Results are reported in UK sterling in 2009 prices. Despite most of the data being highly skewed, all data are reported as means because this is more relevant than medians for decision-makers using the information to plan services. All information is reported in terms of both resources used and costs. Information about resource use is reported correct to two decimal places and information about costs is reported correct to the nearest full penny. All SDs are reported correct to the same number of decimal places as the mean.
Resource use and cost per patient are provided, by item, for all those for whom data are available for that item. However, in order to estimate total per patient cost, the base-case analysis includes only those for whom complete data are available for all items of resource use. Base-case NHS costs use the participants for whom complete NHS cost and QALY data (n = 1272; 57%) were available; base-case personal costs use those for whom there were complete personal cost data (n = 995; 44%); complete cases for valuing lost productivity due to time off work are n = 1334 (59%). Numbers are included in all tables.
Each table of costs first presents the results using all of the data available and then, below, the costs for the 1272 participants with full data on all cost and QALYs (‘complete cases’) are presented.
NHS resource use
Consultations with physiotherapy services
Information about the total number of contacts with different aspects of the physiotherapy services is shown in . The associated costs are shown in . It is clear from these data that the total costs associated with provision of physiotherapy by PhysioDirect were slightly higher than the physiotherapy costs associated with usual care although the cost difference was not large.
Mean number and total duration of physiotherapy consultations
Mean cost (£) per patient for physiotherapy consultations
Primary care and district nurse consultations
Information about the number of primary care consultations is shown in and the cost of those consultations is shown in . There are few differences in cost but, again, overall the primary care costs for patients allocated to receive usual care were slightly smaller than those of patients allocated to receive PhysioDirect.
Mean number of primary care consultations
Mean cost (£) per patient for primary care consultations
Prescribed medication
Information about the amount of prescribed medication in primary care, and the cost of that medication, is shown in . Medication provision through primary care was extremely similar across the two groups with mean medication costs for those allocated to PhysioDirect being very slightly lower than amongst those allocated to usual care.
Mean resource use and cost for medication received by patients within primary care
Hospital costs
Resource use associated with hospital visits, including visits to the A&E department, outpatient visits and associated procedures, and inpatient stays is shown in . Again, costs in the two groups are similar but the hospital costs associated with treatment in the PhysioDirect arm of the trials are slightly higher than those associated with usual care, as shown in .
Mean number of episodes of hospital resource-use consultations
Mean cost (£) per patient for hospital care
summarises the three main elements of NHS costs other than physiotherapy. It shows that using all the data available the NHS costs are almost identical in the two arms. However, in the complete case analysis the NHS costs were higher in the PhysioDirect arm. Further examination of the data indicated that this was due to two patients in the usual care arm who had expensive hospital treatment but who had missing data on other variables and were therefore not included in the complete case analysis.
Summary of NHS costs (£) other than physiotherapy in each arm of the trial
Costs to patients and their families
Cost of telephone calls related to physiotherapy
gives the estimated costs to patients of telephoning the physiotherapy service. This includes calls to the PhysioDirect service for a consultation and calls to the face-to-face service to make an appointment. Costs to those in the PhysioDirect group are nearly twice as much as for the usual-care group, although absolute costs are relatively low.
Cost (£) to patients of calls to the physiotherapy service
Travel costs
contains information about the proportions of patients reporting expenditure on travel to both physiotherapy services and primary care. For each, the associated cost is also included in . Unsurprisingly, travel costs for physiotherapy care were slightly higher for those who were allocated to receive usual care (because of the cost of travelling to the physiotherapy appointment).
Mean resource use for patient travel
Mean costs (£) for patient travel
Out-of-pocket expenditures
The proportions of individuals reporting expenditure on over-the-counter medication, prescriptions and other out-of-pocket expenditure are given in . The costs in are for those who provided detail of expenditure. The biggest area of difference in cost between the groups appears to be in the more costly use of private therapy among those allocated to receive PhysioDirect (although the proportion reporting expenditure on private care is similar and the variation in cost is very wide).
Mean resource use for patient out-of-pocket expenditure
Mean costs (£) for patient out-of-pocket expenditure
Loss of earnings
A number of individuals reported that they had experienced a loss of earnings () as a result of their physiotherapy condition.
Mean resource use and costs associated with loss of earnings
Disability payments
The proportion of participants receiving disability payments due to their physiotherapy condition is shown in , along with mean amount received. The proportion of patients receiving these payments was similar in the two groups but patients in the usual-care group received, on average, considerably more than those in the PhysioDirect group though the variation among all participants is high as shown by the large SDs.
