Methods	This was a GOG multi centre randomised trial comparing doxorubicin-cisplatin (AP) chemotherapy with whole-abdominal irradiation (WAI) in women with stage III or IV endometrial carcinoma having a maximum of 2 cm of postoperative residual disease. A total of 190 were randomly assigned to receive AP and 198 women were assigned to receive WAI
Participants	Four hundred twenty-two women were entered between May 1992 and February 2000, of whom 388 were initially eligible. Women with FIGO stage III or IV endometrial carcinoma of any histology were eligible for this trial. Eligibility required total abdominal hysterectomy and bilateral salpingo-oophorectomy, surgical staging, tumour resection, no single site of residual tumour more than 2 cm and normal haematology and biochemistry. Nodal sampling was optional but women with para-aortic lymph nodes metastasis were required to have negative scalene node biopsies and chest computed tomography scans
Interventions	Chemotherapy consisted of doxorubicin 60 mg/m2 plus cisplatin 50 mg/m2 every 3 weeks for eight cycles. The maximum allowable cumulative dose of doxorubicin was 420 mg/m2. Therefore only cisplatin was to be infused during the eighth cycle. Hydration was maintained by administering normal saline at 500 ml/h for 2 hours before and after the cisplatin dose. Doxorubicin doses were reduced based on pretreatment blood counts, with dose levels reduced from 60 to 15 mg/m2 in 15-mg/m2 increments. Doses were re instituted with recovery of myelosuppression. Treatment interruption caused by myelosuppression exceeding 6 weeks required discontinuation of protocol therapy Whole abdominal radiation therapy (WAI) was delivered with an open-field anterior-posterior/posterior-anterior technique with up to 30 Gray in 20 daily fractions. Kidney blocks were used posteriorly but there was no liver shield. After WAI, women received a boost to the true pelvis or to an extended field encompassing pelvic lymph nodes (PLNs) and para-aortic lymph nodes (PALNs). A boost to both areas was administered to patients with positive PLN and no PALN sampling or patients with neither PLN nor PALN sampling. Pelvic (± para-aortic) boosts were accomplished using a four-field box technique with custom blocking to minimise the treated small-bowel volume. The boost dose was 15 Gray in 8 fractions. All fields were treated once daily, 5 days per week
Outcomes	The primary end point for comparison of the treatment regimens was PFS; OS was a secondary end point
Results (OS & PFS)	At the time of reporting the final analysis, 98 patients (51%) on the chemotherapy arm were alive compared to 76 women (38%) on the WAI arm. Women alive or deceased, without evidence of tumour progression included 97 (50%) and 93 (46%) in the AP and WAI arms, respectively, including 18 and 22 women, respectively, who died before documentation of tumour progression. The majority of deaths were attributed to cancer progression. The progression hazard ratio relative to the WAI arm, adjusted for stage, was 0.71 (95% CI, 0.55 to 0.91; P = .007). This adjusted relative hazard ratio was associated with a predicted increase in PFS at 60 months of 12% (50% for AP versus 38% for WAI). There was also a statistically significant difference in overall survival between the two arms. Subgroup analysis, presented at the IGCS in 2010 showed that this was independent of cervical stromal involvement of lymph node metastasis The hazard ratio of death relative to the WAI arm, adjusted for stage, was 0.68 (95% CI, 0.52 to 0.89; P = .004). This adjusted relative hazard estimate was associated with a predicted increase in survival at 60 months of 13% for patients on the AP arm versus WAI patients (55% versus 42%, respectively). Unadjusted Kaplan-Meier estimates of five-year PFS and OS were 42% and 53%, respectively, in the AP arm compared with 38% and 42%, respectively, in the WAI arm. The unadjusted hazard ratio recalculated from the survival curves of GOG 122 reveal that the overall survival hazard ratio is 0.74 (0.57, 0.96) and the progression-free survival hazard ration is 0.81 (0.63 to 1.05) Treatment hazard ratio with 95% CI by prognostic group is reproduced in Figure 11.
