Factors influencing harmonized health data collection, sharing and linkage in Denmark and Switzerland: A systematic review

Introduction The digitalization of medicine has led to a considerable growth of heterogeneous health datasets, which could improve healthcare research if integrated into the clinical life cycle. This process requires, amongst other things, the harmonization of these datasets, which is a prerequisite to improve their quality, re-usability and interoperability. However, there is a wide range of factors that either hinder or favor the harmonized collection, sharing and linkage of health data. Objective This systematic review aims to identify barriers and facilitators to health data harmonization—including data sharing and linkage—by a comparative analysis of studies from Denmark and Switzerland. Methods Publications from PubMed, Web of Science, EMBASE and CINAHL involving cross-institutional or cross-border collection, sharing or linkage of health data from Denmark or Switzerland were searched to identify the reported barriers and facilitators to data harmonization. Results Of the 345 projects included, 240 were single-country and 105 were multinational studies. Regarding national projects, a Swiss study reported on average more barriers and facilitators than a Danish study. Barriers and facilitators of a technical nature were most frequently reported. Conclusion This systematic review gathered evidence from Denmark and Switzerland on barriers and facilitators concerning data harmonization, sharing and linkage. Barriers and facilitators were strictly interrelated with the national context where projects were carried out. Structural changes, such as legislation implemented at the national level, were mirrored in the projects. This underlines the impact of national strategies in the field of health data. Our findings also suggest that more openness and clarity in the reporting of both barriers and facilitators to data harmonization constitute a key element to promote the successful management of new projects using health data and the implementation of proper policies in this field. Our study findings are thus meaningful beyond these two countries.


Introduction
Technological advances made over the past few years have increased the digitalization of medicine, thus leading to a considerable growth of clinical, research and public health datasets. These data sources are increasingly related to the big data environment and they include, amongst others, genomics and other-omics related-data collections, electronic health records (EHRs), patient registries, medical imaging, administrative data and clinical trials data [1,2]. However, a good part of such datasets are often kept and analysed in silos and not adequately shared [3]. If properly integrated into the clinical life cycle, such collections of data stand to offer a unique opportunity to drive scientific discoveries and improve healthcare research. For example, they may allow a better understanding of the aetiology of illnesses and subsequently help in improving the management, prevention and treatment of diseases [1,2]. This is even more promising in the framework of learning healthcare systems, where clear boundaries between research and care are dissolving and the same data are used both for improving scientific knowledge and providing better care [4].
In this context, developing the harmonization of health data-described as the sum of all "efforts to combine data from different sources and provide users with a comparable view of data from different studies" [5]-is crucial to improve clinical research and practice. Such standardized approach requires not only better quality, re-usability and interoperability of data, but also more open and collaborative communication between the different stakeholders active in the health data environment [6]. The fact that a good percentage of healthcare spending are being wasted as a consequence of under-exploiting data potential in several healthcare systems around the world [7-9] should be considered as one important factor urging for such changes to happen. Harmonized health datasets are also laying the foundation of a new era of biomedical research, where three concepts are currently converging, namely precision medicine, learning healthcare systems and implementation science [7,10].
The harmonization of health data is a complex procedure which involves significant changes in how data are collected, shared and linked. Harmonization can be either prospective, when modifications occur in the study design to subsequently render the pooling of data more straightforward, or retrospective, when pooling is performed with data collected previously according to different study designs [11]. In practical terms, harmonization can be achieved through two distinct but complementary approaches, namely a "stringent" and a "flexible one" [12]. By means of a "stringent" approach, data are harmonized through the use of standard collection tools and standard operating procedures, implementable only in a Informatics" [Mesh] OR "Medical Record Linkage" [Mesh] OR "Information Dissemination"[-Mesh] OR "Data Integration" OR "Data Sharing") AND ("Switzerland" [Mesh]) [filters used are Articles types (Clinical Study, Clinical Trials (including controlled and Phases I to IV), Comparative Study, Evaluation Studies, Journal Article, Multicenter Study, Observational Study, Pragmatic Clinical Trial, Randomized Control Trial, Technical Report and Validation Studies), language (Danish, English, French and German) and species (Human Studies)]. We did not include harmonization as an imperative component in our search strategy since the exact boundaries of this concept are still controversial [25] and the addition of the term "harmonization" as an imperative component drastically reduced the number of publications for each country.
Eligibility criteria for this study were: (i) publications based on health data collection, sharing or linkage projects. There was no restriction on study design and type, i.e. qualitative, quantitative or mixed method studies, and clinical or observational studies were included; systematic reviews were excluded; (ii) there were no restriction on age, gender, disease and ethnic group of participants involved in these studies; (iii) the studies had to involve some health data collection, sharing or linkage at cross-institutional, cross-national or cross-regional levels in at least one of the two countries; (iv) only English, French, German and Danish language articles were included, and (v) publication year of articles ranged from January 2008 to March 2019.

Data extraction and quality assessment
The literature search results were catalogued on EndNote TM X8, a reference manager software. The titles and abstracts of all articles were screened independently by two authors (LDG and AM). The full-texts of the included publications were reviewed by LDG and AM to ensure that they met the eligibility criteria to be included in the systematic review. LDG and AM performed independently the data extraction from the included articles through a standard data extraction form developed progressively by the authors of this review. Additional publications gathered through reference screening went through title and abstract, and independent fulltext screenings and data extraction by MCM. Another review author, TW, validated randomly twenty percent of the publications reviewed by LDG, AM and MCM, to assess the quality of the data extraction process. A disagreement level of less than 10% for the data entries was considered acceptable.
The data extraction form included (i) study information (author(s) and publication year), (ii) sources of health data, (iii) cross-institutional or cross-national nature of the study, (iv) presence or absence of primary and secondary health data collection, analysis and sharing, and lastly (v) the categorization of barriers and facilitators to harmonized health data collection, sharing and linkage. The sources of health data were categorized as having three standard origins, namely health services, public health and research [26]. Other sources of health data falling outside these three categories were classified in a residual category ("Other").
LDG and AM performed a categorization of the identified barriers and facilitators separately, and came to consensus on the final categorization of these elements for accuracy and inclusiveness. Disagreements were solved with the mediation of TW. The identified barriers and facilitators to harmonized health data collection, sharing and linkage were subsequently clustered into main categories, which were then sub-clustered into smaller categories to highlight the most common barriers and facilitators in these main categories (the full clustering/ sub-clustering of barriers and facilitators is shown in Table 1). For the purpose of this systematic review, we defined harmonization techniques as methods which would allow the coherent pooling of different data sources, involving health data collected either prospectively, retrospectively or both. Examples include the use of standard case report forms or data dictionaries, for the period of January 2018 to March 2019 (upon request of the journal) resulting in a total of 425 additional papers. The result of the two searches were combined for each stage of the PRISMA resulting in an overall total of 2353 papers retrieved for the period of January 2008 to March 2019 (Fig 1). Duplicates (n = 170) were removed either automatically using ENDNOTE X8 or manually after reviewing abstracts and their titles. The remaining 2183 papers went through title and abstract screening, which resulted in the exclusion of 1789 papers. In-depth full-text screening was performed for 394 papers, and 115 more papers were excluded for not meeting the inclusion criteria (see Fig 1 for reasons). Reference screening of the 279 included papers, resulted in the identification and inclusion of 66 additional papers which met the eligibility criteria for this systematic review (Fig 1).
The 345 included papers are summarized in Table 2, where they are categorized based on their national (n = 240) or cross-national (n = 105) dimension, their sources of health data and the total number of barriers and facilitators reported for each project. We identified 200 Danish and 40 Swiss national projects, and 105 cross-national projects. Among these crossnational projects, 14 projects involved the use of health data from both Denmark and Switzerland, 51 and 40 projects involved a Danish partner and a Swiss partner respectively. Overall, the number of projects which involved primary health data collection, sharing and analysis were 106 (30.7%), 92 (26.7%) and 106 (30.7%) respectively. Comparatively, the number of projects which involved secondary health data collection and analysis were 283 (82.0%; if a study collected both primary health data and secondary health data, it was counted for both). Of the 345 projects, 199 used health data from health services, 211 from public health sector, 94 from research and 62 from other health data sources. Factors influencing harmonized health data collection, sharing and linkage in Denmark and Switzerland

Overview of barriers
Barriers of an ethical nature were reported 19 times in the included records and they concerned mainly issues related to privacy (n = 9) and respect for autonomy of study participants (n = 6) ( Table 3). As to legal barriers, these were reported 17 times and they included issues associated with national data protection regulations (n = 4), differences in national legislations concerning data security and privacy (n = 4) and "Other" (n = 9) (e.g. legal uncertainty concerning health data collection or sharing, market restriction, etc.). Overall, the type of barriers that were more often reported, however, were those of a technical nature. In the records, 416 technical barriers were mentioned and they were classified as data quality issues (e.g. data incompleteness, potential misclassification of data, etc.) (n = 234), lack of data standards (data structure and semantics, e.g. ambiguous terminologies, temporal evolution of data standards, etc.) (n = 151), limited technical capabilities (e.g. no unique identifier, etc.) (n = 21) and "Other" (n = 10) (e.g. time constraints on physicians preventing the use of standard procedures for data collection). Financial barriers were also reported, but only a limited amount of times (n = 9), and they were principally referring to the unavailability or inadequacy of financial support (n = 8). Only 13 political barriers were found and they comprised institutional/constitutional organization issues (e.g. federalist system and different healthcare systems) (n = 6), mistrust between stakeholders (n = 3), data ownership issues (n = 2) and "Other" (n = 2) (e.g. no official guidelines for data sharing). Studies also reported some motivational barriers, including lack of research incentives (n = 17) (including additional workload imposed on physicians/researchers), data re-use prevented by stakeholders as they are deemed unfit for secondary use (n = 2), stakeholders' competing interests (n = 2) and additional barriers of a diversified content, thus labelled as "Other" (n = 4) (e.g. study participants not showing up for part of the study). Finally, 6 socio-cultural barriers were reported in the included records, half of which were related to a "cultural clash" (n = 3), which we defined as issues resulting from different cultures in data collection, sharing and linkage of the partners involved in the project.  Table 3 shows the distribution of barriers' sub-clusters in national and cross-national Danish and Swiss projects. As such, single-country and multi-national countries are not differentiated. a N is the total number of projects in each country category b n is the total number of reported barriers per sub-cluster c-is the absence of reported barriers per sub-cluster N/A-Not Applicable https://doi.org/10.1371/journal.pone.0226015.t003

Barriers Countries involved in projects
Factors influencing harmonized health data collection, sharing and linkage in Denmark and Switzerland

Overview of facilitators
Facilitators of an ethico-legal nature were reported 582 times in total, and they were classified as official/legal approval of study (e.g. Danish Data Protection Agency) (n = 148), ethical approval by a REC/IRB (n = 135), legislation permitting to proceed with health data collection, sharing and linkage without consent or REC/IRB approval (n = 79), obtaining informed consent from participants (n = 69), health data anonymization (n = 58), the presence of legislation requiring mandatory reporting (n = 41), confidentiality measures (n = 29; e.g. data security audits), project done according to international laws and regulations (n = 8), data access rights for patients (n = 4), clear legislation for data collection, sharing or linkage (n = 3) and "Other" (n = 8) (e.g. study data made available by researchers upon request). Facilitators of a technical nature were reported 981 times in total, which were grouped in three categories, namely techniques for data harmonization (n = 798), data linkage (n = 155) and "Other" (n = 28) (e.g. study allowed the creation of optional and mandatory datasets, whereby a minimum of data are classified as mandatory). Facilitators of a financial nature, especially explaining how funding was successfully secured, were mentioned 12 times. These referred, for example, to public-private partnerships, where both partners would gain some benefits from the collaboration, as a solution for funding issues (n = 3). 169 facilitators related to politics were reported. These referred to the structure of the health system as an advantage for harmonized health data collection, sharing and linkage (n = 139), data access control by the players (n = 11), the presence of a data sharing agreement between the stakeholders (n = 9), building and maintaining stakeholders' trust for collaboration (n = 7) and "other" (n = 4). There were 14 motivational facilitators, which included monetary incentives to incite researchers/stakeholders to abide by standardized procedures for data handling and management (n = 7), improved data collection tool to ease the workload of researchers/stakeholders for data collection/sharing (n = 3), a memorandum of understanding between partners to ensure collaboration till end of study (n = 2) and "other" (n = 2). Lastly, there were 8 socio-cultural facilitators, which included data subjects controlling access to their data (n = 4) and "Other" (n = 4) (e.g. transparent policies for the participants). Country-wise distribution for all six facilitators categories are presented in Table 4.

Barriers and facilitators identified in national Danish and Swiss projects
When considering only national projects (n = 240) involving either Denmark (N = 200) or Switzerland (N = 40) alone, there were 323 identified barriers and 1234 facilitators. Technical barriers and facilitators were most frequently reported. For comparison purposes and compensation for the imbalances in the number of national projects identified in each country, the absolute numbers and the number of barriers and facilitators per 1,000 national projects for each country are illustrated in Table 5. Interestingly, the only identified category of barriers which was comparatively almost equally reported in Swiss and Danish single-country projects was that of technical barriers. Otherwise, ethical, legal, financial, motivational and socio-cultural barriers were reported 5.0, 3.3, 3.3, 5.7 and 10.0 times more in Swiss projects than in Danish projects respectively. On the contrary, a Swiss project reported on average more facilitators than a Danish one (only financial facilitators were reported equally in both countries). Ethico-legal, technical, motivational and socio-cultural facilitators were reported 1.2, 1.3, 4.2 and 1.3 times more in Swiss projects than in Danish projects respectively. Only facilitators related to politics were reported 4.5 times more in Danish projects than Swiss projects.

Barriers and facilitators identified in cross-national Danish and Swiss projects
With respect to cross-national projects (n = 105), there were 182 identified barriers and 532 identified facilitators. Technical barriers and facilitators were more frequently reported than those of another nature. For comparison purposes and compensation for the imbalances in the number of cross-national projects involving each country, the number of barriers and facilitators per 1,000 cross-national projects was calculated (excluding cross-national projects involving both countries) and illustrated in Table 6. Factors influencing harmonized health data collection, sharing and linkage in Denmark and Switzerland

Facilitators Countries involved in projects
Concerning cross-national projects, we observed a reverse tendency as compared to national projects. Studies involving a collaboration with a Swiss partner have, on average, reported 1.8 times less barriers than those involving a Danish partner. More in detail, projects including Switzerland reported 4.7, 1.9, 1.6, 3.1 times less barriers of an ethical, technical, financial and political nature respectively, than those with a Danish partner. However, crossnational projects involving a Swiss partner, reported 2.1 times more barriers of a motivational nature than those with a Danish partner. Comparatively, cross-national collaboration involving either a Swiss or Danish partner reported almost the same number of facilitators. Ethicolegal and political facilitators were identified 1.1 and 2.4 times more in cross-national projects with a Danish partner as opposed to cross-national projects involving a Swiss one. However,  Factors influencing harmonized health data collection, sharing and linkage in Denmark and Switzerland technical facilitators were identified 1.2 times more in cross-national projects with a Swiss partner than in those with a Danish one.

Discussion
This systematic review provides a comprehensive overview of projects from either Denmark or Switzerland which involved the collection, linking or sharing of data and of the barriers and facilitators related to the usage of health data therein reported. Our study includes a broad range of projects relying on data from different sources and contexts (health services, public health, research and other) and it confirms that studies involving the harmonization, linking or sharing of health data still encounter a high number of obstacles, but also underscores that barriers have prompted the development of numerous solutions. We will here address and discuss the findings related to barriers and facilitators of each cluster that was identified.

