Suchergebnisse

ABSTRACTBackgroundData integration comprises methods and tools to aggregate data from disparate sources to various purposes. Heterogeneity and uncertainty are technical challenges in this field. The first involves different data representation or meaning, while the second refers to incomplete data or the expectancy that a data item exists in a data source. Our goal is to design and validate a data integration model and computing tools able to address both problems. Such model and tools will support the setup of a population-based cohort comprised by 100 million individuals and the generation of data marts (domain-specific data) to be used in epidemiological studies within an ongoing cooperation Brazil-UK. Such studies will assess the impact of a conditional cash transfer programme (PBF – Bolsa Família) on the occurrence, severity and mortality of several diseases and health problems (hospitalization, mortality, child health etc) over this cohort.
ApproachWe propose a three-dimensional data model to aggregate information on the cohort, exposition (payments received during the observed period) and health outcomes. We treat heterogeneity based on our existing probabilistic linkage pipeline that provides data quality assessment, data conditioning (standardization, cleansing, blocking, and anonymization), two methods for probabilistic record matching, and accuracy assessment. Through this pipeline, we are able to probabilistically link records from PBF, CadastroÚnico (CADU - socioeconomic information) and healthcare databases from the Unified Health System (SUS). Uncertainty is modeled through "possible worlds", which represent a data instance (record) with a corresponding probability. There exist 2^records possible worlds with the probability distribution being the product of record probabilities. We map the most probable relationships between all the databases involved and create some simulation scenarios in order to validate them. We are seeking for a good balance between the set of possible worlds, not overextending the possibilities, and their proximity to a real scenario.
ResultsThe current implementation comprises the linkage of the 2011 extraction of CADU and healthcare databases to populate the proposed model. Such linkage provides timely execution (up to 9 hours depending on the databases) with high accurate data marts (over 95% of true positive matched pairs) for samples with increasing size (from 1,447,512 to 12,036,010 records).
ConclusionOur model is able to treat heterogeneity aspects present in huge databases. Our tools provide timely execution of probabilistic linkage with high accuracy. We started to model uncertainty in order to perform simulations and decide how to incorporate it in our model.

ObjectivesClustering algorithms are commonly used to identify disease clusters. New Clustering algorithms are benchmarked on synthetic data to assess their accuracy. These datasets lack the complexities of real electronic health record (EHR) producing a partial assessment of the algorithm. We developed a synthetic EHR cluster generator for benchmarking clustering algorithms.
ApproachWe have created a synthetic EHR cluster generator, clustEHR, based on Synthea, a synthetic EHR generator that produces datasets (with parameterized noise and cluster separation) of known clusters and with clinically relevant patient outcomes. We evaluated clustEHR by generating multiple datasets of variable cluster separation and percentage of noise variables to reflect easy and hard clustering problems. We used a linear model to assess the relationship between these parameters and cluster problem difficulty. K-means accuracy was used as a proxy to measure cluster problem difficulty.
ResultsWe have developed a tool for generating synthetic EHR cluster data with clinically relevant outcomes based on the rate of decline of medical observations (e.g. blood pressure). The following parameters are supported: a) number of clusters, b) number of patients in each cluster, c) number and data type of features, d) separation through defining clusters as either diseases such as COPD or dementia (high separability) or inter-disease conditions such as emphysema and chronic bronchitis within COPD (low separability), and e) noise variables through identifying variables not predictive of true cluster outcomes random forest feature importance metric. We show that high cluster separation significantly increases k-means accuracy (coefficient of 0.33). Smaller percent of noise variables increase accuracy though not significantly (coefficient 0.42).
ConclusionClustEHR offers realistic mixed data types as well as outcomes which are frequently used to evaluate clusters when subtyping diseases. The evaluating results suggest that the difficulty of the cluster data can be user determined. The tool can be used to create realistic datasets for evaluating clustering approaches.

