Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Jun 1.
Published in final edited form as: Clin Tradit Med Pharmacol. 2025 Apr 14;6(2):200215. doi: 10.1016/j.ctmp.2025.200215

Baseline predictors of responders to auricular point acupressure in chronic low back pain

Nada Lukkahatai a,#,*, Wanqi Chen b,#, Jennifer Kawi c, Hulin Wu b, Claudia M Campbell d, Johannes Thrul e, Xinran Huang b, Paul Christo d, Constance M Johnson c
PMCID: PMC12095896  NIHMSID: NIHMS2081500  PMID: 40417635

Abstract

Background:

Chronic low back pain (cLBP) is a major cause of disability, with varied patient responses to treatments. Auricular point acupressure (APA) has shown potential as a non-pharmacological intervention, but individual responses may differ significantly.

Objective:

This study aimed to determine the predictability of baseline characteristics, including functional disability, symptom severity, and treatment expectancy, on clinically significant responses to APA in reducing pain and improving function.

Methods:

A secondary analysis was performed using data from a randomized controlled trial with 263 cLBP patients. Participants were randomly assigned to targeted APA (T-APA), non-targeted APA (NT-APA), or to a control group. APA responders were defined as those with at least a 1.5-point reduction in pain intensity or a 2.5-point improvement in the Roland-Morris Disability Questionnaire (RMDQ). Predictors of response were assessed using logistic regression and machine learning models, including the Random Forest and Support Vector Machine (SVM).

Results:

Baseline pain, physical function, sleep disturbance, and treatment expectancy were key predictors. The Random Forest model had the highest accuracy for T-APA; however, logistic regression performed best in NT-APA. SVM was most accurate in the control group, with predictive accuracy varying by group (AUC 60.9%–80%). The Least Absolute Shrinkage and Selection Operator (LASSO) method was found to be overly aggressive, often eliminating important variables.

Conclusion:

This study highlights the variability in APA treatment responses for cLBP. While predictive models provide useful insights, further research with larger datasets is needed to improve prediction accuracy and generalizability, enhancing personalized treatment approaches for cLBP.

Keywords: Auricular point acupressure, Chronic low back pain, Physical function, Pain, Auriculotherapy, APA responders

1. Introduction

Chronic low back pain (cLBP) is one of the leading causes of disability worldwide (Alfalogy et al., 2023; Hallberg et al., 2023; Perrot et al., 2022). Despite the advancement of both pharmacological and non-pharmacological interventions, the outcomes for cLBP have not shown improvement, and the rates of disability continue to escalate (Chowdhury et al., 2021; Patricio et al., 2021). The substantial healthcare expenditures associated with cLBP intensify these public health concerns. Americans allocated an estimated $134.5 billion towards managing low back and neck pain (Dieleman et al., 2020). Managing cLBP often involves a multi-faceted approach to reduce pain and improve function. Nonpharmacological options for cLBP include exercise, yoga, tai chi, psychological therapies, multidisciplinary rehabilitation, acupuncture, and spinal manipulation—all providing small improvements in pain and function (Soares Fonseca et al., 2023; Zhou et al., 2024). Patients with cLBP respond very differently to these treatments in clinical practice (Obbarius et al., 2020; Scholz et al., 2024).

Auricular point acupressure (APA), a traditional Chinese medicine technique, has emerged as a potential intervention for cLBP management. APA involves applying pressure to specific points on the ear believed to correspond to bodily organs and systems. Studies among individuals with cLBP demonstrated that APA significantly reduced pain intensity and improved physical function and quality of life (Yeh et al., 2022; Yeh et al., 2022; Kawi et al., 2025). Conversely, other studies have reported more modest or non-significant effects. For instance, studies found that the impact of APA on pain relief was similar to a placebo in a randomized controlled trial, suggesting that psychological and personal factors may play a substantial role in perceived pain reduction (Mohamed Mohamed et al., 2020; Foster et al., 2021; Zieger et al., 2022). These conflicting results suggest the need for further research to clarify the conditions under which APA may be most effective and to identify patient characteristics that predict positive outcomes.

Clinically, differentiating responders from non-responders to non-pharmacological pain interventions such as APA typically involves assessing self-reported pain intensity using scales such as the Numerical Rating Scale (NRS) or the Visual Analog Scale (VAS), with a reduction of at least 1.5 points considered minimally, clinically significant (Salaffi et al., 2004). Additionally, for the change in physical function (measured by the Roland-Morris Disability Questionnaire [RMDQ]), a decrease of 2–3 points on the RMDQ is commonly regarded as a clinically meaningful improvement in functional disability among individuals with cLBP (Stratford and Riddle, 2016). Based on the RMDQ score change, some responders have typically shown substantial reductions in pain intensity and disability, signaling a positive response, while non-responders may have exhibited insufficient improvement despite treatment.

Factors including participants’ baseline characteristics, symptom severity, functional disabilities, and expectancy of treatment may influence the effectiveness of APA on pain. Baseline characteristics, such as age, gender, and comorbidities, can significantly impact how patients respond to non-pharmacological treatments for chronic pain. For example, factors such as age and underlying health conditions can influence pain sensitivity and the ability to engage in certain therapies, affecting treatment outcomes (Haque et al., 2024). Symptom severity and functional disability are also critical predictors, as patients with higher initial levels of pain and disability may require more intensive interventions and may respond differently to intervention than those with milder symptoms (Wilson et al., 2023; Hassett et al., 2023). Functional limitations may reduce adherence to treatment and limit the effectiveness of nonpharmacological interventions (Sowden et al., 2018).

Furthermore, the expectancy of treatment plays a vital role in outcomes for non-pharmacological interventions. Research suggests that positive treatment expectancy can enhance the perceived effectiveness of interventions such as APA or acupuncture, as patients who believe in the treatment’s potential are more likely to report improvements due to placebo effects and to increase engagement with the therapy (Zieger et al., 2022; Trutnovsky et al., 2018). By considering these factors—baseline characteristics, symptom severity, functional disability, and treatment expectancy—clinicians and researchers can better tailor APA interventions to meet the individual needs of patients with chronic low back pain, thereby improving pain management outcomes and facilitating more personalized treatment approaches.

The present study is a secondary analysis of data from our previously published randomized controlled trial of APA for cLBP (Kawi et al., 2025; Yeh et al., 2020). The aim was to examine whether patient characteristics at baseline (including functional disability, baseline symptoms severity, and expectancy of treatment) predict clinically significant treatment response of APA on pain and disability.

2. Methods

2.1. Data source

We performed a secondary analysis of the data from a prospective randomized controlled study of auricular point acupressure to manage chronic low back pain in older adults. The parent study was a three-arm, randomized controlled trial (RCT). Participants were randomly assigned to targeted auricular point acupressure (T-APA), non-targeted auricular points acupressure (NT-APA), or control groups. The study was approved by the Institutional Review Board of the Johns Hopkins School of Medicine (IRB00175409) and registered in ClinicalTrials.gov (Trial ID: NCT03589703; URL: https://clinicaltrials.gov/ct2/show/record/NCT03589703). The study protocol details were published (Yeh et al., 2020).

