Results of the search

For details of the search strategies, see Appendix 1 and Appendix 2.

Our search identified 2227 unique references, excluding duplicates (Figure 2). At least two review authors (JM, SC) independently screened each abstract in this update of the review; 2083 articles that obviously did not meet the inclusion criteria were excluded at this stage. We retrieved 144 references in full and translated these into English, where appropriate. We found 23 references, reporting on five randomised controlled trials (RCTs), that met our inclusion criteria (Chekman 2015; Fagotti 2016; Kehoe 2015; Onda 2016; Vergote 2010); 12 references reporting on five ongoing trials (Kumar 2009, Mahner 2017  NCT04257786; NCT04515602; SUNNY). We excluded the remaining 108 references (see Excluded studies).

Figure 2

---

---

Study flow diagram of the search (up to October 2020).

Kumar 2009 had reported interim analyses in abstract form, but the outcomes were inadequately reported and the risk of bias profile was unclear, so we briefly discussed this trial in the Agreements and disagreements with other studies or reviews in the Discussion and included it with the list of ongoing studies (see Ongoing studies) rather than give it any weight in the main body of the review. Despite contacting the author, unfortunately, no further data have been provided to date for inclusion in the review.

One full‐text study (Jiang 2018) is awaiting classification (see Characteristics of studies awaiting classification). Jiang 2018 described the study as a retrospective, cross‐sectional study. However, the two groups (NACT versus PDS) were described as 'randomised' and ethical approval and informed consent were sought from study participants. There were significant differences in surgical outcomes between the two groups, but no significant differences in survival outcomes. Despite contacting the author, unfortunately, no further data have been provided to date for inclusion in the review.

Included studies

See Characteristics of included studies.

Chekman 2015 was a randomised controlled trial (RCT), conducted in Algeria between 2008 and 2014. The study enrolled 90 women with FIGO stage IIIc ovarian carcinoma who were randomised to either primary debulking surgery (PDS) followed by chemotherapy or neoadjuvant chemotherapy (NACT) followed by interval debulking surgery (IDS). The same surgeon operated on all women in both intervention arms. It would appear that all women had surgery as well as chemotherapy. Nine women were excluded (reasons not stated) and only data for those who had their disease resected to < 1 cm (including no macroscopic residual disease) were reported, i.e. there did not appear to be an intention‐to‐treat analysis. The diagnosis of stage IIIC ovarian carcinoma was confirmed by laparoscopic exploration in all but three cases. The number of cycles of chemotherapy in the NACT arm was six cycles (Carboplatin AU5/7.5 mg/mL/minute + Paclitaxel 175 mg/m²/3 hours every three weeks) on average with 44% having six cycles (range 3 to 7 cycles). Women in the PDS arm had six cycles of chemotherapy on average (78%) (range: 4 to 9) and followed the same chemotherapy protocol as in the NACT arm. The mean duration of follow‐up was 254.2 months (range: 69 to 480 months). The trial reported on < 1 cm residual tumour nodules (optimal debulk) or macroscopic resection, overall survival (OS), recurrence‐free survival (RFS), morbidity and discussed the role of lumboaortic lymphadenectomy. The study was in abstract form only, but Professor Chekman kindly provided us with more information on request. Unfortunately, survival outcomes could not be analysed, as data for time‐to‐event outcomes were not provided in an appropriate format for inclusion.

Kehoe 2015 (CHORUS) was a multicentre, non‐inferiority phase III RCT, conducted in 87 institutions in the UK and New Zealand. Inclusion criteria were women with clinical or radiological evidence of a pelvic mass with extra‐pelvic disease compatible with stage III or IV ovarian, fallopian tube or primary peritoneal cancer who were fit for surgery and chemotherapy. All women had clinical assessment including serum tumour markers and radiological imaging and 552 women were randomised to undergo treatment; two women were subsequently excluded due to being randomised in error. In the PDS arm, 276 women were assigned to undergo PDS followed by six cycles of platinum‐based chemotherapy within six weeks of surgery. In the PDS arm, women with residual tumour deposits > 1 cm were eligible to undergo an additional cytoreductive surgery after three cycles of chemotherapy. In the NACT arm, 274 women were assigned to undergo NACT for three cycles with platinum‐based chemotherapy and then have IDS and to recommence chemotherapy within six weeks of surgery. Women in the NACT arm had histological or cytological confirmation of diagnosis before commencing chemotherapy. The primary outcome measure was OS; secondary outcomes were progression‐free survival and quality of life (QoL). QLQC‐30 and QLQ‐Ov28 QoL questionnaires were used. The published QoL data provided only the global score at baseline (pretreatment), six months and 12 months post‐treatment.

