OC02.02: Development of new models to predict of pouch of Douglas (POD) obliteration in women with suspected endometriosis

To generate and evaluate new mathematical models to predict POD obliteration in women scheduled for laparoscopy with suspected endometriosis. Multi-centre prospective study (Jan 09–Feb 13). Women with chronic pelvic pain (CPP) +/− history of endometriosis and due for laparoscopy were included. All women underwent standardized history and transvaginal scan (TVS), including assessment for POD obliteration using TVS “sliding sign”. More than 50 historical, clinical and TVS end points were recorded. Two logistic regression (LR) models, LR1 and LR2, were developed on training set and validated on test set. The performances of the models were evaluated using ROC curves. We compared the performance of LR1/LR2 to the TVS “sliding sign” alone. 189 consecutive women with CPP who underwent both TVS and laparoscopy were included: 100 (53%) training set and 89 (47%) test set. 47/189 (25%) had POD obliteration at laparoscopy. LR1 and LR2 included following variables: presence endometriotic nodule (LR1 & LR2), right ovarian fixation (LR1 & LR2), and POD obliteration (i.e. negative ‘sliding sign’) (LR1)/ bowel infiltration on TVS (LR2). LR1 and LR2 gave AUC of 0.970/0.97 for training set and 0.957/0.933 for test set, respectively. LR1 and LR2 accuracy, sensitivity, specificity, PPV, NPV, positive and negative likelihood ratios for training vs. test sets were: 96/95% vs. 95/89%, 93/90% vs. 88/88%, 97/97% vs. 97/89%, 93/93% vs. 88/65%, 97/96% vs. 97/97%, 32.7/31.5 vs. 32.2/8.1, 0.07/0.1 vs. 0.1/0.13, respectively. When the TVS “sliding sign” alone was applied to the training set in the prediction of POD obliteration, the results were: 93.0%, 83.3%, 97.1%, 92.6%, 93.2%, 29.2 and 0.17, respectively. We have generated and evaluated two new LR models to predict the likelihood of POD obliteration. LR1 is the optimal of the two LR models and outperforms the TVS sliding sign alone.