START (STrAtegies for RelaTives) study: a pragmatic randomised controlled trial to determine the clinical effectiveness and cost-effectiveness of a manual-based coping strategy programme in promoting the mental health of carers of people with dementia

Eligibility criteria for participants

We included family carers of patients with dementia referred in the previous year who:

• provided emotional or practical support at least weekly

• identified themselves as the primary carer of someone with dementia not living in 24-hour care.

We excluded family carers who:

• were unable to give informed consent to the trial

• were currently taking part in a RCT in their capacity as a family carer (not just as an informant)

• lived more than 1.5 hours’ travelling time from the researchers’ base.

To help identify these potential participants, we completed the Mini Mental State Examination28 in carers aged ≥ 60 years. Participants who scored < 24 out of 30 were discussed with the investigators GL or CC to determine if the low score was related to cognition, mood or education. If carers were judged to have dementia, they were not included in the study and the referring clinician was informed. This occurred only once in the study.

Randomisation

The assessing team member gave the details of those enrolled to the trial manager, who entered them into the online randomisation system. Randomisation to intervention or TAU was carried out using an online computer-generated randomisation system. Randomisation was stratified by trust using random permuted blocks. An allocation ratio of 2 : 1 (intervention to TAU) was used to allow for potential clustering effects by therapist in the intervention arm. 29 It was set up and maintained by an independent Clinical Trials Unit at the Institute of Psychiatry, King’s College London, and accessed only by the START trial manager and supervising clinical psychologist.

Blinding

Outcome assessors were blinded to randomisation status, but it was not possible to blind study participants. The researchers worked in two teams, each assessing outcomes for approximately half the participants and providing therapy to those allocated to treatment in the half of participants they were not assessing. These teams were in different rooms and had supervision separately so as to remain blinded. Assessors asked participants at the beginning of each interview not to disclose their allocation group.

The trial manager told the intervention team (B if the assessment team was A or vice versa) the result, but the assessment team was not informed of the outcome of randomisation. The participant was then telephoned by a member of the intervention team and informed that they had been allocated either to TAU and would be contacted for a 4-month follow-up (by the assessment team A, if the therapy team was A or vice versa) or to the intervention. If allocation was to therapy, an appointment was made for the first session of the therapy to start. Allocation within the individual team was according to workload.

Location of interventions

Participants were usually seen for assessment and for intervention in their own homes, although, if participants so requested, they were seen in other settings. In the intervention group, most participants were seen at home, but a minority were seen at UCL (n = 7) or the NHS trust’s offices (n = 13). One carer was seen both at work and at a local restaurant and another both at home and in the trust.

Development of the intervention manual

Dolores Gallagher-Thompson gave us permission to adapt the ‘Coping with Caregiving’ manual, a 13-session group therapy. We therefore used it to develop the START manual, an eight-session manual-based individual therapy programme adapted for UK use with individual family carers. The therapy was adapted by the project team, GL, PR, CC and DL, by considering UK usage of language and culture, brevity, and the fact that we were addressing individuals rather than a group.

We began by familiarising ourselves with the structure and content of the manual, which made use of the stress appraisal and coping response model. In this, stress is seen as the mismatch between primary appraisal (perceived demand) and secondary appraisal (perceived ability to cope and available resources and options, and principles from CBT).

Training and delivery

We employed and trained psychology graduates with no clinical training to deliver the intervention. Researchers in the two groups carried out assessments on carers having received training in taking informed consent, good clinical practice and all the assessment tools being administered. There was a strong practical focus in the training programme on how to deliver the therapy, potential clinical dilemmas, working with interpreters, empathic listening skills, effective use of supervision, safe working practice and when to ask for help. Throughout the training, a strong emphasis was placed upon the researchers guiding carers to where they themselves could get answers to their concerns or questions rather than feeling the need to provide specialised advice themselves within the session. We trained therapists to adhere to the manual and required them to demonstrate, by role-play, competence in delivering each session of the intervention.

