Identification of an optimal threshold to define oliguria in critically ill patients: an observational study

The relevance of current consensus threshold to define oliguria has been challenged by small observational studies. We aimed to determine the optimal threshold to define oliguria in critically-ill patients. Cohort study including adult patients admitted within a multi-disciplinary intensive care unit between January 1st 2010 and June 15th 2020. Patients on chronic dialysis or who declined consent were excluded. We extracted hourly urinary output (UO) measurements along with patient's characteristics from electronic medical records and 90-day mortality from the Swiss national death registry. We randomly split our data into a training (80%) and a validation (20%) set. In the training set, we developed multivariable models to assess the relationship between 90-day mortality and the minimum average UO calculated over time windows of 3, 6, 12 and 24 h. Optimal thresholds were determined by visually identifying cut-off values for the minimum average UO below which predicted mortality increased substantially. We tested models' discrimination and calibration on the entire validation set as well as on a subset of patients with oliguria according to proposed thresholds. Among the 15,500 patients included in this analysis (training set: 12,440, validation set: 3110), 73.0% (95% CI [72.3-73.8]) presented an episode of oliguria as defined by consensus criteria (UO 85% for all time windows) discrimination and calibration. The relationship between minimum average UO and predicted 90-day mortality was nonlinear with an inflexion point at 0.2 ml/kg/h for 3 and 6 h windows and 0.3 ml/kg/h for 12 and 24 h windows. Considering a threshold of