BACKGROUND: Optimal management of chest tubes can be challenging, and delays in chest tube removal are common even when surgical experts lead patient care. The duration of chest tube drainage remains a significant driver of length of stay after lung resection. This study aims to evaluate the performance of an artificial intelligence (AI) decision support system (CheLSEA) in generating safe and efficient recommendations for chest tube management.
METHODS: A machine leaning model to determine if chest tube(s) can be safely removed after lung resection was developed using 296 patients from a single institution. Receiver operating characteristic analysis and decision thresholds were used to minimize premature removal recommendations while maximizing efficiency in chest drain management. Clinical status, digital pleural drainage data, and chest X-ray data (pneumothorax, subcutaneous emphysema, and pleural effusion grades) were collected as part of a prospective, single-arm, double-blinded observational (silent mode) trial to evaluate the performance of the system against standard chest tube care. From October 2020 to May 2021, 50 adult elective pulmonary resection patients with at least 24 hours of chest tube drainage were enrolled and 3 were excluded due to incomplete drainage data.RESULTS: Most patients were female (29/47; 62%), smokers (39/47; 83%), with a median age of 73 (IQR = 66-77), a median BMI of 24 (IQR = 22-28) who underwent minimally invasive (44/47; 94%) lobectomy (41/47; 87%) for primary non-small cell lung cancer (35/47; 75%). Median duration of chest tube drainage was 74 hours (IQR = 53-97) and median length of stay was 5 days (IQR = 5-6). ChelSEA was queried in 174 instances for chest tube management recommendations (median of 3/patient (IQR = 3-5). Of these requests, 21% (36/174) triggered CheLSEA’s safeguard subsystem due to severe or increasing subcutaneous emphysema (20/36; 56%) or patients were deemed clinically unstable (10/36; 28%). CheLSEA recommended chest tube removal in 9% of remaining requests (12/13; n=9), 83% of which (10/12) were safe and 17% (2/12) could have resulted in premature chest tube removal by only 6 hours in both cases. The remaining 126 requests were answered with a recommendation to maintain chest tube drainage before and up to the optimal removal time in 76% (97/126), or shortly after the optimal removal time in 24% (29/126; median delay of 17 hours; IQR=17-22). There would have been no delays as a result of CheLSEA’s recommendations to maintain drainage when compared to the actual time when chest tube(s) were removed by the surgical team. In 67% of the maintain recommendations (85/126), a prediction of future chest tube removal time was made by the system and ultimately found to be within +/-12 hours of the optimal removal time in 42% (36/85; median of 1 hour; IQR=-5 to 1).CONCLUSIONS: AI-based clinical decision support systems can provide safe chest tube management recommendations and have potential to enhance care by reliably emulating expert level clinical guidance. Additional prospective trials are required to further enhance safety and efficiency.