ABSTRACTObjectivesWe used data from UK-Biobank that were linked with Hospital Episode Statistics and Office for National Statistics to assess the relationship between low-density lipoprotein cholesterol (LDL-C) and atrial fibrillation (AF). In this study, we applied Mendelian randomization in order to find out whether there is a causal effect of LDL-C to AF.
ApproachWe used data from the UK Biobank (~500,000 subjects) which is linked with electronic health records. At baseline (2006-2010), participants from across the UK took part in this project. They have undergone measures, provided blood, urine and saliva samples for future analysis, detailed information about themselves and agreed to have their health followed. Information in relation to the development of atrial fibrillation was derived from a) the enrollment of the participants (self-reported events), b) their hospitalization before and after their recruitment to UK-Biobank (confirmed events from Hospital Episode Statistics) and c) the death certificates [confirmed events from Office for National Statistics]. We also used genetic data from the analyses of the participants' blood sample that have been stored. We used Mendelian randomization to capture the effect of LDL-C to AF. As instruments, we used a genetic predisposition risk score (GPRs) for LDL-C, which was created as a weighted sum of the 18 most significant SNPs related to LDL-C, as there were documented in Global Lipid Consortium, in 18 out of 22 chromosomes. We ran a logistic regression model, using AF as outcome and GPRs as exposure.
ResultsOur final sample consisted of 144,092 individuals, for which we have valid information for their genetic data. The AF cases in this sample were 3207, most of which were identified from Hospital Episode Statistics (hospitalization of the participants). From the Mendelian randomization study, from our preliminary results, we found a weak positive relationship between GPRs and AF, when we did not adjust for any covariate [OR per one unit increase of GPRs=1.08, 95% CI= (0.95-1.22)] and results remained practically the same when we adjusted for age and sex [OR=1.09, 95% CI= (0.96-1.24 )].
ConclusionWe observed a weak positive association between LDL-C and AF in this study. This is the first Mendelian randomization approach that focuses on this relationship. More Mendelian randomization studies should be performed in order to identify the causal effect of LDL-C to AF. The use of electronic health records will facilitate the conduction of similar studies.

[EN] Record linkage is a technique widely used to gather data stored in disparate data sources that presumably pertain to the same real world entity. This integration can be done deterministically or probabilistically, depending on the existence of common key attributes among all data sources involved. The probabilistic approach is very time-consuming due to the amount of records that must be compared, specifically in big data scenarios. In this paper, we propose and evaluate a methodology that simultaneously exploits multicore and multi-GPU architectures in order to perform the probabilistic linkage of large-scale Brazilian governmental databases. We present some algorithmic optimizations that provide high accuracy and improve performance by defining the best algorithm-architecture combination for a problem given its input size. We also discuss performance results obtained with different data samples, showing that a hybrid approach outperforms other configurations, providing an average speedup of 7.9 when linking up to 20.000 million records. ; This work has been partially supported by CNPq, FAPESB, Bill & Melinda Gates Foundation, The Royal Society (UK), Medical Research Council (UK), NVIDIA Hardware Grant Program, Generalitat Valenciana (Grant PROMETEOII/2014/003), Spanish Government and European Commission through TEC2015-67387-C4-1-R (MINECO/FEDER), and network CAPAP-H. We have also worked in cooperation with the EU-COST Programme Action IC1305, "Network for Sustainable Ultrascale Computing (NESUS) ; Boratto, M.; Alonso-Jordá, P.; Pinto, C.; Melo, P.; Barreto, M.; Denaxas, S. (2019). Exploring Hybrid Parallel Systems for Probabilistic Record Linkage. The Journal of Supercomputing. 75:1137-1149. https://doi.org/10.1007/s11227-018-2328-3 ; S ; 1137 ; 1149 ; 75 ; Andrade G, Viegas F, Ramos GS, Almeida J, Rocha L, Gonçalves M, Ferreira R (2013) GPU-NB: a fast CUDA-based implementation of Naïve Bayes. In: 2013 25th International Symposium on Computer Architecture and High Performance Computing, pp 168–175 ...

