Results of the search

A MEDLINE search identified 598 studies. A similar search of EMBASE, the Cochrane Central Register of Controlled Trials (CENTRAL) and the Specialised Register revealed 941 studies, 263 studies and 253 studies, respectively. Reference lists were checked, and handsearching of journals and congress abstracts yielded five additional studies.

A total of 74 articles were selected from the first sift. After 18 duplicates and 39 other abstracts were removed, 17 possibly eligible studies were retrieved for more detailed analysis. We found eight RCTs (Chiara 1994; Dueñas‐González 2011; Kantardzic 2004; Lorvidhaya 2003; Tang 2012; Vishnevskaia 1999; Wang 2010; Wong 1999). Two RCTs met the inclusion criteria and were included in this review (Dueñas‐González 2011; Lorvidhaya 2003). One study (Dueñas‐González 2011) subsequently reported a subgroup analysis to determine prognostic factors for participants (Dueñas‐González 2012). Six trials were excluded for one or more of the following reasons: Neoadjuvant chemotherapy was given as well as adjuvant chemotherapy (Chiara 1994; Tang 2012; Vishnevskaia 1999); only radiation was used in either of the control arms (Kantardzic 2004; Vishnevskaia 1999; Wang 2010; Wong 1999) or in both arms (Chiara 1994).

The eight remaining non‐randomised studies were excluded for the following reasons: only radiation provided or phase II studies without a comparative arm (Choi 2007; Domingo 2009; Kim 2012; Mahasittiwat 2011; McCaffrey 2011; Vrdoljak 2006); prospective matched case comparison without randomisation (Choi 2010); or a preliminary report of one study (Vrdoljak 2005). See QUOROM (Quality of Reporting of Meta‐analyses) statement flow diagram (Figure 1).

Figure 1

---

---

Study flow diagram.

Included studies

We found three reports from two RCTs (Dueñas‐González 2011; Lorvidhaya 2003). Dueñas‐González 2011 described the results of their subgroup analysis in a subsequent report (Dueñas‐González 2012), details of which are provided in the Characteristics of included studies.

Excluded studies

Of 17 possibly eligible studies, 14 were excluded: six were RCTs, and eight were phase II prospective trials.

Six RCTS were excluded for one or more of the following reasons: neoadjuvant chemotherapy was also given aside from ACT (Chiara 1994; Tang 2012; Vishnevskaia 1999); only radiation was used in either the control arm (Kantardzic 2004; Vishnevskaia 1999; Wang 2010; Wong 1999) or both arms (Chiara 1994).

Of the eight remaining non‐randomised studies, all were excluded for the following reasons: providing radiation only, phase II studies without a comparative arm (Choi 2007; Domingo 2009; Kim 2012; Mahasittiwat 2011; McCaffrey 2011; Vrdoljak 2006), prospective matched case comparison without randomisation (Choi 2010) or a preliminary report of one study (Vrdoljak 2005).

See Characteristics of excluded studies.

Allocation

In Lorvidhaya 2003, randomisation was stratified according to each individual centre among the six centres joining in the trial. No information was provided regarding the method and the centre or site responsible for randomisation as well as allocation concealment. We assessed this as unclear risk.

In Dueñas‐González 2011, participants were randomly assigned to CCRT plus ACT or CCRT alone at week 0 according to the Pocock and Simon algorithm by disease stage (IIB versus III to IVA), tumour diameter (< versus ≥ 5 cm), investigational centre, radiation equipment (cobalt‐60 versus linear accelerator) and age (< versus ≥ 55 years). The random allocation sequence was generated by central telephone of the pharmaceutical company providing the research funding (Eli Lilly, Indianapolis, IN, USA). Allocation was concealed until interventions were assigned. This has been deemed to show low risk of bias.

Blinding

No blinding during treatment was reported in either trial (Dueñas‐González 2011; Lorvidhaya 2003). This is believed to show high risk of bias.

Incomplete outcome data

After randomisation, Lorvidhaya 2003 identified six participants in the CCRT group and seven in the CCRT plus ACT group (approximately 3% both) who had small cell carcinoma and were excluded from the analysis. After a median follow‐up of 89 months, 31 participants in both arms (12% or 5% in the CCRT group and 19% or 8% in the CCRT plus ACT group) were lost to follow‐up and received incomplete treatment. Reasons were not specified. Participants were censored as having failure of treatment.

Among the 515 participants who were randomly assigned in Dueñas‐González 2011, one who was allocated to CCRT was given both CCRT and adjuvant chemotherapy. A total of 53 participants (approximately 10%) discontinued treatment. After a median follow‐up period of 46.9 months, a total of 56 participants (approximately 11%) were lost to follow‐up by the end of the study. Numbers and reasons for loss to follow‐up were similar in the two treatment arms.

We determined both trials as having low risk of bias.

Selective reporting

All outcomes in Lorvidhaya 2003 were reported according to the protocol. Nevertheless, no information was provided on the timing of adverse events.

Alhough the primary endpoint in Dueñas‐González 2011 was changed from OS to PFS after completion of enrolment, all outcomes specified in the protocol were reported. Survival analyses were performed by an intention‐to‐treat approach (including the one participant who had a treatment cross‐over), and toxicities were analysed for safety according to the actual treatment received.

We determined that both trials had low risk of bias.

Other potential sources of bias

No imbalance in baseline characteristics between the two arms was reported in either of the two studies (Dueñas‐González 2011; Lorvidhaya 2003). However, an important difference in CCRT treatment between groups was found in one study (Dueñas‐González 2011). The control arm received only single‐agent chemotherapy (cisplatin), and the study arm was given two drugs (cisplatin and gemcitabine) during CCRT. We determined this to show high risk of bias, as it is likely to have had a significant effect on the outcomes.