In the NACT arm, 253 (92%) of 274 women started treatment as allocated and 217/274 (79%) had IDS. Nineteen of the 274 (6.9%) women in the NACT arm had no treatment; 36 women had no surgery following chemotherapy; 17 women had no postoperative chemotherapy (one of whom had primary surgery). In the PDS arm, 251 (91%) of 276 women started treatment as allocated; 212 (77%) had adjuvant chemotherapy. Ten of the 276 (3.6%) women had no treatment; 11 women had chemotherapy first with no surgery afterwards; 39 women had no postoperative chemotherapy (one of whom had preoperative chemotherapy); one woman had an unknown postoperative treatment status. See Characteristics of included studies for further details.

Vergote 2010 (EORTC 55971/NCIC OV13) was a large, international, multicentre, non‐inferiority phase III RCT. In total, 718 women were enrolled between 1998 and 2006; however, 48 were excluded after randomisation owing to authorisation irregularities at the Argentinian centre. Thus, 670 women with stage IIIc/IV epithelial ovarian cancer (EOC), primary peritoneal cancer or fallopian tube cancer were evaluated. For inclusion, an extra‐pelvic tumour needed to be 2 cm or more and treatment needed to begin within three weeks of the initial biopsy. The experimental group (334 women) were allocated to receive three cycles of platinum‐based NACT, followed by IDS and then at least three more cycles of chemotherapy (CT). The control group (336 women) received 'standard' treatment (i.e. PDS plus at least six cycles of platinum‐based CT ± IDS). The primary outcome was OS. Secondary outcomes were progression‐free survival (PFS), surgical morbidity and mortality, QoL and adverse effects. The investigators performed subgroup analyses on OS with respect to age, FIGO stage and extent of residual tumour. Subgroups of age were: age under 50 years, age 50 to 70 years and age over 70 years; subgroups of extent of residual tumour were: no residual tumour, residual tumour of 1 mm to 10 mm, and residual tumour greater than 10 mm. QoL data from the Vergote 2010 trial were subsequently reported by Greimel 2013 (see nested references in Vergote 2010).

Of the 334 women assigned to NACT, 326 (98%) started chemotherapy and 295 (88%) underwent IDS. Of the 336 women assigned to the PDS group, 315 (94.3%) had PDS and 88.4% started chemotherapy. See Characteristics of included studies for further details.

Onda 2016 (JCOG0602) was a multicentre, non‐inferiority, phase III RCT conducted in Japan. The authors enrolled 301 women between 2006 and 2011. For inclusion, women had stage III/IV ovarian, tubal and peritoneal cancers diagnosed by clinical findings, radiological imaging and cytology. CA125 had to be > 200 U/mL and CEA < 2 ng/mL to exclude malignancies of other anatomical sites. Women assigned to the control group (149) underwent PDS followed by eight cycles of platinum‐based chemotherapy. An additional debulking operation was performed after PDS, if PDS left > 1 cm of residual tumour. An additional debulking operation was mandatory if the uterus, adnexa or omentum had not been removed at PDS, unless disease progression occurred. Women assigned to the experimental group (152) received four cycles of platinum‐based NACT, then underwent IDS followed by a further four cycles of chemotherapy. The primary outcome of the study was OS, with survival data published in a peer reviewed journal in 2020, having previously been presented in conference proceedings. Secondary outcomes were invasiveness of surgery in terms of adverse events; these data have been published. No QoL assessment was performed.

