The minimum data set (MDS) is a collection of data elements to be grouped in a standard approach to allow the use of data for clinical and research purposes. An MDS is designed to capture essential data elements, which can be aggregated, at an individual level, or a combination of both, depending on the specific requirements of the registry and research objectives.

Data standardization allows for the accurate comparability of collected data and, consequently, improved generalization of findings [1]. Many countries use the MDS strategy in their health systems to standardize essential and fundamental data elements to properly record patient information and support public health planning and management. In addition, MDSs facilitate data interoperability between different health services that comprise a national health network [2]. The MDS strategy makes it possible to identify relevant health indicators in health information systems (HISs) that can serve as a basis for the definition of public policies that can influence the monitoring of diseases, organize the resources available, and improve social well-being [2].

With the high volume of fragmented data in different health services, a validated, stable, and safe MDS is essential to support the use of clinical-administrative elements and information to generate national health statistics. As noted by the World Health Organization (WHO), it can help elaborate models of service performance and patient satisfaction to meet the growing health demands of HISs worldwide. Therefore, the development of HISs relevant to public health should be able not only to generate data but also understand and manage the produced information. They should also provide indicators to represent the true health context of a given population and provide subsidies to monitor the quality of treatments offered by health services [3].

However, health data are typically voluminous, complex, and sometimes ambiguous to generate indicators that can provide knowledge and information on health. Thus, raw data must be ordered, interpreted, and transformed into information [4] that must be processed and analyzed. Many digital tools are used for processing and analysis, including machine learning algorithms, artificial intelligence, neural network applications, big data, computational ontologies, and semantic web [5].

This complexity extends further to the rare disease (RD) domain. RDs are pathologies with a low prevalence in the general population, with <50 cases per 100,000 individuals [6]. Although they affect a low percentage of people individually, such diseases are numerous, and together they can affect up to 10% of the world population; thus, they significantly impact health systems [7]. For many RDs, there is no well-structured knowledge about their diagnoses and treatments, increasing the demand for MDS strategies [8].

Many countries have initiated national plans to promote care, research, and technology in RD. These plans focus on enabling health managers to improve the services provided in a contextualized way [9]. In addition, these proposals can increase the accuracy of health decisions and reduce the fragmentation of large volumes of data, creating a solid base of information pertinent to diagnoses, treatments, and processes [10].

Owing to the complexity of the areas of knowledge related to RD, initiatives have been developed to provide informational support to health networks that provide services, care, and research in RD [11]. The best known is Orphanet, a networked platform comprising researchers from European countries. It is funded by the European Commission to increase the available knowledge base of RDs to improve care processes in this domain. Orphanet promotes and provides a structured and comprehensive database of information and knowledge about the RD domain. It also offers a validated ontology with a high standard of quality consistency and manual data auditing performed by specialists [12].

In Brazil, the Brazilian Policy for Comprehensive Care for People with Rare Diseases of the Ministry of Health of Brazil established reference services in RD that offer preventive, diagnostic, and therapeutic actions for people with rare conditions [13]. There is a need for human, technological, and infrastructure resources in the Brazilian Unified Health System, which leads to management and communication problems such as difficulties in transmitting information between the services that comprise the health network, leading to different processes and costs across the public system [14].

MDS is essential for health surveillance, providing services, and generating recommended population indicators. However, the national policy in Brazil does not support the structured use of validated MDS in health services [15]. To address this issue, the Brazilian National Network of Rare Diseases (in Portuguese: Rede Nacional de Doenças Raras [RARAS]) was designed to create an epidemiological surveillance structure. This network encompasses all reference services for RDs enabled in Brazil and brings together university hospitals and reference services for neonatal screening from all regions of the country [16].

The RARAS project is conducting the first nationally representative survey in Brazil on the epidemiology, clinical scenarios, diagnostic and therapeutic resources, and costs related to individuals with RD of genetic and nongenetic origin. Project managers developed their own data collection protocol in the absence of a standardized global MDS reference. Although we identified this information bottleneck in Brazil, the international literature describes a global problem with data collection, recording, and structuring in RD [17].

This study aimed to identify and analyze the MDSs used for RD in health care networks worldwide and compare them against WHO guidelines. The secondary objectives were to verify MDS implementation and study quality, map the collection of data elements, suggest a global fundamental MDS for RD, improve diagnostic and care processes, and optimize public planning and decision-making.

