Unified health database creation: 125 million brazilian cohort from information systems of hospital, outpatient, births, notifications and mortalities

Abstract

MethodsSemantic analysis of data was performed to describe and understand different meanings of different fields existing in the studied bases. In addition, there were four main procedures, executed with database operations tools and PLSQL programming language: cleaning and standardization of databases(document's numbers was checked in the brazilian national people's database, with a string approximator algorithm to decide if the document's number belonged or no the register); registration information extraction, deterministic and probabilistic deduplication thereof. The procedures were first performed on each database separately and after the unification of the records, was held again a deterministic deduplication. Except the probabilistic deduplication which was performed only on the final deterministic deduplicated's database.Performed procedures allowed a decision-making to chose fields used in data model for the unified database creation. Nine database's representative fields related to patients were selected: patient's name; patient mother's name; sex; birth date; state; city; zip code; cpf and cns(brazilian documents). ResultsInitially, the unified registration database resulted in 705.599.785 records, after deterministic deduplication there was a reduction culminating in 198.400.762 records. This reduction is explained because these databases are not fully integrated. Moreover, there is not always agreement between systems' semantics and in some cases changes occur in the data format over the period within the same system. After probabilistic deduplication, the number of unique records decreased to 124.545.186 which is explained by non-linked pairs by deterministic process. This result is guaranteed with a estimate error of at most 3.3% of false positive and at most 12.3% of false negative pairs. ConclusionThe results show that data deduplication is necessary and should be carried out thoroughly. Where the databases had limited patients' registration information, the technique enabled to capture, in more complete basis, additional information. Futhermore, it allowed to identify and assist in the understanding of positive and negative aspects within systems and trace clinical condition of patients, enabling pharmacoeconomic and epidemiological studies that define effectiveness and efficiency of public policies and embedded technologies. As future work, is important ensure the univocity of records and link this database with past period.

The long-term costs for treating multiple sclerosis in a 16-year retrospective cohort study in Brazil

“…The methodological procedures adopted for the relationship were the same ones described by Pereira et al . [ 25 ]. The monetary values were adjusted according to the purchasing power parity index (PPP) of the World Bank [ 26 ].…”

Building the National Database of Health Centred on the Individual: Administrative and Epidemiological Record Linkage - Brazil, 2000-2015

“…This technique represents an important tool to link Brazilian health data, providing wider applicability than that for which they were created. Health record linkage is widely used by international and national scientific studies aiming to develop integrated databases focused on individuals (5)(6)(7)(8)(9)(10)(11)(12). According to these authors, the probabilistic deduplication has been successful in enabling epidemiological and economic studies in the health field.…”