2025 Proffered Presentations
S210: PREDICTING WHO GRADE IN MENINGIOMA SURGERY: DEVELOPMENT OF AN UNBIASED VARIABLE SELECTION AND VALIDATION APPROACH USING PREOPERATIVE CLINICAL AND RADIOGRAPHIC DATA
Philip Ostrov, MD1; Raja N Jani, DO, MHA1; Maunil Mullick1; Niraj Rama1; Amrit Avula2; Arshi Chopra1; Arinan Dourado, PhD1; Isaac Abecassis1; Akshitkumar Mistry1; Brian J Williams, MD1; 1University of Louisville; 2Washington University St. Louis
Introduction: Meningiomas represent the most common intracranial tumor, but their diverse clinical presentations and biological behaviors make it challenging to predict patient outcome and tumor grade. Traditional prediction models often rely heavily on imaging data, which, while informative, may not capture the full spectrum of factors influencing patient prognosis. This study aims to develop an algorithm that integrates clinical, lab, and radiographic data to predict WHO grade using preoperative information.
Methods: We conducted a retrospective analysis of 60 patients surgically treated for meningioma at the University of Louisville from 2020 to 2024. Our dataset included a comprehensive array of variables such as patient demographics, clinical data (comorbidities, monocyte count, platelet to lymphocyte ratio, neurological status), radiographic characteristics (tumor size, location, imaging findings), and pathological features (Ki67 index, brain invasion, WHO grade). First, multiple univariate linear regressions were obtained for each unique variable and outcome in our dataset to assess for statistical significance. After which, the first iteration of our multiple regression model was built via a stepwise method with K-fold cross validation with k = 5 to prevent overfitting.
Results: In the initial stages of developing our model, we conducted several linear regression analyses that revealed significant associations within our dataset. Specifically, we found that age at diagnosis (p = 0.01813, OR 0.876 (0.777, 0.972), Charleson comorbidity index (p = 0.00721, OR 3.038 (1.413, 7.401), presence of preoperative FLAIR signal in adjacent brain(p = 0.01630, OR 11.262 (2.017, 121.722), and lesion volume (p = 0.04465, OR 1.019 (1.002, 1.041) were all associated with WHO grade independently. These findings were a robust foundation for the development of the model, which can predict WHO grade 1 vs WHO grade 2 or 3 with 84.3% accuracy.
Discussion: Our integrative variable selection and validation approach allows for the incorporation of a broader spectrum of data. This finding highlights the importance of integrating diverse data types beyond purely imaging-based models described in the literature. The inclusion of clinical, lab, and radiographic features enhances the model's predictive power. This holistic approach offers an accurate and individualized prediction of pathological findings, which is crucial for personalized treatment planning and resource allocation.
Conclusion: Developing this comprehensive approach furthers our understanding of predictive analytics in the treatment of meningioma. By integrating clinical, radiographic, and lab data, our model provides a more detailed and accurate prediction of WHO grade. This approach can improve clinical decision-making, optimize patient care, and enhance the allocation of healthcare resources.
