We conducted an economic evaluation alongside a randomised controlled trial to establish the cost-effectiveness of PhysioDirect compared with usual care based only on face-to-face treatment over a period of 6 months. The trial and its clinical findings have been reported in full elsewhere.15
The aim of the evaluation was to provide information about the long-run costs and benefits of the alternative methods of running a physiotherapy service for this patient group so with that in mind we excluded the initial set up costs associated with establishing the new telephone service, including the training undertaken by the practitioners.24
As the nature of the intervention suggests there could be an impact on patients’ costs, and as it is known that musculoskeletal conditions account for a considerable amount of time off work14
we chose to use a cost-consequences approach, comparing cost from all three perspectives (healthcare provider, patients and carers, lost productivity) with a range of clinical outcomes.25
However, the perspective of greatest interest to the UK policy makers is that of the health and social care provider27
so we also conducted a cost-utility analysis to compare cost to the NHS with Quality-Adjusted Life Years (QALYs).
Setting and participants
We recruited adults aged 18 and over from four community physiotherapy services in England—Bristol, Somerset, Stoke-on-Trent and Cheshire—which provided diversity in terms of socioeconomic status and a mix of urban and rural communities. All patients referred by their GP, or who referred themselves, for physiotherapy for a non-urgent musculoskeletal problem were invited to take part. Patients were randomised to PhysioDirect or usual care on a two to one basis to increase the chances of the PhysioDirect service being fully utilised.
The intervention has been reported in detail elsewhere.15
Patients randomised to the PhysioDirect service received an invitation to telephone a senior (band 6 or above) specially trained physiotherapist, who assessed their musculoskeletal problem with the aid of previously developed computerised templates.23
These templates were provided by Huntingdonshire Primary Care Trust, which has been operating a similar service since 2001. Patients were then sent appropriate advice leaflets about self-management and exercises to try at home, and invited to phone again and/or make a face-to-face appointment if necessary. If the service was engaged when the patient called, the call was answered by a receptionist who added the patient to a ‘call-back’ list and the physiotherapist would return the call when they were free. Patients randomised to usual care were put on the usual service waiting list for face-to-face assessment and treatment.
We used the EQ-5D-3L28
valued using the UK tariff,29
to estimate QALYs gained for the cost-utility analysis. The primary outcome for the trial was the physical component summary (PCS) measure from the SF-36v2 questionnaire30
and secondary clinical outcomes included: the Measure Yourself Medical Outcomes Profile (MYMOP)31
; a Global Improvement Score—a single question about overall improvement; a composite measure of response to treatment including pain, function and overall improvement (OMERACT OARSI);32
patient satisfaction and waiting time to first treatment advice from a physiotherapist. All outcomes (except the global improvement score and waiting time to first treatment advice) were measured at baseline, 6 weeks and 6 months and were obtained from a self-completed questionnaire administered at these three time points.
The analysis was based on costs related to the reason for which the patient was referred to the physiotherapy service. We identified relevant resources in discussion with participating physiotherapists and service managers. Direct costs to the healthcare provider included: cost of initial and follow-up physiotherapy consultations; primary and community consultations; hospital care and prescribed medication. Patient and carer costs included: telephone calls to the PhysioDirect service; travel; over-the-counter medication; prescription costs; private therapy and purchase of equipment; extra domestic help and loss of earnings. Lost productivity was estimated separately in relation to time off work to attend physiotherapy appointments and time off because of the musculoskeletal condition itself.
Patient level data about all physiotherapy appointments and consultations were recorded either automatically by computer or by the physiotherapist treating the patient. For those in the intervention group, the PhysioDirect assessment software recorded which physiotherapist conducted each telephone call, and the duration of each call. In addition to the time logged on to the system physiotherapists had to carry out administrative activities following each telephone call, such as collating information to send to the patient by post. The time spent on these activities was estimated from information available at one site (Bristol) where manual recording of the entire encounter supplemented the electronic recording.
Physiotherapists assigned to the PhysioDirect service were required to be available throughout the time the service was ‘open’ but they were not usually fully engaged in dealing with patients in the PhysioDirect service during these hours. We conducted an observational time and motion study at each of the four sites to determine how they occupied their non-PhysioDirect contact time in order to apportion costs appropriately. Time and motion data were collected at points in the study when the sites were expected to be fully operational, and across a mix of day, time of day and location. The capital costs required to run a telephone service are potentially less than for a face-to-face service. Each site provided information about space and equipment required to run their telephone service and we used this to estimate an overall percentage reduction of capital costs for these compared with a standard face-to-face service.
Data about all face-to-face appointments were recorded routinely. These data included the length of appointment, the grade of the physiotherapist seen, and information about missed appointments.
Information about other NHS resource use was collected, where possible, from general practice records and supplemented by information gained directly from patients. General practice notes were scrutinised for patient level data on primary care consultations and prescribed medication. We included all consultations at which musculoskeletal condition for which the patient was referred to physiotherapy was mentioned and these were recorded by type of consultation (eg, face-to-face, telephone, out of hours, home visit) and by type of professional seen (eg, GP, nurse). It was not feasible to distinguish between medication prescribed for the condition for which the patient was referred to physiotherapy and any other musculoskeletal problem so we included all medication of a potentially relevant type, defined using British National Formulary (BNF)33
coding. These were: analgesics (chapters 4.7.1–4.7.2); non-steroidal anti-inflammatory drugs (10.1.1); local corticosteroid injections (10.1.2.2) and drugs for the relief of soft-tissue inflammation (10.3).
