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

For this observational study, we have used a high-resolution cohort described in details elsewhere [2]. Briefly, we collected data from all adult (> 18 years old) patients admitted to our tertiary ICU between January 1st 2010 and June 15th 2020. We excluded patients with documented or expressed wishes of non-participation to clinical research, those with end-stage renal disease (ESRD), whose stay lasted for less than 6 h or with missing outcome or AKI defining data. For each patient, we considered only the first eligible ICU admission. Data were extracted from electronic medical records [Metavision®(IMD Soft, Tel Aviv, Israel) and Soarian® (Cerner, North Kansas City, USA)]. In particular, we collected baseline characteristics, comorbidities, pre-admission weight, illness severity scores and hourly UO measurements. The primary outcome was 90-day mortality. This outcome was assessed by cross-referencing our dataset with the Swiss national death registry.

Continuous data are reported as mean (standard deviation, SD) or median (interquartile range, IQR) according to underlying data distribution. Categorical variables are expressed as number (percentage). All statistical analyses and modeling were carried out in R version 4.1.2 [13]. The level of statistical significance was set at 5%.

This study was approved by the Ethics Committee Vaud (CER-VD 2017-00008, Lausanne, Switzerland). In accordance with the Swiss Federal Act on Research involving Human Beings (article 34) [19], retrospective utilization of non-genetic health-related personal data was permitted, provided that the patient (or its legal representative) had not expressed wishes of non-participating to clinical research. This study followed “The Strengthening the Reporting of Observational Studies in Epidemiology” (STROBE) guidelines for reporting observational studies.

Among the 18,314 patients admitted to our ICU during the study period. 2584 were excluded: 1116 declined consent for data utilization for research, 84 were under 18 years old, 747 had an ICU stay of less than 6 h, 217 had ESRD, 45 had no administrative data available, 305 had incomplete UO or sCr data and 70 had unknown 90-day mortality status. Therefore, our main analysis included 15,550 patients (Table 1). Patients were predominantly male (10,283, 66.1%) with a median (IQR) age of 65 (53.0–75.0) years, a median body weight of 75.0 (65.0–87.0) kg, a median Charlson score of 4 (2.0–6.0) and a median corrected SAPS II score of 37 (28–48). Among those, 73.0% (95% CI [72.3–73.8]) presented a minimal average UO below 0.5 ml/kg/h for 6 h and therefore fulfilled criteria for oliguria according to consensus definition. Clinical outcomes and missing values are reported in the Additional file 1 (Additional file 1: Tables S1 and S2). Pre-admission body weight was not available and had to be imputed in 3188 (20.5%) patients (Cf. multiple imputations above).

The training and validation sets included data from 12,440 and 3110 patients, respectively (Table 1).

Figure 1 illustrates the dependence between the minimum average UO (computed over time windows of 3, 6, 12 or 24 h) and 90-day predicted mortality in the different categories of admissions. Predicted mortality is estimated for median values of age, corrected SAPS II score and Charlson index. This figure suggests that adjusted 90-day mortality increases substantially in all categories of admission when the minimum average urine output is below 0.2 ml/kg/h over 3 or 6 h, or below 0.3 ml/kg/h over 12 or 24 h. Unadjusted data are presented in Additional file 1: Figure S1.

Similar associations and thresholds between minimum average UO and mortality were observed when using the same model to predict 30-day mortality (Additional file 1: Figure S2). Similarly, in sensitivity analyses, extreme patients’ weights (< 43 kg or > 130 kg; Additional file 1: Figure S3), gender (Additional file 1: Figures S4 and S5), SAPS II scores (Additional file 1: Figures S6 and S7) and time periods (Additional file 1: Figures S8, S9 and S10) did not appear to influence these results.

