Search

Introduction/Background* The prognosis of advanced stage high grade serous ovarian cancer patients (HGSOC) is multifactorial, and could be accurately predicted by using Machine Learning (ML) algorithms. We designed a study to support the feature selection of selected clinical variables to define their relative survival impact on two-year prognosis prediction in HGSOC patients, who received surgical treatment. Methodology This was a retrospective analysis of 209 FIGO stage III-IV HGSOC women, who were scheduled for cytoreductive surgery in SJUH, Leeds between Jan 2015 to Dec 2018 with curative or life-prolonging intent. The two-year prognosis estimation was formulated as a binary classification problem. Dataset was split into training (80%) and test (20%) cohorts with repeated random sampling until there was no significant difference (p=0.20) between the two cohorts. A ten-fold cross-validation was applied. Various state-of-the-art supervised ML classifiers were tested, including Support-Vector-Machines (SVMs), K-Nearest Neighbors (KNNs), Ensemble Classifiers, and Naive Bayes, based on a set of performance metrics. These results were directly compared to conventional Logistic Regression (LR). For feature selection, multivariate feature ranking using the MRMR method was carried out. Result(s)* Two hundred nine patients were identified. The model’s mean prediction accuracy reached 73%. We demonstrated that SVM and Ensemble Discriminant algorithms outperformed Logistic Regression in accuracy indices. The probability of achieving a cancer-free state was maximized with a combination of primary cytoreduction, good performance status, and maximal surgical effort (AUC 0.63). Standard chemotherapy, performance status, tumor load, and residual disease were consistently predictive of the two-year overall survival (AUC 0.63-0.66) (figure 1). The model recall and precision were greater than 80%. Conclusion* Appropriate feature selection is required when building a HGSOC model for two-year prognosis prediction. For HGSOC prognosis, one should consider not only the patient’s disease burden but also their overall medical status and ability to undergo extensive surgery, resulting in survival benefits alongside with standard chemotherapy.

Introduction/Background* In advanced stage high grade serous ovarian cancer (HGSOC), the introduction of maximal surgical effort without compromising peri-operative management and subsequent recovery to achieve complete cytoreduction, requires Critical Care Unit (CCU) availability. This paradigm shift prompts the development of tools to accurately predict CCU admission following cytoreductive surgery. Modern data mining technology, such as Machine Learning (ML) could be helpful in accurately predicting CCU admissions to improve standards of care. We developed a framework to improve the accuracy of predicting CCU admission in HGSOC patients by use of ML algorithms (figure 1). Methodology A cohort of 291 advanced stage HGSOC patients, who underwent surgical cytoreduction from Jan 2014 to Dec 2019, was selected from the ovarian database. They were randomly assigned to training (60%) and test (40%) sub-cohorts. Forward selection and backward stepwise regression were employed to screen independent pre- and intra-operative variables. Linear (LDA), Quadratic (QDA), and non-linear distance (ANN and KNN) ML models were employed to derive predictive information. These methods were tested against conventional linear regression (LR). Model performance was evaluated by prediction accuracy, sensitivity, specificity, and F1 scores. Result(s)* We identified 56/291(19.2%) CCU admissions. For the outcome of CCU admission, the prediction accuracies were higher for LDA (0.90) and QDA (0.93) compared with LR (0.84) when all the variables were included in the in-built model. Feature selection identified pre-treatment albumin, surgical complexity score, estimated blood loss, operative time, and bowel resection with stoma formation as the most significant prediction features. With feature selection, the prediction accuracies were higher for LDA (0.89) and KNN (0.86) compared with LR (0.82). Admission to CCU was associated with increased length of stay (P = 0.000), and decreased number of postoperative complications (P = 0.001). Conclusion* Herein, ML algorithms accurately predicted HGSOC patients, who required CCU admission following their cytoreductive surgery. Linear discriminant analysis was consistently more predictive than LR for CCU admission, irrespective of the number of features included in the analysis. Limited, potentially modifiable, mostly intra-operative factors contributing to CCU admission were identified and suggest areas for targeted interventions.

Introduction/Background Clarity and precision about the anatomical extent of disease in cancer is essential for prognostication, research, and cancer-control activities. To select effective therapeutic approaches for advanced high-grade serous ovarian cancer (HGSOC), yet the most prevalent and lethal form, it is important to identify stratification factors that could accurately predict prognosis before initial intervention. We hypothesized that women with different tumor dissemination patterns at pre-treatment imaging would have different prognosis. Methodology This was a retrospective analysis of 209 FIGO stage III-IV HGSOC women, who were scheduled for cytoreductive surgery in SJUH Leeds between Jan 2015 to Dec 2018 with curative or life-prolonging intent. CT scans were reported by an MDT radiologist. Three pre-treatment imaging dissemination patterns were identified and verified by final histology. A Cox proportional hazard analysis was used to test the effect of imaging dissemination patterns, age, performance status (PS), timing of surgery (upfront vs delayed cytoreduction), surgical complexity score (SCS), residual disease (RD), disease score, and type of chemotherapy on survival. Kaplan-Meier survival curves were produced using SPSS® 26. Results There were no statistical differences in the cytoreduction rates amongst the three groups (figure 1). The mean progression free survival (PFS) for patients grouped as intraperitoneal (n=137), intraperitoneal and lymphatic (n=56), and intraperitoneal and haematogenous (n = 16) was 26.5 (95% CI 23.4–29.6), 21.3 (95% CI 18.3–24.4) and 19.1 months (95% CI 15.1–22.9), respectively. The mean overall survival (OS) was 45.8 (95% CI 41.5–50.2), 34.8 (95% CI 29.2–40.3) and 30.7 months (95% CI 24.5–36.9), respectively (p=0.05) (figure 2). The mean PFS and OS for the entire cohort was 25 months (95% CI 22.6–27.3) and 41.8 (95% CI 38.3–45.2), respectively. For PFS, Cox regression analysis identified PS (HR 1.23, 95% CI 1.1–1.5, p=0.04), RD (HR 0.69, 95% CI 0.46–0.98, p=0.05) as statistically significant. For OS, Cox regression analysis identified PS (HR 1.47, 95% CI 1.14–1.89, p=0.03), dissemination pattern (HR 1.36, 95% CI 1.02–1.86, p=0.05) as statistically significant. Conclusion For HGSOC prognosis, one should consider not only the patient’s disease burden but also their overall medical status and ability to undergo extensive surgery. Prolonged survival rates were found predominantly in those patients with intraperitoneal only pre-treatment imaging dissemination pattern. Baseline tumor dissemination pattern can be a prognostic factor for overall survival. Classification of such patterns can help counsel patients initially on their prognosis and identify those who might benefit from intraperitoneal chemotherapy. Disclosures No disclosures.

