DataSHIELD: taking the analysis to the data, not the data to the analysis

Privacy-preserving federated machine learning on FAIR health data: A real-world application

Objective

This paper introduces a privacy-preserving federated machine learning (ML) architecture built upon Findable, Accessible, Interoperable, and Reusable (FAIR) health data. It aims to devise an architecture for executing classification algorithms in a federated manner, enabling collaborative model-building among health data owners without sharing their datasets.

Materials and methods

Utilizing an agent-based architecture, a privacy-preserving federated ML algorithm was developed to create a global predictive model from various local models. This involved formally defining the algorithm in two steps: data preparation and federated model training on FAIR health data and constructing the architecture with multiple components facilitating algorithm execution. The solution was validated by five healthcare organizations using their specific health datasets.

Results

Five organizations transformed their datasets into Health Level 7 Fast Healthcare Interoperability Resources via a common FAIRification workflow and software set, thereby generating FAIR datasets. Each organization deployed a Federated ML Agent within its secure network, connected to a cloud-based Federated ML Manager. System testing was conducted on a use case aiming to predict 30-day readmission risk for chronic obstructive pulmonary disease patients and the federated model achieved an accuracy rate of 87%.

Discussion

The paper demonstrated a practical application of privacy-preserving federated ML among five distinct healthcare entities, highlighting the value of FAIR health data in machine learning when utilized in a federated manner that ensures privacy protection without sharing data.

Conclusion

This solution effectively leverages FAIR datasets from multiple healthcare organizations for federated ML while safeguarding sensitive health datasets, meeting legislative privacy and security requirements.

Early childcare arrangements and children's internalizing and externalizing symptoms: an individual participant data meta-analysis of six prospective birth cohorts in Europe

Background

Early childcare attendance may be related to children's internalizing and externalizing symptoms throughout childhood and young adolescence, however evidence from Europe is limited. We aimed to assess this association across multiple population-based birth cohorts of children recruited in different European countries.

Methods

Data come from six parent-offspring prospective birth cohort studies across five European countries within the EU Child Cohort Network. A total of 87,208 parent-child dyads were included in the study. To test associations between childcare attendance (centre-based or informal) anytime between ages 0 and 4 years and children's internalizing and externalizing symptoms in middle childhood and young adolescence (measured at: 5-6 years, 7-9 years, and 10-13 years) a two-stage individual participant data meta-analysis was implemented. Linear regression models were performed in each cohort separately; combined random-effects meta-analysis was then used to obtain overall association estimates. In secondary analyses, we tested interactions between childcare attendance and mother's post-partum depression, low education status, and the child's sex.

Findings

Compared to children who were exclusively cared for by their parents prior to school entry, those who attended centre-based childcare had lower levels of internalizing symptoms in all age groups [5-6 years: β: -1.78 (95% CI: -3.39, -0.16); 7-9 years: β: -0.55 (95% CI: -0.88, -0.73); 10-13 years: β: -0.76 (95% CI: -1.15, -0.37)]. Children who attended informal childcare appeared to have elevated levels of internalizing symptoms between 7-9 and 10-13 years, respectively [β: 1.65 (95% CI: 1.25, 2.06); β: 1.25 (95% CI: 0.96, 1.54)]. Informal childcare attendance was also associated with increased levels of children's externalizing symptoms between 7-9 and 10-13 years, respectively [β: 2.84 (95% CI: 1.41, 4.26); β: 2.19 (95% CI: 0.54, 3.84)].

Interpretation

Early centre-based childcare is associated with decreased levels of children's internalizing symptoms compared to exclusive parental care. For informal childcare, opposite associations were observed. Overall, our results suggest that centre-based childcare attendance may be associated with slight positive impacts on children's emotional development and should be encouraged by public policies. In addition, children from socioeconomically disadvantaged families require special attention, as they may not sufficiently benefit from universal early childhood education and care (ECEC).

Funding

This research was funded by the ERC Consolidator grant RESEDA (Horizon Europe, 101001420).

Meat consumption and incident type 2 diabetes: an individual-participant federated meta-analysis of 1·97 million adults with 100 000 incident cases from 31 cohorts in 20 countries

BACKGROUND

Meat consumption could increase the risk of type 2 diabetes. However, evidence is largely based on studies of European and North American populations, with heterogeneous analysis strategies and a greater focus on red meat than on poultry. We aimed to investigate the associations of unprocessed red meat, processed meat, and poultry consumption with type 2 diabetes using data from worldwide cohorts and harmonised analytical approaches.

METHODS

This individual-participant federated meta-analysis involved data from 31 cohorts participating in the InterConnect project. Cohorts were from the region of the Americas (n=12) and the Eastern Mediterranean (n=2), European (n=9), South-East Asia (n=1), and Western Pacific (n=7) regions. Access to individual-participant data was provided by each cohort; participants were eligible for inclusion if they were aged 18 years or older and had available data on dietary consumption and incident type 2 diabetes and were excluded if they had a diagnosis of any type of diabetes at baseline or missing data. Cohort-specific hazard ratios (HRs) and 95% CIs were estimated for each meat type, adjusted for potential confounders (including BMI), and pooled using a random-effects meta-analysis, with meta-regression to investigate potential sources of heterogeneity.

FINDINGS

Among 1 966 444 adults eligible for participation, 107 271 incident cases of type 2 diabetes were identified during a median follow-up of 10 (IQR 7-15) years. Median meat consumption across cohorts was 0-110 g/day for unprocessed red meat, 0-49 g/day for processed meat, and 0-72 g/day for poultry. Greater consumption of each of the three types of meat was associated with increased incidence of type 2 diabetes, with HRs of 1·10 (95% CI 1·06-1·15) per 100 g/day of unprocessed red meat (I2=61%), 1·15 (1·11-1·20) per 50 g/day of processed meat (I2=59%), and 1·08 (1·02-1·14) per 100 g/day of poultry (I2=68%). Positive associations between meat consumption and type 2 diabetes were observed in North America and in the European and Western Pacific regions; the CIs were wide in other regions. We found no evidence that the heterogeneity was explained by age, sex, or BMI. The findings for poultry consumption were weaker under alternative modelling assumptions. Replacing processed meat with unprocessed red meat or poultry was associated with a lower incidence of type 2 diabetes.

INTERPRETATION

The consumption of meat, particularly processed meat and unprocessed red meat, is a risk factor for developing type 2 diabetes across populations. These findings highlight the importance of reducing meat consumption for public health and should inform dietary guidelines.

FUNDING

The EU, the Medical Research Council, and the National Institute of Health Research Cambridge Biomedical Research Centre.

Distributed non-disclosive validation of predictive models by a modified ROC-GLM

BACKGROUND

Distributed statistical analyses provide a promising approach for privacy protection when analyzing data distributed over several databases. Instead of directly operating on data, the analyst receives anonymous summary statistics, which are combined into an aggregated result. Further, in discrimination model (prognosis, diagnosis, etc.) development, it is key to evaluate a trained model w.r.t. to its prognostic or predictive performance on new independent data. For binary classification, quantifying discrimination uses the receiver operating characteristics (ROC) and its area under the curve (AUC) as aggregation measure. We are interested to calculate both as well as basic indicators of calibration-in-the-large for a binary classification task using a distributed and privacy-preserving approach.

METHODS

We employ DataSHIELD as the technology to carry out distributed analyses, and we use a newly developed algorithm to validate the prediction score by conducting distributed and privacy-preserving ROC analysis. Calibration curves are constructed from mean values over sites. The determination of ROC and its AUC is based on a generalized linear model (GLM) approximation of the true ROC curve, the ROC-GLM, as well as on ideas of differential privacy (DP). DP adds noise (quantified by the ℓ 2 sensitivityΔ 2 ( f ^ ) ) to the data and enables a global handling of placement numbers. The impact of DP parameters was studied by simulations.

RESULTS

In our simulation scenario, the true and distributed AUC measures differ by Δ AUC < 0.01 depending heavily on the choice of the differential privacy parameters. It is recommended to check the accuracy of the distributed AUC estimator in specific simulation scenarios along with a reasonable choice of DP parameters. Here, the accuracy of the distributed AUC estimator may be impaired by too much artificial noise added from DP.

CONCLUSIONS

The applicability of our algorithms depends on the ℓ 2 sensitivityΔ 2 ( f ^ ) of the underlying statistical/predictive model. The simulations carried out have shown that the approximation error is acceptable for the majority of simulated cases. For models with highΔ 2 ( f ^ ) , the privacy parameters must be set accordingly higher to ensure sufficient privacy protection, which affects the approximation error. This work shows that complex measures, as the AUC, are applicable for validation in distributed setups while preserving an individual's privacy.

Green spaces and respiratory, cardiometabolic, and neurodevelopmental outcomes: An individual-participant data meta-analysis of >35.000 European children

Studies evaluating the benefits and risks of green spaces on children's health are scarce. The present study aimed to examine the associations between exposure to green spaces during pregnancy and early childhood with respiratory, cardiometabolic, and neurodevelopmental outcomes in school-age children. We performed an Individual-Participant Data (IPD) meta-analysis involving 35,000 children from ten European birth cohorts across eight countries. For each participant, we calculated residential Normalized Difference Vegetation Index (NDVI) within a 300 m buffer and the linear distance to green spaces (meters) during prenatal life and childhood. Multiple harmonized health outcomes were selected: asthma and wheezing, lung function, body mass index, diastolic and systolic blood pressure, non-verbal intelligence, internalizing and externalizing problems, and ADHD symptoms. We conducted a two-stage IPD meta-analysis and evaluated effect modification by socioeconomic status (SES) and sex. Between-study heterogeneity was assessed via random-effects meta-regression. Residential surrounding green spaces in childhood, not pregnancy, was associated with improved lung function, particularly higher FEV1 (β = 0.06; 95 %CI: 0.03, 0.09 I2 = 4.03 %, p < 0.001) and FVC (β = 0.07; 95 %CI: 0.04, 0.09 I2 = 0 %, p < 0.001) with a stronger association observed in females (p < 0.001). This association remained robust after multiple testing correction and did not change notably after adjusting for ambient air pollution. Increased distance to green spaces showed an association with lower FVC (β = -0.04; 95 %CI: -0.07, -0.02, I2 = 4.8, p = 0.001), with a stronger effect in children from higher SES backgrounds (p < 0.001). No consistent associations were found between green spaces and asthma, wheezing, cardiometabolic, or neurodevelopmental outcomes, with direction of effect varying across cohorts. Wheezing and neurodevelopmental outcomes showed high between-study heterogeneity, and the age at outcome assessment was only associated with heterogeneity in internalizing problems.. This large European meta-analysis suggests that childhood exposure to green spaces may lead to better lung function. Associations with other respiratory outcomes and selected cardiometabolic and neurodevelopmental outcomes remain inconclusive.