Camden and Islington NHS Foundation Trust

Patients and carers in this service were referred by a general practitioner (GP) to the Memory Service for assessment and diagnosis, assessed at home with family carer and, after investigations, offered a diagnostic appointment. Medication was prescribed as was appropriate and, if on anti-dementia medication, followed up by memory clinic nurses. Information and education in dementia was offered initially by nurses but then by dementia advisors situated in offices in and managed by the Alzheimer’s Society. Patients were offered cognitive stimulation therapy. Risk assessments were carried out and risk plans put in place, for example telecare, driving information to the Driver and Vehicle Registry Agency (DVLA), medical identification (ID) bracelets, advice regarding power of attorney and capacity assessment, and social services referral for personal care, day centre and financial advice. Neuropsychiatric symptoms in patients were assessed and managed.

North East London Foundation Trust

We recruited from NELFT’s Admiral Nurse Service. Admiral Nurses are mental health nurses who specialise in dementia and work with family carers and people with dementia in the community and other settings. Family carers are referred to the Admiral Nursing Service through self-referral, directly from the GP, or from community mental health teams and the Memory Service. They assess the needs of both the carer and the person with dementia, and provide information, education and advice about caring for someone with dementia. Admiral Nurses focus on the needs of the family carer and provide emotional and psychological support to families through the transitional phases of the illness, such as diagnosis, when the condition advances, care home placement and end-of-life care. Referrals may be made to other services and they liaise with other health and social care professionals on behalf of the family.

North Essex Partnership University NHS Foundation Trust

Patients and carers in this service were referred by a GP to the Memory Service for assessment and diagnosis, assessed at home with family carer and, after investigations, offered a diagnostic appointment. The diagnosis was given in the clinic with a doctor or at home with nurses. Other interventions included medication as appropriate and, if on antidementia medication, follow-up by memory clinic nurses; risk assessment with a risk plan put in place, for example telecare and driving information to the DVLA; medical ID bracelet; and social services referral for personal care or day centre. Information and education in dementia was offered initially by nurses but thereafter by dementia advisors. Neuropsychiatric symptoms were assessed and managed. Patients were referred to dementia advisors situated in the Alzheimer’s Society and carers were offered support groups.

University College London Hospitals NHS Foundation Trust

Patients were referred by their local GP, memory clinic or specialist to this cognitive disorders clinic. They were assessed in the clinic, with investigations completed during the visit. They were admitted as day case for cerebrospinal fluid examination or for other investigations. A few weeks later they had a diagnostic appointment with a neurologist, when they received the results of any tests, and on the same day saw a nurse consultant for discussion. In this discussion they were given information, education around legal and financial implications and a risk assessment was carried out. All patients were offered the opportunity to take part in clinical trials and other research. Available patient support groups were highlighted to them. This was followed up by telephone consultations. If appropriate, patients returned to the nurse-led therapeutics clinic for medication. Complex cases were followed up in the clinic and all others signposted back to their local services.

Barnet, Enfield and Haringey Mental Health Trust

We did not recruit participants from this trust as they did not agree to extra treatment costs.

Secondary outcomes

Short term (up to 8 months’ follow-up):

1. depression and anxiety caseness and scores on the HADS32

2. carer (HSQ mental health45) and care recipient (QoL-AD43) QoL

3. MCTS score. 36,37

Long term (12–24 months’ follow-up):

1. time to entry of the person with dementia to 24-hour care

2. depression and anxiety caseness and scores on the HADS32

3. carer (HSQ mental health45) and care recipient (QoL-AD43) QoL

4. MCTS score.

Scoring questionnaires

The relevant outcome scores were calculated from the individual items for each instrument according to standard algorithms. Where individual items were missing, standard procedures (where available) for the calculation of these outcome scores were used. Single missing data items on the HADS were imputed using the subscale mean, but cases with more than one missing item in any given subscale were excluded as invalid. 47 Missing cells on the Brief COPE scale were imputed using the carer’s own mean score for that particular subscale if there was one item missing (Charles Carver, University of Miami, 2011, personal communication). For the QoL-AD, if one or two items were missing, the mean of the completed items was used to impute the missing items. For all other outcomes, scores were calculated only where all relevant items had been completed.

