Skip to main content
Fig. 5 | Annals of Intensive Care

Fig. 5

From: Prevalence and risk factors of hemodynamic instability associated with preload-dependence during continuous renal replacement therapy in a prospective observational cohort of critically ill patients

Fig. 5

Forest plot of risk factors for occurrence of HIRRT associated with preload-dependence in multivariate analysis. Bars are 95% confidence interval of odds ratios. CITD cardiac index assessed by thermodilution, HIRRT hemodynamic instability related to renal replacement therapy, ICU intensive care unit, OR odds ratio. *This cut-off value was chosen as it maximized the Youden’s index in univariate analysis. The following variables were entered into the multivariate full model: preload-dependence on the preceding measurement before HIRRT (yes/no), delay since last HIRRT episode > 8 h (yes/no), delay since CRRT onset, cardiac index assessed by thermodilution on the preceding measurement before HIRRT, global end-diastolic volume on the preceding measurement before HIRRT, pulmonary vascular permeability index on the preceding measurement before HIRRT, pulse pressure variation on the preceding measurement before HIRRT, systolic arterial pressure on the preceding measurement before HIRRT, mechanical ventilation status (yes/no) on the preceding measurement before HIRRT, sex male (yes/no), SOFA score at ICU admission, body weight on the day of hemodynamic measurement, SOFA score on the day of hemodynamic measurement; lactate on the day of hemodynamic measurement, base excess on the day of hemodynamic measurement, hemoglobin on the day of hemodynamic measurement, sepsis criteria on the day of hemodynamic measurement (yes/no), septic shock criteria on the day of hemodynamic measurement (yes/no). The following variables were not entered into the multivariate full model because of multicollinearity: cardiac index assessed by pulse contour analysis and global ejection fraction on the preceding measurement before HIRRT (collinearity with cardiac index assessed by thermodilution), stroke volume variation on the preceding measurement before HIRRT (collinearity with pulse pressure variation), mean and diastolic arterial pressure on the preceding measurement before HIRRT (collinearity with systolic arterial pressure), bicarbonate on the day of hemodynamic measurement (collinearity with base excess). No significant interaction was identified between any of the selected variables. C-statistic of the final model: 0.77. Model calibration assessed by the Hosmer–Lemeshow test: p = 0.76

Back to article page