Observational data in development of learning health systems
Development of methods to harness observational data to inform development of learning health systems
For the purposes of this Highlight Notice, “observational data” is used to mean data from non-randomised sources (including non-randomised components of Randomised Controlled Trials (RCTs)). There is a growing availability of quantitative observational data on the effectiveness and safety of interventions, for example from routinely collected electronic health records (EHR) or large cohort studies. These data have the potential to provide information that is different and complementary to that from randomised controlled trials. It is therefore becoming increasingly important for organisations like the National Institute for Health and Care Excellence (NICE) and other government health-related agencies to understand and harness the utility of such data for clinical decisions at both population and patient levels.
The ability of observational data to inform the decision-making process is dependent upon the bias inherent in both the data itself and in its synthesis with other data. Therefore, it is necessary that decision-makers can make an accurate and informed assessment of how observed data generalises to a target population.
A range of new approaches for controlling for unknown biases are starting to become more widely used which may be important in enabling the above. These include instrumental variables (best known in the use of Mendelian randomisation in the bioinformatics arena), propensity scoring and graphical methods for exploring causality (e.g. Tetrad). These may create additional methodological requirements for how observational data is collected and managed.
Advances are needed in two areas of methodological research in order to expand the use of observational data to inform decision making at the level of NHS Trusts or individual patients:
Firstly, new informatics approaches and robust methods for collection of clinical data within routine health record systems and in real-time could improve these data to research standards.
Secondly, methods are required for increasing the use of observational data to inform development of a Learning Health System, i.e. a health system which gathers and uses observational data to inform delivery of care; live data capture and use could inform both research and service improvement.
Ultimately, both ways described above use observational data to augment information from RCTs, or to substitute for them in instances when RCTs are infeasible, to ensure that care is informed by the best available evidence.
MRC/NIHR seeks to support the development of innovative methods for the capture and use of observational/routinely collected data in informing the development of Learning Health Systems.
MRC and NIHR are interested in development of methodologies through multidisciplinary proposals combining expertise in informatics, computer science, clinical research, epidemiology and/or statistical methods.
Applications to develop and validate methodologies to underpin and support the use of observational data to inform learning health systems are invited. Responses to the highlight might cover:
(A) Data quality & standards
- Methods for the linkage and standardisation of health intervention/outcomes data derived from different sources and data models.
- Methods to capture and preserve the provenance of data throughout its analysis or synthesis.
- Methods for the evaluation of data quality and completeness in routine data sources.
(B) Consent & Privacy
- Inform practice to ensure informed consent for both service improvement and generically to future research use; with consent agile for future cross-regulator requirements.
- Methods to set the boundaries of consent within data types and to ensure its preservation within analytics.
- Innovative approaches to privacy protection in the use of routine data, where data is analysed at source.
- Methods for improving the real time collection of well-standardised data from EHR systems. These will include interdisciplinary working with expertise and stakeholders including, e.g., informatics, computer science and industry.
- Methods to improve data utility in nascent and developing health systems (e.g. Health Data Collaborative, Kenya).
- Methods to use observational data to provide timely and integrated decision support at critical points in clinical care and using the best available evidence.
- Efficient approaches of validation and revalidation to underpin advances in precision medicine and machine learning.
- New approaches for deriving estimates of effectiveness from routine data, including EHRs and observational cohorts, going beyond traditional epidemiological approaches and controlling for unknown biases.
- Use of observational data to link and validate population-level modelling to patient-level prediction.
Application process and schedule
Applications will develop near-term methodological advances, but may not necessarily impact on clinical decision making within the lifetime of the project. Applicants must clearly articulate a clear pathway towards longer-term impact.
Applications for projects are invited through the Methodology Research Programme Panel, to its regular deadlines and meetings. These will be in competition with other applications received, but the Panel will be mindful of the strategic importance of this area.
In accordance with the remit of the MRP applications should focus specifically on supporting methods development research where the proposed outputs are generalisable beyond an individual case study and where methods development is the primary purpose of the research. Proposals are encouraged to include exemplar case studies which demonstrate the application of the developed methods.
Contact and guidance
It is essential to discuss your proposals with MRC Head Office at an early stage. All applications must be approved by the Methodology Programme Manager prior to submission. Please contact:
Dr Samuel Rowley