Fagotti 2016 (SCORPION) was a single institution, superiority, phase III RCT. In total, 280 women with advanced ovarian cancer were enrolled into the study but, in order to be eligible for randomisation to the study arms, women had to undergo a staging laparoscopy. This was to obtain histology and confirm diagnosis, as well as assess the tumour load. Tumour load was assessed using a predictive index (PI). Only women with a PI score >/= to 8 and </ = 12, corresponding to a high tumour load, were eligible for randomisation. If it was deemed not possible to perform a staging laparoscopy due to large masses occupying the abdominal cavity infiltrating the abdominal wall or the presence of mesenteric retraction, women were withdrawn from the study. After recruitment reached 110 women in order to achieve statistical power for the analysis of the first co‐primary end point of major perioperative morbidity, further women were recruited to attain statistical power on PFS (more details are given in Risk of bias in included studies). Two hundred and twenty‐five women underwent staging laparoscopy in total, but only 171 went on to be randomised. In the control group, 84 women were assigned to PDS followed by six cycles of platinum‐based chemotherapy started within four weeks of surgery. Once women in the control arm had undergone PDS they were not allowed to have an additional cytoreductive procedure. In the experimental group, 87 women were assigned to three or four cycles of platinum‐based NACT and to undergo surgery within four weeks of the last cycle, if disease progression was excluded on imaging. The final cycles of chemotherapy in the experimental arm were resumed within four weeks of IDS. The mean and median time to the start of chemotherapy was 42.7 (SD = 18.3) and 41 days (range: 18‐169) in the PDS arm, respectively. In the NACT arm, the mean time to chemotherapy was 26.4 days (SD = 11.5) and the median was 26 days (range: 3‐79). The mean and median time to start adjuvant treatment after IDS in the NACT arm was 39 (SD = 10.8) and 37 (range: 14‐71) days, respectively. Co‐primary outcomes were PFS survival and postoperative complications. Secondary outcomes were OS and QoL. Further data were kindly provided by Professor Fagotti. Some outcomes were reported based on the initially published cohort of 110 patients, whereas others were reported for the final 171 participants in the randomised cohort.

Excluded studies

See Characteristics of excluded studies.

One hundred and eight references were excluded for the following reasons:

• Non‐RCTs (77);

• Eleven RCTs without a surgical arm comparison (Bertelsen 1990; Chan 2017; Deval 2003; Dutta 2005; Liu 2017; Lotze 1987; Mackay 2011; Mahner 2006; Polcher 2009; Rutten 2012; Trope 1997);

• Three RCTs of IDS following PDS (Redman 1994; Van der Burg 1995; Varma 1990);

• One RCT of non‐platinum‐based NACT versus surgery (Evdokimova 1982);

• One RCT of chemotherapy plus iliac artery embolisation versus surgery (Liu 2004);

• Fourteen reviews or systematic reviews (Baekelandt 2003; Bristow 2001; Dai‐yuan 2013; Fujiwara 2013; Kumar 2015; Lyngstadaas 2005; Mahner 2014; Makar 2016; Qin 2018; Sato 2014; Schorge 2014; Xiao 2018; Yang 2017; Zeng 2016);

• Two pooled analyses of studies included in the review (Vergote 2018; Vergote 2019);

• One RCT comparing early IDS after 3 cycles of NACT with late IDS after 6 cycles of NACT (Kumari 2020).

Liu 2004, an RCT comparing NACT plus iliac artery embolisation versus PDS, was originally an 'included study' in the 2006 version of this review.  In a previous update of the review, we revised our assessment of this study and excluded it, as the study findings might have been attributable to NACT versus PDS, iliac artery embolisation, or the combination, because NACT versus PDS was not the only variable in the study and iliac artery embolisation was not delivered in both arms.

Allocation

The Chekman 2015 study selection bias was judged to be at high risk, especially when compared to other studies with centralised randomisation, although allocation concealment was unclear due to lack of information. Ninety women with FIGO stage IIIc ovarian carcinoma were enrolled and underwent surgery, but only 82 women were randomised: 41 to PDS/chemotherapy and 41 to NACT/IDS. The randomisation was performed in the operating room by random draw by someone other than the surgeon, once verification of inclusion criteria and resectability under laparoscopy or laparotomy had been confirmed.

The Fagotti 2016 study was deemed to be at low risk of selection bias, albeit from a highly selected population. A centrally‐performed, computer‐generated list for block randomisation (1:1 ratio) was used. Women were randomly (maximum allowable percentage deviation = 10%) allocated to PDS + systemic adjuvant chemotherapy (arm A, control) or to NACT + IDS (arm B, experimental). Women were only eligible for randomisation into the study once they had undergone a staging laparoscopy to assess disease burden. The staging laparoscopy was used as a triage tool to assess eligibility for the study. If a staging laparoscopy was unfeasible, women were removed from the study. If the staging laparoscopy was successful, a predictive index (PI) value was calculated based upon seven parameters: presence or absence of omental cake, extensive carcinomatosis of the peritoneal or diaphragmatic surfaces, mesenteric retraction, infiltration of the stomach, spleen or bowel and or superficial liver metastases. If the PI score was ≥ 8 or ≤ 12, this was considered to be a high tumour load, related to lower chances of optimal cytoreduction and worse prognosis. The PI scoring system was based upon earlier work by the same group (Fagotti 2006; Fagotti 2013; Vizzielli 2014).

