A brief pain management program compared with physical therapy for low back pain: Results from an economic analysis alongside a randomized clinical trial




Guidelines for the management of acute low back pain in primary care recommend early intervention to address psychosocial risk factors associated with long-term disability. We assessed the cost utility and cost effectiveness of a brief pain management program (BPM) targeting psychosocial factors compared with physical therapy (PT) for primary care patients with low back pain of <12 weeks' duration.


A total of 402 patients were randomly assigned to BPM or PT. We adopted a health care perspective, examining the direct health care costs of low back pain. Outcome measures were quality-adjusted life years (QALYs) and 12-month change scores on the Roland and Morris disability questionnaire. Resource use data related to back pain were collected at 12-month followup. Cost effectiveness was expressed as incremental ratios, with uncertainty assessed using cost-effectiveness planes and acceptability curves.


There were no statistically significant differences in mean health care costs or outcomes between treatments. PT had marginally greater effectiveness at 12 months, albeit with greater health care costs (BPM £142, PT £195). The incremental cost-per-QALY ratio was £2,362. If the UK National Health Service were willing to pay £10,000 per additional QALY, there is only a 17% chance that BPM provides the best value for money.


PT is a cost-effective primary care management strategy for low back pain. However, the absence of a clinically superior treatment program raises the possibility that BPM could provide an additional primary care approach, administered in fewer sessions, allowing patient and doctor preferences to be considered.


Back pain in the UK poses a sizeable epidemiologic and economic problem, with direct health care costs estimated to be £1.6 billion in 1998 (1). Early referral to physiotherapy is recommended for patients with nonspecific low back pain in whom first-line treatment by general practitioners (GPs) has not been effective (2, 3). In such patients, evidence for the effectiveness of specific types of physiotherapy is conflicting (4, 5) but supports active intervention above no treatment or sham procedures (6, 7).

The reasons why patients may not recover are complex. Epidemiologic evidence has consistently shown that psychosocial factors are important determinants of outcome in low back pain (8, 9), with psychosocial distress and inappropriate beliefs about back pain leading to an increased risk of chronic disability (10–12). Accordingly, UK clinical guidelines recommend a broad approach to the primary care management of low back pain, with early attention to psychosocial factors (2, 3). Hospital-based multidisciplinary pain management programs based on cognitive behavioral techniques have been shown to reduce disability and increase the likelihood of return to work in patients with chronic low back pain (13, 14). These programs are labor intensive and would be impractical to run for the large number of patients presenting with low back pain in primary care. One approach recently adopted was to take the important key messages and principles from hospital-based biopsychosocial management programs and incorporate them within the physiotherapy management of low back pain in primary care (15). The present study compared the cost effectiveness and cost utility, in primary care, of a brief pain management program (BPM) delivered by physiotherapists with that of a physical therapy (PT) program including spinal mobilization techniques in the treatment of nonspecific low back pain of less than 12 weeks' duration.


Trial design.

Comprehensive details of the trial design have been published elsewhere (15). The study population included adults ages 18–64 years presenting to 1 of 28 general practices in North Staffordshire, UK, for the first or second time, with an episode of nonspecific low back pain of <12 weeks' duration. Of 544 patients assessed for eligibility, 402 (74%) were recruited and randomly assigned to either BPM (n = 201) or PT (n = 201).


Six physiotherapists administered the interventions. Three worked exclusively within the PT arm and 3 worked within the BPM arm. Patients commenced their intervention within 10 working days of randomization. The interventions consisted of one 40-minute assessment/treatment session, plus up to 6 subsequent 20-minute treatment sessions.

The PT program was designed to be consistent with best current manual physiotherapy practice in the UK. All of the physiotherapists administering this program had completed postgraduate training in manual therapy. Treatment approaches were standardized, with emphasis on diagnosing and treating biomechanical dysfunction of the spine with manual therapy techniques (including mobilization, manipulation, or other soft-tissue treatment approaches) and back-specific exercises coupled with ergonomic advice. Psychosocial risk factors were not targeted.

The physiotherapists administering the BPM program were trained to predetermined competencies in the biopsychosocial model of care. The program was designed to identify and address known psychosocial risk factors for persistent or recurrent back-related disability. No manual therapy techniques were permitted. Patients were assessed using a stem and leaf interview (16), with the aim of developing a management plan that included explanation about pain mechanisms, promotion of positive coping strategies, and implementation of a graded return to normal activities through functional goal setting. Individualized exercises focused on increasing overall physical activity and spinal mobility, but were not back specific.

