Article Text


The influence of baseline characteristics on response to a laser acupuncture intervention: an exploratory analysis
  1. Gregory Glazov
  1. University of Western Australia, Perth, Australia
  1. Correspondence to Dr Gregory Glazov, School of Primary, Aboriginal and Rural Health Care, University of Western Australia, 328 Stirling Hwy Claremont WA 6010, Australia; glazog01{at}


Objectives In clinical practice it is known that subjects vary in their response to acupuncture, but there is little data on what predicts the outcome. The aim of this study was to identify such predictors.

Method A secondary analysis was performed on data from 100 participants in a trial of laser and sham laser acupuncture for chronic non-specific low back pain. Multiple regression analysis was used to identify which baseline characteristics predicted pain change in the immediate, short and intermediate term. An analysis of covariance was performed based on these results to re-examine the primary result of the trial.

Results Strong predictors of poor response were receipt of disability support pension, headache, the regular use of analgesics or previous failed back surgery. Higher pain scores or exacerbation of pain at baseline predicted a greater proportionate pain relief after the intervention. Adjusted analysis suggested a clinically important effect of laser compared to sham (p<0.05), at short term follow-up only.

Conclusion The findings of this study suggest which characteristics of patients with chronic low back pain are more likely to respond to laser acupuncture treatment, but require replication in other studies. The findings may not apply in other acupuncture interventions and treatment of different conditions. They may also be used to set selection criteria for future studies, and to aid interpreting the effect of baseline imbalances on trial results.

Statistics from


Studies on back pain have found evidence that demographic features, characteristics of the pain condition and previous and current treatment may influence chronicity of disability or response to treatment.1 2 It has been frequently noticed by acupuncturists that subjects tend to vary in their response to acupuncture. In traditional Chinese medicine, it was considered that a better result was achieved if acupuncture point selection was guided by the patient's traditional Chinese medicine ‘correspondences’. In more recent times, Felix Mann3 identified the concept of the ‘strong responder’.

There have been a few studies of needle acupuncture which have explored the effects of patient variables on outcome, known as response predictors. A small study4 examining treatment of various conditions suggested that neither age nor gender is related to rate of recovery, but that patients with more severe initial conditions, particularly bodily pain, tended to make more rapid improvements. A large multi-centre study5 of needle acupuncture for chronic low back pain (CLBP), analysed the effect of baseline scores of chronicity, pain intensity and depression on treatment outcomes. This trial did not find a clinically important difference in functional improvement after treatment in relation to either baseline intensity or chronicity. The presence of depression at baseline did not affect the improvement in physical health but did predict a clinically significant improvement in mental health after the acupuncture intervention. Another large pragmatic trial6 comparing needle acupuncture with usual care in patients with non-specific CLBP showed few predictors except a greater improvement in symptoms and function if these were more severe at baseline, and in patients who did not use narcotics.

In the current study7 100 subjects with non-specific CLBP were recruited, mainly from notices in community newspapers, into a double blind trial of laser acupuncture, to determine the effectiveness of a 10 mW laser compared to sham. A co-intervention consisting of education and encouragement to exercise was also provided to both groups. The trial found no difference in pain outcomes between the intervention (laser) and the control (sham laser) group. There was a significant pain reduction in the combined groups at the end of treatment (40%) and at follow-up (30%) at 6 weeks and 6 months.

It was decided to do a posthoc subgroup analysis to explore which baseline factors were associated with a better or worse outcome, and whether bias from an imbalance of these factors between the treatment groups may have affected the result of the trial.


Before participants were randomised, comprehensive baseline data were recorded during a structured interview (see box 1). The responses to items were already dichotomous (yes/no), or categorical in which case they were collapsed onto binary form for the present analysis. Continuous variables such as age, body mass index (BMI) and scale scores were categorised. All data were analysed using SPSS V.15.0.

Box 1 Baseline data recorded in the study


  • Gender: male/female

  • Smoking status: yes/no

  • BMI: normal/overweight or obese

  • Alcohol intake: <2 (females) or <4 (males) standard drinks per day/more

  • Employment status: employed/on a pension or unemployed

  • Pension status: on disability support pension/other pension or no pension

Pain characteristics

  • VAS score (average pain in previous week): ≤median (5.9)/>median (5.9)

  • Duration of pain: ≤2 years/>2 years

  • Disability at baseline (Oswestry disability index): nil or minimal (0–20%)/moderate or severe (21–60%)

