A research team in Japan has outlined a newly developed model for predicting the likelihood of intracranial aneurysms recanalising after coil embolisation treatments. Detailing their work in the Journal of Stroke and Cerebrovascular Diseases, researchers propose a “practical, externally validated” scoring system based on four significant risk factors for recanalisation: rupture status, aneurysm volume, Raymond-Roy occlusion classification (RROC), and volume embolisation ratio of the first coil (FVER).
Via a single-centre, retrospective analysis including patients with cerebral saccular aneurysms who underwent initial coil embolisation at their hospital between 2012 and 2023, Ken Aoki, Hiroyasu Nagashima and Yuichi Murayama—all based at the Jikei University School of Medicine in Tokyo, Japan—attempted to provide clarity on the known postoperative recanalisation risks associated with the procedure. Prior analyses have revealed recanalisation rates close to 25% in addition to retreatment rates of roughly 10–20% in these cases.
The researchers excluded cases in which there was less than one year of follow-up, retreatment, or utilisation of bioactive coils, and their key outcomes of interest were postoperative RROC scores in addition to a number of aneurysm characteristics. Univariate and multivariate Cox proportional hazard models were used to identify independent recanalisation predictors, with a simplified risk score being constructed using least absolute shrinkage and selection operator (LASSO) logistic regression and β-coefficients from multivariable analysis. Both internal and external validation of the scoring system were performed.
Some 79 patients were ultimately analysed, with 21 experiencing recanalisation (26.6%) and eight undergoing retreatment during the follow-up period. Based on multivariate analyses, the researchers identified aneurysm rupture, aneurysm size >7mm, neck size >5mm, aneurysm volume >155mm3, immediate postoperative absence of complete aneurysm occlusion (RROC without Class I), FVER <8%, and first coil percentage (FCP) <26%, as potential recanalisation predictors. They also found that, while balloon assistance was more prevalent in recanalised versus non-recanalised patients, this difference did not reach statistical significance.
The researchers subsequently settled on four independent predictors of post-coiling recanalisation: aneurysm rupture, aneurysm size ≥7mm, RROC without Class I, and FVER <8%. An integer-based risk scoring system—ranging from zero to seven in value—was constructed based on these variables.
“In the cohort analysed in this study, the model demonstrated strong discrimination with a C-statistic of 0.87,” the authors note. “Internal validation using 1,000 bootstrap replications yielded a bias-corrected C-statistic of 0.89. External validation was conducted using an independent cohort of 468 patients, of whom 35 (7.5%) experienced recanalisation. The model retained good discriminative ability in this external dataset with a C-statistic of 0.81.”
They go on to detail that calibrations within the external validation cohort revealed a “slight overestimation” of recanalisation risk in patients with higher scores, while the Hosmer-Lemeshow test indicated a statistically significant ‘poor fit’ (p=0.0007) for the model. Risk stratification based on total scores demonstrated “clinically relevant separation”, as recanalisation occurred in 1.8% of patients with low scores (0–2), 13.5% with intermediate scores (3–4), and 41.5% with high scores (5–7), confirming—in the researchers’ view—the utility of their scoring system for individualised risk assessments.
While several prior studies have proposed alternative models for predicting post-coiling aneurysm recurrence, the present approach is distinct from these efforts in that it places greater emphasis on “clinical practicality”, and ultimately offers a “simple” solution to inform risk-based follow-up planning and support clinical decision-making.
“Aneurysm volume had the highest individual AUC [area under the curve] but, due to measurement variability, aneurysm size was used instead,” the authors explain. “FVER was chosen over FCP as it can be computed immediately post-embolisation without procedural dependency. The score reflects effect sizes from the multivariable model, while accounting for predictor collinearity. When applied to our cohort, our model demonstrated superior discriminative power (AUC, 0.89) compared to [other] models.”
The researchers feel that key limitations of their study include its single-centre, retrospective nature and relatively small number of recanalisation events in both cohorts—and, while their model appears able to effectively distinguish between low- and high-risk patients, its predictive probabilities “did not fully align with observed outcomes in the external population”.
“This discrepancy may reflect differences in patient characteristics, procedural techniques or follow-up imaging schedules between cohorts,” they add. “Nevertheless, risk stratification remained clinically meaningful across all score categories. Based on the observed threshold, we propose that patients with a score ≥3 be considered for shorter-interval imaging follow-up due to increased recanalisation risk.”
Concluding their paper, the authors posit that further multicentre validations across diverse populations are necessary to fully establish the scoring system’s generalisability, and—while they included morphological and procedural variables—haemodynamic factors like wall shear stress were not evaluated and should be incorporated into future prediction models utilising larger, prospective datasets.
