Assessing survival and learning curves for EVD procedures in hemorrhagic stroke patients in a New York City hospital.

BACKGROUND

Types of hemorrhagic stroke and other forms of bleeding within the cranial vault, account for a significant portion of neurological emergencies. External ventricular drain (EVD) placement is a common, life-saving intervention for patients presenting with acute hydrocephalus following intracranial hemorrhage. EVDs reduce intracranial pressure (ICP) by draining cerebrospinal fluid (CSF), minimizing the risk of secondary brain injury. Additionally, EVDs allow for continuous ICP monitoring, providing critical data to guide decisions. However, post-procedural complications may occur following EVD placement. While previous studies have examined EVD outcomes using traditional statistical methods, the impact of provider heterogeneity and resident learning curves on these outcomes remains poorly understood. Understanding the impact of these complications is crucial for improving surgical quality and outcomes.

METHODS

To address these critical gaps and advance our understanding of EVD outcomes, we reviewed medical records of patients with subarachnoid hemorrhage (SAH), intracranial hemorrhage (ICH), or interventricular hemorrhage (IVH) who received EVDs for ICP at the Mount Sinai Health System in New York City from 2019 to 2022. We evaluated surgical complications and patient outcomes following EVD placement by neurosurgeons at different stages of their training. Random effect survival models were implemented to account for heterogeneity between providers. This novel methodological approach integrated provider heterogeneity and learning curves into EVD outcome analysis, extending beyond traditional statistical methods used in previous studies.

CONCLUSIONS

We found that in time-to-death Cox models, age was a significant predictor while for time-to-complication, device type and pre-operative modified Rankin score were significant predictors. A Cox frailty model did not show significant predictors, but our unique evaluation of restricted mean survival time with frailty showed that smoking was a significant predictor of survival. Furthermore, we also modeled learning curves amongst neurosurgeons performing EVDs to improve surgical proficiency. We found that employing the frailty modeling improved the learning curve fit.