  • Radiation of pain outside of low back: absent/present

  • Exacerbation pain (worsening of pre-existing pain at baseline): yes/no

  • Presence of degenerative changes on imaging: yes/no or not done or not available

Previous and current treatment

  • Low back surgery done over 2 years previously: yes/no

  • Specialist interventions for LBP (eg, facet block injections, epidurals, etc) done over 3 months previously: yes/no

  • Any type of acupuncture done over 3 months previously: yes/no

  • Current use of analgesics: regular/none or as necessary

  • Current use of NSAIDS etc: regular/none or as necessary

  • Current use of any non-prescription herbs, vitamins or supplements: yes/no

  • Current use of antidepressant medication: yes/no

  • Current use of other physical therapies: yes/no

Presence of other functional symptoms at baseline

  • Headaches often present: yes/no

  • Neck pain often present: yes/no

  • 'Irritable bowel syndrome': yes/no

  • Trochanteric bursitis present: yes/no

Depression Anxiety Stress (DASS-21) subscales

  • Depression: normal to mild (0–13)/moderate to severe (14–28+)

  • Anxiety: normal to mild (0–9)/moderate to severe (10–20+)

  • Stress: normal to mild (0–18)/moderate to severe (19–34+)

First, the baseline comparability of characteristics between the intervention groups was examined. Because of the heterogeneity of the chronic pain population studied and relatively small sample size, it was expected that by chance some baseline characteristics would not be evenly split between the treatment arms.

As recommended in the Revised Consort Satement,8 9 we did not test for significant differences, however as our study was exploratory, and there was little previous data on response predictors to laser acupuncture, we considered it important to list the full range of variables for baseline comparison.

The next step was to determine which baseline characteristics appeared to influence the amount of pain reduction after treatment. As the primary result showed no clear difference between the treatment arms, the overall sample was used in this assessment. The primary outcome of interest in the original trial was the mean pain on a visual analogue scale (0–10) at the last session of treatment (immediate term follow-up). Pain levels at short term (6 weeks after last treatment) and intermediate term follow-up (6 months after last treatment) were also examined as secondary endpoints. We used the percentage pain change from baseline at all three time points in testing for differences in the 90 participants who completed the protocol of a minimum 5–10 treatment sessions. We excluded from the analysis 10 participants who dropped out before completing five sessions, consisting of four who did not start treatment, and six who dropped out before five sessions were attended. These were not further followed up. They were evenly distributed between intervention groups.

A previous consensus10 on minimal important change for low back pain (LBP), concluded that a 30% reduction may represent clinically meaningful improvement. However in our analysis we decided to use a lower figure, to include more possible predictors. Hence a difference of percentage pain change ≥20 between subgroups in response the intervention at any time point was used to select factors to be included in the multiple regression equation.

During follow-up, there was some further loss from the 90 participants who completed treatment, as shown in table 1.

Table 1

Numbers of subjects remaining in the trial by stage of follow-up

As there was uncertainty on the effect of these missing values at follow-up, we imputed missing data in two ways: (a) pain score at last completed assessment and (b) pain score at baseline.

Multiple regression was used to identify which baseline factors were best able to predict improvement in pain after the intervention. A separate model was used for the immediate, 6 week and 6 month outcomes. The independent variables to be included in the final model were:

  1. Factors which were associated with pain change at any time point.

  2. Factors with greatest baseline imbalance, including any not determined in 1.

  3. Intervention group (laser/sham).

For each follow-up time point, all listed variables were included in a stepwise backward selection to determine the model.

Finally, an adjusted analysis was performed to determine whether a significant difference between the intervention/control group in pain outcome may have existed at any time point after controlling for relevant covariates.


Baseline comparison

The distribution of baseline variables by treatment group are presented in tables 2 and 3. The means for continuous variables were comparable except for a higher disability score in the laser group. Inspection of distributions in the categorical comparison table demonstrated that there were higher rates of recipients of disability support pension (15%), radiating pain (20%) and headache (15%), as well as greater use of regular analgesics (23%) and over-the-counter (OTC) medications (15%) in the laser group. Conversely in the sham group more subjects were currently using other forms of physical therapy (24%) and had previously received acupuncture (22%).

Table 2

Continuous variables in groups at baseline

Table 3

Dichotomous and categorical variables in groups at baseline

Subgroup analysis

Mean percentage pain change was calculated at each of the three end points for each of the baseline variables listed in box 1 using both imputation methods for missing values.

All variables for which there was a difference in percentage pain change ≥20 between categories were listed in table 4, labelled with the category with the worse pain outcome (less response). Using these criteria the remaining variables from box 1 did not appear to predict pain outcome following treatment.