Stress and anxiety during pregnancy and length of gestation: a federated study using data from five Canadian and European birth cohorts

While its etiology is not fully elucidated, preterm birth represents a major public health concern as it is the leading cause of child mortality and morbidity. Stress is one of the most common perinatal conditions and may increase the risk of preterm birth. In this paper we aimed to investigate the association of maternal perceived stress and anxiety with length of gestation. We used harmonized data from five birth cohorts from Canada, France, and Norway. A total of 5297 pregnancies of singletons were included in the analysis of perceived stress and gestational duration, and 55,775 pregnancies for anxiety. Federated analyses were performed through the DataSHIELD platform using Cox regression models within intervals of gestational age. The models were fit for each cohort separately, and the cohort-specific results were combined using random effects study-level meta-analysis. Moderate and high levels of perceived stress during pregnancy were associated with a shorter length of gestation in the very/moderately preterm interval [moderate: hazard ratio (HR) 1.92 (95%CI 0.83, 4.48); high: 2.04 (95%CI 0.77, 5.37)], albeit not statistically significant. No association was found for the other intervals. Anxiety was associated with gestational duration in the very/moderately preterm interval [1.66 (95%CI 1.32, 2.08)], and in the early term interval [1.15 (95%CI 1.08, 1.23)]. Our findings suggest that perceived stress and anxiety are associated with an increased risk of earlier birth, but only in the earliest gestational ages. We also found an association in the early term period for anxiety, but the result was only driven by the largest cohort, which collected information the latest in pregnancy. This raised a potential issue of reverse causality as anxiety later in pregnancy could be due to concerns about early signs of a possible preterm birth.

[Interoperability Working Group: core dataset and information systems for data integration and data exchange in the Medical Informatics Initiative]

The interoperability Working Group of the Medical Informatics Initiative (MII) is the platform for the coordination of overarching procedures, data structures, and interfaces between the data integration centers (DIC) of the university hospitals and national and international interoperability committees. The goal is the joint content-related and technical design of a distributed infrastructure for the secondary use of healthcare data that can be used via the Research Data Portal for Health. Important general conditions are data privacy and IT security for the use of health data in biomedical research. To this end, suitable methods are used in dedicated task forces to enable procedural, syntactic, and semantic interoperability for data use projects. The MII core dataset was developed as several modules with corresponding information models and implemented using the HL7® FHIR® standard to enable content-related and technical specifications for the interoperable provision of healthcare data through the DIC. International terminologies and consented metadata are used to describe these data in more detail. The overall architecture, including overarching interfaces, implements the methodological and legal requirements for a distributed data use infrastructure, for example, by providing pseudonymized data or by federated analyses. With these results of the Interoperability Working Group, the MII is presenting a future-oriented solution for the exchange and use of healthcare data, the applicability of which goes beyond the purpose of research and can play an essential role in the digital transformation of the healthcare system.

Machine learning-based health environmental-clinical risk scores in European children

BACKGROUND

Early life environmental stressors play an important role in the development of multiple chronic disorders. Previous studies that used environmental risk scores (ERS) to assess the cumulative impact of environmental exposures on health are limited by the diversity of exposures included, especially for early life determinants. We used machine learning methods to build early life exposome risk scores for three health outcomes using environmental, molecular, and clinical data.

METHODS

In this study, we analyzed data from 1622 mother-child pairs from the HELIX European birth cohorts, using over 300 environmental, 100 child peripheral, and 18 mother-child clinical markers to compute environmental-clinical risk scores (ECRS) for child behavioral difficulties, metabolic syndrome, and lung function. ECRS were computed using LASSO, Random Forest and XGBoost. XGBoost ECRS were selected to extract local feature contributions using Shapley values and derive feature importance and interactions.

RESULTS

ECRS captured 13%, 50% and 4% of the variance in mental, cardiometabolic, and respiratory health, respectively. We observed no significant differences in predictive performances between the above-mentioned methods.The most important predictive features were maternal stress, noise, and lifestyle exposures for mental health; proteome (mainly IL1B) and metabolome features for cardiometabolic health; child BMI and urine metabolites for respiratory health.

CONCLUSIONS

Besides their usefulness for epidemiological research, our risk scores show great potential to capture holistic individual level non-hereditary risk associations that can inform practitioners about actionable factors of high-risk children. As in the post-genetic era personalized prevention medicine will focus more and more on modifiable factors, we believe that such integrative approaches will be instrumental in shaping future healthcare paradigms.

OHDSI-compliance: a set of document templates facilitating the implementation and operation of a software stack for real-world evidence generation

Introduction

The open-source software offered by the Observational Health Data Science and Informatics (OHDSI) collective, including the OMOP-CDM, serves as a major backbone for many real-world evidence networks and distributed health data analytics platforms. While container technology has significantly simplified deployments from a technical perspective, regulatory compliance can remain a major hurdle for the setup and operation of such platforms. In this paper, we present OHDSI-Compliance, a comprehensive set of document templates designed to streamline the data protection and information security-related documentation and coordination efforts required to establish OHDSI installations.

Methods

To decide on a set of relevant document templates, we first analyzed the legal requirements and associated guidelines with a focus on the General Data Protection Regulation (GDPR). Moreover, we analyzed the software architecture of a typical OHDSI stack and related its components to the different general types of concepts and documentation identified. Then, we created those documents for a prototypical OHDSI installation, based on the so-called Broadsea package, following relevant guidelines from Germany. Finally, we generalized the documents by introducing placeholders and options at places where individual institution-specific content will be needed.

Results

We present four documents: (1) a record of processing activities, (2) an information security concept, (3) an authorization concept, as well as (4) an operational concept covering the technical details of maintaining the stack. The documents are publicly available under a permissive license.

Discussion

To the best of our knowledge, there are no other publicly available sets of documents designed to simplify the compliance process for OHDSI deployments. While our documents provide a comprehensive starting point, local specifics need to be added, and, due to the heterogeneity of legal requirements in different countries, further adoptions might be necessary.

Colorectal cancer health and care quality indicators in a federated setting using the Personal Health Train

OBJECTIVE

Hospitals and healthcare providers should assess and compare the quality of care given to patients and based on this improve the care. In the Netherlands, hospitals provide data to national quality registries, which in return provide annual quality indicators. However, this process is time-consuming, resource intensive and risks patient privacy and confidentiality. In this paper, we presented a multicentric 'Proof of Principle' study for federated calculation of quality indicators in patients with colorectal cancer. The findings suggest that the proposed approach is highly time-efficient and consume significantly lesser resources.

MATERIALS AND METHODS

Two quality indicators are calculated in an efficient and privacy presevering federated manner, by i) applying the Findable Accessible Interoperable and Reusable (FAIR) data principles and ii) using the Personal Health Train (PHT) infrastructure. Instead of sharing data to a centralized registry, PHT enables analysis by sending algorithms and sharing only insights from the data.

RESULTS

ETL process extracted data from the Electronic Health Record systems of the hospitals, converted them to FAIR data and hosted in RDF endpoints within each hospital. Finally, quality indicators from each center are calculated using PHT and the mean result along with the individual results plotted.

DISCUSSION AND CONCLUSION

PHT and FAIR data principles can efficiently calculate quality indicators in a privacy-preserving federated approach and the work can be scaled up both nationally and internationally. Despite this, application of the methodology was largely hampered by ELSI issues. However, the lessons learned from this study can provide other hospitals and researchers to adapt to the process easily and take effective measures in building quality of care infrastructures.

Associations of Maternal Educational Level, Proximity to Green Space During Pregnancy, and Gestational Diabetes With Body Mass Index From Infancy to Early Adulthood: A Proof-of-Concept Federated Analysis in 18 Birth Cohorts

International sharing of cohort data for research is important and challenging. We explored the feasibility of multicohort federated analyses by examining associations between 3 pregnancy exposures (maternal education, exposure to green vegetation, and gestational diabetes) and offspring body mass index (BMI) from infancy to age 17 years. We used data from 18 cohorts (n = 206,180 mother-child pairs) from the EU Child Cohort Network and derived BMI at ages 0-1, 2-3, 4-7, 8-13, and 14-17 years. Associations were estimated using linear regression via 1-stage individual participant data meta-analysis using DataSHIELD. Associations between lower maternal education and higher child BMI emerged from age 4 and increased with age (difference in BMI z score comparing low with high education, at age 2-3 years = 0.03 (95% confidence interval (CI): 0.00, 0.05), at 4-7 years = 0.16 (95% CI: 0.14, 0.17), and at 8-13 years = 0.24 (95% CI: 0.22, 0.26)). Gestational diabetes was positively associated with BMI from age 8 years (BMI z score difference = 0.18, 95% CI: 0.12, 0.25) but not at younger ages; however, associations attenuated towards the null when restricted to cohorts that measured gestational diabetes via universal screening. Exposure to green vegetation was weakly associated with higher BMI up to age 1 year but not at older ages. Opportunities of cross-cohort federated analyses are discussed.