Data collection.

The number of treatment sessions attended by BPM and PT patients was obtained from standard proformas completed by the trial physiotherapists. Health outcome data were collected at baseline, 3 months, and 12 months by means of self-reported postal questionnaires. At 12 months, a separate standardized questionnaire was self-completed to collect data on health care resource use for the 12-month period after randomization. Specifically, this questionnaire collected data on back pain–related resource use for hospital inpatient stays, outpatient consultations, and visits to other health care practitioners (both within the UK National Health Service [NHS] and private practice). Information on consultations with NHS primary health care providers (e.g., GP, practice nurse), prescribed medications, and over-the-counter purchases was also collected.

A record review was undertaken in a random sample of 10% of patients to validate responses to questions concerning hospital referrals to NHS health care professionals and additional consultations with primary care. The study nurse collected data from medical records of GP consultations (electronic and paper copies) during the period of followup. Referrals and consultations were categorized separately as yes (had at least 1 referral/consultation) or no (no referral/consultation) for each patient. The self-report data were similarly dichotomized to cross-validate the findings. The study nurse was blind to the self-report questionnaire data. The results indicated good reliability (17), with 89% and 81% agreement and kappa values of 0.66 and 0.63 for hospital referrals and primary care consultations, respectively.

Health outcomes.

Clinical outcomes were utility and change in back-related disability. Disability was measured on the Roland and Morris disability questionnaire (RMDQ), a 24-item back pain–specific disability scale with scores ranging from 0 (best health) to 24 (worst health), and was expressed as changes between baseline and 12-month scores. Utility was measured by the EuroQol (EQ-5D) at baseline, 3 months, and 12 months. This generic health status measure provides utility values for all possible responses to the 5-dimension questionnaire based on health state valuations elicited from a large representative sample of the UK population (the York A1 tariff) (18). These utility values range from 1.00 (patient reports no problems on all dimensions) to −0.59 (patient reports severe or extreme impairment on each dimension). Using area under the curve analysis (19), these health utility values allowed us to estimate the number of quality-adjusted life years (QALYs) that patients experienced during the 12-month period after randomization. In this study, the maximum number of QALYs was equal to 1, equivalent to a year spent in full health, with QALYs <1 reflecting less than perfect health. We assumed that utility changed linearly between consecutive data collection points.

Cost estimation.

To acknowledge important cost implications beyond those of the UK NHS, a health care perspective was adopted for the economic analysis, including both public sector and private sector health care resources. The unit costs of NHS and private health care resources are shown in Table 1, with costs expressed in pounds sterling at 2001/2002 prices. Discounting was not necessary because the followup period was 12 months. NHS care was costed as national averages (20), with inpatient and outpatient episodes costed using the 2002 NHS reference costs (21). Due to the paucity of high-quality unit cost data for private health care consultations or procedures, private care was costed as the NHS equivalent. The British National Formulary was used to cost prescribed medication (22). Because physiotherapists are the largest group of health professionals providing acupuncture in the UK, acupuncture costs were assumed to be equivalent to the costs of physiotherapy.

Table 1. Unit costs of health care resources at 2001/2002 prices*
Health care resourceUnit cost (£)
  • *

    BPM = brief pain management; PT = physical therapy; NHS = National Health Service.

  • Additional hospital-based consultations within the NHS.

  • Consultations with private health care practitioners.

Treatment sessions (20) 
 40-minute assessment for BPM package24.67
 40-minute assessment for PT package24.67
 20-minute consultation for BPM package12.33
 20-minute consultation for PT package12.33
Primary care contacts (20) 
 General practitioner16.00
 General practitioner, home visit49.00
 Practice nurse8.00
 Physiotherapist, home visit40.00
Inpatient episodes (cost per episode) (21) 
 NHS, surgical1,823.00
 NHS, medical1,563.00
 Private, surgical1,823.00
Outpatient attendances (21) 
 NHS, rheumatology: injection92.00
 NHS, rheumatology: new attendance95.00
 NHS, rheumatology: followup68.00
 Private, injection92.00
 Private, new attendance95.00
 Private, followup68.00
 Accident and emergency59.00
Other health care professionals (20) 
 Consultation within the NHS 
  First visit23.33
  Followup visit11.67
 Private consultation 
  First visit23.33
  Followup visit11.67

The 12-month resource use questionnaire also required patients to record days off work due to low back pain measured on a 5-point scale ranging from none to >3 months. There was no significant difference in response between treatment groups (P = 0.57, chi-square test for trend). Although these data were collected, given the health care perspective chosen for this analysis, the number of days off work remained a secondary outcome in the clinical study and were not costed (15).