2.2. Participants

Participants diagnosed with cLBP from the parent study who completed the 4-week intervention program and the one-month follow-up visit with no missing baseline pain and disability scores were included in the present analysis.

2.3. Intervention

After informed consent and baseline assessments, participants were randomly assigned to one of three groups, with randomization conducted in blocks of three or six by a statistician using a random number generator before the study started. Targeted APA (T-APA) group: Participants in this group received weekly sessions where auricular seeds were placed on ear points, specifically targeting the low back areas. These sessions were conducted by trained interventionists. Following the seed placement, participants were instructed to perform self-administered stimulation of the seeds at home for four weeks. Non-Targeted APA (NT-APA) group: Participants received weekly sessions with auricular seed placement and self-administered stimulation at home, similar to the T-APA group. However, the seeds were placed on non-specific ear points unrelated to the low back areas. Control group: Participants received educational materials and comparable time and attention.

2.4. Measures

For the present secondary data analysis, APA responders were categorized based on two primary outcomes: pain intensity and physical function. The pain intensity responders were participants who achieved at least a 1.5-point reduction in worst pain at 1-month after the 4-week intervention from baseline. The physical function responders were participants with at least 2.5 points reduction in the RMDQ at 1-month post-intervention from baseline. These thresholds reflect established minimal clinically important differences for cLBP. A 1.5-point reduction on the NRS is considered the smallest change perceived as beneficial by patients (Salaffi et al., 2004), and a 2–3 point decrease in RMDQ has been widely accepted as clinically meaningful for functional improvement (Stratford and Riddle, 2016). Participants’ baseline characteristics, symptom profiles, and treatment expectancies were included in the analysis as potential predictors of APA treatment response. These variables included:

Participants’ characteristics, including age, biological sex, body mass index, race, educational level, employment status, smoking history, and opioid use, were measured using a demographics questionnaire.

Baseline symptoms were measured using multiple self-reported questionnaires. Pain severity (average, least, and now) was measured on the Likert scale from 0 (no pain) to 10 (worse pain). Neuropathic pain was measured by the painDETECT questionnaire, a validated self-report tool designed to screen for neuropathic components in individuals with back pain. This instrument has shown high internal consistency (Cronbach’s α = 0.86) and effectively distinguishes between neuropathic and nociceptive pain profiles (Freynhagen et al., 2016). The other symptoms (i.e., pain interference, fatigue, sleep disturbance, anxiety, and depression) were measured by the Patient-Reported Outcomes Measurement Information System (PROMIS-29) Adult Profile (Hays et al., 2018). The PROMIS-29 is a self-reported instrument developed by the National Institutes of Health that measures seven health domains (physical function, fatigue, pain interference, depression, anxiety, social participation, and sleep disturbance) using a 1–5 Likert scale and a 0–10 pain scale. The raw scores for each subscale were converted into T-scores. The PROMIS-29 demonstrated strong psychometric properties in this study (Cronbach’s α = 0.84 – 0.93)

Physical disability, including Physical limitation, was measured by RMDQ, a 24-item scale that assesses functional limitations associated with chronic low back pain. Participants responded yes or no to statements about their physical functioning, with higher scores indicating greater disability. The RMDQ demonstrated excellent reliability in this study (Cronbach’s α = 0.91) (Roland and Fairbank, 2000).

Pain catastrophizing was measured by the Pain Catastrophizing Scale (PCS), which assesses how individuals experience exaggerated negative thoughts and feelings in response to pain. This 13-item instrument captures three dimensions—rumination, magnification, and helplessness—and demonstrates strong internal consistency (Cronbach’s α = 0.94) (Hayashi et al., 2022).

Comorbidity was assessed by the Charlson Comorbidity Index, a validated tool that assigns weighted scores to various chronic conditions based on their potential influence on mortality and overall health. In this study, the index showed high reliability (Cronbach’s α = 0.88) (Charlson et al., 1987).

Participants’ treatment expectations and positive outlook were measured by the treatment expectations and positive outlook domains of the Healing Encounters and Attitudes Lists [HEAL] short form. Each domain includes six items scored on a 5-point Likert scale ranging from strongly disagree (1) to strongly agree (5). These measures assess participants’ beliefs about treatment effectiveness and recovery and shows strong internal consistency (Cronbach’s α ≥ 0.88) (Gerger et al., 2019).

Fear/avoidance of physical activity was measured by the Physical Activity subscale in the Fear-Avoidance Beliefs Questionnaire (FABQ-PA). This subscale includes four items, each rated on a 7-point Likert scale from 0 (completely disagree) to 6 (completely agree), with higher total scores reflecting greater avoidance behaviors. The FABQ-PA showed solid reliability in this sample (Cronbach’s α = 0.85) (Panhale et al., 2019).

2.5. Statistical analysis

After including participants who met the inclusion criteria for this analysis, ANOVA or Chi-square tests were used to compare the differences in baseline characteristics among treatment groups. The latent class analysis was conducted for pain phenotyping, wherein individuals were categorized into distinct classes based on their baseline pain intensity, emotional distress, and demographic characteristics. Subsequently, three latent classes were identified and treated as categorical variables in subsequent analyses. To assess the association between baseline factors and APA treatment response, a standard logistic regression model was fitted, incorporating all continuous and categorical variables.

Additionally, we applied various feature selection methods to identify significant variables, including backward selection, stepwise selection, best subset selection, and the Least Absolute Shrinkage and Selection Operator (LASSO) method. The criteria for feature selection included Akaike information criterion (AIC) for backward and stepwise selection, and Bayesian information criterion (BIC) for best subset selection. Cross-validation was utilized to determine the lambda value for LASSO regression, minimizing the mean squared error (MSE). We performed association analysis separately for the T-APA and NT-APA treatment groups, with three categorical variables (smoking status, employment status, and latent class) excluded before analysis for RMDQ responders in the NT-APA group due to limited sample size.

For predictive modeling, the dataset was divided into training and testing sets at a ratio of 7:3. Both statistical logistic models and machine learning models were considered for predictions. The following were employed to predict treatment responses: Logistic regression with backward, stepwise selection, best subset selection, and LASSO, as well as Random Forest and Support Vector Machine (SVM). We applied 10-fold cross-validation to optimize the hyperparameters of the machine learning models. This method involves partitioning the data into 10 subsets, iteratively training the model on 9 subsets, and validating it on the remaining one. This process helps maximize data use and provides a more stable estimate of model performance, which is especially valuable in small sample settings. The number of trees for the Random Forest was determined based on out-of-bag (OOB) error; the kernel for SVM was selected based on model performance. The following R packages were utilized for analysis: “caret ”, “pROC ”, “ggplot2 ”, “MASS ”, “leaps ”, “glmnet ”, “random Forest ”, and “e1071 ”.

