1. Introduction
Pain is defined as an unpleasant sensory and emotional experience, and it is classified as chronic pain if it lasts beyond 3 months.53 The prevalence of chronic pain ranges from 7.1% to 61% among Asian populations.48 And chronic pain occurs annually among 1 in 5 U.S. adults.86 Currently, chronic pain has been one of the leading causes of disability and imposes substantial socioeconomic burdens.26,88 Therefore, treating patients suffering from chronic pain seems to be important in increasing global quality of life and reducing economic costs.
Because most pharmacological treatments for chronic pain have addictive properties or side effects,18,64 nonpharmacological management strategies, such as acupuncture, are commonly used.76,86 A meta-analysis conducted on 39 randomized controlled trials (RCTs) with a total of 20,827 patients has confirmed the effectiveness of acupuncture for treating chronic osteoarthritis, headache, and musculoskeletal pain.77 However, the efficacy varies across populations and patients and seems prone to a multiplicity of influences.23,32,71,76,77 Identifying influencing factors and developing approaches that predict medical benefits for each patient before treatment promise important scientific and clinical applications. Predictive factors may offer opportunities to design personalized medicine, improve treatments, enhance the success of clinical trials, and resolve the controversy of acupuncture effect.
Chronic pain is influenced to varying degrees by psychological and neurological factors.53 Current literature suggests that chronic pain responses to treatments (including acupuncture) are partially determined by baseline psychological factors,5,14,44,72,74,82 such as expectation, emotion, personality, and awareness. For instance, significant associations have been found between better improvement and higher patient expectations on outcomes in 4 RCTs of acupuncture in chronic pain.44 Additionally, accumulating evidence indicates that baseline brain magnetic resonance imaging (MRI) metrics, including functional connectivity, spontaneous activity, and brain volume, can predict the improvement of chronic pain after acupuncture and placebo treatment.66,69,70,74,84,85 A recent study has shown that pain relief after acupuncture in chronic low back pain can be mainly predicted by functional connectivity among default mode network (DMN) and subcortical striatal (ie, caudate and putamen) regions.69 Although both psychological and neurological factors are essential in the efficacy prediction of chronic pain, they have rarely been jointly studied.
This study aims to identify the psychological and neuroimaging metrics collected before acupuncture treatment that would determine acupuncture efficacy in knee osteoarthritis (KOA), whose major characteristic is chronic pain. Recent neuroimaging advances have demonstrated the pivotal role of the cortical regions (DMN and insula) in KOA pathophysiology.16 Besides, patients with higher arthritic pain levels had greater opioid receptor in the striatum.12 We speculated that effects of acupuncture treatment are mainly predetermined by structure and function of key cortical systems (eg, DMN) and subcortical striatum because they have been frequently implicated in chronic pain,12,16,38 pain modulation,7,49 and efficacy prediction of acupuncture.69 And the psychological and neurological factors may be complementary to one another in efficacy prediction.
2. Materials and methods
2.1. Study design
This randomized controlled neuroimaging trial was approved by the ethic committee of Dongzhimen Hospital in Beijing University of Chinese Medicine (DZMEC-KY-2017-53-02), preregistered in Chinese Clinical Trial Registry (ChiCTR1900025807),42 and conducted in accordance with the Declaration of Helsinki. Because of the coronavirus pandemic, the second stage of the intended crossover design was not continued for the participants. The current study consisted of a baseline assessment, 4-week acupuncture or sham-acupuncture treatment, and an assessment after treatment. Written informed consent was signed by all participants before the study. Participants were not compensated, but all MRI scanning and acupuncture treatments were free.
2.2. Participants
Participants with KOA were initially recruited from the general population via social media (WeChat), leaflets, and advertising in hospitals or community service centers. The diagnosis of KOA was based on the American College of Rheumatology clinical criteria.33 Patients were evaluated for eligibility by a screening interview covering age, pain levels/duration/location, comorbid health, current and previous treatment status, MRI safety, and willingness to participate in the trial.
To meet inclusion criteria, participants had to be 45 to 65 years old with a ≥6-month history of knee pain and report a pain level ≥4 (range, 0-10) on the numerical rating scale (NRS) in the past week. Additionally, the enrolled patients should have radiologic confirmation of osteoarthritis (Kellgren–Lawrence score II or III).35 To rule out the influence of handedness on the neuroimaging results, only right-handed patients were included here.
The exclusion criteria included the history of knee injection in the past 6 months or arthroscopy in the previous 12 months, knee surgery or waiting for knee surgery, knee pain caused by other diseases, psychiatric or neurological disorders, coagulation disorders, severe acute or chronic organic diseases, MRI contraindications (eg, cardiac pacemaker, claustrophobia, or other metallic agents embedded within the body), alcohol or drug abuse, pregnancy or lactation, acupuncture treatment in the last 1 month, or participated in other clinical trials in the past 3 months.
2.3. Sample size
Based on our previous randomized controlled clinical trial,71 we anticipated a mean decrease of 2.0 NRS scores after acupuncture treatment and a mean decrease of 1.0 NRS scores after sham-acupuncture treatment, with an estimated SD of 1.3; this results in an effect size estimate of 1.7 for post-pre differences after acupuncture treatment and 0.8 for group differences. Power analyses performed in G*Power version 3.1.9.7, with α = 0.05, 1 − β = 0.8, and a SD of 1.3, indicated that we needed 26 participants per group to detect anticipated effects. Furthermore, a sample size of 26 in each group also permits the detection of medium (∼0.5) correlation effects in each group. For calculating sample size in MRI studies, there is no precise standard currently. Here, the number of patients was based on experience in the previous studies with chronic pain populations and MRI techniques, where a sample size ranging from 13 to 20 is sufficient to detect a significant difference in structural and functional brain metrics.52,66,69,74 Considering the attrition rate and excessive head motion–induced low-quality image, we recruited 30 patients in each group.
2.4. Randomization and blinding
Eligible participants were randomly assigned to the acupuncture group (AG), sham-acupuncture group (SG), and waiting group (WG) with equal proportions. The randomization sequence was generated using SAS software (version 9.3) with a block of 6 by an independent researcher. Acupuncturists were not blinded to the allocation for successfully performing acupuncture treatment. However, they were not allowed to discuss the intervention type with patients. The patients, MRI operators, and outcome evaluators were all blinded to group assignment.
2.5. Interventions
Each patient received 12 sessions of 30-minute acupuncture treatment during 4 weeks, with 3 sessions per week. To ensure that patients can receive treatment on time and improve adherence, the treatment arrangement should consider patients' time. The treatment time for each patient was recorded by researchers for monitoring adherence. In Dongzhimen Hospital, patients were treated in different cubicles to refrain from communication. Registered acupuncturists with more than 5 years of experience performed intervention procedures. At the beginning of the study, all acupuncturists were trained by a qualified researcher based on standard operating procedures about the manipulation of needles and locations of acupoints/nonacupoints. Rescue medication (Paracetamol, Tylenol; Shanghai Johnson & Johnson Pharmaceuticals, Ltd., Shanghai, China) was provided as needed. Adverse events and usage of rescue medication were recorded according to the reports of the participants and acupuncturists during the 4-week treatment.
2.5.1. Acupuncture treatment
The acupuncture prescription was from expert consensus65 in traditional Chinese medicine and our previous multicenter RCT.71 The acupoints included Dubi (ST 35), Neixiyan (EX-LE4), Ququan (LR8), Xiyangguan (GB33), Sanyinjiao (SP6), Xuehai (SP10), Taixi (KI3), and an ashi point (the point with most pain for the patient). The acupoint locations based on WHO Standard Acupuncture Locations are shown in Figure S1A and Table S1 (available as supplemental digital content at https://links.lww.com/PAIN/B778). The disposable needles (0.25 × 25 mm or 0.25 × 40 mm; Guizhou Andi Pharmaceutical Machinery, Ltd., Guizhou, China) were used to penetrate into the skin 5 to 20 mm through the 1.6 × 15-mm sterile adhesive pad (Fig. S1B, available as supplemental digital content at https://links.lww.com/PAIN/B778). De qi sensations (ie, numbness, soreness, distension, and heaviness) were achieved through the manual stimulation of the needles by the acupuncturists. Then, the needles were retained for 30 minutes.