Clinical outcomes

Separate regression analyses were used to estimate group differences in HADS-T score over the short term (using 4- and 8-month follow-ups) and the longer term (using 12- and 24-month follow-ups). In both cases, random-effects models accounted for repeated measurements and therapist clustering in the intervention arm. Adjustments were made for baseline HADS-T scores and centre (on which randomisation was stratified), and also on factors believed from the literature to affect affective symptoms (carer age, sex, carer burden and care recipient neuropsychiatric symptoms). All analyses were carried out by intention to treat, but we excluded. in the short term. carers who had data missing at both 4- and 8-month follow-up and, in the longer term, those with data missing at the 12- and 24-month follow-up. If an individual’s data were available at 4 months but not at 8 months, or vice versa, their partial data were used in the short-term analysis, and similarly with the 12- and 24-month data for the long-term modelling.

Similar approaches were taken for analyses of the secondary outcomes. Random-effects logistic regression was used for binary outcomes. In the short term, entry of the person with dementia to 24-hour care was compared between groups using a simple comparison of proportions (not allowing for clustering) because of small numbers. For the long-term analyses, the effect of the intervention on the time until institutionalisation was examined using parametric shared frailty models to allow for clustering in the intervention arm and adjustment for baseline factors. (The models fitted had a gamma distribution for the shared frailty and a Weibull survival distribution for time until institutionalisation.)

Model assumptions were examined by checking the normality of the residuals and also plotting the residuals compared with the fitted values both at the individual and therapist cluster level. To quantify therapist clustering, unadjusted ICCs at each follow-up within the intervention group were estimated with 95% confidence intervals (CIs).

Sensitivity analyses were used to reanalyse the outcomes in various ways to assess robustness of our conclusions. These analyses considered adjustment for imbalances in baseline characteristics between the randomised groups and the differential effects of treatment over time (treatment by time interaction). In considering missing data, we examined the extent to which missing outcome varied by baseline characteristics using logistic regression. In these analyses, for each outcome a binary variable was created according to whether or not it was missing (1 = missing score, 0 = valid score). This binary outcome was used in logistic regression to determine associations with baseline factors. In a sensitivity analysis, the main analyses for each outcome were repeated adjusting for those factors found to be associated with missingness, thus making missing at random assumptions more plausible. In the longer-term analyses, we also fitted models incorporating all repeated measurements (at 4, 8, 12 and 24 months) and examined the interaction between treatment group and long-/short-term follow-up.

Cost-effectiveness

The primary economic evaluation was a cost-effectiveness analysis comparing differences in treatment costs for carers receiving the START intervention with QALYs computed from the EQ-5D40 and societal weights48 over 1–8 months’ and then 1–24 months’ follow-up. In a secondary analysis, differences in treatment costs were compared with changes in in HADS-T score over the same time periods. Both the primary and secondary economic evaluations were undertaken from the perspective of health and social care agencies.

Data on services used and support received by the carer and the person with dementia were collected using an adapted version of the CSRI as above at baseline (randomisation) and at 4, 8, 12 and 24 months. On each occasion, the carer was asked to report service use over the previous 4 months. For the primary economic analyses over 1–8 months and then 1–24 months, our focus was on service use by the carer.

Total carer-related costs were derived by combining health and social care service use data with estimated unit costs. Costs were calculated for the periods 1–8 months and 1–24 months. We did not collect service use data during the period between the 12- and 21-month data collection time points (on average 8 months). Costs during this period were interpolated from the 12- and 24-month data.

Unit costs were obtained from publicly available sources and set at 2009–10 prices: Department of Health National Schedule of Reference Costs 2008–949 for inpatient (accident and emergency) and outpatient attendances; the Personal Social Services Research Unit (PSSRU) unit cost compendium50 for some inpatient services such as acute adult inpatient services; most other community-based health and social care professional services; and voluntary organisation services for a small number of services used by a few carers. Unit costs of services are listed in Appendix 1.

