data mining approach in prediction of erythropoietin dosage in hemodialysis patients

background and purpose: kidney failure reduces the kidney function and in long term it leads to chronic kidney disease. one of the main complications of this disease is irreversible damage to the kidneys (end-stage kidney disease) and hemodialysis is the main method used to treat advanced kidney failure. the main problem associated with hemodialysis is treating anemia caused by lack of erythropoietin secretion in kidney which is usually treated by synthetic erythropoietin. on the other hand, choosing the right dosage of erythropoietin is important because it is expensive and could have some complications. this research aimed at predicting the dosage of erythropoietin and identifying affecting factors. materials and methods: data was collected from a dialysis center in tehran and data mining methods were used. the input variables were measured in the past 6 months of treating patients with erythropoietin. the sequential data was then converted to the bag of features (bof) format. then support vector machines and random forest were applied on the bof to predict the erythropoietin dosage. results: the amount of medication in previous months was found to be an important factor in determining the appropriate dosage of erythropoietin for the next month. in optimal condition, random forest and svm could predict the erythropoietin dosage with an average accuracy of 90% and 79%, respectively. conclusion: this study identified the factors influencing the treatment and control of anemia in hemodialysis patients. these results could be of great benefit in prescribing the proper dosage of erythropoietin, and reducing the treatment cost and duration. moreover, it helps to prevent the complications caused by excessive use of erythropoietin such as increase in hemoglobin level.