2.5.2. Sham-acupuncture treatment
Nonacupoints were used in the sham-acupuncture group (please see Fig. S1A and Table S2 for location details, available at https://links.lww.com/PAIN/B778). The blunt single-use needles (0.25 × 25 mm or 0.25 × 40 mm; Hwato) were used to penetrate into the adhesive pad without penetrating into the skin (Fig. S1B, available as supplemental digital content at https://links.lww.com/PAIN/B778). Then, the blunt needles were retained for 30 minutes. The combination of needle with blunt tip and adhesive pad is a validated placebo needling method and had a good blinding effect for patients.45,46
2.6. Clinical outcome measures
Patients completed pain-related questionnaires at baseline and after treatment (week 4). The more painful knee at baseline was evaluated in patients with bilateral knee osteoarthritis during the study, whereas the affected knee was evaluated in patients with unilateral osteoarthritis.81 The primary clinical outcome was the changes in pain on the 11-point NRS. The NRS is a classical standard pain measurement used in clinical for pain level assessment. The minimal clinically important improvement (MCII) of NRS was set to 2 points.71 The secondary clinical outcomes were the changes in the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC)8 and short-form McGill Pain Questionnaire (MPQ).47 The short-form MPQ includes sensory and affective facets of pain, whereas the WOMAC measures pain, stiffness, and function. Minimal clinically important improvement on the WOMAC function subscale (68-point) was set to 6 points.71 The changes in clinical outcomes could also be dichotomized into clinical improvement and nonclinical improvement based on the previously validated criteria71: improvement in NRS ≥2 points and improvement in WOMAC function subscale ≥6 points. Minimal clinically important improvement on the MPQ sensory and affective have been estimated as 4.5 and 2.8 in a previous osteoarthritis population.29
2.7. Psychological measures
Patients were asked to complete 4 psychological questionnaires including the Neuroticism Extraversion Openness Five-Factor Inventory (NEO-FFI, Costa & McCrae, available from www.parinc.com), Stanford Expectations of Treatment Scale (SETS),87 Trait Meta-Mood Scale (TMMS),56 and Mindful Attention Awareness Scale (MAAS).11 The NEO-FFI is a 60-item version implementation of the extensively empirically validated 5-factor model of human personality, which includes extraversion, openness, conscientiousness, agreeableness, and neuroticism factors. The SETS is a widely used scale for measuring the outcome expectancy of patients in clinical trials. The TMMS is a measure of emotional intelligence that involves 3 dimensions: emotional attention, clarity of feelings, and emotional regulation. The MAAS is a standard and commonly used self-reported scale to assess dispositional mindfulness as a general tendency to “be attentive to and aware of present-moment experiences.”
2.8. Magnetic resonance imaging data acquisition
The MRI scanning was performed at baseline and after treatments using a 3.0-Tesla scanner (Skyra, Siemens, Erlangen, Germany) in the Beijing Hospital of Traditional Chinese Medicine. High-resolution brain T1-weighted (T1w) MRI was obtained using a sagittal 3D magnetization-prepared rapid gradient echo (MPRAGE) sequence: repetition time (TR)/echo time (TE)/inversion time = 2530/2.98/1100 milliseconds, flip angle = 7°, slice number = 192, matrix = 256 × 256, voxel size = 1 × 1 × 1 mm3, slice thickness = 1 mm, slice gap = 0 mm. Resting-state fMRI was scanned with an echo planar imaging (EPI) sequence: TR = 2000 milliseconds, TE = 30 milliseconds, flip angle = 90°, field of view (FOV) = 224 mm × 224 mm, slice spacing = 4.375 mm, slice number = 32, number of volumes = 240, matrix = 64 × 64, voxel size = 3.5 × 3.5 × 3.5 mm3. Participants were instructed to remain still and keep their eyes open during the acquisition. The procedure was repeated at baseline and after intervention (week 4).
2.9. Magnetic resonance imaging data processing
2.9.1. Image preprocessing
Image preprocessing was performed using DPABISurf V1.5 (http://rfmri.org/DPABISurf),83 which is a user-friendly pipeline based on MATLAB (The MathWorks Inc, Natick, MA), fMRIPrep,20,21 FreeSurfer,24 ANTs,4 FSL,34 AFNI,17 and SPM.3 The DPABISurf pipeline first converts the data into Brain Imaging Data Structure (BIDS) format28 and then calls fMRIPrep to preprocess the structural and functional MRI data automatically. fMRIPrep is a robust tool to prepare human fMRI data for statistical analyses. Many internal realizations of fMRIPrep 20.2.1 use Nipype 1.5.122,27 and Nilearn 0.6.2.1 For more details about the pipeline, please see the corresponding fMRIPrep's documentation. The following text about fMRIPrep's preprocessing steps is based on boilerplate released with the tool under CC0 license (the users could copy and paste this text into the manuscripts unchanged).
2.9.2. Structural image preprocessing
The T1w images were corrected for intensity nonuniformity with N4BiasFieldCorrection,73 distributed with ANTs 2.3.3 and used as T1w reference throughout the workflow. The T1w reference was then skull-stripped with a Nipype implementation of the antsBrainExtraction.sh workflow (from ANTs), using OASIS30ANTs as the target template. Segmentation of the subcortical regions, reconstruction of the cortical surface, and tessellation of the pial surface and gray/white matter boundary were conducted using “recon-all” command implemented in FreeSurfer 6.0.1. The brain mask estimated previously was refined with a custom variation of the method to reconcile ANTs-derived and FreeSurfer-derived segmentations of the cortical gray matter of Mindboggle.36 Volume-based spatial normalization to ICBM 152 Nonlinear Asymmetrical template version 2009c (MNI152NLin2009cAsym) standard space25 was performed through nonlinear registration with antsRegistration (ANTs 2.3.3), using brain-extracted versions of both T1w reference and the T1w template. Brain tissue segmentation of cerebrospinal fluid (CSF), white matter (WM), and gray matter (GM) was performed on the brain-extracted T1w using fast in FSL 6.0.
2.9.3. Cortical thickness and subcortical volume extraction
A direct measure of cortical thickness (mm) was calculated using the shortest distance between the pial surface and gray/white matter boundary at each vertex. Subsequently, mean thickness was extracted for each of the 68 cortical regions (34 per hemisphere) in the Desikan–Killiany (DK) atlas. In addition, the volumes (mm3) of striatum subcortical structures (caudate, putamen, and pallidum) based on DK atlas were extracted for each participant, as well as estimated total intracranial volume (eTICV).
2.9.4. Functional image preprocessing
Each functional data's first 10 volumes were removed to allow for the stabilization of the magnetic field. Then, a reference volume and its skull-stripped version were generated using a custom methodology of fMRIPrep. The BOLD reference was then co-registered (6 degrees of freedom) to the T1w reference using bbregister (FreeSurfer), which implements boundary-based registration.30 Head-motion parameters with respect to the BOLD reference are estimated before any spatiotemporal filtering using MCFLIRT in FSL 6.0. The BOLD runs were slice-time–corrected using 3dTshift from AFNI 20160207. The BOLD time series (including slice-timing correction) were resampled onto their original, native space by applying the transforms to correct for head motion. The BOLD time series were resampled into MNI152NLin2009cAsym space. Volumetric resamplings were performed using antsApplyTransforms (ANTs), configured with Lanczos interpolation to minimize the smoothing effects of other kernels.39 The preprocessed functional images were resampled into a voxel size of 2 × 2 × 2 mm3. Confounding signals, including Friston 24 head-motion parameters, WM, and CSF signals, were removed from the data via linear regression.
2.9.5. Fractional amplitude of low-frequency fluctuation calculation
The fractional amplitude of low-frequency fluctuation (fALFF) is an important resting-state fMRI measure that can reveal the magnitude of local spontaneous activity. The calculation of the fALFF value referred to the procedures described in previous studies.89,90 For the time series in each voxel, fALFF was computed as the sum of the amplitudes within the 0.01 to 0.08 Hz frequency range divided by the sum of amplitudes across the entire 0 to 0.25 Hz frequency range. The voxel-wise fALFF map of each individual was standardized into Z-map (by subtracting the whole-brain mean fALFF and dividing the SD) and then spatially smoothed by a 6-mm FWHM Gaussian kernel. The mean fALFF values in the regions of interest (ROIs) were extracted for further statistical analyses.
2.9.6. Graph-based brain network analysis
Before brain network construction, the preprocessed functional data were bandpass filtered within a frequency range of 0.01 to 0.08 Hz. The time series of the 34 DMN ROIs corresponding to the well-known Dosenbach 160 ROIs template (radius of 5 mm)19 were extracted. Because subcortical striatum structures are believed to play a role in chronic pain, six 5-mm radius ROIs of bilateral caudate, putamen, and pallidum in Tian subcortical template68 were added. The Fisher r-to-z transformed Pearson correlation between time series of all ROI pairs (40 × 40 correlation matrix) within 40 ROIs (34 DMN and 6 striatum) were calculated. To ensure the consistency of the resulting graph metrics (ie, with same numbers of edges) and simplify the brain networks, the sparsity was generally used as the threshold to filter the individual correlation matrix.2 Because there is no golden standard for thresholding, we simplified each original correlation matrix over a relatively wide range of sparsity (10%-30%) with 1% step length. In addition to each individual network, 100 random graphs were constructed with the same number of edges and nodes to serve as a baseline for comparison. A key property of complex network, small-worldness,63,80 was examined for all individual weighted brain network at various sparsities based on the graph theory framework in Brain Connectivity Toolbox.55 A small-worldness value larger than 1 is believed to reflect an optimal balance between segregation and integration.2
2.9.7. Image quality control
The quality of functional images and co-registration were assessed using fMRIPrep's visual reports. Participants with more than 2.5 mm of displacement in x, y, or z or 2.5 degrees of angular head motion were excluded in further analyses.
