The Limits of Averages: A Framework for Contextual Effectiveness in Palliative Care Medicinal Cannabis Research

Introduction: The Gap Between Patient Experience and Clinical Evidence

Clinicians and patients often disagree about whether a treatment “works.” Clinicians typically ground their positions in evidence interpreting observer-independent hypotheses, while patients rely on lived experience.^1,2 When patients with advanced illness report improvement from an intervention, this can conflict with clinical trial findings showing no benefit.

This article argues that such patient-reported improvements can be epistemically justified even when clinical trials report null results.

We illustrate this using data from a qualitative sub-study of a medicinal cannabis trial for advanced cancer.

One participant (Steven [a pseudonym]) described:

Does it make sense to ask, “What is the average treatment effect of medicinal cannabis for Steven”? Or is a more relevant question, “In this specific life context, will cannabis help them feel better?”

We propose a framework of contextual effectiveness to answer such questions. First, we clarify what counts as evidence and knowledge in clinical practice.

Note that the qualitative data presented here comprises quotations drawn from two sociologically informed sub-studies embedded within a series of MC trials conducted by the broader research team.^3,4 Purposively-sampled interviewees included patients who had participated in one of the MC trials for palliative care patients with advanced cancer. Semi-structured interviews were conducted in person or by telephone, with data collection overseen and/or conducted by experienced qualitative researchers who remained part of the investigative team ongoing.

Guiding interview questions prompted patients to reflect on perspectives of MC and general cannabis use; reasons for participating or not in a trial; conceptualizations of well-being; and experiences of MC during and after trial participation (where applicable). Each interview was recorded using a digital voice recorder and transcribed verbatim; with thematic analysis subsequently undertaken. Sociodemographic details reflected a balance of characteristics: the results from interviews accompanying the first MC trial are reported, and the second accompanying study results are being prepared for publication.^1,5

What Counts as Evidence? Epistemology in Palliative Care

Evidence-based medicine (EBM) emphasizes knowledge derived from controlled trials, which works well for objective outcomes like mortality. However, subjective outcomes like overall quality of life (QoL) raise challenges.^6–8

Patients’ specific reports of side effects (“this drug makes me dizzy”) are typically accepted without question, but similar reports of benefit (“it helps me relax socially”) are often dismissed if trial data show no average improvement. Although QoL instruments are built from patients’ subjective ratings, their use in evidence synthesis typically involves aggregating scores across times and contexts. ⁹

In contrast, in clinical encounters clinicians often act on a patient’s specific report of an adverse effect, yet may discount an equally specific report of contextual benefit when it conflicts with averaged trial results. This asymmetry matters because it reveals how subjectivity is treated: individual harms are acted on contextually, while individual benefits are subsumed into averages—exactly the move our contextual effectiveness framework challenges. We argue that a patient’s claim that an intervention helped them is epistemically justified unless undermined by a defeater (a reason that overturns that belief). ¹⁰ Importantly, a negative trial result may not qualify as such a defeater for complex outcomes like “feeling better overall.”

We adopt a post-positivist stance: the belief that a mind-independent reality exists but can only be imperfectly known, as opposed to positivism, which assumes objective truths are fully knowable through empirical observation. ¹¹ Whereas positivism presumes that careful measurement can fully capture objective truths, post-positivism concedes that all observation is theory-laden and partial. Critical realism adds that mechanisms operate differently across contexts. ¹² Our position extends these insights to argue that, for holistic subjective states, the “mechanism” may be inseparable from the context in which it is enacted. Thus, a patient’s situated testimony—when cogent and coherent with their broader narrative—possesses epistemic weight that population statistics cannot automatically override.

This framing supports a broader justification for patient-reported outcomes in practice. Testimony about subjective states—such as relief, calmness, or engagement—carries prima facie justificatory force because these experiences are accessible only to the person living them. Phenomenological accounts of illness and recovery, as well as the literature on patient-reported outcome measures (PROMs), affirm that such first-person accounts are not only valid but often indispensable in evaluating therapeutic benefit.^13,14

Our group has conducted two of the most methodologically rigorous randomized trials of medicinal cannabis in palliative care to date, with a third under way, as well as numerous embedded qualitative substudies.^3–5,15 Yet many participants from those trials subsequently obtained cannabis commercially and reported meaningful, context-specific benefit, findings we documented in a post-trial surveillance report of lower methodological stringency. Rather than attributing the disconnect to unidentified responder subgroups or to patients’ misattribution of placebo effects, we propose a framework that reconciles these apparently contradictory observations—one that both clarifies the limits of population-level trial evidence and guides the generation of richer post-trial data.

