Results of the search

See Characteristics of included studies and Characteristics of excluded studies.

Ten potentially relevant studies were identified from the search strategy described above. The full text for all of these was obtained. These were then assessed for inclusion/exclusion by two reviewers (KJ, NJ) using the stated criteria.

Included studies

Nine trials consisting of 2119 women, representing 100% of eligible women from known randomised trials, contributed data. Patient accrual varied from 20 to 546.

Three United States trials were inter‐group studies with more than 40 participating institutions in each (Alberts 1996; GOG 172; Markman 2001). The Gadducci 2000 study involved the Gruppo Oncologico Nord‐Ovest in Italy and it was unclear how many centres participated. The Zylberberg 1986 study involved three participating centres. In the Kirmani 1994 trial an unclear number of community hospitals participated in the San Diego area of California. In the remaining three trials one centre participated (Polyzos 1999; Yen 2001; Yen 2009).

Zylberberg 1986 is the earliest RCT on IP chemotherapy for the initial management of epithelial ovarian cancer. They randomised 20 women with stage III disease to receive either IV chemotherapy including doxorubicin, fluorouracil, bleomycin, cisplatin, vinorelbine and ifosfamide, or a combination of IV doxorubicin, fluorouracil, bleomycin, cisplatin, vinorelbine, ifosfamide and IP bleomycin, cisplatin, fluorouracil and doxorubicin. Both groups received maintenance intramuscular chemotherapy with 12 monthly courses of ifosfamide, fluorouracil, and methotrexate. Ten women were randomised to each arm and the outcome of each patient was reported. They described a statistically significant increase in the number of women alive and free of disease in the IP group (P < 0.05) but no further statistics were provided. In the IV group, of the 10 randomised, there were five deaths after 11, 18, 24, 38 and 62 months and five free of relapse after 31, 40, 42, 64 and 72 months. In the IP group there was one death after 29 months, two with residual disease resected at second look laparotomy, free of disease after 23 and 54 months, and the other seven were free of disease after 23, 23, 27, 28, 34, 58 and 59 months. More mature data are unfortunately not available and our attempts to contact the lead author were unsuccessful. We were able to construct a Kaplan‐Meier curve for time to death and these data have been included in the survival meta‐analysis.

Kirmani 1994 randomised 87 women with stage IIC‐IV disease over four years to receive IV cisplatin 100 mg/m² and cyclophosphamide 600 mg/m² or IP cisplatin 200 mg/m² and etoposide 350 mg/m². Of these, 68 were evaluable for toxicity, 62 for survival (33 IV and 29 IP), and 46 for response. Ten were excluded from participation after review of the pathology. Seven women refused their assigned treatment arm (4 IP and 3 IV). For various reasons another eight women were not evaluable for response. The two groups were well balanced with respect to age, stage and residual disease, but more in the IV group had an ECOG performance status of 0 (14 versus 5) and fewer had a status of 2 (3 versus 6). This was the only trial comparing direct IV to IP (without additional IV) chemotherapy. Kaplan‐Meier curves show non‐significantly better overall survival in the IV arm. Disease‐free survival included only those with no clinical evidence of disease at the end of treatment, and also showed no significant difference between the two groups. However, only 25 IV and 21 IP women were evaluable for this assessment. There was also no detectable difference in either haematological or non‐haematological toxicity between the high‐dose IP regime and the standard‐dose IV regime.

The North West Oncology Group trial (Gadducci 2000) was run over 7½ years, and randomised 113 women with 100 evaluable (54 IV and 46 IP) participants. They considered stage II‐IV disease with < 2 cm residual disease after debulking surgery. The chemotherapy arms included IV or IP cisplatin 50 mg/m². Both groups also received IV epidoxorubicin 60 mg/m² + IV cyclophosphamide 600 mg/m². Twenty‐two women did not complete the assigned treatment (2 IV and 20 IP). Significant reasons for treatment change in the IP arm included patient refusal (6), bowel perforation (3) and abdominal pain (2). Patient characteristics were well balanced in the two groups except histotype. Seven women with clear cell histology were randomly allocated to the IV arm. Median disease free survival was 25 and 42 months for the IV and IP arms (P = 0.13) and median overall survival was 51 and 67 months (P = 0.14). Similar survival curves were obtained after exclusion of the women with clear cell histology. The only significant toxicity difference was more grade 3‐4 myelotoxicity in the IP group (52% versus 34%).

