Skip to content
Matching techniques go beyond virtual control arms, reminds senior data scientist at Medaffcon

Home > All articles > Matching techniques go beyond virtual control arms, reminds senior data scientist at Medaffcon

Matching techniques go beyond virtual control arms, reminds senior data scientist at Medaffcon

Techniques such as propensity score matching and inverse probability weighting are essential tools for a broad range of pharmaceutical studies, stresses Samuli Tuominen, senior data scientist at Medaffcon. 

Virtual control arms. Also known as external control arms have generated interest worldwide as real-world data-based tools for businesses, research organisations and regulatory authorities to complement the findings of clinical trials. 

Real-world data (RWD) can, however, be utilised also in a variety of non-clinical medical and pharmaceutical studies to compare aspects such as treatment outcomes and resource utilisation between different geographies and treatment options, reminds Samuli Tuominen, senior data scientist at Medaffcon

“The world where matching techniques are applicable is much larger than virtual control arms,” he stresses. 

For example, a pharmaceutical company can utilise real world data (RWD) to carry out a study in conjunction with an expansion to or the launch of a product in a new geography. 

The key to any such study is patient matching, the process of eliminating confounding interference from factors that could influence the treatment outcome, such as patient age, co-morbidities or treatment history. 

Two of the more common matching techniques are propensity score matching and inverse probability weighting, according to Tuominen.  

The former makes use of covariates to assess the propensity of a patient receiving a particular treatment and matches patients with similar scores while eliminating those without a suitable match within the defined caliper. The latter does not eliminate any patients but rather gives higher weight to well-matching patient. 

“Inverse probability weighting is good especially if the patient population is small,” he says. 

The selection of covariates is dictated by disease related clinical insight, but also by the clinical trial or real-world data availability – one may not include row-level patient data; the other may not be consistent across the patient population. Both scenarios can bring about technical challenges, particularly in studies with a small patient population. 

“The research question and data availability guide the choice of methods. More advanced matching techniques require higher number of patients, and you simply need to adjust the techniques to the patient data,” sums up Tuominen. 

“At Medaffcon, we have the methodological expertise to do that,” he adds, citing as an example a recent study that compared the effectiveness of two immunosuppressants in patients with primary sclerosing cholangitis after liver transplantation. “We also have a lot of experience Finnish registers: we know what data they contain and have a good understanding of the matching possibilities. And then we have a lot of experience in processing and analysing the data.” 

This expertise naturally applies to virtual control arms. Tuominen reminds that a virtual control arm can be created for most studies connected to the development or launch of a new pharmaceutical product to examine differences between the trial participants and real-life patients in a region where real-world health care data are available, such as Finland or Sweden. 

Such comparisons can produce insights for pharmaceutical companies and regulatory authorities. Comparisons also support decision-making in various stages of pharmaceutical development, from initial market research to final authorisation and beyond. 

“At the simplest end of the spectrum, a pharmaceutical company could examine how much the inclusion criteria of a clinical trial, when applied to real-life patients, limit the number of patients who are eligible for the treatment,” tells Tuominen. 

“Naturally the inclusion criteria of a clinical trial don’t align perfectly with the treatment’s indication. For example, the age distribution may be such that real-life patients are much older or in much poorer health, possibly with some contraindications. So, in a way it helps to provide an indication about the real market for the upcoming treatment.” 

A virtual control arm could be created for the following reasons: 

  1. It was impractical or unethical to include a control arm to a clinical trial 
  1. A control arm could not be created for reasons such as low disease incidence 
  1. The comparator of the clinical trial is not used in the target region 
  1. The clinical trial patient population does not coincide with real-world patients

The creation of a virtual control arm starts with the extraction of real-world health data based on diagnoses from regional or national registers. The data can then be enriched with other relevant variables, such as drug purchases, causes of death, or sick leaves, before the patient cohort is finalised by applying the inclusion and exclusion criteria of the study.  

Finally, patients in the control group are matched with their counterparts, be it trial participants or other real-life patients, to enable meaningful comparison of the outcomes. 

  • Medaffcon’s expertise in patient matching is applicable beyond virtual control arms 
  • A virtual control arm can be created to complement almost any study related to a new pharmaceutical product 
  • Real-world health data can also be utilised to examine geographical and treatment-specific differences in patient outcomes, with or without an underlying clinical trial 
  • Any real-world evidence study requires careful patient matching to produce reliable and actionable results 
  • Some of the more common matching techniques are propensity score matching and inverse probability weighting 
Back to top