No adjustments for multiple comparisons were applied in this analysis. As a secondary, exploratory study, the statistical tests were conducted to identify potential predictors of treatment response rather than to confirm predefined hypotheses. Therefore, findings should be interpreted as hypothesis-generating and require validation in future confirmatory studies.

3. Results

We included 263 participants who completed the 4-week APA intervention and the 1-month post-intervention visits without missing data. Participants’ demographic and baseline characteristics for each treatment group are summarized in Table 1. The table showed a high degree of balance in these factors across all three groups.

Table 1:

Baseline Demographic and Clinical Characteristics of Participants by Treatment Group

Variables Mean (SD) or Counts (Percentage) P value
T-APA n = 88 NT-APA n = 88 Control n = 87
Age 69.68 (7.08) 68.68 (7.18) 71.02 (7.16) 0.0958
BMI 31.35 (8.54) 31.08 (7.77) 30.64 (6.91) 0.8613
Sex  Female 56 (63.64%) 56 (63.64%) 58 (66.67%) 0.8896
 Male 32 (36.36%) 32 (36.36%) 29 (33.33%)
Employment status  Working 9 (10.23%) 15 (17.05%) 15 (17.24%) 0.3298
 Not Working 79 (89.77%) 73 (82.95%) 72 (82.76%)
Race  White 38 (43.18%) 27 (30.68%) 30 (34.38%) 0.209
 Non-White 50 (56.82%) 61 (69.32%) 57 (65.52)
Smoking status  Current 15 (17.05%) 21 (23.86%) 11 (12.64%) 0.1969
 Quit 39 (44.32%) 41 (46.59%) 37 (42.53%)
 Never 34 (38.64%) 26 (29.55%) 39 (44.83%)
Opioid Use  Yes, or Not Sure 47 (53.4%) 44 (50%) 37 (42.53%) 0.3384
 No 41 (46.59) 44 (50%) 50 (57.47%)
Education level  Bachelor’s degree and higher 32 (36.36%) 24 (27.27%) 30 (34.48%) 0.3985
 Pre-bachelor’s education 56 (63.64%) 64 (72.73%) 57 (65.52%)
Pain severity Average Pain 6.02 (1.75) 6.09 (1.90) 6.22 (1.72) 0.7645
Least Pain 4.24 (2.70) 4.48 (2.73) 4.17 (2.77) 0.7397
Pain Right Now 4.62 (2.65) 4.84 (2.69) 4.44 (2.80) 0.6155
Other Symptomsa (t score) Pain Interference 61.68 (7.70) 63.77 (6.54) 62.58 (5.94) 0.1233
Fatigue 52.74 (10.61) 53.85 (9.39) 51.86 (9.51) 0.4086
Sleep Disturbance 53.77 (8.79) 54.31 (8.31) 54.09 (7.99) 0.9112
Anxiety 51.10 (8.99) 53.29 (10.89) 51.51 (9.80) 0.2992
Depression 50.57 (9.70) 52.90 (8.93) 50.32 (9.03) 0.1256
Physical Function 38.01 (6.83) 36.60 (6.19) 38.24 (7.03) 0.2139
Neuropathic Pain b 15.93 (7.10) 18.32 (8.06) 16.84 (7.74) 0.1147
Pain Phenotyping (Latent Class)  Class 1: moderate pain and emotional distress 25 (28.41%) 29 (32.95%) 25 (28.74%) 0.6594
 Class 2: high pain and emotional distress 34 (38.64%) 39 (44.32%) 36 (41.38%)
 Class 3: low pain and emotional distress 29 (32.95%) 20 (22.73%) 26 (29.89%)
Physical Limitation c 12.36 (5.35) 12.82 (6.02) 11.95 (5.74) 0.6065
Pain Catastrophizing 15.60 (12.49) 19.28 (14.26) 15.54 (12.24) 0.1011
Fear of Physical Activities d 15.43 (8.27) 17.26 (8.45) 15.59 (8.40) 0.2969
Charlson Comorbidity 1.32 (2.87) 1.03 (2.18) 0.97 (1.57) 0.5525
Healing Encounters and Attitudes Lists Treatment Expectancy 21.57 (5.77) 22.57 (5.48) 21.83 (5.46) 0.4673
Positive Outlook 23.58 (6.05) 23.66 (5.21) 23.76 (6.23) 0.9796
Open in a new tab

Note:

a

PROMIS-29 score;

b

PainDETECT score;

c

RMDQ score;

d

FABQ-PA score

3.1. Association between the baseline characteristics and APA responses

We evaluated the association between the baseline factors and APA treatment responses using a standard logistic regression model. We used one-hot encoding to label the treatment groups, with all patients across the T-APA, NT-APA, and Control groups included in the logistic regression analysis. Baseline factors were incorporated into the model for both pain responder and physical function responder outcomes. The results, presented in Table 2, showed that three factors—Group treatment (NT-APA), sleep disturbance, and physical function—emerged as significantly associated with both APA pain response and physical function response. Notably, the type of APA treatment and the physical function displayed positive associations with pain intensity response and RMDQ response, respectively; sleep exhibited negative associations with both outcomes.

Table 2.

Association of Baseline Characteristics with Pain and Physical Function Responses to APA Treatment

Variables Pain Responder n = 190 Physical Function Responder n = 200
Estimate P value Estimate P value
Intercept −2.3841 0.0111e 0.1063 0.9048
Group (ref: Control) Group T-APA 1.2760 0.0071e 0.7889 0.0876
Group NT-APA 1.5041 0.0043e 1.6786 0.0009f
Age −0.1598 0.4770 −0.7817 0.0015e
BMI 0.2118 0.3172 0.0961 0.6426
Sex (ref: female) Male 0.0035 0.9935 −0.3090 0.4763
Employment status (ref: working) Not Working 0.8004 0.2111 0.6038 0.3228
Race (ref: white) Non-White 0.1203 0.7959 −0.2686 0.5421
Educational level (ref: no bachelor’s degree) Bachelor’s degree or higher −0.4198 0.3833 −0.6187 0.1940
Smoking Status (ref: Never)  Current −0.3005 0.6325 −1.6106 0.0131e
 Quit −0.0866 0.8527 −0.5761 0.1938
Opioid Use (ref: no) Yes, or Not Sure 0.4746 0.2923 −0.3934 0.3488
Pain intensity Worst Pain 0.8827 0.0037e −0.1087 0.7133
Average Pain −0.6558 0.0828 0.2048 0.5889
Least Pain 0.1676 0.5834 −0.1833 0.5521
Pain Right Now −0.1893 0.5496 −0.4089 0.1859
Other Symptoms a Pain Interference 0.6687 0.0423e 0.5313 0.1023
Fatigue 0.2197 0.4393 −0.1037 0.7016
Sleep Disturbance −0.5670 0.0128e −0.4839 0.0394e
Depression −1.6176 0.0001f −0.5920 0.0762
Anxiety 0.4197 0.2632 0.6862 0.0587
Physical Function 1.6010 <0.0001f 0.9566 0.0051e
Neuropathic Pain b 0.2488 0.3326 0.2083
Pain Phenotyping: Latent Class (ref: Class 3: low pain and emotional distress)  Class 1: moderate pain and emotional distress −0.5837 0.3729 −0.6850 0.3083
 Class 2: high pain and emotional distress 0.8966 0.2385 −0.9503 0.2049
Physical Limitation c 0.4538 0.1165 1.8959 <0.0001f
Pain Catastrophizing −0.1638 0.5754 −0.0591 0.8400
Fear of Physical Activities d −0.0662 0.7711 −0.4392 0.0510
Charlson Comorbidity 0.0891 0.6884 0.1869 0.3698
Treatment Expectancy 0.6541 0.0051e 0.1488 0.5115
Positive Outlook −0.7367 0.0038e 0.0566 0.8137
Open in a new tab