The initial phase of the Fagotti 2016 study identified 280 women, of whom 14.3% (40) were excluded: seven due to refusal to participate; 15 due to PS score > 2; and 18 due to age > 75 years. A further 15 women (6.25%) had an unsuccessful attempt at a staging laparoscopy, leaving 225 women who underwent a successful staging laparoscopy. Of those 225 women, a further 115 (51.1%) were excluded following staging laparoscopy: 69 due to a PI score < 8; 31 due to mesenteric retraction or PI score > 12; and 15 had non‐EOC histology. This left 110 women, with 55 allocated to each arm of the study. These complexities in trial design introduce potential sources of bias and may limit the applicability to the general advanced ovarian cancer population.

The risk of selection bias in the Kehoe 2015 study was deemed to be low risk as the randomisation was performed centrally using a minimisation method based on randomising centre, largest radiological tumour size, clinical FIGO stage, and prespecified chemotherapy regimen.

The Onda 2016 study was deemed to be at low risk of selection bias. The Japan Clinical Oncology Group (JCOG) data centre randomly assigned treatment to each woman via a minimisation method based on institution, stage (III versus IV), performance status (0 to 1 versus 2 to 3) and age (< 60 versus > 60).

In Vergote 2010, randomisation and allocation were performed centrally and the study appeared to be at low risk of allocation bias, although details of the process of randomisation method and concealment were lacking in published data.

Blinding

The five included studies were open‐label studies and outcome assessment were not blinded. This is probably not an issue for primary outcomes (i.e. survival); however, it may lead to detection bias with regard to other outcomes or subgroups (e.g. extent of debulking achieved). The importance of blinding of outcome assessment in ovarian cancer studies had been raised in a Gynecologic Cancer InterGroup (GCIG) consensus statement (Thigpen 2011). Data for such outcomes are thus to be interpreted with caution and all studies were deemed to be at high risk of bias.

Incomplete outcome data

Chekman 2015 was at unclear risk of attrition bias due to lack of reported details.

Fagotti 2016 was judged to be at high risk of attrition bias. After the recruitment of 110 women was achieved for the analysis of the first co‐primary end point of major perioperative morbidity, further women were recruited to attain statistical power on PFS. The final trial cohort consisted of 171 women, with 84 randomised to PDS and 87 randomised to NACT. Information was not available for two patients who were lost during treatment, one for each arm. The initial published data reported QoL outcomes and short‐term surgical outcomes. There were substantial missing data for QoL outcomes, but relative results (hazard ratios (HRs)) for survival (OS and PFS) were adequately reported and analysed. Of the women included in the analysis, 82/84 women in the PDS arm required upper abdominal surgical procedures compared to 28/74 women who underwent IDS (42.3%). Median duration of entire treatment from randomisation to completion of medical treatment was also longer in the PDS arm (38 weeks versus 28 weeks). This was due to an almost two‐week difference in time to start post‐surgery chemotherapy (median time post‐PDS 40 days; median time post‐IDS 27 days; P = 0.0001).

Kehoe 2015 and Onda 2016 were deemed to be at low risk of attrition bias, as all trial participants were accounted for and the results were analysed on an intention‐to‐treat basis.

In the Vergote 2010 study, data from 48 women from Argentina were excluded owing to "potential authorisation irregularities"; however, the investigators stated that their results were similar when these excluded data were included. The exclusions appeared erroneously as pre‐randomisation exclusions on the published study‐flow diagram. The study was, therefore, at unclear risk of attrition bias.

Selective reporting

Chekman 2015 was at unclear risk of reporting bias due to lack of detail.

Fagotti 2016 was at unclear risk of reporting bias due to the differences in numbers reported for different outcomes, as described above, and lack of quality of life data to date.

In Kehoe 2015, the risk of reporting bias was unclear. All prespecified outcome measures have been reported in some capacity, but QoL data were provided only in the form of a global score at baseline, six months and 12 months post‐treatment.

The potential for reporting bias in the Onda 2016 study is now deemed to be low risk; surgical morbidities were reported in the initial publication and survival outcomes have now been published .

