Prediction of Outcomes After Surgery for Unruptured Intracranial Aneurysms
Accurate preoperative identification of patients at high risk for adverse outcomes would be clinically advantageous, as it would allow enhanced resource preparation, better surgical decision-making, enhanced patient education and informed consent, and potentially even modification of certain modifiable risk factors. The aim of the Prediction of adverse events after microsurgery for intracranial unruptured aneurysms (PRAEMIUM) study is therefore to develop and externally validate a clinically applicable, robust ML-based prediction tool based on multicenter data from a range of international centers.
The Prediction of Adverse Events After Microsurgery for Intracranial Unruptured Aneurysms (PRAEMIUM) Study
Introduction Unruptured intracranial aneurysms (UIAs) are incidentally detected at an increasing rate, mostly owing to the rise in availability of non-invasive cranial imaging. Decision-making in UIAs is complex and requires consideration of many risk factors for aneurysm growth and rupture to balance the benefits and risks of treatment versus observation. This is due to: 1) the high morbidity and case fatality inherent to aneurysmal subarachnoid hemorrhage (SAH) 2) the relatively low rupture rate of unruptured aneurysms; 3) the potential morbidity and mortality rate associated with either microsurgical or endovascular treatment.
Some consistent risk factors for rupture have been identified, including involvement of the posterior circulation, larger diameter, higher age, and some specific populations such as Japanese and Finnish patients. Many other risk factors have been suggested based on varying levels of evidence. However, it is difficult to integrate this considerable number of factors into a single risk assessment and to present a clear clinical decision making algorithm to patients. A range of scoring systems have been developed and validated to approximate the risk of rupture (PHASES) and growth (ELAPSS) or to balance the risks and benefits of microsurgical treatment versus follow-up imaging directly (UIATS) by integrating some of these risk factors. Still, these scores are focused on predicting rupture events instead of neurological outcome. In addition, they usually are focused on solely one outcome, instead of providing a wide range of objective predictive analytics that may then improve shared decision-making.
Machine learning (ML) methods have been extraordinarily effective at integrating many clinical patient variables into one holistic risk prediction tailored to each patient. A previous pilot study has been carried out to assess the feasibility of predicting surgical outcomes after surgery for UIAs in a small single-center sample, and it was found that prediction was feasible with good performance metrics, and the most important factors to be included in such models were also identified. A robust, multicenter, externally validated prediction model or predictive score for surgical outcome after microsurgery for UIAs does not yet exist.
Methods Data will be collected by a range of international centers. Overall, the model will be built and publication will be compiled according to the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) guidelines.
Each center will collect their data either retrospectively, or from a prospective registry, or from a prospective registry supplemented by retrospectively collected variables. Data from patients operated from January 1st 2010 and onwards will be eligible for inclusion. Data collection should be completed, and deidentified data should be sent to the sponsor institution.
A standardized Excel spreadsheet will be provided by the sponsor. The data will be entered in standardized and anonymized form. This spreadsheet will only contain a study-specific patient number. The data set is anonymized source data that includes clinical data extracted from electronic health records (retrospectively or from a prospective registry of already existing data). The data will be anonymized upon entering them into the PRAEMIUM Excel spreadsheet, after which the patients will be numbered consecutively and there will be no way to trace the data back to individual patients. No identifiable data such as date of birth will be included. Whenever the PRAEMIUM Excel spreadsheet is transferred, it will be encrypted using a password and sent through a secure institutional e-mail server. The password will be sent in a separate e-mail. Some missing data is acceptable, but should be kept to a minimum (i.e. must be < 10%)
Endpoint Definitions Models will be developed for the following three endpoints at discharge: Poor neurological outcome (1), as well as presence of (2) new sensorimotor neurological deficits and (3) any complications (surgical or non-surgical). Neurological outcome was assessed by the modified Rankin scale (mRS), and a favorable neurological outcome was defined as mRS 0, 1, or 2. Complications will be assessed using the modified 2009 Clavien-Dindo grading (CDG), and occurrence of a complication was defined as any deviation from CDG 0.The Clavien-Dindo grading system is a classification of surgical complications: Grad 0 signifying no complication, Grade I identifying complications with any deviation from the normal intra- or postoperative course requiring medical treatment, and so forth. Detailed definitions are provided in the Excel spreadsheet. Surgery-related as well as none-surgery-related complications are counted. In case of multiple complications, only the complication with the highest CDG was counted per patient.
Input Feature Definitions All features are measured preoperatively. Recorded baseline variables will include age, gender, maximum aneurysm diameter, anatomical location (artery), total number of aneurysms per patient, if multiple aneurysms were treated during the index session, calcification of the aneurysm wall or neck, aneurysm morphology (saccular, dissecting, fusiform, or other), involvement of critical perforating or branch vessels, and intraluminal thrombosis.
In addition, the investigators will capture prior SAH, mRS at admission, prior aneurysm treatment, presence of anticoagulation/antiplatelet therapy preoperatively, and hypertension, as well as American Society of Anesthesiologists (ASA) grading, the PHASES, ELAPSS, and UIATS scores including the UIATS "pro-repair" and "pro-conservative treatment" subscores. The unruptured intracranial aneurysm treatment score (UIATS) consists of two subscores: One that represents the strength of recommendation for invasive repair of an unruptured aneurysm, and one that represents the strength of recommendation for conservative management of an unruptured aneurysm. The final overall UIATS score is subsequently calculated as the difference between the two subscores. Also included was the surgical approach: minimally invasive or standard approach, and whether a bypass was performed.
Aneurysm, Brain, machine learning, unruptured aneurysm, Intracranial Aneurysm, Aneurysm, Microsurgery
You can join if…
Open to people ages 18 years and up
- Adult patients (18 or older)
- Undergone microsurgical treatment for unruptured intracranial aneurysm
- Patients with prior SAH may only be included when surgical treatment occurred at least 4 weeks after ictus.
- Treated from January 1st 2010 onwards
You CAN'T join if...
- No specific exclusion criteria
not yet accepting patients
San Francisco California 94143 United States
- Stanford University
not yet accepting patients
Stanford California 94305 United States
- accepting new patients at some sites,
but this study is not currently recruiting here
- Start Date
- Completion Date
- University of Zurich
- Study Type
- Expecting 4000 study participants
- Last Updated