ABSTRACTObjectivesElectronic health records (EHR) across primary, secondary, and tertiary care are increasingly being linked for research at a population level. The increasing volume, variety, velocity, and veracity of big biomedical data makes research reproducibility challenging. Research reproducibility and replicability is essential for the external validity and generalizability of scientific findings and the lack of standardized approaches and tools and relative opaqueness of data manipulation methods is detrimental to their integrity. The objective of this study was to explore, evaluate and propose methods, tools and approaches for addressing some of the challenges associated with reproducibility when using linked national electronic health records for research.
ApproachWe systematically searched literature and internet resources for well-established and appropriate methods, tools, and approaches used in related scientific disciplines. The identified techniques were systematically evaluated in terms of their capacity to facilitate reproducible research in routinely collected health data across the life course of a research project: from protocol creation and raw data curation to data transformation and statistical analysis though to finding dissemination and impact. Most importantly, the identified techniques were tested and applied in a contemporary database of linked electronic health records. CALIBER is a research data platform of linked national electronic health records from primary care (Clinical Practice Research Datalink), secondary care (Hospital Episode Statistics), acute coronary syndrome disease registry (Myocardial Ischaemia National Audit Project) and cause-specific mortality (Office for National Statistics) for roughly 2 million adults.
ResultsFirstly, we present the review of methods and approaches which we identified through our search. Secondly, we propose a set of recommendations for applying them within the context of research projects making use of linked routinely collected health data. Focal interests included: a) documentation of data (attributes, relationships, and interpretation), b) data processing (source code, instructions, and parameters), c) results (visualizations, figures), and any supplementary material. Thirdly, we present approaches around a) raw data curation using international metadata standards, b) study protocol encoding, c) provenance and sharing of data transformation and statistical analysis operations, d) public and private data retention, and e) computable EHR-driven phenotypes.
ConclusionThe complexity and size of routinely collected health data is increasing through linkages across distributed data sources. The scientific community benefits from findings which can be replicated. This study presents a number of methods, tools and approaches across the project life course for ensuring that their research studies are reproducible and replicable from the wider scientific community.

ABSTRACT
AimsWe aimed (i) to create algorithms for identifying stroke and acute myocardial infarction (MI) cases in a large prospective population-based study (UK Biobank), using linked data from national hospital admissions data and death registers, combined with self-report data from the baseline assessment; and (ii) to assess validity by examining associations with risk factors and mortality.MethodsUK Biobank is a prospective study of 503,000 participants, aged 40-69 years when recruited in 2006-2010 from centres across England, Scotland and Wales. Participants provided extensive questionnaire data on lifestyle, environment and medical history (with confirmation of self-reported medical conditions during a brief interview with a trained research nurse), had physical measures, and provided biological samples. Follow-up is principally through linkages to national health-related datasets, integrated from different data providers for each country and including cohort-wide data for ICD-coded hospital admissions and registered deaths, with follow-up to 2011 for this report. We used expert opinion and systematic reviews to identify baseline self-report items and ICD codes maximising positive predictive value for identification of stroke, MI and their subtypes. We classified participants with at least one relevant code as being first affected either before (prevalent cases) or after recruitment (incident cases). We compared cases with non-cases (controls) using logistic regression (prevalent cases) and survival analysis (incident cases) to examine associations with vascular risk factors, defined using data from the baseline assessment. We used survival analysis to compare vascular, non-vascular and all-cause mortality for cases versus controls post recruitment (prevalent cases) and post hospitalization (incident cases).ResultsWe identified 8654 stroke cases and 13,479 MI cases. 90% of both were prevalent at recruitment, of which 29% (stroke) and 54% (MI) were identified through both research nurse-confirmed self-report and hospital admission records. During ≈1 million person years of follow-up in those without a prevalent record, we identified 871 incident strokes and 1387 incident MIs, of which 8% (stroke) and 2.9% (MI) were identified in both hospital and death records. Male sex, low socio-economic status, smoking and increased body mass index were all positively associated with both stroke and MI. Compared with age and sex-matched controls, mortality for stroke and MI cases was increased both after recruitment (prevalent cases) and after hospitalization (incident cases).ConclusionInformation from linked coded datasets from different data providers can be combined with information collected at recruitment to identify and validate prevalent and incident acute MI and stroke.