The population, concept, and context (PCC) methodology proposed by the Joanna Briggs Institute was used to define the research question of this systematic review. The PCC strategy is recommended as an alternative to the population, intervention, comparison, and outcome model for investigations without well-defined clinical intervention [18]. A total of four research questions were defined, preceding our central question as follows: (1) Are there standardized MDSs for RD patient records in health networks worldwide? (2) What are the data elements of each MDS? (3) What can we assess of their usefulness? and (4) Can a fundamental MDS be developed for RD networks?

These questions were formulated to help answer the following central question: “What is the minimum data set used in registries for RDs in health networks?”

To improve the transparency and reproducibility of the review, we registered it in the Prospective Register of Systematic Reviews international database under the identifier CRD42021221593 [17].

The inclusion criteria were research papers (eg, full texts and conference papers), national plans, policies, industry reports, position papers, and program reports. Studies published in Portuguese, English, and Spanish were included. Owing to the specificity and novelty of the proposal, we included publications from any time. Only studies that fully described MDS for RD and its practical implementation as a health strategy were included. Furthermore, only studies that answered our research question and provided clear evidence on the subject were included.

Publications without the scientific rigor to answer our research question using clear and objective evidence were excluded from this systematic review. We excluded personal editorials, studies using only theoretical approaches without practical implementation (such as studies defining operational or predictive models), and social media publications. Studies published in languages other than those listed in the inclusion criteria were also excluded. In addition, studies that used an MDS for RD but did not describe all the data elements that comprised the MDS were excluded.

The concepts used to create the search strategy came from PCC methodology and were adapted using Medical Subject Headings, the vocabulary thesaurus of the National Library of Medicine used for indexing articles, and metadata for health sciences [19].

A total of 4 databases were reviewed: PubMed, Scopus, Web of Science, and Literatura Latino-Americana e do Caribe em Ciências da Saúde (LILACS). We also searched for documents from the WHO and government websites for additional studies. These databases were chosen because of their robustness and relevance in the clinical and health arenas. Table 1 lists the keywords and search strings entered into the search strategy for different databases. Logical connectors were used to assign the necessary precision and refine queries.

The Google Scholar database was also used for manual searches and searching for references to theses and dissertations. These documents are part of the gray literature because they are not published in commercial media. Searching through such sources can reduce publication bias, increase the comprehensiveness and timeliness of reviews, and provide a balanced picture of available information [20]. RD clinical experts participated in the evaluation process.

We aimed to combine the findings of similar studies into subcategories using a consensus among specialists. We did not exclude studies with inadequate quality from the data synthesis. Descriptive statistics were used to report the frequency and proportion of the outcome measures.

The subgroup analysis method was used to describe the possible reasons for the heterogeneity of the data extracted from the selected studies and to facilitate the synthesis of information via tabulation and visual arrangement so that we could later carry out the necessary analyses.

When considering the reporting method of a single instrument, proportions indicating the percentage of studies using that instrument were calculated. Next, the selected resources were analyzed and interpreted. The main contents related to the research objective were classified as RD minimum data elements. The sections were then classified into 2 general categories, management data and clinical data, which is an efficient method for categorizing health data [27].

We identified 407 studies in the initial database search phase. After removing duplicates, 337 articles remained for the title and abstract screening stage, 290 of which were excluded because they were irrelevant to this review. We evaluated 47 full-text articles based on the inclusion criteria. Of these, 19 articles were excluded because they did not answer the stipulated research question, and 8 articles were excluded because they only mentioned an MDS strategy for RD and did not describe the data elements or provide information about implementation in practice. The detailed process is shown in Figure 1.

A total of 20 unique studies were included in this systematic review. The general characteristics of these studies are described in Table 2, organized by the first author’s given names in alphabetical order. The specific characteristics of these studies are described in Multimedia Appendix 1.

Of the 20 selected studies, 70% (n=14) referred to a specific health domain and the remaining 30% (n=6) referred to RD in general. Of the studies that referred to a specific domain, 2 studies were dedicated to the composition of MDS for juvenile dermatomyositis [40,41], 2 to metabolic disorders [36,45], 1 to nonmalignant RD [33], and 1 to undiagnosed RDs [28]. Other specific studies included mucopolysaccharidosis type I [34], homocystinuria and methylation defects [42], primary immunodeficiencies [29], cystic fibrosis [38], congenital hypothyroidism [39], adenosine deaminase severe combined immunodeficiency [35], atypical hemolytic uremic syndrome [30], and inflammatory rheumatic diseases [46], with 1 study each.

