FDA's Mike Blum, MD, MPH, Discusses Postmarketing Surveillance of Biosimilars

In a Monday workshop held at the International Society for Pharmacoeconomics and Outcomes Research’s 23rd Annual International Meeting, in Baltimore, Maryland, Mike Blum, MD, MPH, deputy director in the Office of Pharmacovigilance and Epidemiology at FDA, addressed the role of postmarketing surveillance in the US biosimilars market and the potential role of real-world evidence in interchangeable designations for biosimilars.

In a Monday workshop held at the International Society for Pharmacoeconomics and Outcomes Research’s 23rd Annual International Meeting, held in Baltimore, Maryland, Mike Blum, MD, MPH, deputy director in the Office of Pharmacovigilance and Epidemiology at FDA, addressed the role of postmarketing surveillance in the US biosimilars market and the potential role of real-world evidence in interchangeable designations for biosimilars.

Blum began his talk with a reminder of the FDA’s focus on a multi-disciplinary lifecycle approach to tracking all new drugs, in which the agency uses all available data sources and a risk-based approach. However, “For biosimilars and biologics,” he said, “there is additional focus, and that focus is primarily on immunogenicity and medication errors.”

Tracking Immunogenicity and Medication Errors

Tracking immunogenicity of biosimilars using the FDA’s Adverse Event Reporting System (FAERS) is not without its challenges; because terms related to anti-drug antibodies rarely appear in spontaneous FAERS reports, investigators must also look at reports of reduced efficacy or changes in pharmacokinetics; hypersensitivity events that are either immediate or delayed, such as serum sickness and immune complex disease; descriptions of quality issues; or reports that pertain to specific patient populations (for example, pregnant, very young, or elderly patients).

Medication errors are also an area of focus, and Blum underscored the importance of tracking safety events by products using the biosimilar’s nonproprietary name, with its core name and attached 4-letter suffix that is devoid of meaning.

“The reason for the naming convention,” said Blum, “is not just transparency in terms of allowing us to identify and do our robust pharmacovigilance, but also building trust; practitioners and patients want to know that the FDA has the tools available to do biosimilar-specific pharmacovigilance.”

Among the challenges of conducting pharmacovigilance under such a naming convention, however, is the fact that manufactures are required to submit reports of events that have been reported to them, but “that doesn’t mean that it’s their product; it could be unclear when it’s reported to the manufacture what the product is,” and it cannot be assumed that a report pertains to a company’s drug simply because the report was issued by the company under mandatory reporting requirements.

Furthermore, while FAERS data are useful, Blum explained that “these are solely numerator data;” with no comparison of how many people are actually using the biosimilars versus their references. FAERS reports cannot provide a full picture of biosimilar safety versus reference product safety.

Putting FAERS Data Into Context

In order to create a more robust understanding of biosimilar safety, the FDA is also interested in use data to contextualize FAERS reports. But while use data can capture trends in prescribing, it is not granular enough to explain trends; for example, use data will not show how many patients use a biosimilar when they initiate a drug versus how many use a biosimilar when they switch from a reference. It also does not capture a patient’s reasons for switching (ie, medical reasons, cost reasons, or formulary constraints).

Population-based data sources may be useful for further investigation; use studies could help to capture exposure, switching method studies could be useful to inform the design of future studies, and safety or effectiveness studies could be used if a safety signal warrants further investigation.

Challenges to Using Population-Based Data to Support Interchangeability Designations

Blum also raised what he called theoretical challenges to conducting a safety or effectiveness study using a population-based data source as biosimilar developers seek to secure interchangeable designations for their products.

“You can imagine a scenario where a biosimilar has been approved and now it needs to be studied to get an interchangeability claim,” he said. “The question is, how much real-world data [are] necessary to support that claim?”

As the FDA’s current draft guidance holds, in some cases, postmarketing data can be used as one factor in considering an interchangeability claim (together with data from appropriately designed clinical studies).

Challenges to conducting observational studies include data sources; biosimilar approvals and uptake vary widely across countries, and that fact has implications for study size. Approved indications of a biosimilar may also vary by country, and reimbursement and formulary considerations will affect which patients receive which therapies. All of these factors limit generalizability of findings.

Other challenges include the fact that, in terms of outcomes, effectiveness measures are not typically collected in claims data; for example, a disease activity score in a count of 28 joints or a psoriasis area and severity index score may not be available. “It would be very difficult to do an observational study with those endpoints.”

In attempting to design a study, Blum said, particularly with regard to switching, in a new user design in a population-based data source, sample size could be considerably restricted. “You have to entertain the option of studying and including patients who have switched from biologics once, multiple times, switching back-and-forth.”

Including such switched patients raises questions, however, about why they switched; patients who switch to biosimilars may differ in important ways from those who do not. Thus, substantial confounding may be present.

Finally, researchers could attempt to design a study that includes patients who have switched multiple times, but while this design studies the broadest population and is the most inclusive, it “gets analytically very complex.”