Overview

For each participant the total NHS and Personal Social Services (PSS) costs were estimated by summation of the intervention cost and other NHS and PSS costs. Carer input (coated in terms of lost productivity) and certain patient-borne costs were also estimated. Costs were estimated in UK sterling (£) at 2010–11 financial year levels.

Training

Therapists were provided with training in order to deliver the intervention. Three types of training were provided: (1) group training in Nottingham; (2) individual training to a particular site therapist; and (3) within-site cascade training. The time inputs by all members of staff were estimated for each of these training methods, including preparation and travel time. The unit cost of NHS community therapy time, as estimated by Curtis et al. ⁴² was assumed to apply to all time inputs, where this was adjusted to be grade 7 for the individual and group trainer (compared with grade 5 for the cascade trainer and for those receiving the training and subsequently delivering the intervention). Travel distances were also estimated, and assigned a travel cost of 54p per mile.⁴² Total training costs (the sum of staff and travel costs) were apportioned across all participants allocated to the intervention arm of the study.

Therapy contacts

Participants in both arms received a baseline visit; this was assumed to last 1 hour, including preparation, travel time and the writing up of notes, as well as the patient contact. Generally, this was provided by one therapist, although two therapists may undertake this (or subsequent visits) if there were issues relating to handover/safety, thus those who provided the intervention were asked to note which therapist(s) attended each visit (if this was not recorded it was assumed that one therapist provided the intervention). The associated travel costs for the baseline, and all subsequent therapy visits, were estimated in the following manner. After discussion with those who delivered the service, it was assumed that a travel time of 12 minutes would apply to each visit (this is in line with a previous assumption in relation to GP home visits⁴²). Assuming an average speed of 29.9 miles per hour (mph) (the average free-flow vehicle speed in a 30-mph limit,⁴³ this would equate to a total mileage of 5.98 miles per visit. Applying a travel cost of 54p per mile,⁴² this would equate to a travel cost of £3.23 per visit.

Those in the intervention arm received extra intervention visits, where the time associated with these visits (including travel to and from the participant’s home) was prospectively recorded by the therapist providing the intervention. It was additionally assumed that, for each visit, there would be 5 minutes’ preparation time and a further 10 minutes to write up associated patient records. Again for these visits the associated travel time was assumed to be 12 minutes, with a travel cost of £3.23 per visit. Thirty-minute supervisor meetings were also assumed to occur (one per site per month) for the duration of the study, where these were assumed to be 1 : 1 (therapy grade 7 and grade 5). Total supervisor meeting costs were equally apportioned across all visits. No other costs were included, as these were considered negligible (e.g. occasional bus trips with the participant, as part of the intervention).

Levels of resource use

The UK National Institute for Health and Care Excellence (NICE) recommends that costs can be calculated from the perspective of the NHS and PSS.⁴⁴ Accordingly, at both 6 and 12 months post randomisation, participants were asked to complete a resource-use questionnaire and return it by post. In this they were asked the number of times they had received different NHS and PSS services, any other care and certain patient costs that had been incurred. Specific questions with regard to health-care professional visits (and where they took place), hospital attendances and admissions, residential/nursing home admissions, home help (from community care assistant/someone who lives with them/other friends or family), Meals on Wheels and equipment purchased (to help with a health problem) were included.

Assumptions made in order to assign costs to items of resource use

Previously estimated unit costs^{42 , 45} were assigned to levels of resource use, where the following assumptions were made. Within Curtis et al.,⁴² the length of time and/or cost associated with patient contacts (including home visit) is not stated for many health professionals, although the cost per hour of employment is generally available. Costs in terms of per hour of employment, per practice visit and per home visit are available, however, for the GP. Thus, the ratio of 1 hour of employment compared with (1) a practice visit and (2) a home visit can be calculated. This ratio was applied to the costs per hour of employment of other staff to estimate associated practice/hospital visit and home visit costs. Where the cost per hour of employment, for a particular health professional, was not reported within Curtis et al. ⁴² the average cost across the following health professionals was used: general practitioner, practice nurse, district nurse, dietitian, physiotherapist, occupational therapist, social worker, speech and language therapist, for the respective type of visit (GP, home or hospital). If the place of a health professional visit was not reported then it was assumed that the patient travelled to the health professional (patient travel costs were not estimated/included).

With regard to hospital admissions (in the past 6 months), if the length of a hospital admission was not reported the mean length of stay (per admission) for other respondents (who reported both the number of admissions and the accompanying length of stay) was applied to each admission that was reported but had no accompanying length of stay.

