Artificial neural networks based modeling for pharmacoeconomics application

Palliative chemotherapy is one of the major parts among costs of cancer. An aging population, introduction of new technologies and increased patients' healthcare expectations made the pharmacoeconomics analyses one of the mostly used tools in the decision making process. One of the main problems existing in pharmacoeconomics studies is the estimation of the effect. The aim of the study was to develop and validate on a real data artificial neural networks based systems for medical effect prediction and use them as a tool for modeling effect in pharmacoenomics analyses. Analysis was conducted on retrospective data from clinical records of non-small cell lung cancer in IIIB or IV (inoperative) stage patients treated with various therapy schemes. Logistic regression was used as the validation method. Based on the analysis of mean life duration (survival median) the output value in the database was transformed into a binary form. The threshold obtained after scientific, medical literature analysis was set at 35 weeks. Total classification rate as well as classification rate of both classes and AUROC value were used as a determinants of ANN's effectiveness. Best obtained neural model results was 84% of correctly classified records with 76% and 89% of class 1 (survival >35 weeks) and class 0 (survival <35 weeks). The AUROC value was 0.82. After training with using whole dataset, numerical experiments were conducted. Tests with in silico chemotherapeutics replacement was done. The results confirms effectiveness of artificial neural networks based pharmacoeconomics models.