2.10. Statistical analyses
2.10.1. Baseline characteristics
Patients' demographical, clinical, and psychological variables were analyzed using SPSS statistics software. We first performed Shapiro–Wilk test to check the normality of the continuous variable in each group. For continuous data with normal distribution in each group, mean and SD were summarized. For nonnormally distributed continuous data, median and a centile range (25th and 75th centiles) were presented. Numbers and proportions were reported for categorical variables.
2.10.2. Acupuncture treatment effect calculation
For data with normal distribution in both groups, 2-sample t test was applied to compare treatment improvements; for nonnormally distributed improvements, we used nonparametric Mann–Whitney U test to compare the group differences. Six P values in clinical outcome comparisons were corrected for false discovery rate (FDR) using the Benjamini–Hochberg procedure. To further control the influences of baseline pain and other potential prognostic values, we performed analysis of covariance (ANCOVA) with NRS at 4 weeks as dependent variable, and including baseline NRS, age, and sex as covariates.
2.10.3. Relation between clinical measurements and psychological factors
The Pearson correlation and Spearman correlation analyses were used to investigate the relationships between each subscale in the psychological questionnaire data and the change (4-week minus baseline) of each clinical measurement among all 27 patients in the AG. Besides, 2-sample t test or nonparametric Mann–Whitney U test was performed to identify group differences in psychological factors (patients with clinical improvement vs nonclinical improvement within the AG, P < 0.05). All 70 P values were also checked if they survived FDR correction (P < 0.05).
2.10.4. Relation between brain metrics and primary clinical outcome
Because the distribution of NRS changes was not normal in the AG, nonparametric Spearman correlations were used to examine the correlations between brain metrics (subcortical striatum volume, cortical thickness, fALFF, and small-worldness restricted to the interested brain regions) and NRS changes. In addition, the group differences in brain metrics between clinical improvement patients and nonclinical improvement patients within the AG were also tested using Mann–Whitney U test for nonnormally distributed data or 2-sample t test for normally distributed data. All P values in each type of measurement (3 in subcortical volumes, 68 in cortical thickness, 12 in fALFF, and 1 in network metrics) were also checked if they survived FDR correction (P < 0.05).
2.11. Machine learning for predicting and classifying acupuncture improvement
Support vector machine (SVM) with linear kernel was used to distinguish clinical improvement and nonclinical improvement using libsvm MATLAB toolbox. We used the classical leave-one-out cross-validation (LOOCV) procedure where machine learning models were trained in n − 1 participants. Once the optimal SVM model and C regularization parameter were obtained, they were used to classify the left-out participant. The accuracy of each classification analysis was also examined against the null distribution generated using 1000 scrambled data with randomized labels.
Support vector regression (SVR) with linear kernel was used to predict primary clinical outcome (ie, NRS) changes using libsvm MATLAB toolbox and LOOCV procedure. Once the optimal SVR model (regularization parameter C = 1) was obtained, it was used to predict the left-out participant. The correlation and root mean square error (RMSE) between the predicted and true NRS changes were calculated.
The features were selected from questionnaires (psychological and clinical variables) and brain metrics that have significant relation with acupuncture improvement. The selected features were normalized because they were obtained from different measurements.
3. Results
3.1. Participants and baseline characteristics
The total duration of the study lasted 20 months (from October 30, 2019 to July 5, 2021). Figure 1 illustrates the flow of patients through the study. Of the 154 enrolled participants with KOA, 64 failed the screening. The remaining 90 patients were randomized into 1 of 3 groups: acupuncture treatment (n = 30), sham-acupuncture treatment (n = 30), or waiting treatment (no treatment during the experiment, n = 30). In the AG, 3 patients were discontinued from the study; of the SG, 4 patients were discontinued from the study, with an additional patient being excluded from the final analysis because of having head motion larger than 2.5 mm during fMRI scanning; of the WG, 6 patients were discontinued from the study, with 2 additional patients being excluded because of extreme head motion. The final sample size included 27 participants with KOA in the AG, 25 participants with KOA in the SG, and 22 participants in the WG. The baseline characteristics (demographical, clinical, and psychological variables) are shown in Table S3, available at https://links.lww.com/PAIN/B778. The demographics and pain levels at baseline were equivalent among groups (Table S3, available at https://links.lww.com/PAIN/B778). Because only a single-digit number of patients in the WG completed MPQ, NEO-FFI, MAAS, and TMMS questionnaires, we mainly focused on the data analyses in the AG and SG. The correlation matrix of different pain measurements (NRS, WOMAC, and MPQ) across all patients in the AG and SG at baseline is presented in Figure S2, available at https://links.lww.com/PAIN/B778.
;)
Figure 1.: CONSORT flow diagram of the study. CONSORT, Consolidated Standards of Reporting Trials.
3.2. Acupuncture treatment effect
We tested for the effects of acupuncture intervention by comparing the clinical changes between the 27 patients receiving acupuncture treatment and the 25 patients comprising the sham-acupuncture arm. For NRS changes (4-week minus baseline), a significant group difference was found (Cohen d = 0.900, P = 0.002, FDR-corrected P = 0.012 < 0.05; Table 1, Fig. 2A), that is, acupuncture treatment showed a stronger diminution in pain intensity compared with sham-acupuncture (controlling for placebo response or regression to the mean). Importantly, the overall magnitude of NRS changes in the AG was greater than MCII, which is comparable with the efficacy reported in our previous multicenter clinical trial on patients with KOA.71 Moreover, the ANCOVA confirmed the group differences in the primary outcome (ie, NRS) after adjusting for baseline pain, age, and sex (F = 6.124, P = 0.017, partial eta square = 0.115). For intent-to-treat analysis, multiple imputation method was used to deal with missing values, and 5 imputation datasets were analyzed to compare group differences. The overall group differences were still significant (P = 0.003).
Table 1 -
Acupuncture effect on clinical measurements.
| Measurements |
AG (n =27) [4-wk − baseline] |
SG (n =25) [4-wk − baseline] |
Cohen d (95% CI) |
P
|
| NRS score |
−3.00 (−4.00, −2.00) |
−2.00 (−2.00, −1.00) |
0.900* |
0.002† |
| WOMAC pain |
−1.85 (2.44) |
−1.60 (2.53) |
0.101 (−0.443, 0.646) |
0.717 |
| WOMAC stiffness |
0.00 (−1.00, 0.00) |
0.00 (−1.00, 0.00) |
0.089* |
0.731 |
| WOMAC function |
−5.37 (6.55) |
−4.96 (7.46) |
0.059 (−0.485, 0.603) |
0.834 |
| MPQ sensory |
−2.15 (2.46) |
−1.52 (2.04) |
0.277 (−0.270, 0.823) |
0.324 |
| MPQ affective |
−2.03 (1.89) |
−1.00 (−2.00, 0.00) |
0.407* |
0.145 |
We used mean (SD) if the measurements were normally distributed and median (lower quartile to upper quartile) if the measurements were not normally distributed in the Shapiro–Wilk test. For data with normal distribution in both groups, 2-sample t test was applied to compare baseline characteristics.
†P < 0.01 and FDR-corrected P < 0.05.
AG, acupuncture group; CI, confidence interval; MPQ, McGill Pain Questionnaire; NRS, numeric rating scale; SG, sham-acupuncture group; WOMAC, Western Ontario and McMaster Universities Osteoarthritis Index.
Figure 2.: Acupuncture intervention diminishes knee pain. (A) The patients in the acupuncture group (AG) showed higher pain intensity improvements compared with the patients in the sham-acupuncture group (SG). (A and B) Previously validated criteria (absolute NRS change ≥2 and absolute WOMAC function change ≥6) identified clinical improvement in the AG. (C) WOAMC pain, (D) WOMAC stiffness, (E) MPQ sensory, and (F) MPQ affective dissociated clinical improvement (AGci) and nonclinical improvement (AGnci) in the AG but showed nonspecific improvements with exposure to the trial (AG and SG arms equally improving). Only (E) MPQ sensory dissociated clinical improvement (SGci) and nonclinical improvement (SGnci) in the SG, and no treatment and clinical improvement interaction effect was observed (P > 0.05). In (A–F), number of subjects are in parentheses; *P < 0.05; **P < 0.01; ****P < 0.0001. For data with normal distributions in the Shapiro–Wilk tests, 2-sample t test was applied, and for nonnormally distributed data, nonparametric Mann–Whitney U test was used to compare the group differences. For display purposes, bars indicate mean and error bars indicate SEM. MPQ, McGill Pain Questionnaire; NRS, numeric rating scale; ns, not significant; WOMAC, Western Ontario and McMaster Universities Osteoarthritis Index.