Statistical analysis.

Analysis was performed according to the intention-to-treat principle. Multiple imputation was used to address the issue of missing EQ-5D scores, RMDQ scores, and resource use data (23). Resource use data, when assigned appropriate unit costs, allowed us to estimate the costs associated with low back pain (defined by the perspective of the study) for each patient in the trial. Accordingly, we imputed a total health care cost for patients who had not provided adequate resource use data. We generated 5 possible values for each missing value using multiple linear regression models containing age, sex, treatment group, and a number of clinical outcomes as covariates. The final imputed value was the arithmetic mean of the 5 data values created, with reported standard deviations and confidence intervals adjusted to account for the additional between-imputation variance (24).

Cost and QALY data alongside randomized clinical trials are invariably skewed. We generated 95% confidence intervals (95% CIs) around differences in mean costs, EQ-5D scores, QALYs, and RMDQ change scores using conventional parametric methods and bias-corrected and accelerated (BCa) bootstrapping (1,000 replications). Intervals using these 2 approaches were compared, where appropriate, to assess whether parametric methods were robust to the level of skewness in the data. Baseline imbalances between treatment groups for RMDQ scores and utility were controlled for using a multiple regression–based adjustment (25).

The aim of the economic evaluation was to estimate and compare the mean health care costs associated with the 2 treatment programs and relate this to their effectiveness. An incremental approach was used in the analysis, with differences in costs and health outcomes expressed using an incremental cost per 1-point improvement in RMDQ change score (an incremental cost-effectiveness ratio [ICER]) and an incremental cost per QALY (an incremental cost-utility ratio [ICUR]). Due to sampling variation, there was uncertainty around these ratios. If there is a nonnegligible probability that the effect difference is close to zero, confidence interval estimation for these ratios is problematic (26). Therefore, uncertainty was analyzed by generating 5,000 bootstrapped replications of mean cost and outcome differences, and by plotting these pairs on cost-utility and cost-effectiveness planes, where incremental QALYs (or RMDQ change scores) are represented along the x-axis and incremental costs along the y-axis (27). Given that it is difficult to have a prior conception of the monetary value that society would place on a 1-point improvement in RMDQ change scores, a threshold analysis was performed for the ICUR only. This involved using a cost-utility acceptability curve to estimate the probability that the new intervention (BPM) would be cost effective at specific cost-per-QALY thresholds (28, 29). Statistical analysis was performed using SPSS for Windows, version 13.0 (SPSS, Chicago, IL) and Stata software, version 9 (StataCorp, College Station, TX).

Sensitivity analysis.

The methodology adopted in the main analysis addressed many issues of uncertainty (multiple imputation of missing data, confidence interval estimation for mean differences, and controlling for baseline imbalances between treatment groups). As a sensitivity analysis, a complete case analysis was performed to validate the use of a multiple imputation approach to deal with missing data. To be eligible for the complete case analysis, patients were required to provide complete resource use and outcome data during the study period.

A second sensitivity analysis was performed to explore how robust the results were to variation in the cost of private health care. Unit costs for private health care are likely to exceed those of the NHS. If the use of private health care is systematically different between treatment groups, the difference in private health care costs will be underestimated when assuming unit costs equivalent to the NHS. This sensitivity analysis involved simultaneously multiplying the unit costs of private health care by a price premium ranging from 1 (the base case condition where private costs equal NHS costs) to 3 (where private costs are 3 times the unit cost of the NHS equivalent).


Results of the clinical trial have been reported elsewhere (15). Briefly, RMDQ change scores did not differ significantly between groups at 3 months or 12 months. For secondary outcomes, both groups were similar with respect to back pain and function, psychosocial measures, and clinical assessments.

Health care costs and health outcomes.

Complete data on health care resource use were provided by 299 (74%) patients at the 12-month followup. Of the 103 patients with incomplete data, 92 (89%) provided no resource use data at all. For the health outcomes, 284 (71%) patients completed the EQ-5D at baseline, 3 months, and 12 months, and 329 (82%) patients provided RMDQ scores at baseline and 12 months. There were no differences in response rates between treatment groups; nonresponders were younger and more likely to be male (15).

