By DoM Communications | Date published: May 15, 2026

Shihla Shireen Kanamgode Studies Validity of ICU Prediction Models in Recent Publication 

In a recent study published in the Journal of Intensive Care Medicine, Shihla Shireen Kanamgode, MD, a Pulmonary and Critical Care Medicine fellow, explored the validity of ICU prediction models, specifically ProVent14. 

According to Dr. Kanamgode, the ProVent14 score was developed in the early 2010s as an objective bedside tool to estimate 1-year mortality risk in patients requiring prolonged mechanical ventilation. The tool incorporates readily available clinical variables on day 14 of ventilation, including age, platelet count, vasopressor use, need for hemodialysis, and trauma admission status, to support prognostic assessment and potentially inform clinical decision-making. 

“This remains highly relevant in modern critical care, where advances in ICU care have improved short-term survival, yet long-term outcomes for patients with prolonged mechanical ventilation remain poor,” explained Dr. Kanamgode. “Prior studies have shown that only about 1 in 10 patients with prolonged mechanical ventilation regain functional independence at home, nearly three-quarters do not survive to 1 year, and almost half do not survive beyond 90 days after tracheostomy placement. In this context, ProVent14 offers a practical framework to support goals-of-care discussions and guide difficult decisions for clinicians and families.” 

Despite its promise, contemporary implementation of ProVent14 has been limited by the lack of validation in modern ICU cohorts and uncertainty regarding whether the model can be applied earlier in the course of mechanical ventilation, when many important decisions are made. This study addressed these gaps by evaluating the performance of ProVent14 in a contemporary ICU population using the MIMIC-IV database and by testing its utility not only at day 14, but also at earlier time points on days 7 and 10. The team found that ProVent14 maintained stable discrimination and acceptable calibration in contemporary data. Importantly, scores calculated using day 7 and day 10 values showed predictive validity similar to that of the original day 14 model, suggesting that meaningful prognostication may be possible earlier in the clinical course and across a broader group of eligible patients. Dr. Kanamgode and her colleagues also found that model performance remained stable across the years of the study, further supporting the continued relevance of the ProVent framework in current critical care practice.