In a receptive health research system, evidence from real-world data (RWD) and, in particular, from electronic health records (EHRs) is needed as an important complement to (and sometimes a replacement of) randomized clinical trials (RCTs). RWD can be found in regional health or epidemiologic registries, and hospital records, including also genetic/genomic databases. When the scientific question is causal, RCT studies are often preferred to RWD-based studies, thanks to their freedom-from-bias certification. But do we really want to give up the ability of RWD studies to tackle a larger space of scientific questions, and their viability in situations where RCTs are unethical, or unfeasible, or time-consuming? As a matter of fact, both RCT and RWD studies are necessary. But if we want to use RWD to address causal questions, we have to do this with the aid of statistical methods that allow us to identify and attenuate the sources of bias (whether due to unobserved confounding or selection or other) and to pick up the causally meaningful aspects of the obtained evidence. The idea is, in fact, to use RWD in such a way as to emulate the results we would have obtained through an experiment. Examples of such methods are propensity scoring, Mendelian Randomisation, and mechanistic interaction. Also relevant are special sampling methodologies and methods to handle the awkward missing data patterns that arise in EHRs.
We want to encourage the proper use of RWD in health research by presenting examples of causal questions arising in a medical research context and tackled via an analysis of RWD, in a way that correctly addressed the causal aspects of the problem. We welcome papers where RWD are used to address a meaningful causal question in cardiovascular and/or stroke medicine, typically as an input to health policy-making, medical resource prioritization, clinical decision-making, or drug discovery. The paper should illustrate a mix of medical and statistical knowledge. Studies based on the use of causal inference methods are especially welcome.
We are not precluded from accepting papers based on the use of AI and machine-learning methods, provided the causal aspects of the analysis are correctly addressed. Papers illustrating synergism between the RWD and RCT approaches are also welcome.
In this Research Topic, we aim to cover:
- assessment of the causal effect of a patient-level or population-level intervention, or identifying modifiers of that effect
- use of EHR datasets to create contemporaneous external control arms in Early Phase Clinical Trials
- use of RWD to assess treatment outcomes for patients with rare diseases or vulnerabilities
- use of RWD to emulate RCTs
- use of genetic instruments to assess the effect of exposure on a medical outcome
- exploring gene-gene or gene-environment or gene-treatment causative interaction
- use of causal inference methods (eg, Mendelian Randomisation, colocalisation, mediation, mechanistic interaction) to explore the way genetic information modulates disease risk by penetrating the upstream layers of molecular pathways.
- sampling strategies to reduce bias to collect RWD data in a medical research context
In a receptive health research system, evidence from real-world data (RWD) and, in particular, from electronic health records (EHRs) is needed as an important complement to (and sometimes a replacement of) randomized clinical trials (RCTs). RWD can be found in regional health or epidemiologic registries, and hospital records, including also genetic/genomic databases. When the scientific question is causal, RCT studies are often preferred to RWD-based studies, thanks to their freedom-from-bias certification. But do we really want to give up the ability of RWD studies to tackle a larger space of scientific questions, and their viability in situations where RCTs are unethical, or unfeasible, or time-consuming? As a matter of fact, both RCT and RWD studies are necessary. But if we want to use RWD to address causal questions, we have to do this with the aid of statistical methods that allow us to identify and attenuate the sources of bias (whether due to unobserved confounding or selection or other) and to pick up the causally meaningful aspects of the obtained evidence. The idea is, in fact, to use RWD in such a way as to emulate the results we would have obtained through an experiment. Examples of such methods are propensity scoring, Mendelian Randomisation, and mechanistic interaction. Also relevant are special sampling methodologies and methods to handle the awkward missing data patterns that arise in EHRs.
We want to encourage the proper use of RWD in health research by presenting examples of causal questions arising in a medical research context and tackled via an analysis of RWD, in a way that correctly addressed the causal aspects of the problem. We welcome papers where RWD are used to address a meaningful causal question in cardiovascular and/or stroke medicine, typically as an input to health policy-making, medical resource prioritization, clinical decision-making, or drug discovery. The paper should illustrate a mix of medical and statistical knowledge. Studies based on the use of causal inference methods are especially welcome.
We are not precluded from accepting papers based on the use of AI and machine-learning methods, provided the causal aspects of the analysis are correctly addressed. Papers illustrating synergism between the RWD and RCT approaches are also welcome.
In this Research Topic, we aim to cover:
- assessment of the causal effect of a patient-level or population-level intervention, or identifying modifiers of that effect
- use of EHR datasets to create contemporaneous external control arms in Early Phase Clinical Trials
- use of RWD to assess treatment outcomes for patients with rare diseases or vulnerabilities
- use of RWD to emulate RCTs
- use of genetic instruments to assess the effect of exposure on a medical outcome
- exploring gene-gene or gene-environment or gene-treatment causative interaction
- use of causal inference methods (eg, Mendelian Randomisation, colocalisation, mediation, mechanistic interaction) to explore the way genetic information modulates disease risk by penetrating the upstream layers of molecular pathways.
- sampling strategies to reduce bias to collect RWD data in a medical research context