• It is difficult to infer causality from observational data 
  • Non-randomised design and may lack a control group
• Potential for reduced data quality (missing data, data entry/coding errors)
  • May lack standardised approach across settings
  • Visit schedules may not be required/data may not be collected at fixed intervals
• Often subject to bias and confounding factors that require statistical adjustments (eg, propensity-score matching)
  •  Risk factors may change during follow-up

• Can assess risk factors that may be considered unethical in RCTs
• Reflect real-world outcomes, treatment patterns and clinical decision-making
• Less likely to be protocol-driven
• Analyses performed at low cost and over a short time frame
• Registries may have a longer time frame than RCTs (variable, detailed, longitudinal information)
• Large size of databases may identify outcomes of patients with rare events
• More diverse population of patients
• Generalisability of data about treatment impacts and use, costs for health services
• Lack of standardisation and randomisation; patient groups may not be comparable
• Susceptible to sample bias (if specific to one geographic location), observational bias and recall bias
• Risk factors/outcomes may change during follow-up
• Increased cost of data collection with larger number of patients and clinical settings
• Potential for reduced data quality (missing data, data entry/coding errors)
• Lack of statistical power to detect events other than specified key endpoints
• May lack a control or comparator group
• Difficult to distinguish cause and effect

RCT=randomised controlled trial; RWE=real-world evidence
*This is an overview of general strengths and limitations of real-world data sources, which may vary with each specific real-world data source type.

References:
1. Katkade VB, et al. J Multidiscip Healthc. 2018;11:295-3042.
2. Blonde L, et al. Adv Ther. 2018;35(11):1763-17743.
3. Camm AJ, et al. Open Heart. 2018;5(1):e000788.