The cost of the START intervention was calculated using data on time spent by therapists in training and supervision with a clinical psychologist and contacts that therapists had with carers in delivering the intervention. Cost per hour of contact for therapists and supervising clinical psychologist were based on figures in the PSSRU compendium,50 taking the mid-point of the relevant scales and including employer costs (national insurance and superannuation contributions) and appropriate overheads (capital, administration and managerial, including recruitment costs). We added costs for the relaxation CDs based on the market rates for copying and delivering.

We analysed HADS-T and QALY differences between START and TAU interventions using a multilevel, mixed-effects model to account for therapist clustering in the intervention arm and repeated measures at 4, 8, 12 and 24 months. For the HADS-T analysis, we adjusted for baseline HADS-T score, centre and carer age and sex, carer burden (Zarit) and neuropsychiatric symptoms (NPI) of the person with dementia. For the QALY analyses, we adjusted for the same baseline variables, except substituting QALY for HADS-T.

Differences in total health and social care cost between the START and TAU interventions were regressed on treatment allocation, baseline costs, centre, carer age and sex, carer burden (Zarit) and care recipient neuropsychiatric symptoms (NPI). We used a linear multilevel regression model to account for therapist clustering in the intervention arm and repeated measures for each individual. Non-parametric bootstrapping was used to estimate 95% CIs for mean costs. Significance (p < 0.05) was judged where the bias-corrected 95% CIs of between-group change score excluded zero.

All analyses were conducted on an intention-to-treat basis; carers were excluded if data were missing at both 4 and 8 months for the 1–8-month economic analysis, and, for the 1–24-month economic analysis, carers were excluded if data were missing at 4, 8 and 24 months. Because the estimation of QALYs requires data at each time point, only complete cases were included in the cost-effectiveness analysis. No imputation was conducted.

Each incremental cost-effectiveness ratio (ICER) was calculated as the difference in the cost of the START and TAU interventions divided by the difference in outcome (measured by HADS-T score or QALYs). Cost-effectiveness acceptability curves (CEACs) were plotted to locate the findings of the economic evaluation in their wider decision-making context. The CEAC illustrates the probability that the START intervention would be seen as cost-effective compared with TAU across a range of hypothesised values placed on incremental outcome improvements (willingness to pay by health and social care system decision-makers). Each CEAC was derived using a net benefit approach. Monetary values of incremental effects and incremental costs for each case were combined and net monetary benefit (NMB) derived as:

where NMB was net monetary benefit, λ is willingness to pay for a 1-point difference in the outcome measure (HADS-T score or QALYs) and subscripts a and b denote the TAU and START interventions, respectively. We explored a range of λ values for each outcome. We were able to account for sampling uncertainty and make adjustments as necessary in the primary analyses and sensitivity analyses.

For the cost-effectiveness analyses carried out at 24 months’ follow-up, costs and outcomes in the second year were discounted at a rate of 3.5%, as recommended by NICE. 51 The discount rate was varied from 0% to 6% to assess the impact changing the NICE-recommended rate had on costs, outcome and the ICER.

Two sensitivity analyses were conducted to assess the robustness of our results. The first sensitivity analysis adjusted for those variables which predicted missing outcomes, HADS-T score and QALYs, and was investigated separately for each outcome using logistic regression. The first step was to model a binary variable (missing vs. not missing) in bivariate logistic regression with each baseline demographic variable. Those variables identified as significantly associated with missing were then used in multivariate logistic regression to determine which remained significant. The main analyses were then repeated, adjusting for those factors which were found to be associated with ‘missingness’ on each outcome. For the analysis of HADS-T scores, the variables found to be associated with ‘missingness’ were patient living with carer, relationship to carer, carer having children at home, patient ethnicity and COPE dysfunction score. For the QALY outcome, the carer’s work situation (employed vs. unemployed) and ethnicity were associated with ‘missingness’.

A second sensitivity analysis adjusted for imbalances in baseline characteristics between the treatment groups that occurred despite randomisation [i.e. adjusting for carers’ work situation, relationship to patient, and patients’ and carers’ education and living situation (coresident or living separately)]. These analyses were chosen so as to be consistent with those used in the effectiveness analysis. 41

We also plotted the CIs around the NMB to estimate the impact of uncertainty.

All economic analyses were carried out using Stata, version 12 (StataCorp LP, Chicago, IL, USA).