Polyzos 1999 randomised 90 women (46 IV and 44 IP) with stage III disease randomised to receive either IV or IP carboplatin 350 mg/m² all of whom were eligible for analysis. Both groups also received IV cyclophosphamide 600 mg/m². There were no statistically significant differences in the distribution of characteristics of the women between the two chemotherapy groups. There was no maximum tumour diameter permitted for entry into this trial. A residual tumour volume ≥ 2 cm was found in 43% of women. There were no significant differences in time to progression or overall survival in the two groups, even after adjusting for residual tumour volume but we were unable to obtain the survival curves to include data in the meta‐analysis. Significantly more women in the IV group had grade 3 or 4 leukopenia (P < 0.01) and non‐significantly more grade 3 or 4 thrombocytopenia (P < 0.09). However, 10% of women receiving IP chemotherapy were reported to have major catheter‐related morbidity, including three cases of chemotherapy infused directly into the large bowel with ensuing massive diarrhoea, and two cases where fluid was infused between layers of the abdominal wall. Weaknesses in this study include the poor statistical power with the low number randomised. Fewer women had tumour volumes < 2 cm (25 IV and 26 IP). The randomisation technique and allocation concealment was unclear, and analysis by intention‐to‐treat was not described.

Yen 2001 randomised 132 women with stage III disease to receive either IV cisplatin 50 mg/m² or IP cisplatin 100 mg/m² and of these 118 were eligible (55 IV and 63 IP). Both groups also received IV adriamycin or epirubicin 50 mg/m² and IV cyclophosphamide 500 mg/m². There were no significant differences between the two study groups with regard to important prognostic factors. All women initially had optimal debulking surgery with residual disease of ≤ 1 cm. The median survival rates were 43 months for the IP group and 48 months for the IV group (P = 0.469). Significantly more women in the IV group had grade 3 or 4 leukopenia but there was no significant increase in thrombocytopenia or anaemia. Catheter‐related complications in the IP group included pain (41.8%) and obstruction (25.5%).

Alberts 1996 coordinated the first large study incorporating investigators from the SWOG and GOG, randomising 654 women with residual disease of ≤ 2 cm to either IP or IV cisplatin (100 mg/m²); 546 women were finally eligible. Both groups also received IV cyclophosphamide 600 mg/m². The trial investigators included a subgroup analysis in January 1991, four and a half years after the study commenced accrual, after consensus emerged that women with a residual tumour size of ≤ 0.5 cm would be most likely to benefit from IP chemotherapy. The study was then extended for an additional year to achieve an adequate (powered) sample size for analysis of this subgroup. There were no significant differences between the two study groups with respect to important prognostic factors. The IP arm had significantly less myelotoxicity and tinnitus but there were no major differences between the two arms in terms of severe toxicity, treatment‐related deaths and removal of women from the study due to adverse events.

Markman 2001 randomised 532 women with 462 evaluable (227 IV and 235 IP) and compared an IV paclitaxel plus cisplatin regime with an experimental combination of two courses of high dose IV carboplatin followed by six courses of IV paclitaxel plus IP cisplatin. The maximum tumour diameter permitted for entry into this trial was 1 cm. Women in the IP arm had a median disease‐free survival of 28 months and a median overall survival of 63 months, both of which were superior to the median 22‐month progression‐free survival and median 52‐month overall survival associated with IV‐administered drugs (P = 0.02 and P = 0.05, respectively). Neutropenia, thrombocytopenia, and gastrointestinal and metabolic toxicities were significantly greater in the IP group such that 18% of these women actually received ≤ 2 cycles of intraperitoneal cisplatin. This was thought to be a consequence of the two initial high dose IV carboplatin treatments (AUC 9), designed to minimize the residual volume of disease prior to IP treatment.

GOG 172 was a similar trial for women with stage III, optimally de‐bulked ovarian cancer. In this trial 429 women were randomised and 14 were ineligible. This left 210 in the IV arm to receive paclitaxel 135 mg/m² plus cisplatin 75 mg/m² and 205 in the IP arm to receive IV paclitaxel 135 mg/m² plus IP cisplatin 100 mg/m² plus IP paclitaxel 60 mg/m² on day 8, with each regime delivered every 21 days for 6 cycles. The principal endpoint was progression‐free interval, with secondary endpoints being overall survival and toxicity and quality of life scores. Data provided by the study chair from the GOG meeting in January 2005 revealed the following: 429 women had entered, well balanced for prognostic factors and completion of treatment in each arm. There were significantly more grade 3 or 4 toxicities in the IP arm, including leukopenia, thrombocytopenia, gastrointestinal, renal, neurologic, fatigue, infection, metabolic and pain scores. This may be expected due to the use of a higher IV cisplatin dose (100 mg/m² v 75 mg/m²) and the addition of IP paclitaxel on day eight. Nevertheless, the median progression‐free interval and time to death for the IP group was prolonged, with the RR of recurrence 0.79 and a RR of death 0.75, suggesting a therapeutic advantage for IP therapy. These data have since been published (Armstrong 2006).

