 The study investigates the impact of missing data and imputation methods on the performance of machine learning classifiers in both simulated and real-world clinical datasets and introduces a new class of discrepancy scores to assess imputation quality. The results show that the percentage of missingness in the test data has a significant effect on classifier performance and commonly used measures for assessing imputation quality may lead to poorly matched imputed data. The study emphasizes the importance of considering imputation quality when performing downstream classification as it can significantly affect the interpretability and accuracy of classifier models. This article was authored by Toulou Shadbar, Michael Roberts, Jan Stanczuk, and others.