Artificial Intelligence in Upper Gastrointestinal Endoscopy.

BACKGROUND: Over the past decade, several artificial intelligence (AI) systems are developed to assist in endoscopic assessment of (pre-)cancerous lesions of the gastrointestinal (GI) tract. In this review, we aimed to provide an overview of the possible indications of AI technology in upper GI endoscopy and hypothesize about potential challenges for its use in clinical practice. SUMMARY: Application of AI in upper GI endoscopy has been investigated for several indications: (1) detection, characterization, and delineation of esophageal and gastric cancer (GC) and their premalignant conditions; (2) prediction of tumor invasion; and (3) detection of Helicobacter pylori. AI systems show promising results with an accuracy of up to 99% for the detection of superficial and advanced upper GI cancers. AI outperformed trainee and experienced endoscopists for the detection of esophageal lesions and atrophic gastritis. For GC, AI outperformed mid-level and trainee endoscopists but not expert endoscopists. KEY MESSAGES: Application of artificial intelligence (AI) in upper gastrointestinal endoscopy may improve early diagnosis of esophageal and gastric cancer and may enable endoscopists to better identify patients eligible for endoscopic resection. The benefit of AI on the quality of upper endoscopy still needs to be demonstrated, while prospective trials are needed to confirm accuracy and feasibility during real-time daily endoscopy.

Training for Medical Oncologists on Shared Decision-Making About Palliative Chemotherapy: A Randomized Controlled Trial.

BACKGROUND: Systemic treatment for advanced cancer offers uncertain and sometimes limited benefit, while the burden can be high. This study examines the effect of shared decision-making (SDM) training for medical oncologists on observed SDM in standardized patient assessments. MATERIALS AND METHODS: A randomized controlled trial comparing training with standard practice was conducted. Medical oncologists and oncologists-in-training (n = 31) participated in a video-recorded, standardized patient assessment at baseline (T0) and after 4 months (T1, after training). The training was based on a four-stage SDM model and consisted of a reader, two group sessions (3.5 hours each), a booster session (1.5 hours), and a consultation card. The primary outcome was observed SDM as assessed with the Observing Patient Involvement scale (OPTION12) coded by observers blinded for arm. Secondary outcomes were observed SDM per stage, communication skills, and oncologists' satisfaction with communication. RESULTS: The training had a significant and large effect on observed SDM in the simulated consultations (Cohen's f = 0.62) and improved observed SDM behavior in all four SDM stages (f = 0.39-0.72). The training improved oncologists' information provision skills (f = 0.77), skills related to anticipating/responding to emotions (f = 0.42), and their satisfaction with the consultation (f = 0.53). CONCLUSION: Training medical oncologists in SDM about palliative systemic treatment improves their performance in simulated consultations. The next step is to examine the effect of such training on SDM in clinical practice and on patient outcomes. IMPLICATIONS FOR PRACTICE: Systemic treatment for advanced cancer offers uncertain and sometimes limited benefit, while the burden can be high. Hence, applying the premises of shared decision-making (SDM) is recommended. SDM is increasingly advocated based on the ethical imperative to provide patient-centered care and the increasing evidence for beneficial patient outcomes. Few studies examined the effectiveness of SDM training in robust designs. This randomized controlled trial demonstrated that SDM training (10 hours) improves oncologists' performance in consultations with standardized patients. The next step is to examine the effect of training on oncologists' performance and patient outcomes in clinical practice.