Invasive Mechanical Ventilation Duration Prediction using Survival Analysis

Abstract Invasive mechanical ventilation is one of the leading life support machines in the intensive care unit (ICU). By identifying the predictors of ventilation time upon arrival, important information can be gathered to improve decisions regarding capacity planning. In this study, first-day ventilated patients’ ventilation time was analyzed using survival analysis. The probabilistic behaviour of ventilation time duration was analyzed and the predictors of ventilation time duration were determined based on available first-day covariates. A retrospective analysis of ICU ventilation time in Ontario was performed with data from ICU patients obtained from the Critical Care Information System (CCIS) in Ontario between July 2015 and December 2016. As part of the protocol for inclusion, a patient must have been connected to an invasive ventilator upon arrival to the ICU. Parametric survival methods were used to characterize ventilation time and to determine associated covariates. Parametric and non-parametric methods were used to determine predictors of ventilation duration for first-day ventilated patients. A total of 128,030 patients visited the ICUs between July 2015 and December 2016. 51,966 (40.59%) patients received invasive mechanical ventilation on arrival. Analysis of ventilation duration suggested that the log-normal distribution provided the best fit to ventilation time, whereas the log-logistic Accelerated Failure Time model best describes the association between the covariates and ventilation duration. ICU site, admission source, admission diagnosis, scheduled admission, scheduled surgery, referring physician, central venous line treatment, arterial line treatment, intracranial pressure monitor treatment, extra-corporeal membrane oxygen treatment, intraaortic balloon pump treatment, other interventions, age group, pre-ICU LOS, and MODS score were significant predictors of the ICU ventilation time. The results show substantial variability in ICU ventilation duration for different ICUs, patient’s demographics, and underlying conditions, and highlight mechanical ventilation as an important driver of ICU costs. The predictive performance of the proposed model showed that both the model and the data can be used to predict an individual patient’s ventilation time and to provide insight into predictors.