Cost-effectiveness of manual therapy versus physiotherapy in patients with sub-acute and chronic neck pain: a randomised controlled trial

An economic evaluation was conducted alongside a Dutch 52-week pragmatic RCT [8]. The study design and informed consent procedure were approved by the Medical Ethics committee CMO Arnhem-Nijmegen (NL21128.091.08).

Patients could enter the study by directly presenting at one of the participating practices or through GP referral. Additionally, participants were recruited through advertisements in local newspapers [8]. Upon entering the trial, participants provided informed consent and a research assistant performed the baseline measurements. Subsequently, participants were randomised to the MTU or PT group by an independent research assistant using a central computer generated randomisation scheme. Randomisation was performed at the individual level using block randomization (size = 4), with pre-stratification for pain (range: 0 (no pain) to 10 (maximum pain); <7 versus ≥7) and age (<40 versus ≥40 years) [8].

Sixteen primary healthcare centres for manual therapy and PT participated in the study (i.e., patients were recruited and treated in all of these centres). Centres employed at least one manual therapist, one physiotherapist, and one research assistant. Adults (18–70 years) who had neck pain for more than 2 weeks, but no longer than 52 weeks were eligible for inclusion. Neck pain was defined as “pain in the cervical region with or without radiation to the shoulder region or upper extremities, and/or accompanied by headaches as long as the main pain complaint was the neck”. Exclusion criteria were: red flags; previous cervical surgery; pregnancy; whiplash; conditions that seriously impede treatment; insufficient knowledge of the Dutch language; or conservative neck pain treatment during the previous 3 months.

Baseline measurements included participant characteristics, demographic variables, and potential confounding variables.

Perceived recovery (yes/no) and functional status were assessed at 3, 7, 13, 26, 39, and 52 weeks using the global perceived effect (GPE) and Neck Disability Index—Dutch Version (NDI-DV), respectively. The GPE measures a participant’s subjective global improvement using a 7-point scale ranging from “worse than ever”(1) to “completely recovered”(7) [9]. Perceived recovery (yes) was defined as being “completely recovered” or “much improved”; other responses were defined as not recovered (no). The NDI-DV measures a participant’s self-rated physical disability. The NDI-DV score ranges from 0 to 50, with higher scores indicating higher disability levels. The proportion of responders on the NDI-DV was also estimated. Response was defined as improving ≥4 points on the NDI-DV (yes); other responses were defined as not responded (no) [10].

Health-related quality of life was assessed at baseline, 7, and 52 weeks using the SF-6D. The participants’ SF-6D health states were translated into utility scores using the UK tariff [11]. Quality-adjusted life years (QALYs) were calculated by multiplying the participants’ utility scores by their time spent in a health state using linear interpolation between measurement points.

Intention-to-treat analyses were performed. Descriptive statistics were used to compare baseline characteristics between MTU and PT group participants, and those with complete and incomplete data. Missing data were multiply imputed, stratified by treatment group. Using fully conditional specification and predictive mean matching, 15 datasets were created in IBM SPSS (v22, Chicago, IL) (loss-of-efficiency <5 %) [16]. Each dataset was analysed separately as specified below. Pooled estimates were calculated using Rubin’s rules, incorporating both within-imputation variability (i.e., uncertainty about the results from one imputed data set) and between-imputation variability (i.e., reflecting the uncertainty due to the missing information) [16].

Effectiveness at 52 weeks was estimated using linear regression analyses, adjusted for baseline values. To compare costs between groups, linear regression analyses were performed as well. Seemingly unrelated regression (SUR) analyses were performed to estimate total cost and effect differences (i.e., ∆C and ∆E). An advantage of SUR is that two regression equations (i.e., one for ∆C and one ∆E) are modelled simultaneously so that their possible correlation can be accounted for [17]. Incremental cost-effectiveness ratios (ICERs) were calculated by dividing the corrected difference in total costs by that in effects (i.e., ∆C/∆E). Uncertainty surrounding the ICERs and 95 % confidence intervals (95 %CIs) around cost differences were estimated using bias corrected and accelerated (BCA) bootstrapping (5000 replications). Uncertainty was graphically illustrated by plotting bootstrapped incremental cost-effect pairs (CE-pairs) on cost-effectiveness planes (CE-planes) [13]. A summary measure of the joint uncertainty of costs and effects was provided using cost-effectiveness acceptability curves (CEACs). Such curves provide an indication of the probability of MTU being cost-effective in comparison with PT at different values of willingness-to-pay [13]. Unless otherwise stated, data were analysed in STATA (V12, Stata Corp, TX). Statistical significance was set at p < 0.05.

One hundred and eighty one sub-acute and chronic non-specific neck pain patients participated. Of them, 90 were randomized to the MTU group and 91 to the PT group (Fig. 1). At baseline, no relevant differences were found between groups (Table 1). Complete data were obtained from 114 participants (62 %) on the effect measures and 147 participants (81 %) on the cost measures. Some significant differences were observed between participants with complete and incomplete data in both treatment groups (Table 1). These variables were included in the imputation model.

No significant differences were found between the MTU and PT group in perceived recovery (yes/no: β = 0.09; 95 %CI: −0.05–0.24), functional status (continuous: β = −1.03; 95 %CI: −2.55–0.48, yes/no: β = −0.01; 95 %CI: −0.15–0.13), and QALYs (β = −0.01; 95 %CI: −0.04–0.03).

Participants in the MTU group visited a manual therapist on average 3.0 times, while participants in the PT group visited a physiotherapist on average 5.1 times. This resulted in a significant difference in intervention costs between both groups in favour of the MTU group. MTU group participants had significantly lower healthcare costs and significantly higher unpaid productivity costs than their PT group counterparts. All other between-group differences were not significant (Table 2).