We also asked about the number of times a person had been admitted to both a residential home and a nursing home (in the past 6 months). We did not request that participants report the length of any associated stays in such care home we thereby made the assumption that each time a person reported they were admitted to a residential/nursing home they had stayed there for three of the preceding 6 months (the average length of stay in a care home has been estimated to be 801 days⁴⁶ and we assumed that participants were on average admitted half-way through the 6-month period). If more than one admission was reported, it was assumed the participant had been in the home for the whole 6-month period in question.

Participants were asked whether they had attended a day-care centre in the past 6 months, and if they had, how many times per week they attended. The number of reported day centre attendances per week was assumed to apply to all weeks in the 6-month period in question.

Participants were asked to report the number of times (in the past week) they had home help or a visit from a community care assistant, and how long that person stayed. We requested that participants report the average time per visit. If the average time per visit was not reported, the average time per visit for other respondents (who reported both the number of times and the accompanying average time per visit) was applied to each visit that was reported but had no accompanying time per visit. Additionally, some of the responses were higher than what we considered to be a possible visit length. For example, one participant reported 28 visits and a visit length of 960 minutes (16 hours). This equates to more hours than there are in a week. We thereby assumed that this participant, and all others for whom the product of the number of visits and the associated time was greater than the number of hours in a week (a total of four participants at 6 months and five participants at 12 months), had misinterpreted the question and reported the total length of contact in the week, rather than the average per visit. Thus for these participants the reported visit length was assumed to be the total for the whole week. Again, it was assumed that the number of visits/hours reported for the week in question applied to all weeks in the past 6 months. In line with the therapist intervention, the associated travel time for each home help visit was assumed to be 12 minutes, with a travel cost of £3.23 per visit.

Participants were additionally asked how many times they had received help from someone they lived with, in the past week and the average associated length of time. They were also asked to report the level of such help from people they do not live with. In both these questions, we requested that people report the average length of time (per visit) for the help they received. If the average time per visit was not reported, the average time per visit for other respondents (who reported both the number of times and the accompanying average time per visit) was applied to each visit that was reported but had no accompanying time per visit. Also, for some participants, the product of the number of visits and the associated time was greater than the number of hours in a week (with regard to someone they lived with this occurred for a total of two participants at 6 months and two participants at 12 months, and for people they do not live with this occurred for one participant at 6 months). We thereby assumed that these participants had misinterpreted the question and reported the total length of contact in the week, rather than the average per visit. Thus for these participants the reported visit length was assumed to be the total for the whole week. Again it was also assumed that the number of visits/hours reported for the week in question was equivalent to the average per week across all weeks in the past 6 months. Within both these questions, participants were asked whether the person who provided the help had had to take time off work to provide this help. In order to provide an estimate of lost productivity, a cost was only applied to the total number of hours they were estimated to have received (both for people they live with, and do not live with) if they reported that the person that provided the care had to take time of work to provide such care. In this case, the average hourly earnings⁴⁷ was applied to these times, consistent with the human capital approach.⁴⁸ However, as the costing of such care is sometimes considered controversial,⁴⁹ these costs are also reported separately to the aforementioned NHS and PSS costs.

Participants were asked to report if they had Meals on Wheels and, if so, the number that they had received in the past week, and, if they paid for them, the amount they cost. Meals on Wheels that were not reported to be paid for by participants were assigned a previously estimated unit cost⁵⁰ and classified as a PSS cost. The costs for those who reported that they paid for them themselves were classified as patient-borne costs. Again, it was assumed that the number of meals reported for the week in question was equivalent to the average per week across all weeks in the past 6 months.

Participants were also asked to report any equipment they had bought, or been given, to help with a health problem and, if so, to state the equipment, who paid for it (the participant/social services) and the cost of the item. Items that were not reported to be paid for by social services were classified as a patient-borne cost. When a cost was not stated, where possible, previously estimated unit costs were assigned to items (where these were taken from, e.g. Curtis et al. ⁴²). When a unit cost for an item could not be identified, or the type of item was not reported, either the cost of what was considered to be a similar item or the average cost of all items for which a unit cost was identified was assigned to the item in question. This question did not specify the time frame over which it was interested in equipment purchases; consequently, even although all other questions specified the previous 6 months, it is possible that some of the reported items may have been purchased before the participant joined the study. The potential impact of this was reduced, however, as the equivalent annual cost of equipment purchases was calculated,⁴⁸ for which the discount rate was assumed to be 3.5% per annum and the lifespan of the equipment was assumed to be 7 years.

