Hybrid machine learning improves lost-circulation prediction in small
A small pilot study found that one session of real instrument-assisted chiropractic care was linked to greater changes in communication between different areas of the brain than a sham treatment. These changes were measured while participants rested. The differences appeared right after treatment in the delta brain-wave range and one week later in the theta range. They occurred across a group of connected brain systems that the researchers call the “extended triple-network.” However, the researchers found no difference between the groups in heart rate variability, a measure linked to the activity of the parasympathetic nervous system. The study included only 30 adults, did not assign participants randomly, and was not designed to prove that the treatment works. Therefore, the brain findings are preliminary. They do not show that spinal manipulation improves brain function, mental health, or physical health.
The study took place at the Center for Chiropractic Research at Life University in Marietta, Georgia. It included 30 participants: 15 received real chiropractic care, and 15 received a sham treatment. Participants were assigned to the groups in alternating order rather than randomly. The study used a single-blind design, meaning that participants did not know which treatment they received. Checks showed that the blinding appeared to work. For the real treatment, the chiropractor used an Activator instrument to deliver quick, short pushes at selected sites. For the sham treatment, the same assessment, positioning, and instrument contact were used, but the instrument was triggered beside the skin rather than against it. The researchers recorded brain activity using electroencephalography (EEG) and heart activity using electrocardiography (ECG) before treatment, immediately afterward, and one week later. The study was registered in advance at ClinicalTrials.gov under the number NCT05249530.
The researchers used 64 electrodes placed on the scalp to record EEG signals. They estimated where the signals came from in the brain using a method called a linearly constrained minimum variance beamformer. They then measured functional connectivity using the multivariate interaction measure, which detects relationships between the timing and phases of signals from different brain regions without showing which region influences the other. To locate the signals, they used the Desikan–Killiany atlas, which divides the brain's outer layer, or cortex, into 68 regions. Their planned network analysis focused on 36 regions on both sides of the brain. They studied five ranges of brain-wave frequencies: delta, from 1 to 4 hertz; theta, from 4 to 8 hertz; alpha, from 8 to 13 hertz; beta, from 13 to 30 hertz; and gamma, from 30 to 40 hertz. A statistical method called the network-based statistic identified groups of connections that differed between the two treatment groups more than would be expected by chance in permutation tests. The selected regions represented the medial frontoparietal, dorsolateral frontoparietal, and mid-cingulo-insular networks. Together, these networks include the default mode, central executive, salience, dorsal attention, and ventral attention systems.
Immediately after treatment, the researchers found one statistically significant group of connections in the delta frequency range. It contained 28 brain regions, called nodes, and 52 connections, called edges. The result had a family-wise error-corrected p-value below 0.001, meaning that it remained highly significant after the statistical correction used for this group of tests. Every connection in this group showed a greater increase from the starting measurement in the real-treatment group than in the sham group. At the one-week assessment, the researchers found a separate significant group of connections in the theta frequency range. This group contained 35 nodes and 109 edges, also with a family-wise error-corrected p-value below 0.001. Of the 109 connections, 108 showed larger increases from the starting measurement after real treatment. One connection between the left and right middle temporal regions showed a slightly smaller increase in the real-treatment group. The difference in the multivariate interaction measure was −0.013134. The researchers found no significant groups of connections in the alpha, beta, or gamma ranges at either time after the starting measurement.
These network-based findings do not mean that every individual connection within either group was independently statistically significant. The statistical correction applied to each group as a whole. The connected groups show the overall pattern of connections that passed the initial statistical threshold. The way the researchers interpreted each network was also limited by the brain atlas and the region assignments they used. Although the Desikan–Killiany atlas divides the cortex into 68 regions, there is no published map that clearly assigns every region to one specific network. The authors therefore had to use their own judgment, and some regions belong to more than one functional system. The insula was not divided into front and back sections, while parts of the superior frontal, inferior parietal, and supramarginal regions are linked to different networks. The analysis included only regions in the cortex, not deeper brain structures. Also, methods for estimating the sources of brain signals are generally less reliable for regions near the middle of the brain than for regions closer to the sides.
The heart measurements did not show a similar difference between the treatment groups. The researchers measured the root mean square of successive differences, or RMSSD. This is a measure of heart rate variability that is often used to estimate parasympathetic activity affecting the heart. They used a statistical model that accounted for starting values and repeated measurements over time. Immediately after treatment, the estimated difference between the real-treatment and sham groups in change from the starting measurement was 1.99 milliseconds. The 95% confidence interval ranged from −3.89 to 7.87 milliseconds. After one week, the estimated difference was 1.49 milliseconds, with a 95% confidence interval from −4.45 to 7.44 milliseconds. Both intervals included zero, meaning the results did not establish a clear difference between the groups. These analyses were exploratory, and the study was not large enough to test treatment effectiveness. The study therefore found a preliminary link with changes in brain activity but no detectable difference in heart rate variability. It did not demonstrate a pathway connecting spinal adjustment to changes in brain networks.
