Neurological intervention transition model for dynamic prediction of good outcome in spontaneous subarachnoid haemorrhage

Abstract

Abstract Deterioration of neurovascular conditions can be rapid in patients with spontaneous subarachnoid haemorrhage (SAH) and often lead to poor clinical outcomes. Therefore, it is crucial to promptly assess and continually track the progression of the disease. This study incorporated baseline clinical conditions, repeatedly measured neurological grades and haematological biomarkers for dynamic outcome prediction in patients with spontaneous SAH. Neurological intervention, mainly aneurysm clipping and endovascular embolisation, was also incorporated as an intermediate event in developing a neurological intervention transition (NIT) joint model. A retrospective cohort study was performed on 701 patients in spontaneous SAH with a study period of 14 days from the MIMIC-IV dataset. A dynamic prognostic model predicting outcome of patients was developed based on combination of Cox model and piecewise linear mixed-effect models to incorporate different types of prognostic information. Clinical baseline covariates, including cerebral oedema, cerebral infarction, respiratory failure, hydrocephalus and vasospasm, as well as repeated measured Glasgow Coma Scale (GCS), glucose and white blood cell (WBC) levels were covariates contributing to the optimal model. Incorporation of neurological intervention as an intermediate event increases the prediction performance compared with baseline joint modelling approach. The average AUC of the optimal model proposed in this study is 0.7783 across different starting points of prediction and prediction intervals. The model proposed in this study can provide dynamic prognosis for spontaneous SAH patients and significant potential benefits in critical care management.