Note:

a

PROMIS-29 score;

b

PainDETECT score;

c

RMDQ score;

d

FABQ-PA score.

e

Significant (P<0.05);

f

Highly significant (P<0.0001) without multiple testing adjustments.

Further analysis indicated that the factors positively associated with pain responses included the baseline worst pain score, treatment expectancy, and baseline pain interference, whereas depression and positive outlook were negatively associated with pain responses. Conversely, for physical function responses, the factors that were positively associated included baseline physical limitation (RMDQ) score, while age and current smoking showed negative associations.

3.2. Prediction results

The prediction of patient responses to APA treatment demonstrates variability across individuals, underscoring the need for personalized treatment approaches. In the analysis of pain intensity responders, the prediction performance varied among the treatment groups. The key variables retained under each feature selection method across different treatment groups are provided in Appendix A, tables S1–S4, to support reproducibility and highlight model-specific priorities. We used 10-fold cross-validation to evaluate the predictive performance of various models. For the T-APA group, the Random Forest model yielded the best results (Table 3), with an area under the curve (AUC) of 60.9%, a false positive (FP) rate of 19.0%, and a false negative (FN) rate of 38.1%. In the NT-APA group, the best subset regression LR model performed optimally, with an AUC of 72.0%, an FP rate of 47.1%, and an FN rate of 11.8%. The control group also showed the best performance with the best subset regression LR model, which achieved an AUC of 61.5%, an FP rate of 31.6%, and an FN rate of 0.

Table 3.

Predictive Model Performance for Pain Intensity Response Across Treatment Groups

Testing False Positive Testing False Negative Testing AUC
T-APA group, sample size = 69, 48 in training set, 21 in testing set
Backward 0.143 0.286 0.577
Stepwise 0.143 0.429 0.441
Best Subset 0.190 0.381 0.518
LASSO (min) - - -
Random Forest 0.190 0.381 0.609a
SVM 0.286 0.19 0.527
NT-APA group, sample size = 58, 41 in training set, 17 in testing set
Backward 0.353 0.118 0.561
Stepwise 0.412 0.235 0.652
Best Subset 0.471 0.118 0.720b
LASSO (min) 0.588 0.118 0.652
Random Forest 0.294 0.235 0.439
SVM 0.588 0.059 0.636
Control group, sample size = 63, 44 in training set, 19 in testing set
Backward 0.211 0.211 0.487
Stepwise 0.316 0 0.5
Best Subset 0.316 0 0.615c
LASSO (min) - - -
Random Forest 0.316 0 0.590
SVM 0.316 0.053 0.538
Open in a new tab

Note:

a

The highest AUC in T-APA group;

b

The highest AUC in NT-APA group;

c

The highest AUC in Control group.

Similarly, predicting responders based on the physical function measured by RMDQ also revealed varying performance levels across models and treatment groups (Table 4). For the T-APA group, the Random Forest model again provided the highest predictive accuracy, with an AUC of 76.7%, an FP error rate of 13.6%, and an FN error rate of 9.1%. In the NT-APA group, Random Forest remained the best-performing model, with an AUC of 65.5%, an FP rate of 31.6%, and an FN rate of 10.5%. The control group, however, demonstrated superior performance with the SVM model, which achieved an AUC of 80.0%, an FP rate of 26.3%, and an FN rate of 5.3%.

Table 4.

Predictive Model Performance for Physical Function Response Across Treatment Groups

Testing False Positive Testing False Negative Testing AUC
T-APA group, sample size = 72, 50 in training set, 22 in testing set
Backward 0.091 0.318 0.608
Stepwise 0.091 0.273 0.733
Best Subset 0.136 0.273 0.675
LASSO (min) - - -
Random Forest 0.136 0.091 0.767a
SVM 0.182 0.318 0.617
NT-APA group, sample size = 63, 44 in training set, 19 in testing set
Backward 0.263 0.263 0.435
Stepwise 0.105 0.263 0.619
Best Subset 0.263 0.158 0.560
LASSO (min) - - -
Random Forest 0.316 0.105 0.655b
SVM 0.316 0.105 0.571
Control group, sample size = 65, 46 in training set, 19 in testing set
Backward 0.263 0.158 0.393
Stepwise 0.211 0.211 0.614
Best Subset 0.158 0.158 0.600
LASSO (min) 0.263 0 0.600
Random Forest 0.158 0.105 0.743
SVM 0.263 0.053 0.800c
Open in a new tab

Note:

a

The highest AUC in T-APA group;

b

The highest AUC in NT-APA group;

c

The highest AUC in Control group.

3.3. ROC curve analysis

To further illustrate model performance, Fig. 1 presents ROC curves for all the statistical and machine learning models we used to predict pain and physical function responders in different treatment groups. The curves show the trade-off between true positive rate (sensitivity) and false positive rate (1– specificity), supporting the AUC values reported earlier. In Fig. 1(F), the ROC curves for the LASSO and best subset methods overlap due to identical prediction rankings despite differences in the false positive and false negative rates.

Fig. 1.

Fig. 1.

Open in a new tab

ROC curves for pain and physical function response across treatment groups: (A) Pain responder in the T-APA group; (B) Pain responder in the NT-APA group; (C) Pain responder in the control group; (D) Physical function responder in the T-APA group; (E) Physical function responder in the NT-APA group; (F) Physical function responder in the control group.

4. Discussion

This secondary analysis of an RCT examined baseline predictors of clinically meaningful response to APA among individuals with cLBP. Consistent with the existing literature (Chen et al., 2023; Deyo et al., 2015; Nicol et al., 2023), emphasizing that individualized factors are crucial in determining nonpharmacological treatment efficacy. We found that baseline physical function, treatment expectancy, and sleep disturbance emerged as key predictors of response to APA. Participants with more severe functional impairment at baseline showed greater improvement, suggesting APA may benefit those with higher disability.

