Real-world evidence (RWE) studies strongly complement the conventional randomized controlled trials (RCTs), offering crucial insights into the performance of interventions in routine clinical settings. However, one of the most persistent practical concerns in RWE studies is the issue of confounding, which occurs owing to the inherent biases in real-world data (RWD). Unlike RCTs, where randomization factors in both known and unknown covariates amongst study groups, RWE studies are usually based on observational data, where treatment allocation is not arbitrary.[1] This results in confounding variables, i.e. factors that are influenced by both the treatment and the outcome, which can distort the estimated effects of interventions.[2]
The first step in addressing confounding in RWE studies is a robust study design. Researchers must be cautious while evaluating the data source, conditions for cohort selection, and timing of assessment of covariates to ensure that potential confounders are well-defined. It is essential to recognize a distinct temporal connection between exposure, confounders, and outcome. Mispositioning in these time points can result in biased associations, especially if covariates are influenced by the treatment itself or are quantified post-exposure. Careful design selection can lower this risk before applying any statistical adjustment.[2, 3]
Statistical methods play a key role in addressing confounding in RWE. Techniques like multivariable regression, inverse probability of treatment weighting (IPTW), instrumental variable analysis, and propensity score matching (PSM) are commonly applied; each method has its set of assumptions and limitations. For example, PSM can compare observed covariates between treatment groups, but they cannot justify unmeasured confounding. Instrumental variable methods, while theoretically strong, need even robust instruments, which may not be widely available in real-world datasets. These techniques seek to improve causal inference by simulating the balance achieved in RCTs. The choice of an appropriate method relies on the nature of data, the credibility of assumptions, and the research question.[2, 4, 5]
Sensitivity analyses are important tools in assessing the strength and validity of findings in the presence of residual confounding. With variable assumptions, such as the robustness of unmeasured confounding or the model specifics, researchers can evaluate how much their results might be influenced by factors not included directly. Quantitative bias analysis, E-values, and negative control outcomes are some methods that can improve the reliability of study findings. These methods do not remove confounding but help analyse the potential extent of bias.[2, 6]
Finally, transparent reporting is also essential for addressing confounding in RWE studies. Researchers should precisely define their methods for identifying, measuring, and adjusting for confounders, including the reasoning behind selected techniques and any limitations in the data. Communicating code lists, model specifications, and sensitivity analyses improves robustness and enables others to evaluate the authenticity of the findings. Established reporting guidelines, such as the STRengthening the Reporting of OBservational studies in Epidemiology (STROBE) statement [7] for observational studies, provide a solid basis for transparency of findings. For studies considering routinely collected RWD, the REporting of Studies Conducted Using Observational Routinely-Collected Health Data (RECORD) statement, which is an extension of the STROBE statement, provides additional guidance particularly to RWE complexities.[8] Implementing such frameworks facilitates clearer communication of study design and results, making RWE more reliable and actionable.[6, 8]
With the growing use of RWE in regulatory, clinical, and policy decision-making, carefully addressing confounding will be vital for ensuring reliable and actionable evidence generation.
Become A Certified HEOR Professional – Enrol yourself here!
References
- Tashkin DP, Amin AN, Kerwin EM. Comparing Randomized Controlled Trials and Real-World Studies in Chronic Obstructive Pulmonary Disease Pharmacotherapy. Int J Chron Obstruct Pulmon Dis. 2020 Jun 2;15:1225-1243.
- Wang SV, Schneeweiss S. Assessing and Interpreting Real-World Evidence Studies: Introductory Points for New Reviewers. Clin Pharmacol Ther. 2022 Jan;111(1):145-149.
- Laurent T, Lambrelli D, Wakabayashi R, et al. Strategies to Address Current Challenges in Real-World Evidence Generation in Japan. Drugs Real World Outcomes. 2023 Jun;10(2):167-176.
- European Network of Centres for Pharmacoepidemiology and Pharmacovigilance. Chapter 6: Methods to address bias and confounding. Available at: https://encepp.europa.eu/encepp-toolkit/methodological-guide/chapter-6-methods-address-bias-and-confounding_en
- Chandramouli R. Statistical Methodologies in Real-World Evidence (RWE) for Medical Product Development. 2024. Available at: https://www.linkedin.com/pulse/statistical-methodologies-real-world-evidence-rwe-medical-r-bbglc
- Assimon MM. Confounding in Observational Studies Evaluating the Safety and Effectiveness of Medical Treatments. Kidney360. 2021 May 14;2(7):1156-1159.
- Von Elm E, Altman DG, Egger M, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Preventive medicine. 2007;45(4):247–51.
- Nicholls SG, Quach P, von Elm E, et al. The REporting of Studies Conducted Using Observational Routinely-Collected Health Data (RECORD) Statement: Methods for Arriving at Consensus and Developing Reporting Guidelines. PLoS One. 2015 May 12;10(5):e0125620.