Yen 2009 randomised Stage III ovarian cancer patients to IP versus IV cisplatin or carboplatin after giving all IV paclitaxel. The objective of the trial and published manuscript was to construct a prognostic nomogram addressing the following risk factors: age, CA‐125, IP/IV delivery, stage, histology, and upper abdominal metastases. Out of 367 women recruited, 298 were analysed (152 in IV group and 146 in IP group). Chemotherapy consisted of a three hour infusion of paclitaxel (175mg/m²) on day one to all women. On day two, platinum (either 100 mg/m² cisplatin or 300 mg/m² carboplatin) was administered by either the IP or IV route. This was repeated every three weeks for six cycles provided serum creatinine was less than or equal to 2 mg/dl, WCC > 3000/mm³, and platelet count was > 80,000/mm³. IP therapy was discontinued and shifted to IV chemotherapy if catheter problems occurred. Published data consisted of a nomogram to predict survival, including an odds ratio (OR) for IV versus IP of 2.14 (95% CI 1.93 to 2.41) in favour of IP therapy. The investigators also showed that they could identify the subset of women who are least likely to benefit from IP chemotherapy, i.e. age > 80 years or with baseline CA 125 > 3000 U/ml or Stage IIIc disease or clear cell carcinoma or upper abdominal tumour metastases or colon resection, and may, therefore, be spared the potential complications. We obtained additional methodological information and unpublished data, including survival data and adverse effects (Grade 3/4), from the trial investigators.

Overall survival data were extracted from the reports and unpublished data of all but one of the trials (Polyzos 1999). Five trials provided HRs and covariates in Cox regression analysis (Alberts 1996; GOG 172; Markman 2001; Yen 2001; Yen 2009[unpublished data]). Most included age, tumour type and grade, performance status and residual disease as covariates prior to commencing chemotherapy. Six trials displayed Kaplan‐Meier survival curves (Alberts 1996; Gadducci 2000; GOG 172; Kirmani 1994; Markman 2001; Yen 2001) and we were able to construct a survival curve for the Zylberberg 1986 trial from individual patient data presented. One of these presented sufficient data to allow calculation of the HR using Parmar's methods (Gadducci 2000). The HR of death for Kirmani 1994 and Zylberberg 1986 was estimated from the Kaplan‐Meier survival curves. Only one of the trials (Alberts 1996) included subgroup analysis data, and so meta‐analysis on subgroup data could not be performed.

Disease‐free survival data and Kaplan‐Meier disease‐free survival curves were available from five trials (including unpublished data from Yen 2009). Two presented the calculated HRs (GOG 172; Markman 2001), two trials presented sufficient data to allow calculation of the HR using Parmar's methods (Gadducci 2000; Yen 2009) and the HR of recurrence for Kirmani 1994 was estimated from the Kaplan‐Meier diease‐free survival curves. The definitions of recurrence varied amongst these trials. For example, Kirmani 1994 used the Eastern Cooperative Oncology Group definition, Markman 2001 used date of entry to date of appearance of disease (clinical or radiological) or equated it to survival, Gadducci 2000 used date of entry to date of first progression and GOG 172 used date of randomisation to progression, death, or the date of last contact, whichever came first.

Eight of the nine included trials reported adverse effects with the exception of Zylberberg 1986. We obtained unpublished data from Yen 2009. Toxicity criteria included SWOG, Common toxicity criteria and WHO, all of which were comparable for the toxicities described. Gadducci 2000 used the WHO criteria but included 4 grades of alopecia and these data, therefore, could not be assimilated because the WHO criteria divide alopecia into only 3 grades.

QOL scores were reported in the GOG 172 trial using the Functional Assessment of Cancer Therapy scale with General, Neurotoxicity, Pain and Ovarian cancer sub‐scales (Wenzel 2004; Wenzel 2007).

Only GOG 172 assessed the effect of chemotherapy on QOL issues. Women receiving the higher dose IP therapy experienced more QOL disruption, pain and neurotoxicity when compared with those who received more conventional IV therapy. No other studies addressed this issue and, therefore, no meta‐analysis could be performed.