Introduction/Background Accurate prediction of patient prognosis can be especially useful for the selection of best treatment protocols. Machine Learning (ML) can serve this purpose by making predictions based upon generalizable clinical patterns embedded within learning datasets. We hypothesised that use of ML algorithms could improve prognosis estimation in advanced high grade serous ovarian (HGSOC) patients. We aimed to compare the performance of two ML prediction methods for HGSOC prognosis, based on Area Under Curve (AU-ROC) performance for a 2-year prognosis period. Methodology This was a retrospective analysis of 209 FIGO stage III-IV HGSOC women, who were scheduled for cytoreductive surgery in SJUH, Leeds between Jan 2015 to Dec 2018 with curative or life-prolonging intent. Support-Vector-Machine (SVM) and K-Nearest Neigbors (K-NN) were employed to model prognosis. The prognosis estimation problem was formulated as a binary classification problem. For the 2-year prognosis period, two groups were defined using patient survival information; patients who did not relapse or survived beyond two years were labelled in the positive class and patients who relapsed or died before reaching that period were considered in the negative class. The study was restricted to the most common prognostic variables and focused on predictive model comparisons. Dataset was split into training and test cohorts with repeated random sampling until there was no significant difference (p=0.20) between the two cohorts with respect to all variables. Results 172 out of 209 patients with fully curated data were eligible for 2-year prognosis prediction analysis. 104/172 (60%) and 55/172 (32%) patients had disease recurrence or died of disease within two years, respectively. The variable importance for the 2-year progression free survival (PFS) and overall survival (OS) is shown in figure 1. A combination of good performance status, upfront cytoreduction and increased surgical complexity score best predicted 2-year PFS with an accuracy of 63% and 62.1% for the SVM and K-NN classifiers, respectively. SVM best predicted 2-year OS by a combination of Carboplatin/Taxol chemotherapy, low disease score, no residual disease, increased surgical complexity score, and upfront cytoreduction with an accuracy of 71.6% ( AU-ROC: 0.66) (figure 2). Conclusion ML appears to be promising for accurate estimation of HGSOC prognosis. We provide evidence as to what combination of prognosticators leads to the largest impact on the HGSOC two-year prognosis. The cohort is currently expanding to further examine the short term vs long term contribution of the clinical variables from the comparative models Disclosures No disclosures.

Introduction/Background The foundation of modern ovarian cancer care is cytoreductive surgery to remove all macroscopic disease (R0). Identification of R0 resection patients may help individualise treatment. Machine learning and AI have been shown to be effective systems for classification and prediction. For a disease as heterogenous as ovarian cancer, they could potentially outperform conventional predictive algorithms for routine clinical use. We investigated the performance of an AI system, the k-nearest neighbor (k-NN) classifier to predict R0, comparing it with logistic regression. Methodolog A cohort of patients diagnosed with high grade serous advanced ovarian, tubal and primary peritoneal cancer (HGSOC), undergoing surgical cytoreduction from 2015–2019, was selected from the ovarian database. Performance variables included age, BMI, Charlson Comorbidity Index (CCI), timing of surgery, surgical complexity and disease scores. The k-NN algorithm classified R0 vs non-R0 patients using 3–20 nearest neighbors. Prediction accuracy was estimated as percentage of observations in the training set correctly classified. The relative importance of the selected variables was quantified by calculating the prediction accuracy/error rate in relation to the number of predictors included in the models. Results 154 patients were identified, with mean age of 64.4 + 10.5 yrs, BMI of 27.2 +5.8 and mean SCS of 3 + 1 (1–8). Complete and optimal cytoreduction was achieved in 62% and 88% patients. The mean predictive accuracy was 66% compared to 63.4% by logistic regression. R0 resection prediction of true negatives was as high as 90% using k=20 neighbors. From the variables tested to contribute in R0 prediction, only disease score was statistically significant (p = 0.0006). For a given neighborhood size k=15, R0 resection was best predicted by a kNN model that included age and CCI (figure 1). Conclusion The k-NN algorithm is a versatile and promising tool for R0 resection in HGSOC patients, which outperforms logistic regression. The model, which is very much reflective of ‘previous clinical experience’ can be directly available to clinicians and is expected to improve accuracy with data expansion. Disclosures No disclosures.