DistSNE: Distributed computing and online visualization of DNA methylation-based central nervous system tumor classification

The current state-of-the-art analysis of central nervous system (CNS) tumors through DNA methylation profiling relies on the tumor classifier developed by Capper and colleagues, which centrally harnesses DNA methylation data provided by users. Here, we present a distributed-computing-based approach for CNS tumor classification that achieves a comparable performance to centralized systems while safeguarding privacy. We utilize the t-distributed neighborhood embedding (t-SNE) model for dimensionality reduction and visualization of tumor classification results in two-dimensional graphs in a distributed approach across multiple sites (DistSNE). DistSNE provides an intuitive web interface (https://gin-tsne.med.uni-giessen.de) for user-friendly local data management and federated methylome-based tumor classification calculations for multiple collaborators in a DataSHIELD environment. The freely accessible web interface supports convenient data upload, result review, and summary report generation. Importantly, increasing sample size as achieved through distributed access to additional datasets allows DistSNE to improve cluster analysis and enhance predictive power. Collectively, DistSNE enables a simple and fast classification of CNS tumors using large-scale methylation data from distributed sources, while maintaining the privacy and allowing easy and flexible network expansion to other institutes. This approach holds great potential for advancing human brain tumor classification and fostering collaborative precision medicine in neuro-oncology.

Green space exposure and blood DNA methylation at birth and in childhood - A multi-cohort study

Green space exposure has been associated with improved mental, physical and general health. However, the underlying biological mechanisms remain largely unknown. The aim of this study was to investigate the association between green space exposure and cord and child blood DNA methylation. Data from eight European birth cohorts with a total of 2,988 newborns and 1,849 children were used. Two indicators of residential green space exposure were assessed: (i) surrounding greenness (satellite-based Normalized Difference Vegetation Index (NDVI) in buffers of 100 m and 300 m) and (ii) proximity to green space (having a green space ≥ 5,000 m2 within a distance of 300 m). For these indicators we assessed two exposure windows: (i) pregnancy, and (ii) the period from pregnancy to child blood DNA methylation assessment, named as cumulative exposure. DNA methylation was measured with the Illumina 450K or EPIC arrays. To identify differentially methylated positions (DMPs) we fitted robust linear regression models between pregnancy green space exposure and cord blood DNA methylation and between cumulative green space exposure and child blood DNA methylation. Two sensitivity analyses were conducted: (i) without adjusting for cellular composition, and (ii) adjusting for air pollution. Cohort results were combined through fixed-effect inverse variance weighted meta-analyses. Differentially methylated regions (DMRs) were identified from meta-analysed results using the Enmix-combp and DMRcate methods. There was no statistical evidence of pregnancy or cumulative exposures associating with any DMP (False Discovery Rate, FDR, p-value < 0.05). However, surrounding greenness exposure was inversely associated with four DMRs (three in cord blood and one in child blood) annotated to ADAMTS2, KCNQ1DN, SLC6A12 and SDK1 genes. Results did not change substantially in the sensitivity analyses. Overall, we found little evidence of the association between green space exposure and blood DNA methylation. Although we identified associations between surrounding greenness exposure with four DMRs, these findings require replication.

Health data space nodes for privacy-preserving linkage of medical data to support collaborative secondary analyses

Introduction

The potential for secondary use of health data to improve healthcare is currently not fully exploited. Health data is largely kept in isolated data silos and key infrastructure to aggregate these silos into standardized bodies of knowledge is underdeveloped. We describe the development, implementation, and evaluation of a federated infrastructure to facilitate versatile secondary use of health data based on Health Data Space nodes.

Materials and methods

Our proposed nodes are self-contained units that digest data through an extract-transform-load framework that pseudonymizes and links data with privacy-preserving record linkage and harmonizes into a common data model (OMOP CDM). To support collaborative analyses a multi-level feature store is also implemented. A feasibility experiment was conducted to test the infrastructures potential for machine learning operations and deployment of other apps (e.g., visualization). Nodes can be operated in a network at different levels of sharing according to the level of trust within the network.

Results

In a proof-of-concept study, a privacy-preserving registry for heart failure patients has been implemented as a real-world showcase for Health Data Space nodes at the highest trust level, linking multiple data sources including (a) electronical medical records from hospitals, (b) patient data from a telemonitoring system, and (c) data from Austria's national register of deaths. The registry is deployed at the tirol kliniken, a hospital carrier in the Austrian state of Tyrol, and currently includes 5,004 patients, with over 2.9 million measurements, over 574,000 observations, more than 63,000 clinical free text notes, and in total over 5.2 million data points. Data curation and harmonization processes are executed semi-automatically at each individual node according to data sharing policies to ensure data sovereignty, scalability, and privacy. As a feasibility test, a natural language processing model for classification of clinical notes was deployed and tested.

Discussion

The presented Health Data Space node infrastructure has proven to be practicable in a real-world implementation in a live and productive registry for heart failure. The present work was inspired by the European Health Data Space initiative and its spirit to interconnect health data silos for versatile secondary use of health data.

Multi-omics subgroups associated with glycaemic deterioration in type 2 diabetes: an IMI-RHAPSODY Study

Introduction

Type 2 diabetes (T2D) onset, progression and outcomes differ substantially between individuals. Multi-omics analyses may allow a deeper understanding of these differences and ultimately facilitate personalised treatments. Here, in an unsupervised "bottom-up" approach, we attempt to group T2D patients based solely on -omics data generated from plasma.

Methods

Circulating plasma lipidomic and proteomic data from two independent clinical cohorts, Hoorn Diabetes Care System (DCS) and Genetics of Diabetes Audit and Research in Tayside Scotland (GoDARTS), were analysed using Similarity Network Fusion. The resulting patient network was analysed with Logistic and Cox regression modelling to explore relationships between plasma -omic profiles and clinical characteristics.

Results

From a total of 1,134 subjects in the two cohorts, levels of 180 circulating plasma lipids and 1195 proteins were used to separate patients into two subgroups. These differed in terms of glycaemic deterioration (Hazard Ratio=0.56;0.73), insulin sensitivity and secretion (C-peptide, p=3.7e-11;2.5e-06, DCS and GoDARTS, respectively; Homeostatic model assessment 2 (HOMA2)-B; -IR; -S, p=0.0008;4.2e-11;1.1e-09, only in DCS). The main molecular signatures separating the two groups included triacylglycerols, sphingomyelin, testican-1 and interleukin 18 receptor.

Conclusions

Using an unsupervised network-based fusion method on plasma lipidomics and proteomics data from two independent cohorts, we were able to identify two subgroups of T2D patients differing in terms of disease severity. The molecular signatures identified within these subgroups provide insights into disease mechanisms and possibly new prognostic markers for T2D.

Urban environment in pregnancy and postpartum depression: An individual participant data meta-analysis of 12 European birth cohorts

BACKGROUND

Urban environmental exposures associate with adult depression, but it is unclear whether they are associated to postpartum depression (PPD).

OBJECTIVES

We investigated associations between urban environment exposures during pregnancy and PPD.

METHODS

We included women with singleton deliveries to liveborn children from 12 European birth cohorts (N with minimum one exposure = 30,772, analysis N range 17,686-30,716 depending on exposure; representing 26-46 % of the 66,825 eligible women). We estimated maternal exposure during pregnancy to ambient air pollution with nitrogen dioxide (NO2) and particulate matter (PM2.5 and PM10), road traffic noise (Lden), natural spaces (Normalised Difference Vegetation Index; NDVI, proximity to major green or blue spaces) and built environment (population density, facility richness and walkability). Maternal PPD was assessed 3-18 months after birth using self-completed questionnaires. We used adjusted logistic regression models to estimate cohort-specific associations between each exposure and PPD and combined results via meta-analysis using DataSHIELD.

RESULTS

Of the 30,772 women included, 3,078 (10 %) reported having PPD. Exposure to PM10 was associated with slightly increased odds of PPD (adjusted odd ratios (OR) of 1.08 [95 % Confidence Intervals (CI): 0.99, 1.17] per inter quartile range increment of PM10) whilst associations for exposure to NO2 and PM2.5 were close to null. Exposure to high levels of road traffic noise (≥65 dB vs. < 65 dB) was associated with an OR of 1.12 [CI: 0.95, 1.32]. Associations between green spaces and PPD were close to null; whilst proximity to major blue spaces was associated with increased risk of PPD (OR 1.12, 95 %CI: 1.00, 1.26). All associations between built environment and PPD were close to null. Multiple exposure models showed similar results.

DISCUSSION

The study findings suggest that exposure to PM10, road traffic noise and blue spaces in pregnancy may increase PPD risk, however future studies should explore this causally.

CODA: an open-source platform for federated analysis and machine learning on distributed healthcare data

OBJECTIVES

Distributed computations facilitate multi-institutional data analysis while avoiding the costs and complexity of data pooling. Existing approaches lack crucial features, such as built-in medical standards and terminologies, no-code data visualizations, explicit disclosure control mechanisms, and support for basic statistical computations, in addition to gradient-based optimization capabilities.

MATERIALS AND METHODS

We describe the development of the Collaborative Data Analysis (CODA) platform, and the design choices undertaken to address the key needs identified during our survey of stakeholders. We use a public dataset (MIMIC-IV) to demonstrate end-to-end multi-modal FL using CODA. We assessed the technical feasibility of deploying the CODA platform at 9 hospitals in Canada, describe implementation challenges, and evaluate its scalability on large patient populations.

RESULTS

The CODA platform was designed, developed, and deployed between January 2020 and January 2023. Software code, documentation, and technical documents were released under an open-source license. Multi-modal federated averaging is illustrated using the MIMIC-IV and MIMIC-CXR datasets. To date, 8 out of the 9 participating sites have successfully deployed the platform, with a total enrolment of >1M patients. Mapping data from legacy systems to FHIR was the biggest barrier to implementation.

DISCUSSION AND CONCLUSION

The CODA platform was developed and successfully deployed in a public healthcare setting in Canada, with heterogeneous information technology systems and capabilities. Ongoing efforts will use the platform to develop and prospectively validate models for risk assessment, proactive monitoring, and resource usage. Further work will also make tools available to facilitate migration from legacy formats to FHIR and DICOM.