The remaining 6 studies that presented MDSs focused on the domain of RD in general [2,31,32,37,43,44], that is, they used data elements capable of representing patients with any RD in its health context. However, even these studies have demonstrated limitations and restrictions because of their geographic coverage and the particular local characteristics of the regions in which they were applied. As mentioned in the Discussion section, the proposed fundamental MDS seeks to address these limitations.

All the data collected and compiled, as well as their occurrences in the MDSs of the studies included in this systematic review, have been described in Multimedia Appendix 2.

In the eligibility category, the term “patient consent” appeared the most; of the 17 terms in this category, 13 were unique. In the identification category, “patient’s name” was the most common term, and 5 unique terms were observed among the 25 included in this group. Of the 148 terms in the diagnostic category, “diagnosis” was the most frequent (appearing 11 times), and 32 were unique. In the treatment category, “date treatment started” was the most common term, appearing 6 times; of the 72 terms in this category, 29 were unique. In the medical consultation category, “anthropometric data” was the term that appeared the most; of the 180 terms, 78 were unique.

Of the 17 terms, “disease malignancy” was the most common in the comorbidity category, with 8 unique terms. In hospitalization, of the 5 terms, “history of hospitalization” appeared most frequently (3 times). In the examination category, “general and specialized laboratory tests” were the most common among the 42 terms, of which 29 were unique. In the outcome category, “date of death” was the term that appeared the most, comprising 11 of the 57 terms, and in the others category, “the patient having previously provided a biological sample for research” was the most frequent term, comprising 2 of the 26 terms used in this category.

By combining these findings with the extraction, categorization, and synthesis techniques mentioned earlier, we used data science methods and clinical experience to design, structure, and recommend a fundamental global MDS for RD patient records in health care networks. It aims to comprehensively cover the data needed in clinical and management contexts. These summarized results can be found in Multimedia Appendix 3.

The results of the risk of bias evaluations for the 20 included studies are presented in Multimedia Appendix 4. We reported the risk of bias evaluation using the STROBE statement standards and checklist [25]. The recommendations regarding the numerical indices can be found in tab 2 of Multimedia Appendix 4, STROBE statement recommendations. We used the “Unclear risk of bias” classification for cases where the recommendation did not apply to the study.

Using a matrix of 440 cells generated by crossing the 22 STROBE recommendations with the 20 selected studies, we found that 48.4% (213/440) of the items evaluated in all studies showed a low risk of bias and 39.1% (172/440) of these items had an unclear risk of bias. The remaining 12.5% (55/440) of the items assessed across all the studies had a high risk of bias.

In addition, the STROBE recommendations with the highest number of low risk of bias assessments for this set of studies were item 5, “Setting,” and item 14, “Descriptive data,” with 17 occurrences. The recommendation with the highest number of unclear risk of bias assessments was item 12, “Statistical methods,” with 14 occurrences. The STROBE recommendation that received the highest number of high-risk bias assessments was item 3, “Objectives,” with 12 occurrences.

We carried out this systematic review to elucidate the MDSs used for RD in health care network records in health care systems worldwide. The results showed a lack of terminological standardization of the concepts used in these MDSs. Different health systems may use different terms for the same concepts, hindering the interoperability, integration, and sharing of highly relevant information and knowledge to describe phenomena and generate health indicators.

The WHO recommends that health systems and networks use standardized terminology to exchange data, information, knowledge, and intelligence in health. Thus, health observatories that integrate population and epidemiological data require standard semantics to collect and organize this information to enable the generation of indicators capable of positively influencing public health policies [47]. Orphanet’s initiative also points to the need for a formal vocabulary to map and classify processes in RD. The lack of structured knowledge about RDs leads to problems, such as increased waiting time for a diagnosis and, consequently, difficulties in establishing an adequate treatment for these patients [48,49].

We encountered difficulties in the standardization and classification of the MDSs findings. Notably, the selected studies did not follow the same standard structure for classifying the data from their MDSs. Thus, we classified all data into 10 categories.

This process allowed us to summarize the information, verify its frequencies, perform statistical analysis, and analyze and identify synonyms (items with the same meaning, but described using different terms in specific MDSs). Clinical experts in RD and information and data scientists met in groups to verify and analyze these synonyms.