Mean resource use and mean costs for the observed data (i.e., patients that provided complete resource use data prior to imputation [n = 299]) are reported in Tables 2 and 3 in a disaggregated format to allow readers and decision makers to select the information most relevant to their perspective. BPM patients attended significantly fewer treatment sessions (difference in means −0.72; 95% CI −1.24, −0.20) at a significantly lower treatment session cost (mean cost difference −£9.57; 95% CI −£16.40, −£2.75). Following imputation, the total mean health care cost per patient for BPM was £142.33 compared with £194.52 for PT; this cost saving was not statistically significant (Table 3).

Table 2. Health care resource use per patient by treatment group*
Health care resourceResource use per patient
BPM (n = 145)PT (n = 154)
  • *

    Values are the mean ± SD. See Table 1 for definitions.

Treatment sessions  
 BPM treatment sessions3.61 ± 2.30.00 ± 0.0
 PT treatment sessions0.00 ± 0.04.32 ± 2.2
Primary care contacts  
 General practitioner1.14 ± 2.70.94 ± 2.2
 General practitioner, home visits0.03 ± 0.30.01 ± 0.1
 Practice nurse0.01 ± 0.10.02 ± 0.1
 Physiotherapist0.16 ± 1.10.00 ± 0.0
 Physiotherapist, home visits0.03 ± 0.40.00 ± 0.0
 Acupuncturist0.00 ± 0.00.08 ± 1.0
Inpatient episodes  
 NHS, surgical0.00 ± 0.00.02 ± 0.1
 NHS, medical0.01 ± 0.10.01 ± 0.1
 Private, surgical0.01 ± 0.10.01 ± 0.1
Outpatient attendances  
 NHS0.11 ± 0.40.13 ± 0.5
 Private0.04 ± 0.30.13 ± 0.7
 Accident and emergency0.00 ± 0.00.01 ± 0.1
Other health care professionals  
 Acupuncturist, NHS0.08 ± 0.80.29 ± 2.4
 Acupuncturist, private0.01 ± 0.20.08 ± 1.0
 Chiropractor, private0.19 ± 1.10.32 ± 3.2
 Physiotherapist, NHS1.97 ± 6.01.03 ± 4.1
 Physiotherapist, private0.28 ± 2.00.85 ± 4.0
 Osteopath, private0.08 ± 0.50.40 ± 2.7
 Other, NHS0.00 ± 0.00.01 ± 0.1
 Other, private0.05 ± 0.40.12 ± 0.7
Prescribed medications0.55 ± 1.00.48 ± 0.9
Over-the-counter purchases0.48 ± 0.60.42 ± 0.7
Table 3. Health care costs per patient by treatment group*
Health care resourceCosts (£) per patientMean difference (95% CI)
BPM (n = 145)PT (n = 154)
  • *

    Values are the mean ± SD unless otherwise indicated. 95% CI = 95% confidence interval; see Table 1 for additional definitions.

  • Difference = BPM − PT. Confidence intervals were generated using conventional parametric methods. For the individual cost components, the only significant cost difference was for the treatment sessions (see text).

  • P = 0.15.

Treatment sessions   
 BPM treatment sessions55.46 ± 31.00.00 ± 0.0 
 PT treatment sessions0.00 ± 0.065.03 ± 28.9 
Primary care contacts   
 General practitioner18.32 ± 42.415.06 ± 35.2 
 General practitioner, home visits1.69 ± 16.70.64 ± 5.6 
 Practice nurse0.06 ± 0.70.16 ± 1.1 
 Physiotherapist2.38 ± 17.00.00 ± 0.0 
 Physiotherapist, home visits1.38 ± 16.60.00 ± 0.0 
 Acupuncturist0.00 ± 0.01.27 ± 15.7 
Inpatient episodes   
 NHS, surgical0.00 ± 0.035.51 ± 252.8 
 NHS, medical10.78 ± 129.820.30 ± 177.5 
 Private, surgical12.57 ± 151.411.84 ± 146.9 
Outpatient attendances   
 NHS outpatient9.74 ± 34.810.90 ± 44.0 
 Private outpatient3.56 ± 23.410.21 ± 52.9 
 Accident and emergency0.00 ± 0.00.77 ± 6.7 
Other health care professionals   
 Osteopath, private1.29 ± 7.85.23 ± 32.9 
 Chiropractor, private2.57 ± 14.94.24 ± 39.0 
 Acupuncturist, NHS1.05 ± 10.83.64 ± 29.5 
 Acupuncturist, private0.24 ± 2.90.98 ± 12.2 
 Physiotherapist, NHS24.70 ± 72.913.36 ± 50.3 
 Physiotherapist, private3.46 ± 25.310.61 ± 49.4 
 Other, NHS0.00 ± 0.00.15 ± 1.9 
 Other, private0.72 ± 6.21.97 ± 9.9 
Prescribed medications4.15 ± 11.22.67 ± 6.0 
Over-the-counter treatments4.00 ± 10.65.37 ± 14.4 
Total health care cost for observed data (n = 299)158.11 ± 282.2219.90 ± 449.6−61.79 (−146.76, 23.18)
Total health care cost with imputed data (n = 402)142.33 ± 261.3194.52 ± 445.6−52.19 (−123.62, 19.22)