Western Ontario and McMaster Universities Osteoarthritis Index and MPQ were used as secondary pain outcomes. The WOMAC subscales and MPQ affective subscale significantly correlated with our primary pain outcome (ie, NRS; Fig. S2, available at https://links.lww.com/PAIN/B778). However, although there were trends of higher improvement in the AG, these measurements did not significantly differentiate between treatment groups (Table 1, Fig. 2), they (except WOMAC stiffness) were significantly changed after intervention in both groups (for more details, please see Fig. S3, available at https://links.lww.com/PAIN/B778).
Exploratorily, the clinical changes were dichotomized into clinical improvement and nonclinical improvement based on the validated criteria71: |NRS change| ≥ 2 points and |WOMAC function change| ≥ 6 points. We found WOMAC pain, WOMAC stiffness, MPQ sensory, and MPQ affective could significantly dissociate clinical improvement and nonclinical improvement in the AG (all P < 0.05; Figs. 2C–F). Therefore, despite variability across these clinical measures, the clinical improvement effects of acupuncture were globally concordant. However, only MPQ sensory dissociated clinical improvement and nonclinical improvement in the SG (Fig. 2E). In the following sections of the study, we concentrate on unraveling the mechanisms of the acupuncture-induced decrease in knee pain.
In AG, 11.1% of participants (3/27) reported acupuncture-related adverse events (1 subcutaneous congestion and 2 postneedling sensation) compared with 4% of participants (1/25, subcutaneous congestion) in the SG. These adverse events could recover without treatment. No serious adverse event occurred. All 52 participants completed more than 10 of 12 treatment sessions.
Besides, we compared the clinical improvements in the AG with the WG to reveal the effects of “sit-and-wait” approaches. We found stronger improvements in NRS, WOMAC pain, and WOMAC function after acupuncture treatment compared with no treatment (Table S4, available at https://links.lww.com/PAIN/B778).
3.3. Baseline psychological factors relating to acupuncture effect
First, we sought to identify psychological factors predisposing patients with KOA to greater symptom improvements from a battery of 4 questionnaires with 10 subscales collected at baseline. Univariate statistics were used to assess correlations with the magnitude of clinical changes and group differences among clinically improved and nonclinically improved patients. Here, NEO-FFI extraversion, TMMS emotional attention, and SETS positive expectation were correlated with the magnitude of clinical improvements (Figs. 3A–E; P < 0.05 or marginally significant P < 0.06 in both the Pearson correlation and Spearman correlation analyses). We also found that SETS positive expectation was different between clinically improved and nonclinically improved patients in the AG (t(25) = 2.087, P = 0.047; Fig. 3F).
Figure 3.: Psychological factors were related to the magnitude of clinical changes. (A) The Pearson correlation r matrix and Spearman correlation rho matrix show the relationships between each subscale in the psychological questionnaire data and each clinical measurement. Extraverted patients experienced less acupuncture-induced WOMAC function improvement (B), whereas patients with positive expectations experienced more WOMAC function improvements (C). Patients with higher TMMS emotional attention scores had larger improvements in MPQ sensory (D) and WOMAC pain (E). (F) T-tests showed SETS positive expectation was significantly different (t (25) = 2.087, P = 0.047) between clinical improvement and nonclinical improvement in the AG. **P < 0.01; *P < 0.05; $0.05 ≤ P < 0.06. AG, acupuncture group; MAAS, Mindful Attention Awareness Scale; MPQ, McGill Pain Questionnaire; NEO-FFI, Neuroticism Extraversion Openness Five-Factor Inventory; NRS, numeric rating scale; SETS, Stanford Expectations of Treatment Scale; TMMS, Trait Meta-Mood Scale; WOMAC, Western Ontario and McMaster Universities Osteoarthritis Index.
3.4. Structural properties relating to acupuncture effect
We secondly sought for baseline structural properties predisposing patients with KOA to greater symptom improvements. The volumes of the caudate, putamen, and pallidum were first examined because these striatum structures are important factors for motion function and pain modulation. Spearman correlation analyses were performed between the primary clinical outcome (ie, NRS change) and the 3 subcortical volumes. Results in Figure 4A showed that the caudate volume was correlated with NRS change (Spearman rho = −0.480, P = 0.011, FDR-corrected P < 0.05), whereas the putamen and pallidum volumes were not linked to NRS changes (Spearman rhoputamen = −0.144, P = 0.475; Spearman rhopallidum = −0.139, P = 0.491). When controlling for age and sex, the correlation between NRS change and caudate volume still remained (Spearman rho = −0.431, P = 0.025). Moreover, the volumes of caudate, putamen, and pallidum were significantly different between clinically improved and nonclinically improved patients in the AG (all P < 0.05 with Mann–Whitney U test, Fig. 4B). The locations of the aforementioned subcortical structures are shown in Figure 4C. The structural properties of the cortex were assessed with cortical thickness. Among all 68 cortical regions in the DK atlas, mean thickness of the left posterior cingulate cortex (PCC) was correlated with NRS changes (Spearman rho = 0.410, P = 0.034; Spearman rho = 0.421, P = 0.029 when controlling for age and sex) and also was different between clinically improved and nonclinically improved patients in the AG (Mann–Whitney U test, P = 0.048; Fig. 4D). The aforementioned structural differences were also observed in the 4-week MRI data (Figs. 4B and E). The identification of brain structural features, present before treatment and persisting throughout the study, provides evidence for acupuncture propensity stemming, in part, from stable neurobiology.
Figure 4.: Primary outcome changes and clinical improvement of acupuncture intervention are predetermined by subcortical striatum volume and posterior cingulate cortex thickness. (A) The caudate volume correlated with the magnitude of the primary outcome (numeric rating scale [NRS]) changes, whereas putamen and pallidum volumes were not significantly correlated with NRS changes. (B) The difference in subcortical volumes between patients with clinical improvement (AGci) and nonclinical improvement (AGnci) in the acupuncture group (AG) was observed at baseline and after treatment. (C) Brain map displays the location of the caudate, putamen, and pallidum. (D) Whole-brain neocortical region-wise analyses showed both significant NRS-thickness correlation and thickness of AGnci greater than AGci in the left posterior cingulate cortex (PCC). This effect was consistent across baseline and 4-week scans (E). *P < 0.05. Error bars indicate SEM. SG, sham-acupuncture group. Unit of volumes: mm3, unit of cortical thickness: mm.
3.5. Functional properties predisposing to acupuncture effect
Thirdly, we tested whether baseline functional properties provided useful information to determine primary outcome (ie, NRS) changes. First, we mainly examined the relationship between mean fALFF in the regions with significant structural–clinical links. We found that the fALFF value in the PCC/precuneus (Fig. 5A, from Yeo's 7-network template67), a major region of the DMN, correlated with NRS changes in the AG (Spearman rho = −0.570, P = 0.003, FDR-corrected P < 0.05; Spearman rho = −0.628, P = 0.001, when removing one outlier in fALFF value [lower than Q1 − 3IQR]; Spearman rho = −0.510, P = 0.007, when adjusting age, sex, and head motion [mean root-mean-square]; Fig. 5B). Clinically improved patients in the AG displayed higher fALFF values than nonclinically improved patients in the PCC/precuneus (t(24) = 2.185, P = 0.039, with one outlier removed), but this difference disappeared after treatment (Fig. 5C). For the subcortical striatum structures, the pallidum fALFF correlated with the NRS changes (Spearman rho = −0.391, P = 0.044; Spearman rho = −0.392, P = 0.043, when adjusting age, sex, and head motion), whereas caudate and putamen fALFF were not significantly correlated with NRS changes (Fig. 5D). Additionally, only putamen fALFF values were significantly different between clinically improved and nonclinically improved patients at baseline (t(25) = 2.546, P = 0.017), whereas the group differences dissipated at 4 weeks (Fig. 5E).
Figure 5.: Primary outcome changes and clinical improvement of acupuncture intervention are related to functional properties of the default mode network and subcortical regions. (B) The fALFF in the PCC/precuneus (A), a major region of the default mode network (DMN), correlated with numeric rating scale (NRS) changes (Spearman rho = −0.570, P = 0.003) in the acupuncture group (AG). The black dotted circle marks an outlier (lower than Q1 − 3IQR) on the PCC/precuneus fALFF. (C) The difference of PCC/precuneus fALFF between patients with clinical improvement (AGci) and nonclinical improvement (AGnci) in the AG was observed at baseline but not after intervention (when removing one outlier). (D) The pallidum fALFF correlated with the NRS changes, whereas caudate and putamen fALFF were not significantly correlated with NRS changes. (E) The difference of putamen fALFF between AGci and AGnci was observed at baseline but not after intervention. (F) DMN (orange spheres) and subcortical (caudate, putamen, and pallidum; yellow spheres) regions were used to construct brain network of interest. (G) Small-worldness measure (Sigma) in the graph-based analyses across different graph sparsities (10% to 30%). (H) The area under the curve (AUC) of Sigma was positively correlated with NRS change. (I) The difference in Sigma AUC between AGci and AGnci was observed at baseline but not at 4 weeks. *P < 0.05; $0.05 ≤ P < 0.06. Error bars indicate SEM. fALFF, fractional amplitude of low-frequency fluctuations; PCC, posterior cingulate cortex; SG, sham-acupuncture group.
