AI Assistance in GI Endoscopy Recovery Assessment

Overview

We have developed and validated an AI model to assess endoscopy recovery status based on 400 voice recordings from 200 patients. This model has a mean accuracy of 84.14% with a mean area under the curve (AUC) of 0.91. To further enhance the performance of this AI model, we plan to collect additional voice recordings to retrain it. We also plan to develop a mobile application of this AI model for effectiveness evaluation in a pilot randomized controlled trial (RCT) setting. Endoscopy nurses in Hong Kong were invited to participate in a survey study. Therefore, we believe implementation of AI model in clinical practice will be well accepted by endoscopy nurses in Hong Kong.

Globally, cancer is one of the leading causes of death. 70% of this global burden of cancer is attributed to premature mortality, of which 70% of these deaths are preventable and 30% are treatable Globally, cancer is one of the leading causes of death. 70% of this global burden of cancer is attributed to premature mortality, of which 70% of these deaths are preventable and 30% are treatable . Among the ten most common causes of cancer death worldwide, one third are digestive cancers. These can be diagnosed by outpatient gastrointestinal endoscopy. These include colorectal cancer (via colonoscopy), as well as esophageal and gastric cancer (via esophagogastroduodenoscopy). In addition to diagnosis, outpatient endoscopy is also widely used for cancer screening and surveillance. With the global increase in the aging population and heightened awareness of cancer screening, the demand for endoscopy services for cancer screening, diagnosis and surveillance is rapidly increasing to achieve the goal of early cancer detection and treatment. To reduce patients' fear and anxiety regarding endoscopy and to relieve the associated pain and discomfort, most endoscopies are performed under sedation. It is known that the sedative effect lasts longer than needed for diagnostic or basic therapeutic colonoscopies, which can mostly be done within 30 minutes. Patients are closely monitored at the recovery room after the completion of the endoscopy, and the recovery nurse assesses the consciousness of the patient after a fixed period of time, typically 60 minutes, or every 10 minutes until the standardized discharge criteria are met. In an outpatient setting, it is important to determine if patients are fully recovered from the sedative effect and have reached a clinically stable state before discharging them from the hospital with the accompany of a responsible adult. The assessment of the standardized discharge criteria includes the followings: 1) return of consciousness to baseline level; 2) vital signs are within normal limits; 3) respiratory status is not compromised; and 4) pain and discomfort have been addressed. A standardized discharge assessment scoring, such as, the modified Aldrete's score, and the modified post-anaesthesia discharge scoring system (mPADSS), were recommended. The mean recovery time required by both systems was reported to be 60 minutes, which is quite time-consuming. International guidelines on sedation in gastrointestinal endoscopy recommend a 1:1 nursing ratio to closely monitor patients following moderate or deep sedation to enhance patient safety. With this 1:1 ratio, recovery nurses can assess the consciousness level of patient every 10 minutes by standardized discharge assessment scoring, which facilitates a shorter recovery time. However, assessment every 10 minutes is time-consuming and labour-intensive and such recovery nursing ratio may not be practicable in resource-limited countries. In Hong Kong, the usual recovery nursing ratio is 1:10, therefore, the current standard practice is to assess patient's consciousness after 60 minutes. As a result, the number of endoscopies arranged in each session is limited by the recovery time (i.e. patient turnover rate), the recovery space and nursing manpower. Moreover, the decision of the recovery nurse on whether a patient is dischargeable can be interfered by a series of contextual factors, such as heavy workload, the availability of recovery space and the demand of patient. A fast, convenient, and reliable assessment system is warranted to reduce the recovery time (i.e. to increase the turnover rate) because of the anticipated increasing demand of sedated endoscopy which leads to the requirement for space and nursing manpower for patient recovery. To our best knowledge, no interventional trial has been conducted to reduce the recovery time by AI technology without increasing the nursing manpower. In the past decade, artificial intelligence (AI) technology has emerged and been successfully implemented in various clinical settings, particularly in the field of gastrointestinal endoscopy. AI models trained from endoscopic images have been proven to be effective in detecting and diagnosing gastrointestinal diseases and cancers. Human voice can be transferred to image and used to train AI models to assist in disease diagnosis. For example, AI has been trained to effectively detect Alzheimer's disease and predict its severity solely