Median length of follow‐up was reported in eight of the nine trials: Alberts 1996 > 60 months; Gadducci 2000 60 months; GOG 172 > 60 months; Kirmani 1994 46 months; Markman 2001 > 60 months; Yen 2001 74 months; Yen 2009 62 months; and Zylberberg 1986 50 months. Polyzos 1999 did not provide data on length of follow‐up.

Excluded studies

Piccart 2003 compared four doses of maintenance IP chemotherapy compared to observation in 153 women who had had a complete response to conventional platinum‐based intravenous chemotherapy. After a median follow‐up of 8 years, 80 women (52%) had progressed with no difference in the pattern of relapse between the two groups. The respective hazard ratios for PFS and OS with 95% CI were 0.89 (0.59 to 1.33) and 0.82 (0.52 to 1.29). The authors concluded that this was suggestive of a treatment benefit. However, as maintenance therapy did not fit this review's eligibility criteria we excluded this trial.

Allocation

Three United States multicentre trials (Alberts 1996; GOG 172; Markman 2001) followed guidelines for the conduct of inter‐group trials with adequate randomisation and allocation concealment. Randomisation and allocation concealment was also adequate for the Gadducci 2000 study. Yen 2001 described the use of random number tables but did not confirm a method of allocation concealment. Yen 2009 randomised women using a computer generated table but did not describe allocation concealment. Both randomisation and concealment of allocation were unclear in Kirmani 1994; Polyzos 1999 and Zylberberg 1986.

Blinding

Personnel and patient blinding after allocation was not possible because of the nature of the trials. Therefore, both recipients of care and care providers were aware of their assigned intervention. Assessor blinding was performed in Alberts 1996; GOG 172 and Markman 2001.

Selective reporting

Six studies reported intention‐to‐treat analyses (Alberts 1996; Gadducci 2000; GOG 172; Markman 2001; Yen 2001; Yen 2009). Most trials reported expected outcomes except for Yen 2009 where the stated objective was to construct a nomogram for survival, but additional data was provided by the investigators for the purposes of this review. Zylberberg 1986 did not report adverse effects.

Other potential sources of bias

We sub‐grouped the Alberts 1996; Gadducci 2000; GOG 172; Markman 2001; Yen 2001 and Yen 2009 trials as high quality, and the Kirmani 1994; Polyzos 1999 and Zylberberg 1986 trials as low quality and data from all are included in the meta‐analysis. It was not possible to comprehensively document the numbers of, or reasons for, withdrawal from treatment protocol. The number who received all, part or none of the planned treatment is recorded for each treatment arm in Table 1.

Overall survival (time to death)

Eight trials (2026 women) were included in this meta‐analysis. There was a significant improvement in overall survival in the IP intervention group (HR 0.81, 95% CI 0.72 to 0.90; Analysis 1.1; Figure 2). This remained when only high quality trials were considered (six trials; HR 0.80, 95% CI 0.72 to 0.90). Data were homogeneous and there was no significant difference between subgroups. Some experts have argued that the excluded trial (Piccart 2003) should be included on the basis that the addition of IP chemotherapy after completion of initial surgery and IV chemotherapy may still be described as part of primary treatment for ovarian cancer. If these data are included, the HR for time to death is similar. Furthermore, if the analysis is restricted to trials that used the same chemotherapy regimes in each arm, the HR remains similar (three trials; 0.79, 95%CI 0.67 to 0.92; Analysis 1.2)

Figure 2

---

---

Forest plot of comparison: 1 IP component therapy versus IV therapy, outcome: 1.1 Time to death.

Progression‐free survival (time to recurrence)

Five trials (1311 women) were included in the meta‐analysis of time to recurrence. There was a significant improvement in progression‐free survival in the IP intervention group (HR 0.78, 95% CI 0.70 to 0.86; Analysis 1.3; Figure 3). Data were homogeneous. Polyzos 1999 provided limited survival data (dichotomous) that were unusable in the meta‐analysis (33/46 = 72% of controls versus 33/44 = 75% of IP arm); time to progression was 19 (8 to 62+) and 18 months (6 to 72+) with the median survival advantage of one month in favour of the intraperitoneal arm.

Figure 3

---

---

Forest plot of comparison: 1 IP component therapy versus IV therapy, outcome: 1.3 Time to recurrence.

