How side effects can improve treatment efficacy: a randomized trial

The study challenges the common notion that ideal treatments should lack side effects, noting that the expectation of side effects can increase their occurrence (nocebo effects). The authors hypothesize that mild side effects may act as cues indicating treatment potency (e.g., powerful drugs often have side effects), thereby enhancing positive treatment expectations which in turn can improve outcomes via placebo mechanisms. Prior observations, such as the efficacy of active placebos and the frequent use of impure placebos by general practitioners, motivate this hypothesis. The research question is whether side effects can improve treatment outcomes by augmenting treatment expectations and through what neural mechanisms. The authors preregistered the hypothesis that side effects would recruit the descending pain modulatory system and modulate coupling between rostral anterior cingulate cortex (rACC) and periaqueductal gray (PAG).

Background literature frames side effects as potential contributors to expectancy effects and placebo responses. Studies on active placebos suggest larger placebo effects than inert placebos in pain settings. Surveys indicate clinicians more commonly use impure (active) placebos than inert ones. Placebo analgesia literature identifies expectation as a key driver of non-specific therapeutic effects and implicates neural circuits including rACC and PAG, with inconsistent findings for DLPFC across individuals. The additivity assumption in randomized clinical trials may be violated if side effects differentially shape expectations and placebo responses between arms, potentially inflating apparent treatment effects of active drugs.

Design: Preregistered, multi-session experimental placebo paradigm with independent manipulation of side effects and treatment expectations, including fMRI.

Participants: 104 healthy participants enrolled; exclusions: 9 due to nasal spray issues, 1 technical problem, 2 due to pain floor effects (mean < VAS10), 15 after preregistered manipulation check; final N=77 (Expectancy group: n=42, age 24.3±3.8 [18–38], 12 male; No-expectancy group: n=35, age 24.8±5.4 [18–43], 14 male). Health confirmed via medical interview. Ethics approval obtained; written informed consent provided.

Initial visit: Informed about study and fentanyl nasal spray context, including indication (cancer pain) and potential side effects (including nasal burning). Vital signs and drug screening obtained. Questionnaires completed, including belief that stronger treatments have more side effects.

Pain calibration: Thermal pain via thermode (PATHWAY System, Medoc). VAS 0–100 for pain ratings. Individual temperatures calibrated to VAS 40, 60, 70.

Experimental visits (Sessions 1 and 2): Participants believed each of three nasal sprays per session had 50% chance of containing fentanyl; in reality, no fentanyl was used. One of three sprays (second or last) contained capsaicin 0.15 µg/puff to induce a mild nasal burning side effect; others contained saline (no side effect). After each spray, side effects were rated on a 4-point scale (none at all/minimally/a bit/clearly), including nasal burning.

Trial structure per run: Thermode location moved before each run; three 20s warm-up stimuli. Each run: 24 trials with anticipation (1.5–2.5 s), pain stimulation (6.5 s; 4 s plateau), VAS pain rating (8 s), ITI 7–9 s.

Conditions (Session 1):

• Sham control: saline spray without capsaicin; pain at VAS 70.
• Active placebo: saline + capsaicin side effect; reduced pain (VAS 40) to mimic treatment benefit.
• Inert placebo: saline without capsaicin; reduced pain (VAS 40) to mimic benefit. Order of active vs inert placebo counterbalanced and concealed from participants and experimenters. After runs, participants guessed treatment vs control received and rated confidence.

Expectation manipulation before fMRI (Session 2): Randomization (custom MATLAB script) to:

• Expectation group: continued to believe fentanyl could be present; same procedure repeated.
• No-expectation group: debriefed that no fentanyl was ever present; prior pain reductions due to reduced temperatures. Session 2 in MRI repeated the same paradigm; during active/inert runs pain corresponded to VAS 60. Sprays applied outside the scanner between runs.

Follow-up: Participants reinvited 7±1 days later for repetition (details in Supplementary Information). Expectation group debriefed at the end; no participant withdrew data.

Behavioral analysis: SPSS 27. Repeated-measures ANOVAs with condition (within) and group (between) on pain ratings per phase. Paired t-tests or Wilcoxon Signed-Ranks as appropriate with Bonferroni correction. Moderated mediation: compute differences (active–inert) for side effects, treatment expectations (scored: guess treatment=2; no guess=1; guess control=0), and pain. Belief about side effects measured by agreement (1–5) with "stronger treatments have more side effects." PROCESS model 8 with 5000 bootstrap samples; significance P<0.05 two-tailed.

fMRI preprocessing and analysis: SPM12. Slice timing, motion correction, coregistration to T1, DARTEL normalization (CAT12 IX 1555 template), 6-mm FWHM smoothing. First-level GLM with regressors for cue, pain, rating; boxcars convolved with canonical HRF. T-contrasts between conditions. ROI analyses (per preregistration): rACC (10 mm), DLPFC (10 mm), insula (8 mm), S2 (8 mm), PAG (4 mm). Medial ROIs centered at x=0. Combined ROI mask per contrast; significance p<0.05 FWE-corrected.

