More cost-effective management of patients with musculoskeletal disorders in primary care after direct triaging to physiotherapists for initial assessment compared to initial general practitioner assessment

The study population and setting have been described in more detail earlier [11]. Participants were recruited from three Swedish primary care clinics with registered patients from varying sociodemographic levels, such as are representative for the urban Swedish population. These clinics all had several years’ experience working according to the triage model described above. Patients with MSD were first assessed by triage nurses as suitable for triaging directly to physiotherapists. Clinical triage instructions were to book all patients seeking help for MSD and with no apparent immediate need for GP services directly to a physiotherapist [22]. Focus was to be placed on MSDs with recent debut or recent flare-up. Other inclusion criteria were being of working age (16–67 years) and having sufficient command of either Swedish or English to fill out the questionnaires. Patients who required home visits or primarily needed medical aids were excluded. Patients who had ongoing treatment for the current MSD were not eligible, nor were those seeking help for chronic conditions with unchanged symptoms the last three months and who had already tried physiotherapeutic treatment for this condition. The inclusion and exclusion criteria were formulated to mirror as accurately as possible the patients which the nurses normally triaged directly to physiotherapists, to make it easier for them to follow the study protocol and to ensure that study results would be representative for the clinical setting. Participants received no financial or other compensation.

The intervention consisted only of determining which profession would make the initial assessment. All visits and services after the initial visit were treatment as usual. Eligible patients agreeing to participate in the study were randomised to a first visit with either a physiotherapist (intervention) or a GP (treatment as usual (TAU)). The healthcare providers were unaware of study participation and were free to provide the treatment and services they thought appropriate. Participants were followed for 1 year from inclusion regarding health-related quality of life, healthcare resource use and absence from work for MSD. A cost-efficiency analysis was performed from a societal perspective where formal and informal healthcare costs as well as production effects for one year following inclusion were compared for the group initially assessed by a physiotherapist and the group initially assessed by a GP. As the relevant time period was 1 year, no discounting was necessary. A power analysis was performed for the original RCT requiring 63 participants per group [11]. Recruiting was, however, discontinued early when impending organizational changes threatened to hinder study conditions with physiotherapist placement at the primary healthcare centres.

Health-related quality of life (HRQoL) was chosen as the health benefit in the evaluation as a generic measure of health improvement. This was measured at inclusion and at 2, 12, 26 and 52 weeks with Euroqol 5 dimensions-3 L (EQ5D) [27]. Questionnaires were sent home to participants with reminders as necessary. Quality-adjusted life-years (QALYs) were calculated based on index values for EQ5D, assessed using the Dolan tariff, and the relevant time after inclusion [28]. QALYs were calculated for each participant based on linear interpolation between each measurement point and then combining the “areas under the curve”. Regression analysis was used to adjust for baseline differences in HRQoL [29].

The participants were given paper diaries to record healthcare services and visits and were instructed to record only services and visits which were relevant for the MSD for which they were included in the study. Diaries were collected and replaced at each follow-up time together with EQ5D. Healthcare resources in the form of visits to GPs and physiotherapists, referrals to secondary care specialists and to radiological examinations, prescription details and the number of days of sick-leave were reported. Data from the diaries was supplemented, when available, by information in the medical records.

Formal healthcare costs for GP and physiotherapist visits and referred services were collected from the healthcare organisation. Capital and operating costs and social fees were included in mean hourly rates each for physiotherapists and GPs, which were then used to calculate total costs for each participant’s visits. Prescribed pain medication was recorded per patient during the follow-up period and then linked to Swedish market prices.

There are several accepted methods used to determine how to value production costs in cost-effectiveness analyses [30]. Here, the human capital approach was used to value production loss [30, 31]. This includes all work hours lost due to health problems and healthcare treatment. Informal costs for patients’ time use and production loss due to healthcare visits and sick-leave during the follow-up period were based on mean gross wages (including social fees) in the area. The number of sick-leave days was self-reported, double-checked in the medical records and includes the first week of absence from work which, in Sweden, does not require a physician’s sick-note. Production loss for visits to healthcare providers was based on a presumed 30-min visiting time for GPs, 45 min for physiotherapists, 1 h for referred visits and examinations plus 1-h travel/waiting time for all visits. If the patient was on sick-leave at the time of the visit, then compensation level for unpaid work with net mean wage after tax was used.