Mean RMDQ scores, EQ-5D scores, and QALYs are presented in Table 4 for each treatment group. Controlling for baseline RMDQ and EQ-5D scores increased the difference in mean RMDQ change scores and QALYs between groups at 12 months, indicating a baseline imbalance that favored the BPM group. PT resulted in a larger mean QALY and RMDQ change score at 12 months, although these differences were not statistically significant. Confidence intervals generated using BCa bootstrapping provided similar results, indicating that the parametric intervals were robust to the level of skewness in the data (bootstrapped intervals not reported).

Table 4. Health outcomes over 12 months: main analysis and complete case analysis*
Outcomes (EQ-5D, QALYs, RMDQ)BPMPTMean difference (95% CI)
  • *

    Values are the mean ± SD unless otherwise indicated. EQ-5D = EuroQol; QALYs = quality-adjusted life years; RMDQ = Roland and Morris disability questionnaire; 95% CI = 95% confidence interval; see Table 1 for additional definitions.

  • Difference = BPM − PT. 95% CIs were generated using conventional parametric methods. There were no statistically significant differences between the groups for each of the analyses (P ≥ 0.05).

  • Values are predicted scores from the multiple regression when controlling for baseline imbalances.

  • §

    P = 0.35 (2 dp).

  • P = 0.57 (2 dp).

  • #

    P = 0.25 (2 dp).

Main analysis (n = 402)   
 Baseline EQ-5D0.702 ± 0.30.696 ± 0.30.006 (−0.05, 0.06)
 3-month EQ-5D0.758 ± 0.40.789 ± 0.3−0.031 (−0.10, 0.04)
 12-month EQ-5D0.770 ± 0.30.785 ± 0.3−0.014 (−0.08, 0.05)
 QALYs0.756 ± 0.30.776 ± 0.2−0.020 (−0.07, 0.03)
 QALYs (controlled for baseline EQ-5D)0.7550.777−0.022 (−0.07, 0.02)§
 Baseline RMDQ13.771 ± 4.813.289 ± 4.90.483 (−0.47, 1.43)
 12-month RMDQ5.062 ± 6.24.558 ± 6.80.503 (−0.76, 1.77)
 RMDQ change score8.709 ± 6.68.730 ± 7.3−0.021 (−1.39, 1.35)
 RMDQ change score (controlled for baseline RMDQ)8.5538.887−0.334 (−1.49, 0.82)
Complete case analysis (n = 264)   
 Baseline EQ-5D0.710 ± 0.30.706 ± 0.30.004 (−0.06, 0.07)
 3-month EQ-5D0.751 ± 0.30.779 ± 0.2−0.028 (−0.09, 0.03)
 12-month EQ-5D0.774 ± 0.20.788 ± 0.2−0.013 (−0.07, 0.04)
 QALYs (controlled for baseline EQ-5D)0.7540.774−0.020 (−0.06, 0.02)§
 Total health care cost (£)167.75 ± 295.6221.32 ± 447.6−53.56 (−145.92, 38.80)#

Cost-utility analysis.

The point estimates for mean health care costs and mean QALYs suggested that PT was both more effective and more costly, with a cost-per-QALY ratio of £2,362. The cost-utility plane with 5,000 bootstrapped incremental cost-utility pairs comparing BPM with PT is shown in Figure 1. The majority of replications are located in the lower-left quadrant (86%), which demonstrates that on average patients randomized to BPM gained fewer QALYs, but at a lower mean health care cost compared with the PT group.

