Klinik deney verileri ve makine öğrenmesi araçları kullanılarak ilaçların potansiyel ciddi advers etkilerinin bulunması

Healthcare is improving day by day and these developments make healthcare more accessible and this leads to production of large amount of data. The interpretation of these data, making assumptions and revealing significant results by using data analysis methods are important here as in every field that produces big data. Analysed data that are collected during clinical trials have great effect in ensuring developments in healthcare. Adverse event reports are one of the important parts of the clinically studied drugs or treatments. Evaluating the safety of an anticancer treatment through serious adverse events in clinical practice can provide important information for future cancer treatments. In this study, we propose a method to discover the links between serious adverse events and drugs, and between serious adverse events themselves. Our hypothesis is that, it can be possible to estimate the drug specific serious adverse events that occur together by using the clinical trial data that are transformed into a table structure for turning data into information to provide significant insights. We used ClinicalTrials.gov to download the clinical trial results that reported serious adverse events and in particular studied the anticancer drugs Cytarabine, Sorafenib and Doxorubicin. We used the MeSH and the CTCAE thesaurus to assign unique IDs to the serious adverse events to handle the inconsistency in the reports of serious adverse events. t-SNE and DBSCAN combination was used to find similar serious adverse events. To cluster the serious adverse events based on the drugs, we used spectral co-clustering. With the help of the hierarchical structure of the thesaurus, the p-values of the root and parent events were calculated to find significant ones that are encountered more or less, relatively, in a specific drug. Most of the results of this study are compatible with the available sources in the literature and the approach provided could predict the serious adverse events that are specific to new treatment options.