Therapists

We trained 10 psychology graduates with no further clinical training – seven women and three men – who were ethnically white British (eight) and black and minority ethnic (two) in the age range 23–33 years. They worked with between 11 and 32 participants each (mean 17.3 participants) (Table 4).

Adherence with therapy

Of the eight therapy sessions offered, five or more were attended by 130 (75.1%) carers in the intervention group (Figure 2 shows details of number of sessions attended). Seven (4.0%) of those in the intervention group withdrew before taking part in any therapy sessions. Adherence (attending five or more sessions) was better in those of white ethnicity [110 (78.0%) vs. 19 (61.3%)] than in those of other ethnicity and slightly better for male carers [46 (80.7%) vs. 84 (72.4%) of female carers] and those with at least A-level education [56 (80.0%) vs. 74 (71.8%)]. Adherence was similar by age group [aged < 60–75 years (77.3%) vs. 55 years (73.3%)] and work situation [in paid work 49 (77.8%) vs. other 81 (73.6%)].

FIGURE 2.

Number of sessions completed by participants.

Fidelity

Double rating was carried out for the first 13 participants; these had a Cohen’s kappa value of 0.77, which is substantial agreement. 52 After this, we judged it was necessary to have only a single rating for the remaining participants. Fidelity rating was carried out for 128 (78%) of the 165 participants who received one or more intervention(s). The remaining 38 either refused to be audiotaped (10) or withdrew before carrying out the session which had been randomly selected for assessment (n = 28). For 100 (78%) participants fidelity was rated as 5, for 20 (16%) it was rated as 4, for five it was rated as 3 and for three it was rated as 2. Overall mean fidelity was calculated by adding each average rating (rating 1 plus rating 2, divided by 2) of the first 13 participants to the ratings of the remaining 105 participants, and dividing by the overall number of participants with available fidelity ratings. The mean fidelity score was 4.7 (SD 0.66).

Languages, translation and interpreters

Four eligible participants used translators at the recruitment stage, in Turkish (one), Bengali (two) and Farsi (one), and all four consented and were randomised: three to intervention and one to TAU. One participant allocated to intervention changed his or her mind and withdrew from intervention before session 1 but not follow-ups; one completed five sessions and one completed eight sessions.

Missing outcome data

The proportion of missing data for the main trial outcomes is given in Table 5 by group and follow-up time. The primary outcome (HADS-T score) was missing for 35 (13.5%) patients at 4 months and for 56 (21.5%) patients at 8 months, with a slightly higher proportion missing in the intervention group: 40 (23.0%) in the intervention group and 16 (18.0%) in the TAU group. Two hundred and twenty-nine (88.1%) had HADS-T data for at least one of the 4- and 8-month points and, so, could be included in the primary analyses. Using logistic regression models, we identified baseline factors that were associated with missing outcome data.

Missingness of HADS-T score at 4 months was associated with the patient living with carer (p = 0.007) and for the 8-month outcome with having dependent children at home (p = 0.003), patient ethnicity (p = 0.013), patient living with carer (p = 0.002), patient relationship to carer (p = 0.011) and the COPE dysfunction score (p = 0.004).

Primary outcome

Analysis of HADS-T, adjusting for trust and baseline score and for factors related to outcome (carer age and sex, NPI and Zarit scores), showed a mean difference of –1.80 points (95% CI –3.29 to –0.31 points; p = 0.02) in favour of the intervention (Table 6). If the model did not include factors relating to outcome, then the results were similar, with an average decrease in score of –1.46 points (95% CI –2.89 to –0.03 points; p = 0.05). There was little therapist clustering: ICC at 4 months was 0.02 (95% CI 0.00 to 0.09) and at 8 months was 0.00 (95% CI 0.00 to 0.08).

Model assumptions were examined by checking the normality of the residuals and also plotting the residuals compared with the fitted values. These were done at the individual and therapist cluster level. These plots are shown in Figure 3 and do not indicate concerns about the model fit.

FIGURE 3.