PPI: Seed time series from rACC (3 mm sphere at [-6 33 -1.5]) during pain; interaction term = seed × psychological predictor (pain vs no pain). New first-level model with seed, psychological predictor, PPI term; PAG ROI (4 mm) tested for modulation. For visualization, maps shown at p<0.005 uncorrected (figures). Coordinates in MNI space.

Session 1 (behavioral):

• Side effects: Active placebo induced more side effects than inert placebo (p<0.001; F(1,75)=1138). Reported nasal burning ratings markedly higher for active vs inert (e.g., 2.74±0.59 vs 0.13±0.38 in Figure 2A).
• Pain relief: Pain ratings were lower after active placebo than inert placebo (p=0.002; F(1,75)=10.7; VAS 26.2±2.0 vs 30.2±2.0), demonstrating that experiencing a side effect enhanced analgesia.
• Expectations: Most participants believed the active placebo contained fentanyl (94.8%), while only 16.9% believed the inert placebo contained fentanyl; sham was rated as control by 93.5%.
• Mediated by expectations and beliefs: Full moderated mediation observed: the belief that side effects indicate a more potent treatment moderated the effect of experienced side effects on treatment expectation, and expectation mediated the relationship between side effects and pain relief (indirect effect a*b = -1.57 (0.88), 95% CI [-3.83, -0.44]); no direct effect of side effects on pain after accounting for expectations.

Session 2 (fMRI session with expectation manipulation):

• Behavior: Significant interaction between side effects and expectation manipulation (p=0.04; F(1,75)=4.3). Expectation group showed lower pain after active vs inert placebo (VAS 38.7±3.4 vs 43.0±3.2), whereas the no-expectation group did not (VAS 45.0±3.3 vs 43.5±3.3). Side effects continued to be reported only after active placebo, with no group differences (p<0.001; e.g., 2.71±0.58 vs 0.05±0.22).
• Neural mechanisms: Interaction contrast revealed reduced BOLD in rACC during active vs inert in the expectation group relative to the no-expectation group (T=4.5, p=0.003; peak at [-6, 33, -1.5]). No significant DLPFC effect. PPI showed increased rACC-PAG coupling for active vs inert in the expectation vs no-expectation group (T=4.37, p<0.001; PAG peak [0, -30, -12]), indicating recruitment of the descending pain modulatory system.

Follow-up (7±1 days):

• Pain relief persisted: Main effect of side effects with lower pain for active vs inert placebo (p=0.004; F(1,65)=9.0); group difference non-significant (p=0.15); no interaction (p=0.93). Debriefed no-expectation group reestablished treatment benefit expectations.

Findings demonstrate that mild, benign side effects can enhance analgesic outcomes by amplifying treatment expectations, and that this effect depends on individuals’ beliefs linking side effects to treatment potency. The moderated mediation indicates side effects shape expectations, which then drive pain relief; side effects did not directly reduce pain absent expectancy changes. fMRI results align with established placebo analgesia pathways, showing rACC modulation and increased rACC–PAG coupling during enhanced analgesia, consistent with activation of the descending pain modulatory system; DLPFC modulation was not observed, aligning with recent meta-analytic variability in DLPFC effects. Conditioned pain modulation is unlikely to explain results given the low-intensity capsaicin and the session 2 interaction where nasal burning was identical across groups but analgesic enhancement depended on expectation. Clinically, framing benign side effects as signals that a treatment is active may reduce nocebo effects and increase positive expectations, potentially improving outcomes. Methodologically, in randomized trials, differential side effect profiles can bias expectations and thereby differentially alter placebo effects between arms, challenging the additivity assumption and potentially inflating apparent drug efficacy; active placebos or alternative designs could mitigate this confound.

Mild treatment side effects can act as cues that increase positive treatment expectations and thereby enhance analgesic outcomes via recruitment of the descending pain modulatory system (rACC–PAG circuitry). The effect is contingent on patients’ beliefs about side effects and is mediated by expectations rather than a direct sensory effect. These insights suggest two avenues: in clinical practice, careful, ethically appropriate framing of benign side effects as signs of treatment action may improve outcomes while minimizing nocebo; in research, accounting for side effect-driven expectancy differences (e.g., via active placebos) is crucial to preserve trial validity. Future work should replicate these effects in clinical patient populations, test diverse side effect modalities and intensities, evaluate open-label, non-deceptive implementations, and refine trial designs to control expectancy-related confounds.

• Generalizability: Study in healthy volunteers with experimentally induced heat pain; results may not directly translate to clinical patients or chronic pain.
• Side effect modality and intensity: Only a mild nasal burning induced by low-dose capsaicin was tested; effects may differ with other side effects or intensities.
• Deception/expectancy manipulation: Deceptive elements, later debriefed; although ethically managed, such procedures differ from typical clinical contexts.
• Exclusions and attrition: Multiple exclusions (e.g., manipulation check, technical issues); potential selection bias; follow-up sample size reduced (as implied by degrees of freedom).
• ROI-focused fMRI: A priori ROI approach increases sensitivity but may miss effects outside predefined regions; DLPFC modulation not observed.
• Self-reported beliefs and expectations: Mediation relies on self-report measures, subject to bias and measurement error.