Published in:
Open Access 01-03-2022 | Intracranial Aneurysm | Original Article - Vascular Neurosurgery - Aneurysm
Current decision support tools fail to agree or predict therapeutic decisions in a single cohort of unruptured intracranial aneurysms
Published in: Acta Neurochirurgica | Issue 3/2022
Login to get accessAbstract
Background
There is limited evidence to direct the management of unruptured intracranial aneurysms. Models extrapolated from existing data have been proposed to guide treatment recommendations. The aim of this study is to assess whether a consensus-based treatment score (UIATS) or rupture rate estimation model (PHASES) can be used to benchmark UK multi-disciplinary team (MDT) practice.
Methods
Prospective data was collected on a consecutive series of all patients with unruptured intracranial aneurysms (UIAs) presenting to a major UK neurovascular centre between 2012 and 2015. The agreement between the UIATS and PHASES scores, and their sensitivity and specificity in predicting the real-world MDT outcome were calculated and compared.
Results
A total of 366 patients (456 aneurysms) were included in the analysis. The agreement between UIATS and MDT recommendation was low (weighted kappa 0.26 [95% CI 0.19, 0.32]); sensitivity and specificity were also low at 36% and 52% respectively. Groups that the MDT allocated to treatment, equipoise or no treatment had significantly different PHASES scores (p = 0.004). There was no significant difference between the two scores when predicting patients for whom MDT outcome was to recommend aneurysm treatment, but the UIATS score was superior in predicting patients who received an MDT recommendation of ‘treatment-equipoise’, or ‘not-for-treatment’ (AUC of 0.73 compared to 0.59 for PHASES).
Conclusions
The models studied failed to agree with the consensus view of multi-disciplinary team in a major neurovascular centre. We conclude that decision support tools such as the UIATS and PHASES scores should not be blindly introduced in respective institutions without prior internal validation, as they may not represent the local reality.