Arterial Interventions and Peripheral Arterial Disease (PAD)
Takafumi Ouchi, MD (he/him/his)
Medical staff
Mie University Hospital, Radiology
Disclosure(s): No financial relationships to disclose
Noriyuki Kato, n/a
Professor
Mie University Hospital, Radiology
Hiroaki Kato, n/a
Clinical Fellow
Mie University Hospital, Radiology
Takatoshi Higashigawa, n/a
Assistant Professor
Mie University Hospital, Radiology
Hisato Ito, n/a
Senior Lecturer
Mie University Hospital, Thoracic and Cardiovascular Surgery
Ken Nakajima, n/a
Attending physician
Ise Red Cross Hospital, Radiology
Shuji Chino, n/a
Manager
Ise Red Cross Hospital, Radiology
Toshiya Tokui, n/a
Manager
Ise Red Cross Hospital, Thoracic Surgery
Kensuke Oue, n/a
Manager
Kochi Health Sciences Center, Cardiovascular Surgery
Toru Mizumoto, n/a
Manager
Anjo Kosei Hospital, Cardiovascular Surgery
Hajime Sakuma, n/a
Professor
Mie University Hospital, Radiology
Identifying preoperatively patients who may obtain a survival benefit from elective thoracic endovascular aortic repair (TEVAR) for thoracic aortic aneurysm (TAA) is important. The aim of this study was to derive and validate a machine learning model that could predict mid-term all-cause mortality after elective TEVAR for TAA.
Materials and Methods:
This study enrolled 257 consecutive patients aged 60 years or more who underwent elective TEVAR for TAA from January 2007 to December 2020. Their medical records were retrospectively reviewed. The patients were randomly divided into training and testing datasets with a ratio of 4:1 to perform 5-fold cross-validation. Before training, synthetic minority over-sampling technique was used to synthesize data from the training set and to compensate a small sample size. Random survival forest was adopted as a machine learning algorithm to predict death from any cause occurring within 8 years of TEVAR. Features were selected from preoperative clinical and imaging characteristics. The performance was evaluated with the validation set using time-dependent receiver operating characteristic curve. The whole cohort was stratified using the machine learning model and the survival difference was assessed using the log rank test.
Results:
The median patient age was 75 years (interquartile range [IQR], 70 to 81 years). Sixty-seven patients died during the follow-up period (median, 4.4 years; IQR, 2.0 to 6.8 years). The 8-year overall survival rate was 59% (95% confidence interval [CI], 51 to 68%) for the whole cohort. The machine learning model resulted in an area under the curve of 0.83 (95% CI, 0.70 to 0.95) at 1-year and 0.78 (95% CI, 0.70 to 0.86%) at 5-year. The most important predictor of mortality was the number of comorbidities. The whole cohort was stratified into three groups, low-risk, moderate-risk, and high-risk groups. The 8-year overall survival rates were 94% (95% CI, 86 to 100%) for the low-risk, 64% (95% CI, 50 to 83%) for the moderate-risk, and 4% (95% CI, 1 to 23%) for the high-risk group (P < .001).
Conclusion:
The machine learning model showed good performance to predict mid-term all-cause mortality after elective TEVAR for TAA. The application of TEVAR in the high-risk group will need to be a careful decision.