A Federated Database for Obesity Research: An IMI-SOPHIA Study

Obesity is considered by many as a lifestyle choice rather than a chronic progressive disease. The Innovative Medicines Initiative (IMI) SOPHIA (Stratification of Obesity Phenotypes to Optimize Future Obesity Therapy) project is part of a momentum shift aiming to provide better tools for the stratification of people with obesity according to disease risk and treatment response. One of the challenges to achieving these goals is that many clinical cohorts are siloed, limiting the potential of combined data for biomarker discovery. In SOPHIA, we have addressed this challenge by setting up a federated database building on open-source DataSHIELD technology. The database currently federates 16 cohorts that are accessible via a central gateway. The database is multi-modal, including research studies, clinical trials, and routine health data, and is accessed using the R statistical programming environment where statistical and machine learning analyses can be performed at a distance without any disclosure of patient-level data. We demonstrate the use of the database by providing a proof-of-concept analysis, performing a federated linear model of BMI and systolic blood pressure, pooling all data from 16 studies virtually without any analyst seeing individual patient-level data. This analysis provided similar point estimates compared to a meta-analysis of the 16 individual studies. Our approach provides a benchmark for reproducible, safe federated analyses across multiple study types provided by multiple stakeholders.

Strengths and limitations of non-disclosive data analysis: a comparison of breast cancer survival classifiers using VisualSHIELD

Preserving data privacy is an important concern in the research use of patient data. The DataSHIELD suite enables privacy-aware advanced statistical analysis in a federated setting. Despite its many applications, it has a few open practical issues: the complexity of hosting a federated infrastructure, the performance penalty imposed by the privacy-preserving constraints, and the ease of use by non-technical users. In this work, we describe a case study in which we review different breast cancer classifiers and report our findings about the limits and advantages of such non-disclosive suite of tools in a realistic setting. Five independent gene expression datasets of breast cancer survival were downloaded from Gene Expression Omnibus (GEO) and pooled together through the federated infrastructure. Three previously published and two newly proposed 5-year cancer-free survival risk score classifiers were trained in a federated environment, and an additional reference classifier was trained with unconstrained data access. The performance of these six classifiers was systematically evaluated, and the results show that i) the published classifiers do not generalize well when applied to patient cohorts that differ from those used to develop them; ii) among the methods we tried, the classification using logistic regression worked better on average, closely followed by random forest; iii) the unconstrained version of the logistic regression classifier outperformed the federated version by 4% on average. Reproducibility of our experiments is ensured through the use of VisualSHIELD, an open-source tool that augments DataSHIELD with new functions, a standardized deployment procedure, and a simple graphical user interface.

Twenty years of network meta-analysis: Continuing controversies and recent developments

Network meta-analysis (NMA) is an extension of pairwise meta-analysis (PMA) which combines evidence from trials on multiple treatments in connected networks. NMA delivers internally consistent estimates of relative treatment efficacy, needed for rational decision making. Over its first 20 years NMA's use has grown exponentially, with applications in both health technology assessment (HTA), primarily re-imbursement decisions and clinical guideline development, and clinical research publications. This has been a period of transition in meta-analysis, first from its roots in educational and social psychology, where large heterogeneous datasets could be explored to find effect modifiers, to smaller pairwise meta-analyses in clinical medicine on average with less than six studies. This has been followed by narrowly-focused estimation of the effects of specific treatments at specific doses in specific populations in sparse networks, where direct comparisons are unavailable or informed by only one or two studies. NMA is a powerful and well-established technique but, in spite of the exponential increase in applications, doubts about the reliability and validity of NMA persist. Here we outline the continuing controversies, and review some recent developments. We suggest that heterogeneity should be minimized, as it poses a threat to the reliability of NMA which has not been fully appreciated, perhaps because it has not been seen as a problem in PMA. More research is needed on the extent of heterogeneity and inconsistency in datasets used for decision making, on formal methods for making recommendations based on NMA, and on the further development of multi-level network meta-regression.

Alpha-1-antitrypsin-deficiency is associated with lower cardiovascular risk: an approach based on federated learning

BACKGROUND

Chronic obstructive pulmonary disease (COPD) is an inflammatory multisystemic disease caused by environmental exposures and/or genetic factors. Inherited alpha-1-antitrypsin deficiency (AATD) is one of the best recognized genetic factors increasing the risk for an early onset COPD with emphysema. The aim of this study was to gain a better understanding of the associations between comorbidities and specific biomarkers in COPD patients with and without AATD to enable future investigations aimed, for example, at identifying risk factors or improving care.

METHODS

We focused on cardiovascular comorbidities, blood high sensitivity troponin (hs-troponin) and lipid profiles in COPD patients with and without AATD. We used clinical data from six German University Medical Centres of the MIRACUM (Medical Informatics Initiative in Research and Medicine) consortium. The codes for the international classification of diseases (ICD) were used for COPD as a main diagnosis and for comorbidities and blood laboratory data were obtained. Data analyses were based on the DataSHIELD framework.

RESULTS

Out of 112,852 visits complete information was available for 43,057 COPD patients. According to our findings, 746 patients with AATD (1.73%) showed significantly lower total blood cholesterol levels and less cardiovascular comorbidities than non-AATD COPD patients. Moreover, after adjusting for the confounder factors, such as age, gender, and nicotine abuse, we confirmed that hs-troponin is a suitable predictor of overall mortality in COPD patients. The comorbidities associated with AATD in the current study differ from other studies, which may reflect geographic and population-based differences as well as the heterogeneous characteristics of AATD.

CONCLUSION

The concept of MIRACUM is suitable for the analysis of a large healthcare database. This study provided evidence that COPD patients with AATD have a lower cardiovascular risk and revealed that hs-troponin is a predictor for hospital mortality in individuals with COPD.

Will it run?-A proof of concept for smoke testing decentralized data analytics experiments

The growing interest in data-driven medicine, in conjunction with the formation of initiatives such as the European Health Data Space (EHDS) has demonstrated the need for methodologies that are capable of facilitating privacy-preserving data analysis. Distributed Analytics (DA) as an enabler for privacy-preserving analysis across multiple data sources has shown its potential to support data-intensive research. However, the application of DA creates new challenges stemming from its distributed nature, such as identifying single points of failure (SPOFs) in DA tasks before their actual execution. Failing to detect such SPOFs can, for example, result in improper termination of the DA code, necessitating additional efforts from multiple stakeholders to resolve the malfunctions. Moreover, these malfunctions disrupt the seamless conduct of DA and entail several crucial consequences, including technical obstacles to resolve the issues, potential delays in research outcomes, and increased costs. In this study, we address this challenge by introducing a concept based on a method called Smoke Testing, an initial and foundational test run to ensure the operability of the analysis code. We review existing DA platforms and systematically extract six specific Smoke Testing criteria for DA applications. With these criteria in mind, we create an interactive environment called Development Environment for AuTomated and Holistic Smoke Testing of Analysis-Runs (DEATHSTAR), which allows researchers to perform Smoke Tests on their DA experiments. We conduct a user-study with 29 participants to assess our environment and additionally apply it to three real use cases. The results of our evaluation validate its effectiveness, revealing that 96.6% of the analyses created and (Smoke) tested by participants using our approach successfully terminated without any errors. Thus, by incorporating Smoke Testing as a fundamental method, our approach helps identify potential malfunctions early in the development process, ensuring smoother data-driven research within the scope of DA. Through its flexibility and adaptability to diverse real use cases, our solution enables more robust and efficient development of DA experiments, which contributes to their reliability.

Reconciling the biomedical data commons and the GDPR: three lessons from the EUCAN ELSI collaboratory

The coming-into-force of the EU General Data Protection Regulation (GDPR) is a watershed moment in the legal recognition of enforceable rights to informational self-determination. The rapid evolution of legal requirements applicable to data use, however, has the potential to outstrip the capabilities of networks of biomedical data users to respond to the shifting norms. It can also delegitimate established institutional bodies that are responsible for assessing and authorising the downstream use of data, including research ethics committees and institutional data custodians. These burdens are especially pronounced for clinical and research networks that are of transnational scale, because the legal compliance burden for outbound international data transfers from the EEA is especially high. Legislatures, courts, and regulators in the EU should therefore implement the following three legal changes. First, the responsibilities of particular actors in a data sharing network should be delimited through the contractual allocation of responsibilities between collaborators. Second, the use of data through secure data processing environments should not trigger the international transfer provisions of the GDPR. Third, the use of federated data analysis methodologies that do not provide analysis nodes or downstream users access to identifiable personal data as part of the outputs of those analyses should not be considered circumstances of joint controllership, nor lead to the users of non-identifiable data to be considered controllers or processors. These small clarifications of, or modifications to, the GDPR would facilitate the exchange of biomedical data amongst clinicians and researchers.

Sleep duration in preschool age and later behavioral and cognitive outcomes: an individual participant data meta-analysis in five European cohorts

Short sleep duration has been linked to adverse behavioral and cognitive outcomes in schoolchildren, but few studies examined this relation in preschoolers. We aimed to investigate the association between parent-reported sleep duration at 3.5 years and behavioral and cognitive outcomes at 5 years in European children. We used harmonized data from five cohorts of the European Union Child Cohort Network: ALSPAC, SWS (UK); EDEN, ELFE (France); INMA (Spain). Associations were estimated through DataSHIELD using adjusted generalized linear regression models fitted separately for each cohort and pooled with random-effects meta-analysis. Behavior was measured with the Strengths and Difficulties Questionnaire. Language and non-verbal intelligence were assessed by the Wechsler Preschool and Primary Scale of Intelligence or the McCarthy Scales of Children's Abilities. Behavioral and cognitive analyses included 11,920 and 2981 children, respectively (34.0%/13.4% of the original sample). In meta-analysis, longer mean sleep duration per day at 3.5 years was associated with lower mean internalizing and externalizing behavior percentile scores at 5 years (adjusted mean difference: - 1.27, 95% CI [- 2.22, - 0.32] / - 2.39, 95% CI [- 3.04, - 1.75]). Sleep duration and language or non-verbal intelligence showed trends of inverse associations, however, with imprecise estimates (adjusted mean difference: - 0.28, 95% CI [- 0.83, 0.27] / - 0.42, 95% CI [- 0.99, 0.15]). This individual participant data meta-analysis suggests that longer sleep duration in preschool age may be important for children's later behavior and highlight the need for larger samples for robust analyses of cognitive outcomes. Findings could be influenced by confounding or reverse causality and require replication.