The incremental cost-effectiveness ratio (ICER) was calculated as the ratio between the difference in mean costs and mean QALYs (∆Cost/∆QALYs) between the physiotherapist group and the TAU group from randomisation to the 12-month follow-up. The ICER can be interpreted as the price tag per one-unit change of mean QALYs [32]. Multiple imputation (MI) was used to deal with missing data. A multivariate MI approach with 20 sequential imputations based on chained regressions using Stata v.15 was implemented, with baseline HRQoL, age, sex, comorbidities and treatment status used as factors.

Sampling uncertainty was assessed using non-parametric bootstrapping, where ICERs were estimated based on 1000 bootstrap resamples and summarised in a cost-effectiveness plane (CE-plane) and in a cost-effectiveness acceptability curve (CEAC).

The CE-plane is typically divided into four quadrants (north-east, north-west, south-east, and south-west). The south-east quadrant represents a situation where the intervention dominates over the control leading to both health improvements and monetary savings. The north-east and south-west quadrants represent the “trade-off” quadrants where increases in costs must be compared to improvements in health (north-east quadrant) or cost-savings to reduced health (south-west quadrant). Lastly, the north-west quadrant represents a situation where the intervention is dominated by the control leading to reduced health at greater costs.

The CEAC shows the probability that the intervention is cost-effective compared to the control for a range of different maximum acceptable ICERs. This is a useful illustration considering that the de facto “threshold value” (or willingness-to-pay per QALY) differs between different healthcare contexts and jurisdictions.

Of the 55 included patients, 2 were dropped because they were missing EQ5D data at baseline as well as at all subsequent measurement points. Among the remaining 53 patients, data was missing for costs and EQ5D between 0% (EQ5D at baseline) and 56% (EQ5D at the final measurement). There was no statistically significant difference between the intervention and TAU groups with respect to missing cost data (p-value = 0.17) or missing EQ5D data (p-values ranging from 0.15 to 0.7). Missing data was slightly more common in younger patients, but not related to the sex, country of birth or comorbidities.

Table 3 shows the difference in mean costs, mean QALYs and the incremental cost-effectiveness ratio for the intervention compared to TAU. The point estimates show a lower cost as well as a higher number of QALYs in the intervention group compared to TAU, both from a societal perspective including healthcare costs, production loss and patient’s time costs as well as from a healthcare perspective including only formal healthcare costs. However, the differences in mean cost and QALYs are not statistically significant using the standard rule of thumb with 95% confidence intervals. The point estimates imply that the ICER represents an intervention that dominates care as usual, i.e. less expensive and better.

In Fig. 1, the uncertainty of the ICER is assessed based on 1000 bootstrapped replicates plotted on the CE-plane. The point estimate of the ICER (Table 3) is shown in the south-east quadrant (black diamond). This and the large mass of replicated ICERs in this quadrant represent a situation where the intervention dominates TAU (less costly and higher QALYs). A smaller share of the bootstrapped ICERs are located in the “trade-off” north-east (higher QALYs, higher costs) and south-west (lower QALYs, lower costs) quadrants. Finally, a small share of ICERs (7–8%) are also scattered in the north-west quadrant, representing a situation where the intervention is dominated by treatment as usual.

The cost-effectiveness acceptability curve (CEAC) shows the probability that the intervention is cost-effective compared to TAU for a range of different maximum acceptable ICERs (Fig. 2). At a willingness-to-pay threshold of 20,000 €, the probability that the intervention is cost-effective is approximately 85%. The probability that the intervention is cost-effective never reaches higher than 92–93% irrespective of the willingness-to-pay threshold because there is a small probability (approximately 7–8%), as seen in Fig. 1, that the intervention is dominated by TAU.

Patients who were triaged to a physiotherapist had generally higher costs for physiotherapy and lower costs for GP visits and referrals. The difference between groups in health outcomes was small but nonetheless favoured the physiotherapy-triage group. This combined with reduced sick-leave and production loss for the physiotherapist-triaged group led to the assessment of high probability of cost-efficiency. The difference in production loss was the single largest factor affecting cost-efficiency. It is possible that immediate active treatment of MSD speeds recovery, leading to less time off work. The RCT on which this economic evaluation was based indicated tendencies to better clinical course after initial triaging to physiotherapists [11].

