Article Text
PDF
Abstract
Introduction/Background P-POSSUM scale has been widely used in predicting peri-operative morbidity and mortality. There is paucity of evidence on P-POSSUM performance in OC cytoreductive surgery. We evaluate P-POSSUM performance in predicting morbidity for OC cytoreductive surgery. We also improve PPSOSSUM performance using a new model incorporating PPOSSUM and additional covariates.
Methodology Data were retrospectively collected on consecutive patients undergoing cytoreductive surgery for OC from a single tertiary gynaecological oncology cancer network. Data included demographics, P-POSSUM morbidity and mortality scores, Edmonton Frail Scale (EFS) score, pre-operative serum albumin level, observed post-operative 30-day morbidity classified using Clavien-Dindo(CD) scale and mortality. We performed inferential univariate statistics and data were calibrated for expected versus observed morbidity and mortality in deciles. Receiver Operating Characteristic curve (ROC) was used to validate the performance of P-POSSUM scales (sensitivity, specificity). We used step-wise regression-analysis, to build a new model to improve PPOSSUM morbidity performance by incorporating additional pre-operative co-variates.
Results We included 161 patients with mean age 66.4(95CI:60–75). Ninety-five(59%) underwent primary, N =45(28%) interval, and N=21(13%) delayed cytoreductive surgery, with mean duration-of-surgery of 223mins(95CI:142–279). Sixty-five(40.3%) patients experienced at least one post-operative complication, with two reported deaths. Of the complications, CD4=4(6.1%), CD3=10(15.3%), CD2=38(58.5%) and CD1=11(16.9%). The predicted (P-POSSUM) vs. observed mean-morbidity was 59.5%(95CI:30–88%) vs. 9.5%(CI:1.13–19.37%) and for mean mortality this was 5.86%(95CI:1.31–14.9%) vs. 2.35, CI:0.65–2.78%. P-POSSUM overpredicts morbidity at>35% probability-threshold, and underpredicts, at< 25% probability-threshold(p=0.001). The AUC for P-POSSUM morbidity and mortality were 0.539(p=0.401) and 0.569(p=0.137) respectively. The number of deaths was too small to draw any meaningful inference. Our new model of PPOSSUM+EFS+albumin had AUC=0.624(Delong’s Z=-1.3442, p=0.1789). The optimal sensitivity/specificity maximising the Youden’s-Index=0.585/0.656.
Conclusion The P-POSSUM shows poor performance in predicting morbidity. It overpredicts at high-probability and underpredicts at low-probability thresholds. We explore and suggest a new model which improves morbidity prediction. This needs to be prospectively validated in future studies.
Disclosures None declared.
