Machine Learning and Causal Inference for the estimation of the effect of tacrolimus on kidney rejections

More Info
expand_more

Abstract

Tacrolimus is an immunosuppressive drug given to kidney transplant patients. A low concentration of this drug can lead to kidney rejection, but to our knowledge no research has been done to causally connect the two. This paper investigates the causal effect of tacrolimus concentration on kidney rejection occurrence using predictive analysis and a marginal structural model. The data utilized in this study was obtained from a randomized clinical trial conducted at the Erasmus Medical Center, Rotterdam. The challenges posed by limited data availability and class imbalance were carefully considered in designing the model structures. To investigate the predictive properties of tacrolimus related variables we compared results of Logistic Regression and XGBoost models on different sets of variables, yielding inconclusive results. To measure the causal effect of tacrolimus concentrations on the rejection probability, a marginal structural model was developed to estimate the causal effect of the percentage of hours spent within the target tacrolimus concentration range on the probability of kidney rejection. While a large amount of uncertainty remains, our estimates tentatively indicate a decrease as the percentage in rejection probability as the percentage of hours on target increased. Future studies are recommended to explore alternative datasets to enhance the confidence of the findings.