Factors affecting the performance of brain arteriovenous malformation rupture prediction models

Machine Learning Approach for Analyzing 3-Year Outcomes of Patients with Brain Arteriovenous Malformation (AVM) after Stereotactic Radiosurgery (SRS)

A cerebral arteriovenous malformation (AVM) is a tangle of abnormal blood vessels that irregularly connects arteries and veins. Stereotactic radiosurgery (SRS) has been shown to be an effective treatment for AVM patients, but the factors associated with AVM obliteration remains a matter of debate. In this study, we aimed to develop a model that can predict whether patients with AVM will be cured 36 months after intervention by means of SRS and identify the most important predictors that explain the probability of being cured. A machine learning (ML) approach was applied using decision tree (DT) and logistic regression (LR) techniques on historical data (sociodemographic, clinical, treatment, angioarchitecture, and radiosurgery procedure) of 202 patients with AVM who underwent SRS at the Instituto de Radiocirugía del Perú (IRP) between 2005 and 2018. The LR model obtained the best results for predicting AVM cure with an accuracy of 0.92, sensitivity of 0.93, specificity of 0.89, and an area under the curve (AUC) of 0.98, which shows that ML models are suitable for predicting the prognosis of medical conditions such as AVM and can be a support tool for medical decision-making. In addition, several factors were identified that could explain whether patients with AVM would be cured at 36 months with the highest likelihood: the location of the AVM, the occupation of the patient, and the presence of hemorrhage.

Endovascular Management of Brain Arteriovenous Malformations

Due to the risk of cerebral hemorrhage, and its related morbidity-mortality, brain arteriovenous malformations (bAVMs) are a rare and potentially life-threatening disease. Despite this, there is only one randomized controlled trial on bAVM management, A Randomized trial of Unruptured Brain Arteriovenous malformations (ARUBA). The results of the ARUBA trial favor a noninterventional approach in the case of an unruptured bAVM; however, implementation of these findings is challenging in daily practice. Instead, management of bAVM relies on multidisciplinary discussions that lead to patient-specific strategies based on patient preferences, local expertise, and experience in referral centers. Considering the diverse patterns of presentation and numerous treatment modalities, implementing standardized guidelines in this context proves challenging, notwithstanding the recommendations or expert opinions offered. Endovascular treatment (EVT) of bAVM can be curative, or can serve as an adjunct treatment prior to surgery or radiosurgery ("pre-EVT"). EVT practice is in constant evolution (i.e., venous approach, combination with surgery during the same anesthesia, etc.). Liquid embolic agents such as ethylene vinyl alcohol (EVOH) copolymer and cyanoacrylates (CYA), and their method of injection to increase bAVM occlusion have also benefited from technical evolutions such as the use of adjunctive flow arrest techniques (mini balloons, pressure cooker technique, and multiple catheters). Further research is necessary to evaluate the advantages and disadvantages of EVT for bAVM.

Development and Validation of a Scoring System for Hemorrhage Risk in Brain Arteriovenous Malformations

IMPORTANCE

The dilemma between natural rupture risk and adverse outcomes of intervention is of major concern for patients with unruptured arteriovenous malformations (AVMs). The existing risk score for AVM rupture includes factors that are controversial and lacks prospective validation.

OBJECTIVE

To develop and robustly validate a reliable scoring system to predict the rupture risk of AVMs.

DESIGN, SETTING, AND PARTICIPANTS

This prognostic study developed a prediction model derived from a single-center cohort (derivation cohort) and validated in a multicenter external cohort (multicenter external validation cohort) and a cohort of patients receiving conservative treatment management (conservative treatment validation cohort). Patients were recruited from a nationwide multicenter prospective collaboration registry in China. A total of 4135 patients were enrolled in the registry between August 1, 2011, and September 1, 2021. Of those, 3962 patients were included in the study (3585 in the derivation cohort and 377 in the multicenter external validation cohort); 1028 patients from the derivation cohort who had time-to-event data and prerupture imaging results were included in the conservative treatment validation cohort. Data were analyzed from March 10 to June 21, 2022.