There was an unclear risk of selective reporting bias for QoL data in the Vergote 2010 study. Vergote 2010 (including Greimel and colleagues) subsequently published the QoL data from the Vergote 2010 study (see additional reference under Vergote 2010). They reported that compliance for all women was too restrictive and changes to the protocol‐defined analysis plan were made. The dataset for QoL data was then restricted to institutions with the best compliance. The authors stated that the sample size of the Vergote 2010 was overpowered to detect clinically meaningful differences in QoL between the two study arms and they therefore decreased the sample size for QoL data to 400 participants. They further restricted QoL data collection to institutions that had 50% compliance at baseline and at least 35% on further follow‐up over all enrolled women. Twenty‐seven institutions out of 59 contributed 404 women (60.3% of the total 670 trial participants). The participants in institutions that were included in the QoL data had statistically significant differences compared to those participants not included: they had larger tumours (P < 0.01) and optimal debulking rates were 20% higher (P = 0.001). Those participants in institutions selected for inclusion in QoL data analysis had a greater median OS (nine months longer; P = 0.001) and a greater median progression‐free survival (PFS) (2.4 months longer; P < 0.001) than the participants in the institutions that were not included in the QoL data collection. In addition, as well as selecting institutions with the highest compliance with QoL data, the overall compliance from those institutions was still relatively poor over time. Compliance rates were 83.4% at baseline, 58.7% at chemotherapy cycle 3, 74% at chemotherapy cycle 6, 59.4% at six‐month follow‐up and 45.7% at 12‐month follow‐up.

The authors concluded that there were no differences in the QoL functioning or symptoms scales, other than for pain and dyspnoea.. At baseline, the PDS group had higher pain scores (P = 0.046; PDS mean 36.7; NACT mean 29.9) and lower dyspnoea scores (P = 0.049; PDS mean 22.9; NACT mean 27.9). As the difference between the groups was less than 10 points, the authors concluded that this did not represent a "clinically relevant difference".

There was, therefore, unclear risk of reporting bias for the QoL data, given the differences in disease that those participants selected for measurement of this outcome had in comparison with participants in the institutions not selected.

Other potential sources of bias

Due to lack of detail, Chekman 2015 was judged to be at unclear risk of other potential sources of bias.

The complexity of the inclusion criteria in Fagotti 2016, as described above, mean that we were unclear about other potential sources of bias and the study design limits the applicability of the study to a wider, less selected, cohort of women with ovarian cancer.

Supplementary data in Kehoe 2015 table 7 show that hysterectomy/bilateral salpingo‐oophorectomy (BSO) and omentectomy were performed in varying proportions in the different arms. It is unclear what effect this might have on outcomes and this could be a potential source of bias.

In the Onda 2016 study, 14 women (one in PDS and 13 in NACT) underwent some type of additional surgery (off‐protocol treatment). These off‐protocol operations were not included as PDS or IDS in the analysis. There appeared to be more off‐protocol surgery in the NACT group. No intention‐to‐treat analysis was performed. These issues could be another potential source of bias.

Overall survival (OS) (Analyses 1.1 to 1.4)

Meta‐analysis of four studies (Fagotti 2016; Kehoe 2015; Onda 2016; Vergote 2010), assessing 1692 participants, demonstrated little or no difference in OS between neoadjuvant chemotherapy (NACT) and primary debulking surgery (PDS) for initial treatment in advanced ovarian cancer (hazard ratio (HR) = 0.96, 95% CI 0.86 to 1.08; high‐certainty evidence); Analysis 1.1; Figure 3 and Figure 4).

Figure 3

---

---

Forest plot of comparison: 1 NACT vs PDS, outcome: 1.1 Overall survival.

Figure 4

---

---

In the PDS group 757 people out of 1000 had died over 4 years compared to 743 (95% CI 704 to 783) out of 1000 for the NACT group.  Green = alive at 4 years with PDS/chemo; yellow = additional people alive at 4 years with NACT/IDS; red = people who had died by 4 years with either NACT/PDS or PDS/chemo.

The results were also robust (i.e. no meaningful difference between subgroups) in terms of OS when three trials (Fagotti 2016; Kehoe 2015; Vergote 2010) were subgrouped by age (< 50, 50 to 70 and 70+ years) (Analysis 1.2), and extent of residual disease in two studies (Kehoe 2015; Vergote 2010) (up to 0.5 mm, 0.5‐1 cm, > 1 cm) (Analysis 1.3). The results were also robust when three trials (Kehoe 2015; Onda 2016; Vergote 2010) were subgrouped by stage (III and IV) (Analysis 1.4). Survival data by stage were not yet available for one study (Fagotti 2016).

We were not able to extract time‐to‐event data for OS from the Chekman 2015 study. However, in total, 24 women died during the study period; 15 women (62.5%) in the PDS arm compared to nine women (37.5%) in the NACT arm.