We then examined the small-worldness (Sigma) of the brain network constructed from 40 preselected ROIs (34 DMN and 6 subcortical, Fig. 5F). The area under the curve (AUC) of Sigma across different graph sparsities (10% to 30% with steps of 1%, Fig. 5G) was positively correlated with NRS change (Spearman rho = 0.407, P = 0.035; Spearman rho = 0.424, P = 0.028, when adjusting age, sex, and head motion; Fig. 5H). Small-worldness of the networks across all sparsities were larger than 1 (Fig. 5G), reflecting an optimal balance between segregation and integration here. The difference in Sigma AUC between clinically improved and nonclinically improved patients in the AG was observed at baseline (t(25) = −2.127, P = 0.044) but not at 4 weeks (Fig. 5I). These functional results, therefore, demonstrate the existence of a DMN and subcortical striatum network, whose local spontaneous activity and global topological property transiently determine the acupuncture effect.
We then examined the variability of the aforementioned structural and functional brain metrics in patients of the SG. We observed no changes across the visits, indicating the stability of the metrics without true acupuncture effects (Figs. 4B and E, Figs. 5C, E, and I, all P > 0.05). We further tested if the mere exposure to acupuncture intervention, regardless of the improvement, impacted these brain metrics by comparing the AG with the SG. These analyses revealed the absence of acupuncture intervention effects on structural and functional brain metrics. The results suggest that the changes observed in the brain metrics were primarily driven by the actual acupuncture improvement rather than by true or sham acupuncture exposure.
3.6. Classifying acupuncture improvement using support vector machine
We used SVM with LOOCV procedure to determine whether acupuncture improvement in the AG could be predicted from baseline questionnaires and brain imaging data. Within each n − 1 patients training sample set, the normalized scores of the features were used to build the SVM classifier. We initially used questionnaires data to classify the patients into clinically improved group or nonclinically improved group. Here, we selected SETS positive, extraversion, emotional attention, disease duration, and body mass index (BMI) as features (Fig. 6A). Obesity is one of the important risk factors for developing KOA, which is also associated with increased pain-related disability and reduced physical functioning. A previous cross-sectional survey among 6524 elderly individuals in China has found the significant correlation between BMI and legs/feet/back chronic pain.41 Consequently, BMI was included in the prediction modeling here. Support vector machine classification achieved an accuracy of 66.67% and an AUC of 65.93% in classifying acupuncture response (P = 0.092 comparing with null distribution, Figs. 6B and C).
Figure 6.: Machine learning classifies acupuncture improvement (clinical improvement or nonclinical improvement). A support vector machine (SVM) classifier was applied to the baseline questionnaire data (A) or brain metrics (D) with a leave-one-out cross-validation (LOOCV) procedure to classify participants into clinical improvement or nonclinical improvement in the acupuncture group (AG). (B and E) The accuracy of the classification analysis is displayed against the null distribution generated using 1000 scrambled data with randomized labels. (G) The accuracy of the classification model using combining features (both baseline questionnaire data and brain metrics) was 81.48% and significantly higher than null distribution. However, the classification model constructed from the combining features failed to predict acupuncture response in the patients of SG (J). (H and K) Classification performance was also shown as confusion matrix. (C, F, I, and L) The receiver operating characteristic (ROC) curve shows true-positive rate and false-positive rate of the model. AUC, area under curve; BMI, body mass index; eTICV, estimated total intracranial volume; fALFF, fractional amplitude of low-frequency fluctuations; PCC, posterior cingulate cortex; SETS, Stanford Expectations of Treatment Scale.
We next trained a classifier using brain metrics. Specifically, caudate, putamen, pallidum, eTICV volumes, left PCC thickness, PCC/precuneus fALFF, and small-worldness (Sigma AUC of the brain network constructing from 40 preselected ROIs) were selected as features (Fig. 6D). Support vector machine classification achieved an accuracy of 62.96% and an AUC of 71.98% in classifying acupuncture improvement (P = 0.130 comparing with null distribution, Figs. 6E and F). Therefore, the above 2 models failed at classifying clinically improved and nonclinically improved patients in AG above the randomly scramble level (P > 0.05).
When combining questionnaires and brain imaging features, the model can successfully classify acupuncture improvement with an accuracy of 81.48% (P = 0.004, Figs. 6G and H) and an AUC of 88.46% (Fig. 6I). These results indicate that questionnaires and brain metrics are complementary to one another.
3.7. Predicting primary outcome changes using support vector regression
We used SVR with LOOCV procedure to determine whether primary outcome changes in the AG could be predicted from baseline psychological and brain imaging data. Specifically, 5 features were selected (4 brain metrics [caudate volume, left PCC thickness, PCC/precuneus fALFF, and small-worldness] and 1 psychological factor [emotional attention] with higher weights in the modeling) to control overfitting of the regression model. The correlation between the predicted and actual NRS changes is significant (r = 0.628, P < 0.001; RMSE = 0.980).
3.8. The specificity of the correlation results and machine learning models
We finally tested whether our correlations and models were specific for acupuncture improvement. First, we performed correlation analyses in the SG. For psychological variables correlating with clinical improvement in the AG, we found none of them were significant in the SG (Fig. S4, available at https://links.lww.com/PAIN/B778). In addition, we also performed correlation analyses in the sham-acupuncture arm on each brain metric correlating with the improvement of primary clinical outcome (ie, NRS) in the AG, and none of them were significant or marginally significant (Fig. S5, available at https://links.lww.com/PAIN/B778).
The classifier accuracy using combining features identical to the AG was considered nonsignificant in the SG to classify clinically improved and nonclinically improved patients (accuracy = 68.00%, P = 0.886, AUC = 64.04%, Figs. 6J–L). Moreover, the predicted and actual NRS changes correlation in the SVR predictive model with 5 brain and psychological features was not significant in the SG (rpredicted − actual = 0.140, P = 0.486; RMSE = 1.224) and was significantly lower than the AG (r = 0.628 in the AG vs r = 0.140 in the SG, z = 2.021, P = 0.022). Therefore, the correlations and machine learning models were specific to acupuncture.
4. Discussion
This is an integral study designed to identify the psychological and neurological factors collected before acupuncture treatment that would determine acupuncture efficacy in chronic pain. First, we examined the effects of 4-week acupuncture treatment in patients with KOA. The clinical statistics revealed that patients receiving acupuncture treatment showed larger NRS changes compared with patients in the sham-acupuncture arm, indicating that acupuncture successfully induced analgesia that could not be explained by the placebo effect. Second, NEO-FFI extraversion, TMMS emotional attention, and SETS positive expectation were correlated with the magnitude of clinical improvements, and SETS positive expectation was different between clinically improved and nonclinically improved patients in the AG. Third, the identified neurological properties encompassed striatal volumes, striatal fALFF, PCC cortical thickness, PCC fALFF, and small-worldness of the DMN and striatal regions. Specifically, functional metrics predisposing patients to acupuncture response changed as a consequence of acupuncture treatment while structural metrics remained stable. Machine learning applied to questionnaire and brain features could jointly predict acupuncture response. Our study provides insight into the psychobiological mechanism underlying the acupuncture analgesia effect, and suggests the potential of questionnaire and brain markers in recognizing individuals sensitive to acupuncture intervention.
The primary clinical finding that acupuncture can relieve pain intensity (NRS) in patients with KOA is consistent with our previous multicenter, randomized, sham-controlled trial (480 patients)71 and further confirms the acupuncture effect on KOA chronic pain.77 Although not significantly different in WOMAC pain, we did observe greater pain relief in AG compared with SG (1.85 vs 1.60, Table 1). Besides, no significant difference in WOMAC function or WOMAC stiffness score between AG and SG is also in line with our previous clinical trial.71 Given the consistent clinical effects of acupuncture, the present study mainly focused the psychological and neurological relevance.
Associating psychological factors with acupuncture effect in patients with KOA departs from the literature of chronic pain. Outcome expectation refers to the expected consequence that follows an intervention and is one of the most investigated factors. A pooled analysis of 4 acupuncture RCTs in 864 patients with migraine, headache, chronic low back pain, and KOA have shown that patients with higher positive expectations had more likely to experience clinical improvements than patients with lower expectations.44 As expected, positive expectation was correlated with the magnitude of WOMAC function improvement as well as able to differentiate clinical improvement and nonclinical improvement in our patients. In addition, acupuncture response was also driven by extraversion. Personality characteristics are widely examined in chronic pain and related psychological health fields,13,62,74,75,79 and higher extraversion is often associated with better health outcome.62,75 Conversely, we found that patients with less extroverted personality had higher WOMAC function improvement. Extraverts usually lead fast-paced lives and are involved in many activities, which may have more negative influences on their knee pain or function than other types of chronic pain. Moreover, patients paying more attention to their emotions had better pain sensory relief. Emotional attention, clarity of feelings, and emotional regulation constitute emotional intelligence. Previous studies have indicated that clarity of feelings and emotional regulation play critical role in facilitating positive adjustment to chronic pain,15 our findings further emphasized the role of emotional attention in managing pain sensory. Broadly, these evidences support the importance of emotional intelligence in development and management of chronic pain. These findings are vital because questionnaires are easy to acquire and may provide indispensable information to predict intervention effects in chronic pain.
