– Written by Sreekanth Reddy Yasa, Manager Statistical Programming, RWS Department at IQVIA

Please note, this blog post is the opinion of the author and does not represent PHUSE nor necessarily the opinion of the author’s employer.

Introduction

As the field of clinical research increasingly embraces real-world evidence (RWE), statisticians and programmers are finding themselves at the centre of a new era of innovation and discovery. RWE studies leverage data from outside the traditional clinical trial setting – electronic health records, health insurance databases and patient registries, among other sources – to generate insights into health intervention effectiveness, safety and quality. However, as promising as these data sources are, they also present significant challenges. In this blog post, we’ll explore the key statistical programming obstacles that arise in real-world evidence trials and how we might navigate them.

Classification of Real World Evidence Studies

Real-world trials, also known as real-world studies or real-world evidence studies, can be classified as:

• Cohort Studies: Researchers follow a group of people over time to see the effect of certain variables or risk factors.
• Case-Control Studies: Researchers compare a group of people with a specific condition to a group without the condition.
• Cross-Sectional Studies: Researchers analyse data from a population at a specific point in time.

Alternatively, RWE studies can be classified as:

Retrospective Studies: These studies look back and examine exposure to suspected risk or protection factors in relation to an outcome.

Prospective Studies: Unlike retrospective studies, these look forward and watch for outcomes.

Hybrid Retrospective/Prospective (Omnispective) Studies: These studies are both retrospective and prospective.

Furthermore, RWE studies can be classified according to whether the data are newly generated:

Primary Data Usage: The study data is newly generated for the specific research question at hand.

Secondary Data Usage: These studies use existing data sources such as hospital charts, electronic health records (EHRs), insurance databases or public health records to examine questions about diseases, treatments, comparative effectiveness, etc.

Finally, RWE studies can be listed by their purpose and specific design:

Safety Studies: These investigate the safety of a treatment in an RWE setting, potentially with many more patients than would be available in a Phase III setting. Pregnancy safety studies belong to this class of studies.

Registry Studies: These studies collect data for a specific population in a systematic way, often over a long period of time. They can be disease-specific or product-specific.

Drug Utilisation Studies: These studies find out how a treatment is applied in an RWE setting, which may be different to a Phase III setting.

Natural History of Disease Studies: These intend to describe the course of a disease which is not yet treatable, to understand the disease and to identify suitable endpoints for potential clinical trials.

Pragmatic Clinical Trials (PCTs): These studies are designed to determine the effectiveness of interventions in real-world routine practice conditions. Unlike traditional clinical trials, which are conducted in very controlled conditions, PCTs aim to reflect more typical care settings and patient populations.

There are even more RWE study types – RCT extension studies, expanded access programme studies, external comparator arm studies – each with its own challenges and opportunities. Below are some key statistical programming obstacles that arise in real-world evidence trials.

Data Structure and Quality

Unlike data from randomised clinical trials, which are carefully collected according to a pre-specified plan, real-world data (RWD) is usually variable in data format, structure and quality. As a result, statistical programmers may have to deal with unstructured data, missing or inconsistent data entries, or variables defined differently across data sources.

To overcome this, it’s essential to develop robust data management and cleaning procedures. Innovative approaches, such as natural language processing (NLP) for unstructured data, may be beneficial. In addition, programmers can leverage standards like the Observational Medical Outcomes Partnership (OMOP) Common Data Model, which enables systematic analysis and data quality checks across diverse datasets.

Data Privacy and Security

Dealing with real-world data often means handling sensitive patient information, which raises substantial privacy and security issues. Compliance with regulations such as GDPR in Europe and HIPAA in the US is paramount.

Statistical programmers need to be skilled in techniques for de-identification and anonymisation of data, if there is no separate department dedicated to these procedures. In this case, they would also need to work closely with data governance and legal teams to ensure full compliance with privacy regulations and be versed in secure programming practices to protect data from potential breaches.

Methodological Challenges

RWE studies often deal with observational data, which introduces a host of methodological issues, such as missingness in baseline covariates, confounding and bias. Statistical programmers must apply sophisticated techniques, such as multiple imputation, propensity score matching, instrumental variables or multivariable regression, to adjust for these issues.

Keeping up to date with the latest methodologies and being proficient in implementing these in a statistical programming language (such as R or SAS) is crucial. Investing time in learning these techniques can pay dividends in the ability to generate valid, reliable and meaningful RWE.

Scalability and Efficiency

With the sheer volume and complexity of RWD, traditional data processing methods may not be sufficient. Statistical programmers are challenged to develop solutions that are both scalable and efficient.

Leveraging big data technologies and advanced computing infrastructures such as cloud-based services can be beneficial. Familiarity with languages designed for big data analysis – Python or Scala for example – alongside traditional statistical programming languages, can also help programmers manage and analyse large datasets more efficiently.

Conclusion

While the challenges associated with statistical programming in RWE trials are significant, they are not insurmountable. With the right mix of technical skills, methodological knowledge and innovative thinking, statistical programmers can turn these challenges into opportunities.

After all, overcoming these hurdles will open the door to more informed decision-making in healthcare – from understanding disease patterns to evaluating treatments in diverse patient populations. As we navigate this exciting terrain, the role of statistical programmers in shaping the future of clinical research and healthcare cannot be overstated. As the adage goes, “Every challenge is an opportunity in disguise.” This rings particularly true in the realm of real-world evidence trials.