Validation and Extension of the Prolonged Mechanical Ventilation Prognostic Model (ProVent) Score for Predicting 1-Year Mortality after Prolonged Mechanical Ventilation.

Abstract

Prognostic models can inform management decisions for patients requiring prolonged mechanical ventilation. The Prolonged Mechanical Ventilation Prognostic model (ProVent) score was developed to predict 1-year mortality in these patients. External evaluation of such models is needed before they are adopted for routine use. The goal was to perform an independent external validation of the modified ProVent score and assess for spectrum extension at 14 days of mechanical ventilation. This was a retrospective cohort analysis of patients who received prolonged mechanical ventilation at the University of Iowa Hospitals. Patients who received 14 or more days of mechanical ventilation were identified from a database. Manual review of their medical records was performed to abstract relevant data including the four model variables at Days 14 and 21 of mechanical ventilation. Vital status at 1 year was checked in the medical records or the social security death index. Logistic regressions examined the associations between the different variables and mortality. Model performance at 14 to 20 days and 21+ days was assessed for discrimination by calculating the area under the receiver operating characteristic curve, and calibration was assessed using the Hosmer-Lemeshow goodness-of-fit test. A total of 180 patients (21+ d) and 218 patients (14-20 d) were included. Overall, 75% were surgical patients. One-year mortality was 51% for 21+ days and 32% for 14 to 20 days of mechanical ventilation. Age greater than 65 years was the strongest predictor of mortality at 1 year in all cohorts. There was no significant difference between predicted and observed mortality rates for patients stratified by ProVent score. There was near-perfect specificity for mortality in the groups with higher ProVent scores. Areas under the curve were 0.69 and 0.75 for the 21+ days and the 14 to 20 days cohorts respectively. P values for the Hosmer-Lemeshow statistics were 0.24 for 21+ days and 0.22 for 14 to 20 days. The modified ProVent model was accurate in our cohort. This supports its geographic and temporal generalizability. It can also accurately identify patients at risk of 1-year mortality at Day 14 of mechanical ventilation, but additional confirmation is required. Further studies should explore the implications of adopting the model into routine use.