Identifying key predictors of MSI/MMR status in colorectal cancer: Insights from a real-world clinical dataset.

e15718 Background: Microsatellite instability (MSI) and mismatch repair (MMR) testing is critical for guiding therapeutic decisions in colorectal cancer (CRC). Despite their clinical importance, routine MSI/MMR testing faces significant challenges, including high costs, long turnaround times, and pathology workforce shortages. Artificial intelligence (AI) offers opportunities to overcome these barriers through data-driven solutions. To maximize the clinical utility and performance of AI-based approaches, it is crucial to identify and analyze the key predictors of MSI/MMR status. Such analysis not only aids in developing more accurate predictive models, but also supports efforts to adapt diagnostic tools to diverse patient populations. To this end, we conducted a comprehensive study with a real-world clinical dataset of retrospective CRC cases to identify the critical factors associated with MSI/MMR status. Methods: Clinical and histopathological data from 800 CRC cases at St James’s University Hospital (UK) were analyzed using a random forest (RF) classifier to identify the most important predictors for determining the overall MSI/MMR status. MSI/MMR testing was done as part of routine clinical care, with 11.9% of cases classified as MMR-deficient/MSI-high (n = 95). Normalized importance values were computed for each feature to quantify their contribution. Pairwise correlation analysis was conducted using Cramer's V and Chi-squared tests to evaluate interdependencies among features. The cohort included 308 patients (38.5%) under 65 years of age and 492 patients (61.5%) aged 65 or older, with a higher proportion of males (n = 453, 56.6%) compared to females (n = 347, 43.4%). The majority of patients were diagnosed with Stage II and III cancers (n = 557, 69.6%) and had tumors graded as moderately differentiated (n = 620, 78.6%). The primary tumor site was the colon (n = 528, 66.0%) and the most common histological subtype was adenocarcinoma (n = 686, 85.8%). Results: The classifier identified age as the most influential feature associated with MSI/MMR status, with an importance of 52.9%. Stage and grade were the next most significant contributors, accounting for 17.0% and 12.2%, respectively. Histological subtype and tumor site were less influential, with contributions of 7.8% and 6.0%, respectively. Gender was the least impactful factor, with an importance of 4.1%. Pairwise correlation analysis showed weak associations among individual features, with all Cramer's V values below 0.26 and Chi-squared tests indicating statistical significance (p < 0.001). Conclusions: Our analysis highlights the role of age, stage, and grade in determining MSI/MMR status, with minimal interdependencies among features. These results provide valuable insights for refining predictive models and advancing the development of reliable, generalizable diagnostic tools for diverse CRC patient populations.