The sensorial function of the striatum (usually consisting of caudate, putamen, and pallidum) relates to pain modulation.7 Cortical regions modulate nociception partly by projecting to the basal ganglia including the striatum, which also receives nociceptive inputs from the spinal cord via the globus pallidus.10 Striatal nuclei are the most densely populated regions for opioid receptors in the brain, and a positron emission spectrometry study has found that patients who reported higher levels of arthritic pain had greater opioid receptors in the caudate.12 Recently, a randomized, placebo-controlled, crossover study suggests that caudate nucleus of the striatum is a key structure underlying the pain-modulating effects of neuropeptide oxytocin in patients with chronic low back pain.9 Our results on striatal volume, spontaneous activity, and network properties relating to acupuncture effect are complementary to the aforementioned striatal pain modulation literature.
Default mode network is one of the most important networks in the human brain, mainly including PCC/precuneus, medial prefrontal cortex (mPFC), and angular gyrus,19,67 and has been identified as a biomarker for several chronic pain conditions.40 Previous studies have already demonstrated that the integrity of the DMN is disrupted in patients with chronic back pain,6 indicating the important role of DMN in chronic pain. Moreover, the functional connectivity strength within DMN regions (mPFC and angular gyrus) and between DMN and striatum regions (mPFC-putamen and mPFC-caudate) have been found correlating with acupuncture efficacy in chronic low back pain.69 The present finding extends these previous studies by showing that the overall topology (ie, small-worldness) of the network combining DMN and striatum regions contributed to the acupuncture effect prediction in KOA.
The specificity of the results in the AG was also tested in the present study. First, the significant correlations in the AG were not significant or near to significant in the SG. Second, the machine learning models in the AG could not successfully predict the clinical improvement in the SG. Therefore, we may infer that the results are specific to acupuncture and do not generalize to the sham-acupuncture arm. That is, different psychological and neurological mechanisms may underlie acupuncture and sham-acupuncture. For instance, sham-acupuncture treatment may reduce clinical pain via the affective pain pathway69 other than DMN and striatum.
Some limitations of the present work create future research opportunities. First, because acupuncture improvement-related structural and functional metrics were mainly centered at DMN and striatum, the present graph-based network analysis was focused on the network of DMN and striatal regions. Beyond DMN, chronic pain also involves sensorimotor network, salience network, reward system, and antinociceptive system,16,31,37,38,49,50,57 where different networks may engage in different pain stages.37 In the future, it is needed to fully investigate the relation between chronic pain and within- and cross-network functional connectivity of the DMN longitudinally. Second, during the hard time of the coronavirus pandemic, this study had difficulties in recruiting patients; further studies with larger sample size may properly calibrate error rates and increase the robustness of the prediction models.51,54 Third, social factors are also a part of the biopsychosocial model of chronic pain.72 The present study investigated the biological and psychological factors of this model; the follow-up studies integrating social factors, such as social economic status and social support, may contribute to the accuracy of the efficacy prediction. Fourth, this study only completed parallel group trial; further N-of-1 trials, which repeatedly conduct multiple treatments in the same patient and also include same treatment multiple times,61 are essential for estimating variances within and between individuals.59–61,78 Fifth, although we mainly used MCII for interpreting findings, the dichotomization of improvements decreases statistical power and efficiency on the group level and should be treated cautiously in the future.58,78 Finally, some of the predictors did not pass multiple comparisons correction, but we chose to present all significant correlations and select potential features based on uncorrected p74 to ensure the performances of the predicting models.
In conclusion, our results demonstrate the psychological, brain functional, and structural determinants of the acupuncture response and suggest that these properties may offer opportunities to save medical resources and personalize clinical practices that lead to better outcomes for patients. The combination of psychological and neurological factors accompanying with artificial intelligence methods may open up a new avenue for predicting prognosis before treatment and making treatment decisions.43
Conflict of interest statement
The authors have no conflicts of interest to declare.
Appendix A. Supplemental digital content
Supplemental digital content associated with this article can be found online at https://links.lww.com/PAIN/B778.
Acknowledgments
This study was supported by the Beijing Municipal Science and Technology Commission (grant D171100003217003) and Beijing Municipal Administration of Hospitals (grant XMLX201607). The funders had no role in study design, data collection, data analyses, results interpretation, or manuscript preparation. The present neuroimaging analysis plan was not preregistered in an independent, institutional registry.
Author contribution: Study concept and design: C.-Z. Liu and X. Wang. Acquisition, analysis, or interpretation of data: X. Wang, J.-L. Li, X.-Y. Wei, G.-X. Shi, N. Zhang, J.-F. Tu, C.-Q. Yan, Y.-N. Zhang, Y.-Y. Hong, J.-W. Yang, L.-Q. Wang, and C.-Z. Liu. Drafting of the manuscript: X. Wang. Critical revision of the manuscript for important intellectual content: C.-Z. Liu. Obtaining of funding: C.-Z. Liu.
Data availability: The data that support the present findings are available from the corresponding author.
References
[1]. Abraham A, Pedregosa F, Eickenberg M, Gervais P, Mueller A, Kossaifi J, Gramfort A, Thirion B, Varoquaux G. Machine learning for neuroimaging with scikit-learn. Front Neuroinform 2014;8:14.
[2]. Achard S, Bullmore E. Efficiency and cost of economical brain functional networks. PLoS Comput Biol 2007;3:e17.
[3]. Ashburner J. SPM: a history. Neuroimage 2012;62:791–800.
[4]. Avants BB, Epstein CL, Grossman M, Gee JC. Symmetric diffeomorphic image registration with cross-correlation: evaluating automated labeling of elderly and neurodegenerative brain. Med Image Anal 2008;12:26–41.
[5]. Baischer W. Acupuncture in migraine: long-term outcome and predicting factors. Headache 1995;35:472–4.
[6]. Baliki MN, Geha PY, Apkarian AV, Chialvo DR. Beyond feeling: chronic pain hurts the brain, disrupting the default-mode network dynamics. J Neurosci 2008;28:1398–403.
[7]. Barceló AC, Filippini B, Pazo JH. The striatum and pain modulation. Cell Mol Neurobiol 2012;32:1–12.
[8]. Bellamy N, Buchanan WW, Goldsmith CH, Campbell J, Stitt LW. Validation study of WOMAC: a health status instrument for measuring clinically important patient relevant outcomes to antirheumatic drug therapy in patients with osteoarthritis of the hip or knee. J Rheumatol 1988;15:1833–40.
[9]. Boll S, Ueltzhoeffer K, Roth C, Bertsch K, Desch S, Nees F, Grinevich V, Herpertz SC. Pain-modulating effects of oxytocin in patients with chronic low back pain. Neuropharmacology 2020;171:108105.
[10]. Braz JM, Nassar MA, Wood JN, Basbaum AI. Parallel “pain” pathways arise from subpopulations of primary afferent nociceptor. Neuron 2005;47:787–93.
[11]. Brown KW, Ryan RM. The benefits of being present: mindfulness and its role in psychological well-being. J Personal Soc Psychol 2003;84:822–48.
[12]. Brown CA, Matthews J, Fairclough M, McMahon A, Barnett E, Al-Kaysi A, El-Deredy W, Jones AKP. Striatal opioid receptor availability is related to acute and chronic pain perception in arthritis: does opioid adaptation increase resilience to chronic pain? PAIN 2015;156:2267–75.
[13]. Bucourt E, Martaillé V, Mulleman D, Goupille P, Joncker-Vannier I, Huttenberger B, Reveillere C, Courtois R. Comparison of the big five personality traits in fibromyalgia and other rheumatic diseases. Joint Bone Spine 2017;84:203–7.
[14]. Chester R, Jerosch-Herold C, Lewis J, Shepstone L. Psychological factors are associated with the outcome of physiotherapy for people with shoulder pain: a multicentre longitudinal cohort study. Br J Sports Med 2018;52:269–75.
[15]. Condon SE, Parmelee PA, Smith DM. Examining emotional intelligence in older adults with chronic pain: a factor analysis approach. Aging Ment Health 2021;25:213–8.
[16]. Cottam WJ, Iwabuchi SJ, Drabek MM, Reckziegel D, Auer DP. Altered connectivity of the right anterior insula drives the pain connectome changes in chronic knee osteoarthritis. PAIN 2018;159:929–38.
[17]. Cox RW. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 1996;29:162–73.
[18]. da Costa BR, Reichenbach S, Keller N, Nartey L, Wandel S, Jüni P, Trelle S. Effectiveness of non-steroidal anti-inflammatory drugs for the treatment of pain in knee and hip osteoarthritis: a network meta-analysis. Lancet 2017;390:e21–33.
