Validation of a Risk Assessment Model for Postoperative Delirium Based on Artificial Intelligence
Validation of a Risk Assessment Model for Postoperative Delirium Based on Artificial Intelligence
Postoperative delirium (POD) is a frequent postoperative complication in the elderly, characterised by fluctuating disturbances in attention, awareness, and cognition. Identifying the patients at highest risk of developing POD was the aim of the artificial intelligence (AI)-based algorithm PIPRA. This prospective cohort study is to externally validate the AI-based PIPRA algorithm. The primary endpoint is the performance (AUC) of the PIPRA algorithm in predicting POD. The secondary endpoint is the performance (AUC) of the clinicians in predicting POD (and how it compares with the performance of the PIPRA algorithm).
Perioperative neurocognitive disorders (PND) include postoperative delirium (POD) and postoperative neurocognitive disorder or postoperative cognitive dysfunction (POCD). POD is recognised as a frequent postoperative complication in the elderly, occurring in 10% to 50% of older patients after major surgical procedures. POD usually occurs in the early postoperative period and is defined as an acute neuropsychiatric disorder. It is characterised by fluctuating disturbances in attention, awareness, and cognition. The American Society of Enhanced Recovery and Perioperative Quality Initiative Joint Consensus Statement on Postoperative Delirium Prevention recommend focusing on identifying those patients at highest risk of developing POD. Identifying these highest risk patients was the aim of the artificial intelligence (AI)-based algorithm PIPRA, which was created based on an individual participant data (IPD) meta-analysis including more than 2500 patients. This risk-prediction algorithm uses standard data (i.e. age, height, weight, history of delirium, cognitive impairment, ASA status, number of medications, preoperative C reactive protein (CRP), surgical risk and laparotomy), which are routinely collected before surgery. PIPRA was internally validated with an area under the curve (AUC) of 0.837 with 95% confidence interval 0.808 to 0.865, when plotting the true positive rate against the false positive rate. The aim of this prospective cohort study is to externally validate the AI-based PIPRA algorithm.
First, the anaesthesiologist in charge will be asked to evaluate, based on his/her experience (quantified in years of anaesthesia practice), the risk for the included patient to develop POD (categorised as low, intermediate, high or very high). Next, an investigator will assess included patents in a systematic and reproductible manner. After surgery, an investigator will visit the patient twice daily from postoperative day 1 to 5 or until hospital discharge (whichever occurs first) to screen for delirium using the 4AT or the ICDSC. The PIPRA score will be calculated separately by the coordinating study centre.
Inclusion Criteria:
Exclusion Criteria: