CHAPTER 4 198 CHECK-HF in the Netherlands, Swedish heart failure registry and BIOSTAT-CHF and ASIAN-HF represent quality sources of patient data that were specifically designed for HF research. Registries have detailed and structured information on HF severity measures, medical history, electrocardiogram, echocardiography, clinical and laboratory measurements which were important for case-mix adjustment, given the complexity and heterogeneity of patients with this condition.64–67 Although international registries allow us to understand cross-border practice, it is necessary to recognize that patient and site selection do take place. Study sites in registries are typically academic centres or hospitals and investigators involved usually have specific interest in heart failure.68 On the other hand, EHRs offer broader reflection of everyday patients but lacks uniformity in data on disease severity and requires substantial effort to pre-process and structure free-text clinical notes into scalable, computable formats. Currently, disease-specific registries and electronic records complement one another as they each bring unique advantages in terms of data completeness and uniformity and spectrum of HFrEF severity. To leverage on growing quantities and dimensions of biomedical data, largescale data pooling can be done by combining data from different organizations into a single large data set and analyzing by individual-level meta-analysis (ILMA). Although this approach offers convenience for analytics, it is typically not possible owing to ethical and legal constraints on third-party data transfer.69 For this reason, a data federation framework or decentralized model is proposed to link multiple disparate data repositories across institutional and cross-jurisdictional boundaries to a central analytic computer.70 This way data shall remain geographically localized but accessible by data queries. Federated data systems require agreed and shared technological infrastructure, data and metadata interoperability, legal and governance policies and an example is CanDIG, a Canadian federated data system for research on genomic data.71 Mapping of terminologies to a common data model for cardiovascular research was also undertaken by BigData@Heart consortium partners. While this is in progress, data providers in the consortium agreed to first share data via an approach that minimally aggregates data to preserve some granularity while assuring privacy. For small datasets, an issue with low table cell counts of between one and three patients poses a risk of identifying patients. We circumvented this by assigning a central number of two72 and tested the extent of information loss from aggregation. From this, we demonstrated insignificant loss of
RkJQdWJsaXNoZXIy MjY0ODMw