[Preliminary establishment of weaning prediction model].

Objective: To establish a model that can predict weaning failure from ventilation through hemodynamic and fluid balance parameters.

Methods: A retrospective analysis was conducted. The patients who underwent invasive mechanical ventilation for more than 24 hours and having spontaneous breathing test admitted to intensive care unit (ICU) of Tianjin Third Central Hospital from January 1st, 2017 to December 31st, 2018 were enrolled. The information was collected, which included the baseline data, hemodynamic parameters by pulse indicator continuous cardiac output (PiCCO) monitoring, B-type natriuretic peptide (BNP), urinary output, fluid balance in first 24 hours when patients admitted to ICU, and hemodynamic parameters by PiCCO monitoring, BNP, urinary output, fluid balance, diuretic usage, noradrenalin usage within 24 hours before weaning as well as usage of continuous renal replacement therapy (CRRT) during mechanical ventilation. According to weaning success or failure, the patients were divided into weaning success group and weaning failure group, and the statistical differences between the two groups were calculated. Variables with statistical significance within 24 hours before weaning were included in the multivariate Logistic regression analysis to establish weaning failure prediction model and find out the possible risk factors of weaning failure.

Results: A total of 159 patients were included in this study, which included 138 patients in the weaning success group and 21 patients in the weaning failure group. There were no statistical differences in all hemodynamic parameters by PiCCO monitoring, BNP, urinary output, fluid balance within 24 hours into ICU between two groups. There were statistical differences in BNP (χ = 9.262, P = 0.026), central venous pressure (CVP; χ = 7.948, P = 0.047), maximum rate of the increase in pressure (dPmx; χ = 10.486, P = 0.015), urinary output (χ = 8.921, P = 0.030), fluid balance (χ = 9.172, P = 0.027) within 24 hours before weaning between two groups. In addition, variable about cardiac index (CI; χ = 7.789, P = 0.051) was included into multivariate Logistic regression model to improve the prediction model and enhance the accuracy of model. Finally, variables included in the multivariate Logistic regression model were BNP, CVP, CI, dPmx, urinary output, fluid balance volume, and the accuracy of the weaning failure prediction model was 92.9%, the sensitivity was 100%, and the specificity was 76.8%. When the model was adjusted by variables of age and noradrenalin usage, the accuracy of model to predict failure of weaning was 94.2%, the sensitivity was 100%, the specificity was 81.2%.

Conclusions: Weaning failure prediction model based on hemodynamic parameters by PiCCO monitoring and variables about liquid balance has high accuracy and can guide clinical weaning.