Evaluating The Impact Of Ventilator Associated Parameters On Ventilation-Free Days And In-hospital Mortality Of Non-ARDS Patients In ICU

Background:It is normal for critically ill patients without ARDS to be mechanically ventilated,what kind of ventilation parameters is the optimal ventilation strategy for non-ARDS ICU patients?Methods:A retrospective cohort study involved MIMIC-III database for non-ARDS patients who underwent invasive ventilation for at least 48 consecutive hours from 2001to 2012,performed univariate,multivariate regression analysis,covariate balancing propensity score?CBPS?and inverse–probability–of–treatment weighting?IPTW?,and different machine learning?ML?models to predict various outcomes.The included predicted factors were four parameters of mechanical ventilation?Driving pressure?DP?,tidal volumes?VT?,positive end-expiratory pressure?PEEP?and mechanical power?MP??,and age,OASIS,ventilated day 2 of p H,Sp O₂,MBP,Pa CO₂,temperature.The primary outcome was the number of ventilator-free days?VFDs?at day 28,while the secondary outcomes included in-hospital mortality,ICU,30-day,and 1-year mortality,and ICU and hospital length of stay?LOS?.By establishing various ML models,we ranked the importance of relevant factors,and evaluated the prediction of models on different outcome variables.Results:The study included 2932 patients,low DP?OR-0.62?95%CI-0.70,-0.54?,P-Value<0.0001?,low PEEP?OR 0.65?95%CI0.58,0.72?,P-Value<0.0001?and low MP?OR-0.11?95%CI-0.14,-0.09?,P-Value<0.0001?for non-ARDS patients were related with prolonged VFDs at day 28,and reduced ratio of in-hospital mortality?OR 1.02?95%CI1.01,1.03?,P-Value<0.0001?for DP,?OR 0.81?95%CI 0.80,0.85?,P-Value<0.0001?for PEEP,?OR 0.01?95%CI0.002,0.02?P-Value<0.01?for MP and length of hospital stay,improved the prognosis of patients.However,VT?OR 0?95%CI-0.11,0.09?,P-Value=0.88?had no prognostic significance for the patients.The population was divided into a training set and a test set according to the proportion of 90/10.The Gini coefficient,coefficient of variance,and LASSO regularization were used to identify clinical features top relevant to the prognosis of patients,and different ML models were established to predict various clinical outcomes.Among ML models with VFDs at day 28,the randomforest had the lowest RMSE prediction,which was 43.68,and the model with the gradient boost works best for in-hospital mortality,with the area under the curve?AUC?was 0.93,error rate was 0.09,f score was 0.87.Conclusions:For non-ARDS patients who received MV for at least 48 consecutive hours,low DP,low PEEP and low MP were beneficial to the patients.However,the effect of VT was inconclusive.The prediction models of ML were good.