A questionnaire was administered to participants at 6 weeks and 6 months after randomisation to obtain resource use data not available elsewhere. The questionnaire was designed specifically for this study but was similar in content and structure to others used for the same purpose.34
Questions included information about hospital care related to the condition for which the patient was referred to physiotherapy: visits to accident & emergency, outpatient appointments, and inpatient stays. Information about personal expenditure relevant to the patient's musculoskeletal condition was also gained from the questionnaire at 6 weeks and 6 months. We asked about the cost of travel to physiotherapy and other healthcare appointments, expenditure on over-the-counter medication, prescription costs, use of private therapies and their cost, expenditure on equipment or devices and extra help at home. In addition, participants were asked about any time off work, and the associated loss of earnings, because of their condition or to attend healthcare appointments relating to the condition including usual care physiotherapy and PhysioDirect.
Valuation of resource use
gives the unit costs and sources used to value the healthcare resources. We used Curtis35
to value primary and community healthcare and Department of Health reference costs36
for all hospital-based care. The cost of prescribed medication was estimated from that published in the BNF,33
adjusted to allow for the discount available to the NHS, and the professional fee and container allowance in accordance with the Drug Tariff for England.37
Personal expenditure was reported directly by the participants, the exception being travel by car, which was reported as mileage and costed using the AA schedule of motoring costs.38
Time off work was valued using the median gross weekly earnings by age and sex.39
The cost of face-to-face physiotherapy consultations was estimated by adapting the methods of Curtis35
to obtain a different unit cost for each band of staff at each site. National median pay rates, by band,42
were adjusted to allow for National Insurance, superannuation, and overheads, as per Curtis, then further adjusted to allow for band and site-specific non-contact time. Information about the proportion of time physiotherapists on each grade typically spend in direct contact with patients was provided by the four physiotherapy service managers. This provided us with a cost per hour for each band of staff at each site.
The unit cost of physiotherapists working in the PhysioDirect service was estimated in a similar way, but allowing for the reduced cost of capital and overheads; information from the site managers indicated this to be about 50%. To obtain a cost per hour of telephone contact we used information from the computerised records of the PhysioDirect service, which identified the proportion of time spent by physiotherapists actually dealing with patients in the PhysioDirect service. We then combined this with data from the time and motion study, which identified activities undertaken during non-contact time, for example, administration relating to face-to-face appointments or general administration, to give a cost per hour for each band of staff at each site.
All costs were valued in £ sterling at 2009 prices, adjusted for inflation where necessary.35
We investigated the amount of each resource used by patients in each group using frequencies, means and medians. Mean total cost per participant was derived by combining resource use with unit costs.
QALYs were derived from responses to the EQ-5D-3 L at baseline, 6 weeks and 6 months using valuations from the UK general population.29
These values, representing health-related quality of life on a scale between 0 (death) and 1 (best imaginable health), were used to compute QALYs experienced over the 6-month period using the area under the curve approach and adjusting for any difference between the groups at baseline.43
A cost-consequences matrix was constructed using all available data. We compared costs from all three perspectives (healthcare provider, patients and carers, lost productivity) with the SF-36v2 PCS, MYMOP, Global Improvement Score, OMERACT OARSI, patient satisfaction, waiting time and QALYs.
The cost-utility analysis was carried out using data on all patients for whom we had complete NHS cost and QALY data. An incremental cost-effectiveness ratio (ICER) was constructed, comparing the difference in mean total cost per patient with mean difference in QALYs, thus the lower the ICER, the greater the cost-effectiveness and the better the value for money. Uncertainty around the ICER was captured using the bootstrapping technique: 5000 replicates of the cost and QALY data were created by sampling from the original data, with replacement. The range and spread of the 5000 ICERs was used to construct a cost-effectiveness acceptability curve (CEAC) to indicate the likelihood of the intervention being cost-effective. The net monetary benefit (NMB) of the intervention was estimated from the point estimate of the ICER for values of societal willingness-to-pay of £20 000 and £30 000/QALY. If the NMB is positive at a given level of willingness to pay, the intervention is regarded as cost-effective. CIs around the NMB were formed from the bootstrapped estimates.
We used the multiple imputation by chained equation procedure to address the issue of missing cost and EQ-5D data.44
This technique uses a regression model to estimate missing values from known values. In addition to cost and EQ-5D-3 L variables the imputation model also included randomisation group, age, sex and SF36v2 PCS. Stata V.1245
was used to generate five datasets using 10 switching procedures.
Discounting was not carried out because the analysis was restricted to costs and outcomes over a period of less than a year. All analyses were conducted using Microsoft Excel and Stata V.12.45
We addressed three areas of uncertainty using four one/two-way sensitivity analyses. First, we estimated the cost of running the PhysioDirect service if it was operating at full capacity. It is likely that this was not achieved during the trial because of low demand due to exclusions and non-participation in the trial; inflexible staffing levels to ensure consistency throughout the trial period; and the ‘one-way’ system generally used, where physiotherapists waited for patients to call them but did not routinely contact patients themselves (notwithstanding some limited use of answer-machines). Data from the Bristol service, which continued to operate beyond the trial period and was then able to tailor staffing levels to demand, were used to estimate the cost of running a more ‘efficient but feasible’ PhysioDirect service once the trial had ended.
The second area of uncertainty addressed hospital costs. Patients in the trial were recruited from primary care and for these, use of secondary care is infrequent but relatively expensive and this can have a disproportionate effect on mean total cost. We tested this by excluding hospital costs from the total.
The third area of uncertainty tested the effect of using imputed data rather than complete cases; the third sensitivity analysis used trial data with missing values imputed.
Finally, in a two-way sensitivity analysis, we re-estimated the results of the first, (mimicking an ‘efficient but feasible’ service) in this instance using the imputed dataset.