But first: What exactly are RWD and RCD?
- RWD (Real-World Data): This is not a scientifically standardized term, but rather a descriptive umbrella term for all data generated outside the controlled setting of clinical trials (RCTs). It is a very broad concept and has been criticized by expert bodies such as the IQWiG due to its vagueness and its often promotional (“marketing-heavy”) use.
- RCD (Routinely Collected Data): This is the more scientifically precise term. It refers to a specific subset within RWD. The defining characteristic is its origin: these data are “byproducts” of administrative processes and clinical documentation requirements.
‍
The IQWiG's Position: Methodological Excellence Over Data Volume
The IQWiG takes a very clear stance in this debate: While observational studies based on routine data are a necessary complement (for example, for rare events or long-term effects following approval), they cannot replace randomized controlled trials (RCTs).
The reason is methodological: In an RCT, randomization ensures that patients are comparable. With routine data, however, the choice of treatment is usually targeted: a doctor prescribes a medication based on specific symptoms or preexisting conditions. When comparing the outcomes of these patients, one often measures not the drug’s efficacy but rather the differences in the baseline characteristics of the patient groups. Without complex adjustment procedures, this leads to massive distortions, known as “bias.”
The Role of AI in a Broader Context
Artificial Intelligence (AI) acts as an “amplifier.” On the one hand, machine learning provides tools for cleaning large, heterogeneous datasets and identifying more complex, nonlinear relationships within the data. On the other hand, researchers warn of risks:
- Statistical “shortcuts”: AI models tend to identify correlations in the data that are based on administrative processes (e.g., “Testing is more common in better-equipped hospitals”) rather than on medical effects.
- The "black box" problem: Because many algorithms do not disclose their decision-making processes, methodological errors or discrimination against certain population groups often go unnoticed.
- Overinterpretation: AI is no substitute for a study design. Without clinical expertise that understands the context of the data, algorithms may produce results that are statistically significant but clinically irrelevant or even misleading.
The debate is not an “either/or” situation
Experts in the field, including those at IQWiG, see the solution in a hybrid strategy:
- Strengthening RCTs: Greater efforts should be made to conduct traditional studies in everyday clinical practice in a way that is simpler, faster, and more cost-effective.
- Targetedly closing gaps in the evidence: Routine data should be used primarily in areas where prospective studies are hardly feasible (e.g., rare diseases, benefit assessment after market entry).
- Methodological rigor: Analyses of routine data require interdisciplinary collaboration.
- Methodological standards must remain high to prevent erroneous decisions in health care.
"Whether they are statisticians, clinicians, epidemiologists, HTA experts, or AI experts—they all agree: The use of routine data presents unique challenges..."
Tim Mathes, Head of the IQWiG Health Economics Division‍
He adds that “targeted strategies are needed to make effective use of them. We show that in many cases, conducting an RCT that uses routine data is the better approach.”