Checking model assumptions: residuals and fitted values from the models for HADS-T (based on the model-adjusted only for centre and baseline score). (a) Normality of residuals (individual level); (b) residuals vs. fitted values (individual level); (c) normality of residuals (cluster level); and (d) residuals vs. fitted values (cluster level).

Sensitivity analysis

For the sensitivity analysis we adjusted for baseline factors associated with missingness of the HADS-T outcome. Missingness of the 4-month HADS-T was associated with patient living with carer (p = 0.007) and for the 8-month outcome with having dependent children at home (p = 0.003), patient ethnicity (p = 0.01), patient living with carer (p = 0.002), patient relationship to carer (p = 0.010) and the COPE dysfunction score (p = 0.004). Refitting the main models adjusting for these significant predictors of missing values did not have a significant impact on the results (mean difference –1.53, 95% CI –2.96 to –0.10). We also carried out a sensitivity analysis to adjust for baseline imbalances, namely carer’s work situation, relationship to carer and patient and carer education and living situation, which again gave similar conclusions (mean difference –1.78, 95% CI –3.30 to –0.27). Models including an interaction with time showed no evidence of a differential effect of the intervention between the 4- and 8-month time points (p = 0.90).

Secondary outcomes

Service use and costs

Carers used a wide range of health and social care services over the 8-month period, as can be seen in Table 7. We did not impute individual items of service use and means are presented for non-missing cases only. Outpatient hospital and GP services were used by quite high proportions of participants.

The patterns of service use were weighted by their unit costs. Mean costs – grouped into outpatient, community and other services – are given in Table 8. The table distinguishes between the START and TAU groups, and reports figures for two time periods: the 4-month period between baseline and the 4-month assessment, and the 4-month period between the 4- and 8-month assessments. Although the number of users of outpatient services was higher in the intervention group (see Table 8), their average outpatient service cost was lower as they used outpatient services less frequently than people in the TAU group.

The right-hand column of Table 8 shows the difference in costs between the START and TAU groups across the whole evaluation period of 8 months. Excluding the direct cost of the intervention itself, mean costs over the study period (1–8 months) were £558 in the START group and £625 in the TAU group. After adjustment for baseline characteristics (see above) the standardised difference was £14, with the 95% CI (–£239 to £211) suggesting that there was not a significant difference in costs between the two groups. For purposes of comparison, scores on the two outcome measures used in the economic evaluation are included towards the bottom of Table 8.

The calculation of cost of the therapy is based on the time spent by the 10 therapists in delivering one-to-one therapy to carers, their own training sessions (40 sessions of 2.5 hours over a 6-week period), time spent making telephone calls to participants, time spent writing up notes and supervision of the therapists by the clinical psychologist (1.5 hours each per week for 8 weeks). Looking at the average time spent in each session with the carer, excluding those who did not have any sessions and those who used translators (n = 163), the median time per session was 1 hour 16 minutes (76 minutes). The mean time per session was 1 hour 17 minutes (77 minutes, SD 20 minutes). For those who used translators, the mean time per session was 3 hours 47 minutes (SD 1 hour 36 minutes), with a range of 2 hours 40 minutes to 4 hours 55 minutes. We calculated that the mean cost per session per carer was £36. Adding in the cost of the relaxation CDs (which totalled £284), the overall mean direct intervention cost averaged £232.15 per carer.

Including the cost of intervention itself in the comparison between the groups and adjusting for baseline variables as we did in the clinical effectiveness calculations, costs for the START group were slightly but not significantly higher than for the TAU group. The mean cost difference was £252 (95% CI –£28 to £565) for sample members on the EQ-5D, and £247 (95% CI £0 to £569) for sample members on the HADS-T measure (Table 9).

Cost-effectiveness

Results from the net benefit regression using the two outcomes examined in the economic evaluation (QALYs and HADS-T score) are summarised and the ICERs reported in Table 10. The cost and outcome differences are obtained after adjustment for baseline characteristics and are influenced slightly by size of sample with complete data for each outcome.