Several results suggested that a larger study could be carried out. All 30 participants returned for the one-week follow-up. Tests of whether participants could identify their assigned treatment were consistent with successful blinding. The Bang Blinding Index was 0.20 in both groups, with 95% confidence intervals from −0.30 to 0.70. Of the 90 EEG recordings collected, 82% were still suitable for analysis after data cleaning. For the 90 ECG recordings, 96% remained suitable. The researchers enrolled 30 of 91 eligible adults, or 33%. They judged that participant retention and ECG data quality were good enough to proceed without changes, but EEG data quality and recruitment needed improvement. Possible ways to improve EEG recordings include better preparation to reduce movement and muscle noise, more consistent placement of electrodes, and stricter checks of recording quality while the measurements are being collected.
The study design also had important weaknesses that prevent firm conclusions about cause and effect or clinical benefits. Alternating the assignments was intended to balance the number and locations of contacts between the groups. However, because the researchers did not use random assignment or hide the allocation process, they could not rule out selection bias or ensure that the groups were balanced for other factors that might affect the results. Although the measured characteristics of the groups were similar at the start, the sample was too small to assess other important factors reliably. The sham treatment controlled for whether the instrument delivered a quick, forceful push. But it could not show whether the results depended on treating a vertebral site that the chiropractor had identified as needing treatment, rather than delivering the same kind of push at a similar site. The researchers said that a third group receiving a push at a site not identified as needing treatment would be necessary to separate these possibilities. The study also involved generally healthy adults and did not measure symptoms or ask participants to report changes in their health. It therefore could not establish whether the changes in brain connectivity were linked to meaningful health benefits.
Changes in alertness are another possible explanation that the pilot study could not rule out. Delta and theta brain waves can change when people become sleepy. The researchers recorded EEG signals while participants had their eyes closed, after they had been lying face down for treatment. These conditions could make people less alert. The researchers kept the timing of each participant's sessions roughly consistent but did not standardize the timing across all participants. They also did not calculate formal measures of alertness or absolute brain-wave power in each frequency range. As a result, the observed pattern might reflect differences in how awake participants were rather than a specific effect of the adjustment. The statistical results also need to be treated carefully because the researchers tested 10 combinations: five frequency ranges at two time points after the starting measurement. The network-based correction controlled the chance of false-positive findings within each comparison, but not across all 10 comparisons. This increased the risk that some results could appear significant by chance. Both major groups of connections remained significant when the researchers used stricter thresholds to decide which connections to include. However, the number of connections in the delta group became much smaller as the threshold became stricter.
The findings are best viewed as an early signal that needs further testing, not as proof of a new clinical benefit. The study showed that researchers could carry out an EEG protocol with a sham treatment, keep all participants through the follow-up, and achieve results consistent with successful blinding. It also showed which parts of the procedure need improvement. A key question is whether a physical treatment applied to the spine might be linked to measurable changes in the way large groups of brain regions work together, even when measurements of the autonomic nervous system show no difference. This idea remains uncertain, especially when it comes to emotions, thinking, and memory, because the study did not measure these outcomes and some brain-network assignments relied on the authors' judgment. The researchers are planning a fully powered randomized controlled trial to compare real and sham adjustments in adults. Even then, confirming the findings, getting reliable estimates of the size of the effects, and determining whether they matter for health will require larger and more carefully designed studies. These studies should include standardized monitoring of alertness, clearer definitions of brain networks, and outcomes that can show whether brain measurements are linked to meaningful changes in health.
Subject of Research: An exploratory study of resting-state brain connectivity and heart rate variability after real versus sham chiropractic adjustment in adults.
Article Title: Differences between groups in resting-state functional connectivity within the brain’s extended triple-network after real and sham chiropractic adjustments: a quasi-experimental pilot trial
Article References: Sullivan, S., Sliwka, M., Drake, E., & Perez, T. M. (2026). Differences between groups in resting-state functional connectivity within the brain’s extended triple-network after real and sham chiropractic adjustments: a quasi-experimental pilot trial. Frontiers in Human Neuroscience, 20, Article 1984309. https://doi.org/10.3389/fnhum.2026.1984309
Image Credits: AI Generated
DOI: 10.3389/fnhum.2026.1984309
Keywords: chiropractic adjustment, resting-state connectivity, electroencephalography, sham-controlled pilot study, extended triple-network, heart rate variability, nonrandomized trial, brain networks, differences between groups, resting-state functional connectivity