Evidence Brief August 28, 2026

This week’s strongest evidence centers on three issues with direct implications for evidence strategy: long-term survival extrapolation in immuno-oncology, how DCE design choices shape apparent patient preferences, and the distinction between preference instability and genuine decision burden.

Peer-Reviewed Evidence

1. Kim et al. The application of excess hazard models for plausible long-term extrapolation in cost-effectiveness: a re-evaluation of immunotherapy in advanced non-small cell lung cancer using the KEYNOTE-024 trial. European Journal of Health Economics.

The authors re-evaluated the cost-effectiveness of pembrolizumab using both immature and mature KEYNOTE-024 survival data. They compared conventional parametric survival models with excess-hazard models that incorporate expected mortality in the general population.

Standard exponential models were overly pessimistic, while conventional log-normal models produced implausibly long survival tails. An excess-hazard log-normal model fitted to the original immature data more closely reproduced the survival subsequently observed with mature follow-up and generated more stable cost-effectiveness estimates.

Why it matters

Long-term survival extrapolation in immuno-oncology is not simply a technical modeling choice. It can materially determine the resulting ICER and potentially the reimbursement conclusion.

Where durable survival is plausible, analysts should consider whether conventional parametric extrapolation adequately represents long-term mortality or whether relative-survival or excess-hazard approaches provide more credible projections.

2. Jang et al. Variation in Attribute Prioritization and Design of Discrete Choice Experiments Across Pharmaceuticals and Medical Devices Patient Preference: A Systematic Review. Patient Preference and Adherence.

This systematic review examined 534 patient-preference DCEs published between 2004 and 2025.

Side effects, effectiveness, and convenience were among the most frequently included attributes, while cost appeared much less consistently. Effectiveness was identified as the highest-priority attribute in approximately half of all studies and in 73% of oncology studies. Attribute selection also varied by geography and funding source.

Why it matters

A DCE does not simply reveal a pre-existing hierarchy of patient preferences. The results are partly conditional on the attributes investigators choose to include, the levels used to describe them, their framing, and the analytical approach.

The finding that effectiveness dominates many oncology DCEs should therefore be interpreted cautiously. It describes the published evidence base, but it does not establish that efficacy intrinsically dominates every oncology treatment decision.

Cross-study comparisons of relative attribute importance are particularly vulnerable to this problem because the underlying choice environments often differ substantially.

3. Huijgens et al. Clinicians’ Perception and Management of Cancer Patients’ Decision Making Burden. Journal of General Internal Medicine.

Through interviews with breast- and prostate-cancer clinicians, the investigators examined circumstances in which patients appeared burdened by treatment decisions.

Clinicians identified uncertainty, information complexity, probabilistic risk information, emotional vulnerability, and mismatch between the decisional role offered to a patient and the role that patient actually wanted. Clinicians also differed in whether they viewed explicit treatment recommendations as supportive of, or potentially inconsistent with, shared decision-making.

Why it matters

Difficulty making a treatment decision should not automatically be interpreted as evidence of poorly formed or unstable preferences.

In oncology preference studies, opt-outs, inconsistent choices, apparent lexicographic behavior, or changes in stated preferences may sometimes reflect genuine decision burden rather than respondent error.

This distinction becomes particularly important when DCEs incorporate survival probabilities, treatment risks, progression, genetic information, or end-of-life outcomes.

4. Alagoz et al. Conversations in rectal cancer treatment: multidisciplinary clinician perspectives on non-operative management decision-making. Supportive Care in Cancer.

Clinicians described decisions surrounding non-operative management as longitudinal processes rather than isolated choice events. Information, uncertainty, treatment response, and patient priorities evolved over time and through repeated multidisciplinary encounters.

Why it matters

Patient preferences in oncology may legitimately change because the underlying decision has changed.

A single preference-elicitation exercise conducted at one arbitrary time point may therefore incompletely represent treatment decisions that unfold over months and depend on response to therapy.

Researchers should distinguish true preference instability from rational preference adaptation to changing clinical circumstances.

Preprints and Emerging Evidence

No new oncology DCE, DCE-TTO, or health-economic modeling preprint published this week met our threshold for inclusion.

The Evidence Brief deliberately favors selectivity over completeness. Preprints are included when they introduce methods or findings with a plausible near-term impact on study design, interpretation, evidence generation, or payer strategy.

What This Means for Evidence Strategy

The papers selected this week point toward a common methodological problem: the observed result is partly conditional on how the decision is represented.

Survival projections depend on the extrapolation architecture. Apparent patient priorities depend on the attributes and levels presented in a DCE. Treatment choices are influenced by uncertainty, decision burden, timing, and evolving clinical circumstances.

Increasing analytical sophistication does not remove these dependencies.

Strong evidence generation therefore begins before model selection. It requires careful specification of the decision itself, the population and context in which it occurs, the outcomes that genuinely matter, and the assumptions through which those outcomes will be interpreted.

Worth Reading First

For health-economic modeling, the Kim et al. paper is the most immediately actionable.

For patient-preference research, Jang et al. provides the strongest methodological warning: preference estimates cannot be separated entirely from the experimental architecture used to elicit them.

Together, the papers reinforce a broader principle: study design does not merely measure the decision problem. It helps define the version of that problem that ultimately becomes evidence.