Abstract: Objective To explore the risk factors of subsyndromal delirium (SSD) in patients after cardiac surgery, to construct a prediction model for the occurrence of SSD in this group, and verify its performance. Methods A total of 549 surgical patients in the cardiac surgery department of a tertiary Grade-A hospital in Hunan Province from August 2023 to July 2024 were selected. They were divided into a non-delirium group and a delirium group based on the occurrence of SSD during their ICU stay. The patients were divided into a training set and a test set at a ratio of 7:3. A prediction model was developed using the random forest algorithm on the training set, validated in the testing set, and subjected to variable importance ranking. Results Among the 549 patients, 63 were diagnosed with SSD. There were no statistically significant differences between the non-delirium and delirium group in terms of sex, height, place of residence, and mode of admission (P>0.05). The variables selected by Lasso regression were incorporated into the model constructed by random forest algorithm, demonstrating that the predictive efficacy of the model in the test set was 0.959. The variables with importance ranking from high to low were the rate of adrenaline decline, the rate of noradrenaline decline, first-day urine output, mean corpuscular hemoglobin concentration (MCHC), body weight, age, post-MVR (mitral valve replacement), post-CABG (coronary artery bypass grafting), and post-AVR (aortic valve replacement). Conclusion The influencing factors for the occurrence of SSD in patients after cardiac surgery, with importance ranking from high to low are the rate of adrenaline decline, the rate of noradrenaline decline, first-day urine output, MCHC, body weight, age, post-MVR, post-CABG, and post-AVR. The model constructed by random forest algorithm has good predictive efficacy.

Key words: random forest algorithm, postoperative cardiac surgery, subsyndromal delirium, prediction model