Amudha Thangavelu

Background/Objectives: Malnutrition is common among women undergoing gynaecologic oncology (GO) surgery and is associated with increased morbidity, prolonged hospitalisation, and reduced survival. Nevertheless, the optimal nutritional screening tools remain uncertain. Methods: We conducted a narrative review of commonly used nutritional screening and assessment tools in surgical GO patients. To highlight practical challenges in accurately identifying at risk individuals, we incorporated findings from our clinical audit. Results: There was a considerable variation between tools. While many tools were associated with adverse outcomes, their clinical value in this population was unclear. The presence of ascites and rapid deterioration in oral intake may contribute to under-recognition of at-risk patients, as illustrated by our audit findings. Emerging strategies including determining body composition from routine pre-operative Computed Tomography (CT) scans, which have shown statistical associations with survival and toxicity in observational studies, but their clinical utility is not yet established. Conclusions: Although several screening tools were statistically associated with adverse outcomes, robust data on their clinical utility are lacking. Current tools may inadequately factor for the specific considerations when screening this group. Consequently, nutritionally vulnerable surgical GO patients requiring nutritional intervention may be missed. As no gold standard currently exists for this population, bespoke, objective approaches and prospective studies are urgently needed to address disease-specific nutritional considerations.

Background/Objectives: Nutritional risk screening is critical in the management of gynaecologic oncology (GO) surgical patients. Malnutrition is prevalent in this population and is associated with poorer surgical outcomes, including increased morbidity, prolonged hospital stays, and reduced survival rates. Nevertheless, the optimal nutritional screening tools for this patient group remain undefined. Methods: We conducted a narrative review to critically appraise commonly used nutritional screening and assessment tools in surgical GO patients. To highlight practical challenges in accurately identifying at-risk individuals, we incorporated findings from our recent clinical audit. Results: Several nutritional screening and assessment tools were identified. The results varied considerably between tools. The presence of ascites and rapid deterioration in oral intake were frequently overlooked, leading to under-recognition of malnutrition. These issues were corroborated by our audit findings. Emerging strategies including determining body composition from routine preoperative CT scans show promise. Conclusions: Accurate nutritional assessment is imperative to improve surgical outcomes in surgical GO patients. As currently no gold standard currently exists for this population, bespoke approaches to address disease-specific nutritional considerations are urgently needed to identify those at risk and allow for timely nutritional interventions. Integrating CT-based body composition analysis can provide an objective solution, thus requiring further investigation.

Background/Objectives: The advancement of natural language processing (NLP) technologies has transformed various sectors. However, their application in the healthcare domain, particularly for analysing clinical notes, remains underdeveloped. We investigated the use of deep neural networks, specifically transformer-based models, to predict intraoperative and post-operative outcomes related to advanced-stage epithelial ovarian cancer cytoreduction (aEOC) using unstructured surgical notes. Methods: We evaluated the performance of RoBERTa, a general-purpose language model, and GatorTron, a domain-specific model, across eight binary classification tasks using the same dataset. The dataset consisted of 560 surgical records from patients with aEOC who underwent cytoreductive surgery at a tertiary UK reference centre. Predictive outcomes were converted into binary features to facilitate classification tasks. To enhance the contextual information available to the models, textual data from “operative findings” and “operative notes” were concatenated. Results: Our findings highlight the tangible benefits of employing domain-specific language models for clinical text analysis. GatorTron generally outperformed RoBERTa across most predictive tasks, underscoring the advantages of domain-specific pretraining for understanding medical terminology and context. Both models struggled to predict certain outcomes, particularly those involving post-operative events like major complications and length of hospital stay, despite adjustments in hyperparameters and training strategies. This limitation suggests that operative text alone may not sufficiently capture the complexities of post-operative recovery. Conclusions: These findings have valuable implications for developing medical AI systems to improve the delivery of modern aEOC healthcare.

The advancement of natural language processing (NLP) technologies has transformed various sectors. However, their application in the healthcare domain, particularly for analysing clinical notes, remains underdeveloped. We investigated the use of deep neural networks, specifically transformer-based models, to predict intraoperative and postoperative outcomes related to advanced-stage epithelial ovarian cancer cytoreduction (aEOC) using unstructured surgical notes. We evaluated the performance of RoBERTa, a general-purpose language model and GatorTron, a domain-specific model, across eight binary classification tasks using the same dataset. The dataset consisted of 560 surgical records from patients with aEOC who underwent cytoreductive surgery at a tertiary UK reference centre. Predictive outcomes were converted into binary features to facilitate classification tasks. To enhance the contextual information available to the models, textual data from “operative findings” and “operative notes” were concatenated. Our findings highlight the tangible benefits of employing domain-specific language models for clinical text analysis. GatorTron generally outperformed RoBERTa across most predictive tasks, underscoring the advantages of domain-specific pretraining for understanding medical terminology and context. Both models struggled to predict certain outcomes, particularly those involving postoperative events like major complications and length of hospital stay, despite adjustments in hyperparameters and training strategies. This limitation suggests that operative text alone may not sufficiently capture the complexities of postoperative recovery. These findings have valuable implications for developing medical AI systems to improve the delivery of modern aEOC healthcare.

