Real-World Data can be considered the Holy Grail of potential evidence. There are too many real-world questions, concerns and possible evidence gaps that may work as the driving factor to access it and extract the most information possible. Faced with a research question or evidence gaps that you need to tackle you should think about the data you need to achieve your outcome.

‍

‍

The first step should be mapping your data sources. There are multiple options, but not all will respond directly to your question. Ideally, one would be able to merge datasets in one to get the maximum information being considered. However, this may not be possible when dealing with anonymized data, for example.

‍

This leads to the second step, which is data feasibility - i.e., can it address our research question and generate enough evidence to eliminate the gap on the literature or provide us a clear idea to deal with previously existing divergent standpoints? Ideally, your answer should be “Yes”.

‍

Let me remind you that I have used the word “enough” and not “perfect”. Please keep in mind that studies present levels of associated uncertainty that derive from possible limitations associated to the data.

‍

‍

Let me provide you with a real-world example, from personal experience.

‍

I aimed to study prescription and adherence patterns associated to diabetes, and I was kindly provided access to the Portuguese e-prescription dataset (SPMS, EPE). This is a rich administrative dataset that covers the physician – prescription – patient interaction.

‍

At first, I thought I could approach any question with this powerful information in hands! Still, as you probably expected, this is not entirely true...

‍

Within the context of study, I considered this data to be relevant, i.e., I considered the evidence on outcomes, covariates, representativeness, and duration of follow-up as key elements to be considered within the scope of this study. Anonymized information at the prescription, patient, healthcare provider, and pharmacy levels were provided.

‍

Although I could answer my research questions there was always this latent interrogation whether I should have controlled for other issues. I could, eventually – if I had the information to do that.

‍

‍

Here’s three examples of issues that you may find with this and other similar data.

‍

1. Anonymized information enable us to merge this data with other available evidence. This limits already the opportunity of increase the possible outcomes to be achieved and control measures to be used.

‍

‍

2. Prescriptions contain limited information on patients’ demographics, but do not contain any information on socioeconomic characteristics. For this reason, we are unable to answer any question that could be income-related.

‍

We also lack information on the patient’s location, so possible heterogeneities that arise from regional disparities, e.g., will also remain unanswered.

‍

At last, we were not able to access information on patient’s health status, which may introduce optimistic results towards disease progression or prescription differentials. And again, we cannot eliminate this issue.

‍

‍

3. Prescriptions provide no direct information on sociodemographic aspects of the physician as well as we lack information on their specific location while prescribing. This limits our approach to their behaviour, as well as to possible interactions with other physicians and health professionals within their work context.

‍

Accounting for these gaps, we should focus on the following:

- Assumptions (and inherent limitations) are necessary to uphold validity;

- The presence of this information could change the scope of the analysis;

- Information can be extrapolated by using pre-existing results from other datasets;

- We should step away from idealized assumptions (what we want vs. what we get);

- Research requires creativity, consistency, and ways of figuring it out.

‍

I could not end this note without saying that we should not stick to the perfect idea of “I would have the answer… If I just had the data”.

‍

‍