Treatment expectancy was positively associated with pain response, reinforcing the well-documented role of expectancy effects in complementary and integrative therapies (Efverman, 2020; Zheng et al., 2015). Patients who believed in the effectiveness of APA were more likely to report significant reductions in pain. In this study, we incorporated all treatment arms—including T-APA, NT-APA, and control—into the logistic regression models to examine the association between expectancy and treatment response. This approach allowed us to directly assess whether psychological factors such as treatment expectancy predicted clinical improvements, even in the control group, thereby accounting for potential placebo effects. This highlights the importance of addressing patients’ beliefs and attitudes about treatment during clinical encounters. Enhancing treatment credibility through psychoeducation and trust-building could be a low-cost way to optimize outcomes in APA and similar interventions.

Baseline sleep disturbance was negatively associated with treatment outcomes, reducing the likelihood of improvement in both pain and function. Poor sleep worsens pain sensitivity and emotional regulation, which may interfere with the body’s ability to benefit from APA (Wilson et al., 2023). These findings suggest that addressing sleep disturbance through parallel interventions or behavioral strategies may be necessary to realize the full benefits of APA. Furthermore, the analysis revealed other predictors of treatment response. Higher baseline pain interference and worst pain scores were associated with greater reductions in pain, likely due to more room for improvement. However, depressive symptoms and lower levels of positive outlook were linked to poorer responses, underscoring the impact of psychological well-being on the effectiveness of non-pharmacologic treatments. This is consistent with prior studies that demonstrate depression can attenuate treatment effects and reduce engagement with therapies (Hayashi et al., 2022; Gerger et al., 2019).

From a predictive analytics perspective, the present study identifies Random Forest and logistic regression with the best subset selection as effective tools for predicting treatment response. This represents a meaningful contribution to the growing body of research leveraging predictive analytics in pain management. However, the reported predictive accuracy—with AUC values ranging from 60.9% to 80%—is moderate compared to other studies using advanced machine learning techniques. Previous studies have demonstrated similar or superior predictive performance using methods such as SVM and neural networks (Niederer et al., 2024; Zheng et al., 2023). One factor contributing to these moderate results may be the relatively small and homogeneous sample size, which can reduce generalizability and increase the risk of overfitting, particularly in more complex algorithms. These findings suggest that predictive performance depends not only on model complexity but also on sample size, data variation, and feature structure across treatment groups.

The variability in model performance across treatment groups likely reflects differences in both intervention characteristics and data structure. Random Forest models, which excel at detecting nonlinear relationships, performed best in the T-APA group—where targeted stimulation may have introduced more nuanced patterns (Rajkomar et al., 2019; Agarwal et al., 2023). In contrast, logistic regression, a simpler and more interpretable method, achieved optimal performance in the NT-APA group, suggesting a more linear relationship between predictors and outcomes in this condition (Christodoulou et al., 2019). Interestingly, the SVM model performed best in the control group, possibly due to the homogeneity of this group and the SVM’s capacity to establish optimal decision boundaries in well-separated datasets (Li et al., 2023; O’Bryant et al., 2018; Hong et al., 2022; Llorian-Salvador et al., 2023; Liew et al., 2022). Feature selection and variable importance also played a significant role in model effectiveness. Predictors such as pain severity, sleep disturbance, and psychological distress likely contributed differently across treatment groups. Furthermore, data issues such as class imbalance, outliers, and distributional variability may have influenced model performance, reinforcing the need to match modeling strategies to the specific characteristics of the dataset and intervention (Nemati et al., 2018; O’Brien and Ishwaran, 2019; Cho et al., 2022).

To strengthen the robustness of our findings, we used several variable selection methods, including backward selection, stepwise regression, best subset selection, and LASSO. The key variables retained under each method are detailed in Appendix A, tables S1–S4, providing transparency and allowing readers to compare the relative importance of predictors across techniques. While LASSO is known for reducing model complexity by penalizing less predictive variables, its aggressive shrinkage can lead to excluding moderately important predictors. More recent studies have reinforced these concerns, showing that LASSO may perform suboptimally when the true model includes groups of correlated features or moderate-effect predictors (Alhamzawi and Ali, 2018; Mallick et al., 2021). As such, LASSO results should be interpreted with caution. Alternative regularization techniques—such as ridge regression, which handles multicollinearity by shrinking coefficients without exclusion, or elastic net, which combines the strengths of LASSO and ridge—may provide more stable and balanced solutions. Additionally, Bayesian variable selection approaches offer probabilistic frameworks for identifying relevant predictors, potentially improving model interpretability and robustness in future studies.

Despite the strengths of this analysis—such as the use of a well-characterized sample, diverse modeling strategies, and thorough feature selection—several limitations must be acknowledged. First, this is a secondary analysis of a randomized controlled trial, and while the parent study was rigorously conducted, the secondary nature of the analysis limits causal inference. All findings are observational and hypothesis-generating. Second, our sample included only participants who completed the 4-week APA intervention and 1-month follow-up assessments. This may introduce selection bias, as individuals who dropped out might differ systematically from those who completed the study. Third, the sample size—while adequate for initial exploration—remains relatively small for training and validating machine learning models. This limitation can lead to model instability, reduced external validity, and increased susceptibility to overfitting, especially when applying complex models like Random Forest or SVM. Additionally, we did not apply a correction for multiple comparisons due to the exploratory nature of this analysis. As a result, the risk of Type I error may be elevated, and findings should be interpreted with caution and validated in future confirmatory studies.

Another important limitation is the lack of adherence data for the APA self-stimulation protocol. Given that APA relies on consistent stimulation of ear seeds by participants outside of clinical sessions, variation in adherence could significantly influence outcomes. Without objective adherence measures, it’s difficult to fully interpret the effectiveness of the intervention. Additionally, we did not examine mechanistic variables, such as neurophysiological responses, biomarkers, or changes in brain activity, which could help explain how APA exerts its effects. Future studies should incorporate biological and behavioral mechanisms to strengthen causal inferences and deepen our understanding of APA’s therapeutic pathways.

Clinically, these findings offer valuable insights into how APA might be applied more effectively in real-world practice. By screening for baseline physical limitations, psychological symptoms, sleep quality, and treatment expectancy, clinicians may be able to identify patients most likely to respond to APA. For instance, individuals with poor physical function, high pain interference, and strong belief in the treatment’s efficacy appear to derive the greatest benefit, while those with significant sleep disturbance or depression may require adjunctive strategies to optimize outcomes. Integrating cognitive-behavioral therapy for insomnia or mood disorders could help address these barriers and enhance the overall effectiveness of APA. Furthermore, predictive models—particularly Random Forest—could eventually support the development of clinical decision-support tools that enable more personalized treatment planning for individuals with cLBP.