Figure 1.

Incremental cost-utility plane comparing brief pain management (BPM) with physical therapy (PT). QALYs = quality-adjusted life years.

The presence of bootstrapped pairs in all 4 quadrants of a cost-utility plane denotes that any ICUR should be interpreted with caution. Acceptability curves provide a clear way of interpreting these results (Figure 2). Each point on the curve represents the probability that BPM is cost effective at specific cost-per-QALY thresholds. The curve does not follow the familiar “textbook” shape because the new intervention in this study (BPM) was associated with fewer QALYs and a lower cost (30). It does not pass through the origin because there are bootstrapped cost-utility pairs associated with cost savings attributable to the BPM program (95% of the 5,000 pairs are in the lower-left or lower-right quadrant in Figure 1). Similarly, the curve does not asymptote to 1 because there are cost-utility pairs associated with a lower mean QALY in the BPM group (91% of the pairs are in the upper-left or lower-left quadrant). The higher the willingness-to-pay threshold, the less likely BPM is to be considered cost effective. Reading from the curve shows that if the cost-per-QALY threshold was a conservative £10,000 per QALY gained (31, 32), the chance that BPM is cost effective is only 17% (i.e., an 83% chance that PT is the preferred option).

Figure 2.

Cost-utility acceptability curve comparing brief pain management (BPM) with physical therapy. QALY = quality-adjusted life year.

Cost-effectiveness analysis.

On average, patients randomized to the PT group had a greater change score on the RMDQ at 12 months, although the difference in mean change scores was not significant (Table 4). Point estimates for mean costs and RMDQ change scores provided an ICER of £156 per unit RMDQ change. This suggests that the additional cost of a 1-point improvement in RMDQ change score at 12 months, resulting from PT compared with BPM, is £156. The cost-effectiveness plane provided similar findings to the cost-utility analysis, with 70% of the bootstrapped pairs in the lower-left quadrant.

Analysis of sensitivity.

As in the main analysis, there were no statistically significant differences between mean costs, QALYs, or RMDQ change scores for all further analyses. Results of the sensitivity analysis for the cost-utility analysis are reported below.

A total of 264 (66%) patients provided sufficient data to be included in the complete case analysis. The incremental cost and QALY point estimates resulted in an ICUR of £2,701 (Table 4), with 72% of the bootstrapped replications in the lower-left quadrant. Assuming a cost-per-QALY threshold of £10,000, BPM had a 26% chance of being the most cost-effective treatment strategy. Comparing P values with those generated in the main analysis demonstrated that multiple imputation reduced the uncertainty surrounding the difference in mean total health care costs (Tables 3 and 4).

The second sensitivity analysis introduced a price premium on all private health care unit costs to examine the robustness of the results to such variation (because we were considering variation in costs, the outcomes remained the same). Table 2 shows that private health care resource use was more frequent in the PT group. Varying the premium between 1 and 3 had little effect on the findings from the main analysis. Assuming private health care unit costs 3 times that of their NHS equivalent (premium = 3), the total mean health care cost per patient for BPM was £197.92 compared with £302.35 for PT. This cost saving of £104.43 (resulting in an ICUR equal to £4,727) was not statistically significant (95% CI −£242, £33). The lower-left quadrant contained 85% of the bootstrapped pairs, with BPM associated with a 31% chance of being the more cost-effective treatment option when assuming a £10,000 cost-per-QALY threshold.


An increasing number of economic evaluations are being performed to help decision makers provide efficient ways to allocate health care resources. In the context of low back pain, there is considerable uncertainty as to what the most cost-effective management strategy is for this large patient group (33). This study presents new cost-utility and cost-effectiveness data, providing an important contribution to the literature. We report the results of an economic evaluation comparing a BPM program, administered by physiotherapists, with a PT program in the treatment of nonspecific low back pain.

Both outcome measures used in this study (RMDQ change scores and QALYs) had similar results, with point estimates suggesting a slightly greater clinical benefit in the PT group but lower mean health care costs for BPM. These results did not change considerably in the sensitivity analysis, which considered a complete case analysis and variation in the unit costs of private health care. The estimated ICUR of £2,362 lies comfortably within previous recommendations by the National Institute for Health and Clinical Excellence (31). This incremental ratio is relatively small, therefore, the recent literature concerning appropriate decision rules for the lower-left quadrant does not prevent us from making clear conclusions (34, 35). Following the quantification of uncertainty around the cost-per-QALY ratio, we conclude that PT is likely to be cost effective.

