x

Virtual Poster Details

Name
10618 - Surgical Based Decision Tree Analysis Modelling to Predict AAGL Endometriosis Surgical Complexity Levels
Presenting Author
Jason Mak
Affiliation
Central Coast Local Health District
Abstract
Study Objective: Develop surgical-based endometriosis severity staging models to predict AAGL surgical complexity levels.

Design: Univariate analysis used to identify surgical features of endometriosis that best predict surgical complexity and build decision tree models of staging

Setting: Multicentre

Patients or Participants: Pre-existing dataset (n= 640) with comprehensive surgical data, collected prospectively

Interventions: Observational

Measurements and Main Results: 4 stage (model 1) and 3 stage (model 2) surgical-based decision tree analysis models were developed. Data were divided into training (n=448 (70%)) and test data (n=192 (30%)). Univariate analysis via chi-squared test was performed on 36 surgical finding categories. The C4.5 algorithm used to identify optimal features for inclusion in decision trees. Models built on training data, applied to test data. Pruning and tuning parameters applied. Concordance of endometriosis staging models and AAGL surgical complexity level was assessed using kappa, weighted kappa coefficients, sens, spec, PPV and NPV. Model 1 correlated with surgical complexity levels (A, B, C and D) and model 2 correlated with (A, B+C and D). Model 1 identified six surgical features (any DE, bowel DE, endometrioma, bowel superficial endometriosis, POD complete obliteration & bladder DE); Model 2 identified six surgical features (any DE, bowel DE, endometrioma, bowel superficial endometriosis, right uterosacral ligament DE & partial POD obliteration). Model 1 (4 stage) accuracy predicting AAGL surgical complexity level (95%CI’s): Kappa 0.60 (0.52-0.69), Weighted Kappa 0.66 (0.57-0.74). Sens/spec/PPV/NPV for A (97.73/87.88/72.88/99.15), B (75.86/92.37/83.02/88.62), C (28.57/94.49/66.67/77.42), D (92.00/86.75/53.49/98.50). Model 2 (3 stage) accuracy predicting AAGL surgical complexity level: Kappa 0.74 (0.65-0.83), Weighted Kappa 0.78 (0.71-0.85). Sens/spec/PPV/NPV for A (100/93.39/87.30/100), B+C (81.32/87.06/87.06/81.32), D (63.33/93.84/67.86/92.57).

Conclusion: Decision tree models for surgical based endometriosis severity staging built on univariate analysis has a high level of agreement with AAGL surgical complexity levels. The 3-stage system outperformed the 4-stage system.

Authors

Mak, JN*. Department of Obstetrics and Gynaecology, Central Coast Local Health District, Holden St, Gosford, NSW, Australia, Eathorne, A. Medical Research Institute of New Zealand, Wellington, New Zealand, Wellington, New Zealand, McClenahan, P. Nepean Hospital, NSW, Australia, New South Wales, Australia, Gil, A. Nepean Hospital, New South Wales, Australia, Condous, G. OMNI Ultrasound and Gynaecological Care, Sydney, Australia

Primary Category
Endometriosis
Secondary Category
Endometriosis
Close