Background/Objectives: The incidence of acute kidney injury (AKI) following advanced epithelial ovarian cancer (EOC) surgery has not been extensively studied. This study aimed to investigate the incidence of AKI and identify preoperative and intraoperative predictors in patients undergoing advanced EOC cytoreduction using both traditional statistics and Artificial Intelligence (AI) modelling. Methods: Retrospective data was collected for 134 patients with a suspected or confirmed diagnosis of advanced EOC (FIGO Stage III-IV) who underwent surgical cytoreduction between January 2021 and December 2022 at a UK tertiary referral centre. AKI was diagnosed according to the KDIGO criteria. Data on 22 patient variables were extracted, including age, Charlson Comorbidity Index (CCI), procedure length, surgical complexity, and length of hospital stay. Logistic regression analysis was used for feature selection to identify AKI predictors, and an extreme gradient boost (XGBoost) model was applied to all variables related to AKI events. Results: The incidence of postoperative AKI was 6.72% (n=9). Predictive factors for AKI included younger age (OR = 0.942, p=0.037), lower CCI (OR = 0.415, p=0.015), longer procedure duration (OR = 1.006, p=0.019), and greater surgical effort (OR = 1.427, p=0.007). Patients with perioperative AKI experienced a doubling in the length of hospital stay (p=0.008). Mortality rates were similar between patients with and without AKI. AI-driven algorithms highlighted the complexity of AKI prediction and provided individual risk profiles, enabling future stratification and prompting different frequencies of AKI monitoring following cytoreduction. Conclusions: Predicting AKI is a complex task. This study found a lower-than-expected incidence of AKI following advanced EOC cytoreductive surgery. AKI is linked to heightened surgical risk-taking, underscoring the need for improved guidelines focusing on postoperative monitoring for targeted patients. Artificial Intelligence offers the potential for personalized AKI prediction.

There is no well-defined threshold for intra-operative blood transfusion (BT) in advanced epithelial ovarian cancer (EOC) surgery. To address this, we devised a Machine Learning (ML)-driven prediction algorithm aimed at prompting and elucidating a communication alert for BT based on anticipated peri-operative events independent of existing BT policies. We analyzed data from 403 EOC patients who underwent cytoreductive surgery between 2014 and 2019. The estimated blood volume (EBV), calculated using the formula EBV = weight × 80, served for setting a 10% EBV threshold for individual intervention. Based on known estimated blood loss (EBL), we identified two distinct groups. The Receiver operating characteristic (ROC) curves revealed satisfactory results for predicting events above the established threshold (AUC 0.823, 95% CI 0.76–0.88). Operative time (OT) was the most significant factor influencing predictions. Intra-operative blood loss exceeding 10% EBV was associated with OT > 250 min, primary surgery, serous histology, performance status 0, R2 resection and surgical complexity score > 4. Certain sub-procedures including large bowel resection, stoma formation, ileocecal resection/right hemicolectomy, mesenteric resection, bladder and upper abdominal peritonectomy demonstrated clear associations with an elevated interventional risk. Our findings emphasize the importance of obtaining a rough estimate of OT in advance for precise prediction of blood requirements.

The Surgical Complexity Score (SCS) has been widely used to describe the surgical effort during advanced stage epithelial ovarian cancer (EOC) cytoreduction. Referring to a variety of multi-visceral resections, it best combines the numbers with the complexity of the sub-procedures. Nevertheless, not all potential surgical procedures are described by this score. Lately, the European Society for Gynaecological Oncology (ESGO) has established standard outcome quality indicators pertinent to achieving complete cytoreduction (CC0). There is a need to define what weight all these surgical sub-procedures comprising CC0 would be given. Prospectively collected data from 560 surgically cytoreduced advanced stage EOC patients were analysed at a UK tertiary referral centre.We adapted the structured ESGO ovarian cancer report template. We employed the eXtreme Gradient Boosting (XGBoost) algorithm to model a long list of surgical sub-procedures. We applied the Shapley Additive explanations (SHAP) framework to provide global (cohort) explainability. We used Cox regression for survival analysis and constructed Kaplan-Meier curves. The XGBoost model predicted CC0 with an acceptable accuracy (area under curve [AUC] = 0.70; 95% confidence interval [CI] = 0.63–0.76). Visual quantification of the feature importance for the prediction of CC0 identified upper abdominal peritonectomy (UAP) as the most important feature, followed by regional lymphadenectomies. The UAP best correlated with bladder peritonectomy and diaphragmatic stripping (Pearson’s correlations > 0.5). Clear inflection points were shown by pelvic and para-aortic lymph node dissection and ileocecal resection/right hemicolectomy, which increased the probability for CC0. When UAP was solely added to a composite model comprising of engineered features, it substantially enhanced its predictive value (AUC = 0.80, CI = 0.75–0.84). The UAP was predictive of poorer progression-free survival (HR = 1.76, CI 1.14–2.70, P: 0.01) but not overall survival (HR = 1.06, CI 0.56–1.99, P: 0.86). The SCS did not have significant survival impact. Machine Learning allows for operational feature selection by weighting the relative importance of those surgical sub-procedures that appear to be more predictive of CC0. Our study identifies UAP as the most important procedural predictor of CC0 in surgically cytoreduced advanced-stage EOC women. The classification model presented here can potentially be trained with a larger number of samples to generate a robust digital surgical reference in high output tertiary centres. The upper abdominal quadrants should be thoroughly inspected to ensure that CC0 is achievable.

