Clinical Failures Persist Because They're Structural, Not Random

By Mikhail Evteev, founder, Evteev Advisory

When Bayer’s Factor XIa program failed, it may have ended a single asset, but it also triggered a reassessment of an entire therapeutic strategy. Factor XIa inhibition had been positioned as a next-generation anticoagulation approach, promising efficacy without the bleeding risks associated with existing therapies. Early data across multiple sponsors supported that narrative. Programs scaled quickly into large, expensive cardiovascular outcome trials. They failed.

In turn, sponsors paused development, investors repriced the entire class, and competing programs were forced to reevaluate their assumptions. Clinical failures are no longer contained within a single program; they propagate across portfolios, companies, and therapeutic areas, influencing regulatory interpretation, capital allocation, and trial design. The relevant question is no longer why a specific trial failed but why similar failures continue to emerge across different programs and disease areas.

Clinical Failure Follows Recurring Structures

Despite the complexity of drug development, most failures follow a limited set of recurring structures. Some emerge early, when the underlying biology does not support the mechanism of action. Others appear in Phase 2, when early signals weaken under broader patient populations and more rigorous endpoints. The most expensive failures occur in Phase 3, where accumulated assumptions are tested at scale and small weaknesses become amplified.

These patterns repeat across therapeutic areas. Oncology, cardiovascular disease, and neurodegeneration differ in biology but share similar failure dynamics. This reflects the structure of clinical development itself. Organizations that recognize these patterns treat failure as an embedded signal. Those that do not, treat it as an isolated event and tend to repeat the same mistakes.

Phase 2 Determines Decision Quality

The most important inflection point in development is Phase 2. This is where uncertainty must be reduced to a level that supports capital allocation decisions. It is also where ambiguity is most common. Signals are often present but inconsistent, subgroups respond differently, and endpoints move without translating into clear clinical benefit. The central issue is not whether activity exists but whether the evidence is strong enough to justify escalation.

Many organizations advance programs on partial signals. The cost of stopping is high, and internal incentives often favor continuation. The result is that unresolved uncertainty is carried forward. The development of KRAS G12C inhibitors illustrates this dynamic. Early studies showed measurable responses in heavily pretreated patients, creating a credible signal. Multiple sponsors advanced monotherapy strategies into randomized Phase 2 trials, but the results were consistent: Response did not translate into meaningful survival.

The signal was there but insufficient. Some teams stopped and redirected toward combination strategies. Others continued investing, effectively converting uncertainty into cost. The core failure mode of Phase 2 is not about missing data but about misinterpreting its strength. Phase 2 is a filter. Organizations that use it that way reduce downstream risk, while those that do not carry unresolved uncertainty into Phase 3.

Phase 3 Exposes Embedded Assumptions

By Phase 3, most critical decisions are already fixed. Patient populations, endpoints, dosing strategies, and statistical assumptions are locked in, while operational complexity increases significantly. At this stage, flexibility is limited, and trials primarily test whether earlier decisions hold under scale rather than generating fundamentally new insights.

When Phase 3 fails, the cause is rarely isolated. The failure of Factor XIa programs demonstrates how early assumptions can break under real-world conditions. Phase 2 signals suggested a favorable balance between efficacy and safety, which supported large outcome trials. At scale, those assumptions did not hold. The issue was already present but not fully recognized, as Phase 2 signals overestimated the balance between efficacy and safety and did not fully capture how these effects would translate under broader patient populations and real-world clinical conditions.

A similar dynamic has been observed in Alzheimer’s disease, where programs built around amyloid reduction advanced despite inconsistent evidence linking biomarker changes to clinical outcomes. Regulatory approval did not resolve that uncertainty. Real-world data forced a reassessment of both the biological hypothesis and the development strategy. In both cases, Phase 3 exposed those shortcomings. This reflects path dependence: Once a development trajectory is established, it becomes difficult to reverse, even when signals weaken.

Failure Reflects Decision Quality Across The System

Clinical failure is often analyzed as a series of isolated events, but in practice it reflects decision quality across the entire development life cycle. Early safety failures indicate misalignment with biology. Phase 2 failures reflect weaknesses in how uncertainty is interpreted and how progression decisions are made. Phase 3 failures reflect accumulated strategic errors, where early assumptions were not sufficiently stress-tested before scaling.

These failure modes are connected. A weak decision in Phase 2 increases the probability of failure in Phase 3, and early misinterpretations can propagate through the entire program. Organizations that treat failure as a system-level signal can identify these connections, compare patterns across programs, and refine their decision frameworks. Over time, this creates a structural advantage in capital allocation and pipeline resilience.

Strategies For Development Teams

Development teams should design Phase 2 trials to resolve uncertainty, not extend optionality, with predefined thresholds that guide progression decisions. Trial design should prioritize decision quality by focusing on clinically meaningful endpoints, representative patient populations, and robust statistical assumptions. Early signals should be stress-tested against scale to ensure they hold under broader conditions.

Equally important is alignment of incentives. Teams should be rewarded for making correct stopping decisions, not only for advancing programs. Failure should be integrated into the decision process as expected feedback rather than treated as an exception. Clinical development, after all, is about generating data and interpreting it correctly at the right time.

In practice, this requires more explicit decision frameworks. For example, in the case of Factor XIa inhibitors, earlier stress-testing of safety signals under broader patient populations and longer follow-up periods may have revealed limitations before large-scale Phase 3 investment. Similarly, for KRAS G12C inhibitors, earlier recognition that response rates in heavily pretreated populations may not translate into survival benefit could have accelerated the shift toward combination strategies.

More broadly, development teams should define clear criteria not only for progression but also for termination, including predefined thresholds for efficacy, durability, and clinical relevance. Without these constraints, ambiguity in Phase 2 is often reinterpreted as optionality, leading to capital being deployed before uncertainty is resolved.

Lessons Must Be Learned

The structure of failure is consistent. Across therapeutic areas, the same patterns repeat: Early signals are over-interpreted, Phase 2 ambiguity is tolerated, and assumptions are scaled before they are fully tested. These are not exceptions but are embedded in the system.

For sponsor companies, the challenge is improving how decisions are made. That means defining clear thresholds for progression, aligning biology with clinical strategy, and terminating programs before uncertainty compounds into cost. Phase 2 is the critical control point. If uncertainty is carried forward, Phase 3 will amplify it.

Sponsors that outperform are not those that avoid failure, but those that learn from it. They fail earlier, interpret signals correctly, and act decisively. Over time, this leads to more efficient capital allocation and more predictable outcomes. Clinical development will always involve uncertainty. The advantage lies in how it is managed.

About The Author:

Mikhail Evteev is the founder of Evteev Advisory.