Risk Analysis of Intracranial Aneurysm Rupture Through Social Determinants of Health

Intracranial aneurysms (IA), the most common intracranial arterial malformation, affect approximately 3% of the global population. They are characterized by a dilation of the vascular wall within the intracranial region, typically at a bifurcation. The most feared complication of IA is rupture, which is associated with high morbidity and mortality. About 30% of patients face death or severe disability, and even among those with no visible sequelae, up to 50% suffer from invisible disabilities such as cognitive problems. This explains the significant economic burden of ruptured IAs, estimated at approximately £168 million annually in the UK.

One way to reduce this morbidity and societal burden is to identify patients at high risk of rupture to determine those most likely to experience disease progression. Various initiatives, such as the ELAPSS, PHASES, and UCAS scores, have been developed to assess rupture risk in routine clinical practice. However, these scores largely rely on well-known modifiable and non-modifiable risk factors, such as ethnicity, hypertension, sex, age, smoking, and history of subarachnoid hemorrhage. Despite their utility, they are insufficient for reliably stratifying rupture risk, underestimating it by approximately 30%, which limits their impact in daily practice.

Recent research has identified emerging risk and protective factors, such as sleep apnea syndrome, lipid control, and the use of antidiabetic, lipid-lowering, or antihypertensive treatments. Therefore, it is necessary to revisit IA rupture risk by considering these factors in relation to patients' living environments.

Social determinants of health (SDOH) play a critical role in this context. Defined as the conditions of life that influence individual and population health, SDOH include broad factors such as age, ethnicity, and gender, as well as addiction behaviors, income levels, environmental exposure, and social status. They are estimated to contribute to over 80% of the population's health outcomes. In vascular diseases, SDOH -such as low education levels, limited income, social isolation, and low physical activity-are known to impact disease severity beyond traditional risk factors. The WHO estimates that vascular diseases are the leading global cause of mortality, with 17.7 million deaths annually, half of which are preventable. Understanding and addressing SDOH is therefore crucial for public health prevention efforts.

Initial studies on the impact of SDOH on IA rupture are limited, often focusing on individual determinants. A more comprehensive approach to SDH is needed, especially given the lack of consensus guidelines for IA management. Current therapeutic options, while advanced, still carry perioperative morbidity. Multivariable modeling and weighted composite scores can help address the complexity of clinical and environmental factors.

SDH data are routinely collected in medical records and increasingly digitized through hospital information systems. Health data warehouses (HDWs), such as the one at Nantes University Hospital, integrate structured and unstructured data from millions of patients, enabling their reuse for research purposes. The advent of artificial intelligence (AI) has made it possible to extract and analyze this data, revealing insights from previously inaccessible information.

Natural language processing (NLP), a branch of AI, enables automated extraction of information from textual medical records, unlocking the potential of unstructured data within HDWs. State-of-the-art models, such as BERT, ChatGPT, Llama, and Mistral, can process and analyze large volumes of text. However, the application of NLP to non-English languages is constrained by a lack of annotated corpora, privacy concerns, and data sovereignty regulations.

The ARAMISS project aims to investigate the interaction between SDH and known risk factors for IA rupture, comparing patients with unruptured IAs to those with ruptured IAs. Using SDH, the project will classify patients into risk groups and spatially analyze patient data through APIs like DataGouv. Additionally, the project will focus on developing FAIR (Findable, Accessible, Interoperable, Reusable) methodologies to share its algorithms and training corpora with the biomedical community while maintaining regulatory compliance.