Table 4

Variables predicting Percentage Pain Change (PPC) ≥20 between categories at any time point after completion of treatment

For each time point, the variables which appeared to affect pain outcome (table 4), as well as four other variables which showed the greatest imbalance between groups (previous acupuncture, use of OTC medications, current use of other physical treatments, radiation of pain present) and the treatment group variable were entered into the regression model. Those that were found to be statistically significant in predicting pain change are listed in table 5.

Table 5

Predictors of PPC at three time points after intervention

At immediate follow-up, subjects with baseline headaches had worse pain outcome. There also was a trend towards less improvement in subjects on disability support pension and those with previous back surgery. At short term (6 weeks) follow-up subjects on disability support pension had less improvement in pain, while those with higher pain scores at baseline had a greater proportionate pain reduction. There was a trend to worse outcome in headache sufferers and regular users of analgesics. At 6 months follow-up, those on regular analgesics did worse and subjects who described an exacerbation of pain at baseline had a greater proportionate pain reduction.

When different methods of imputation were used for missing values, the results were similar. They were identical at immediate follow-up (as there were no missing values at this time). At short term follow-up for headaches the statistical relationship was stronger when value for pain at baseline was imputed. At intermediate term follow-up this applied to the group with exacerbation at baseline when value for pain at last end point was imputed.

Adjusted analysis

A one-way analysis of covariance between groups was conducted to compare the effectiveness of laser against sham in producing pain reduction at the three endpoints in this trial. The independent variable was the type of intervention (laser/sham), and the dependent variable was percentage pain change at the respective end point. Four baseline characteristics which were selected as covariates had been shown to predict pain outcome from table 5, and also were unevenly distributed between the treatment arms of the trial. These were baseline presence of headaches, pension status, analgesic use and occurrence of previous back surgery.

After adjustment for these covariates (see table 6), a statistically significant difference in percentage pain change between groups was only found at 6 weeks follow-up. At best this represented a 23% difference in pain reduction effect in favour of laser, approaching a medium effect size according guidelines by Cohen.10 The imputation methods used would have included scenarios of (a) good and (b) poor outcomes in dropouts with missing values.

Table 6

Results comparing laser and sham laser using analysis of covariance at short term follow up (6 weeks after end of treatment)


This study showed that certain baseline subcategories of patients with non-specific CLBP responded differently in pain relief after laser acupuncture treatment. Receiving a disability support pension or being prone to habitual headaches resulted in less pain reduction in the immediate and short term, and regular use of analgesics predicted less pain reduction in the short and intermediate term. Previous back surgery or specialist pain relief blocks to the back predicted a poorer response to laser acupuncture treatment, although the statistical relationship for this was weaker.

There are a number of explanations postulated for these findings. The findings are rational as these factors are likely to indicate more severe pathology group or structural change. There may be secondary gain for patients on some type of pensions to remain ill. A similar effect may occur with chronic back pain resulting from work injury or motor vehicle accidents although these cases were excluded from this study. Previous failed treatment including anaesthetic blocks and surgery damaging nerves and causing other changes such as scarring or instability may make back pain subsequently unresponsive to acupuncture. Patients may be dependent on certain analgesics such as codeine, and large frequent doses of analgesics may create less scope to experience benefit or may overwhelm the appreciation of a smaller effect from acupuncture. Our trial excluded patients taking strong opioid analgesics but allowed use of compound tablets containing up to 30 mg codeine or tramadol. This result is consistent with the study by Sherman et al6 who reported better response in ‘non-narcotic users’. It remains to be seen if regular use of simple analgesics only still depresses the laser acupuncture response. Interestingly, the use of non-steroidal anti-inflammatory drugs did not predict response in our study.

Headache was a strong predictor of less pain reduction in this study (even though the treatment protocol allowed treatment of concurrent headache). In patients with central neurogenic sensitisation11 12 various dysfunctional states may occur together. It may be possible that a patient would develop resistance to acupuncture as they ascend the ladder of severity in sensitisation. Presence of habitual headaches in a patient with CLBP could be a marker of this progression.

Other common baseline conditions associated with LBP such as bowel dysfunction, depression, neck pain were not shown to be response predictors in this study. Subjects with widespread body pain were mostly excluded from this study.

Another factor which has been shown to be a strong predictor of improvement is baseline pain severity,4 6 13 with higher baseline pain more likely to lead to greater improvement with treatment. This was demonstrated in our trial in the short term. This may be related to the phenomenon of regression to the mean or to the fact there is greater room for improvement in those with initially high pain scores.

