Article Preview
TopIntroduction
Uterine fibroids occur in women of reproductive age between 4.5% and 30% of the time, with over 50% of those affected claiming it negatively impacted their life (Fuldeore, & Soliman, 2017; Jayakumaran, et al., 2017; Zimmermann, et al., 2012). Myomectomy is a surgical procedure to remove fibroids that leaves the uterus intact for future pregnancies. Myomectomies may be conducted using a variety of surgical techniques including both minimally invasive techniques and standard surgical non-minimally invasive techniques. However, myomectomies are not without risk. Perioperative morbidity occurs in as high as 39% of patients (Andrade, et al., 2017; Frederick, et al., 2002; Sawin, et al., 2000) and although much less likely mortality still occurs (Choo, Yeo, & Thomas, 1998; Hamoda, Tait, & Edmonds, 2009; Jin, et al., 2009; Katz, et al., 2001).
Estimated blood loss (EBL) during surgery has been shown to be a statistically significant (p < 0.05) indicator of all perioperative morbidities using a univariate logistic regression model and also an indicator of serious perioperative morbidities using a multivariate logistic regression (Carson, et al., 2002). In addition to indicating perioperative morbidity, EBL has also been shown to be a strong indicator for other surgical outcomes including hospital length of stay (Jarnagin, et al., 2002; Sørensen, et al., 2005) and recurrence of a carcinoma (Katz, et al., 2009). The ability to preoperatively predict blood loss would allow for improved implementation of blood conservation techniques and would alert surgeons to the likelihood of related complications (Yu, et al., 2013). However, EBL is extremely difficult to measure accurately (Algadiem, et al., 2016; Eipe, & Ponniah, 2006; Guinn, et al., 2013). Prior research emphasizes the need to develop more accurate mathematical models to determine EBL (Brecher, Monk, & Goodnough, 1997).
Actual blood loss, estimated by EBL, is the most significant reason for operative and postoperative transfusions (Chang et al., 2001; Stoller, Wolf, & St. Lezin, 1994). Transfusions help to reduce mortality in patients with significant intraoperative blood loss (Wu, et al., 2010; Wu et al., 2012) and have been associated with increased postoperative morbidities (Frederick, et al., 2002). Therefore, preoperative knowledge of EBL and transfusion requirements of myomectomy patients will enable surgeons and clinicians to be better prepared to deal with probable postoperative morbidities and possible mortality.
Predictive models in medicine are most commonly developed using regression (Xie, et al., 2017). Other popular machine learning-based predictive modelling methods for clinical decision making include artificial neural networks (ANNs) and decision trees (Kourou, et al., 2015; Xie, et al., 2017). Predicting EBL is correlated with predicting transfusions and should be amenable to similar predictive informatics strategies. Prior research shows that ANNs are both a frequently used and an effective problem solving method in medicine (Cruz, & Wishart, 2006; Cunningham, Carney, & Jacob, 2000; Dreiseitl, & Ohno-Machado, 2002; Shaikhina, & Khovanova, 2017; Walczak, 2018) and for this reason ANNs are selected to try and model both EBL and transfusions.
This article examines the use of separate ANNs to predict intraoperative EBL and also perioperative transfusions and additionally introduces the novel approach in medicine of developing an ensemble ANN to predict both EBL and transfusions simultaeneously. This addresses a gap in current research where ANNs have been used to predict transfusion requirements, but primarily for cardio-circulatory surgeries (Covin et al., 2003; Walczak & Scharf, 2000). Prior research has not utilized ANNs to predict EBL or transfusions for gynecologic surgeries. A positive outcome may be able to improve surgeon and other clinician decision making regarding the intraoperative and postoperative care of these patients to help reduce morbidities and mortalities.