P5: Predicting post-operative infections in CRS/HIPEC patients undergoing splenectomy with a novel machine learning model

Sunday, February 18, 2024

8:45 AM - 8:50 AM AST

Location: Miramar I & II

Oral Poster Presenter(s)

Shannon N. Radomski, MD (she/her/hers)

Resident Physician
Johns Hopkins University School of Medicine
Baltimore, Maryland, United States

Submitter(s)

Shannon N. Radomski, MD (she/her/hers)

Resident Physician
Johns Hopkins University School of Medicine
Baltimore, Maryland, United States

Author(s)

Shannon N. Radomski, MD (she/her/hers)

Resident Physician
Johns Hopkins University School of Medicine
Baltimore, Maryland, United States
NW

Nolan Winicki, MS

Medical Student
University of California Riverside, United States
Yusuf Ciftci, BS

Medical Student
Johns Hopkins University School of Medicine
Baltimore, Maryland, United States
Fabian M. Johnston, MD, MHS

Associate Professor of Surgery
Division of Gastrointestinal Surgical Oncology, Johns Hopkins University School of Medicine, Baltimore, MD, United States
Baltimore, Maryland, United States
Jonathan B. Greer, MD

Assistant Professor of Surgery
Division of Gastrointestinal Surgical Oncology, Department of Surgery, Johns Hopkins University School of Medicine
Baltimore, Maryland, United States

Disclosure(s):

Shannon N. Radomski, MD: No financial relationships to disclose

Yusuf Ciftci, BS: No financial relationships to disclose

Introduction: Hematologic changes following splenectomy make it difficult to differentiate between infection and a normal post-operative response. Heterogeneity in baseline laboratory values due to neoadjuvant chemotherapy or hyperthermic intraperitoneal chemotherapy (HIPEC) can further complicate assessment. The aim of this study was to develop a machine learning model to discriminate between leukocytosis related to a post-operative infection vs. normal hematologic changes following splenectomy in patients undergoing CRS/HIPEC.

Methods:

Patients who underwent splenectomy during CRS/HIPEC from 7/2015 to 12/2022 at a single center were identified. Demographic, comorbidity, daily laboratory (white blood cell [WBC], platelets, lymphocyte, and neutrophil counts/percentages), microbiology, and radiographic data were collected. Patients were divided into infected and non-infected cohorts and the post-operative course was limited to 14 days. Machine learning (XGBoost) was used to model the development of a post-operative infection based on all variables collected.

Results: A total of 96 patients were included in the study, with 67 in the training group (70%) and 29 in the validation group (30%). The median age was 58.5 years (IQR 48.5-66) and 65 (68%) were women. The majority of patients underwent CRS/HIPEC for appendiceal tumors (n=56, 58%), followed by colorectal cancer (n=27, 28%). The remaining 14% (n=13) were for peritoneal mesothelioma, gastric, ovarian, or small bowel cancer. Median WBC levels and the platelet:WBC ratios were not significantly different between the infected and non-infected groups (Figure 1). XGBoost exhibited excellent prediction accuracy to rule out serious, culture-positive, post-operative infection, with an area under the curve of 0.85, negative predictive value of 92%, positive predictive value of 57%, sensitivity of 80% and specificity of 79%. For all other infections, XGBoost displayed high diagnostic capability with an area under the curve of 0.80, negative predictive value of 92%, positive predictive value of 50%, sensitivity of 50% and specificity of 92%.

Conclusions: By utilizing deep machine learning, we developed a novel algorithm to predict the development of a post-operative infection in splenectomy patients. In the future, this tool can be used to not only aid in the early diagnosis of serious infections (prior to culture positivity or imaging confirmation) but also prevent unnecessary imaging by ruling out the development of any infection in post splenectomy CRS/HIPEC patients.

Learning Objectives:

describe post operative trajectories of hematologic laboratory values in patients undergoing splenectomy during CRS/HIPEC
describe how to use a machine learning model to rule out the development of a serious post operative infection in post splenectomy CRS/HIPEC patients based on daily laboratory values
describe how to use a machine learning model to predict the development of any post operative infection in post splenectomy CRS/HIPEC patients based on daily laboratory values