Patients who were experiencing an exacerbation of back pain at the time of assessment also showed a greater improvement. This is understandable, though this effect was not demonstrated until the 6 month follow-up.

There were some other findings which had less statistical significance, were inconsistent or were not biologically plausible (see table 4). Surprisingly, an increased responsiveness in those with higher alcohol intake was noted. High intake was found in a small fraction (8%) of the total group. There was also some inconsistency of response depending on baseline score of psychological distress with those with higher anxiety doing worse, while subjects with high stress levels responded better.

Another observation was that subjects who had no evidence of degeneration on available imaging did worse. This is consistent with evidence of poor correlation between imaging changes and symptoms of CLBP.14 Although of interest, such findings may also have arisen by chance and warrant confirmation in further studies.

In the few previous trials which have examined the effect of baseline characteristics on outcome in acupuncture, lack of predictors has been a common finding. In this study characteristics such as age group, gender, BMI status, smoking status, attributed cause of onset and total duration of back pain, pain radiation, previous acupuncture (over 3 months prior), use of concurrent OTC medicines or other physical therapies and level of disability did not appear to affect the response to treatment. Participants on aged persons' or other pensions did as well as those participants who were in employment.

A further aim of this study was to determine whether an imbalance in baseline characteristics resulted in a biased conclusion in the primary analysis. This bias is of particular importance in trials testing the efficacy of acupuncture interventions where there is a large placebo effect and a likely smaller specific effect of treatment. It is postulated that a trial may be apparently adequately powered to detect a difference in theory, but with small or moderately sized trials chance baseline inequality involving strong predictors of outcome can easily result in bias despite randomisation. This may be a common explanation of frequent inconsistency in results of many acupuncture randomised controlled trials. The implication is that to establish efficacy in acupuncture interventions one must select very large sample sizes or have smaller trials with strict inclusion criteria to reduce the entry of poor responders.

This trial showed an apparently negative result on primary analysis, but the adjusted analysis provided a suggestion of a clinically important effect of laser acupuncture in participants after correction for baseline imbalances between treatment arms. However, this occurred only at short term (6 weeks follow-up) and may suggest a biological effect of laser which produces maximal effect on pain reduction with a time delay after a course of treatment. Posthoc subgroup and adjusted analysis however needs to interpreted with caution and the primary result of this clinical trial deserves most emphasis. The analysis suggests the need to repeat the trial with tighter exclusion criteria removing subgroups which respond poorly to acupuncture. With these restrictions, the external validity for a trial with a positive result would be reduced but could suggest efficacy of laser acupuncture in treating a less severe spectrum of chronic back pain. A positive result would also encourage further efficacy studies of laser acupuncture in other conditions and basic research to determine mechanisms of action of low level laser in pain reduction.

It must be stressed that the results of this study may not be generalisable to acupuncture using needles or other modalities. The results of this study also only relate to the overall ‘laser acupuncture intervention’ since the analysis was done on the whole group because of sample size restrictions. It remains possible that a specific laser effect may produce different responses according to subject characteristics.

A major limitation of the approach in this study was that it was a posthoc exploration after the primary analysis was performed. A selection of baseline factors were tested across a number of time points with associated problems associated with multiple testing. The use of a stricter criterion for significance with α level of 0.01 instead of 0.05 would result in a more conservative interpretation. The sample size of this trial was relatively small making its conclusions less reliable. A strength of the analysis was that that investigation clearly described how predictors of outcome were selected and how missing values were dealt with.

The general conclusion of this investigation is that it may be possible to quantify and predict how certain groups of patients may respond to acupuncture. The generalisability of these findings needs to be confirmed in further larger observational studies.


  • In a trial of laser acupuncture intervention on chronic back pain, certain baseline characteristics in participants were found to predict pain response.

  • Response predictors have led to bias in the primary analysis of the study.

  • Response predictors should be used in selecting patients for acupuncture studies and considered in their analysis.


I would like to thank Sarah Bolt, Research Officer, UWA for her help with statistical aspects of this paper and all the participants of the trial for contributing to this research.


View Abstract


  • Funding Sources of support: Australian Medical Acupuncture College.

  • Competing interests None.

  • Ethics approval This study was conducted with the approval of the Monash SCERH.

  • Provenance and peer review Not commissioned; not externally peer reviewed.

  • Detail has been removed from this case description/these case descriptions to ensure anonymity. The editors and reviewers have seen the detailed information available and are satisfied that the information backs up the case the authors are making.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.