Working Smarter, Not Harder: The Effects of a Digital Therapeutic on Exercise Performance in an Activated RCT

Study at a glance

• Authors: Harleen Rai, Daniëlle M. M. Bruel, Philippa Hood, Catherine Sabiston, Bechara J. Saab, Norman A. S. Farb
• Year: 2026
• Journal: Mindfulness
• DOI: 10.1007/s12671-025-02750-1
• N: 70
• Design: Randomized controlled trial (3-week smartphone intervention), two active arms (‘activated RCT’); preregistered (OSF 57rct); linear mixed models + exploratory factor analysis

Key findings

• Manipulation check succeeded: the digital mindfulness therapeutic raised self-reported interoceptive awareness more than the active control (condition×session β=1.83, 95% CI [0.06, 3.59], p=0.044) — validating the app as a form of mindfulness training.
• H1 (performance): a small but significant advantage of mindfulness training on high-intensity exercise endurance — time to voluntary exhaustion (condition×session β=0.21, 95% CI [0.01, 0.39], p=0.041). A session×gender interaction (β=0.23, p=0.022) showed females, who started with lower endurance, improved more.
• H2 (wellbeing): no significant effect of the intervention on affective wellbeing during the bike test (β=−0.55, 95% CI [−1.16, 0.07], p=0.083).
• H3 (mechanism): no evidence for mediation. None of the preregistered candidate mediators (mental health, physical health, mindfulness, attentional focus, mental toughness) showed a significant condition×session effect, so the Baron & Kenny mediation was not run — the performance benefit could not be traced to any measured mechanism, including interoception.
• Exploratory factor analysis reduced the battery to four factors (symptom burden; awareness & attitude; effort & distress tolerance; exercise performance). The intervention moved only the effort & distress-tolerance factor (β=0.33, 95% CI [0.02, 0.64], p=0.039), not the awareness/attitude, symptom-burden, or performance factors.
• At the person level, MAIA was uncorrelated with either performance measure (MAIA↔TVE r=−0.02 ns; MAIA↔performance factor r=0.05 ns), while distress tolerance was the factor that tracked performance (r=0.38, p<0.001). So training raised interoceptive awareness and raised performance, but the two did not co-vary — distress tolerance, not interoceptive awareness, is the candidate route.

Limitations

• No passive control (‘activated RCT’): both arms improved at retest, and the authors note the study itself — goal-setting, familiarity, motivation on the second bike test — may have acted as training, shrinking any arm difference. Social-cognitive self-regulation predicts exactly this second-assessment gain.
• Randomization failed to equate the groups: the mindfulness arm started with higher baseline endurance and the sexes were unevenly distributed (χ²=11.91, p<0.01, more males in mindfulness), forcing gender interaction terms into every model.
• Only 3 weeks of training (vs the typical 8) with high required adherence (81%), so the study may be at the minimal effective dose and cannot examine training dose as a moderator; only two measurement points, so no nonlinear training dynamics.
• The MAP bike test loads leg strength specifically; generalization to other exercise modalities (arm cranking, running, resistance training) is untested.
• Preregistered composite variables for mental/physical health and mindfulness had poor factor loadings, forcing a post-hoc EFA — the reported four-factor structure is exploratory, not confirmatory.
• Declared commercial conflict: one coauthor co-founded and holds ~21% of Mobio Interactive (maker of the AmDTx therapeutic tested); the senior author is an advisor holding <1% stock.

Connections

• Methods: Multidimensional Assessment of Interoceptive Awareness (MAIA), body-focused contemplative training
• Concepts: Distress tolerance, Physical activity, sedentary behaviour and interoception, Interoceptive exposure, Core affect, Interoceptive taxonomy (Farb et al. Box 1), Anxiety sensitivity
• Debates: Is more interoceptive awareness better?, contemplative training increase objective interoceptive accuracy?
• Related studies: Eusebio et al. (2022), Farb et al. (2022), Wallman-Jones et al. (2023), Weng et al. (2021)

A Farb-group randomized trial (Rai first author, Farb senior) testing whether a smartphone mindfulness therapeutic can improve how people perform and feel during high-intensity exercise (HIE). Seventy undergraduates (M age 26.5, 58.5% female) trained for three weeks with either the research predecessor of AmDTx — a Mobio Interactive app delivering guided meditation, body scans, open-monitoring practice and didactic content — or an active control: a cognitive-training app built on the 2048 puzzle game, matched for daily engagement, mood/stress check-ins and a camera-based heart-rate snapshot. Both arms are active, which the authors call an “activated RCT.” The endurance outcome was time to voluntary exhaustion (TVE) on a Maximal Athletic Performance (MAP) stationary-bike test, with affective wellbeing sampled every two minutes during the test.