In summary, this study identifies key predictors of APA treatment response and supports the potential of integrating predictive modeling into the personalized management of chronic low back pain. While this study focused on APA, it is important to consider these findings within the broader context of non-pharmacological treatments. Existing approaches such as exercise therapy, acupuncture, and multidisciplinary rehabilitation have also demonstrated small-to-moderate effects on pain and physical function (Soares Fonseca et al., 2023; Zhou et al., 2024). However, few have incorporated predictive modeling to tailor treatment based on individual characteristics. Our study contributes to this evolving field by demonstrating how machine-learning approaches can enhance the personalization of APA and position it as one of several viable options in individualized cLBP care. These findings require cautious interpretation due to the small sample and lack of adherence data. Further validation in larger, more diverse populations is needed to confirm model reliability. Until such validation is completed, the current predictive models should be applied with care in clinical settings. Future studies should not only confirm these results but also explore mechanisms of action and optimize model performance to improve precision and real-world applicability.

5. Conclusion

In conclusion, this secondary analysis highlights key baseline factors—such as physical function, treatment expectancy, and sleep disturbance—that can predict clinically meaningful responses to auricular point acupressure in individuals with chronic low back pain. Integrating predictive modeling, particularly Random Forest and logistic regression offers promising avenues for advancing personalized pain management. However, the moderate predictive performance, limited sample size, and absence of adherence data underscore the need for caution in interpreting these findings and applying these models clinically. Future research should focus on validating these predictors in larger, more diverse populations, incorporating adherence and mechanistic measures, and refining model performance to support the development of clinically useful, personalized treatment strategies.

Supplementary Material

Table S1- S4

Acknowledgement

We dedicate this work to our esteemed Principal Investigator, Professor Dr. Chao Hsing Yeh’s memory with immense respect, gratitude, and a commitment to advancing the pursuit of knowledge in her honor. Her influence is evident in every aspect of this manuscript, which is a tribute to her remarkable legacy. We would also like to thank Mr. Martin Blair for his editing assistance with this manuscript and express our heartfelt gratitude to all the study participants and research staff for their invaluable contributions to this research. Their involvement made this study possible, and we are deeply thankful for their time and effort.

Funding

This study was supported by the National Institute on Aging under grant number R01AG056587.

Footnotes

Ethical approval

The study protocol was reviewed and approved by the Institutional Review Board at the Johns Hopkins School of Medicine (IRB00175409). All participants provided written informed consent prior to enrollment.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Credit authorship contribution statement

NL and WC contributed equally to this work. Both were involved in the conceptualization, methodology, data analysis, and manuscript drafting. NL also provided supervision and led the review and editing process. WC additionally led data curation and visualization. JK supported resource acquisition and participated in manuscript review and editing. HW provided statistical oversight and supervision. CC contributed methodological guidance and manuscript review. JT provided supervision and critical manuscript feedback. XH supported data management and analysis. PC offered clinical oversight and contributed to the manuscript review. CJ assisted with project administration, securing funding, supervision, and manuscript editing.

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.ctmp.2025.200215.

Data availability

The datasets generated and analyzed during the current study are not publicly available due to privacy or ethical restrictions but are available from the corresponding author on reasonable request. During the preparation of this work, the author(s) used ChatGPT (OpenAI) and Grammarly to help improve the clarity of the language and correct grammar. After using these tools, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