Progression‐free survival (PFS) (Analysis 1.5)

Meta‐analysis of four studies (Fagotti 2016; Kehoe 2015; Onda 2016; Vergote 2010), assessing 1692 participants, found there is probably little or no difference in risk of disease progression between NACT and PDS for initial treatment in advanced ovarian cancer (HR = 0.98, 95% CI 0.88 to 1.08; I² = 0%; moderate‐certainty evidence) (Analysis 1.5; Figure 5 and Figure 6).

Figure 5

---

---

Forest plot of comparison: 1 NACT vs PDS, outcome: 1.4 Progression‐free survival.

Figure 6

---

---

In the PDS group 858 people out of 1000 had ovarian cancer that had recurred by 2 years compared to 852 (95% CI 821 to 879) out of 1000 for the NACT group.  Green = not had recurrent disease by 2 years with PDS/chemo; yellow = additional people without recurrent disease by 2 years with NACT/IDS; red = peopel with recurrent disease by 2 years with either NACT/PDS or PDS/chemo.

From the Chekman 2015 study, we were not able to extract time‐to event data for PFS. However, there were 36 recurrences (44%); 20 participants with progressive disease (55.5%) in the control arm (PDS) and 16 (44.5%) in the experimental (NACT) arm.

Of the 12 women in Chekman 2015 who were still alive with confirmed recurrence, five (41.6%) were in the PDS arm and seven (58.3%) were in the NACT arm. Peritoneal recurrence was reported to be most common. Further details about recurrence are given in the table Characteristics of included studies.

Extent of residual disease

In Kehoe 2015, 79/219 women (36%) and 39/255 women (15%) had no macroscopic residual disease in the NACT and PDS arms, respectively; 68/219 (31%) and 57/255 (22%) had 'optimal debulking' (defined as 0.1 cm to 1 cm residual disease) in the NACT and PDS arms, respectively; and 54/219 (25%) and 137/255 (54%) had suboptimal debulking (defined as > 1 cm) in the NACT and PDS arms, respectively. Overall, 147/219 (67%) women and 96/255 (38%) women in the NACT and PDS arms, respectively, had < 1 cm residual disease. Data on degree of resection were missing for 18 women in the NACT group and 22 in the PDS group.

In the NACT arm, 55/274 (20%) women did not have debulking surgery. In the PDS arm, 251 women had PDS and another four had surgery after NACT, so 21 of the 276 allocated to PDS women did not have debulking surgery (7.6%).

In Vergote 2010, of those who had debulking surgery, 151/295 women (51.2%) and 61/315 women (19.4%) had no macroscopic residual disease in the NACT and PDS arms, respectively; 87/295 (29.5%) and 70/315 (22.2%) had 1 mm to 10 mm residual disease in the NACT and PDS arms, respectively; and 52/295 (17.6%) and 167/315 (53%) had suboptimal debulking (> 1 cm residual disease) in the NACT and PDS arms, respectively. Data on debulking status were stated as missing for five (1.7%) women in the NACT group and 17 (5.4%) women in the PDS group. See Characteristics of included studies table for further details. Therefore, of those who had NACT and interval debulking surgery (IDS), 238 women (80.7%) had debulking to < 1 cm residual disease compared to 131 women (41.6%) who had PDS.

Of those assigned to NACT, 326/334 (98%) started chemotherapy and 295/334 (88%) went on to have IDS. In the PDS group, 315 (94.3%) had PDS and 88.4% started chemotherapy.

In Fagotti 2016, 57/74 women (77%) and 40/84 women (47.6%) had no macroscopic residual disease in the NACT and PDS arms, respectively; 16/74 (21.6%) and 38/84 (45.2%) had residual disease 0.1 cm to 1 cm in the NACT and PDS arms, respectively. Therefore, debulking to < 1 cm was achieved for 73/74 (98.6%) and 78/84 (92.8%) in the NACT and PDS arms, respectively; 1/74 (1.4%) and 6/84 (7.2%) had suboptimal debulking (residual disease > 1 cm) in the NACT and PDS arms, respectively (13 participants in the NACT arm did not undergo IDS). This is despite extensive pre‐assessment and intraoperative exclusion (laparoscopic assessment), which differed significantly from the Kehoe 2015 and Vergote 2010 studies.

