P122 Machine learning for prediction of intra-abdominal abscesses in patients with Crohn’s disease visiting the emergency department
Levartovsky, A.(1);Barash, Y.(2,3);Ben-Horin, S.(1);Ungar, B.(1);Klang, E.(2,3);Soffer, S.(3);Amitai, M.M.(2);Kopylov, U.(1);
(1)Sheba Medical center, Department of Gastroenterology, Tel Hashomer, Israel;(2)Sheba Medical center, Department of Diagnostic Imaging, Tel Hashomer, Israel;(3)Sheba Medical center, DeepVision Lab, Tel Hashomer, Israel
Background
Intra-abdominal abscess is an important clinical complication of Crohn’s disease (CD), which can be diagnosed using computed tomography (CT) or magnetic resonance imaging (MRI). However, a high index of clinical suspicion is needed to diagnose an abscess as abdominal imaging is not routinely used during hospital admission. This study aimed to identify clinical predictors of an intra-abdominal abscess among hospitalized patients with CD.
Methods
We created an electronic data repository of all patients with CD who visited the emergency department (ED) of our tertiary medical center between 2012 and 2018. Data included tabular demographic and clinical variables, as well as CT and MRI imaging outcomes. We searched the data repository for the presence of an abscess on abdominal imaging within seven days from the ED visit. Machine learning models were trained to predict the presence of an abscess. A logistic regression model was compared to a random forest model. The area under the receiver operator curve (AUC) was used as a metric. To establish statistical significance, bootstrapping of 100 experiments with random 80/20 training/testing splits was performed. We included only patients who were hospitalized due to complaints that can be attributed to CD exacerbation. Patients presenting within 30 days from an abdominal surgery were excluded.
Results
Overall, 1556 patients with CD visited the ED, of those 555 patients with a CD exacerbation. Of them, 339 patients were hospitalized and underwent abdominal imaging within 7 days from the ED visit. Forty-two patients (12.1%) were diagnosed with an abscess on abdominal imaging. The average length of the abscess was 32 mm (IQR 21.5, 43.5), mainly in the mesentery adjacent to the small bowel (38.1%). On multivariate analysis, high CRP values (64.97 mg/L, aOR 14.42 [95% CI 4.93–42.13]), high platelet count (322.5 K/microL, aOR 4.01 [95% CI 1.97–8.15]), leukocytosis (10.55 K/microL, aOR 3.83 [95% CI 1.71–8.56]) and higher heart rate (over 87.5 beats per minute, aOR 2.58 [95% CI 1.22–5.46]) were independently associated with an intra-abdominal abscess. Overall, random forest and logistic regression showed similar performance. The random forest model showed an AUC of 0.824±0.065 with eight features (CRP, Hemoglobin, WBC, age, current biologic medical treatment, BUN, current immunomodulatory medical treatment, gender).
Conclusion
In our large tertiary center cohort, the machine-learning model identified features associated with the presentation of an intra-abdominal abscess. Such a decision support tool may assist in triaging CD patients for imaging to exclude this potentially life-threatening complication.