Comparing the results of this study with other back pain–related economic evaluations is difficult due to the different patient populations. For example, the mean health care costs reported here are considerably lower than those reported in the UK Back Pain Exercise and Manipulation (BEAM) trial of physical treatments for back pain in primary care, where the manipulation group had a mean health care cost of £541 (32). However, the UK BEAM trial collected generic health care costs, which could account for an increased use of health care resources. Our findings are consistent with previous economic evaluations of alternative low back pain treatments in identifying greater clinical effects, and greater mean health care costs, within a manipulation-based treatment program (although not statistically significant) (32, 36).

Relying on patients to recall health care resource use introduces 2 types of bias: recall bias (where patients forget the resources they have used) and telescoping (the tendency to remember events as having occurred more recently than they did) (37). We carried out a small validation exercise, based on a random sample of 10% of patients, to check the reliability of patients' recall of primary care consultations and hospital referrals. Although based on a small sample and a simple data acquisition method, the analysis demonstrated good reliability between patient questionnaire responses and GP consultation records. More generally, because economic evaluations are ultimately concerned with incremental costs and consequences, the influence of bias in data collection methods is likely to be nominal, as it will not systematically differ between groups.

Missing data in clinical trials raise questions about the internal validity of a study. Although response rates were good for both health care resource use and health outcomes, we used multiple imputation techniques to overcome any potential bias due to missing data. Comparing the conclusions of the main analysis and the complete case analysis demonstrated that a multiple imputation approach can reduce the degree of uncertainty around key parameters despite the additional between-imputation uncertainty. Conversely, it should be acknowledged that this additional uncertainty is not adequately captured when applying bootstrapping techniques to generate cost-utility planes and acceptability curves.

Patients randomized to BPM reported fewer inpatient stays and outpatient consultations than the PT group (Table 2). Due to the small number of episodes reported, it was not possible to determine whether hospital referrals were significantly lower in the BPM group, or if there was a significant cost saving. These episodes are considerable cost drivers in economic evaluations of treatments for musculoskeletal disorders. Larger adequately powered trials are needed to further investigate whether a pain management program would result in significantly fewer hospital referrals.

The findings from this study are applicable to a particular group of patients with low back pain treated within a UK health care system. As with all economic evaluations, some degree of caution should be exercised when applying the results to different health care systems. Although unit costs are likely to be different for other health care systems, the key issue is whether the cost difference between treatment groups would increase sufficiently as to shift the ICUR beyond a decision makers willingness to pay for additional QALYs. Given that we have demonstrated PT to be cost effective when considering a conservative cost-per-QALY threshold and explored the implications of increased private health care unit costs, we believe the results to be generalizable to other health care systems.

Evidence from this economic evaluation has demonstrated that if the UK NHS values an additional QALY in excess of £5,000, PT provides the best value for money. However, the absence of a clinically superior treatment program does raise the option of providing a BPM service in significantly fewer clinical sessions, although further research is necessary to explore the potential benefits. This might be an attractive option to health care providers given the capacity constraints in primary care, the high prevalence of low back pain, and the possibility of fewer hospital referrals. Furthermore, multiple treatment options would allow additional factors such as patient and doctor preferences to be considered when choosing a back pain management strategy (38).


Professor Hay had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Whitehurst, Lewis, Hay.

Acquisition of data. Whitehurst, Lewis, Mullis.

Analysis and interpretation of data. Whitehurst, Lewis, Yao, Bryan, Raftery.

Manuscript preparation. Whitehurst, Lewis, Bryan, Raftery, Hay.

Statistical analysis. Whitehurst, Lewis, Yao, Bryan, Raftery.


We thank the following organizations and individuals who were directly involved with the study for their support, without which this work would not have been possible: the Primary Care Trusts of North Stoke, South Stoke, Newcastle-under-Lyme, and Staffordshire Moorlands; the musculoskeletal physiotherapy service of the University Hospital and the community physiotherapy service of North Staffordshire; the North Staffordshire General Practitioners Research Network and all of the GPs who referred patients to the trial; the Health Informatics team and Administration team from the Primary Care Musculoskeletal Research Centre; and those members of the public who took part in the study. We also thank the anonymous reviewers for their comments.