Performance of Risk Scores in Predicting Intracranial Aneurysm Instability

Several risk scores targeting different end points-rupture, growth, or rupture after growth-are used to guide prophylactic treatment decisions, despite not all being designed to predict aneurysm rupture directly. We evaluated how population, hypertension, age, size, earlier subarachnoid hemorrhage, site (PHASES), earlier subarachnoid hemorrhage, location, age, population, size, shape (ELAPSS), triple-S, and Juvela scores perform in identifying rupture-prone aneurysms and predicting instability in intracranial aneurysms deemed to have low enough rupture risk to justify conservative treatment.

This retrospective study included all patients with ruptured or conservatively managed unruptured intracranial aneurysms diagnosed from 2005 to 2020 in Tampere University Hospital. Risk scores were calculated from clinical and imaging data at diagnosis. For ruptured aneurysms, score distributions were analyzed to assess sensitivity. For conservatively managed unruptured aneurysms with radiological follow-up of ≥3 months, discrimination for composite instability end point (rupture or growth of ≥1.0 mm) was evaluated using receiver operating characteristic curves, calibration plots, and decision-curve analysis.

A total of 2258 aneurysms were analyzed: 1180 ruptured and 1078 unruptured. Among conservatively managed unruptured intracranial aneurysms, 29 ruptured (3%), including 7 fatal cases identified through a nationwide cause-of-death search. Of the 519 unruptured intracranial aneurysms (48%) with radiological follow-up (median, 3.4 years; total, 2256 aneurysm-years), 71 (14%) demonstrated instability (52 growth, 19 rupture). In the ruptured cohort, PHASES demonstrated the highest sensitivity. ELAPSS and triple-S achieved the best discrimination for instability (area under the curve, 0.78 [95% CI, 0.71-0.85] and 0.79 [0.72-0.85]). Decision curve analysis indicated that ELAPSS and triple-S provided the highest net benefit within clinically relevant thresholds (6%-20%).

In the ruptured aneurysm cohort, PHASES showed the highest sensitivity for identifying rupture-prone aneurysms. In the conservatively managed cohort, ELAPSS and triple-S best discriminated instability, and all models except Juvela showed high sensitivity for rupture. All models had low positive predictive value and limited specificity, indicating that current risk scores alone are insufficient for treatment selection and require further refinement.