Predictive Models of Morbidity and Mortality in Gynaecological Cancer Surgery: A Systematic Review

Abstract

Background/Objectives: Gynaecological–oncological surgery involves substantial perioperative morbidity and mortality risks. Reliable preoperative risk stratification could inform consent, shared decision-making and perioperative optimisation. Several prediction models exist, but their performance and methodological quality in this population remain uncertain. We aimed to provide a comprehensive synthesis of existing evidence to guide the selection and development of prediction tools in this population. Methods: We searched MEDLINE and EMBASE using a prospective protocol (PROSPERO CRD420251179495) and included studies developing or validating multivariable prediction models or risk scores for graded postoperative morbidity/mortality after gynaecological–oncological surgery. Risk of bias and applicability were assessed using the Prediction model Risk Of Bias Assessment Tool (PROBAST) at the study (publication) level. Findings were synthesised narratively. Results: Twenty-eight studies (75,999 patient records, including overlapping cohorts) were included (23 retrospective, five prospective; sample sizes 100–21,359). Models were multifactorial (n = 17), including generic and gynaecological–oncological-specific tools, frailty-based (n = 8) models or biomarker and index-based (n = 3) models. A numerical c-statistic was reported in 20 studies (binary c-statistics 0.50–0.92 across heterogeneous outcome definitions), calibration was reported in nine studies, and internal validation was reported in thirteen. No frailty-based model reported calibration or validation. One model specific to ovarian cytoreductive surgery was independently externally validated, retaining discrimination (c-index 0.715) with reduced calibration. Generic tools performed poorly, (c-statistics ranged from 0.529-0.644). Twenty studies were at high risk of bias, predominantly within the analysis domain. Conclusions: Existing models show, at best, moderate discrimination, limited calibration and infrequent external validation within the gynaecological–oncological surgical population. No model can currently be recommended for routine clinical practice. Further independent external validation of specialty-specific models in local settings, with full calibration reporting, is needed to support preoperative counselling.