The ICER for perceived recovery was −1024, indicating that an additional recovered patient in the MTU group compared with the PT group was associated with a societal cost saving of €1024 (Table 3; Fig. 2-1a). The CEAC in Fig. 2-2a indicates that the maximum probability of cost-effectiveness was 0.88. To reach this probability, societal decision-makers should be willing to pay €39,000 per additional recovered patient in the MTU group compared with the PT group.

When functional status was evaluated as a continuous outcome, an ICER of 92 was found. This indicates that an additional 1-point improvement on the NDI-DV was associated with a societal cost saving of €92 (Table 3; Fig. 2-1b). The CEAC presented in Fig. 2-2b indicates that the maximum probability of cost-effectiveness was 0.91 if societal decision-makers are willing to pay €4000 per additional 1-point improvement on the NDI-DV in the MTU group compared with the PT group.

When functional status was evaluated as a dichotomous outcome (yes/no), an ICER of 7314 was found. This indicates that €7314 was saved by society per participant less with an improvement of ≥4 points on the NDI-DV (Table 3). The CEAC (not shown) indicated that the maximum probability of MTU being cost-effective compared with PT was low (≤0.54), irrespective of the willingness-to-pay.

For QALYs, an ICER of 14,561 was found, indicating that one QALY lost was associated with a societal cost saving of €14,561 (Table 3). The CEAC (not shown) indicated that the maximum probability of MTU being cost-effective in comparison with PT was low (≤0.54), irrespective of the willingness-to-pay.

In line with the main analysis, between-group differences in total costs and effects were not significant in all sensitivity analyses. The overall conclusion of this study would not change when using any of the assumptions of the sensitivity analyses.

In comparison with PT, MTU was associated with significantly lower intervention costs and healthcare costs, whereas unpaid productivity costs were significantly higher. The latter was due to the fact that more MTU group patients reported to experience unpaid productivity losses compared with their PT group counterparts (X ² = 18.5; p = 0.000). Total costs did not significantly differ between groups, nor did all other between-group differences in costs and effects. For QALYs and functional status (yes/no), the maximum probability of MTU being cost-effective in comparison with PT was low (≤0.54). For perceived recovery and functional status (continuous), large amounts of money must be paid by society per additional unit of effect to reach a reasonable probability of cost-effectiveness (e.g., €39,000 per additional recovered patient for a probability of 0.88). Sensitivity analyses confirmed these results. Therefore, MTU cannot be regarded as cost-effective in comparison with PT from a societal perspective. As no clear total societal cost and effect differences were found between MTU and PT, the decision about what intervention to administer, reimburse, and/or implement can be based on the preferences of the patient and the decision-maker at hand.

There may be several potential explanations for the finding that MTU was not cost-effective in comparison with PT. First, only one form of manual therapy (i.e., MTU) was compared to PT, whereas a more eclectic approach to manual therapy may be more likely to be (cost-)effective. This explanation is underscored by the study of Korthals-de Bos et al. (2003) who found manual therapy to be cost-effective in comparison with PT when it incorporated several techniques used in Western Europe, North America, and Australia, including those described by Cyriax, Kaltenborn, Maitland, and Mennel [6]. Second, the participants’ baseline functional status scores were relatively favourable, leaving only a small margin for improvement. This reasoning is underscored by a post hoc analysis, indicating that a larger mean difference in functional status (continuous) was found between the MTU and PT group when participants with a mild or no disability (i.e., NDI-DV <15) were excluded from the analyses (β−1.83; 95 %CI: −5.44–1.78; n = 52) [10]. Third, MTU was only compared to PT, whereas recent evidence indicates that a combination of manual therapy and PT is most likely to be effective [19]. As economic evaluations are sparse in the field of manual therapy [20] and evidence regarding the relative cost-effectiveness of a combination of manual therapy and PT versus an alternative strategy is lacking, future research into this topic is warranted.

A first strength of this study is its pragmatic RCT design. Such a design is acknowledged as the best vehicle for economic evaluations, as its pragmatic approach improves the findings’ generalisability to daily clinical practice, whereas the randomisation of participants reduces the possible influence of selection bias [13]. A second strength is that cost-effectiveness was analysed using SUR analyses, which made it possible to correct for the possible correlation between costs and effects [17]. A third strength is that not only QALYs and functional status were used as an outcome measure in the economic evaluation, but also perceived recovery as measured by the GPE. This is important because the GPE was found to capture chronic neck pain patient perceptions of chance in domains that may not be captured by other outcome instruments [21].

The present study also has some limitations. First, cost and effect measure values were collected using self-report, which may have caused “social desirability” and/or “recall bias”. Nonetheless, as it seems unlikely that the extent of impairment in recall and/or the degree to which participants gave social desirable answers systematically differed between groups, we do not expect that our reliance on them severely biased the results. Second, even though reduced on-the-job productivity (i.e., presenteeism) was found to account for the biggest share of productivity-related costs due to pain complaints [22], presenteeism costs were not included. This may have resulted in an underestimation of the societal costs. Another limitation concerns the amount of missing data. To deal with this limitation, missing cost and effect data were multiply imputed. Multiple imputation is currently considered the most appropriate method for imputing cost data, as it accounts for the uncertainty about the missing data by creating several imputed data sets [16]. Also, some significant differences in baseline characteristics were observed between patients with complete and incomplete data in both treatment groups, indicating that the complete-case analysis is likely biased by self-selection of participants.