Is the Placebo Effect Getting Stronger? What the Data Shows

placebo effect - 1

Short answer: In 84 neuropathic-pain drug trials run between 1990 and 2013, pain relief in the placebo groups rose steadily, reaching an average 30% reduction by 2013, and only in trials conducted wholly in the U.S. That rise is a fact about trial arms, not proof that placebos work better. Most of the evidence points to how trials are built and measured, and the antidepressant version of the story mostly collapses once trial design is adjusted for. medicalxpress


The number that started the story

In 2015, a McGill University team led by Jeffrey Mogil and Alexander Tuttle published a meta-analysis in Pain. Trials in Europe or Asia showed no change in placebo responses over the same period. Meanwhile U.S. trials grew longer, from an average of four weeks in 1990 to 12 weeks in 2013, and larger, from under 50 patients to more than 700. medicalxpresseurekalert

The consequence was practical. More than 90% of candidate drugs for neuropathic and cancer pain had failed in late-stage trials over the preceding ten years. A bigger placebo arm makes it harder for a drug to beat it. nature

Then there is the detail almost every retelling skips. The rise appeared in the placebo groups but not in the drug groups, whose average response stayed stable. Nature’s report drew the obvious inference: drug and placebo responses may not be strictly additive. iasp-painnature

What actually grew

“Placebo response” and “placebo effect” are different things, and the viral version of this story treats them as one.

The improvement measured in a dummy-pill group is the change from baseline. It combines the placebo response with regression to the mean, roughly the get-better-anyway effect. Add the natural course of the illness and the way patients report symptoms to people they like, and you get the picture below. painscience <div class=”svg-container” style=”width:100%; max-width:800px; margin:2rem auto; text-align:center;”> <svg xmlns=”http://www.w3.org/2000/svg” viewBox=”0 0 800 380″ role=”img” aria-labelledby=”t d” style=”width:100%; height:auto; background:#0f172a; border-radius:8px; font-family:system-ui, sans-serif;”> <title id=”t”>What is inside the improvement seen in a placebo arm</title> <desc id=”d”>A bar split into four labeled parts: natural history, regression to the mean, reporting and expectation bias, and the genuine placebo effect. Only the last part is the placebo effect. Segment widths are illustrative, not measured.</desc> <defs> <pattern id=”hatch” width=”8″ height=”8″ patternUnits=”userSpaceOnUse” patternTransform=”rotate(45)”> <rect width=”8″ height=”8″ fill=”#334155″/><line x1=”0″ y1=”0″ x2=”0″ y2=”8″ stroke=”#94a3b8″ stroke-width=”2″/> </pattern> <pattern id=”dots” width=”8″ height=”8″ patternUnits=”userSpaceOnUse”> <rect width=”8″ height=”8″ fill=”#1e3a5f”/><circle cx=”4″ cy=”4″ r=”1.6″ fill=”#7dd3fc”/> </pattern> </defs> <text x=”400″ y=”38″ fill=”#f8fafc” font-size=”20″ font-weight=”700″ text-anchor=”middle”>What a “placebo arm improvement” is made of</text> <text x=”400″ y=”62″ fill=”#94a3b8″ font-size=”13″ text-anchor=”middle”>Illustrative proportions, not measured data</text> <rect x=”50″ y=”95″ width=”700″ height=”70″ fill=”none” stroke=”#e2e8f0″ stroke-width=”2″ rx=”4″/> <rect x=”50″ y=”95″ width=”210″ height=”70″ fill=”url(#hatch)”/> <rect x=”260″ y=”95″ width=”170″ height=”70″ fill=”url(#dots)”/> <rect x=”430″ y=”95″ width=”190″ height=”70″ fill=”#4c1d95″/> <rect x=”620″ y=”95″ width=”130″ height=”70″ fill=”#f59e0b”/> <text x=”155″ y=”136″ fill=”#f8fafc” font-size=”14″ font-weight=”600″ text-anchor=”middle”>1. Natural history</text> <text x=”345″ y=”136″ fill=”#f8fafc” font-size=”14″ font-weight=”600″ text-anchor=”middle”>2. Regression</text> <text x=”525″ y=”128″ fill=”#f8fafc” font-size=”14″ font-weight=”600″ text-anchor=”middle”>3. Reporting and</text> <text x=”525″ y=”146″ fill=”#f8fafc” font-size=”14″ font-weight=”600″ text-anchor=”middle”>expectation bias</text> <text x=”685″ y=”136″ fill=”#0f172a” font-size=”14″ font-weight=”700″ text-anchor=”middle”>4. Placebo</text> <line x1=”685″ y1=”170″ x2=”685″ y2=”205″ stroke=”#f59e0b” stroke-width=”2″/> <text x=”685″ y=”226″ fill=”#fbbf24″ font-size=”14″ font-weight=”600″ text-anchor=”middle”>The only part that is a</text> <text x=”685″ y=”244″ fill=”#fbbf24″ font-size=”14″ font-weight=”600″ text-anchor=”middle”>placebo effect</text> <text x=”50″ y=”205″ fill=”#cbd5e1″ font-size=”13″>Parts 1 to 3 also appear in a no-treatment group.</text> <text x=”50″ y=”226″ fill=”#cbd5e1″ font-size=”13″>A drug-vs-placebo trial cannot separate them from part 4.</text> <rect x=”50″ y=”290″ width=”700″ height=”54″ fill=”#1e293b” rx=”6″/> <text x=”400″ y=”313″ fill=”#f8fafc” font-size=”14″ text-anchor=”middle”>Trial length, size, site practices and outcome scales can inflate any of parts 1 to 3.</text> <text x=”400″ y=”333″ fill=”#94a3b8″ font-size=”13″ text-anchor=”middle”>Only a no-treatment arm isolates part 4.</text> </svg> </div>