Ethico-legal barriers and facilitators
Although ethico-legal factors are often described as some of the most problematic elements when it comes to linking and sharing health-related data [9, 373,374], our results show that barriers of this nature are rarely reported. The small amount of ethico-legal barriers identified might either mean that such barriers were rarely present or that they were present but underreported. In our view, the latter option is more probable for at least two reasons. Firstly, as the records included in this review were all published articles, the explicit mentioning of ethicolegal complications might have been avoided to bypass problems related to publication. Secondly, ethico-legal factors are often less tangibile and transparent in comparison-for examplewith technical ones [15] and they are thus more likely to be superseded. Moreover, underreporting would confirm that ethico-legal aspects related to processing of health data are still underappreciated, which is a major obstacle to the final success of research projects [2]. This also suggests that there is some resistency by authors to openly disclose and discuss ethicolegal problematics. For the future, a less cautious approach would be much more beneficial, since it would allow new research projects to build on the issues encountered by old ones. Ethico-legal facilitators were more widely mentioned. Results show that Swiss projects are still predominantly anchored to the "consent or anonymise" approach, according to which the solution to solve ethico-legal problematics concerning health data is to either anonymize information or to require explicit authorization by data subjects [1]. Differently, Danish projects have made vaster use of alternative solutions, such as relying on specific confidentiality tools, and, more importantly, exploiting regulation that allows-upon certain conditions-to share and link health-related data without the need of obtaining consent by data subjects or REC approval. This demonstrates that the development of proper regulations to facilitate the harmonization and linking of health data offers practical solutions that projects developers are then willing to use. In this framework, another important finding concerns the role of the data protection authority. Whereas in Switzerland this public office-although existing-does not play a defined role with respect to research, results show that Danish studies have a more active interaction with the Data Protection Agency, as they need to apply for permission to use health data. The nature of the application to the national Data Protection Agency that Danish projects need to file is not explicitly described in the records reviewed, but it has been presented elsewhere [375,376] as a less demanding procedure, resembling a simple duty of notification. Thus, many Danish projects dealing exclusively with health data-in accordance with national regulation-do not need to apply for full ethical review from a REC or IRB, an often demanding and lengthy process, but simply have to obtain clearance from the Data Protection Agency. This institutionalized interaction with the public authority responsible to ensure compliance with data processing rules can be an important factor helping project developers, since it incentivizes to proactively tackle privacy concerns. This interaction could thus be considered as a model to inspire changes in the regulatory framework in Switzerland.

Technical barriers and facilitators
In this systematic review, data quality issues were the most commonly reported barriers, followed by the lack of data standards and limited technical capabilities. Although Denmark has a developed health data infrastructure, numerous identified projects described that data quality problems still affect health services, public health and research datasets [38,79,86,98,119,143,149,151]. This is confirmed by other studies, such as a review on the Danish National Patient Registry (DNRP) where the authors concluded that data incompleteness and heterogeneous validation methods of data limited the research potential of this registry [377]. Although relevant, data quality issues can be mitigated in a system like the Danish one, since linkage between data from different registries can be easily performed using the personal identification number (CPR) provided to all Danish citizens at birth and to stable residents [270]. Comparatively, Swiss projects and projects involving a Swiss partner also reported slightly more issues related to data quality than to data standards. However, in comparison to their Danish counterparts which reported almost twice more issues related to data quality than data standards, the difference in reporting of data standards and data quality issues was smaller in Swiss projects. This more equivalent reporting could imply that data standard issues are considered as important as data quality issues for the success of Swiss projects. Indeed, the high levels of data-heterogeneity in the Swiss healthcare context might stem from the fragmented nature of the healthcare system, where each of the 26 cantons [federal states] has a high degree of autonomy and where more than 55 health insurers are active [378].
These findings underline how technical issues are interconnected with the context where projects are carried out, and that also external systemic factors-and not simply internal complications of the projects themselves-affect the emerging of these barriers. In Denmark, for example, the presence of nation-wide registries fosters the development of studies relying on secondary use of routinely collected data, where researchers are more likely faced with issues about the quality of data, since the latter was originally collected for a different purpose. On the contrary, in a country like Switzerland-where data are more often prospectively collected -issues about the absence of common standards because of fragmentation are also likely to be evident, on top of data quality issues.
Our findings suggest therefore that even technical issues concerning data are strongly embedded in the surrounding where projects are conceived. This should induce project developers to communicate and learn from each others, since the barriers they will encounter and the solutions they will find are more likely to be dependent also on the context where they act, and not only on the specific features of their research. For example, since Switzerland's healthcare sector does not use a universal personal identification number because of privacy concerns [379], linkage of data will almost certainly represent a technical challenge, regardless of the features the single project or the data that it aims at using.