Results (site of recurrence)	The initial site of recurrence was limited to the pelvis in 27 women (13%), within the abdomen in 33 women (16%), and extra-abdominal or liver metastases in 45 patients (22%). Four women allocated whole abdominal radiotherapy and in two allocated chemotherapy arm, sites of initial recurrence were unknown. Of the 194 women given chemotherapy, 97 (50%) had documented tumour recurrence. The initial site of recurrence was limited to the pelvis in 34 patients (18%); 27 patients (14%) experienced disease recurrence within the abdomen, and, in 34 patients (18%), the first recurrence included extra-abdominal or liver metastases
Toxicity	Comparisons of grade 3 to 4 haematologic toxicities between AP and WAI were: WBC (62% versus 4%, respectively), absolute neutrophil count (< 85% versus 1%, respectively), platelets (21% versus 3%, respectively), and maximum haematologic toxicity (defined as percentage of patients who developed at least one grade 3 or 4 haematologic toxicity of any type; 88% versus 14%, respectively). The second most commonly reported acute toxicity was grade 3 to 4 GI toxicity, which was reported in 20% versus 13% of patients in the AP and WAI arms, respectively, and hepatic toxicity was reported in 1% versus 3% of patients in AP and WAI arms, respectively. Grade 3 to 4 cardiac and neurologic toxicities were seen in 15% and 7% of patients, respectively, receiving AP compared with 0% and less than 1% of patients, respectively, receiving WAI. Other infrequent grade 3 to 4 toxicities included hip osteonecrosis, vaginal vault necrosis, and dehydration (one patient each) in the WAI arm and ventral hernia (one patient) in the AP arm Quality of life was reported separately using the Fatigue Scale (FS), Assessment of Peripheral Neuropathy (APN), Functional Alterations due to Changes in Elimination (FACE), and Functional Assessment of Cancer Therapy-General (FACT-G), QoL measured at: pre-treatment, end of treatment (EOT), and 3 and 6 months post-treatment. Of 396 eligible patients, 317 provided a baseline QoL assessment. The AP arm produced a statistically significant survival benefit along with greater toxicities, including peripheral neuropathy persisting up to 6 months. WAI patients reported worse FS (P < 0.001) and FACE (P < 0.001) scores at EOT and poorer FACE scores 3 months post-treatment (P = 0.004) compared to AP patients. APN scores were significantly worse among AP patients at EOT, and 3 and 6 months post-treatment (P < 0.001 for all). There is no indication that FACT-G scores differed between the two arms at any assessment point
Notes	Treatment probably contributed to eight deaths on the AP arm (two patients had sepsis, two patients had congestive heart failure, and one patient each had sepsis plus left ventricular/aortic thrombus, hypoglycemic shock with myelosuppression, stroke secondary to congestive heart failure, and renal failure with severe thrombocytopenia) and five deaths on the WAI arm (one patient each had veno-occlusive liver disease, disease progression with hepatomegaly, aspiration and liver necrosis, renal and liver failure secondary to sepsis with severe ascites, and sepsis and liver failure)
Risk of bias
Bias	Authors’ judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	The GOG Statistical and Data Center randomly assigned therapy to each patient with equal probability of assignment to each treatment regimen. A balanced block randomisation was used to balance assigned treatment regimens within each institution
Allocation concealment (selection bias)	Low risk	The sequence of treatment assignments was concealed from institutions and patients until telephone registration with verification of eligibility
Blinding (performance bias and detection bias) All outcomes	High risk	The trial was supervised by the GOG Data Monitoring Committee without knowledge of treatment results. However, treatment allocation could not be concealed for the trial participants
Incomplete outcome data (attrition bias) All outcomes	Low risk	Reasons for the exclusion of 34 patients (15 on the WAI arm and 19 on the AP arm) included wrong stage (n = 3), double primary (n = 8), wrong cell type (n = 4), prior malignancy (n = 1), residual disease more than 2 cm (n = 1), incomplete lymph node sampling or laparoscopic surgery (n = 8), registration error (n = 1), and inadequate documentation of pathology (n = 8). The eight patients (four on each arm) deemed ineligible because of unilateral lymph node sampling or use of laparoscopic surgery are included in the analyses. Since this study was initiated, acceptance and use of laparoscopic surgery has widened, and these patients are otherwise eligible. Their inclusion does not change the study results
Selective reporting (reporting bias)	High risk	It was suggested at the ASCO annual meeting by Thomas 2003 that the data were originally presented with a focus on survival time when mortality reached 50% exaggerating the real differences between the two groups The GOG 122 data were reported to show a survival advantage to adjuvant chemotherapy over wide field whole abdominal external beam radiotherapy. Both at the initial data presentation and final publication two interpretations of the data were given. The data as randomised gave five-year outcomes: PFS raw RT 38% versus chemo 42% (P not given), stage-adjusted 38% versus 50% (SS); raw OS RT 42% versus chemo 53% (P not given), stage-adjusted (despite it being randomised trial) 42% versus 52% (SS). Since the rate of recurrence between the two arms differed by only 4% (54% versus 50%) it is unclear whether the differences are due to action on cancer or confounding effects
Other bias	High risk	The trial took 10 years to complete from 1992. As a consequence, the inclusion criteria were very broad. The study population included women with an exceptionally good prognosis after hysterectomy, others has an exceptionally bad prognosis. The adjuvant therapies are not optimal by current standards. A dose of 30 g in 20 fraction has only a 10% chance of eradicating upper abdominal disease. The survival curves converge and the benefit for chemotherapy is early and applies only to distant metastases. There was no stratification for risk, particularly site and volume of residuum and we do not know how well balanced the groups were with respect to specific risk factors. These criticisms were eloquently delivered by Thomas 2003 in a plenary session discussion available on the ASCO media player The authors of the definitive manuscript (GOG 122) say that the treatment arms were balanced in terms of patient characteristics but there are slight imbalances with respect to mixed cell type and FIGO stage. If anything, there were slightly more poor prognosis pathologies and more advanced stage cancer in the group allocated chemotherapy
Main problems with the trial as perceived by this analysis	High risk	This is predominantly a trial of advanced disease. Few participants had stage 1 or 2 cancer and the data may not be applicable to early disease The applicability of the “control” arm remains uncertain as this trial tests whole abdominal radiotherapy. The radiotherapy regimen used in GOG-122 was tested in only two prospective phase 2 studies, one of which reported after GOG-122 had completed randomisation. The first was GOG 9001 with ten women having whole abdominal radiotherapy with concomitant chemotherapy for resected stage pT3 to pT4 disease in 1990-1992. The second was GOG 94 and this is a description of 21 women with stage pT1 to pT4 endometrial cancer. In 2010, whole abdominal radiotherapy in remains outside the standard treatments suggested by major guideline groups (IOG (UK), National Cancer Institute, USA, National Comprehensive Cancer Network, Lukka 2006). These two factors may explain why the study did not change standard practice. Nevertheless this study fulfils the search criteria for the inclusion in this meta-analysis using the original unadjusted data