MAIN OUTCOMES AND MEASURES

A scoring system was developed based on risk factors identified from a literature review and a robust selection process. Patients were stratified into different risk groups based on scores to calculate hemorrhage-free probability in future years, and Kaplan-Meier curves were plotted to visualize risk stratification. Receiver operating characteristic curves were used to assess the discrimination of models. Univariable analyses (logistic regression analysis for descriptive data and Cox regression analysis for survival data) were used to compare baseline information and assess bias.

RESULTS

Among 3962 patients (2311 men [58.3%]; median [IQR] age, 26.1 [14.6-35.5] years), 3585 patients (2100 men [58.6%]; median [IQR] age, 25.9 [14.6-35.0] years) were included in the derivation cohort, and 377 patients (211 men [56.0%]; median [IQR] age, 26.4 [14.5-39.2] years) were included in the multicenter external validation cohort. Thirty-six hemorrhages occurred over a median (IQR) follow-up of 4.2 (0.3-6.0) years among 1028 patients in the conservative treatment validation cohort. Four risk factors were used to develop the scoring system: ventricular system involvement, venous aneurysm, deep location, and exclusively deep drainage (VALE). The VALE scoring system performed well in all 3 cohorts, with areas under the receiver operating characteristic curve of 0.77 (95% CI, 0.75-0.78) in the derivation cohort, 0.85 (95% CI, 0.81-0.89) in the multicenter external validation cohort, and 0.73 (95% CI, 0.65-0.81) in the conservative treatment validation cohort. The 10-year hemorrhage-free rate was 95.5% (95% CI, 87.1%-100%) in the low-risk group, 92.8% (95% CI, 88.8%-97.0%) in the moderate-risk group, and 75.8% (95% CI, 65.1%-88.3%) in the high-risk group; the model discrimination was significant when comparing these rates between the high-risk group and the low- and moderate-risk groups (P < .001 for both comparisons).

CONCLUSIONS AND RELEVANCE

In this prognostic study, the VALE scoring system was developed to distinguish rupture risk among patients with AVMs. The stratification of unruptured AVMs may enable patients with low risk of rupture to avoid unnecessary interventions. These findings suggest that the scoring system is a reliable and applicable tool that can be used to facilitate patient and physician decision-making and reduce unnecessary interventions or unexpected AVM ruptures.

External validation of brain arteriovenous malformation haemorrhage scores, AVICH, ICH and R2eD

PURPOSE

To externally validate the arteriovenous malformation-related intracerebral haemorrhage (AVICH), intracerebral haemorrhage (ICH), and novel haemorrhage presentation risk score (R2eD) in brain arteriovenous malformations.

METHODS

Adult patients diagnosed radiologically with an arteriovenous malformation (AVM) at a tertiary neurosurgical centre between 2007 and 2018 were eligible for inclusion. Both the AVICH and ICH scores were calculated for AVM-related symptomatic haemorrhage (SH) and compared against the modified Rankin scale (mRS) at discharge and last follow-up, with unfavourable outcome defined as mRS > 2. R2eD scores were stratified based on presentation with SH. External validity was assessed using Harrel's C-statistic.

RESULTS

Two hundred fifty patients were included. Mean age at diagnosis was 46.2 years [SD = 16.5]). Eighty-seven patients (34.8%) had a SH, with 83 included in the analysis. Unfavourable mRS outcome was seen in 18 (21.6%) patients at discharge and 18 (21.6%) patients at last follow-up. The AVICH score C-statistic was 0.67 (95% confidence interval [CI], 0.53-0.80) at discharge and 0.70 (95% CI, 0.56-0.84) at last follow-up. The ICH score C-statistic was 0.78 (95% CI 0.67-0.88), at discharge and 0.80 (95% CI 0.69-0.91) at last follow-up. The R2eD score C-statistic for predicting AVM haemorrhage was 0.60 (95% CI, 0.53-0.67).

CONCLUSIONS

The AVICH score showed fair-poor performance, while the ICH score showed good-fair performance. The R2eD score demonstrated poor performance, and its clinical utility in predicting AVM haemorrhage remains unclear.