In around 600 BC, ancient Indians tested for diabetes by determining if ants were attracted to a person’s urine. Though rudimentary, the experiment formed the mainstay of the extremely accurate diagnostics we use for the disease today.

In this sense, the presence of glucose in body fluids can be considered a biomarker: a characteristic that can serve as an objective indicator of a biological process and be used to measure disease progression or response to a therapeutic drug. Thanks to recent developments in genetic sequencing, a new wave of biomarkers are emerging, forming the basis of personalised medicine.

By identifying disease-related genes that can predict how a person might respond to treatment, the goal of creating products that align with pharma’s new mantra of ‘the right drug for the right patient at the right time’ becomes closer to being realised. But before such therapies can show their worth on the market and in standard practice, clinical trials must be conducted to determine their efficacy and safety.

And while the advent of personalised medicine sounds great for patients, it brings significant challenges for clinical trial professionals. For starters, there are various definitions of ‘biomarker’ in the research literature, each serving multiple purposes, as Pei He, a statistical scientist at Genentech, points out.

"Most definitions of biomarkers can be really wild," she says. "From diagnostic to preventative, there’s a bunch of different biomarkers with different definitions. But in clinical trials, most often people are talking about predictive biomarkers. That refers to some DNA trait or protein expression level that can forecast a clinical outcome of a patient in responding to a specific drug."

A good example of a predictive biomarker in targeted cancer therapy is the KRAS gene, which codes for a protein that performs an essential function in normal tissue signalling. Two EGFR inhibitors − cetuximab and panitumumab − have been approved for colon cancer patients bearing the wild-type version of the gene. Even though KRAS is not the target of either drug, it plays an important role in predicting the treatment outcome, so people with the mutant form of the gene typically do not respond to these treatments.

The challenges of biomarker trials

Due to the many biomarkers and their diverse functions, designing clinical trials for personalised therapies can be much more complicated than the usual one-size-fits-all approach. Not all trials that rely on biomarkers have a similar purpose. Some are designed to explore the properties of biomarkers – for example, which marker to choose and how to categorise patients into groups based on a numerical biomarker value − while others are to validate the chosen biomarker and its cut point, which identifies the patient population that will benefit from the treatment.

And because biomarker clinical trials require the recruitment of an often niche patient population, numbers can be a major challenge when it comes to performing statistical analysis later on.

"Normally, a clinical trial requires a specific sample size to have a specific power," He explains. "When you’re using a biomarker, you must have biomarker- positive populations and biomarker-negative patient populations. And if only the biomarker-positive population responds to the drug, then you’ll have a reduction of sample size. And that obviously depends on the biomarker’s prevalence."

Another big hurdle is that at the very early stages of clinical development, scientists don’t always know the exact properties of the biomarker they are testing for. In 2009, the European Medicines Agency (EMA) published guidelines defining biomarker tests used for diagnosing or monitoring a disease as ‘diagnostic agents’ and established a development process to licence such assays. It’s an additional complication of the drug development process and requires a lot of preparation work.

"If you are choosing one biomarker over the other, of if you were finding different cut-off points, it can be tricky," says He. "You need to do all these exploratory things first but that will need validation in the future because you’re going to need it for the next phase of the trial.

"With all your trials you need to explore a little more competently to ensure you understand the biomarker and the disease. So that’s an extra task."

An additional worry is ensuring that the assay that tests for the biomarker is refined enough. At the beginning of the trial, and particularly at early phases, the assay may not be very well established. But when the study completes to phase III, you have to submit the assay separately for regulatory approval. This can cause problems if the biomarker test has changed significantly throughout the trial’s progression.

"Because biomarker clinical trials require the recruitment of an often niche patient population, numbers can be a major challenge when it comes to performing statistical analysis later on. " 

"And the last major challenge I can think of is sometimes the biomarkers can vary over time," reveals He.

"So protein expression level, for example, may change in the late stage of the disease. And often you have to measure the biomarker at a specific location. If you have multiple samples they may have different biomarker expression levels, so that makes a trial design very challenging."

Intelligent design

Choosing the correct design for your biomarker trial is vital and can help alleviate some of the challenges. From validating the biomarker, to selecting the subgroup, to confirming the efficacy of the tested treatment, different approaches can be selected. So before designing a clinical study, investigators need to clarify its purpose and scale.

"It’s clearly a matter of what clinical development stage you are at, and what biomarker assay development stage you are at. And what’s your purpose? That determines which kind of design you should consider," emphasises He.