Background: To effectively prevent, detect, and treat health conditions that affect people during their lifecourse, health-care professionals and researchers need to know which sections of the population are susceptible to which health conditions and at which ages. Hence, we aimed to map the course of human health by identifying the 50 most common health conditions in each decade of life and estimating the median age at first diagnosis. Methods: We developed phenotyping algorithms and codelists for physical and mental health conditions that involve intensive use of health-care resources. Individuals older than 1 year were included in the study if their primary-care and hospital-admission records met research standards set by the Clinical Practice Research Datalink and they had been registered in a general practice in England contributing up-to-standard data for at least 1 year during the study period. We used linked records of individuals from the CALIBER platform to calculate the sex-standardised cumulative incidence for these conditions by 10-year age groups between April 1, 2010, and March 31, 2015. We also derived the median age at diagnosis and prevalence estimates stratified by age, sex, and ethnicity (black, white, south Asian) over the study period from the primary-care and secondary-care records of patients. Findings: We developed case definitions for 308 disease phenotypes. We used records of 2 784 138 patients for the calculation of cumulative incidence and of 3 872 451 patients for the calculation of period prevalence and median age at diagnosis of these conditions. Conditions that first gained prominence at key stages of life were: atopic conditions and infections that led to hospital admission in children (<10 years); acne and menstrual disorders in the teenage years (10-19 years); mental health conditions, obesity, and migraine in individuals aged 20-29 years; soft-tissue disorders and gastro-oesophageal reflux disease in individuals aged 30-39 years; dyslipidaemia, hypertension, and erectile dysfunction in individuals aged 40-59 years; cancer, osteoarthritis, benign prostatic hyperplasia, cataract, diverticular disease, type 2 diabetes, and deafness in individuals aged 60-79 years; and atrial fibrillation, dementia, acute and chronic kidney disease, heart failure, ischaemic heart disease, anaemia, and osteoporosis in individuals aged 80 years or older. Black or south-Asian individuals were diagnosed earlier than white individuals for 258 (84%) of the 308 conditions. Bone fractures and atopic conditions were recorded earlier in male individuals, whereas female individuals were diagnosed at younger ages with nutritional anaemias, tubulointerstitial nephritis, and urinary disorders. Interpretation: We have produced the first chronological map of human health with cumulative-incidence and period-prevalence estimates for multiple morbidities in parallel from birth to advanced age. This can guide clinicians, policy makers, and researchers on how to formulate differential diagnoses, allocate resources, and target research priorities on the basis of the knowledge of who gets which diseases when. We have published our phenotyping algorithms on the CALIBER open-access Portal which will facilitate future research by providing a curated list of reusable case definitions. Funding: Wellcome Trust, National Institute for Health Research, Medical Research Council, Arthritis Research UK, British Heart Foundation, Cancer Research UK, Chief Scientist Office of the Scottish Government Health and Social Care Directorates, Department of Health and Social Care (England), Health and Social Care Research and Development Division (Welsh Government), Public Health Agency (Northern Ireland), Economic and Social Research Council, Engineering and Physical Sciences Research Council, National Institute for Social Care and Health Research, and The Alan Turing Institute.

The Center for Data and Knowledge Integration for Health (CIDACS) was created in 2016 in Salvador (Bahia, Brazil). This paper aims to present a profile of CIDACS, including its current databases. CIDACS aims to conduct interdisciplinary studies and research, develop new scientific methodology and promote professional training using linked large-scale databases and high-performance computational resources in a secure environment. Administrative data is at the core of the activities conducted by CIDACS. The advantages of administrative data include significantly larger sample sizes, an inherent longitudinal structure and high-quality information. The center's research projects are primarily focused on enhancing the understanding surrounding the impact of social protection policies (e.g., public cash-transfer and housing programs) on health outcomes in low-income populations throughout Brazil. CIDACS' primary data source is citizens who register with the Cadastro Único program, which encompasses individuals eligible to receive benefits from over 20 governmental social programs. CIDACS has two separate environments for data handling: 1) Data Production Center, a secure room housing the computational infrastructure for ingesting, storing, cleaning, processing and linking original databases, as well as extracting research-ready datasets and 2) Data Analysis Environment, a computational infrastructure based on data safe haven principles, which allows researchers to access and process requested datasets. Brazil has a large public health community that uses national health and social databases for research programs, and the linkage of different databases has been a widely employed practice in the country. CIDACS is the result of efforts by researchers, policymakers and public health officials to use and improve the quality of Brazilian health databases. CIDACS is expected to be an important resource for researchers and policymakers interested in improving the evidence base in different aspects of health, as well as with regard to the social determinants of health and the effects of social and environmental policies on health in general.

Information is increasingly digital, creating opportunities to respond to pressing issues about human populations in near real time using linked datasets that are large, complex, and diverse. The potential social and individual benefits that can come from data-intensive science are large, but raise challenges of balancing individual privacy and the public good, building appropriate socio-technical systems to support data-intensive science, and determining whether defining a new field of inquiry might help move those collective interests and activities forward. A combination of expert engagement, literature review, and iterative conversations led to our conclusion that defining the field of Population Data Science (challenge 3) will help address the other two challenges as well. We define Population Data Science succinctly as the science of data about people and note that it is related to but distinct from the fields of data science and informatics. A broader definition names four characteristics of: data use for positive impact on citizens and society; bringing together and analyzing data from multiple sources; finding population-level insights; and developing safe, privacy-sensitive and ethical infrastructure to support research. One implication of these characteristics is that few people possess all of the requisite knowledge and skills of Population Data Science, so this is by nature a multi-disciplinary field. Other implications include the need to advance various aspects of science, such as data linkage technology, various forms of analytics, and methods of public engagement. These implications are the beginnings of a research agenda for Population Data Science, which if approached as a collective field, can catalyze significant advances in our understanding of trends in society, health, and human behavior.