Disability payments to patients
Societal costs of lost production
Societal costs of lost production include the value of time off work to attend physiotherapy and time off work because of the condition. In we show the proportion of participants who reported any time off work during the 6 months and the value of that lost productivity. Data on the total cost of lost productivity were available for 1334 (59%) participants.
Mean (SD) resource use and cost to society associated with lost production from time missed from work by patients due to physiotherapy consultations, and through the condition itself
Quality-adjusted life-years
gives the EQ-5D scores at baseline and the two follow-up points, along with QALYs over the 6-month period. We present results for all available patients, which vary according to the time period, and for complete cases (n = 1272). QALYs have been adjusted to allow for the difference in baseline EQ-5D between the two groups. Patients in both groups improved during the trial period. Patients in the PhysioDirect group had a slightly lower EQ-5D score at baseline and a slightly higher score at the end of the 6-month period than those in the usual-care group; QALYs in this group were therefore slightly higher.
Quality of life and QALYs over the 6-month period
Summary of findings and conclusion
Cost consequences
gives the cost–consequence matrix. Here, total cost per patient by category is given for all of the available data. Consequences are represented by each of the primary and secondary outcomes.
Cost-effectiveness analysis
summarises the cost-effectiveness analysis using resource-use data from the NHS perspective. These data show that, although the costs to the NHS of the two services are remarkably similar, the cost of the PhysioDirect service is slightly higher overall than the cost associated with usual care. The incremental cost per QALY gained, obtained by dividing the difference in cost by the difference in QALYs, is £2889; the net monetary benefit (NMB) is £117 if society's willingness to pay for a QALY (λ) is valued at £20,000. Uncertainty around this value is illustrated in and . The cost-effectiveness plane shows the 5000 estimated incremental cost-effectiveness ratio (ICER) replicates produced using the bootstrapping technique. The CEAC shows the probability that the PhysioDirect service is cost-effective at different levels of willingness to pay for a QALY. At £20,000 per QALY gain the probability is 0.88 and at £30,000 it is 0.90.
Cost-effectiveness from NHS perspective
Cost-effectiveness plane for PhysioDirect (base-case analysis), showing 5000 bootstrapped replicates of the ICER: complete case data, adjusted QALYs.
Cost-effectiveness acceptability curve for the cost of PhysioDirect at different levels of willingness to pay, showing the probability that the intervention is cost-effective at different levels of willingness to pay: complete case data, adjusted QALYs. (more...)
Sensitivity analysis
Impact of mimicking an efficient service
If the PhysioDirect services had run during the trial in the more efficient way as seen in Bristol following the trial, the amount of non-contact physiotherapist time would have been reduced in all four centres. During the trial, the physiotherapists spent about 35% of their time during PhysioDirect sessions on the telephone or dealing with directly related administration. In Bristol, after the trial, this was increased to 57%. The cost implications of this are shown in .
Cost-effectiveness: sensitivity analysis mimicking an efficient service
Under this scenario, cost per patient in the PhysioDirect group is £14.53 less than under trial conditions and £2.11 less per patient in the usual-care group. The ICER is therefore lower, at £1045, and the NMB correspondingly higher at £127 (λ = £20,000). The probability of PhysioDirect being cost-effective at λ = £20,000 is 0.89; at £30,000 it is 0.90. This is illustrated in and .
Cost-effectiveness plane for PhysioDirect (sensitivity analysis on cost of PhysioDirect mimicking an efficient service): complete case data, adjusted QALYs; showing 5000 bootstrapped replicates of the ICER.
Cost-effectiveness acceptability curve for PhysioDirect (sensitivity analysis on cost of PhysioDirect mimicking an efficient service), showing the probability that the intervention is cost-effective at different levels of willingness to pay; complete (more...)
Impact of excluding hospital costs
Hospital costs accounted for 75% of all NHS costs yet only 19% participants reported using any secondary care. The main element of hospital costs was inpatient admissions but only 15 patients were admitted. The effect of removing hospital costs from the analysis is shown in and and .
Cost-effectiveness: sensitivity analysis excluding hospital costs
Cost-effectiveness plane for PhysioDirect (sensitivity analysis excluding hospital costs), showing 5000 bootstrapped replicates of the ICER.
Cost-effectiveness acceptability curve for PhysioDirect (sensitivity analysis excluding hospital costs) showing the probability that the intervention is cost-effective at different levels of willingness to pay.
This analysis is based on all patients for whom we had primary care and medication costs and QALY data: n = 1317 (59%). There is almost no difference in NHS costs between the two groups; the cost difference is due entirely to the difference in the cost of physiotherapy. The ICER is therefore reduced (£1084) and the NMB higher at £142 (λ = £20,000) and £217 (λ = £30,000). The probability of PhysioDirect being cost-effective at λ = £20,000 is 0.93; at £30,000 it is 0.94.