Based on these findings, we propose to consider a minimum average UO of < 0.2 ml/kg/h for 6 h as the new threshold to define oliguria. In our cohort, 24.7% [95% CI 24.0–25.4] of the patients fulfilled these revised criteria for oliguria (vs 73.0% [95% CI 72.3–73.8]) for consensus criteria). Unlike consensus criteria, this stricter definition is significantly associated with a higher 90-day adjusted mortality (OR 1.98 [95% CI 1.57–2.49] vs 1.27 [95% CI 0.95–1.70]). As shown in Fig. 2, patients with a minimum average UO of < 0.2 ml/kg/h for 6 h had a lower 3-month survival compared to other groups of patients throughout categories of illness severity and age, and this persisted at 12 months, (Additional file 1: Figure S11).

Our prognostic models achieved good discrimination on validation data, with AU-ROC exceeding 85% for all time windows (Table 2). However, prior univariate analyses identified the corrected SAPS II score as the strongest predictor of 90-day mortality, with AU-ROC already reaching 82%. As a consequence, only minor improvements in discrimination were observed when accounting for the minimal average urine output in prognostic models over the base model (Table 2).

When considering all validation data, calibration performance was very good (Additional file 1: Figure S12) both for the base model and the model including minimal average UO as a predictor of mortality. On the other hand, calibration belts presented in Fig. 3 suggest that, when restricting the attention to patients complying with the proposed oliguria definitions, the calibration of predicted mortalities improved substantially when accounting for the minimal average urine output. This was further supported by statistically significant (p < 0.001) results returned by Hosmer–Lemeshow tests on validation data for all models that omitted urine output from the list of predictors of mortality, suggesting that predicted probabilities of death derived from such models were miscalibrated in patients with oliguria. On the other hand, models which included UO as a predictor of mortality showed no evidence of miscalibration in these patients, with p values > 0.05 for 3, 6, 12 and 24 h time windows.

We conducted a large observational study including more than 15,000 patients admitted to a tertiary ICU during a 10-year period. We assessed the association between minimum average UO (calculated over time windows of 3, 6, 12 and 24 h) and 90-day mortality after controlling for age on admission, comorbidities and illness severity. When assessed on validation data, prognostic models including minimum average UO as a predictor of 90-day mortality resulted in good calibration and discrimination. We observed a nonlinear relationship between minimum average UO and 90-day mortality. Our results suggest that the widely used cut-off for oliguria of 0.5 ml/kg/h over 6 or 12 h may be too conservative. We found that values of 0.2 ml/kg/h (3 and 6 h time-windows) and 0.3 ml/kg/h (12 and 24 h time-windows) had higher prognostic implications. These values were similar for patients admitted for medical or surgical reasons.

Only a handful of studies have evaluated the relationship between oliguria and clinical outcomes. Two have considered the occurrence of AKI (defined as a sCr rise) as an outcome [10, 11]. In those studies, oliguria, defined as a UO < 0.5 ml/kg/h for 6 h, only had a weak association with AKI and very limited positive predictive value. Intrinsically, however, this design implies that only a rise in sCr could define true AKI while UO only represents either an associated finding or a biomarker. This is in contradiction with recent data which suggest that oliguria might be associated with mortality irrespective of sCr rise [2, 3]. Therefore, patients’ centered outcomes such as mortality or need for renal replacement therapy (RRT) appear more suitable to assess the relevance of oliguria.

Our data are in agreement with other smaller observational studies conducted in other health systems such as New Zealand or Finland [6, 12]. Ralib et al. have determined optimal UO thresholds for prediction of mortality or the need for RRT for time windows of 1 to 12 h. These cut-offs were derived from AUC-ROC curves at a univariate level. For the 3- 6- and 12-h time windows, identified cut-offs were, respectively, 0.2, 0.3 and 0.5 ml/kg/h. Our analyses, based on a much larger dataset and modeling UO as a continuous predictor with smooth functional forms while accounting for age, comorbidities and severity score, lead to slightly different values particularly at 12 h. However, we confirm the authors’ impression that current definition of oliguria is too liberal. In an analysis of the FINNAKI cohort, episodes of severe (< 0.1 ml/kg/h) oliguria lasting more than 3 consecutive hours were independently associated with the development of sCr-AKI or RRT. In this analysis, consistent with our data, oliguria defined as a UO between 0.3 and 0.5 ml/kg/h for 6 h, was not significantly associated with adjusted 90-day mortality (OR 1.65 [95% CI 1.0–2.72]). The shortest periods of consecutive oliguria associated with an increased risk for 90-day mortality on multivariate analyses were 6 h (for 0.1 to 0.3 ml/kg/h) and 3 h (for < 0.1 ml/kg/h).