Enrichment design is one approach. This type of trial involves only biomarker-positive patients who are then treated as a fixed sample size. Patients are tested for the biomarker before enrolment. The main purpose of such trials is to evaluate the safety and clinical efficacy of the treatment within the biomarker-positive group. This type of trial is usually implemented when sufficient clinical evidence has been established to indicate that only patients with the positive biomarker would benefit from the treatment.

This design has been proven to work best for drugs such as Herceptin, which is only effective for HER2-positive breast cancer patients. Another example is panitumumab, the aforementioned colon cancer drug, which He used to work on in her previous role at Amgen.

"The label is for wild type KRAS," she explains. "So the patient has to have the KRAS gene to be wild type, not mutated, to use that drug. In the late stage of our phase II trial as we developed the assay, we were very clear that wild type patients would benefit from our drug, so we used enrichment design."

It’s a preferable approach when you have a clear understanding of the biomarker before patients are recruited, she says, but there are limitations too. If trials only enrol biomarker-positive patients, the major drawbacks would be unanswered questions for the general population. The assay accuracy and reproducibility of the companion test would also remain uncertain.

"And I think a major challenge would be recruitment would be slower because there might not be that many patients out there, depending on the biomarker," He adds.

If you don’t want to restrict your trial to biomarker-positive patients, all-comer design might be worth a try. Here, patients are enrolled, regardless of being positive or negative for a particular biomarker. However, their status must be known at the start of the trial because one of the goals of such a design is to explore the interaction of the biomarker with the treatment effect. As He explains though, there are a number of pros and cons for the strategy.

"With all-comer design the idea is to enrol everyone," she says. "You then test your volunteers for the biomarker and stratify them. The advantage is you get all the information, but the cost is that depending on how you design it, it may not be enough statistical power for your biomarker-positive population. Because, first of all, when you design it, you lay down the overall population and prevalence rate you get for the biomarker and that rate may well be wrong.

"I think the major cost is the large sample size and the hypothesis selection. Are you really testing with regard to the full sample size or overall population, or are you only testing for the biomarker? It’s hard to determine at the early stages at that time."

Biomarker strategy design is similar to all-comer design. Patients are usually enrolled before their biomarkers test, and randomised into a biomarker-strategy group and a control group. In the first group, biomarker-positive patients receive the experimental treatment, while the negative volunteers receive standard of care.

"The good thing is you have the complete information and you can also test the companion diagnostic for false positive, false negative, those kinds of things," says He. "But the bad thing is the cost and the scale."

If a new treatment were only effective in the biomarker-positive group, more patients would need to be recruited than in enrichment design to achieve the same statistical power.

The three types of design discussed have their pros and cons, and it’s up to researchers to weigh the trade-off between the cost/sample size of a trial and the comprehensive information a study can gather. Sometimes, a more flexible approach might be preferred.

In these cases, adaptive enrichment design might be worth a shot.

In such cases, the trial is separated into two stages. In the first, the whole population is enrolled. Afterwards a decision is made to choose the all-comer or enrichment design for the second stage. There is interim adjustment for either the sample size, randomisation ratio or even the change of end points.

"A good example of a predictive biomarker in targeted cancer therapy is the KRAS gene, which codes for a protein that performs an essential function in normal tissue signalling."

"You divide the trial into two or three parts depending on your needs," He explains. "At the first stage, you get some results, and based on these you refine your trial in the next phase. This is really fit for the biomarker setting where you haven’t a specific idea about where the cut point is, but this way you can explore that in your earlier stage."

Regulatory requirements

However, it could cause a headache when submitting information to the relevant regulatory agency.

"The major drawback of this type of design is whether the regulatory agency would buy it or not," He opines. "In general, FDA is very conservative and would prefer a fixed sample size for a clinical trial."

But as drugs become more personalised and trial sample sizes become smaller, flexibility in clinical trials may well become more important.

"The challenges will be how the regulatory agency responds to that," concludes He. "I think in general they’re warming towards flexibility because with recent developments, more innovative treatments have emerged, so they want more ways of delivering those to the patient. But they also want to make sure trials are designed properly and efficacy results are actually confirmed.

"So I think there will have to be greater interaction between the pharmaceutical industry and regulators to expedite the process."

Disclaimer: The views expressed in this article are the personal opinions of the interviewee and may not be understood as being made on behalf of Genentech or Roche.