Fewer patients who initiate treatment with a physiotherapist received referrals for radiological investigations and specialist assessments. It is known that, for prevalent sub-groups of MSDs such as low back pain and early osteoarthritis, excessive and unnecessary referrals are common and that patients are often active in requesting these services [33‐35]. As physiotherapists do not usually have the right to make direct referrals, it requires an extra step, after a triage visit, for the patient to acquire a referral. If the information and treatment from the physiotherapist have already fulfilled many of the patient’s needs and reduced worry about long-term effects, it may affect the number of patients who feel it necessary to continue the process involved to acquire a referral.

Few management-based interventions are studied extensively. It is not uncommon to implement decisions and reorganise healthcare based only on presumed or calculated advantages for the healthcare organisation. This triage model has been studied from several perspectives and has not been found to lead to any disadvantages. The effects on health aspects, healthcare utilization, accessibility, work environment, patient satisfaction and now cost-efficiency all favour triaging to physiotherapist over traditional management [11, 19, 20, 22]. Involved stakeholders – patients, care-givers, the healthcare organisation and the community – all seem to win by reversing the order of involved care-givers. The results concerning the cost-efficiency of direct triaging to physiotherapist are corroborated by those few studies which have, in some way, compared the economic effects of primary physiotherapist and GP contacts in primary care [19‐21, 36].

Changing standard praxis is never easy. However, the potential gains for doing so likely compensate for any difficulties. Continuing with primary GP contact for all conditions is less efficient [24] and often difficult to provide [37]. Enabling patients to consult physiotherapists directly on their own is not the same as actively triaging suitable patients to physiotherapists. Experience in Austria has indicated that unregulated access to healthcare leads to overutilization of predominantly the most specialized care as patients tend to prefer the care level they deem to have the highest competence but leading to an inefficient use of professional competence [38]. This study showed that, when the GP is the primary contact for patients with MSDs assessed as suitable for primary physiotherapeutic contact, less than 40% of patients actually continued on to see a physiotherapist. This may indicate that primary GP consultation not only delays (comparing referral to physiotherapist with direct triaging to physiotherapist) but often replaces active physiotherapeutic treatment which may be necessary for the most optimal long-term results [39]. It is known that early active treatment leads to advantages regarding clinical course and need for care for low back pain, which is the largest sub-group of MSDs [40‐42]. Primary physiotherapeutic management of MSD may also lead to reduced focus on passive management aspects such as medication and sick-leave and to greater focus on self-responsibility and secondary prevention [39].

It seems that the triage nurses select relatively healthy patients for triaging to physiotherapists as the levels of sick-leave, medication and comorbidity are fairly low in both groups. It may be that the results of this study are most applicable to those patients with uncomplicated MSDs. This sub-group of patients likely has less need of GP services. However, even this sub-group needs optimal management to prevent development of chronic symptoms. This study also included patients who were in need of treatment by both GPs and physiotherapists. The favourable outcome applies to them as well. The order in which patients are assessed by different professions does seem to affect both clinical course and resource use.

There was a considerable level of missing data in the study which was dealt with using multiple imputation. This is the best-practice recommended approach for dealing with missing data in cost-effectiveness analyses, as well as in other forms of applied statistical analyses of health data [43].

There are divided opinions about the most appropriate perspective to use in cost-effectiveness analyses [44]. Here, the focus was on the working-age population and on non-fatal conditions. It was, therefore, important to include societal costs for production loss and the human capitol approach was chosen to value production loss. For illnesses causing primarily short-term or no sick-leave, there is relatively little difference between the predominant methods to value production loss (human capitol approach and friction cost approach) [31]. A more restricted healthcare perspective and the associated ICER have also been presented for comparative purposes, indicating that triaging to physiotherapists leads to economic savings for the healthcare organization.

There are few and relatively small studies, with primarily observational study designs, investigating the effects of triaging to physiotherapists in primary care on resource use [19‐21, 25]. This study had a robust randomised design adapted pragmatically to the clinical environment but was based on a small population. The study was underpowered due to practical difficulties, which should theoretically increase the risk for type II errors. A type II error means accepting the null-hypothesis (that there are no differences between groups) when the null-hypothesis is false. In other words, an underpowered study will increase the risk for underestimating the effects of the intervention. However, a significant difference between groups for referral frequency was found despite the small group size. Non-parametric bootstrapping was employed to reduce sampling uncertainty regarding cost-effectiveness. This statistical method has been found to provide accurate estimates of the mean even with small samples with skewed data [32]. Larger randomised studies will, however, be needed to corroborate this study’s findings.