Imputation of missing data
The results of imputing missing NHS cost and QALY data are shown in and –. The mean cost of physiotherapy was lower for all patients after imputation than for those for whom we had complete data (overall mean £72.60 vs £84.04). This was true of patients in both groups, meaning that our base-case estimate of the cost of physiotherapy is conservative. Conversely, the imputed NHS cost data suggest that the complete case analysis may underestimate the true cost: the estimated cost in all patients after imputation was £131.48 compared with the base case of £108.38. The effect of imputing NHS costs is more marked in the usual-care group, which reduces the difference between the two groups. The overall effect is a reduction of the incremental cost of PhysioDirect compared with usual care from £19.30 (complete cases) to £4.46 (imputed cost data).
Cost-effectiveness after imputing missing data
Cost-effectiveness plane for PhysioDirect (sensitivity analysis after imputing missing data, adjusted QALYs), showing 5000 bootstrapped replicates of the ICER.
Cost-effectiveness acceptability curve for PhysioDirect (sensitivity analysis after imputing missing data and with efficient service scenario, adjusted QALYs), showing the probability that the intervention is cost-effective at different levels of willingness (more...)
Cost-effectiveness acceptability curve for PhysioDirect (sensitivity analysis after imputing missing data, adjusted QALYs), showing the probability that the intervention is cost-effective at different levels of willingness to pay.
Cost-effectiveness plane for PhysioDirect (sensitivity analysis after imputing missing data and with efficient service scenario, adjusted QALYs), showing 5000 bootstrapped replicates of the ICER.
Sensitivity analysis using imputed cost data in the ‘more efficient service’ scenario suggests a possible cost saving of £6.02 per patient.
The effect of imputing missing EQ-5D data is to reduce QALYs gained from 0.3299 (complete cases) to 0.3216 (imputed data). The difference is greatest in the PhysioDirect group, giving a reduced incremental QALY gain of 0.0020 as against 0.0067.
Using imputed cost and QALY data the probability that the PhysioDirect service is cost-effective at £20,000 per QALY is 0.66 and at £30,000 it is 0.67. The probability under the ‘efficient service’ scenario of the sensitivity analysis is 0.72 at both £20,000 and £30,000 per QALY.
Use of Short Form questionnaire-6 Dimensions rather than European Quality of Life-5 Dimensions
It is possible to use the data from the SF-36 questionnaire to estimate QALYs using the SF-6D measure.96 In a further sensitivity analysis (not prespecified a priori but in response to the request of a peer reviewer) we compared the findings using the SF-6D rather than the EQ-5D to generate QALYs. The results are shown in and and .
Cost-effectiveness using SF-6D rather than EQ-5D to generate QALYs
Cost-effectiveness using SF-6D rather than EQ-5D to generate QALYs, imputed data
Cost-effectiveness acceptability curves: sensitivity analyses comparing use of EQ-5D or SF-6D to generate QALYs, with or without imputation of missing data, showing the probability that the intervention is cost-effective at different levels of willingness (more...)
In order to compare the findings using the SF-6D or the EQ-5D it is necessary to include only patients who have data on both outcomes, as well as complete cost data. This reduced the denominator from 1272 patients to 1124 patients. In this group of patients, the cost-effectiveness of PhysioDirect assessed using the SF-6D was slightly lower than when assessed using the EQ-5D (see ). The probabilities that PhysioDirect was cost-effective at the £20,000 and £30,000 willingness-to-pay thresholds were 76% and 80%, respectively.
After imputing missing data, PhysioDirect again appeared to be slightly less cost-effective when QALYs were assessed using SF-6D rather than EQ-5D. The probabilities that PhysioDirect was cost-effective at the £20,000 and £30,000 willingness-to-pay thresholds were 55% and 57%, respectively (see ).
Summary of sensitivity analyses
summarises the findings from the various sensitivity analyses that were conducted to assess the robustness of the conclusions under a range of different scenarios and assumptions.
Cost-effectiveness: summary of findings from sensitivity analyses
Several trends are apparent from the above sensitivity analyses. The incremental cost per QALY (ICER) for the intervention is low. However, because the benefits in terms of QALYs and the costs are both very small, the NMB is also small (and the confidence estimates overlap zero). Under the most optimistic scenario (based on the more efficient service as observed outside the trial and after imputing missing data), the intervention is both more effective and less expensive. Under the least optimistic scenario (using the SF-6D to generate QALYs and imputing missing data), the NMB from PhysioDirect is negligible. It is notable that in all scenarios the probability that PhysioDirect is cost-effective is > 50%.