Motivational and financial barriers and facilitators
With respect to motivational and financial factors, our findings are partly in line with the literature. Previous research had underscored that the key motivational and financial aspects concerned the lack of research incentives from resource-limited institutions, the fears of being 'robbed' of data before publication or of losing reputation because others might identify errors in the data, the reluctance to facilitate access due to potential inappropriateness of further uses, the need to secure resources for data sharing activities and the necessity to make arrangements between institutions for data management costs [15, 380,381].
Overall, national and cross-national Swiss projects combined reported more frequently motivational and financial facilitators than their Danish counterparts. This suggests that in a country with a less institutionalised system of data sharing and where studies often have a prospective design, more strategies are elaborated to deal with financial and motivational issues related to data, since-with a lower systemic support-single project developers have to make a greater effort. In contrary, in a context like Denmark-with the high prevalence of studies with retrospective design and the reliance on secondary uses of routinely collected health data-the need for financial and motivational facilitators might be lower. In fact, when health data harmonization is prevalently retrospective, a lower number of actors is involved [382]since primary data collectors are rarely included-thus reducing the urgency to create motivational or financial incentives for a large number of collaborators.
Another important finding related to financial aspects is that the presence of economic constraints can be the source of additional barriers related to data harmonization, such as data quality issues. For instance, the Swiss project AMIS Plus-concerning a register for acute coronary syndrome-could not envisage systematic site visits to assess data quality or more indepth questionnaires due to resource limitation [290]. In Denmark, similarly, with the Copenhagen School Health Records Register-a health examination register for schoolchildren containing data on more than 350,000 individuals-financial constraints made it impossible for the authors to computerize the entire health card, thus limiting the understanding of potential confounding variables [47]. This indicates even more that barriers of different natures are interconnected and that new projects need to acknowledge this interconnectedness of the barriers to successfully address them.

Political barriers and facilitators
Danish national projects did not report any barriers of a political nature, whereas crossnational collaborations mentioned a few, such as data ownership and organizational issues [44,85,95]. This suggests that an institutionalization of data processing practices, similar to what occurs in Denmark [383], helps to remove political obstacles. Moreover, the presence of a centralized healthcare system structure also proves helpful, because it reduces the number of actors involved and thus the presence of competing interests. Political issues, however, might re-emerge when projects are cross-national and thus abandon the relatively safe-haven created at the national level.
In a context like the Swiss one, on the contrary, political barriers seem to be more relevant for national projects, because these fuel internal conflicts related to the diversity of interests within healthcare and to the difficulty of implementing uniform and centralized policies [132]. In fact, the two most mentioned political facilitators in Switzerland-building trust amongst stakeholders [132, 361] and stakeholders retaining control over data access [132,290]-are both related to the attempt to coordinate the numerous different parties operating in the health data field and accommodate their competing interests. This might also explain why less political barriers are reported for Swiss cross-national projects. In fact, when projects from a context like the Swiss one go to a supra-national level, the chances of disputes related to in-country political antagonism to emerge is lower.
Our results are thus in line with the literature, where mistrust between stakeholders, absence of comprehensive guidelines for data sharing and lack of legal accountability were identified as major political issues [2, 7, 15]. However, our results further show that the incidence of political barriers seems quite different in single-country studies if compared to cross national ones. This finding is particularly important since it underlines that sometimes the choice of a national or cross-national design might have an impact on the number of political issues encountered.

Socio-cultural barriers and facilitators
Barriers and facilitators of a socio-cultural nature were rarely mentioned in the included records. Comparatively, the incidence of cultural barriers seems to be higher for Switzerland, where cultural clashes were mentioned more often than for Danish projects. Such difference could be due to the higher degree of fragmentation of the Swiss healthcare system in comparison to the Danish one, which is centralized and state-funded [18]. In fact, one Swiss study [132] reported that the choice for a distributed model in the managing of data was based on prior failures to implement centralized systems of health data and public mistrust towards the concept of centralization. Socio-cultural facilitators were mostly related to the involvement of data-subjects by allowing them to retain control of data access. For instance, the Swiss project reported that data subjects had the possibility to decide which part of their medical records could be considered "stigmatizing", and thereafter blinded to healthcare professionals, other than their designated and trusted physician [132]. The designated and trusted physician would have access to the full record.
It is naturally impossible to determine whether socio-cultural barriers were actually overlooked or simply not reported. In either case, the limited mentioning of these factors signals an underappreciation of their importance. On the contrary, socio-cultural aspects should be carefully considered by project developers, since the harmonization of health data cannot ignore the cultural peculiarities of the single contexts from where data are pooled [384]. Harmonization, linking and sharing do not happen in a vacuum and opening up the dialogue between data processors and society at large can be an important success factor for the harmonization of health data in the long run.

