Drug utilization research (DUR) aims to examine patterns of medication use and adherence to treatments and to assess determinants of utilization (Godman et al., 2016; Wettermark et al., 2016) The history of DUR is described elsewhere (World Health Organization, 1993; World Health Organization, 2003b; Wettermark, 2013; Wettermark et al., 2016). Over the years, the scope of DUR has expanded; methods have improved, and the use of secondary data has increased. Nonetheless, additional work is required, particularly with regard to the quality of available data (Evans, 2012; Schneeweiss, 2019).

Secondary data that are used for pharmacoepidemiology research are usually derived from information routinely collected for administrative purposes and as part of patient care (Eriksson and Ibáñez, 2016), such as drug sales, medical billing, and prescriptions (Shalini et al., 2010). Given the cost and difficulty of primary data collection, electronic healthcare databases (EHD) are commonly used in many countries to study drug safety (Pacurariu et al., 2018). Linkage of data on medication use with diagnostic, mortality, and other health databases has become routine in Europe, North America, and Asian countries (Wettermark, 2013), but not in low- and middle-income countries, notably, in Latin America (de Castro, 1999; de Castro, 2000; World Health Organization, 2003a; Baldoni, 2011; Coelho and Santos, 2012).

While high-income countries are leveraging the use of Real-World Evidence to inform regulatory decision-making (European Medicines Agency (EMA), 2018; Health Canada, 2019; Food and Drug Administration (FDA), 2020), in Latin America initiatives are incipient and limited to few settings (Durán et al., 2016; Salas et al., 2018). In Brazil, efforts related to “open data” have improved the prospects for creating systematic approaches to the use of secondary data, not only for decision-making but also for research (Controladoria Geral da União, 2020).

Despite awareness of the value of existing databases, and observed expansion of DUR in Brazil using secondary data, a mapping of databases to evaluate their potential, as well as their characteristics and applications, has not been undertaken.

The present work aimed, therefore, to identify, catalogue, and characterize secondary data sources for DUR in Brazil.

The expert network identified 62 data sources. After application of the exclusion criteria, 39 sources were included. Two of them (SIASG-Sistema Integrado de Administração de Serviços Gerais and SISME-Sistema de Minuta de Empenho) were related to the same drug-purchasing system and were grouped as one data source. Thus, the final selection consisted of 38 data sources, which underwent further characterization (Figure 1). Six rounds of discussion took place among the national health experts in order to achieve consensus and define the final list (Supplementary Table S2).

Figure 2 shows how the data sources were grouped. Twenty-three (60%) were classified as automated health care databases; four (11%) as EMRs; four (11%) as adverse event reporting systems; three (8%) as national surveys or datasets; and four (11%) as other types. The description of each data source, as well as the rationale for grouping it in a particular category, is provided in the supplementary material (Supplementary Table S3).

Based on the analysis of each data source, 18 (47%) were classified as “publicly and conveniently accessible online,” 15 of which (88%) were accessible through the DataSUS, with the Brazilian Ministry of Health as custodian. All publicly available online data sources provided national information; most of them had more than 5 years of coverage and both individual- and aggregate-level information. Twenty data sources (53%) were known to collect individual-level data, and three (PNAUM, SIA-SUS, and SIVEP-Gripe) were available for download. Table 1 displays the data sources, grouped by accessibility, geographic granularity, type, setting, and initial year of release. The detailed classification, which allows comparability among the data sources is provided in the supplementary material (Supplementary Table S4).

URLs for the “publicly and conveniently accessible online” data sources are shown in Table 2, as well as the file formats. The URLs for all data sources selected are provided in the supplemental material (Supplementary Table S5). Access through the FTP directory is provided for limited data sources and is also provided in the supplemental material (Supplementary Table S6).

In Brazil, six different ways of assigning codes to medicines were found. Drug coding is not uniform; each data source has its own coding system, depending on the purpose of the data. The drug coding systems employed in Brazil are shown in Table 3, with examples of data sources that use each system. This information was not available for all data sources, an indication of the need for further work on characterization.

The literature review was part of the initial process for mapping Brazilian databases. We found publications related to 23 of the 38 data sources, including reports, manuals, and other documents available online. Scientific articles had been published in national and in international journals. Examples of studies that used some of the selected data sources are presented in Table 4.

This study provides an overview of data sources that are used or have the potential to be used for DUR in Brazil. A total of 38 sources were identified, almost half of which are publicly available and provide national information. Nineteen sources collect individual-level data, but few provide it for download. Those classified as “other sources” were generally related to Ministry of Health administrative processes, as medicines purchases and prices. Further characterization to determine the types of research questions they might address is needed. In Brazil, six different ways of assigning codes to medicines are employed, none of which is recognized internationally. Brazilian data sources have the potential to answer research questions related to medication use, adherence to treatments, purchases, and safety. However, currently mapped sources comprise a mix of databases, of unknown quality, centralized by the national government, but decentralized in terms of research and their usability and purposes for decision-making and post-market surveillance.