What the study actually shows

The paper reports four results that need to be held apart, because the headline (“mindfulness improved exercise performance”) is the weakest of them and the most interesting finding is a null.

1. The manipulation check worked. The app raised MAIA interoceptive awareness more than the control (β=1.83, p=0.044). This is what licenses calling the app “mindfulness training” at all, and it replicates the same DTx’s earlier effect on subjective wellbeing (Walsh et al. 2019).
2. Performance moved, barely. Mindfulness produced a small significant advantage on TVE (β=0.21, p=0.041), concentrated in the women, who began with lower endurance.
3. Wellbeing did not move (β=−0.55, p=0.083). The affective-valence hypothesis — the paper’s own motivating idea, that mindfulness makes HIE feel better and so gets repeated — failed.
4. Nothing mediated the performance gain. Not one preregistered mediator showed a group×time effect, so the mediation analysis could not proceed past its first step. Whatever produced the modest TVE benefit, the study could not attribute it to interoception, mindfulness, attentional focus, mental toughness, or mental/physical health.

The load-bearing wrinkle: awareness rose, performance rose, and they did not co-vary

The intervention raised MAIA and raised TVE, which invites the natural reading that heightened interoceptive awareness drove the performance gain. The correlations refuse it. At the person level MAIA was uncorrelated with both performance measures — MAIA↔TVE r=−0.02 (ns), MAIA↔performance factor r=0.05 (ns) — while the factor that did track performance was effort and distress tolerance (r=0.38 with the performance factor, r=0.25 with raw TVE). MAIA’s strong correlate was instead attitude (r=0.74), inside the self-report cluster.

(A note on the source: the results text on p.342 reads “significant positive correlations between MAIA, r=0.74” with the second variable dropped; Fig. 3 makes clear the r=0.74 partner is attitude, not TVE, and that both MAIA↔performance cells are non-significant. The figure values are authoritative here.)

So the causal chain the title gestures at is broken in the middle. Training raised interoceptive awareness; training raised endurance; the two increases were unrelated across individuals; and the EFA moved only the distress-tolerance factor, not the awareness factor. The most defensible reading — and roughly the one the discussion reaches — is that mindfulness helped people tolerate the effort and discomfort of HIE, an indirect Distress tolerance route, with interoceptive awareness a correlate of the training rather than the mechanism of the benefit. See Interoceptive exposure: HIE is itself a naturally occurring exposure to aversive bodily sensation, and tolerance of that sensation is the plausible lever.

Why this belongs to a pattern, not a one-off

This is the third Farb-group trial in the wiki to compare a mindfulness/interoceptive arm against a serious active comparator and find the nominated interoceptive mechanism not carrying the result:

• Farb et al. (2022): MBCT vs Well-Being CT — equal relapse, shared prefrontal biomarker, no mindfulness-specific effect.
• Eusebio et al. (2022): interoceptive vs movement yoga — equal MAIA and symptom change, mechanism visible only as attentional transfer.
• Rai et al. (2026): mindfulness vs cognitive training — MAIA rose but did not mediate or even correlate with the modest performance gain.

The recurring shape (see Norman A. S. Farb, and the central caveat on contemplative interoceptive training for well-being): when this group tests its own interoceptive account against an active control rather than a waitlist, self-reported interoception moves as advertised, the nominated outcome moves weakly or not at all, and the interoceptive mechanism is not demonstrated. Rai et al. is the cleanest of the three on the mechanism point, because it measured a mediator and reported its failure plainly.

For the reader deciding what to take away

The honest summary is the one the limitations section supports: a short, scalable, digitally delivered mindfulness intervention produced a small endurance benefit in an activated RCT with an unequal randomization and no passive control, and that benefit is not shown to work through interoception — the interoceptive-awareness increase was real but inert with respect to performance, and distress tolerance is the better candidate. As a datum for Is more interoceptive awareness better?, it is a case where more self-reported interoceptive awareness was induced and bought nothing measurable on the behavioural outcome. As a datum for contemplative training increase objective interoceptive accuracy?, it is silent: no accuracy task was administered, so the MAIA rise is a change in sensibility only (see Interoceptive taxonomy (Farb et al. Box 1)).