Desenvolvimento de ferramenta para auxiliar a determinação de intervalo de referência em parâmetros laboratoriais a partir de um banco de dados de grande porte

Abstract

The determination of the reference interval (IR) comes from an extensive population study, however it is of great importance for medical decision-making. This study aims to evaluate the possibility of determining reference intervals (RI) for clinical laboratory examination, using indirect methodology from a large database with information technology resources. The direct methodology, despite presenting more reliable results, is a more costly and complex method than the indirect methodology, requiring recruitment, clinical evaluation and cost of materials with the patient. The possibility of evaluating the RI using the indirect methodology, allows routine exams where only the patient's report is considered, to belong to a pool of information where the data becomes information, much more quickly and economically, enabling improvement in clinical criteria used for diagnosis, indirect (retrospective) estimation methods, which calculate the RI from information in the database, are an alternative to the direct methodology. Using the large volume of data stored in the database of the clinical laboratory, Big Data, we can perform extraction activities, Data Mining, where we managed, by adjusted parameters, to filter the clinically healthy population, Data Cleanring, allowing to work the subsequent data only from a population where they meet the restriction criterion. This study is framed in Law No. 12,527 / 2011, which seeks to conduct the research in a way that does not reveal the patient's identification and make it anonymous, no data beyond what was necessary was and will be used in order to individualize the patient, allowing the use of given under the cover of the law. The selected samples were randomized and all performed on the same platform, Cobas 6000 Roche, in the full period of 2018, both sexes and covering from zero to 99 years of age as a validation sample of the statistical model and exclusion criteria. Regarding the statistical methodology, the homogeneity test of variances is performed using the Shapiro P test, subsequently evaluating the relationship of the analyte in its various systems of origin a to verify if there is a difference from the Kolmogorov-Smirnov test, then proceeded to the exclusion of extremes (outlies) taking as values above and below 3 standard deviations, finally the statistical model of Bhattacharya was applied to obtain the reference value. In all tests, the level of significance adopted was 5%. Statistical analyzes were performed using the Linux operating system, Mint 18.1 distribution, R programming language and PostgreSQL database. The values ​​found for total glucose, in the 75th and 95th percentiles were: 73 to 108 mg / dL, respectively. For insulin, in these percentiles were 1.2 to 21 mcU / mL, respectively. The values ​​of the parameters evaluated here, defined in different age groups of Brazilians in the city of Rio de Janeiro, may represent decision limits for the Brazilian pediatric population, contributing to improve the diagnosis in our country