Some of the data sources presented here had been used by researchers. Ali et al. have described the linkable databases currently available for evaluating health technology assessment in Brazil (Ali et al., 2019). For example, the CIDACS initiative uses SINAN, SIH-SUS, SINASC (Live Births Information System), and SIM (Mortality Information System) to assess outcomes of major social programs (Barreto et al., 2019). SIM (not included in our inventory because it presents only data related to ICD-10 codes for drug poisoning mortality (Mota et al., 2012)) and SINASC are important sources of data for evaluating health outcomes and indicators. The quality of data in both systems has improved over time (Szwarcwald et al., 2019; França et al., 2020); however, health outcomes of medication exposure (not related to poisoning) remain unexplored for most classes of medicines.

Junior et al. linked SIH, SIA, SIM, SINASC and SINAN (Guerra Junior et al., 2018) and created a National Database of Health for longitudinal studies. Freire et al. linked SIM and SIH, including information from APACs (Authorization of High Complexity Procedures of the Outpatient), provided by the SIA-SUS system, and were able to describe the trajectory of patients in the health care network, and cancer-related hospital admissions (Freire et al., 2015). In fact, the APAC reports are among the most important sources of information on medication dispensing in Brazil. However, the information pertains only to drugs dispensed free of charge; that is, only medications supplied by SUS under the APACs are recorded and available through DataSus systems. Moreover, the generation and consolidation of APACs to make the data available for DUR are complex. Few research groups have the expertise required to link the different data sources and prepare the data for longitudinal analysis (Soares and Silva, 2013).

Exposure to medications among the Brazilian population is complicated by the structure of health care delivery, where a private system co-exists with a public system, and no overall control is in place for dispensing most medicines. Consequently, only studies using data from APACs for biological agents, chemotherapy, and other high-cost medicines have the potential to correctly ascertain exposure (Prestes, 2017; Junior et al., 2018).

Other automated health care databases, some of which were identified by Ali et al. (Ali et al., 2019), could be valuable for DUR, but not without an extensive evaluation of the quality of the data they contain. Notable examples are Horus , Farmacia Popular and BNAFAR. Interfaces among the systems that generate these databases are known, but nothing is known about their quality, coverage, and completeness. These data sources, specifically the BNAFAR and the Horus, were not available for research (Ministério da Saúde, 2018). Infrastructure issues are familiar limitations, and at least partially explain why data on drug dispensing are so difficult to obtain in our country (Herrett et al., 2015; Hallas et al., 2017). Pharmacoepidemiology research perspectives in Brazil suffer constraints not due to lack of data, but to lack of linked data and cross-validated secondary data (de Castro, 1999; Junior et al., 2018; da Saúde, 2018).

The Sistema Nacional de Gerenciamento de Produtos Controlados (SNGPC) (Agência Nacional de Vigilância Sanitária, 2019), which monitors dispensing of narcotic and psychotropic medications, and since 2013, antibiotics, is an important data source for controlling the purchase and dispensing of medicines. An “open data” initiative launched by ANVISA has yielded data for DUR. The expectation is that data provided by ANVISA might allow assessing, for example, policy impact of medicines regulation. However, a complete characterization of these data sources for understanding the quality of provided data, and what research questions would be answered using the open data are still lacking.

SIVEP-Gripe and SI-PNI, among other automated health care databases (Table 1), record information on medication use, but the quality, temporality and feasibility for linkage of these data have not been adequately explored for DUR. SIVEP-Gripe is available and provides individual-level data, but the incompleteness of certain variables and lack of temporality in recording medication use, render the information useless for examining, for instance, the effectiveness of medication use. SIVEP-Gripe is an epidemiologic surveillance system that was designed for other purposes, but with properly recorded information, it could help answer important research questions and support other voluntary reporting systems in evaluating adverse drug effects (Melo et al., 2021). As well, non-prescription drugs recorded in surveillance systems such as SINAN, and SIVEP-Gripe are often taken during the onset of a disease—an upturn in sales may serve as an early indicator of an outbreak or epidemic (Das et al., 2005; Edge et al., 2006).

The Electronic Medical Record (EMR) of the Management Application for University Hospitals–AGHU currently covers 30 hospitals across the country (Ministério da Educação, 2019). It is the standard management system for all federal university hospitals provided by the Empresa Brasileira de Serviços Hospitalares (Ebserh) network and is a potential data source for DUR. University hospitals treat both in- and outpatients. The creation of a large cohort of patients receiving different levels of care would allow for follow-up of short- and long-term effects of medication on several outcomes. e-SUS AB might be used for the same purpose. However, no single DUR study was found to have used the Ebserh data.

