The interoception scale moved and the registered outcome did not
- Randomised controlled trial run at six outpatient opioid treatment clinics in western Washington State, recruiting from August 2019 to January 2023 and published in Drug and Alcohol Dependence on 5 August 2025 by Cynthia Price, Kenneth Pike and Joseph Merrill of the University of Washington. Funded by NIH grant R33AT009932; the author manuscript is free in PubMed Central as PMC12365928. Recorded on ClinicalTrials.gov as NCT04082637, submitted 3 September 2019 and first posted 9 September 2019, with the enrolment start date listed as 14 August 2019. The registry names a single primary measure, collected by structured 90-day recall interview.
- 303 people were randomised, 155 to the interoceptive training plus medication and 148 to medication alone. Median age 40 (range 21–73), 157 women (52%), 172 people with chronic pain of three months or longer (57%), 249 on buprenorphine and 35 on methadone. Entry required a stable medication dose, which for buprenorphine meant at least four weeks of treatment.
- Neither primary outcome separated the arms across twelve months. Days abstinent from non-prescribed opioids differed by -0.06 (95% CI -1.65 to 1.03), p = .91, d = -.04 (-.29 to .21); the omnibus test was χ² = 0.83, p = .93. Days abstinent from all substances differed by 0.01 (-0.53 to 0.54), p = .98, omnibus χ² = 1.76, p = .78. Both arms sat above 94% opioid-abstinent days at every assessment.
- The largest effect in the trial sits on the interoception questionnaire. The MAIA gave an omnibus χ² of 24.37, p < .001, and a twelve-month between-arm difference of 0.38 (0.21 to 0.54), p < .001, d = .62 (.37 to .88). Emotion regulation difficulties fell 2.53 points further in the training arm (-5.03 to -0.03), p < .05, and mindfulness skills rose 2.18 (0.36 to 4.01), p = .02.
- One clinical measure did separate the arms: PTSD symptoms on the PCL-5 fell 4.62 points further in the training arm (-8.27 to -0.98), p = .01, on an uncorrected twelve-month comparison; the omnibus test across the five assessments did not reach significance. Depression and anxiety scores fell in both arms and the arms did not part on either, although 54% of the training arm crossed below the anxiety screening cut-off by twelve months against 29% of controls.
The outcome this trial named as primary in its registry did not move. Price, Pike and Merrill entered one measure on ClinicalTrials.gov, the Time-Line Follow-back over 90 days, and published it as two figures. Neither arm shifted two percentage points across the year: opioid-abstinent days ran 96.9 to 95.3 with the protocol added and 97.8 to 96.9 without it. The interval around the twelve-month gap is under three percentage points wide, against the 11-point difference the trial had been powered to find, and the second measure gave the same answer.
Some of that is arithmetic. Entry required people already stable on buprenorphine or methadone, so the trial opened at a ceiling: an 11-point gain has nowhere to sit above 96%. The second primary measure did have room, with both arms above 89% rather than above 96% of days abstinent, and it stayed flat too. Price and colleagues attribute the null to the medication stability they required for entry, and their own limitations section says as much.
What the 155 people in the training arm received is set out closely enough to repeat. MABT is a manualised protocol of eight weekly individual sessions of 75 minutes, run over eight to twelve weeks in the patient's clinic. Stage one asks the person to name a bodily sensation, stage two to reach it by directed attention, stage three to hold that attention on a region of the body, with touch as the anchor. Massage therapists trained in the protocol delivered it, all with prior psychotherapy or mindfulness training, and fidelity was checked by weekly review of audio recordings. Those reporting use at six months were offered six more, and 109 people, 71% of the arm, finished at least three quarters of the programme.
Outcomes that did move
Rank what the trial produced and the questionnaire comes first. The MAIA effect is about twice the size of anything else in the paper, d = .62 against a band of .31 to .35 for the other measures that moved, and it is the only measure whose omnibus test cleared p < .001. What it records is how a person rates their own attention to their body; no behavioural task and no physiological measure was taken beside it at any assessment.
The clinical measures split, and PTSD is where the scope of the claim gets fixed. Symptoms fell further in the training arm on the twelve-month comparison, p = .01, while the omnibus test across all five assessments came to χ² = 8.6, p = .07, with no correction for multiple testing across some fourteen secondary measures on four intervals. Read at that strength, what stayed flat is the registered primary measure and not clinical outcomes as a class. Depression and anxiety moved in both arms and separated by -0.35 (-1.72 to 1.02), p = .62 and -1.07 (-2.38 to 0.25), p = .11. Physical symptom frequency reached significance across the twelve months, χ² = 11.23, p = .02, and not at the twelve-month comparison.
The comparison arm received medication treatment and nothing else: no extra hours, no second clinician, no sham procedure. Both patients and practitioners knew the allocation. Eight individual hours with a trained person are the intervention as delivered, and nothing here isolates the part of them spent on the body.
Eight 75-minute sessions changed how these patients rated their attention to their own bodies more than they changed anything else measured, and left the abstinence figure the registry named as primary sitting where medication alone had already put it.
The null on both primary outcomes cannot be read as evidence that the protocol fails to affect substance use, because the design could not have shown such an effect: entry required medication stability and baseline abstinence stood at 96.9% and 97.8% of days, against a power calculation written for an 11 percentage-point difference. The authors name this themselves. The comparison arm received medication alone, so the trial does not separate the interoceptive content of the eight sessions from the time, the touch and the individual attention of a trained practitioner. Patients and practitioners knew their allocation. Interoception was assessed only by questionnaire, with no behavioural or physiological measure. The paper's secondary outcomes are numerous and reported without correction for multiple testing. On two measures the omnibus test and the twelve-month comparison disagree in direction of significance, physical symptom frequency in one direction and emotion regulation difficulties in the other. Missing data were handled by listwise deletion with no imputation; assessment samples fell from 303 to roughly 245 by twelve months. The sample was mostly white (79%) and drawn from a low-income population in one American state, and the authors state that generalisation to higher socioeconomic groups is unknown. 57% of participants had chronic pain and the analysis does not separate them from those who did not.