Multiple Imputation Strategies in Biomedical Research: Statistical Methods and Clinical Applications

Resumen

Abstract – This study examines the issue of missing data in biomedical research and evaluates the effectiveness of different imputation strategies. Multiple imputation is highlighted as a robust statistical method for improving the validity of analyses, compared to traditional approaches such as eliminating incomplete cases or imputing missing values with the mean, which can introduce bias and reduce statistical accuracy. Simulations and comparative analyses were conducted on biomedical databases to assess the impact of various imputation methods on the preservation of variability and the accuracy of predictive models. The results indicate that K-Nearest Neighbors (KNN) imputation better preserves the original data structure compared to mean imputation, which tends to reduce value dispersion. Additionally, challenges such as the correct specification of imputation models and the integration of machine learning algorithms in these processes are examined. Finally, recommendations are provided to enhance the implementation of multiple imputation in clinical and epidemiological studies, ensuring more accurate and reliable management of missing data in biomedical research.