The arm that received no sessions

• A randomised controlled pilot trial at the Technische Universität Dresden gave twelve patients with a first diagnosis of functional dizziness six twenty-minute sessions of heart rate variability biofeedback, one a day across two weeks, breathing at six cycles a minute with the amplitude of their heart rate variability drawn on screen as a rising and falling balloon. The other twelve received no device, no sessions and no sham. Both arms were assessed at baseline, at the end of the two weeks, and again at three and six weeks from baseline, which is one and four weeks after the course ended. The investigator who carried out the autonomic and clinical examination was not blinded to allocation, and the paper carries no trial registration number.
• The primary outcome was not a level but a change, defined as the change in RMSSD from the end of the course to the last visit four weeks later. In the biofeedback arm RMSSD stood higher at the end of the course than at that last visit, 0.56, 95% CI 0.1 to 1.01, p = 0.014, and the authors report no such contrast in the control arm. SDNN across the same two points gave 0.45, 95% CI 0.2 to 0.69, p = 0.001. Both are contrasts within one arm across time. A Bonferroni correction set the significance level at p ≤ 0.0167, and the paper does not say across how many comparisons.
• Low-frequency power at rest began at 1511 ± 1892 ms² in the biofeedback arm and 1387 ± 2361 ms² in the control arm. At the end of the course it read 1854 ± 1117 against 403 ± 374. At the two later assessments the arms sat together again, 760 ± 880 against 848 ± 967 and then 785 ± 752 against 838 ± 953. These are the printed group values; the paper reports no test between the arms for them.
• The Vertigo Symptom Scale, which counts the presenting complaint, did not change in either arm. The Vertigo Handicap Questionnaire came down in both, but against the corrected threshold of 0.0167 only the control arm's fall clears it, at p = 0.002, while the biofeedback arm's stands at p = 0.027. Depression on the BDI-II came down in the biofeedback arm alone, at p = 0.024, which is on the same side of that threshold as the VHQ result. No measure of interoception was taken anywhere in the trial: no MAIA, no heartbeat counting task, no body awareness scale.

Low-frequency power at rest read 1854 ± 1117 ms² in the biofeedback arm against 403 ± 374 ms² in the control arm at the end of the course, and 785 ± 752 against 838 ± 953 four weeks after the last session.

This is a single-centre pilot with twelve patients in each arm. The control condition was the absence of an intervention, not a sham and not an equal amount of clinician contact, so the trained arm received six one-to-one sessions with a study physician that the other arm did not. The investigator performing the autonomic and clinical examination was not blinded to allocation. The paper reports ethics approval but no trial registration number, so the primary outcome cannot be checked against a protocol lodged in advance. The primary outcome was itself an autonomic quantity, the change in RMSSD between the end of the course and the last visit, and the two measures that reach the patient, the BDI-II and the VHQ, returned p values of 0.024 and 0.027 against the adjusted threshold of 0.0167 the authors set for themselves. The p values reported for RMSSD, SDNN and the depression score are contrasts within one arm across time; no test between the two arms is reported for any of them, so the group means printed for the end of the course cannot be read as a demonstrated difference between the arms. Nothing here licenses the opposite reading either: with twelve patients per arm and no sham, the unchanged Vertigo Symptom Scale is a result the design could not have distinguished from a moderate effect, and it is not evidence that biofeedback does not act on dizziness. Patients with diabetes, coronary heart disease or heart failure were excluded, as were those on a tricyclic antidepressant within three months or a beta-blocker within fourteen days, so the sample is narrower than a dizziness clinic. Interoception was not measured at any point, so the trial reports what happened to an autonomic index and to symptom scores, and not what happened to a patient's perception of bodily signals. Recruitment ended in July 2019 and the paper appeared in 2025.