Other studies have proposed MDSs for specific groups of RDs [28-30,34-36,38-42,45,46], populations, and territories [2,29,33,43,44] based on different methodologies. For example, previous studies have proposed MDSs for RD in national and continental contexts. The European Commission’s Common Data Elements for Rare Disease Registration was developed for use by the RD registries across Europe. Similarly, the French national MDS was built nationally and defined through a national consensus for use in French RD centers [2]. However, to our knowledge, this is the first study to propose a data set for RD based on a systematic review that mapped MDSs used in different studies and could allow the application of the resulting data set to all RDs in unspecified populations.

Europe also has some examples of registries developed nationally and applied to an international consortium, such as the Fabry Disease Registry [50] and REGISTRY (for Huntington disease) [51]. However, neither provides open access to information regarding the data elements and structural evaluation. This limits researchers and practitioners who are not consortium members to access and adapt the registry to their context. In addition, these registries may limit the generalizability of the results to the entire affected population, because registered patients tend to have more access to medical care and treatment centers than the general population, especially in rural regions or countries with less developed health systems.

The Italian National Rare Diseases Registry [52] and Italian Neuromuscular Registry [53] are nationally developed registries intended to collect information on the prevalence and geographic distribution of rare and neuromuscular disorders, respectively. Both are considered effective models; however, their representativeness and generalizability of their data to other populations and countries may need to be improved.

A Spanish study addressed the importance of linking data to strengthen national RD registries [54]. However, the results were based on only 1 RD (amyotrophic lateral sclerosis, a motor neuron disease) and cannot be generalized to other diseases or datasets. In addition, this study highlights that coding errors and other inconsistencies may have affected the validity of the results. This is an important observation, because the accuracy of the results depends on the quality and consistency of the data. To address these limitations, researchers may adopt measures such as cross-checking data, manually reviewing selected cases, and statistical analysis. This evaluation was possible because the authors performed an association of records from different data sources, which occurred exclusively in this study.

Other studies have developed MDSs for RD designed for international applications; however, the MDSs were determined by a national committee. This is the case for the National Institutes of Health, National Center for Advancing Translational Sciences, and Global Rare Diseases Patient Registry Data common data elements [55]. All these studies present MDSs with limitations or restrictions because they cover only a specific RD, or their scope of application is restricted to a specific context or geographic region.

Therefore, the most innovative aspect of this study is the compilation of all the data used in MDSs for RD published worldwide. The fundamental MDS developed based on the analysis of this compilation by clinical and health data specialists can benefit services, assistance, research, and management in RD. Most national health systems and RD policies can also contribute to developing methods and processes and producing information. Thus, defining an MDS may improve data reliability to align strategies to enhance and manage health planning [11]. Although many policies describe comprehensive care in a network without a structured and standardized MDS, health centers cannot work in a coordinated manner.

Thus, the literature indicates the need to identify and implement common data elements [56] to improve care quality [57] and enable collaboration across different health care systems [58]. Our findings can help identify common standards and data elements to minimize the duplication of efforts and enhance the quality of patient records and data. This would lead to greater effectiveness of HISs and consequently, better patient outcomes. One of the recommendations for improving the quality of RD registries is to facilitate harmonization among the many institutions that collect patient information. Thus, it is essential to plan, design, and set up an MDS-based national information system and database for RDs that can provide and evaluate health indicators, promote network research, and foster public policies for RD care [11].

MDS is usually developed by reviewing the scientific literature, consulting with experts, and considering existing guidelines and recommendations. Although some of our findings report that they followed these processes in a combined or individual way, all were performed considering only a specific geographic context or addressed aspects related to a condition or subgroup of RD. Although a systematic review is an approach to synthesizing comprehensive scientific evidence, developing MDS involves reviewing the scientific literature, consulting with experts, and considering existing guidelines to identify essential data for RDs. Our systematic review approach provided a rigorous and transparent methodology to ensure the reliability and validity of the information collected and synthesized.

Identifying common trends and patterns can further simplify and streamline the information-gathering process, avoiding the need to develop and implement different data sets for each RD. This mapping saves time and resources, allowing for more efficient data collection. Sharing resources and knowledge plays a key role in supporting the research and development of therapies for rare conditions.