In Onda 2016, 83/150 women (55%) and 45/147 women (31%) had no macroscopic residual disease in the NACT and PDS arms, respectively; 24/150 (16%) and 47/147 (32%) had residual disease 0.1 cm to 1 cm in the NACT and PDS arms, respectively; and 23/150 (15%) and 55/147 (37%) had residual disease > 1 cm in the NACT and PDS arms, respectively. Overall, 107/150 women (71%) and 92/147 women (63%) had optimal debulking (defined as debulking to no residual disease > 1 cm) in the NACT and PDS arms, respectively. Higher optimal debulking rates than Kehoe 2015 and Vergote 2010 may be due to lower initial disease burden, since the entry criteria included all stage III disease, not just bulky stage IIIc, and 9 (6%) in the PDS and 10 (6.6%) in the NACT groups had no measurable disease (presumably by RECIST criteria (Eisenhauer 2009) but not stated) at outset.

Severe adverse effects (SAEs) (Analyses 1.6)

The trial of Fagotti 2016 reported major perioperative morbidity, initially when the trial had randomised 110 participants. Some level of granularity in adverse events was not given in the follow‐up publication, so some analyses were based on the initial cohort (n = 110), whereas analyses included in the follow‐up publication included all 171 women.

Some studies reported all SAEs during the study period (Kehoe 2015; Onda 2016; Vergote 2010), whereas some reported surgically‐related SAEs only (Chekman 2015; Fagotti 2016). The following grade 3/4 (CTCAE 2017) SAEs were reported (Analysis 1.6):

Other SAEs (Analyses 1.6.6 to 1.6.14 and 1.6.17)

Overall postoperative G3+ SAEs from two studies (Fagotti 2016; Onda 2016) found that the number of patients who had a G3+ SAE in the postoperative period was probably lower in the NACT group (RR 0.22, 95% CI 0.13 to 0.38; participants = 435; studies = 2; I² = 0%; moderate‐certainty evidence) (see Analysis 1.6.18 and Figure 7).

Figure 7

---

---

In the PDS group 29 people out of 100 had G3+ post op serious adverse events (SAE) compared to 6 (95% CI 4 to 20) out of 100 for the NACT group.  Green = no post op G3+ SAE with PDS/chemo; yellow = additional people who were better with NACT/IDS; red = people with G3+ SAEs with either NACT/PDS or PDS/chemo.

The proportion of remaining SAEs that were assessed was low. There was probably little or no difference between arms for risk of urinary/vaginal fistula, nausea, vomiting, diarrhoea, neutropenia, neurotoxicity, thrombocytopenia, anaemia, febrile neutropenia and renal toxicity (see analyses 1.6.6 to 1.6.14; 1.6.17; all low‐certainty evidence). IDS may be associated with less risk of stoma formation, bowel resection, and postoperative grade 3+ events than PDS.

In the Chekman 2015 study, there were a total of 17 complications: 12/41 women in the PDS arm; 5/41 women in the NACT‐IDS arm (intraoperative incidents). We were careful not to over interpret this result from a trial of low numbers in each arm, with issues regarding imprecision and unclear risk of bias.

The authors reported that eight re‐operations (9.8%) were performed, mainly for abdominal and vascular complications; six (7.3%) in the PDS arm and two (2.4%) in the NACT‐IDS arm.

Stoma formation (Analysis 1.6.15)

Women were less likely to require formation of a stoma (colostomy or ileostomy) in the NACT arm versus the PDS arm, although data were only presented in two of the studies (Fagotti 2016; Kehoe 2015) (RR 0.43, 95% CI 0.26 to 0.72; participants = 581; studies = 2; I² = 0%; moderate‐certainty evidence) (Analysis 1.6.15 and Figure 8).

Figure 8

---

---

In the control group 20 people out of 100 had stoma formation following initial surgery compared to 6 (95% CI 2 to 15) out of 100 for the active treatment group. Green = no stoma with PDS/chemo; yellow = additional people who didn't require a stoma with NACT/IDS; red = people who required a stoma with either NACT/PDS or PDS/chemo.

Perioperative/postoperative mortality (Analysis 1.7)

Meta‐analysis of five studies (Chekman 2015; Fagotti 2016; Kehoe 2015; Onda 2016; Vergote 2010), assessing the 1625 participants who had surgery, found women in the NACT arm had less risk of perioperative/postoperative mortality than in the PDS arm (Analysis 1.7; RR 0.16, 95% CI 0.06 to 0.46; participants = 1623; studies = 5; I² = 0%; high‐certainty evidence). Three out of 787 (0.4%) women died within a month of surgery in the NACT arm compared to 26 out of 836 (3.1%) deaths in the PDS arm and, overall, 30/836 (3.6%) due to postoperative complications. There were an additional four deaths in Fagotti 2016 due to postoperative complications in women who survived more than 30 days after surgery, although these deaths were directly related to surgery. Overall, postoperative mortality was therefore 3.6% (30/836) in the PDS group versus 0.4% (3/787) in the NACT group (Analysis 1.7 and Figure 9).