Survival

Lorvidhaya 2003 found no benefit for ACT over CCRT alone. Using the Parmar method to estimate hazard ratio, we found that the HR for OS was 1.309 (95% confidence interval (CI) 0.795 to 2.157) and for DFS was 1.125 (95% CI 0.799 to 1.586) among participants who received adjuvant chemotherapy (ACT).

Dueñas‐González 2011 showed significant improvement in PFS and OS among those who received CCRT plus ACT compared with those given CCRT alone (HR 0.68, 95% CI 0.49 to 0.95 for PFS, P value 0.027; HR 0.68, 95% CI 0.49 to 0.95 for OS, P value 0.022). The benefit of ACT was demonstrated in all subgroups. Subsequent analyses (Dueñas‐González 2012) showed that benefit from ACT, PFS in particular, was greater in women with more advanced disease, larger tumour size and age younger than 55 years. The hazard ratio of PFS among participants with stage III to IVA disease who were given CCRT and ACT compared with those who received only CCRT was 0.59 (95% CI 0.37 to 0.97, P value 0.036).

Data on survival from both studies are summarised in Additional tables: Table 1.

No meta‐analysis was performed. Data on progression‐free survival for each individual trial are shown in Figure 2 and on overall survival in Figure 3.

Figure 2

---

---

Forest plot of comparison: 1 Survival, outcome: 1.1 Progression‐free survival.

Figure 3

---

---

Forest plot of comparison: 1 Survival, outcome: 1.2 Overall survival.

Failure after treatment

Lorvidhaya 2003 did not demonstrate that the addition of chemotherapy could decrease total failure rate or loco‐regional or distant failure after CCRT. Total failure rates were 29.0% among participants who received CCRT (arm 3), and 31.2% among those given CCRT plus ACT (arm 4). No significant differences in loco‐regional or distant recurrence were noted among participants who were treated with CCRT alone or CCRT plus ACT: 14.3% versus 17.6% (P value not given) for local recurrence, and 17.7% versus 19.5% (P value not given) in arm 3 and arm 4, respectively.

Rates of local failure in Dueñas‐González 2011 were not different (11.2% versus 16.4%, P value 0.097). However, distant failure rate was lower in the ACT group: 8.1% versus 16.4% (P value 0.005). Differences in chemotherapy during CCRT should be noted; it is not possible to determine whether differences were due to ACT given alone.

Data on recurrences from both studies are summarised in Additional tables: Table 2.

No meta‐analysis was performed. Data on treatment failure at all sites for each individual trial are shown in Analysis 2.1 (Figure 4), local failure in Analysis 2.2 (Figure 5) and distant failure in Analysis 2.3 (Figure 6).

Figure 4

---

---

Forest plot of comparison: 2 Failure after treatment, outcome: 2.1 Total failure after treatment.

Figure 5

---

---

Forest plot of comparison: 2 Failure after treatment, outcome: 2.2 Local failure.

Figure 6

---

---

Forest plot of comparison: 2 Failure after treatment, outcome: 2.3 Distant failure.

Response rate

Only Dueñas‐González 2011 reported response rates in the two treatment arms. These did not differ (95.8% in the CCRT plus ACT arm versus 93.4% in the CCRT arm, P value 0.249) (Additional tables: Table 3). No information on response rates was available for Lorvidhaya 2003.

Adverse events

Lorvidhaya 2003 reported that one participant in the CCRT arm died from persistent thrombocytopenia and leucopenia. Gastrointestinal symptoms were greater in the CCRT arm, but study authors stated that most women tolerated the treatment well (actual numbers not given). Haematological toxicities were not increased by ACT after CCRT. Anaemia of grade 2 was 24.2% among participants given CCRT plus ACT and 20.6% in those treated with only CCRT. Respective grade 3 and 4 leucopenia was 2.9% and 4.1%, and grade 3 and 4 thrombocytopenia was 1.5% and 2.3% (Additional tables: Table 4). No differences in late side effects were observed among studied arms (Additional tables: Table 5).

Dueñas‐González 2011 reported detailed toxicities in both treatment arms. It should be noted that chemotherapy during CCRT differed when added to ACT. More women who had CCRT plus ACT were hospitalised: 30 women versus 11 who had CCRT only (P value 0.003). Women who received CCRT plus ACT also had a greater number of grade 3 or 4 overall toxicities (87.0% vs 46.0%, P value < 0.001). Rates of grade 3 and 4 toxicities in women who received CCRT and ACT versus CCRT alone were 71.9% versus 23.9% for haematological and 34.6% versus 10.6% for gastrointestinal symptoms (nausea, vomiting, diarrhoea, abdominal pain/cramping, anorexia and proctitis), respectively. More women in the study arm had received one or more blood transfusions: 49.2% versus 27.5% (P value < 0.001). Overall, three deaths occurred in the CCRT plus ACT arm during or 30 days after treatment. Two deaths were attributed to the ACT. When study authors assessed toxicities from the preadjuvant and adjuvant treatments separately, they found that the CCRT plus ACT arm had greater toxicity than the CCRT alone arm during the preadjuvant phase. However, the incidence of new toxicity during ACT was generally low.

No meta‐analysis was performed. Data on anaemia and thrombocytopenia from each individual study are shown in Analysis 3.1 (Figure 7) and Figure 8, respectively.

Figure 7

---

---

Forest plot of comparison: 3 Adverse events, outcome: 3.1 Anaemia.

Figure 8

---

---

Forest plot of comparison: 3 Adverse events, outcome: 3.2 Thrombocytopenia.

Other outcomes

No information on QoL was reported in either study (Dueñas‐González 2011; Lorvidhaya 2003).