# P-POSSUM Falls Short: Predicting Morbidity in Ovarian Cancer (OC) Cytoreductive Surgery

## TL;DR

The P-POSSUM scale is not accurate at predicting complications after ovarian cancer surgery, but adding frailty and BMI improves predictions slightly.

## Contribution

The study evaluates and improves upon the P-POSSUM scale by incorporating frailty and BMI for better morbidity prediction in ovarian cancer surgery.

## Key findings

- P-POSSUM overestimates morbidity and has poor discrimination (AUC 0.54) in predicting complications after ovarian cancer surgery.
- Adding Edmonton Frail Scale (EFS) and BMI to P-POSSUM improves predictive accuracy (AUC 0.66), though not statistically significant.
- Most postoperative complications were classified as Clavien–Dindo II, with a 40.3% overall morbidity rate.

## Abstract

We evaluate the performance of the P-POSSUM scale in predicting perioperative morbidity in women undergoing cytoreductive surgery (CRS) for ovarian cancer (OC). Data from 161 consecutive patients were retrospectively analysed, including demographics, frailty (Edmonton Frail Scale—EFS), preoperative albumin, and surgical outcomes. Postoperative morbidity occurred in 40.3% of patients, with most complications graded as Clavien–Dindo II. P-POSSUM significantly overestimated morbidity (predicted 59.5% vs. observed 40.3%) and demonstrated poor discrimination (AUC 0.54). Mortality prediction was also suboptimal, though the small number of deaths limited interpretation. Step-wise regression with bootstrapping identified EFS and BMI as additional significant predictors of morbidity. Incorporating these into a combined model (P-POSSUM + EFS + BMI) improved predictive accuracy (AUC 0.66), though these improvements did not reach statistical significance. Overall, P-POSSUM alone is inadequate for morbidity prediction in OC CRS. A revised model integrating frailty and BMI shows promise; however, it requires prospective validation.

Objective: The P-POSSUM scale is widely used in predicting perioperative morbidity and mortality. Evidence on the performance of P-POSSUM in predicting outcomes after cytoreductive surgery (CRS) for ovarian cancer (OC) is limited. In this study, we assess how well P-POSSUM predicts morbidity in OC CRS and explore whether incorporating additional clinical variables can enhance its predictive accuracy. We retrospectively collected data on consecutive patients undergoing OC CRS within a tertiary gynaecologic oncology network. The collected information included demographic characteristics, P-POSSUM morbidity and mortality scores, Edmonton Frail Scale (EFS) scores, preoperative serum albumin levels, and observed 30-day postoperative morbidity and mortality, classified using the Clavien–Dindo (CD) scale. The predictive performance of P-POSSUM was evaluated using receiver operating characteristic (ROC) curves to calculate sensitivity and specificity. A stepwise regression analysis was then applied to identify additional variables that could improve model performance, incorporating preoperative covariates. The final model incorporated parameters chosen through bootstrap investigation of the model variability (stepAIC). Predicted versus observed morbidity was calibrated and performance compared between P-POSSUM and the final model. Results: Of 161 sequential OC patients, 95 (59%) underwent primary, 45 (28%) interval, and 21 (13%) delayed CRS. The mean age was 66.4 (95%CI: 60–75) and duration of surgery was 223 mins (95%CI: 142–279). Sixty-five (40.3%) patients had ≥1 postoperative complication. Two deaths were reported. Among the observed complications, 4 patients (6.1%) experienced CD4, 10 patients (15.3%) CD3, 38 patients (58.5%) CD2, and 11 patients (16.9%) CD1 events. The mean P-POSSUM-predicted morbidity and mortality were 59.5% (95%CI: 56.7–62.3%) and 5.86% (95%CI: 5.02–6.70%), respectively. The area under the curve (AUC) for P-POSSUM in predicting morbidity and mortality was 0.539 (p = 0.401) and 0.569 (p = 0.137), respectively. Given the small number of deaths, no robust conclusions regarding mortality are possible. EFS and BMI emerged as significant predictors of observed morbidity using a stepwise-model selection process. The AIC of this final model was 211.44. Our final model of PPOSSUM + EFS + BMI had AUC = 0.6551 (Delong’s Z = 1.8845, p-value = 0.05949). Conclusions: The P-POSSUM scale shows poor performance for predicting morbidity in OC CRS. New validated and accurate model(s) are necessary for predicting surgical morbidity. Our proposed model incorporates additional variables to improve P-POSSUM’s performance. This requires further development and validation.