Health and social care system costs were slightly higher for carers who received the START intervention (in addition to TAU) but the difference was not statistically significant, and carers in this group enjoyed significantly better outcomes, whether measured in terms of health-related QoL (QALY) or affective symptoms (HADS-T). Whether or not these results imply that START is cost-effective compared with TAU depends on the decision-maker’s willingness to pay for these gains in QoL and affective symptoms. To aid discussion of willingness to pay, we computed the ICERs. We also plotted the associated CEACs and examined the CIs around NMB.

Looking first at QALY as the outcome, the mean cost per QALY gained was £6000. The CEAC is shown in Figure 4, illustrating the probability of cost-effectiveness for each of a number of different hypothesised values of willingness to pay. At the £20,000 per QALY threshold associated with NICE recommendations, the probability that the START intervention would be seen as cost-effective was 93% and at the higher NICE threshold of £30,000 it was 99%. 53

FIGURE 4.

Cost-effectiveness acceptability curve: START intervention (manual-based coping strategy therapy) vs. TAU; health and social care perspective, with effectiveness measured in QALY gain over 8 months.

The 95% CIs around NMB suggest that there is a strong likelihood that the START intervention is cost-effective at the £30,000 threshold.

For the other outcome measure, the HADS-T measure of affective symptoms, the mean cost per 1-point difference on the HADS-T was £118. The CEAC for this outcome measure is shown in Figure 5. We are not aware of any previously suggested monetary thresholds for gauging cost-effectiveness on the HADS. However, if we assumed a willingness to pay of £500, the probability that the START intervention would be seen as cost-effective would be 95%. We can also refer to a previous suggestion that a minimally important clinical difference on the HADS-T is 1.6. The mean cost of achieving such a change with the START intervention would be £189.

FIGURE 5.

Cost-effectiveness acceptability curve: START intervention (manual-based coping strategy therapy) vs. TAU; health and social care perspective, with effectiveness measured in HADS gain over 8 months.

Sensitivity analyses

The first sensitivity analysis adjusted for significant baseline differences on demographic and clinical predictors of missing values. The results were similar to those from the primary analyses and are summarised in the first column of figures in Table 10.

The mean ICER values are now £5452 per additional QALY and £107 per 1-point difference in HADS-T score. Figure 6 shows the CEAC with QALY as the outcome measure; at the lower-bound NICE threshold of £20,000, the START intervention has an approximately 95% likelihood of being seen as cost-effective, rising to 98% at the £30,000 threshold.

FIGURE 6.

Sensitivity analysis of CEAC with QALY as the outcome measure over 8 months.

The second sensitivity analysis adjusted for imbalances in baseline characteristics. The results were again quite similar to those from the primary analyses and are summarised in the second column of figures in Table 11. The mean ICER values from this further analysis are £5756 per additional QALY, and £112 per 1-point difference in HADS-T score.

The CEAC with QALY as the outcome measure is shown in Figure 6. At the lower-bound NICE threshold of £20,000, the START intervention has a 93% likelihood of being seen as cost-effective, rising to 98% at the £30,000 threshold.

The CIs around NMB for these two sensitivity analyses are shown in Appendix 2. Taking into account uncertainty in the estimation suggests that a degree of caution should be exercised in concluding that START is necessarily cost-effective.

Missing outcomes

The proportion of missing data for the main trial outcomes is given in Table 11 by group and follow-up time.

Clinical outcomes

Primary and secondary clinical outcomes are shown in Table 12. This summarises average scores at months 12 and 24 and gives the estimated effect of therapy compared with TAU for primary and secondary outcomes.

Analysis of HADS-T scores, adjusting for trust and baseline score and for factors related to outcome (carer age and sex, NPI and Zarit scores), showed a mean difference of –2.58 points (95% CI –4.26 to –0.90 points; p = 0.003) in favour of the intervention. If the model did not include factors relating to outcome then the results were similar, with an average decrease in score of –1.84 (95% CI –3.50 to –0.17; p = 0.03). The therapist ICC at 12 months was 0.00 (95% CI 0.00 to 0.07) and at 24 months was 0.00 (95% CI 0.00 to 0.07).

