Bioinformatics, Computational and Systems Biology
Integration of Blood Cytokine Levels and Medical Record Data for Predicting Pelvic Organ Prolapse Surgical Outcome
Mihyun Lim Waugh (she/her/hers)
Ph.D. Student
University of South Carolina
cayce, South Carolina, United States
Tyler Mills (he/him/his)
Medical Student
University of South Carolina School of Medicine
Columbia, South Carolina, United States
Nicholas Boltin
Instructor
University of South Carolina, United States
Melissa Moss
Professor and Department Chair
University of South Carolina, United States
Pelvic organ prolapse (POP) encompasses the protrusion of the pelvic organs into the vaginal wall due to weakened pelvic muscles and increased abdominal pressure. Surgical procedures involving the use of polypropylene mesh have demonstrated complications via mesh exposure into the vaginal wall. The resulting corrective reinterventions lead to additional medical costs, surgical risks, and emotional distress. The current landscape provides no standard preoperative method to predict vaginal mesh exposure following surgical intervention. Published methods describe the usage of immune response data or patient medical records for predicting post-surgical complications in a variety of procedures but rarely utilize the combination of both data. Our previous study applied an immunoassay, focusing on biomaterial-induced blood cytokine levels from patients who have undergone POP repair surgery, to predict post-surgical mesh exposure into the vaginal wall. The blood cytokine data demonstrated an ability to predict surgical outcomes using supervised learning models. In our current study, we explored the hypothesis that a combination of biomaterial-induced blood cytokine levels with patient-specific medical record data could augment the predictive accuracy of post-surgical mesh exposure into the vaginal wall.
Samples of blood as well as medical records were collected from 20 female patients who had experienced prior surgical intervention for POP using transvaginal polypropylene mesh. Of these subjects, 10 patients had experienced exposure of the mesh through the vaginal wall following surgery and 10 had not. Blood samples from each patient were split into three equal incubation aliquots (1) with 20 ng/mL of inflammatory agent lipopolysaccharide (LPS) (positive control), (2) with sterile polypropylene mesh (experimental), and (3) without intervention (negative control). After incubation for 24h at 37°C, levels of 13 cytokines were measured in the plasma using multiplex assay. Data were combined with standardized, patient-specific medical record information including vital signs, previous diagnoses, and social history. Following integration, 70% and 30% of the data were split, respectively, into training and testing sets for machine learning models. The models were performed under these conditions to predict the presence or absence of post-surgical mesh exposure.
Results and Discussion: The integration of the cytokine blood data with the medical record information demonstrated a greater predictive accuracy in all four previous models (Table I). When considering variable importance across each model, we were able to visualize relatively even distributions between model weight of cytokine versus medical record data. Both of these observations support our hypothesis that models combining both immune response assays and patient-specific medical record data can improve our ability to predict mesh exposure following surgery.
Conclusions: We were able to improve on our previous models and increase our ability to predict mesh exposure post-POP surgery through data integration. Further application of these models to a larger sample size will be pursued to confirm these results. If confirmed, these models could provide an essential tool for surgeons to make more informed recommendations for POP repair surgery candidates.