Severe adverse effects

The following severe adverse effects (AEs grade 3/4) were more likely to occur in the IP group:

1. fever (five trials, 1797 women; RR 1.64, 95% CI 1.13 to 2.38; I² = 20%; Analysis 1.10; Figure 4);

2. fatigue (three trials, 1171 women; RR 2.32, 95% CI 1.06 to 5.07; I² = 71%; Analysis 1.11);

3. gastrointestinal AEs (five trials, 1339 women; RR 1.70, 95% CI 1.28 to 2.26; I² = 48%; Analysis 1.12; Figure 5). In the 'high quality trials' subgroup, I² = 0% (four trials, 1271 women; RR 1.90, 95% CI 1.57 to 2.30);

4. infection (three trials, 1171 women; RR 3.34, 95% CI 2.06 to 5.43; I² = 0%; Analysis 1.13; Figure 6);

5. metabolic AEs (two trials, 873 women; RR 4.45, 95% CI 2.72 to 7.26; I² = 0%; Analysis 1.14);

6. pain (three trials, 1235 women; RR 7.47; 95% CI 4.41 to 12.67; I² = 0% Analysis 1.16; Figure 7).

Figure 4

---

---

Forest plot of comparison: 1 IP component therapy versus IV therapy, outcome: 1.10 Adverse effects ‐ fever (G3‐4).

Figure 5

---

---

Forest plot of comparison: 1 IP component therapy versus IV therapy, outcome: 1.12 Adverse effects ‐ gastrointestinal (G3‐4).

Figure 6

---

---

Forest plot of comparison: 1 IP component therapy versus IV therapy, outcome: 1.13 Adverse effects ‐ infection (G3‐4).

Figure 7

---

---

Forest plot of comparison: 1 IP component therapy versus IV therapy, outcome: 1.16 Adverse effects ‐ pain (G3‐4).

Hearing loss was more common in the IV group (three studies, 1009 women, RR 0.67; 95% CI 0.46 to 0.99; Analysis 1.17). There were no significant differences between interventions for haematological AEs: anaemia (Analysis 1.4); thrombocytopenia (Analysis 1.5) and leukopenia (Analysis 1.6). However, substantial heterogeneity (I² > 60%) was noted in these meta‐analyses. Substantial heterogeneity also occurred in the meta‐analyses of renal, neurological and pulmonary AEs (no significant difference between interventions): Analysis 1.7, I² = 57% Analysis 1.15; I² = 69% and Analysis 1.8, I² = 79%, respectively. This could not be explained by subgroup differences.

Quality of life

Only the GOG 172 trial assessed QOL as an outcome measure (Wenzel 2007) and, therefore, no meta‐analysis was performed. Women who received higher dose IP therapy (paclitaxel 135 mg/m² IV, cisplatin 100 mg/m² IP, paclitaxel 60 mg/m² IP) experienced more QOL disruption when compared to those who received the more conventional dose of IV therapy (paclitaxel 135 mg/m² IV, cisplatin 75 mg/m² IV). Women in the IP arm reported worse QOL and pain prior to the fourth chemotherapy cycle (P < 0.001) and worse QOL 3 to 6 weeks post‐treatment (P = 0.009). The GOG 172 protocol exposed women in the IP arm to additional taxol and neurotoxicity was significantly worse in the IP arm 3 to 6 weeks after completing chemotherapy (P = 0.0004), and one year later (P = 0.0018). There were no significant QOL or PAIN score differences between arms one year post‐treatment.

Catheter‐related complications of IP drug administration were discussed in all of the trials but data retrieved from them were insufficient to produce a meta‐analysis. Catheter blockage was described in Kirmani 1994 (10%), Gadducci 2000 (8.7%), Yen 2001 (25.5%), and GOG 172 (in 7% of those who stopped IP therapy this was the primary reason). Catheter‐related infection was reported in Yen 2001 (12.8%), Markman 2001 (severe infection noted in 4.7%) and GOG 172 (in 19.5% of those who stopped IP therapy this was the primary reason). In GOG 172, of the 205 women in the IP arm 118 (58%) did not complete 6 cycles of IP therapy. Intraperitoneal therapy was stopped in 34 (28.8%) for reasons unrelated to the peritoneal access device, but catheter failures occurred in 49 (41.5%) women. Catheter complications were the primary cause for discontinuing IP therapy in 39% participants. Descending colon or rectosigmoid colon resection appeared to decrease the number of cycles successfully delivered (P = 0.022). Also, data from the trial suggested that IP administration of paclitaxel might in itself increase complications.