An innovative technological infrastructure for managing SARS-CoV-2 data across different cohorts in compliance with General Data Protection Regulation

Background

The ORCHESTRA project, funded by the European Commission, aims to create a pan-European cohort built on existing and new large-scale population cohorts to help rapidly advance the knowledge related to the prevention of the SARS-CoV-2 infection and the management of COVID-19 and its long-term sequelae. The integration and analysis of the very heterogeneous health data pose the challenge of building an innovative technological infrastructure as the foundation of a dedicated framework for data management that should address the regulatory requirements such as the General Data Protection Regulation (GDPR).

Methods

The three participating Supercomputing European Centres (CINECA - Italy, CINES - France and HLRS - Germany) designed and deployed a dedicated infrastructure to fulfil the functional requirements for data management to ensure sensitive biomedical data confidentiality/privacy, integrity, and security. Besides the technological issues, many methodological aspects have been considered: Berlin Institute of Health (BIH), Charité provided its expertise both for data protection, information security, and data harmonisation/standardisation.

Results

The resulting infrastructure is based on a multi-layer approach that integrates several security measures to ensure data protection. A centralised Data Collection Platform has been established in the Italian National Hub while, for the use cases in which data sharing is not possible due to privacy restrictions, a distributed approach for Federated Analysis has been considered. A Data Portal is available as a centralised point of access for non-sensitive data and results, according to findability, accessibility, interoperability, and reusability (FAIR) data principles. This technological infrastructure has been used to support significative data exchange between population cohorts and to publish important scientific results related to SARS-CoV-2.

Conclusions

Considering the increasing demand for data usage in accordance with the requirements of the GDPR regulations, the experience gained in the project and the infrastructure released for the ORCHESTRA project can act as a model to manage future public health threats. Other projects could benefit from the results achieved by ORCHESTRA by building upon the available standardisation of variables, design of the architecture, and process used for GDPR compliance.

Regional and temporal differences in the associations between cardiovascular disease and its classic risk factors: An analysis of 49 cohorts from 11 European countries

AIMS

The regional and temporal differences in the associations between cardiovascular disease (CVD) and its classic risk factors are unknown. The current study examined these associations in different European regions over a 30-year period.

METHODS

The study sample comprised 553818 individuals from 49 cohorts in 11 European countries (baseline: 1982-2012) who were followed up for a maximum of 10 years. Risk factors (sex, smoking, diabetes, non-HDL [high-density lipoprotein] cholesterol, systolic blood pressure [BP], and body mass index [BMI]) and CVD events (coronary heart disease or stroke) were harmonized across cohorts. Risk factor-outcome associations were analysed using multivariable-adjusted Cox regression models, and differences in associations were assessed using meta-regression.

RESULTS

The differences in the risk factor-CVD associations between central Europe, northern Europe, southern Europe, and the United Kingdom were generally small. Men had a slightly higher hazard ratio (HR) in southern Europe (p = 0.043 for overall difference) and those with diabetes had a slightly lower HR in central Europe (p = 0.022 for overall difference) compared with the other regions. Of the six CVD risk factors, minor HR decreases per decade were observed for non-HDL cholesterol (7% per mmol/L; 95% confidence interval [CI], 3-10%) and systolic BP (4% per 20 mmHg; 95% CI, 1-8%), while a minor HR increase per decade was observed for BMI (7% per 10 kg/m2; 95% CI, 1-13%).

CONCLUSION

The results demonstrate that all classic CVD risk factors are still relevant in Europe, irrespective of regional area. Preventive strategies should focus on risk factors with the greatest population attributable risk.

Federated causal inference based on real-world observational data sources: application to a SARS-CoV-2 vaccine effectiveness assessment

INTRODUCTION

Causal inference helps researchers and policy-makers to evaluate public health interventions. When comparing interventions or public health programs by leveraging observational sensitive individual-level data from populations crossing jurisdictional borders, a federated approach (as opposed to a pooling data approach) can be used. Approaching causal inference by re-using routinely collected observational data across different regions in a federated manner, is challenging and guidance is currently lacking. With the aim of filling this gap and allowing a rapid response in the case of a next pandemic, a methodological framework to develop studies attempting causal inference using federated cross-national sensitive observational data, is described and showcased within the European BeYond-COVID project.

METHODS

A framework for approaching federated causal inference by re-using routinely collected observational data across different regions, based on principles of legal, organizational, semantic and technical interoperability, is proposed. The framework includes step-by-step guidance, from defining a research question, to establishing a causal model, identifying and specifying data requirements in a common data model, generating synthetic data, and developing an interoperable and reproducible analytical pipeline for distributed deployment. The conceptual and instrumental phase of the framework was demonstrated and an analytical pipeline implementing federated causal inference was prototyped using open-source software in preparation for the assessment of real-world effectiveness of SARS-CoV-2 primary vaccination in preventing infection in populations spanning different countries, integrating a data quality assessment, imputation of missing values, matching of exposed to unexposed individuals based on confounders identified in the causal model and a survival analysis within the matched population.

RESULTS

The conceptual and instrumental phase of the proposed methodological framework was successfully demonstrated within the BY-COVID project. Different Findable, Accessible, Interoperable and Reusable (FAIR) research objects were produced, such as a study protocol, a data management plan, a common data model, a synthetic dataset and an interoperable analytical pipeline.

CONCLUSIONS

The framework provides a systematic approach to address federated cross-national policy-relevant causal research questions based on sensitive population, health and care data in a privacy-preserving and interoperable way. The methodology and derived research objects can be re-used and contribute to pandemic preparedness.

FAIR, ethical, and coordinated data sharing for COVID-19 response: a scoping review and cross-sectional survey of COVID-19 data sharing platforms and registries

Data sharing is central to the rapid translation of research into advances in clinical medicine and public health practice. In the context of COVID-19, there has been a rush to share data marked by an explosion of population-specific and discipline-specific resources for collecting, curating, and disseminating participant-level data. We conducted a scoping review and cross-sectional survey to identify and describe COVID-19-related platforms and registries that harmonise and share participant-level clinical, omics (eg, genomic and metabolomic data), imaging data, and metadata. We assess how these initiatives map to the best practices for the ethical and equitable management of data and the findable, accessible, interoperable, and reusable (FAIR) principles for data resources. We review gaps and redundancies in COVID-19 data-sharing efforts and provide recommendations to build on existing synergies that align with frameworks for effective and equitable data reuse. We identified 44 COVID-19-related registries and 20 platforms from the scoping review. Data-sharing resources were concentrated in high-income countries and siloed by comorbidity, body system, and data type. Resources for harmonising and sharing clinical data were less likely to implement FAIR principles than those sharing omics or imaging data. Our findings are that more data sharing does not equate to better data sharing, and the semantic and technical interoperability of platforms and registries harmonising and sharing COVID-19-related participant-level data needs to improve to facilitate the global collaboration required to address the COVID-19 crisis.

Developing and testing a protocol using a common data model for federated collection and analysis of national perinatal health indicators in Europe

Context: International comparisons of the health of mothers and babies provide essential benchmarks for guiding health practice and policy, but statistics are not routinely compiled in a comparable way. These data are especially critical during health emergencies, such as the coronavirus disease (COVID-19) pandemic. The Population Health Information Research Infrastructure (PHIRI) project aimed to promote the exchange of population data in Europe and included a Use Case on perinatal health. Objective: To develop and test a protocol for federated analysis of population birth data in Europe. Methods: The Euro-Peristat network with participants from 31 countries developed a Common Data Model (CDM) and R scripts to exchange and analyse aggregated data on perinatal indicators. Building on recommended Euro-Peristat indicators, complemented by a three-round consensus process, the network specified variables for a CDM and common outputs. The protocol was tested using routine birth data for 2015 to 2020; a survey was conducted assessing data provider experiences and opinions. Results: The CDM included 17 core data items for the testing phase and 18 for a future expanded phase. 28 countries and the four UK nations created individual person-level databases and ran R scripts to produce anonymous aggregate tables. Seven had all core items, 17 had 13-16, while eight had ≤12. Limitations were not having all items in the same database, required for this protocol. Infant death and mode of birth were most frequently missing. Countries took from under a day to several weeks to set up the CDM, after which the protocol was easy and quick to use. Conclusion: This open-source protocol enables rapid production and analysis of perinatal indicators and constitutes a roadmap for a sustainable European information system. It also provides minimum standards for improving national data systems and can be used in other countries to facilitate comparison of perinatal indicators.

Rare Diseases in Hospital Information Systems-An Interoperable Methodology for Distributed Data Quality Assessments

BACKGROUND

Multisite research networks such as the project "Collaboration on Rare Diseases" connect various hospitals to obtain sufficient data for clinical research. However, data quality (DQ) remains a challenge for the secondary use of data recorded in different health information systems. High levels of DQ as well as appropriate quality assessment methods are needed to support the reuse of such distributed data.

OBJECTIVES

The aim of this work is the development of an interoperable methodology for assessing the quality of data recorded in heterogeneous sources to improve the quality of rare disease (RD) documentation and support clinical research.

METHODS

We first developed a conceptual framework for DQ assessment. Using this theoretical guidance, we implemented a software framework that provides appropriate tools for calculating DQ metrics and for generating local as well as cross-institutional reports. We further applied our methodology on synthetic data distributed across multiple hospitals using Personal Health Train. Finally, we used precision and recall as metrics to validate our implementation.

RESULTS

Four DQ dimensions were defined and represented as disjunct ontological categories. Based on these top dimensions, 9 DQ concepts, 10 DQ indicators, and 25 DQ parameters were developed and applied to different data sets. Randomly introduced DQ issues were all identified and reported automatically. The generated reports show the resulting DQ indicators and detected DQ issues.

CONCLUSION

We have shown that our approach yields promising results, which can be used for local and cross-institutional DQ assessments. The developed frameworks provide useful methods for interoperable and privacy-preserving assessments of DQ that meet the specified requirements. This study has demonstrated that our methodology is capable of detecting DQ issues such as ambiguity or implausibility of coded diagnoses. It can be used for DQ benchmarking to improve the quality of RD documentation and to support clinical research on distributed data.

