Prediction of postoperative final degree and recurrence of pectus excavatum using machine learning algorithms
Kwanyong Hyun1, Jae Jun Kim2, Won Kyu Choi3
1Department of Thoracic and Cardiovascular Surgery, St. Vincent's Hospital, College of Medicine, The Catholic University of Korea, Gyeonggi-do, Republic of Korea.
Background:
Chest wall re-depression after bar removal (BR) in pectus excavatum (PE) is insufficiently investigated. However, it is not easy to investigate chest wall re-depression due to its multifactorial characteristics. Herein, we investigated chest wall re-depression after BR using machine learning algorithms. To the best of my knowledge, this is the first study of chest wall re-depression after BR using machine learning algorithms.
Methods:
We retrospectively reviewed 199 consecutive subjects who underwent both minimally invasive repair of pectus excavatum (MIRPE) and BR at a single hospital from March 2012 to June 2020. We investigated attributes of chest wall re-depression and risk factors for recurrence after BR, predicted final degree and recurrence of PE after BR, and suggested the optimal age at the time of MIRPE based on recurrence. Data for the chest wall re-depression were analyzed to discover differences according to age group [<10 years (early repair group; EG) vs. ≥10 years (late repair group; LG)].
Results:
We observed no significant difference between the Haller index and radiographical pectus index (RPI) (P=0.431) and a significant correlation between Haller index and RPI (P<0.001). RPI significantly increased for the first 6 months after BR in both age groups (both P<0.001) and was maintained at 1 year after BR. RPI value of the LG were significantly higher than those of the EG for the entire period after MIRPE (P=0.041). Recurrence of PE in the LG was significantly more frequent than in the EG (P<0.001). RPI values before and after MIRPE and age group were identified as independent risk factors for recurrence after BR (P<0.001, P=0.007, and P=0.001, respectively). The linear regression model outperformed for final RPI with performance scores of mean squared error 0.198, root mean squared error 0.445, mean absolute error 0.336, and R2 0.415. In addition, the logistic regression model outperformed for predicting recurrence with performance scores of 0.865 the area under the curve, 0.884 accuracy, 0.859 F1, 0.865 precision, and 0.884 recall.
Conclusions:
The present study shows that machine learning algorithms can provide good estimates for postoperative results in PE. An approach integrating machine learning models and readily available clinical data can be used to create other models in the thoracic surgery field.
