No abstract

No abstract

BACKGROUND Coronavirus disease 2019 (COVID-19) infection is a risk factor for delirium that must be predicted and prevented to avoid adverse outcomes. OBJECTIVE We developed a machine learning (ML) model to predict delirium in hospitalized patients with COVID-19, and to identify modifiable factors to prevent delirium. METHODS The ML model was developed using training data from 757 patients at three medical centers and externally validated in 121 patients from a fourth medical center. The extreme gradient boosting (XGBoost) algorithm was used. A stratified K-fold approach was used to select model hyperparameters and predictor variables. The area under the curve (AUC) of the receiver operating characteristic (ROC) curve was selected as the evaluation metric. RESULTS The incidence of in-hospital delirium was 6.9% in the training cohort. Selected predictor variables for delirium were age, mechanical ventilation, medication (opioids, sedatives, antipsychotics, ambroxol, ceftriaxone, and piperacillin/tazobactam), sodium ion concentration, and white blood cell count (all p < 0.05). The stratified 5-fold AUC values for the training and test cohorts were 0.856 (95% confidence interval [CI] = 0.804–0.908) and 0.998 (CI = 0.989–1.000), respectively. CONCLUSIONS We developed and externally validated the ML model to predict delirium in COVID-19 inpatients. The model identified modifiable factors associated with the development of delirium and could be clinically useful for the prediction and prevention of delirium in COVID-19 inpatients.

BackgroundDelirium is a neuropsychiatric condition strongly associated with poor clinical outcomes such as high mortality and long hospitalization. In the patients with Coronavirus disease 2019 (COVID-19), delirium is common and it is considered as one of the risk factors for mortality. For those admitted to negative-pressure isolation units, a reliable, validated and contact-free delirium screening tool is required.Materials and methodsWe prospectively recruited eligible patients from multiple medical centers in South Korea. Delirium was evaluated using the Confusion Assessment Method (CAM) and 4‘A’s Test (4AT). The attentional component of the 4AT was modified such that respondents are required to count days, rather than months, backward in Korean. Blinded medical staff evaluated all patients and determined whether their symptoms met the delirium criteria of the Diagnostic and Statistical Manual of Mental Disorders 5 (DSM-5). An independent population of COVID-19 patients was used to validate the 4AT as a remote delirium screening tool. We calculated the area under the receiver operating characteristic curve (AUC).ResultsOut of 286 general inpatients, 28 (9.8%) inpatients had delirium. In this population, the patients with delirium were significantly older (p = 0.018) than the patients without delirium, and higher proportion of males were included in the delirium group (p &lt; 0.001). The AUC of the 4AT was 0.992 [95% confidence interval (CI) 0.983–1.000] and the optimal cutoff was at 3. Of the independent COVID-19 patients, 13 of 108 (12.0%) had delirium. Demographically, the COVID-19 patients who had delirium only differed in employment status (p = 0.047) from the COVID-19 patients who did not have delirium. The AUC for remote screening using the 4AT was 0.996 (0.989–1.000). The optimal cutoff of this population was also at 3.ConclusionThe modified K-4AT had acceptable reliability and validity when used to screen inpatients for delirium. More importantly, the 4AT efficiently screened for delirium during remote evaluations of COVID-19 patients, and the optimal cutoff was 3. The protocol presented herein can be used for remote screening of delirium using the 4AT.