Impact of Sex- and Gender-Related Factors on Length of Stay Following Non-ST-Segment-Elevation Myocardial Infarction: A Multicountry Analysis

Background Gender-related factors are psycho-socio-cultural characteristics and are associated with adverse clinical outcomes in acute myocardial infarction, independent of sex. Whether sex- and gender-related factors contribute to the substantial heterogeneity in hospital length of stay (LOS) among patients with non-ST-segment-elevation myocardial infarction remains unknown. Methods and Results This observational cohort study combined and analyzed data from the GENESIS-PRAXY (Gender and Sex Determinants of Cardiovascular Disease: From Bench to Beyond Premature Acute Coronary Syndrome study), EVA (Endocrine Vascular Disease Approach study), and VIRGO (Variation in Recovery: Role of Gender on Outcomes of Young AMI [Acute Myocardial Infarction] Patients study) cohorts of adults hospitalized across Canada, the United States, Switzerland, Italy, Spain, and Australia for non-ST-segment-elevation myocardial infarction. In total, 5219 participants were assessed for eligibility. Sixty-three patients were excluded for missing LOS, and 2938 were excluded because of no non-ST-segment-elevation myocardial infarction diagnosis. In total, 2218 participants were analyzed (66% women; mean±SD age, 48.5±7.9 years; 67.8% in the United States). Individuals with longer LOS (51%) were more likely to be White race, were more likely to have diabetes, hypertension, and a lower income, and were less likely to be employed and have completed secondary education. No univariate association between sex and LOS was observed. In the adjusted multivariable model, age (0.62 d/10 y; P<0.001), unemployment (0.63 days; P=0.01), and some of countries included relative to Canada (Italy, 4.1 days; Spain, 1.7 days; and the United States, -1.0 days; all P<0.001) were independently associated with longer LOS. Medical history mediated the effect of employment on LOS. No interaction between sex and employment was observed. Longer LOS was associated with increased 12-month all-cause mortality. Conclusions Older age, unemployment, and country of hospitalization were independent predictors of LOS, regardless of sex. Individuals employed with non-ST-segment-elevation myocardial infarction were more likely to experience shorter LOS. Sociocultural factors represent a potential target for improvement in health care expenditure and resource allocation.

Ten Topics to Get Started in Medical Informatics Research

The vast and heterogeneous data being constantly generated in clinics can provide great wealth for patients and research alike. The quickly evolving field of medical informatics research has contributed numerous concepts, algorithms, and standards to facilitate this development. However, these difficult relationships, complex terminologies, and multiple implementations can present obstacles for people who want to get active in the field. With a particular focus on medical informatics research conducted in Germany, we present in our Viewpoint a set of 10 important topics to improve the overall interdisciplinary communication between different stakeholders (eg, physicians, computational experts, experimentalists, students, patient representatives). This may lower the barriers to entry and offer a starting point for collaborations at different levels. The suggested topics are briefly introduced, then general best practice guidance is given, and further resources for in-depth reading or hands-on tutorials are recommended. In addition, the topics are set to cover current aspects and open research gaps of the medical informatics domain, including data regulations and concepts; data harmonization and processing; and data evaluation, visualization, and dissemination. In addition, we give an example on how these topics can be integrated in a medical informatics curriculum for higher education. By recognizing these topics, readers will be able to (1) set clinical and research data into the context of medical informatics, understanding what is possible to achieve with data or how data should be handled in terms of data privacy and storage; (2) distinguish current interoperability standards and obtain first insights into the processes leading to effective data transfer and analysis; and (3) value the use of newly developed technical approaches to utilize the full potential of clinical data.

An integrative approach for the analysis of risk and health across the life course: challenges, innovations, and opportunities for life course research

Life course epidemiology seeks to understand the intricate relationships between risk factors and health outcomes across different stages of life to inform prevention and intervention strategies to optimize health throughout the lifespan. However, extant evidence has predominantly been based on separate analyses of data from individual birth cohorts or panel studies, which may not be sufficient to unravel the complex interplay of risk and health across different contexts. We highlight the importance of a multi-study perspective that enables researchers to: (a) Compare and contrast findings from different contexts and populations, which can help identify generalizable patterns and context-specific factors; (b) Examine the robustness of associations and the potential for effect modification by factors such as age, sex, and socioeconomic status; and (c) Improve statistical power and precision by pooling data from multiple studies, thereby allowing for the investigation of rare exposures and outcomes. This integrative framework combines the advantages of multi-study data with a life course perspective to guide research in understanding life course risk and resilience on adult health outcomes by: (a) Encouraging the use of harmonized measures across studies to facilitate comparisons and synthesis of findings; (b) Promoting the adoption of advanced analytical techniques that can accommodate the complexities of multi-study, longitudinal data; and (c) Fostering collaboration between researchers, data repositories, and funding agencies to support the integration of longitudinal data from diverse sources. An integrative approach can help inform the development of individualized risk scores and personalized interventions to promote health and well-being at various life stages.

Moving Toward Findable, Accessible, Interoperable, Reusable Practices in Epidemiologic Research

Data sharing is essential for reproducibility of epidemiologic research, replication of findings, pooled analyses in consortia efforts, and maximizing study value to address multiple research questions. However, barriers related to confidentiality, costs, and incentives often limit the extent and speed of data sharing. Epidemiological practices that follow Findable, Accessible, Interoperable, Reusable (FAIR) principles can address these barriers by making data resources findable with the necessary metadata, accessible to authorized users, and interoperable with other data, to optimize the reuse of resources with appropriate credit to its creators. We provide an overview of these principles and describe approaches for implementation in epidemiology. Increasing degrees of FAIRness can be achieved by moving data and code from on-site locations to remote, accessible ("Cloud") data servers, using machine-readable and nonproprietary files, and developing open-source code. Adoption of these practices will improve daily work and collaborative analyses and facilitate compliance with data sharing policies from funders and scientific journals. Achieving a high degree of FAIRness will require funding, training, organizational support, recognition, and incentives for sharing research resources, both data and code. However, these costs are outweighed by the benefits of making research more reproducible, impactful, and equitable by facilitating the reuse of precious research resources by the scientific community.

Patient data for commercial companies? An ethical framework for sharing patients' data with for-profit companies for research

BACKGROUND

Research using data from medical care promises to advance medical science and improve healthcare. Academia is not the only sector that expects such research to be of great benefit. The research-based health industry is also interested in so-called 'real-world' health data to develop new drugs, medical technologies or data-based health applications. While access to medical data is handled very differently in different countries, and some empirical data suggest people are uncomfortable with the idea of companies accessing health information, this paper aims to advance the ethical debate about secondary use of medical data generated in the public healthcare sector by for-profit companies for medical research (ReuseForPro).

METHODS

We first clarify some basic concepts and our ethical-normative approach, then discuss and ethically evaluate potential claims and interests of relevant stakeholders: patients as data subjects in the public healthcare system, for-profit companies, the public, and physicians and their healthcare institutions. Finally, we address the tensions between legitimate claims of different stakeholders in order to suggest conditions that might ensure ethically sound ReuseForPro.

RESULTS

We conclude that there are good reasons to grant for-profit companies access to medical data if they meet certain conditions: among others they need to respect patients' informational rights and their actions need to be compatible with the public's interest in health benefit from ReuseForPro.

Profile of the multicenter cohort of the German Cancer Consortium's Clinical Communication Platform

Treatment concepts in oncology are becoming increasingly personalized and diverse. Successively, changes in standards of care mandate continuous monitoring of patient pathways and clinical outcomes based on large, representative real-world data. The German Cancer Consortium's (DKTK) Clinical Communication Platform (CCP) provides such opportunity. Connecting fourteen university hospital-based cancer centers, the CCP relies on a federated IT-infrastructure sourcing data from facility-based cancer registry units and biobanks. Federated analyses resulted in a cohort of 600,915 patients, out of which 232,991 were incident since 2013 and for which a comprehensive documentation is available. Next to demographic data (i.e., age at diagnosis: 2.0% 0-20 years, 8.3% 21-40 years, 30.9% 41-60 years, 50.1% 61-80 years, 8.8% 81+ years; and gender: 45.2% female, 54.7% male, 0.1% other) and diagnoses (five most frequent tumor origins: 22,523 prostate, 18,409 breast, 15,575 lung, 13,964 skin/malignant melanoma, 9005 brain), the cohort dataset contains information about therapeutic interventions and response assessments and is connected to 287,883 liquid and tissue biosamples. Focusing on diagnoses and therapy-sequences, showcase analyses of diagnosis-specific sub-cohorts (pancreas, larynx, kidney, thyroid gland) demonstrate the analytical opportunities offered by the cohort's data. Due to its data granularity and size, the cohort is a potential catalyst for translational cancer research. It provides rapid access to comprehensive patient groups and may improve the understanding of the clinical course of various (even rare) malignancies. Therefore, the cohort may serve as a decisions-making tool for clinical trial design and contributes to the evaluation of scientific findings under real-world conditions.

Life course of retrospective harmonization initiatives: key elements to consider

Optimizing research on the developmental origins of health and disease (DOHaD) involves implementing initiatives maximizing the use of the available cohort study data; achieving sufficient statistical power to support subgroup analysis; and using participant data presenting adequate follow-up and exposure heterogeneity. It also involves being able to undertake comparison, cross-validation, or replication across data sets. To answer these requirements, cohort study data need to be findable, accessible, interoperable, and reusable (FAIR), and more particularly, it often needs to be harmonized. Harmonization is required to achieve or improve comparability of the putatively equivalent measures collected by different studies on different individuals. Although the characteristics of the research initiatives generating and using harmonized data vary extensively, all are confronted by similar issues. Having to collate, understand, process, host, and co-analyze data from individual cohort studies is particularly challenging. The scientific success and timely management of projects can be facilitated by an ensemble of factors. The current document provides an overview of the 'life course' of research projects requiring harmonization of existing data and highlights key elements to be considered from the inception to the end of the project.