Categorisation of costs

The above enabled a cost to be assigned to each of the resource-use questions. These were then categorised, as follows, where all costs relate to the 12-month post-intervention period: the costs associated with health professional or home help visits; visits to accident and emergency, walk-in centres, outpatients, day centres; and admissions to hospital, residential homes and nursing homes.

Meals on Wheels and equipment, for which the participant did not pay, were summed to estimate other NHS and PSS costs. Table 17 provides details of the resources monitored within each question. These were, in turn, added to the intervention costs to provide an estimate of total NHS and PSS costs (base-case analysis). The costs associated with help from people they live or do not live with were added together to provide an estimate of lost productivity. Similarly, the costs associated with Meals on Wheels and equipment, for which the participant did pay, provided an estimate of the costs borne by the patient. Finally, overall costs were estimated by the summation of intervention costs, other NHS and PSS costs, lost productivity and patient-borne costs.

TABLE 17

Description of the costs associated with the intervention and control

Overall and incremental costs

For each of the aforementioned cost categories, the mean incremental cost of the intervention (over the 12-month follow-up period) was calculated by subtracting the mean cost for the control group from the mean cost for intervention group.

Measuring outcomes

To estimate the impact on health-related quality of life, participants were asked to complete the EQ-5D⁵¹ at baseline, 6 and 12 months post randomisation. The EQ-5D has five questions, through which the respondent is asked to report the level of problems they have (no problems, some/moderate problems, and severe/extreme problems) with regard to mobility, self-care, usual activities, pain and anxiety/depression.³⁰ The three-level version of the EQ-5D (EQ-5D-3L) was used. Responses to these five dimensions are converted into one of 243 different EQ-5D health-state descriptions, which range between no problems on all five dimensions (11111) and severe/extreme problems on all five dimensions (33333). A utility score (a scale where death = 0 and full health = 1) was assigned to each of these 243 health states using the York A1 tariff⁵² (associated EQ-5D scores range between –0.594 and 1.00). Completion of the EQ-5D enabled a cost–utility analysis to be undertaken, in which the benefits of different health-care treatments can be compared on a common utility scale.⁵³ The area-under-the-curve method⁵³ was used to estimate the mean quality-adjusted life-year (QALY) gain/loss over the 12-month trial period for both groups. Within these QALY calculations, those who died within the study period were assigned a utility score of ‘0’ upon death.

In a similar way, responses to 11 of the questions on the SF-36⁵⁴ were used to estimate a score on the SF-6D.²⁵ The SF-6D is composed of six dimensions (physical functioning, role limitations, social functioning, pain, mental health and vitality), which have between four and six levels. A non-parametric model (which uses Bayesian methods)⁵⁵ was used to estimate SF-6D health-state utility values for each of 18,000 potential health states (associated SF-6D scores range between 0.203 and 1.00). QALY gains/losses were again calculated, as for the EQ-5D, and those who died were, again, assigned a score of ‘0’.

As the EQ-5D and SF-6D utility measures are based on both different health-state descriptions and use different valuation techniques, they could produce different utility scores for the same group of patients. This study therefore sought to explore the impact the choice of utility measure had on estimates of cost–utility, as further research has been argued to be necessary in this area.⁵⁶ It should be noted, however, that NICE currently recommends that the EQ-5D be used within the reference case analysis,³⁹ and thus this constituted our main base-case analysis (see below).

Cost-effectiveness

After checking that dominance was not apparent (this would occur if one intervention were less costly and more effective than another),⁵³ the incremental cost per QALY gain (ICER) associated with the intervention was calculated (mean incremental cost/mean incremental QALY gain). In line with NICE guidance⁴⁴ we compared the ICER to a cost-effectiveness threshold (λ) of £20,000 per QALY.

Decision uncertainty

The bootstrap technique⁵⁸ (with 5000 replications) was used to estimate the 95% CIs surrounding the incremental cost and incremental effect (where appropriate), the 95% CI was estimated using the percentile method.⁵⁹ As the ICER has the potential to be misinterpreted,⁶⁰ we also estimated the incremental net benefit (INB) (and associated 95% CI) at a threshold of £20,000 per QALY. A negative INB would indicate that the intervention was not cost-effective at this threshold. The bootstrap samples were also used to estimate the cost-effectiveness acceptability curve (CEAC) for each group, where the CEAC depicts the probability that an intervention is cost-effective at different levels of λ.⁶¹ The probability of the intervention being cost-effective was specifically estimated at the (λ) of £20,000 per QALY.