We classified four data sources as adverse event report systems: NOTIVISA/VIGIMED, SINAN, SINITOX, and DATATOX. Recently, ANVISA published implementation of the VigiFlow (named Vigimed in Brazil) (Vogler et al., 2020) as a substitute for the NOTIVISA in an effort to enhance the usability of the national system. But no information is available about how different pharmacovigilance systems across the country could be integrated. In 2021, part of Vigimed aggregated data was available on the Anvisa website by drug, adverse reaction (MedDRA SOC/Preferred Term), severity, age group, gender, state of the case report, for example. Clinical trial reports are also recorded in the same database (Notivisa EC) but are not available given the need for data confidentiality.

Spontaneous reporting systems constitute a major resource for detecting adverse drug effects and have made important contributions to pharmacoepidemiology (Strom and Carson, 1990). Systems for active surveillance and projects for detecting signals and monitoring recently approved medications (Racoosin et al., 2012) have been established in other countries. Recent studies involving disproportionality analysis for safety signal screening in children (Vieira et al., 2020) and breast cancer patients (Barcelos et al., 2019) using Notivisa were conducted, demonstrating the potential of this data source. However, Brazil lags behind in terms of research initiatives and decision-making using automated administrative data.

The only national-level drug utilization study that has been conducted in Brazil was based on primary data (Mengue et al., 2016a). The National Survey on Access, Use and Promotion of Rational Use of Medicines (PNAUM) was a cross-sectional, population-based study focusing on urban households. Fieldwork was carried out between September 2013 and February 2014. In total, 41,433 interviews were carried out. The survey examined medication use for chronic health conditions. However, the PNAUM has not been repeated, and the cross-sectional data do not allow evaluation of outcomes. Also, this was the only study to collect population-level data about over-the-counter medication use. Currently, no information about over-the-counter is available in any of the automated databases (Arrais et al., 2016). Other important surveys (cross-sectional) were included in our inventory—PNAD and Vigitel—although their purpose is to assess other characteristics of the Brazilian population and do not provide medication details.

Brazil has no formal policy on setting priorities and using administrative data to evaluate the effectiveness and safety of medications. However, many systems contain information for managing logistics and drug expenditures. APURASUS, SIGAF and SIASG are used by different levels of government to control costs and transmit information from local systems to the national level to plan acquisition and distribution. For example, SIASG made it possible to explore expenditures, pricing and judicial demands for a variety of drugs and drug classes, and it has been important for decision-making about the incorporation of drugs in the national list and the sustainability of provision programs (Luo et al., 2014; Chaves et al., 2017; Chama Borges Luz et al., 2017; Magarinos-Torres et al., 2017; dos Santos Teodoro et al., 2017; Alves et al., 2018; Caetano et al., 2020; dos Santos Dias et al., 2020, 2009–2017; Matos et al., 2020). However, the safety profile of medicines and outcomes in the population cannot be examined with these data.

Despite efforts made by the Ministry of Health to harmonize the recording of information, health institutions’ data collection processes differ considerably. Because of the structure of the healthcare system, patients typically seek care from a variety of providers at several institutions with nonlinked electronic health record systems. Combining data from these systems is a challenge. One of the most important issues to emerge from this study is the lack of unique key identifiers for individuals. These factors, in addition to technological infrastructure and skilled human resource constraints, limit the usefulness of routinely collected data in generating evidence to support clinical and policy decisions and in answering epidemiological questions (Ali et al., 2019).

Another important finding is the heterogeneity of drug-coding systems in Brazil. Federal Law No. 9,787/99 requires that, within the scope of the SUS, purchases of medicines, under any type of acquisition, as well as medical and dental prescriptions for medicines, adopt the DCB (Brazilian Non-proprietary name) or, in their absence, the International Non-proprietary Name (INN). However, this does not apply to administrative databases. For each data source, it is necessary to know the types of codes that are employed, how they are constructed, and why they are used, but no clear definitions are provided.

The limitations of this inventory of Brazilian databases that contain medication-related information are mainly related to the design of the study and the difficulty of assembling a group of experts with an in-depth knowledge of each data source. We may have missed data sources and relevant studies. The literature search was conducted using the names of the data sources, but if a name was unknown, studies could not be found, and the data source was not included. Moreover, this is only an inventory; full characterization of each database has yet to be done.

The main value of this study is to provide an overview with a focus on data sources for DUR. The methodology used by the LatAm project may be highlighted as one of the main strengths of our study, an original multi-phase approach allowing to map national data sources for DUR. The next step is to fully characterize each database using pre-established checklists (Hall et al., 2012), and thereby provide information that will help researchers determine which sources may be may be useful for specific types of studies; what research questions can feasibly be addressed; how the data can be accessed; and what quality may be expected from the data.

Based on this comprehensive and structured inventory, we provided an overview of the several types of data sources for DUR in Brazil. Our findings demonstrated that a uniform system for drug classification, data quality evaluation, and the extent of population covered by year are lacking in the mapped data sources. National administrative health databases are provided mainly through the DataSus and contain information about the population covered by the SUS. Further work is required to assess the reliability of Brazilian data for DUR.