Federated analyses of multiple data sources in drug safety studies

PURPOSE

Studies of rare side effects of new drugs with limited exposure may require pooling of multiple data sources. Federated Analyses (FA) allow real-time, interactive, centralized statistical processing of individual-level data from different data sets without transfer of sensitive personal data.

METHODS

We review IT-architecture, legal considerations, and statistical methods in FA, based on a Swedish Medical Products Agency methodological development project.

RESULTS

In a review of all post-authorisation safety studies assessed by the EMA during 2019, 74% (20/27 studies) reported issues with lack of precision in spite of mean study periods of 9.3 years. FA could potentially improve precision in such studies. Depending on the statistical model, the federated approach can generate identical results to a standard analysis. FA may be particularly attractive for repeated collaborative projects where data is regularly updated. There are also important limitations. Detailed agreements between involved parties are strongly recommended to anticipate potential issues and conflicts, document a shared understanding of the project, and fully comply with legal obligations regarding ethics and data protection. FA do not remove the data harmonisation step, which remains essential and often cumbersome. Reliable support for technical integration with the local server architecture and security solutions is required. Common statistical methods are available, but adaptations may be required.

CONCLUSIONS

Federated Analyses require competent and active involvement of all collaborating parties but have the potential to facilitate collaboration across institutional and national borders and improve the precision of postmarketing drug safety studies.

A. Sørensen T, Elhakeem A, Santos AC, Pinot de Moira A, Heude B, Iñiguez C, Pizzi C, Simons E, Voerman E, Corpeleijn E, Zariouh F, Santorelli G, Inskip HM, Barros H, Carson J, Harris JR, Nader JL, Ronkainen J, Strandberg-Larsen K, Santa-Marina L, Calas L, Cederkvist L, Popovic M, Charles MA, Welten M, Vrijheid M, Azad M, Subbarao P, Burton P, Mandhane PJ, Huang RC, Wilson RC, Haakma S, Fernández-Barrés S, Turvey S, Santos S, Tough SC, Sebert S, Moraes TJ, Salika T, Jaddoe VWV, Lawlor DA, Nybo Andersen AM. Gestational age at birth and body size from infancy through adolescence: An individual participant data meta-analysis on 253,810 singletons in 16 birth cohort studies

BACKGROUND

Preterm birth is the leading cause of perinatal morbidity and mortality and is associated with adverse developmental and long-term health outcomes, including several cardiometabolic risk factors and outcomes. However, evidence about the association of preterm birth with later body size derives mainly from studies using birth weight as a proxy of prematurity rather than an actual length of gestation. We investigated the association of gestational age (GA) at birth with body size from infancy through adolescence.

METHODS AND FINDINGS

We conducted a two-stage individual participant data (IPD) meta-analysis using data from 253,810 mother-child dyads from 16 general population-based cohort studies in Europe (Denmark, Finland, France, Italy, Norway, Portugal, Spain, the Netherlands, United Kingdom), North America (Canada), and Australasia (Australia) to estimate the association of GA with body mass index (BMI) and overweight (including obesity) adjusted for the following maternal characteristics as potential confounders: education, height, prepregnancy BMI, ethnic background, parity, smoking during pregnancy, age at child's birth, gestational diabetes and hypertension, and preeclampsia. Pregnancy and birth cohort studies from the LifeCycle and the EUCAN-Connect projects were invited and were eligible for inclusion if they had information on GA and minimum one measurement of BMI between infancy and adolescence. Using a federated analytical tool (DataSHIELD), we fitted linear and logistic regression models in each cohort separately with a complete-case approach and combined the regression estimates and standard errors through random-effects study-level meta-analysis providing an overall effect estimate at early infancy (>0.0 to 0.5 years), late infancy (>0.5 to 2.0 years), early childhood (>2.0 to 5.0 years), mid-childhood (>5.0 to 9.0 years), late childhood (>9.0 to 14.0 years), and adolescence (>14.0 to 19.0 years). GA was positively associated with BMI in the first decade of life, with the greatest increase in mean BMI z-score during early infancy (0.02, 95% confidence interval (CI): 0.00; 0.05, p < 0.05) per week of increase in GA, while in adolescence, preterm individuals reached similar levels of BMI (0.00, 95% CI: -0.01; 0.01, p 0.9) as term counterparts. The association between GA and overweight revealed a similar pattern of association with an increase in odds ratio (OR) of overweight from late infancy through mid-childhood (OR 1.01 to 1.02) per week increase in GA. By adolescence, however, GA was slightly negatively associated with the risk of overweight (OR 0.98 [95% CI: 0.97; 1.00], p 0.1) per week of increase in GA. Although based on only four cohorts (n = 32,089) that reached the age of adolescence, data suggest that individuals born very preterm may be at increased odds of overweight (OR 1.46 [95% CI: 1.03; 2.08], p < 0.05) compared with term counterparts. Findings were consistent across cohorts and sensitivity analyses despite considerable heterogeneity in cohort characteristics. However, residual confounding may be a limitation in this study, while findings may be less generalisable to settings in low- and middle-income countries.

CONCLUSIONS

This study based on data from infancy through adolescence from 16 cohort studies found that GA may be important for body size in infancy, but the strength of association attenuates consistently with age. By adolescence, preterm individuals have on average a similar mean BMI to peers born at term.

[Sex- and Regionalspecific Differences in Child Abuse and Violence Before the German Reunification. Results from GESA, a Multi-Cohort Study]

Results from a population-based study suggest sex-specific patterns of self-reported child maltreatment, more frequently reported in former West than East Germany. Aim of the current study was to examine these patterns in two regional samples of the former East- (SHIP, 2008 - 2012) and West German (KORA, 2013 - 2014) population. Child maltreatment was assessed using the Childhood Trauma Screener (CTS). Overall, child maltreatment was less often reported in the East German sample, compared to the West German sample. The most prominent differences were identified in self-rated emotional violence (east 6.1%, west 8.7%), physical violence (east 5.7%, west 10.3%) and physical neglect (east 10.0%, west 19.2%). However, we could not find differences in sex-specific patterns between the East and West German samples. Results were discussed within a historical context, since the events took place before the German reunification in two oppose political systems.

Software Application Profile: ShinyDataSHIELD—an R Shiny application to perform federated non-disclosive data analysis in multicohort studies

Motivation

DataSHIELD is an open-source software infrastructure enabling the analysis of data distributed across multiple databases (federated data) without leaking individuals’ information (non-disclosive). It has applications in many scientific domains, ranging from biosciences to social sciences and including high-throughput genomic studies. R is the language used to interact with (and build) DataSHIELD. This creates difficulties for researchers who do not have experience writing R code or lack the time to learn how to use the DataSHIELD functions. To help new researchers use the DataSHIELD infrastructure and to improve the user-friendliness for experienced researchers, we present ShinyDataSHIELD.

Implementation

ShinyDataSHIELD is a web application with an R backend that serves as a graphical user interface (GUI) to the DataSHIELD infrastructure.

General features

The version of the application presented here includes modules to perform: (i) exploratory analysis through descriptive summary statistics and graphical representations (scatter plots, histograms, heatmaps and boxplots); (ii) statistical modelling (generalized linear fixed and mixed-effects models, survival analysis through Cox regression); (iii) genome-wide association studies (GWAS); and (iv) omic analysis (transcriptomics, epigenomics and multi-omic integration).

Availability

ShinyDataSHIELD is publicly hosted online [https://datashield-demo.obiba.org/ ], the source code and user guide are deposited on Zenodo DOI 10.5281/zenodo.6500323, freely available to non-commercial users under ‘Commons Clause’ License Condition v1.0. Docker images are also available [https://hub.docker.com/r/brgelab/shiny-data-shield ].

Associations between exploratory dietary patterns and incident type 2 diabetes: a federated meta-analysis of individual participant data from 25 cohort studies

PURPOSE

In several studies, exploratory dietary patterns (DP), derived by principal component analysis, were inversely or positively associated with incident type 2 diabetes (T2D). However, findings remained study-specific, inconsistent and rarely replicated. This study aimed to investigate the associations between DPs and T2D in multiple cohorts across the world.

METHODS

This federated meta-analysis of individual participant data was based on 25 prospective cohort studies from 5 continents including a total of 390,664 participants with a follow-up for T2D (3.8-25.0 years). After data harmonization across cohorts we evaluated 15 previously identified T2D-related DPs for association with incident T2D estimating pooled incidence rate ratios (IRR) and confidence intervals (CI) by Piecewise Poisson regression and random-effects meta-analysis.

RESULTS

29,386 participants developed T2D during follow-up. Five DPs, characterized by higher intake of red meat, processed meat, French fries and refined grains, were associated with higher incidence of T2D. The strongest association was observed for a DP comprising these food groups besides others (IRR_pooled per 1 SD = 1.104, 95% CI 1.059-1.151). Although heterogeneity was present (I² = 85%), IRR exceeded 1 in 18 of the 20 meta-analyzed studies. Original DPs associated with lower T2D risk were not confirmed. Instead, a healthy DP (HDP1) was associated with higher T2D risk (IRR_pooled per 1 SD = 1.057, 95% CI 1.027-1.088).

CONCLUSION

Our findings from various cohorts revealed positive associations for several DPs, characterized by higher intake of red meat, processed meat, French fries and refined grains, adding to the evidence-base that links DPs to higher T2D risk. However, no inverse DP-T2D associations were confirmed.

dsMTL - a computational framework for privacy-preserving, distributed multi-task machine learning

MOTIVATION

In multi-cohort machine learning studies, it is critical to differentiate between effects that are reproducible across cohorts and those that are cohort-specific. Multi-task learning (MTL) is a machine learning approach that facilitates this differentiation through the simultaneous learning of prediction tasks across cohorts. Since multi-cohort data can often not be combined into a single storage solution, there would be the substantial utility of an MTL application for geographically distributed data sources.