References

  1. Agarwal D, Berbis MA, Luna A, Lipari V, Ballester JB, de la Torre-Diez I, 2023. Automated medical diagnosis of alzheimer s disease using an efficient net convolutional neural network. J. Med. Syst 47, 57. doi: 10.1007/s10916-023-01941-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Alfalogy E, Mahfouz S, Elmedany S, Hariri N, Fallatah S, 2023. Chronic low back pain: prevalence, impact on quality of life, and predictors of future disability. Cureus. 15, e45760. doi: 10.7759/cureus.45760. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Alhamzawi R, Ali HTM, 2018. The Bayesian adaptive lasso regression. Math. Biosci 303, 75–82. doi: 10.1016/j.mbs.2018.06.004. [DOI] [PubMed] [Google Scholar]
  4. Charlson ME, Pompei P, Ales KL, MacKenzie CR, 1987. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J. Chronic. Dis 40, 373–383. doi: 10.1016/0021-9681(87)90171-8. [DOI] [PubMed] [Google Scholar]
  5. Chen JA, Anderson ML, Cherkin DC, Balderson BH, Cook AJ, Sherman KJ, Turner JA, 2023. Moderators and nonspecific predictors of treatment benefits in a randomized trial of mindfulness-based stress reduction vs cognitive-behavioral therapy vs usual care for chronic low back pain. J. Pain 24, 282–303. doi: 10.1016/j.jpain.2022.09.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Cho B, Geng E, Arvind V, Valliani AA, Tang JE, Schwartz J, Dominy C, Cho SK, Kim JS, 2022. Understanding artificial intelligence and predictive analytics: a clinically focused review of machine learning techniques. JBJS. Rev 10. doi: 10.2106/JBJS.RVW.21.00142. [DOI] [PubMed] [Google Scholar]
  7. Chowdhury AR, Graham PL, Schofield D, Cunich M, Nicholas M, 2021. Cost-effectiveness of multidisciplinary interventions for chronic low back pain: a narrative review. Clin. J. Pain 38, 197–207. doi: 10.1097/AJP.0000000000001009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Christodoulou E, Ma J, Collins GS, Steyerberg EW, Verbakel JY, Van Calster B, 2019. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. J. Clin. Epidemiol 110, 12–22. doi: 10.1016/j.jclinepi.2019.02.004. [DOI] [PubMed] [Google Scholar]
  9. Deyo RA, Dworkin SF, Amtmann D, Andersson G, Borenstein D, Carragee E, Carrino J, Chou R, Cook K, DeLitto A, et al. , 2015. Report of the NIH task force on research standards for chronic low back pain. Int. J. Ther. Massage BodyWork 8, 16–33. doi: 10.3822/ijtmb.v8i3.295. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Dieleman JL, Cao J, Chapin A, Chen C, Li Z, Liu A, Horst C, Kaldjian A, Matyasz T, Scott KW, et al. , 2020. US health care spending by payer and health condition, 1996–2016. JAMa 323, 863–884. doi: 10.1001/jama.2020.0734. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Efverman A, 2020. Treatment expectations seem to affect bowel health when using acupuncture during radiotherapy for cancer: secondary outcomes from a clinical randomized sham-controlled trial. Complement. Ther. Med 52, 102404. doi: 10.1016/j.ctim.2020.102404. [DOI] [PubMed] [Google Scholar]
  12. Foster NE, Vertosick EA, Lewith G, Linde K, MacPherson H, Sherman KJ, Witt CM, Vickers AJ, 2021. Acupuncture trialists, C. identifying patients with chronic pain who respond to acupuncture: results from an individual patient data meta-analysis. Acupunct. Med 39, 83–90. doi: 10.1177/0964528420920303. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Freynhagen R, Tolle TR, Gockel U, Baron R, 2016. The painDETECT project - far more than a screening tool on neuropathic pain. Curr. Med. Res. Opin 32, 1033–1057. doi: 10.1185/03007995.2016.1157460. [DOI] [PubMed] [Google Scholar]
  14. Gerger H, Buergler S, Sezer D, Grethler M, Gaab J, Locher C, 2019. The healing encounters and attitudes lists (HEAL): psychometric properties of a german version (HEAL-D) in comparison with the original HEAL. Front. Psychiatry 10, 897. doi: 10.3389/fpsyt.2019.00897. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Hallberg S, Rolfson O, Karppinen J, Schiottz-Christensen B, Stubhaug A, Rivano Fischer M, Gerdle B, Toresson Grip E, Gustavsson A, Robinson RL, et al. , 2023. Burden of disease and management of osteoarthritis and chronic low back pain: healthcare utilization and sick leave in Sweden, Norway, Finland and Denmark (BISCUITS): study design and patient characteristics of a real world data study. Scand. J. Pain 23, 126–138. doi: 10.1515/sjpain-2021-0212. [DOI] [PubMed] [Google Scholar]
  16. Haque R, Alam K, Gow J, Neville C, Keramat SA, 2024. Age and gender differences in the relationship between chronic pain and dementia among older australians. Value in Health doi: 10.1016/j.jval.2024.07.022. [DOI] [PubMed] [Google Scholar]
  17. Hassett AL, Williams DA, Harris RE, Harte SE, Kaplan CM, Schrepf A, Kratz AL, Brummett CM, Kidwell KM, Tsodikov A, et al. , 2023. An Interventional Response Phenotyping Study in Chronic Low Back Pain: protocol for a Mechanistic Randomized Controlled Trial. Pain Med. 24, S126–S138. doi: 10.1093/pm/pnad005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Hayashi K, Ikemoto T, Shiro Y, Arai YC, Marcuzzi A, Costa D, Wrigley PJ, 2022. A systematic review of the variation in pain catastrophizing scale reference scores based on language version and country in patients with chronic primary (non-specific) pain. Pain. Ther 11, 753–769. doi: 10.1007/s40122-022-00390-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Hays RD, Spritzer KL, Schalet BD, Cella D, 2018. PROMIS(R)-29 v2.0 profile physical and mental health summary scores. Qual. Life Res 27, 1885–1891. doi: 10.1007/s11136-018-1842-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Hong Y, Li Y, Ye M, Yan S, Yang W, Jiang C, 2022. Identifying an optimal machine learning model generated circulating biomarker to predict chronic postoperative pain in patients undergoing hepatectomy. Front. Surg 9, 1068321. doi: 10.3389/fsurg.2022.1068321. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Kawi J, Yeh CH, Lukkahatai N, Wu H, Morone NE, Glick R, Schlenk EA, Campbell C, Thrul J, Huang X, et al. , 2025. Auricular point acupressure for older adults with chronic low back pain: a randomized controlled trial. Pain medicine doi: 10.1093/pm/pnaf035. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Li Y, Gao J, Zheng X, Nie G, Qin J, Wang H, He T, Wheelock A, Li CX, Cheng L, et al. , 2023. Diagnostic Prediction of portal vein thrombosis in chronic cirrhosis patients using data-driven precision medicine model. Brief. Bioinformatics 25. doi: 10.1093/bib/bbad478. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Liew BXW, Kovacs FM, Rugamer D, Royuela A, 2022. Machine learning versus logistic regression for prognostic modelling in individuals with non-specific neck pain. Eur. Spine J 31, 2082–2091. doi: 10.1007/s00586-022-07188-w. [DOI] [PubMed] [Google Scholar]
  24. Llorian-Salvador O, Akhgar J, Pigorsch S, Borm K, Munch S, Bernhardt D, Rost B, Andrade-Navarro MA, Combs SE, Peeken JC, 2023. The importance of planning CT-based imaging features for machine learning-based prediction of pain response. Sci. Rep 13, 17427. doi: 10.1038/s41598-023-43768-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Mallick H, Alhamzawi R, Paul E, Svetnik V, 2021. The reciprocal Bayesian LASSO. Stat. Med 40, 4830–4849. doi: 10.1002/sim.9098. [DOI] [PubMed] [Google Scholar]
  26. Mohamed Mohamed WJ, Joseph L, Canby G, Paungmali A, Sitilertpisan P, Pirunsan U, 2020. Are patient expectations associated with treatment outcomes in individuals with chronic low back pain? A systematic review of randomised controlled trials. Int. J. Clin. Pract 74, e13680. doi: 10.1111/ijcp.13680. [DOI] [PubMed] [Google Scholar]
