11 February 2025

Professor Rick Grobbee, Professor of Clinical Epidemiology at the Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht (the Netherlands), and Chief Scientific Officer at Julius Clinical, (Academic CRO), was interviewed by Marjolein Jansen, Director of Digital Innovation, Clinical Services and Project Management at Julius Clinical.

 

Q: Let’s set the scene: can you define Real World Evidence ( RWE)?

RWE refers to the evaluation of how a drug is used and performs  in a real world setting, that is to say, within the population intended to use the drug. This contrasts with clinical trials where strict patient selection criteria can either overestimate or underestimate the true treatment effect or risk profile of a drug. While clinical trials provide essential insights, RWE helps capture the full picture of a drug’s impact in real-world settings.

 

Q: Why use RWE?

RWE serves a variety of purposes. A prime driver for RWE is enhancing generalizability, making findings more applicable to real-world populations.  Interest in RWE is  growing among all stakeholders, including regulatory agencies such as FDA and EMA. A typical view of RWE is that it primarily involves observational data, but this has limitations, especially in determining drug efficacy. However, RWE can also be derived from randomized trials, provided they are conducted outside the constraints of  traditional, highly controlled clinical trials.

RWE can help with fully characterising the safety profile of a drug. RWE is often required to determine safety in patient populations that were excluded from participating in clinical trials, for example pregnant women or the elderly.

Broadly speaking, safety concerns fall into two categories:  those that are difficult to predict in individual patients and those that are expected based on the known pharmacological profile of the drug.

 

Q: Can you provide an example of how RWE is used in safety assessments?

For the first category of safety concerns, those that are difficult to predict in individual patients, consider bone marrow depression which is linked to certain painkillers. This could be particularly serious if the rates are high, so then we use RWE to estimate the frequency of this side effect, for example by creating a cohort of drug users and linking claims data to outpatient hospitalization data. Our data science team regularly conducts such analyses using a federated approach.  On a side note, probabilistic data linking can be an effective alternative, in the absence of unique identifiers. I’ve seen studies reach up to 95% accuracy with this method. While it can serve as a reliable substitute for linked data in safety and comparators, it is less suitable for efficacy assessments.

For the second category of safety concerns, where side effects are expected based on a drug’s mechanism of action, RWE can help determine whether particular patient groups are at higher risk than others. This allows for targeted preventive measures such as the use of anticoagulants and NSAIDs, where RWE can help assess the risk of excessive blood thinning in specific populations. As a drug becomes more widely prescribed and better understood, users at risk will not be prescribed the drug, leading to gradual reduction in the overall risk profile. However, in the early stages of a drug’s use, looking at this data in large cohorts is valuable for assessing risk as later analyses may underestimate the actual incidence due to this evolving prescribing behaviour.

 

Q: When should a company start planning an RWE study?

Planning begins prior to registration. The study then begins once a drug is on the market and being used by sufficient number of patients, ensuring that sufficient real world data is available for meaningful analysis. However, in some cases, conducting an RWE study will be mandatory. Regulatory agencies may require a long term follow-up or legacy studies, which could include patients who previously participated in a registration trial. These long-term follow-up studies can also reveal additional benefits. For example, I have been involved in a 20 year follow-up study using the PROSPER database, which demonstrated that pravastatin not only reduced stroke risk in elderly individuals with obesity and liver fibrosis but also potentially lowered all-cause mortality in subjects with lean weight and liver fibrosis (Reference: Risk of cardiovascular disease in elderly subjects with obesity and liver fibrosis and the potential benefit of statin treatment – PubMed).

 

Q: How do you frame your research question?

A research question typically defines two key elements: a determinant and an outcome. In the context of RWE, the determinant is often drug use, while the outcome is something of interest. And then you need to operationalise the measurements. Measurements generally include three components: drug use (how do you measure that, how do you find that?), the outcome (how do you define that, is it mortality, is it something you need to measure in blood…) and then you always have to be sure that if the data relates the determinant to the outcome, that it’s not due to confounding factors. This is the crux of the issue and why we randomise.

The exclusion of alternative explanations is the hardest part of drawing conclusions from the results and where we may fall short in observational studies. For example, patients at a particular risk may be more likely to receive a specific drug, making it appear as though the drug is creating a risk when, in reality, the underlying condition is the driver. A real-world example is a study I was involved in on third-generation oral contraceptives. These pills with a proven lower risk of thrombosis were given to women with a history of thrombosis (the so-called channeling of drugs) who had higher thrombosis rates in the study, skewing the results. We could call this a prescription bias or confounding by contra indication I suppose.

 

Q: How does one prevent confounding factors then?

In the case of the 3rd generation pill above, the researchers were able to review the data in more detail and discover alternative explanations (more on the subject here: https://pubmed.ncbi.nlm.nih.gov/9753310/). When designing a RWE study, it is useful to try in advance to determine the many ways in which confounding factors can interfere with the outcome, and then work backwards from there.

One way to minimise this type of risk is to run a pragmatic trial, a RWE study in which participants are randomised (versus an observational study) but is more inclusive than a typical clinical trial.

There are advocates of using observational data in real world studies to show drug benefit. Due to the confounding risk however, it is almost impossible to prove this, and a larger study/cohort is not the answer, where numbers do not compensate for errors.

This point is demonstrated by the fact that Regulatory Bodies may accept Real World Data (RWD) for safety but generally not for efficacy. Equally stakeholders may accept RWD to learn about the use of the drug (and respond to questions about equity and equality, guideline adherence etc). An interesting study showed that among new statin users, women are less likely to be up titrated compared to men and to achieve cholesterol target levels (Sex differences in the intensity of statin prescriptions at initiation in a primary care setting – PubMed).

RWE is also important when negotiating reimbursement or market access, and the regulatory bodies may require data against a different competitor than the comparator used in the clinical trial. Routine care can be a misnomer, and then of course it’s country dependant.

 

Q: Have you ever set up a research project and gotten an unexpected outcome?

Yes, I was involved in a long-term study in Sweden looking at women with breast implants and side effects. We found increased mortality in the breast implant group compared to the group of women that had undergone a breast reduction surgery, and the reason for the increased mortality was suicide (Total and cause specific mortality among Swedish women with cosmetic breast implants: prospective study – PMC). So we asked the question, were there were possibly mental health factors to consider in this patient population in Sweden and could treatment practices take this into account?

 

Q: How does it all go wrong?

When you look at the wrong comparison group for example violating comparability requirements, or if you try to answer the wrong question for example trying to prove benefit with the wrong methodology. Eventually it’s all about confounding.