Approach | Potential strengths | Potential limitations |
---|---|---|
Systematic review and meta-analysis | - Methodology applicable to RCT, observational studies and animal studies - Pooling results increases power to detect modest but clinically meaningful effects - Identify early evidence of harms vs. benefits of treatments and provide evidence-based recommendations | - Quality of summary result is dependent on the quality of the included studies - Often substantial between study heterogeneity - Not all available evidence is published and this may give biased results |
Individual patient level data from completed trials | - Pooling data increases statistical power - Greater opportunity and flexibility to explore new research questions than traditional aggregate meta-analysis - RCT level quality control of source data with standardised, validated, monitored data points - Cost effective approach, saving resources and time | - Limited generalizability - Risk of bias of different RCT populations - Analyses restricted to available data - Inconsistency in tools used to assess outcomes may limit potential, this is a particular issue in studies of cognition |
Individual participant level data from cohort studies | - Opportunity for large multicentre research platforms - ‘last word’ science - Strong epidemiological focus | - Limitations inherent to observational studies - Reverse causality - Confounding |
Big data informatics | - Large data volume – cohort size, registries - Heterogeneity of data - multimodality - Potential for semi-automated data analyses | - Patient data confidentiality to be protected - Development of data sharing mechanisms - Complex computational methods and support required |
Data linkage and use of routinely recorded data | - Resource-efficient - variety of differing data sources available, not solely limited to traditional health settings - Cross-sectional studies can turned into longitudinal, e.g. follow up a cohort of people with dementia for hospital admissions, death etc. | - Selection bias depending on type of data used (e.g. hospital admissions) - Quality of data might vary - Non-standardised outcome measures |