Limitations
The limitations of this systematic review include choices that we made regarding the number of databases used for our search, the fact that we did search using English key words, and that only 20 percent of included papers went through double checking for data extraction consistency. We could have thus missed valuable studies that were published only in Danish, French, and German which we could have found if key words had been in those languages. Given the high number of papers included and resources related constraints, we were unable to double check for all information recorded, but in light of low discrepancies found in the portion of records which were double-checked, we are confident in our output. A reporting bias of barriers and facilitators identified in the included papers cannot be excluded as published papers are focused mostly on the effectiveness of their interventions rather than on the implementation phase. It is possible that our results are thus biased towards barriers and facilitators more likely to be reported in the papers (e.g. those of a technical nature). Given the low numbers of certain types of reported barriers and facilitators, it is difficult to compare the situation in the two countries without under-or over-exaggerating their presence or absence in the two countries. However, the main objective of this systematic review was to identify barriers and facilitators to harmonized health data collection, sharing and linkage in Denmark and Switzerland. Causal inference was not part of this review's primary objectives.

Conclusion
This systematic review gathered evidence from Switzerland and Denmark to map and describe barriers and facilitators concerning data harmonization, sharing and linkage. Given the focus of this review on Switzerland and Denmark, part of the findings has specific relevance for these two countries. In particular, for Switzerland it has emerged that fragmentation in the health data environment is a key challenge for harmonizing, sharing and linking of data. Since the implementation of more centralized governance systems-which are of great use in Denmark -might not be a viable option for Switzerland because of the political structure of the country, a distributed governance model, which emphasises interoperability of health data, seems to be the preferable way forward. The introduction of Blockchain technology for patient records, which insures security and respects decentralization [385,386], is reportedly an auspicious technology as its use in the Estonian healthcare system described by Mettler [387] suggests. This review outlined that the existing data infrastructure at the national-level in Denmark incentivizes the completion of retrospective registry-based studies relying on data reuse. Although barriers are still reported, the existence and comprehensiveness of this data infrastructure confirms that past efforts to improve the health data framework have proven successful. For the future, efforts should focus on easing projects involving cross-national collaborations.
However, other findings are meaningful well beyond the borders of the two countries specifically considered. In particular, in this review it has emerged that, although a great number of barriers and facilitators are mentioned by the projects involving health data harmonization, sharing and linking, reporting focusses predominantly on specific aspects-above all technical ones. Whereas technical aspects are certainly important, the reluctancy to mention also issues of other natures is detrimental to the more general effort of the scientific community to favour the harmonization of health data. Referring more openly to the difficulties encountered at the ethico-legal level, for example, might be of help both for new projects to develop appropriate approaches and for policy makers to gather evidence on which regulatory interventions are needed. The under-appreciation of ethico-legal, socio-cultural and other context-specific complexities is a faux-pas, since the trust of both data-subjects and society at large is indispensable for the success a community in improving the health data context, like the experience of Iceland has demonstrated in the past [388]. There, the project to build a national "health sector database" with health information of all citizens imported from their medical records failed also due to the underappreciation of ethico-legal issues (e.g. informed consent and privacy). Specifically, the population complained that inclusion of personal medical records into the database was supposed to happen without consent by individuals or the possibility to opt out. This was felt like a violation of privacy, because of the risk of re-identification and also due to the fact that the database was supposed to be run by a private company [389]. A privacy complaint was brought in front of the national high court, who ruled against the project to build the database. For this reason, the project was definitely aborted [390]. In summary, the success of current and future projects is likely to depend on a better understanding and appreciation of the complexities associated with harmonizing, sharing and linking health data. In the same line, proposed solutions to harmonization issues should not underestimate the contextual particularities of the country, in which such health data processes occur.
Supporting information S1 Text. PRISMA checklist. (DOCX) not involved in any other aspect of this review and the views expressed are those of the authors and not necessarily those of the funder. This systematic review is part of a broader project funded by the SNSF titled "Advancing Smart solutions and eliminating barriers for the Acquisition, Analysis and Sharing of Health data in Switzerland (SMAASH)". MCM was an independent researcher when the work for this systematic review was carried out and she is currently affiliated with the Institute for Social and Preventive Medicine (ISPM) at the University of Bern.

Author Contributions
Conceptualization: Tenzin Wangmo, Bernice Simone Elger.       Factors influencing harmonized health data collection, sharing and linkage in Denmark and Switzerland