Background: Hematological toxicities are common in colorectal cancer patients taking rnchemotherapy. The development of a valid predictive model would go a long way in rnpreventing for such toxicities. rnObjectives: To develop and validate a prediction model for severe hematological rntoxicities in adult colorectal cancer patients taking chemotherapy at Tikur Anbessa rnSpecialized Hospital rnMethods: A retrospective cohort study was conducted using medical records of adult rncolorectal cancer patients who received chemotherapy from 2017 to 2021 at Tikur rnAnbessa Specialized Hospital. The model was developed using uni-variable and rnmultivariable logistic regression analyses, and it was validated using bootstrapping. The rnmodel was developed for severe neutropenia only because of limited number of outcomes rnfor anemia and thrombocytopenia. Discrimination and calibration were used to determine rnthe modelâ€™s prediction accuracy. All statistical tests were two-sided and P-value < 0.05 rnwas considered statistically significant. rnResults: A total of 224 colorectal cancer patients were considered for analysis. About rn rn114 (50.9%), 25 (11.2%) and four (1.8%) patients developed severe neutropenia, anemia rnand thrombocytopenia respectively. Age, polychemotherapy, type of therapy, and longer rnduration of chemotherapy were predictors of severe neutropenia. The model had rnsensitivity of (71.05% versus 73.24%) and specificity of (71.82% versus 68.49%) in the rnderivation and validation cohorts respectively. The area under receiver operating curve rnwas 0.7995 for the derivation and 0.7741 for the validation cohorts. rnConclusions: Neutropenia was the most common hematological toxicity. The application rnof the developed model could help to identify high-risk patients for severe neutropenia rnand to institute preventive strategies before neutropenia develops.