What Would Your Oncologist Think If…

What would your oncologist colleague think if you, a pathologist, said that you would order one of three different assays for her hormone receptor-positive breast cancer patient depending on whether she planned to treat with an aromatase inhibitor (for example, exemestane), tamoxifen, or fulvestrant? Would she perhaps wonder why each treatment required a different test? After all, are you not asking the same questions about the patient’s health regardless of the treatment the oncologist selects?

What would the oncologist think if you then tell her that, for each different assay, you are going to use a different method of scoring? Some scoring systems combine tumor cells with immune cells, whereas others score only tumor cells, and still others only immune cells. You might even add that each assay has a different cutoff for positive or negative results – so a patient assessed with scoring system X is considered positive, but the same patient under scoring system Y is considered negative. And then, on top of that, each assay has a different sensitivity (some three times more sensitive than others) – even though the antibodies used are very similar. In short, depending solely on the assay you use, the patient’s results – and therefore treatment – may change.

Imagine that your oncologist contacts you to ask for the expected positive rate for a biomarker in the population she wants to treat in a clinical trial – and your response is, “It all depends on the assay and scoring system you use, so I can’t answer.” To complicate matters further, it’s unlikely that a laboratory would implement different antibodies for the same analyte, so if you have antibodies for assay X, you’ll need to outsource assays Y and Z to other institutions. All of this creates delays in treating patients in urgent need – and those delays are only exacerbated when healthcare practitioners disagree with one another on the assay and scoring system to be used.

Now imagine that your oncologist has asked you to perform a PD-L1 assay on a solid tumor within a compassionate use or medical need program, but tells you, “I don’t know what assay or scoring system to use.” Currently, pembrolizumab is FDA-approved in 10 different cancer types, but in five of those (urothelial, gastric, cervical, head and neck squamous cell carcinoma, and non-small cell lung cancer), its approval is restricted to PD-L1 positive cancers – and the PD-L1 positivity threshold of each cancer is different.

By this point, I’m sure you must be asking, “How can this ever work in our daily pathology practice?” In the European Union (EU), unlike in the US, we aren’t obligated to use a companion diagnostic if the oncologist wants to treat a patient with a specific treatment that has proven clinical utility in a phase III trial. Reimbursement agencies will point out that a PD-L1 assay should be used, but not specify a vendor or clone. Accordingly, a breast cancer patient in the US may be tested differently than in the EU, potentially with a different result. So how can we develop international guidelines to direct our pathology colleagues and serve our patients by implementing technologies that guide optimal treatment? Your oncologist might say, “Companion diagnostics are not meant for the convenience of the pathologist, right?” – and yet, it still falls to us to identify the right patients to undergo each treatment.

The question that we – pathologists, oncologists, the drug and assay industry, regulators, and patients – need to evaluate thoroughly is this: “Does a positive phase III trial using a particular assay overrule the notion that it may not be implementable in daily practice?” In my opinion, the answer is a clear no, but the solutions are not straightforward. I emphasize that we should not doubt the results of phase III trials that have proven clinical utility – but just because an assay is theoretically useful does not mean it is practically so.

If we don’t act now, the problems we currently have with PD-L1 testing will spread to gene expression assays, tumor mutational burden assays, and other future tests – so it’s clear that we need a new drug and biomarker development paradigm (1). I propose to introduce a risk management approach for the implementation of biomarkers in clinical trials (2). A risk analysis must be performed before the trial is activated, evaluating the performance of the assay in comparison with other assays – or at least demonstrating reproducibility in a powered study and assessing concordance in a set of samples that is representative of the patient population of interest. Ultimately, a drug is either given or not – a binary decision that can be linked to a binary assay. The cut-point for the binary assay can be tuned depending on whether it is an objective, continuous assay or a subjective assay. If the assay is subjective, the cut-point must be chosen in a manner that can be reproduced around the world without exceptional skill or extensive training.

In daily practice, recurrences are frequently biopsied to retest ER, PR, and HER2 for treatment selection. How should we deal with PD-L1 in this setting? Should PD-L1 be tested in the metastatic biopsy or in the primary tumor? What recommendations should we give to pathologists in their daily practice? How do findings from a clinical trial using checkpoint inhibition extrapolate to a real-life setting if the trial practices are different to those routinely used in the lab? We need to convene with oncologists, pathologists, industry, and regulatory bodies to find answers to these questions.

The proposed risk management strategy stipulates that users must ensure that training yields reliable test results before the assay is introduced into daily practice. Concordance rates should be assessed using a statistically powered number of pathologists (always more than two or three), mimicking a real-life situation. This is a departure from the current practice in which a limited number of pathologists at a central testing site evaluate the biomarker of interest, assuming that this will extrapolate to the full pathology community. Conceptually, we may even question the need to have different companion diagnostics for PD-L1 (given that they all target the same analyte) or even the need to have a PD-L1 companion diagnostic at all (given that we don’t have companion diagnostics for all other protein biomarkers we use in our daily practice).

All this is to say that a positive phase III trial should not be taken as a guarantee that the assay used in the trial can be implemented in daily practice. It is our job as pathologists to select the most suitable patients for each treatment using the best assays available. Finding the assays to select those patients is a responsibility that regulatory bodies, academia, and industry all share – and we must form partnerships to identify those assays now for the sake of our current and future patients.