Sensitivity analysis

We assessed how robust conclusions were to the following changes:

1. Received six or more intervention visits:
   1. On the assumption that those who received more visits might benefit more, here, only intervention participants who were included in the base-case analysis and had six or more visits were included in the analyses. The control group was the same as for the base case.
2. MI:
   1. To impute missing data in this data set, we used regression methods to predict these values based on their relationship with other covariates (age, sex, residential status, cost and utility data). Imputation took place in 10 cycles, the estimates from which were then pooled and calculated using Rubin’s Rules. All MI was performed for incomplete cost and outcomes components at the patient level using the mi impute mvn procedure in Stata 11.
3. Winsorising:
   1. As outlined previously, for the clinical data, we replaced data values below the 5th percentile with the 5th percentile value and to data values above the 95th percentile with the 95th percentile value. This was applied to the cost and QALY data for those individuals included in the base-case analysis. One reason for undertaking this analysis was the wide variation in some of the resource-use data that was reported, e.g. home help. This approach reduces the influence of extreme values and may partially test whether or not some of the assumptions in relation to these costs were correct.
   2. Different cost perspective.

Results were re-estimated from an overall cost perspective.

Training

Training costs were as follows (see Table 17 ). Total trainer time associated with group training (provided to eight sites at once) and individual training (to a further nine sites) was estimated to be 35.75 hours (including travel and preparation). Total trainer time for the cascade training (provided at seven sites) and time for receipt of all types of training (including travel) was estimated to be 14 and 62 hours, respectively. Trainer (group and individual) unit costs were estimated to be £45, compared with £31 per hour for cascade trainers and trainees. The total cost of all staff time associated with all aspects of training was thereby estimated to be £3938.89, with the addition of £918 for travel costs to sites (eight people attending the group training and nine site visits, with an average return trip of 100 miles), this gave a total training cost of £4856.89. Apportioned across all 287 participants in the intervention arm, this equates to £16.92 per participant.

Therapy contacts

All participants received a baseline visit. This was the only contact for those in the control arm and the associated mean total cost (staff time and travel costs) was estimated to be £34.78 per participant (see Table 17 ). (It should be noted that one participant in the control received two intervention visits – no associated intervention costs were assigned to this participant as this was provided in error.) The costs associated with therapist visits to those in the intervention arm are summarised in Table 17 . On average, participants in the intervention arm received a further 6.76 visits (range 0–12). The associated time for these visits was recorded for all but 1 of the 1939 visits, for which the mean time was 96.61 minutes (assuming a travel time of 12 minutes, this equates to an average contact time of 84.61 minutes). This mean value was assumed to apply to the visit where the time was not recorded. Each supervisor meeting was estimated to cost £37.87, one per month per site were estimated to occur across the 18-month period for which the intervention was provided. This equates to a total cost of £10,153.09 across all sites: £5.24 per visit undertaken. Visit, preparation and records’ write-up time were each costed at £31 per hour and, after adding the supervision cost (£5.24 per visit) and travel cost (£3.23 per visit), the mean cost of the intervention was estimated to be £492.92 per participant, where this increased to £509.84 after including the aforementioned training costs. The incremental intervention cost, compared with the baseline visit provided to the control arm (cost £34.78) was thereby estimated to be £475.07 per participant.

Other NHS and Personal Social Services costs

The 6-month resource-use questionnaire was returned by 259 out of 287 participants in the intervention arm and 235 out of 281 in the control arm; the numbers at 12 months were 230 and 209, respectively. However, not all returned questionnaires were fully completed and we accordingly note the response rate to each of the individual questions in Table 18 . (Note: A response was required at both 6 and 12 months in order for the 12-month cost to be estimated.) Mean levels of resource use, relating to each of the cost questions, are shown in Table 18 , in which it can be seen that participants frequently visit health professionals, outpatients and receive home help. The unit costs applied to the reported levels of resource use are summarised in Table 19 . Subsequently, costs were categorised into the following groups: other NHS and PSS costs, total NHS and PSS costs, lost productivity, and overall costs (see Table 18 for details of which questions contributed to each cost category). The mean cost (per participant) for the intervention and control groups, for each of these cost categories, are presented in Table 20 , in which it can be seen that, for each of these cost categories, the mean costs are estimated to be higher for the intervention group. It should be noted, however, that the number of participants for whom complete cost data are available falls when a broader perspective is taken. This can be explained largely by the fact that responses are required to a greater number of resource-use questions.

TABLE 18

Estimated levels of resource use and associated cost (mean per participant over the 12-month period)

TABLE 19

Unit costs attached to different items of resource use, with associated source

TABLE 20

Estimates of the mean cost (£) and QALYs associated with each intervention over the 12-month study period