  27. Nemati S, Holder A, Razmi F, Stanley MD, Clifford GD, Buchman TG, 2018. An interpretable machine learning model for accurate prediction of sepsis in the ICU. Crit. Care Med 46, 547–553. doi: 10.1097/CCM.0000000000002936. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Nicol V, Verdaguer C, Daste C, Bisseriex H, Lapeyre E, Lefevre-Colau MM, Rannou F, Roren A, Facione J, Nguyen C, 2023. Chronic low back pain: a narrative review of recent international guidelines for diagnosis and conservative treatment. J. Clin. Med 12. doi: 10.3390/jcm12041685. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Niederer D, Schiller J, Groneberg DA, Behringer M, Wolfarth B, Gabrys L, 2024. Machine learning-based identification of determinants for rehabilitation success and future healthcare use prevention in patients with high-grade, chronic, nonspecific low back pain: an individual data 7-year follow-up analysis on 154,167 individuals. Pain. 165, 772–784. doi: 10.1097/j.pain.0000000000003087. [DOI] [PubMed] [Google Scholar]
  30. O’Brien R, Ishwaran H, 2019. A random forests quantile classifier for class imbalanced data. Pattern. Recognit 90, 232–249. doi: 10.1016/j.patcog.2019.01.036. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. O’Bryant SE, Zhang F, Johnson LA, Hall J, Edwards M, Grammas P, Oh E, Lyketsos CG, Rissman RA, 2018. A precision medicine model for targeted NSAID therapy in alzheimer’s disease. J. Alzheimers Dis 66, 97–104. doi: 10.3233/JAD-180619. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Obbarius A, Fischer F, Liegl G, Obbarius N, van Bebber J, Hofmann T, Rose M, 2020. A step towards a better understanding of pain phenotypes: latent class analysis in chronic pain patients receiving multimodal inpatient treatment. J. Pain. Res 13, 1023–1038. doi: 10.2147/JPR.S223092. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Panhale VP, Gurav RS, Shah P, Nayak N, 2019. Psychometric properties of the hindi version of the fear-avoidance beliefs questionnaire in patients with chronic non-specific low back pain. Spine (Phila Pa 1976) 44, E908–E913. doi: 10.1097/BRS.0000000000003019. [DOI] [PubMed] [Google Scholar]
  34. Patricio P, Roy JS, Rohel A, Gariepy C, Emond C, Hamel E, Masse-Alarie H, 2021. The effect of noninvasive brain stimulation to reduce nonspecific low back pain: a systematic review and meta-analysis. Clin. J. Pain 37, 475–485. doi: 10.1097/AJP.0000000000000934. [DOI] [PubMed] [Google Scholar]
  35. Perrot S, Doane MJ, Jaffe DH, Dragon E, Abraham L, Viktrup L, Bushmakin AG, Cappelleri JC, Conaghan PG, 2022. Burden of chronic low back pain: association with pain severity and prescription medication use in five large European countries. Pain Pract. 22, 359–371. doi: 10.1111/papr.13093. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Rajkomar A, Dean J, Kohane I, 2019. Machine Learning in Medicine. N. Engl. J. Med 380, 1347–1358. doi: 10.1056/NEJMra1814259. [DOI] [PubMed] [Google Scholar]
  37. Roland M, Fairbank J, 2000. The roland-morris disability questionnaire and the oswestry disability questionnaire. Spine (Phila Pa 1976) 25, 3115–3124. doi: 10.1097/00007632-200012150-00006. [DOI] [PubMed] [Google Scholar]
  38. Salaffi F, Stancati A, Silvestri CA, Ciapetti A, Grassi W, 2004. Minimal clinically important changes in chronic musculoskeletal pain intensity measured on a numerical rating scale. Eur. J. Pain 8, 283–291. doi: 10.1016/j.ejpain.2003.09.004. [DOI] [PubMed] [Google Scholar]
  39. Scholz C, Schmigalle P, Plessen CY, Liegl G, Vajkoczy P, Prasser F, Rose M, Obbarius A, 2024. The effect of self-management techniques on relevant outcomes in chronic low back pain: a systematic review and meta-analysis. Eur. J. Pain 28, 532–550. doi: 10.1002/ejp.2221. [DOI] [PubMed] [Google Scholar]
  40. Soares Fonseca L, Pereira Silva J, Bastos Souza M, Gabrich Moraes Campos M, de Oliveira Mascarenhas R, de Jesus Silva H, Souza Maximo Pereira L, Xavier Oliveira M, Cunha Oliveira V, 2023. Effectiveness of pharmacological and non-pharmacological therapy on pain intensity and disability in older people with chronic nonspecific low back pain: a systematic review with meta-analysis. Eur. Spine J 32, 3245–3271. doi: 10.1007/s00586-023-07857-4. [DOI] [PubMed] [Google Scholar]
  41. Sowden G, Hill JC, Morso L, Louw Q, Foster NE, 2018. Advancing practice for back pain through stratified care (STarT Back). Braz. J. Phys. Ther 22, 255–264. doi: 10.1016/j.bjpt.2018.06.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Stratford PW, Riddle DL, 2016. A roland morris disability questionnaire target value to distinguish between functional and dysfunctional states in people with low back pain. PhysiOther Can. 68, 29–35. doi: 10.3138/ptc.2014-85. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Trutnovsky G, Schneeweiss M, Pregartner G, Reif P, Lang U, Ulrich D, 2018. Prenatal acupuncture: women’s expectations and satisfaction and influence on labor. Birth 45, 210–216. doi: 10.1111/birt.12336. [DOI] [PubMed] [Google Scholar]
  44. Wilson AT, Riley JL 3rd; Bishop MD, Beneciuk JM, Cruz-Almeida Y, Markut K, Redd C, LeBlond N, Pham PH, Shirey D, et al. , 2023. Pain phenotyping and investigation of outcomes in physical therapy: an exploratory study in patients with low back pain. PLoS. One 18, e0281517. doi: 10.1371/journal.pone.0281517. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Yeh CH, Li C, Glick R, Schlenk EA, Albers K, Suen LK, Lukkahatai N, Salen N, Pandiri S, Ma W, et al. , 2020. A prospective randomized controlled study of auricular point acupressure to manage chronic low back pain in older adults: study protocol. Trials 21, 99. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Yeh CH, Kawi J, Ni A, Christo P, 2022. Evaluating auricular point acupressure for chronic low back pain self-management using technology: a feasibility study. Pain. Manage Nurs 23, 301–310. doi: 10.1016/j.pmn.2021.11.007. [DOI] [PubMed] [Google Scholar]
  47. Yeh CH, Kawi J, Grant L, Huang X, Wu H, Hardwicke RL, Christo PJ, 2022. Self-guided smartphone application to manage chronic musculoskeletal pain: a randomized, controlled pilot trial. Int. J. Environ. Res. Public Health 19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Zheng H, Huang W, Li J, Zheng Q, Li Y, Chang X, Sun G, Liang F, 2015. Association of pre- and post-treatment expectations with improvements after acupuncture in patients with migraine. Acupunct. Med 33, 121–128. doi: 10.1136/acupmed-2014-010679. [DOI] [PubMed] [Google Scholar]
  49. Zheng X, Reneman MF, Preuper RHS, Otten E, Lamoth CJ, 2023. Relationship between physical activity and central sensitization in chronic low back pain: insights from machine learning. Comput. Methods Programs Biomed 232, 107432. doi: 10.1016/j.cmpb.2023.107432. [DOI] [PubMed] [Google Scholar]
  50. Zhou LP, Zhang RJ, Shang J, Kang L, Zhang ZG, Zhang B, Wang JQ, Jia CY, Zhao CH, Zhang HQ, et al. , 2024. Comparative effectiveness of nonpharmacological interventions in reducing psychological symptoms among patients with chronic low back pain. Int. J. Surg 110, 478–489. doi: 10.1097/JS9.0000000000000798. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Zieger A, Kern A, Barth J, Witt CM, 2022. Do patients’ pre-treatment expectations about acupuncture effectiveness predict treatment outcome in patients with chronic low back pain? A secondary analysis of data from a randomised controlled clinical trial. PLoS. One 17, e0268646. doi: 10.1371/journal.pone.0268646. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Table S1- S4
NIHMS2081500-supplement-Table_S1-_S4.docx (38.2KB, docx)

Data Availability Statement

The datasets generated and analyzed during the current study are not publicly available due to privacy or ethical restrictions but are available from the corresponding author on reasonable request. During the preparation of this work, the author(s) used ChatGPT (OpenAI) and Grammarly to help improve the clarity of the language and correct grammar. After using these tools, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

RESOURCES