

Three years after surgical resection, tumour recurrence occurs in approximately 50% of cases.13, 17, 18 The prevention of recurrence constitutes one of the most important challenges to improve the efficacy of surgery. The absence of a consensus concerning adjuvant therapies justified the present overview using a literature-based meta-analysis.19 Of the 21 references included,31, 63–81 only four therapeutic modalities were evaluated in two studies or more: pre-operative transarterial chemotherapy,63–71 post-operative transarterial chemotherapy,31, 72–77 systemic chemotherapy (5-FU effect)75, 78 and systemic chemotherapy in addition to transarterial chemotherapy.79–81 The present study indicated that only post-operative transarterial chemotherapy constitutes an attractive approach.

Meta-analysis often searches for a moderate treatment effect not detected by several trials. In general, only RCTs are included in meta-analyses.19, 36, 37 However, meta-analytical techniques including NRCTs can be used.20, 21 In the present study, we performed a sensitivity analysis with the inclusion of NRCTs because of the insufficient sample size and restricted number of RCTs. The lack of precision and reliability of NRCTs and the likelihood of an inappropriate selection of patients can lead to incorrect results and spurious associations. However, a recent study has cautioned the classical view that considers only RCTs as evidence-based medicine.82 This study observed that the conclusions derived from NRCTs had a 20-year truth survival (87%) similar to that of those derived from RCTs (85%). Thus, from a consideration of these data, it can be concluded that the inclusion of NRCTs may help in the evaluation of adjuvant therapy. Interestingly, we showed that the inclusion of NRCTs did not modify the conclusions derived from ‘pure’ meta-analyses which included only RCTs.

Our meta-analysis has methodological limits. For example, in the meta-analysis of post-operative transarterial chemotherapy, three RCTs had a quality score of ≤ 1073, 74, 81 and only two RCTs had a score higher than 10.31, 72 Important heterogeneity in survival between control groups constituted another weakness of our study. Such a strong heterogeneity between studies could be due to a heterogeneity in pathological factors associated with recurrence, such as tumour size, venous invasion, resection margin, daughter tumours and the detection of multiple tumours on resected specimens. Unfortunately, most of the studies lacked a precise description and evaluation. Thus, we were not able to assess the influence of these factors.

Another weakness of our study was that we considered meta-analysis as the gold standard of evidence-based medicine. However, the validity of meta-analysis has recently been called into question.82 This study observed that meta-analysis might not have long-term utility in evidence-based medicine.82 The 20-year survival of the conclusions derived from meta-analysis (55%) was significantly lower than that of those derived from NRCTs (87%) and RCTs (85%). Therefore, our method using meta-analysis is subject to discussion, and our conclusions may not stand the test of time.

Pre-operative transarterial chemotherapy did not have an effect on survival or recurrence.63–71 We noted a wide discrepancy in the objectives of the studies. In the study by Wu et al., the authors analysed whether pre-operative transarterial chemotherapy permitted curative resection in patients with large tumours,63 whereas other studies included patients with tumours of less than 10 cm in most cases. As curative resection cannot be achieved for large tumours, surgical resection may not be considered as curative in the control patients of the study by Wu et al. Nevertheless, our data strongly preclude the future evaluation of pre-operative transarterial chemotherapy in patients with curative resection, as the conclusions reached were similar even after the exclusion of the study by Wu et al. (data not shown).

Systemic chemotherapy alone,75, 78 or in combination with transarterial chemotherapy,79–81 did not have an effect on survival or the cumulative probability of no recurrence. Moreover, in the core analysis, systemic chemotherapy in combination with transarterial chemotherapy had a significant deleterious effect in terms of the cumulative probability of no recurrence.

In the present study, only post-operative transarterial chemotherapy31, 72–77 was associated with a significant effect on survival and tumour recurrence. The benefit of post-operative transarterial chemotherapy was observed in the ‘pure’ meta-analysis of RCTs,31, 72–74 but also after the inclusion of NRCTs.75–77 Moreover, the magnitude of the effect was not affected by the inclusion of NRCTs. In terms of the overall survival and disease-free survival, differences between post-operative transarterial chemotherapy and control groups were close to 15% and 30%, respectively. We were unable to identify the sub-group of patients who benefited most from post-operative transarterial chemotherapy. The insufficient description of tumour staging and underlying liver disease did not enable an assessment to be made of the influence of these variables on the effect of post-operative transarterial chemotherapy. In addition, most studies were performed in Asian patients and the data cannot be extrapolated to the non-Asian population.

In conclusion, the present meta-analysis reveals that, after curative resection, post-operative transarterial chemotherapy may improve survival and the cumulative probability of no recurrence. However, a firm conclusion concerning the benefit of post-operative transarterial chemotherapy, drawn from our results, is hampered by the limitations of the included studies. Future studies evaluating post-operative transarterial chemotherapy are strongly recommended.

Ancillary