based on patients' voice data. Another AI model has been developed based on voice analysis to distinguish major psychiatric disorders, including bipolar, depressive, anxiety and schizophrenia spectrum disorders. Given these promising results, we have developed and validated an AI model to assess endoscopy recovery status based on 400 voice recordings from 200 patients. This model has a mean accuracy of 84.14% with a mean area under the curve (AUC) of 0.91. To further enhance the performance of this AI model, we plan to collect additional voice recordings to retrain it. We also plan to develop a mobile application of this AI model for effectiveness evaluation in a pilot randomized controlled trial (RCT) setting. Endoscopy nurses in Hong Kong were invited to participate in a survey study. Therefore, we believe implementation of AI model in clinical practice will be well accepted by endoscopy nurses in Hong Kong.1). Among the ten most common causes of cancer death worldwide, one third are digestive cancers. These can be diagnosed by outpatient gastrointestinal endoscopy. These include colorectal cancer (via colonoscopy), as well as esophageal and gastric cancer (via esophagogastroduodenoscopy). In addition to diagnosis, outpatient endoscopy is also widely used for cancer screening and surveillance. With the global increase in the aging population and heightened awareness of cancer screening, the demand for endoscopy services for cancer screening, diagnosis and surveillance is rapidly increasing to achieve the goal of early cancer detection and treatment. To reduce patients' fear and anxiety regarding endoscopy and to relieve the associated pain and discomfort, most endoscopies are performed under sedation. It is known that the sedative effect lasts longer than needed for diagnostic or basic therapeutic colonoscopies, which can mostly be done within 30 minutes. Patients are closely monitored at the recovery room after the completion of the endoscopy, and the recovery nurse assesses the consciousness of the patient after a fixed period of time, typically 60 minutes, or every 10 minutes until the standardized discharge criteria are met. In an outpatient setting, it is important to determine if patients are fully recovered from the sedative effect and have reached a clinically stable state before discharging them from the hospital with the accompany of a responsible adult. The assessment of the standardized discharge criteria includes the followings: 1) return of consciousness to baseline level; 2) vital signs are within normal limits; 3) respiratory status is not compromised; and 4) pain and discomfort have been addressed (10). A standardized discharge assessment scoring, such as, the modified Aldrete's score, and the modified post-anaesthesia discharge scoring system (mPADSS), were recommended. The mean recovery time required by both systems was reported to be 60 minutes, which is quite time-consuming. International guidelines on sedation in gastrointestinal endoscopy recommend a 1:1 nursing ratio to closely monitor patients following moderate or deep sedation to enhance patient safety. With this 1:1 ratio, recovery nurses can assess the consciousness level of patient every 10 minutes by standardized discharge assessment scoring, which facilitates a shorter recovery time. However, assessment every 10 minutes is time-consuming and labour-intensive and such recovery nursing ratio may not be practicable in resource-limited countries. In Hong Kong, the usual recovery nursing ratio is 1:10, therefore, the current standard practice is to assess patient's consciousness after 60 minutes. As a result, the number of endoscopies arranged in each session is limited by the recovery time (i.e. patient turnover rate), the recovery space and nursing manpower. Moreover, the decision of the recovery nurse on whether a patient is dischargeable can be interfered by a series of contextual factors, such as heavy workload, the availability of recovery space and the demand of patient. A fast, convenient, and reliable assessment system is warranted to reduce the recovery time (i.e. to increase the turnover rate) because of the anticipated increasing demand of sedated endoscopy which leads to the requirement for space and nursing manpower for patient recovery. To our best knowledge, no interventional trial has been conducted to reduce the recovery time by AI technology without increasing the nursing manpower. In the past decade, artificial intelligence (AI) technology has emerged and been successfully implemented in various clinical settings, particularly in the field of gastrointestinal endoscopy. AI models trained from endoscopic images have been proven to be effective in detecting and diagnosing gastrointestinal diseases and cancers. Human voice can be transferred to image and used to train AI models to assist in disease diagnosis. For example, AI has been trained to effectively detect Alzheimer's disease and predict its severity solely based on patients' voice data. Another AI model has been developed based on voice analysis to distinguish major psychiatric disorders, including bipolar, depressive, anxiety and schizophrenia spectrum disorders.