Improving patient stratification is frequently a more cost-effective lever than almost anything else available to a struggling drug development program, because a trial testing an effective drug in the wrong population fails for a reason that has nothing to do with the drug’s actual biology. Knowing a spatial signature predicts response is only the first half of the problem. The second half, a clinical trial design question rather than a biomarker discovery one, is deciding exactly how a trial will act on that knowledge.
Key takeaways
|
Why stratification determines trial success
A drug that only works in a specific patient subpopulation, tested in an unselected population, produces a diluted result: the responders’ benefit gets averaged against the non-responders’ lack of benefit, and the trial can fail to reach statistical significance even though the drug genuinely works well in the patients it was meant to help. That failure mode has nothing to do with the drug’s mechanism and everything to do with trial design, which is precisely why stratification strategy deserves the same rigor as the underlying biomarker science.
Our earlier coverage in AI in clinical trials: Patient selection, adaptive design, and the translational gap covers this problem from the angle of composite, often continuously updated artificial intelligence risk scores combining many data types into a single response-probability estimate. This guide addresses a different, more specific case: a spatial signature, of the kind covered in Predicting Response to Immunotherapy With Spatial Signatures, which found a spatial proximity score outperforming PD-L1 testing directly in the same non-small cell lung cancer patients, used specifically as a trial stratification tool rather than as one input folded into a broader composite score.
Spatial signatures as stratifiers
A spatial signature brings the same basic requirement as any stratification biomarker: it must be measurable prospectively, on a real specimen, within a timeframe that fits the trial’s enrollment workflow, not just retrospectively on archived, already-characterized samples after the trial has concluded.
That requirement is where spatial signatures carry a genuinely distinct burden relative to a single-molecule biomarker. A proximity or co-expression measurement depends on the same assay standardization and cross-site reproducibility questions covered in From Spatial Biomarker to Companion Diagnostic: The Development Path, and a trial that intends to stratify patients prospectively using a spatial signature needs that validation work substantially complete before the trial opens, not planned as a parallel workstream to be finished later.
Trial enrichment strategies
A biomarker known to predict response can be built into a trial in several structurally different ways, and choosing among them is a genuine design decision rather than a formality.
Design | How it uses the biomarker | Main tradeoff |
Enrichment | Restricts enrollment to biomarker-positive patients only | Smaller, more homogeneous trial population, but the drug’s effect in biomarker-negative patients may never be characterized |
Stratified | Enrolls both biomarker-positive and biomarker-negative patients, with randomization stratified by biomarker status | Preserves information on both subgroups at the cost of a larger required sample size |
All-comers, retrospective | Enrolls without regard to biomarker status; biomarker effect assessed only afterward via subgroup analysis | Simplest to run, but overall trial results can appear diluted if the drug’s effect is genuinely confined to one subgroup |
Basket | Groups patients by biomarker status across multiple tumor types or indications rather than by a single disease | Efficient for testing one biomarker-driven hypothesis broadly, but individual indication-level evidence can be thinner |
Table 1. Four standard biomarker-driven trial design categories and their central tradeoff. None is universally correct; the choice depends on how confident the program already is that the biomarker matters.
An all-comers, retrospective design is typically the right choice earliest in a program’s life, precisely when the biomarker’s predictive value is still a hypothesis rather than an established fact, since committing to enrichment before that evidence is solid risks excluding a population the drug might actually help. As evidence accumulates, moving toward a stratified or enrichment design becomes progressively more defensible.
Prospective vs. retrospective use
Deciding whether to build a spatial signature into a trial’s prospective design, rather than measuring it only for retrospective analysis after the trial concludes, is a specific, checkable decision rather than a matter of general enthusiasm for biomarker-driven trials.
A specific test for whether prospective enrichment is worth itA methodological review of adaptive enrichment designs in oncology sets out three conditions that should hold before a biomarker-driven prospective design offers a real advantage over a simple randomized trial followed by retrospective subgroup analysis. The primary endpoint must be observable quickly relative to the pace of patient accrual. A sample size sufficient to detect a clinically meaningful subgroup effect must be achievable within a reasonable timeframe. And the experimental treatment must already carry sufficiently strong preliminary evidence of a mechanism of action tied specifically to the candidate biomarker. If any of these three conditions does not hold, for a specific spatial signature this typically means the mechanistic evidence linking the signature to drug response is not yet strong enough, an all-comers design with planned retrospective analysis is very likely the more defensible choice, since the added operational complexity of prospective stratification is not yet justified by the strength of the underlying evidence. |
The right target for prospective use also shifts across a program’s development phase. Early-phase adaptive biomarker designs are explicit that, at that stage, the goal is not to precisely define the target population, but specifically to avoid missing an efficacy signal that might be limited to a biomarker subgroup. That is a materially different objective from a late-phase confirmatory trial, where the goal shifts toward precisely defining the population a therapy should be approved for. A spatial signature strategy built for one phase’s objective should not simply be carried forward unchanged into the other.
Operational challenges
Even a spatial signature with excellent statistical performance and solid mechanistic grounding faces a set of practical, operational hurdles specific to using it prospectively in a live trial.
- Assay turnaround time against enrollment pace. A spatial assay that takes longer to process and report than a trial’s enrollment pace can absorb becomes a bottleneck on recruitment itself, independent of the biomarker’s predictive value.
- Site-level assay availability. A prospective stratification design assumes every enrolling site can actually run, or rapidly ship samples for, the specific spatial assay in question, which is a logistical requirement as demanding as the assay’s underlying science.
- Sample adequacy for a spatial measurement specifically. A biopsy adequate for a conventional single-marker stain may not provide sufficient, well-preserved tissue architecture for a multiplex spatial assay, which can affect eligibility screening and screen-fail rates in ways a program should anticipate rather than discover mid-trial.
Detailed treatment of how a spatial biomarker moves toward the kind of clinical trial infrastructure and workflow this stratification strategy ultimately depends on is covered in Translating Spatial Biology Into the Clinic: Digital Pathology and Beyond.
For the spatial evidence base this stratification strategy is built on, see Predicting Response to Immunotherapy With Spatial Signatures, for the regulatory and assay development path a stratification biomarker eventually follows, Spatial Biomarkers and Companion Diagnostics: The Next Frontier, and for where biomarkers sit within the full spatial biology pipeline, Spatial Biology in Drug Discovery: From Target Discovery to Translational Medicine.
This article was produced under Drug Discovery News’s AI editorial policies.
