Artificial intelligence-assisted risk prediction of postoperative pulmonary complications in non-small cell lung cancer surgery

Comprehensive Summary

Surgery for non-small cell lung cancer (NSCLC) is the most common treatment option, and, with this, the primary post-operative complication involves the lungs, ranging from 20-60%. The pulmonary problems involve air leaks, pneumonia, atelectasis, secretion retention, bronchopleural fistula, and empyema, all of which primarily stem from previous lung health, smoking history, muscle weakness, and pain. With the use of a combination of artificial intelligence (AI) and deep learning, the aim of this study is to accurately predict patient who will undergo postoperative pulmonary complications using a Fully Connected Neural Network (FCNN) deep learning model. The data gathered from the patients’ included demographics, laboratory data, respiratory function, type of surgery, PET-CT scan, and type of pulmonary disease outcome which totaled to 24 variables being incorporated into the FCNN model. The data of 953 patients was used to train the model to an sensitivity rate of 66.4%, a positive predictive value of 89.8%, and an accuracy rate of 88.6%. Meanwhile, the test dataset was used to determine the same metrics which recorded a score of 65.4% sensitivity, 100% PPV, and 90.4% accuracy with an area under the curve of .84, which indicated a strong ability to predict negative pulmonary outcomes. An F1 score was calculated which combines both sensitivity and specificity, which resulted in a rate of 84.4% for the training data and 86.4% for the test data, indicating a very good performance. Overall, this model strongly predicted outcomes and can be used to identify high risk patients’ post-operation which can lead to better pulmonary health and survival outcomes.

Outcomes and Implications

This FCNN model proved to be accurate in determining post-operative risks for pulmonary disease following non-small cell lung cancer surgical resection. This is vital in improving the overall outcome of these surgeries as it allows the healthcare team to prioritize the high-risk patients, preventing increased incidence of pulmonary disease following surgery. With the successful integration of artificial intelligence into post-operative care for non-small cell lung cancer, the ultimate goal could be having an AI model for every surgery looking at every risk, notifying the healthcare team when a patient has become high risk for a surgery and allowing the healthcare team with the patient make a decision about the surgery itself and what post-operative rehabilitation may look like. The integrated AI models can be added into health charts, taking existing data from PET-CT scans, demographics, and other health metrics while becoming adaptive to the patient’s current health, providing important warnings to physicians about the risks of undergoing surgery. While artificial intelligence should always be alongside the judgement of the physician, it can lead to an increase in overall patient satisfaction, quality of life, and survival following surgeries.