From average to context: Three frameworks of effectiveness

Three conceptual frameworks help clarify how treatment effectiveness is interpreted and are illustrated in Figure 1 and more formally stated in Supplementary Appendix SA1.

OPEN IN VIEWER

FIG. 1.

Three perspectives on treatment effectiveness. The first column applies a single universal threshold ( θ * ) across all patients, determining effectiveness solely by average response and dismissing individual variation as statistical noise. We don’t hold that this simplistic view is common among clinician-investigators, though it may be more commonly held than researchers appreciate. The central column acknowledges patient heterogeneity through individualized thresholds (each dotted line is a subgroup threshold) based on patient-specific factors, yet still categorizes patients into “responder” and “non-responder” groups. The right column is what patients implicitly hold to: that what makes them feel better or worse overall is context dependent. A deceptively simple example may be as follows: wine with some friends may be excellent, but with other friends you’d have a much better time sticking to coffee. A patient, knowing how wine and coffee will make them feel, is in a position to judge this, but such high resolution distinctions are lost as noise in clinical trials. The equations link to the mathematical outline of our argument provided in the Supplementary Appendix SA1.

Framework 1: Universal threshold

This model assumes a single underlying treatment effect that applies to all patients. A treatment is considered effective only if the average change in symptom scores exceeds a fixed threshold, such as the minimal clinically important difference. Individual variation is treated as noise, and effects are evaluated purely in terms of population-level means.

Framework 2: Subgroup heterogeneity

This framework acknowledges that treatment effects differ across patient subgroups due to relatively stable traits such as baseline status, genetics, or comorbidities. While still using a population approach, it allows for post hoc or pre-specified identification of “responders” and “non-responders.” However, these categories are typically fixed and linked to individual characteristics rather than context.

Critically, moving from heterogeneity between patients to variation within the same patient marks a categorical shift: the locus of explanation relocates from fixed traits to unfolding contexts, demanding methods attuned to lived temporality.

Framework 3: Contextual effectiveness

This model recognizes that the same patient may have different treatment effects depending on time, setting, emotional state, or goals. What makes them feel better in one situation may have no effect or even be counterproductive in another. Unless there is what we term “contextual collapse” (see “Conceptual clarification” below), trial results are simply an accidental sampling of contexts.

This framework aligns more closely with how patients describe their experiences. As one participant (Jenny [a pseudonym]) noted:

“If I’m in the house and not going out, then it’s fine to take a bigger dose… It makes me go a bit soft in the head.”

This final framework implies a different logic of study design, leading into the next section.

Reframing Clinical Epistemology: Humility and Co-Decision

Clinicians often assume that trial data with standardized PROM outcomes provides the highest-quality knowledge, while qualitative reporting is more susceptible to error. ⁶ But for subjective experiences, patients may in fact be best placed to evaluate their own outcomes.

One patient explained, “I’m so happy that it fixed up the sleeping thing… I’m not so depressed, I’m not tired. Now I say yes [to invitations to social engagement].” This kind of change in interpretative stance—from withdrawal to engagement—can be invisible to symptom scales but is deeply meaningful.

Such claims are not mere anecdotes to be overturned by group averages. They are provisional beliefs, open to revision through further experience, not abstract statistics. Practicing epistemic humility means respecting that patients may have access to valid knowledge inaccessible through traditional means. For instance, it is well-documented that some Buddhist patients value mental clarity over analgesia—a preference our framework recognizes as epistemologically valid. ²⁰ The fact that this value context is easily recorded does not diminish the epistemic status of more ephemeral contexts (a looming deadline, a family gathering) that may be harder to capture yet equally shape whether an intervention helps or harms.

Practically, epistemic humility translates into explicit invitations for patients to articulate when and why they believe an intervention helps, coupled with planned re-evaluation: “Let’s trial this for two weeks, log contexts and effects, and reassess whether it aligns with your goals.” Such dialogic monitoring shifts the clinician’s role from arbiter of truth to curator of evidence streams—statistical and experiential—assembled around the individual.

A parallel can be seen with opioids for refractory dyspnea: despite conflicting high-quality evidence, clinicians routinely use patient-centered dose-and-re-assess cycles, guided by accumulated practice wisdom and consensus statements. ²¹ The same epistemic humility can—and should—be extended to newer or legally complex interventions such as medicinal cannabis. Our framework provides a clear theoretical justification for doing so.

Authors’ Contributions

T.G. conceived the original concept, developed the theoretical framework, and authored the entire first draft of the article. R.O., C.S., and Prof. P.D.G. provided substantial intellectual input, critical review, and valuable feedback to subsequent drafts. A.S. provided substantial intellectual input to revised versions of the article as well as obtaining the qualitative data quoted. All authors have reviewed and approved the final article.