[19]. Dosenbach NUF, Nardos B, Cohen AL, Fair DA, Power JD, Church JA, Nelson SM, Wig GS, Vogel AC, Lessov-Schlaggar CN, Barnes KA, Dubis JW, Feczko E, Coalson RS, Pruett JR Jr., Barch DM, Petersen SE, Schlaggar BL. Prediction of individual brain maturity using fMRI. Science 2010;329:1358–61.
[20]. Esteban O, Markiewicz CJ, Blair RW, Moodie CA, Isik AI, Erramuzpe A, Kent JD, Goncalves M, DuPre E, Snyder M, Oya H, Ghosh SS, Wright J, Durnez J, Poldrack RA, Gorgolewski KJ. fMRIPrep: a robust preprocessing pipeline for functional MRI. Nat Methods 2019;16:111–6.
[21]. Esteban O, Ciric R, Finc K, Blair RW, Markiewicz CJ, Moodie CA, Kent JD, Goncalves M, DuPre E, Gomez DEP, Ye Z, Salo T, Valabregue R, Amlien IK, Liem F, Jacoby N, Stojić H, Cieslak M, Urchs S, Halchenko YO, Ghosh SS, De La Vega A, Yarkoni T, Wright J, Thompson WH, Poldrack RA, Gorgolewski KJ. Analysis of task-based functional MRI data preprocessed with fMRIPrep. Nat Protoc 2020;15:2186–202.
[22]. Esteban O, Markiewicz CJ, Burns C, Goncalves M, Jarecka D, Ziegler E, Berleant S, Ellis DG, Pinsard B, Madison C, Waskom M, Notter MP, Clark D, Manhães-Savio A, Clark D, Jordan K, Dayan M, Halchenko YO, Loney F, Ghosh S. nipy/nipype: 1.7.1. Zenodo 2022. DOI: 10.5281/zenodo.6415183.
[23]. Fan AY, Zhou K, Gu S, Ming Li Y. Acupuncture is effective for chronic knee pain: a reanalysis of the Australian acupuncture trial. Altern Ther Health Med 2016;22:32–6.
[24]. Fischl B. FreeSurfer. Neuroimage 2012;62:774–81.
[25]. Fonov VS, Evans AC, McKinstry RC, Almli CR, Collins DL. Unbiased nonlinear average age-appropriate brain templates from birth to adulthood. NeuroImage 2009;47:S102.
[26]. GBD 2019 Diseases and Injuries Collaborators. Global burden of 369 diseases and injuries in 204 countries and territories, 1990-2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet 2020;396:1204–22.
[27]. Gorgolewski K, Burns CD, Madison C, Clark D, Halchenko YO, Waskom ML, Ghosh SS. Nipype: a flexible, lightweight and extensible neuroimaging data processing framework in python. Front Neuroinform 2011;5:13.
[28]. Gorgolewski KJ, Auer T, Calhoun VD, Craddock RC, Das S, Duff EP, Flandin G, Ghosh SS, Glatard T, Halchenko YO, Handwerker DA, Hanke M, Keator D, Li X, Michael Z, Maumet C, Nichols BN, Nichols TE, Pellman J, Poline JB, Rokem A, Schaefer G, Sochat V, Triplett W, Turner JA, Varoquaux G, Poldrack RA. The brain imaging data structure, a format for organizing and describing outputs of neuroimaging experiments. Sci Data 2016;3:160044.
[29]. Grafton KV, Foster NE, Wright CC. Test-retest reliability of the Short-Form McGill Pain Questionnaire: assessment of intraclass correlation coefficients and limits of agreement in patients with osteoarthritis. Clin J Pain 2005;21:73–82.
[30]. Greve DN, Fischl B. Accurate and robust brain image alignment using boundary-based registration. Neuroimage 2009;48:63–72.
[31]. Hemington KS, Rogachov A, Cheng JC, Bosma RL, Kim JA, Osborne NR, Inman RD, Davis KD. Patients with chronic pain exhibit a complex relationship triad between pain, resilience, and within- and cross-network functional connectivity of the default mode network. PAIN 2018;159:1621–30.
[32]. Hinman RS, McCrory P, Pirotta M, Relf I, Forbes A, Crossley KM, Williamson E, Kyriakides M, Novy K, Metcalf BR, Harris A, Reddy P, Conaghan PG, Bennell KL. Acupuncture for chronic knee pain: a randomized clinical trial. JAMA 2014;312:1313–22.
[33]. Hochberg MC, Altman RD, Brandt KD, Clark BM, Dieppe PA, Griffin MR, Moskowitz RW, Schnitzer TJ. Guidelines for the medical management of osteoarthritis. Part I. Osteoarthritis of the hip. American College of Rheumatology. Arthritis Rheum 1995;38:1535–40.
[34]. Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, Smith SM. FSL. NeuroImage 2012;62:782–90.
[35]. Kellgren JH, Lawrence JS. Radiological assessment of osteo-arthrosis. Ann Rheum Dis 1957;16:494–502.
[36]. Klein A, Ghosh SS, Bao FS, Giard J, Häme Y, Stavsky E, Lee N, Rossa B, Reuter M, Chaibub Neto E, Keshavan A. Mindboggling morphometry of human brains. PLoS Comput Biol 2017;13:e1005350.
[37]. Kucyi A, Davis KD. The dynamic pain connectome. Trends Neurosci 2015;38:86–95.
[38]. Kucyi A, Moayedi M, Weissman-Fogel I, Goldberg MB, Freeman BV, Tenenbaum HC, Davis KD. Enhanced medial prefrontal-default mode network functional connectivity in chronic pain and its association with pain rumination. J Neurosci 2014;34:3969–75.
[39]. Lanczos C. Evaluation of noisy data. J Soc Ind Appl Math Ser B Numer Anal 1964;1:76–85.
[40]. Letzen JE, Robinson ME. Negative mood influences default mode network functional connectivity in patients with chronic low back pain: implications for functional neuroimaging biomarkers. PAIN 2017;158:48–57.
[41]. Li J, Chen J, Qin Q, Zhao D, Dong B, Ren Q, Yu D, Bi P, Sun Y. Chronic pain and its association with obesity among older adults in China. Arch Gerontol Geriatr 2018;76:12–8.
[42]. Li JL, Liu CZ, Zhang N, Yan CQ, Tu JF, Wang LQ, Qi YS, Liu JH, Wang X. Neurological and psychological mechanisms of the specific and nonspecific effects of acupuncture on knee osteoarthritis: study protocol for a randomized, controlled, crossover trial. Trials 2020;21:989.
[43]. Li F, Sun H, Biswal BB, Sweeney JA, Gong Q. Artificial intelligence applications in psychoradiology. Psychoradiology 2021;1:94–107.
[44]. Linde K, Witt CM, Streng A, Weidenhammer W, Wagenpfeil S, Brinkhaus B, Willich SN, Melchart D. The impact of patient expectations on outcomes in four randomized controlled trials of acupuncture in patients with chronic pain. PAIN 2007;128:264–71.
[45]. Liu B, Xu H, Ma R, Mo Q, Yan S, Liu Z. Effect of blinding with a new pragmatic placebo needle: a randomized controlled crossover study. Medicine (Baltimore) 2014;93:e200.
[46]. Liu Z, Liu Y, Xu H, He L, Chen Y, Fu L, Li N, Lu Y, Su T, Sun J, Wang J, Yue Z, Zhang W, Zhao J, Zhou Z, Wu J, Zhou K, Ai Y, Zhou J, Pang R, Wang Y, Qin Z, Yan S, Li H, Luo L, Liu B. Effect of electroacupuncture on urinary leakage among women with stress urinary incontinence: a randomized clinical trial. JAMA 2017;317:2493–501.
[47]. Melzack R. The short-form McGill pain questionnaire. PAIN 1987;30:191–7.
[48]. Mohamed Zaki LR, Hairi NN. A systematic review of the prevalence and measurement of chronic pain in asian adults. Pain Manag Nurs 2015;16:440–52.
[49]. Niu X, Zhang M, Liu Z, Bai L, Sun C, Wang S, Wang X, Chen Z, Chen H, Tian J. Interaction of acupuncture treatment and manipulation laterality modulated by the default mode network. Mol Pain 2017;13:174480691668368.
[50]. Pfannmöller J, Lotze M. Review on biomarkers in the resting-state networks of chronic pain patients. Brain Cogn 2019;131:4–9.
[51]. Poldrack RA, Huckins G, Varoquaux G. Establishment of best practices for evidence for prediction: a review. JAMA Psychiatry 2020;77:534–40.
[52]. Qiu K, Jing M, Sun R, Yang J, Liu X, He Z, Yin S, Lan Y, Cheng S, Gao F, Liang F, Zeng F. The status of the quality control in acupuncture-neuroimaging studies. Evid Based Compl Altern Med 2016;2016:3685785.