Variations in documentation practices in health systems can also lead to inconsistencies in data collection and standardization related to RD. These variations may include differences in the terminology used, coding systems, and data recording policies. Gaps in capturing these essential data elements can arise because of a lack of consensus on the most relevant information for understanding and studying RD. Existing data sets do not encompass all the aspects necessary for a comprehensive investigation, leading to the need for an MDS that identifies and incorporates essential data elements for scientific research in this field. This will allow for better comparison between different studies and registries.

Our study has some potential limitations. Among these, we highlight the methodological aspects of choosing the quality assessment method, publications, and generalization bias. In addition, we emphasize the potential for further investigations and possible outcome variations in different populations, contexts, and types of RD.

Owing to the originality and nature of the proposal, we did not define restrictions regarding the study design in our inclusion and exclusion criteria. Thus, all study designs were considered for inclusion in this systematic review. Therefore, selecting a method for assessing the quality of these studies is a nontrivial task, as such methods are usually designed to assess specific types of study designs. To mitigate this limitation, it was necessary to adopt the STROBE method in our proposal so that we could assess the quality of each of the selected studies using a standard tool. In addition, in the sequence, we cite possible biases and describe how to mitigate them.

This review may be affected by publication bias, as studies with positive results are more likely to be published than those with negative or inconclusive results. This can lead to a distorted view of the existing MDSs for RD. Therefore, we took several steps to reduce the risk of publication bias affecting our results, such as searching for multiple sources, recording the review, and contacting specialists. Recording the review process increases transparency, reduces the risk of publication bias, and promotes a more rigorous and impartial evaluation of available evidence. This helps avoid omitting relevant studies or data that may not align with the desired results. By keeping a comprehensive record, reviewers and readers can assess the potential impact of any publication bias and ensure that the review is conducted objectively and impartially [59].

Measures adopted to minimize the risk of excluding studies that were not published in scientific journals included the use of gray literature databases, prepublication repositories, clinical trial records, and scientific conferences. Finally, we contacted experts in the field to identify unpublished or ongoing studies that were not found in the systematic search.

Possible biases that may affect the generalizability and comparability of the findings must also be mentioned. Although there was no specific tool to mitigate the issue of generalization bias, it is worth highlighting that the studies included in this analysis were selected based on clear and well-defined criteria. During the data analysis process, we considered the differences between studies, such as population characteristics, study design, and data quality. Ultimately, our goal was to present the findings of our work clearly and transparently using valid instruments from the literature (Multimedia Appendix 5). We also aimed to highlight the limitations of our study and identify potential variations in the results based on robust evidence to provide valuable and relevant information for clinical practice and health decision-making.

Our work discusses the difficulties in standardizing and classifying findings from MDSs for RD because of differences between studies. Clinical experts and data scientists have defined categories based on WHO guidelines to address this issue. This study aimed to compile MDS data from around the world and suggest a fundamental MDS for use in RD services, assistance, research, and management. The fundamental RD MDS designed in this study comprehensively covers the data needs in the clinical and management sectors.

The results can also help public policy makers achieve other aspects of their policies. For example, analyzing the state data produced by HISs is essential for qualifying and quantifying the care provided to people with RDs. It can also allow comparisons with local data regarding preventive actions provided to a community. This can improve decision-making at the managerial and local levels and contribute data that can inform strategic decisions at the national level [60].

In conclusion, the solid base of information compiled regarding MDSs for RD is a technical and social contribution to improving the health network’s ability to map its demands and better understand the public health scenario regarding rare conditions. Although the proposal of a fundamental MDS for RDs is highly relevant and, to our knowledge, unprecedented in the literature, we also suggest collecting data elements to be used in addition to this fundamental MDS, if necessary, for each group or type of RD, to increase the completeness and specificity of the data structure.

Owing to the high complexity of care processes involving RDs, structured information can significantly impact the quality of services offered to the population. A curated description of the methodology for developing an MDS for RD in low- and middle-income countries has not yet been published. We encourage further research in this context.

National data gathering for RD based on standard data sets to encourage interoperability by disseminating agreed-upon data-sharing guidelines can facilitate semantic data standardization. On an organizational level, it can assist institutions in establishing a registry, sharing deidentified data with research networks, and building specific and rich databases. On the basis of these results and the proposal of a fundamental MDS, we aimed to provide evidence-based subsidies to assist managerial and clinical RD processes in health systems.