A preliminary prediction model using a deep learning software program for postoperative cerebrospinal fluid drainage after thoracic and thoracoabdominal aortic surgery

P0019 

Abstract Submission 

Tomohiro Nakajima (1), Tsuyoshi Shibata (1), Kei Mukawa (1), Keitaro Nakanishi (1), Takakimi Mizuno (1), Ayaka Arihara (1), Shuhei Miura (1), JUNJI NAKAZAWA (1), Yutaka Iba (1), Nobuyoshi Kawaharada (1)

(1) Sapporo Medical University CardioVascular Sugery, Sapporo, Hokkaido,Japan

Tomohiro Nakajima    -  Contact Me
Sapporo Medical University CardioVascular Sugery

Tsuyoshi Shibata    -  Contact Me
Sapporo Medical University CardioVascular Sugery

Kei Mukawa    -  Contact Me
Sapporo Medical University CardioVascular Sugery

Keitaro Nakanishi    -  Contact Me
Sapporo Medical University CardioVascular Sugery

Takakimi Mizuno    -  Contact Me
Sapporo Medical University CardioVascular Sugery

Ayaka Arihara    -  Contact Me
Sapporo Medical University CardioVascular Sugery

Shuhei Miura    -  Contact Me
Sapporo Medical University CardioVascular Sugery

Junji Nakazawa    -  Contact Me
Sapporo Medical University CardioVascular Sugery

Yutaka Iba    -  Contact Me
Sapporo Medical University CardioVascular Sugery

Nobuyoshi Kawaharada    -  Contact Me
Sapporo Medical University CardioVascular Sugery

Tomohiro Nakajima    -  Contact Me
Department of Cardiovascular Surgery, Sapporo Medical University

Objectives: Various prevention and treatment options exist to prevent paraplegia during thoracic aortic surgery. Postoperative cerebrospinal fluid drainage (CSFD) is one treatment option when paraplegia occurs. To evaluate the neurological efficacy of postoperative cerebrospinal fluid drainage (CSFD) in patients undergoing thoracic and thoracoabdominal aortic surgery. In addition, we will use Machine Learning to analyze how many hours postoperatively inserted CSFD is effective.

Methods: This retrospective study included 85 patients who underwent perioperative CSFD in thoracic and thoracoabdominal aortic surgery from January 1, 2006 to December 12, 2022.A total of 61 patients (72%) underwent preoperative CSFD and 24 (28%) postoperative CSFD. Perioperative neurological data were analyzed with a focus on perioperative changes. Machine learning by Prediction One (Sony Network Communications Inc., Tokyo, Japan) was also used to analyze the effect of CSFD placement by how many hours postoperatively.

Results: In the postoperative CSFD group, the manual muscle testing (MMT) score before CSFD was 0.8,2.4 immediately after CSFD, and 3.0 at discharge. Thus, postoperative CSFD was associated with improved MMT scores. Scores improved compared to preoperative scores. Once surgery was completed, the postoperative CSFD was conducted after mean 9.8 hours. Machine learning analysis showed that postoperative CSFD was more effective in patients who had a tendency to improving paraplegia at 2 to 3 hours after surgery and who had a CSFD implanted. The area under curve and the accuracy of the model in the validation data were 0.956 and 96%, respectively. Six of the patients (25%) who underwent early postoperative CSFD remained paralyzed without improvement.

Conclusions: Machine learning analysis showed that patients with postoperative paraplegia after thoracic and thoracoabdominal aortic aneurysm surgery could avoid postoperative paraplegia if the CSFD could be implanted by 3 hours after surgery. Based on these results, we recommend early awakening to improve postoperative paraplegia and placement of a CSFD within 3 hours in patients who need it.

Spinal Cord Protection