There is no well-defined threshold for intra-operative blood transfusion (BT) in advanced epithelial ovarian cancer (EOC) surgery. We developed a Machine Learning (ML)-driven prediction algorithm to trigger and explain a communication alert for BT based on anticipated peri-operative events irrespective of BT policies. We analysed prospectively collected data from 403 EOC patients who underwent cytoreductive surgery between 2014 and 2019. We calculated the estimated blood volume (EBV) using the formula EBV = weight x 80 and set off 10%EBV as threshold for individual intervention. Based on the known estimated blood loss (EBL) we identified two groups. We employed Receiver operating characteristic (ROC) curves for performance metrics. The model performance for the above threshold prediction was satisfactory (AUC 0.823, 95% CI 0.76-0.88). The top feature commonly shared between interrogators was operative time (OT). Intra-operative blood loss of at least 10%EBV was associated with OT>250 minutes, primary surgery, serous histology, performance status 0, R2 resection and surgical complexity score >4. Large bowel resection, stoma formation, ileocecal resection/right hemicolectomy, mesenteric resection, bladder and upper abdominal peritonectomy were amongst sub-procedures clearly associated with increased intervention risk. Precise prediction of blood requirements is not possible unless a rough estimate of OT is known in advance.

The Surgical Complexity Score (SCS) has been widely used to describe the surgical effort during advanced stage epithelial ovarian cancer (EOC) cytoreduction. Referring to a variety of multi-visceral resections, it best combines the numbers with the complexity of the sub-procedures. Nevertheless, not all potential surgical procedures are described by this score. Lately, the European Society for Gynaecological Oncology (ESGO) has established standard outcome quality indicators pertinent to achieving complete cytoreduction (CC0). There is a need to define what weight all these surgical sub-procedures comprising CC0 would be given. Explainable Artificial Intelligence (XAI) could explain the impact of real-time features on the CC0 prediction. We analyzed prospectively collected data from 560 consecutive patients with FIGO-stage III-IV who underwent cytoreductive surgery between Jan 2014 and Dec 2019 at a UK tertiary referral centre. Following adaptation of the structured ESGO ovarian cancer report template, we employed the eXtreme Gradient Boosting (XGBoost) algorithm to model an exhaustive list of surgical sub-procedures. We applied the Shapley Additive explanations (SHAP) framework to provide global (cohort) explainability. We used Cox regression for survival analysis and constructed Kaplan-Meier curves. The XGBoost model predicted CC0 with an acceptable accuracy (area under curve [AUC] = 0.70; 95% confidence interval [CI] = 0.63–0.76). Visual quantification of the feature importance for the prediction of CC0 identified upper abdominal peritonectomy (UAP) as the most important feature, followed by regional lymphadenectomies. The UAP best correlated with bladder peritonectomy and diaphragmatic stripping (Pearson’s correlations > 0.5). Clear inflection points were shown by pelvic and para-aortic lymph node dissection and ileocecal resection/right hemicolectomy, which increased the probability for CC0. When UAP was solely added to a composite model comprising of engineered features, it substantially enhanced its predictive value (AUC = 0.80, CI = 0.75-0.84). The UAP was predictive of poorer progression-free survival (HR=1.76, CI 1.14- 2.70, P:0.01) but not overall survival (HR=1.06, CI 0.56-1.99, P:0.86). The SCS did not have significant survival impact. Machine Learning allows for operational feature selection by weighting the relative importance of those surgical sub-procedures that appear to be more predictive of CC0. Our study identifies UAP as the most important procedural predictor of CC0 in surgically cytoreduced advanced-stage EOC women. The classification model presented here can potentially be trained with a larger number of samples to generate a robust digital surgical reference in high output tertiary centres. The upper abdominal quadrants should be thoroughly inspected to ensure that CC0 is achievable.