Establishing of big data clinical dataset in brain vessel aneurysm research

Abstract

Variability and heterogeneity of digital medical data requires establishing of modern algorithms which provide appropriate data processing. The aim of the study was to delineate the main steps in formation of a clinical dataset of patients with brain aneurysms from the stage of producing primary mining specifications to formation of a final version.Material and methods. Data collection, crosschecking of the cases and analyses of dataset has been carried out in Turku University Hospital. Within last two decades available medical data at our hospital have been stored in digital data lake thus allowing automatized data mining. In frame of our study, data mining was performed by a data scientist utilizing R software. Inclusion criteria were based on a set of diagnosis which were coded in medical charts according to international classification of diseases (ICD 10).Resutls and Discussion. Primary data mining identified 3850 patients with brain aneurysms treated at our hospital from January 2000 till May 2018. After independent manual crosschecking of medical charts of these patients, we found 1218 (32 %) cases, which had no aneurysm (false-positive). Data of remaining true aneurysm-cases were divided into clinical and intensive care unit subsets where every event linked to particular date of treatment was defined as an info-unit. All the data in both subsets were structured into separate Excel files and presented in chronological order for each particular patient. Altogether, dataset included 70 000 000 rows of info-units found in 2632 patients.Conclusions. Data mining allowed establishment of detailed clinical dataset of patients with brain aneurysms. Produced mining algorithm had limitation regarding false-positive cases (32 % patients). Based on that, we recommend manual crosschecking of automatically collected dataset before statistical analysis.