Figure 9

---

---

In the PDS group 36 people out of 1000 died in the post‐operative period compared to 6 (95% CI 2 to 17) out of 1000 for the NACT group. Green = alive at the end of the post‐operative period with  PDS/chemo; yellow = additional people who were alive with NACT/IDS; red = people who died in the post‐operative period with either NACT/PDS or PDS/chemo.

In Chekman 2015, no deaths were recorded postoperatively (0 to 30 days), but one death was recorded after a second course of neoadjuvant chemotherapy (prior to surgery).

Chemotherapy‐related toxicity (Analysis 1.8)

Chemotherapy‐specific‐related toxicity was not specifically reported in Vergote 2010 as all SAEs were reported together. However, median time to re‐start chemotherapy after surgery was 18 days (range 5 to 55) and 19 days (range 0 to 84) in the NACT and PDS groups, respectively. In Fagotti 2016, the median time to start chemotherapy following surgery was lower in the NACT group (NACT = 27 days (range 16 to 37 days) versus PDS = 40 days (range 17 to 120 days); P < 0.0001)) for the initial 110 patient cohort.

Two trials (Kehoe 2015; Onda 2016), assessing 768 participants, found that there may be little or no difference in chemotherapy‐related SAEs between arms, although we have low certainty in these results (Analysis 1.8; OR 0.88, 95% CI 0.57 to 1.36, I² = 54%; low‐certainty evidence).

Quality of life (QoL) (Analyses 1.9 to 1.10)

Three studies (Kehoe 2015; Fagotti 2016; Vergote 2010), assessing 524 participants, reported on QoL at six months using the EORTC QLQ‐C30 questionnaire. In two studies, individual symptoms were reported (Fagotti 2016; Vergote 2010). We did not interpret pooled results for individual symptoms due to heterogeneity in results and the summary effects are merely displayed in forest plots to demonstrate the heterogeneity. Results were either inconsistent or there did not appear to be any differences in QoL measures in individual domains between arms. The global health domain was the only domain to demonstrate a numerically significant difference between arms, but the magnitude of the difference was so small, it would be very unlikely to be clinically meaningful. Vergote 2010 and Kehoe 2015 also reported QoL at 12 months with similar results, but due to the high dropout rate, especially by 12 months, these results were of very low‐certainty and should be interpreted with caution (Analysis 1.9; Analysis 1.10). Previously, the Kehoe 2015 results were reported separately, due to uncertainty in which QoL data were reported in the original paper. However, following clarification, we have been able to amalgamate these data. Further data from Kehoe 2015 for individual QoL parameters are awaited and it may be possible to combine further QoL data in future updates.

Duration of operation

Mean operating times in Chekman 2015 were 233 minutes (range 69 minutes to 360 minutes) and 273 minutes (range 144 minutes to 480 minutes) in the NACT and PDS groups, respectively. Mean operating times in the Fagotti 2016 study for IDS after NACT and PDS were 253.2 minutes (SD = 101.4) and 460.6 minutes (SD = 102.6), respectively.

In Vergote 2010, the median operating times were 180 minutes (range 30 minutes to 560 minutes) and 165 minutes (range 10 minutes to 720 minutes) in the IDS and PDS arms, respectively. Kehoe 2015 reported that the median operation time was 120 minutes in both groups (interquartiles ranges were 80 to 161 and 90 to 155 in the PDS and NACT arms, respectively; the overall range was 12 to 450 mins). Onda 2016 found that median operating time, when accounting for the main procedure only (not counting an additional debulking procedure in the PDS group) was 302 minutes in the NACT group and 240 minutes in the PDS group (P < 0.001). However, if the subsequent operative procedures were accounted for in both groups, median operating times were 270 minutes and 347 minutes in the NACT and PDS groups, respectively (P < 0.001). Due to disparities in the data collected, we are not able to combine these in a meta‐analysis.

Length of stay following surgery

Fagotti 2016 reported mean length of hospital stay; in the NACT group, the mean was 6.7 days (SD = 3.9 days) and 14.8 days (SD = 11.3) in the PDS group (P < 0.001). In Kehoe 2015, length of stay was provided as follows: "fewer women were discharged from hospital within 14 days after surgery in the primary‐surgery group compared with primary chemotherapy (198/249, 80% versus 197/211, 93%, P < 0·0001)". Data were not amenable to meta‐analysis. These data were not available for Chekman 2015, Onda 2016 or Vergote 2010.