RESULTS

Here, we describe the development of "dsMTL", a computational framework for privacy-preserving, distributed multi-task machine learning that includes three supervised and one unsupervised algorithms. First, we derive the theoretical properties of these methods and the relevant machine learning workflows to ensure the validity of the software implementation. Second, we implement dsMTL as a library for the R programming language, building on the DataSHIELD platform that supports the federated analysis of sensitive individual-level data. Third, we demonstrate the applicability of dsMTL for comorbidity modeling in distributed data. We show that comorbidity modeling using dsMTL outperformed conventional, federated machine learning, as well as the aggregation of multiple models built on the distributed datasets individually. The application of dsMTL was computationally efficient and highly scalable when applied to moderate-size (n < 500), real expression data given the actual network latency.

AVAILABILITY

dsMTL is freely available at https://github.com/transbioZI/dsMTLBase (server-side package) and https://github.com/transbioZI/dsMTLClient (client-side package).

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Better together: Advancing life course research through multi-cohort analytic approaches

Longitudinal cohorts can provide timely and cost-efficient evidence about the best points of health service and preventive interventions over the life course. Working systematically across cohorts has the potential to further exploit these valuable data assets, such as by improving the precision of estimates, enhancing (or appropriately reducing) confidence in the replicability of findings, and investigating interrelated questions within a broader theoretical model. In this conceptual review, we explore the opportunities and challenges presented by multi-cohort approaches in life course research. Specifically, we: 1) describe key motivations for multi-cohort work and the analytic approaches that are commonly used in each case; 2) flag some of the scientific and pragmatic challenges that arise when adopting these approaches; and 3) outline emerging directions for multi-cohort work in life course research. Harnessing their potential while thoughtfully considering limitations of multi-cohort approaches can contribute to the robust and granular evidence base needed to promote health and wellbeing over the life span.

Infrastructure platform for privacy-preserving distributed machine learning development of computer-assisted theragnostics in cancer

INTRODUCTION

Emerging evidence suggests that data-driven support tools have found their way into clinical decision-making in a number of areas, including cancer care. Improving them and widening their scope of availability in various differing clinical scenarios, including for prognostic models derived from retrospective data, requires co-ordinated data sharing between clinical centres, secondary analyses of large multi-institutional clinical trial data, or distributed (federated) learning infrastructures. A systematic approach to utilizing routinely collected data across cancer care clinics remains a significant challenge due to privacy, administrative and political barriers.

METHODS

An information technology infrastructure and web service software was developed and implemented which uses machine learning to construct clinical decision support systems in a privacy-preserving manner across datasets geographically distributed in different hospitals. The infrastructure was deployed in a network of Australian hospitals. A harmonized, international ontology-linked, set of lung cancer databases were built with the routine clinical and imaging data at each centre. The infrastructure was demonstrated with the development of logistic regression models to predict major cardiovascular events following radiation therapy.

RESULTS

The infrastructure implemented forms the basis of the Australian computer-assisted theragnostics (AusCAT) network for radiation oncology data extraction, reporting and distributed learning. Four radiation oncology departments (across seven hospitals) in New South Wales (NSW) participated in this demonstration study. Infrastructure was deployed at each centre and used to develop a model predicting for cardiovascular admission within a year of receiving curative radiotherapy for non-small cell lung cancer. A total of 10417 lung cancer patients were identified with 802 being eligible for the model. Twenty features were chosen for analysis from the clinical record and linked registries. After selection, 8 features were included and a logistic regression model achieved an area under the receiver operating characteristic (AUROC) curve of 0.70 and C-index of 0.65 on out-of-sample data.

CONCLUSION

The infrastructure developed was demonstrated to be usable in practice between clinical centres to harmonize routinely collected oncology data and develop models with federated learning. It provides a promising approach to enable further research studies in radiation oncology using real world clinical data.

Position paper on management of personal data in environment and health research in Europe

Management of datasets that include health information and other sensitive personal information of European study participants has to be compliant with the General Data Protection Regulation (GDPR, Regulation (EU) 2016/679). Within scientific research, the widely subscribed'FAIR' data principles should apply, meaning that research data should be findable, accessible, interoperable and re-usable. Balancing the aim of open science driven FAIR data management with GDPR compliant personal data protection safeguards is now a common challenge for many research projects dealing with (sensitive) personal data. In December 2020 a workshop was held with representatives of several large EU research consortia and of the European Commission to reflect on how to apply the FAIR data principles for environment and health research (E&H). Several recent data intensive EU funded E&H research projects face this challenge and work intensively towards developing solutions to access, exchange, store, handle, share, process and use such sensitive personal data, with the aim to support European and transnational collaborations. As a result, several recommendations, opportunities and current limitations were formulated. New technical developments such as federated data management and analysis systems, machine learning together with advanced search software, harmonized ontologies and data quality standards should in principle facilitate the FAIRification of data. To address ethical, legal, political and financial obstacles to the wider re-use of data for research purposes, both specific expertise and underpinning infrastructure are needed. There is a need for the E&H research data to find their place in the European Open Science Cloud. Communities using health and population data, environmental data and other publicly available data have to interconnect and synergize. To maximize the use and re-use of environment and health data, a dedicated supporting European infrastructure effort, such as the EIRENE research infrastructure within the ESFRI roadmap 2021, is needed that would interact with existing infrastructures.

dsSynthetic: synthetic data generation for the DataSHIELD federated analysis system

OBJECTIVE

Platforms such as DataSHIELD allow users to analyse sensitive data remotely, without having full access to the detailed data items (federated analysis). While this feature helps to overcome difficulties with data sharing, it can make it challenging to write code without full visibility of the data. One solution is to generate realistic, non-disclosive synthetic data that can be transferred to the analyst so they can perfect their code without the access limitation. When this process is complete, they can run the code on the real data.

RESULTS

We have created a package in DataSHIELD (dsSynthetic) which allows generation of realistic synthetic data, building on existing packages. In our paper and accompanying tutorial we demonstrate how the use of synthetic data generated with our package can help DataSHIELD users with tasks such as writing analysis scripts and harmonising data to common scales and measures.

dsSurvival: Privacy preserving survival models for federated individual patient meta-analysis in DataSHIELD

OBJECTIVE

Achieving sufficient statistical power in a survival analysis usually requires large amounts of data from different sites. Sensitivity of individual-level data, ethical and practical considerations regarding data sharing across institutions could be a potential challenge for achieving this added power. Hence we implemented a federated meta-analysis approach of survival models in DataSHIELD, where only anonymous aggregated data are shared across institutions, while simultaneously allowing for exploratory, interactive modelling. In this case, meta-analysis techniques to combine analysis results from each site are a solution, but an analytic workflow involving local analysis undertaken at individual studies hinders exploration. Thus, the aim is to provide a framework for performing meta-analysis of Cox regression models across institutions without manual analysis steps for the data providers.

RESULTS

We introduce a package (dsSurvival) which allows privacy preserving meta-analysis of survival models, including the calculation of hazard ratios. Our tool can be of great use in biomedical research where there is a need for building survival models and there are privacy concerns about sharing data.

Does social support prevent suicidal ideation in women and men? Gender-sensitive analyses of an important protective factor within prospective community cohorts

BACKGROUND

Suicidal ideation and behavior constitute important public mental health issues. In this study, we examined whether social integration prevents suicidal ideation over time and whether gender modifies this association.

METHODS

Data from the Gutenberg Health Study (population-based representative community sample in midwest Germany) and the Study of Health in Pomerania (population-based cohort study in northeast Germany) were used. Participants reporting low social support were compared to those receiving middle or high social support. Within a longitudinal study design, we calculated multiple logistic regression models including interaction terms and relevant covariates to test whether gender modified the association of social support and suicidal ideation.

RESULTS

Suicidal ideation was present in 7.4% (N = 982) of the pooled cohorts' 13,290 participants. More women (8.6%, N = 565) than men (6.2%, N = 417) reported suicidal ideation. Middle or high social support was associated with a lower probability to report suicidal ideation five years later after controlling for sociodemographic factors, living situation, and cohort (OR = 0.42, 95%-CI = 0.34-0.52). Male gender was negatively related to suicidal ideation, but no statistically significant interaction of gender and social support was found (ratio of ORs = 1.00, 95%-CI = 0.73-1.35).

LIMITATIONS

The number of people reporting suicidal ideation in the SHIP study was small, especially for men. Suicidal ideation was measured using a single item.

CONCLUSIONS

Social support is an important protective factor in preventing suicidal ideation for both women and men. Future research should further clarify gender-specific effects of family variables in suicidal ideation and test similar predictive models of suicidal behavior.

Privacy-preserving federated neural network learning for disease-associated cell classification

Training accurate and robust machine learning models requires a large amount of data that is usually scattered across data silos. Sharing or centralizing the data of different healthcare institutions is, however, unfeasible or prohibitively difficult due to privacy regulations. In this work, we address this problem by using a privacy-preserving federated learning-based approach, PriCell, for complex models such as convolutional neural networks. PriCell relies on multiparty homomorphic encryption and enables the collaborative training of encrypted neural networks with multiple healthcare institutions. We preserve the confidentiality of each institutions’ input data, of any intermediate values, and of the trained model parameters. We efficiently replicate the training of a published state-of-the-art convolutional neural network architecture in a decentralized and privacy-preserving manner. Our solution achieves an accuracy comparable with the one obtained with the centralized non-secure solution. PriCell guarantees patient privacy and ensures data utility for efficient multi-center studies involving complex healthcare data.

•

We enable collaborative and privacy-preserving model training between institutions

•

Training under encryption does not degrade the utility of the data

•

We apply our solution to the single-cell analysis in a federated setting

•

Our method is generalizable to other machine learning tasks in the healthcare domain

High-quality medical machine learning models will benefit greatly from collaboration between health care institutions. Yet, it is usually difficult to transfer data between these institutions due to strict privacy regulations. In this study, we propose a solution, PriCell, that relies on multiparty homomorphic encryption to enable privacy-preserving collaborative machine learning while protecting via encryption the institutions' input data, the model, and any value exchanged between the institutions. We show the maturity of our solution by training a published state-of-the-art convolutional neural network in a decentralized and privacy-preserving manner. We compare the accuracy achieved by PriCell with the centralized and non-secure solutions and show that PriCell guarantees privacy without reducing the utility of the data. The benefits of PriCell constitute an important landmark for real-world applications of collaborative training while preserving privacy.