[53]. Raja SN, Carr DB, Cohen M, Finnerup NB, Flor H, Gibson S, Keefe FJ, Mogil JS, Ringkamp M, Sluka KA, Song XJ, Stevens B, Sullivan MD, Tutelman PR, Ushida T, Vader K. The revised International Association for the Study of Pain definition of pain: concepts, challenges, and compromises. PAIN 2020;161:1976–82.
[54]. Riley RD, Ensor J, Snell KIE, Harrell FE Jr., Martin GP, Reitsma JB, Moons KGM, Collins G, van Smeden M. Calculating the sample size required for developing a clinical prediction model. BMJ 2020;368:m441.
[55]. Rubinov M, Sporns O. Complex network measures of brain connectivity: uses and interpretations. Neuroimage 2010;52:1059–69.
[56]. Salovey P, Mayer JD, Goldman SL, Turvey C, Palfai TP. Emotional attention, clarity, and repair: exploring emotional intelligence using the Trait Meta-Mood Scale. Emotion, disclosure, & health. Washington, DC: American Psychological Association, 1995. p. 125–54.
[57]. Seixas D, Palace J, Tracey I. Chronic pain disrupts the reward circuitry in multiple sclerosis. Eur J Neurosci 2016;44:1928–34.
[58]. Senn S, Julious S. Measurement in clinical trials: a neglected issue for statisticians? Stat Med 2009;28:3189–209.
[59]. Senn S. Individual response to treatment: is it a valid assumption? BMJ 2004;329:966–8.
[60]. Senn S. Mastering variation: variance components and personalised medicine. Stat Med 2016;35:966–77.
[61]. Senn S. Statistical pitfalls of personalized medicine. Nature 2018;563:619–21.
[62]. Seto A, Han X, Price LL, Harvey WF, Bannuru RR, Wang C. The role of personality in patients with fibromyalgia. Clin Rheumatol 2019;38:149–57.
[63]. Stam CJ. Characterization of anatomical and functional connectivity in the brain: a complex networks perspective. Int J Psychophysiol 2010;77:186–94.
[64]. Sullivan MD, Howe CQ. Opioid therapy for chronic pain in the United States: promises and perils. PAIN 2013;154(suppl 1):S94–100.
[65]. Sun N, Wang LQ, Shao JK, Zhang N, Zhou P, Fang SN, Chen W, Yang JW, Liu CZ. An expert consensus to standardize acupuncture treatment for knee osteoarthritis. Acupunct Med 2020;38:327–34.
[66]. Tétreault P, Mansour A, Vachon-Presseau E, Schnitzer TJ, Apkarian AV, Baliki MN. Brain connectivity predicts placebo response across chronic pain clinical trials. PLoS Biol 2016;14:e1002570.
[67]. Thomas Yeo BT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollinshead M, Roffman JL, Smoller JW, Zöllei L, Polimeni JR, Fischl B, Liu H, Buckner RL. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol 2011;106:1125–65.
[68]. Tian Y, Margulies DS, Breakspear M, Zalesky A. Topographic organization of the human subcortex unveiled with functional connectivity gradients. Nat Neurosci 2020;23:1421–32.
[69]. Tu Y, Ortiz A, Gollub RL, Cao J, Gerber J, Lang C, Park J, Wilson G, Shen W, Chan ST, Wasan AD, Edwards RR, Napadow V, Kaptchuk TJ, Rosen B, Kong J. Multivariate resting-state functional connectivity predicts responses to real and sham acupuncture treatment in chronic low back pain. NeuroImage Clin 2019;23:101885.
[70]. Tu Y, Zeng F, Lan L, Li Z, Maleki N, Liu B, Chen J, Wang C, Park J, Lang C, Yujie G, Liu M, Fu Z, Zhang Z, Liang F, Kong J. An fMRI-based neural marker for migraine without aura. Neurology 2020;94:e741–51.
[71]. Tu JF, Yang JW, Shi GX, Yu ZS, Li JL, Lin LL, Du YZ, Yu XG, Hu H, Liu ZS, Jia CS, Wang LQ, Zhao JJ, Wang J, Wang T, Wang Y, Wang TQ, Zhang N, Zou X, Wang Y, Shao JK, Liu CZ. Efficacy of intensive acupuncture versus sham acupuncture in knee osteoarthritis: a randomized controlled trial. Arthritis Rheumatol 2021;73:448–58.
[72]. Turk DC, Okifuji A. Psychological factors in chronic pain: evolution and revolution. J Consult Clin Psychol 2002;70:678–90.
[73]. Tustison NJ, Avants BB, Cook PA, Zheng Y, Egan A, Yushkevich PA, Gee JC. N4ITK: improved N3 bias correction. IEEE Trans Med Imaging 2010;29:1310–20.
[74]. Vachon-Presseau E, Berger SE, Abdullah TB, Huang L, Cecchi GA, Griffith JW, Schnitzer TJ, Apkarian AV. Brain and psychological determinants of placebo pill response in chronic pain patients. Nat Commun 2018;9:3397.
[75]. Vendrig AA, Derksen JJL, de Mey HR. MMPI-2 Personality Psychopathology Five (PSY-5) and prediction of treatment outcome for patients with chronic back pain. J Personal Assess 2000;74:423–38.
[76]. Vickers AJ, Linde K. Acupuncture for chronic pain. JAMA 2014;311:955–6.
[77]. Vickers AJ, Vertosick EA, Lewith G, MacPherson H, Foster NE, Sherman KJ, Irnich D, Witt CM, Linde K. Acupuncture for chronic pain: update of an individual patient data meta-analysis. J Pain 2018;19:455–74.
[78]. Vigotsky AD, Tiwari SR, Griffith JW, Apkarian AV. What is the numerical nature of pain relief? Front Pain Res (Lausanne) 2021;2:756680.
[79]. Wang X, Zhen Z, Xu S, Li J, Song Y, Liu J. Behavioral and neural correlates of social network size: the unique and common contributions of face recognition and extraversion. J Personal 2022;90:294–305.
[80]. Watts DJ, Strogatz SH. Collective dynamics of 'small-world' networks. Nature 1998;393:440–2.
[81]. Witt CM, Jena S, Brinkhaus B, Liecker B, Wegscheider K, Willich SN. Acupuncture in patients with osteoarthritis of the knee or hip: a randomized, controlled trial with an additional nonrandomized arm. Arthritis Rheum 2006;54:3485–93.
[82]. Witt CM, Vertosick EA, Foster NE, Lewith G, Linde K, MacPherson H, Sherman KJ, Vickers AJ. The effect of patient characteristics on acupuncture treatment outcomes: an individual patient data meta-analysis of 20,827 chronic pain patients in randomized controlled trials. Clin J Pain 2019;35:428–34.
[83]. Yan CG, Wang XD, Lu B. DPABISurf: data processing & analysis for brain imaging on surface. Sci Bull 2021;66:2453–5.
[84]. Yang X-J, Liu L, Xu Z-L, Zhang Y-J, Liu D-P, Fishers M, Zhang L, Sun J-B, Liu P, Zeng X, Wang L-P, Qin W. Baseline brain gray matter volume as a predictor of acupuncture outcome in treating migraine. Front Neurol 2020;11:111.
[85]. Yin T, Sun G, Tian Z, Liu M, Gao Y, Dong M, Wu F, Li Z, Liang F, Zeng F, Lan L. The spontaneous activity pattern of the middle occipital gyrus predicts the clinical efficacy of acupuncture treatment for migraine without aura. Front Neurol 2020;11:588207.
[86]. Yong RJ, Mullins PM, Bhattacharyya N. Prevalence of chronic pain among adults in the United States. PAIN 2022;163:e328–32.
[87]. Younger J, Gandhi V, Hubbard E, Mackey S. Development of the Stanford Expectations of Treatment Scale (SETS): a tool for measuring patient outcome expectancy in clinical trials. Clin Trials 2012;9:767–76.
[88]. Zhou M, Wang H, Zeng X, Yin P, Zhu J, Chen W, Li X, Wang L, Wang L, Liu Y, Liu J, Zhang M, Qi J, Yu S, Afshin A, Gakidou E, Glenn S, Krish VS, Miller-Petrie MK, Mountjoy-Venning WC, Mullany EC, Redford SB, Liu H, Naghavi M, Hay SI, Wang L, Murray CJL, Liang X. Mortality, morbidity, and risk factors in China and its provinces, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet 2019;394:1145–58.
[89]. Zou QH, Zhu CZ, Yang Y, Zuo XN, Long XY, Cao QJ, Wang YF, Zang YF. An improved approach to detection of amplitude of low-frequency fluctuation (ALFF) for resting-state fMRI: fractional ALFF. J Neurosci Methods 2008;172:137–41.
[90]. Zuo XN, Di Martino A, Kelly C, Shehzad ZE, Gee DG, Klein DF, Castellanos FX, Biswal BB, Milham MP. The oscillating brain: complex and reliable. Neuroimage 2010;49:1432–45.