In contrast, an analysis from the MIMIC-2 cohort suggested that mortality increased rapidly as UO decreased < 0.5 ml/kg/h [20]. These analyses used multivariate logistic regression analysis and contour plots. The difference in appreciation with our study is most likely explained by different modeling strategies for UO. Indeed, the authors have dichotomized UO whereas it was treated as a continuous predictor in the present study. Due to the nonlinear effect of UO on 90-day mortality (observed in our data), a dichotomous treatment of UO would likely lead to an overestimation of the optimal oliguria threshold. Indeed, a dichotomous threshold of 0.5 ml/kg/h to define oliguria would already lead to a difference in estimated mortality between oliguric and non-oliguric patients while this mortality signal is only driven by the fraction of patients with UO < 0.2 ml/kg/h. Hence, we argue that treating the minimal average UO as a continuous predictor is a preferable strategy to define meaningful oliguria thresholds.

This study is the largest (> 15,000 patients) attempting to assess the relationship between minimal average UO and mortality. The sample includes the vast majority of patients admitted to a large multidisciplinary ICU. Hence, we were able to assess the influence of the type of admission (medical vs surgical) on the described relationship. In addition, the large sample size enabled to separate our cohort into a derivation and validation sub-cohorts and test the discrimination and calibration of our models. We used all UO data points across the entire patient stay. Hence, our data is representative of the entire patients’ admission. The inclusion of illness severity and comorbidity scores as well as age in our models enabled to control for the effect of important confounders on the relationship between UO and 90-day mortality. Finally, in contrast to previous studies of comparable size [12, 20], we have flexibly modeled UO as a continuous predictor of 90-day mortality.

Our study, however, has limitations. As a single center study, the external validity of our findings may be challenged. However, although relying on different types of statistical modeling, our findings are in line with previous smaller or un-adjusted studies and can be viewed as confirmatory. We have elected to restrict our analyses to convenience time windows of 3, 6, 12 and 24 h. Perhaps different time windows would have provided different results. However, due to the similarity of data obtained across all reported time windows, this appears unlikely. In addition, proposed time windows are similar to those currently utilized (except for the 3 h window) and appear clinically relevant and applicable. We have not considered other outcomes of potential importance such as increased sCr or need for RRT. Indeed, previous analyses have demonstrated that severe (stage 2 or 3) oliguria was an independent risk factor for mortality irrespective of sCr [2]. We believe that the demonstration of an impact on 90-day mortality is the most meaningful way to define oliguria. In addition, KDIGO criteria and the sCr criteria in particular were validated through their relevance on mortality. As this is a real-life study, we have not used an automated urimeter which might have improved the accuracy of the data collection. However, the low rate of missing values for UO suggests a high level of validity in our data. Lastly, UO was normalized to patients' body weight, a parameter known to be highly inaccurate in critical illness. This was minimized by the primary consideration of pre-admission weight which was available in 79.5% of the patients and imputed for the remaining 20.5% of patients. In addition, sensitivity analyses suggested that extreme weights did not impact our results [21].

Our data provide an explanation for the perceived lack of relevance of oliguria as defined by a minimum average UO of less than 0.5 ml/kg/h for 6 h. Indeed, this definition appears to be too liberal and to include many patients for which oliguria has no prognostic implication. By considering UO as a continuous predictor of 90-day mortality, we propose to revise the current definition of oliguria and use 0.2 ml/kg/h for 6 h as a cut-off. Our data further suggests that such cut-off might be applied indifferently to patients with medical and surgical (emergent and elective) admissions despite the fact that these populations have very different baseline adjusted mortality rates.

Furthermore, we show that an increased mortality can already be identified when averaging UO over 3 h while using the same cut-off for oliguria (< 0.2 ml/kg/h). Using a time window of three hours to monitor urine output may thus allow early identification of patients at risk of excess mortality.