This work was supported by Health Data Research UK (HDR-9006; CFC0110) and the Medical Research Council (MR/S027750/1). Health Data Research UK is funded by: UK Medical Research Council; Engineering and Physical Sciences Research Council; Economic and Social Research Council; National Institute for Health Research (England); Chief Scientist Office of the Scottish Government Health and Social Care Directorates; Health and Social Care Research and Development Division (Welsh Government); Public Health Agency (Northern Ireland); British Heart Foundation and Wellcome Trust. ; Introduction Multimorbidity is widely recognised as the presence of two or more concurrent long-term conditions, yet remains a poorly understood global issue despite increasing in prevalence. We have created the Wales Multimorbidity e-Cohort (WMC) to provide an accessible research ready data asset to further the understanding of multimorbidity. Our objectives are to create a platform to support research which would help to understand prevalence, trajectories and determinants in multimorbidity, characterise clusters that lead to highest burden on individuals and healthcare services, and evaluate and provide new multimorbidity phenotypes and algorithms to the National Health Service and research communities to support prevention, healthcare planning and the management of individuals with multimorbidity. Methods and analysis The WMC has been created and derived from multisourced demographic, administrative and electronic health record data relating to the Welsh population in the Secure Anonymised Information Linkage (SAIL) Databank. The WMC consists of 2.9 million people alive and living in Wales on the 1 January 2000 with follow-up until 31 December 2019, Welsh residency break or death. Published comorbidity indices and phenotype code lists will be used to measure and conceptualise multimorbidity.Study outcomes will include: (1) a description of multimorbidity using published data phenotype algorithms/ontologies, (2) investigation of the associations between baseline demographic factors and multimorbidity, (3) identification of temporal trajectories of clusters of conditions and multimorbidity and (4) investigation of multimorbidity clusters with poor outcomes such as mortality and high healthcare service utilisation. Ethics and dissemination The SAIL Databank independent Information Governance Review Panel has approved this study (SAIL Project: 0911). Study findings will be presented to policy groups, public meetings, national and international conferences, and published in peer-reviewed journals. ; Publisher PDF ; Peer reviewed

This work was supported by Health Data Research UK (HDR-9006; CFC0110) and the Medical Research Council (MR/S027750/1). Health Data Research UK is funded by: UK Medical Research Council; Engineering and Physical Sciences Research Council; Economic and Social Research Council; National Institute for Health Research (England); Chief Scientist Office of the Scottish Government Health and Social Care Directorates; Health and Social Care Research and Development Division (Welsh Government); Public Health Agency (Northern Ireland); British Heart Foundation and Wellcome Trust. ; Introduction Multimorbidity is widely recognised as the presence of two or more concurrent long-term conditions, yet remains a poorly understood global issue despite increasing in prevalence. We have created the Wales Multimorbidity e-Cohort (WMC) to provide an accessible research ready data asset to further the understanding of multimorbidity. Our objectives are to create a platform to support research which would help to understand prevalence, trajectories and determinants in multimorbidity, characterise clusters that lead to highest burden on individuals and healthcare services, and evaluate and provide new multimorbidity phenotypes and algorithms to the National Health Service and research communities to support prevention, healthcare planning and the management of individuals with multimorbidity. Methods and analysis The WMC has been created and derived from multisourced demographic, administrative and electronic health record data relating to the Welsh population in the Secure Anonymised Information Linkage (SAIL) Databank. The WMC consists of 2.9 million people alive and living in Wales on the 1 January 2000 with follow-up until 31 December 2019, Welsh residency break or death. Published comorbidity indices and phenotype code lists will be used to measure and conceptualise multimorbidity. Study outcomes will include: (1) a description of multimorbidity using published data phenotype algorithms/ontologies, (2) investigation of the associations between baseline demographic factors and multimorbidity, (3) identification of temporal trajectories of clusters of conditions and multimorbidity and (4) investigation of multimorbidity clusters with poor outcomes such as mortality and high healthcare service utilisation. Ethics and dissemination The SAIL Databank independent Information Governance Review Panel has approved this study (SAIL Project: 0911). Study findings will be presented to policy groups, public meetings, national and international conferences, and published in peer-reviewed journals. ; Publisher PDF ; Peer reviewed