Chang 2012a.

Study characteristics
Methods	Retrospective review of medical records. The decision to perform simple or radical procedures was determined by the surgeon.
Participants	Consecutive women with FIGO stage IIIC and IV primary epithelial ovarian, fallopian tube or peritoneal cancer who underwent primary cytoreductive surgery at Ajou University Hospital, Republic of Korea (enrolment 1 January 2000 to 31 December 2011). Age: median 54 years; range 30–78 years FIGO stage IIIC: 189 (93.1%); IV: 14 (6.9%) Tumour cell type: serous 167 (82.3%), mucinous: 4 (2.0%), endometrioid: 5 (2.5%), clear cell: 9 (4.4%), mixed: 18 (8.9%) ASA score 1–2: 114 (56.2%); 3–4: 80 (39.4%); 9 not available Tumour grade 1: 26 (12.8%), 2: 72 (35.5%), 3: 100 (49.3%), unknown: 5 Ascites > 100 mL: 92 (54.7%) Peritoneal carcinomatosis 149 (73.4%) Residual disease: no gross visible: 63 (31.0%); 0–1 cm: 67 (37.9%); > 1 cm: 63 (31.0%) Median BMI: 23.3 (range 11.7–35.2) Carcinomatosis: 149 (73.4%) Baseline details not presented according to type of surgery.
Interventions	Intervention: radical cytoreductive procedures included radical oophorectomy with or without rectosigmoid colectomy, total omentectomy, multiple bowel resections, diaphragm peritonectomy or resection, liver resection, splenectomy, distal pancreatectomy, and gastric resection. Comparison: simple surgery included total abdominal hysterectomy, bilateral salpingo‐oophorectomy, peritoneal biopsies or excisions, infracolic omentectomy, pelvic lymphadenectomy, para‐aortic lymphadenectomy, and segmental resection of small bowel. After surgery, all participants received adjuvant platinum‐based chemotherapy plus paclitaxel for 6–9 cycles.
Outcomes	Overall survival, progression‐free survival and adverse events Ultra‐radical versus standard surgery Median operative time (minutes): 307 versus 235; P < 0.01 Median estimated blood loss (mL): 800 versus 500; P = 0.03 Intra‐ or postoperative blood transfusion: 25% versus 17.6%; P = 0.01 Median stay in intensive care unit (days): 1.5 versus 0.8; P < 0.01 Postoperative mortality within 30 days: 1 versus 0 Any postoperative morbidity: 38% versus 11.8%; P < 0.01 Postoperative morbidity defined as infected lymphocyst, thromboembolism, intestinal obstruction, anastomotic leakage, ureteral injury, sepsis, intra‐abdominal abscess, pneumothorax or postoperative death within 30 days.
Notes	Follow‐up: median 43 months; range 1–124 months Retrospective non‐randomised study. The decision to perform simple or radical procedures was determined by the surgeon. Confounding by indication could not be excluded. Participant and disease characteristics not reported per type of surgery. Blinding not reported. Adjusted HRs were derived from a prognostic model. No details on how modelling was performed, but this seems to have been done based on significance testing (and not on including putative confounders in the analysis, irrespective of statistical significance).
Risk of bias
Bias	Authors' judgement	Support for judgement
1. Bias due to confounding (a–d)	High risk	Domain had a critical risk of bias: (quote) "The decision to perform simple or radical procedures was determined by the surgeon's discretion". Confounding by indication could not be excluded. Also, no known prognostic factors that had potential for confounding of the effect on intervention. Information was collected retrospectively.
2. Selection bias (a)	Low risk	Intervention and follow‐up start were simultaneous as a rule for cytoreductive surgery. No evidence of selection into the study due to variables measured after the intervention since participants were included retrospectively.
3. Classification of interventions (a–b)	Low risk	Well‐defined surgical interventions based on type of surgical procedure (simple versus radical).
4. Deviation from intended interventions (a–c)	Unclear risk	No evidence of any deviations from interventions or usual practice – which may either be an error of omission or that deviations did not happen.
5. Bias due to missing data (a–b)	High risk	Domain had a moderate‐to‐high risk of bias: all selected participants seem to have been included in the analyses. No differential follow‐up or missing data reported; no participant selection due to missing data reported. There was no reason to believe there was a serious bias due to missing data as the study was sound for a non‐randomised study with regard to this domain but could not be considered comparable to a well‐performed randomised trial. Therefore, it was sensible to judge the missing data domain at moderate‐to‐high risk of bias.
6. Measurement of outcomes (a–b)	High risk	Domain had a critical risk of bias: adjusted HRs were derived from a prognostic model. No details on how modelling was performed, but this seems to have been done based on significance testing (and not on including putative confounders in the analysis, irrespective of statistical significance). Also, adjustment were made for residual disease and this was likely to distort the estimate of survival as this adjustment was made after surgery and was a key prognostic factor.
7. Reporting bias (a–c)	Unclear risk	Difficult to judge. No protocol available. All outcomes mentioned in the methods section seemed to have been reported in the results section.