PURPOSE: Although new-onset atrial fibrillation (AF) is frequently observed in the intensive care unit (ICU), the incidence and predictors for sustained new-onset AF have not been investigated, except for cardiac surgery patients. We have evaluated potential predictors for sustained new-onset AF in a mixed ICU. METHODS: In this retrospective observational study, we screened non-cardiac surgery patients who were admitted to the ICU between January 2010 and December 2013 and had been hospitalized for > 24 h in the ICU. We collected information about heart rhythm 6 h after the onset of AF. We compared detailed patient characteristics between patients with sinus rhythm (SR) and those with sustained AF at 6 h after the onset of AF. Additionally, we applied variable selection using backward elimination based on Akaike's Information Criterion (AIC). Calibration was performed based on the Hosmer-Lemeshow test. RESULTS: New-onset AF occurred in 151 of 1718 patients and 99 patients converted to SR at 6 h. Backward elimination identified predictors as follows (AIC = 175.3): CHADS2 score, elective surgery, infection on ICU admission, serum potassium > 4.0 mmol/L, male sex, mechanical ventilation, and diagnostic grouping. The model showed good calibration for sustenance of AF at 6 h after the onset using the Hosmer-Lemeshow Chi-square value of 4.36 (degrees of freedom = 4, p = 0.360) indicating a good fit. CONCLUSIONS: These predictors might be useful in future interventional studies to identify patients who are likely to sustain new-onset AF.