The diagram is a conceptual aid. Nobody has measured these proportions for the Tuttle dataset. Its job is to show which comparison is missing. The IASP’s commentary on the paper argued the same point: longer trials follow more patients over time and so give a more accurate view of natural history, and earlier publication bias could also have favored trials with poorer placebo outcomes. iasp-pain

The antidepressant test case, and a real disagreement

If placebo effects were strengthening in general, antidepressant trials should show it too. They have been the main battleground, and here the sources disagree.

Strongest evidence against a general rise: Furukawa and colleagues analyzed 252 placebo-controlled antidepressant trials from 1978 to 2015, covering 26,324 patients on placebo, published and unpublished. They found a structural break in 1991, after which placebo response rates stayed between 35% and 40%. A 2018 reanalysis found no increase in any dataset once design features were adjusted for, or when only post-1990s studies were used. The earlier apparent increase was confounded by methodological factors. boris.unibe.ch +3

The dissent: Lisa Holper’s 2020 meta-reanalysis modeled the continuous symptom-score outcome rather than the binary responder rate. She found placebo efficacy rising gradually and linearly across decades, and attributed it to small-study effects. nih

I weigh these asymmetrically. The Furukawa dataset is larger, includes unpublished trials, and handles design confounders explicitly. Holper’s result is still instructive, because a “rise” that vanishes when you change from responder rates to score changes, or adjust for small studies, is a statistical artifact until shown otherwise. I rank the antidepressant case as unsupported as a general rise, with an open methodological argument.

What a placebo does when you compare it to nothing

The cleanest test is to compare placebo against no treatment at all. The Cochrane review by Hróbjartsson and Gøtzsche pooled 202 trials across 60 healthcare problems. The pain effect was a standardized mean difference of −0.28 (95% CI −0.36 to −0.19). That is a modest effect that is statistically distinguishable from zero. cochranecochrane

It also carries serious caveats. Larger effects were associated with small trials, trials designed to study placebo, and trials that did not tell patients a placebo might be used. Only 16 of the trials (8%) had low risk of bias. The effect on pain ranged from large to non-existent, even in well-conducted trials. The review also notes it is difficult to distinguish patient-reported effects of placebo from biased reporting. CD003974 placebo interventions all clinical conditions +3

The same features that inflate placebo effects in small studies (patient-reported outcomes, expectation, enthusiastic staff) are the ones you would expect to creep into large, long, heavily staffed industry trials.

Four explanations, ranked

HypothesisWhat supports itWhat weakens it
Longer, larger trialsMogil says the data link longer and larger trials to bigger placebo responses. Only U.S. trials changed format.Observational and correlational. Trial format and country are tangled together.
Site professionalizationU.S. sponsors rely more on contract research organizations than academic centers, which differ in staff turnover, training and subject compensation.Reported as a difference between settings, not a tested mechanism.
Expectation and marketingResearchers noted direct-to-consumer drug advertising exists in the U.S.Rising placebo responses were also reported in antidepressant and antipsychotic trials.
Measurement artifactRegression to the mean, natural history and publication bias all inflate placebo-arm change.Cannot explain why the rise is U.S.-specific by itself. News Release +3

This ranking is my analysis, not a finding from any single study. None of these has been tested head-to-head.

The distinction most coverage misses

The unit that grew is the placebo arm’s change score. That is a property of the trial, and the thing that shrank is the trial’s ability to tell a drug from nothing. Drug arms did not move. A bigger placebo number is therefore not evidence of a stronger biological effect. It is evidence of a noisier or more expectancy-laden measurement setup.

That reframing also kills the comforting corollary. The IASP piece cites a belief in the pain field that drugs failing trials may actually work but the trials cannot show it. If drug and placebo responses are not additive, a large placebo arm does not simply mask a real drug effect. Some drugs may be failing because they fail. A failed trial in this setting is weak evidence of efficacy hidden by placebo, and it is also weak evidence of inefficacy. It is weak evidence. painscience

What to do with this

  1. Reading a trial headline. Check trial length, sample size, whether the outcome is self-reported, and where it ran. A 12-week, 700-patient U.S. pain trial is more exposed to placebo-arm inflation than a short, tightly run academic one.
  2. Asking for the third arm. Only a no-treatment group separates part 4 from parts 1 to 3. Without one, do not call a placebo-arm improvement a placebo effect.
  3. Using placebo-based approaches. The Cochrane pain estimate is modest and wildly variable. Treat it as a possible supplement, never a substitute for a treatment with demonstrated effect.
  4. Judging “failed” analgesic trials. Read the placebo arm’s trajectory before concluding anything about the drug. Prior failure in this area is not a verdict on efficacy.

The question that stays open

Why is the rise confined to U.S. trials? Nobody has answered this, and I can’t either. Resolving it would take individual-patient data from U.S. sites split by contract-research versus academic sponsor, alongside measured patient expectations and a no-treatment arm. Until someone runs that comparison, the U.S. pattern is a well-documented observation without a demonstrated cause.

Leave a Comment