Sensitivity analyses adjusting for significant demographic and clinical predictors of missing values, namely patient living with carer and COPE dysfunction score, still gave a result with significant difference between groups (difference in means –2.69, 95% CI –4.39 to –0.98; p = 0.002; n = 200), as did sensitivity analyses adjusting for factors imbalanced at baseline (carer work, carer education, patient education, relationship with carer, lives with carer; difference in means –2.37, 95% CI –4.11 to –0.63, p = 0.008).

Models including an interaction with time showed no evidence of a differential effect of the intervention between the 4- and 8-month time points or between the 12- and 24-month time points for HADS-T score (p = 0.92) and indicated a similar long-term treatment effect.

Model assumptions were examined by checking plots assessing the normality of the residuals and also plotting residuals versus fitted values. These were done for residuals at both the individual and the therapist cluster level. These plots did not indicate concerns about the model fit.

Secondary outcomes

Results of models considering the long-term effect of intervention on secondary outcomes are also given in Table 12.

Time to care home admission

Two hundred and fifty-nine (87 in TAU and 172 in the intervention group) individuals were included in the analysis. We found that one person with dementia was included twice as two different carers (spouse and daughter) had been included and gave different dates of birth of the patient. The individual was included only once and from the first recruitment date (spouse carer’s recruitment). Both carers had been in the intervention group.

For four people (three in TAU and one in the intervention group), the baseline date and the last known at home date were the same; thus, they were censored at time = 0 days.

Of the remaining 84 in the TAU group, 17 (20.2%) were admitted to a care home, and in the intervention group, 32 (18.7%) of the remaining 171 were admitted, within the 24-month follow-up period.

Figure 8 shows a Kaplan–Meier plot for time until admission to a care home for each randomised group.

FIGURE 8.

Time until admission to a care home for each randomised group over 24 months’ follow-up.

A parametric shared frailty model assuming a gamma distribution for the shared frailty with a Weibull survival distribution was fitted for the time to institutionalisation. The model included trust and randomisation group and additionally adjusted for carer age, carer sex, baseline NPI total and baseline Zarit total. This model included 242 people with dementia. The estimated hazard ratio (HR) was 0.83 with 95% CI 0.44 to 1.56 (p = 0.56), showing no significant evidence of an effect of the intervention compared with TAU. Sensitivity analyses fitted the same model with addition adjustment for baseline imbalances (carer work, carer education, patient education, relationship with carer, lives with carer). This gave HR 0.62 (95% CI 0.31 to 1.23; p = 0.17; n = 232).

Modified Conflict Tactics Scale for significant abuse

There was no evidence for a decrease in abusive behaviour (adjusted OR 0.83, 95% CI 0.36 to 1.93) (Figure 9).

FIGURE 9.

Modified Conflict Tactic Scale for significant abuse over 24 months in intervention and control groups.

Service use and costs

Economic data for the 24-month follow-up period were available for 196 carers (75%): 64 in the TAU group and 132 in the intervention group. The number of carers using services in the 24-month follow-up period is detailed in Table 13. As can be seen in Table 13, carers used a wide range of health and social care services. GP services were used by a high proportion of participants.

Mean health and social care costs are reported in Table 14. The table distinguishes the START and TAU groups and reports figures for the 4-month period from 9 to 12 months and the 4-month period from 21 to 24 months. In contrast to the outpatient service costs over the 4-month period between baseline and the 4-month assessment and the 4-month period between the 4- and 8-month assessments, outpatient service costs over 9–12 months and 21–24 months were higher in the intervention group as participants in this group used outpatient services more frequently than those in the TAU group.

Table 14 shows the difference in costs between the START and TAU groups across the whole evaluation period of 24 months. To calculate costs over 24 months, we summed costs over the study periods 1–8 months, 9–12 months, 13–20 months and 21–24 months. Costs during the period between the 12- and 21-month data collection time points (on average 8 months) were interpolated from 12- and 24-month data. Mean costs over the study period (1–24 months) were £1492 in the START group and £1578 in the TAU group. After adjustment of baseline characteristics, the standardised difference was £173, with the 95% CI (–£115 to £460) suggesting that there was no significant difference in costs between the two groups. The scores on the two outcome measures used in the economic evaluation are included at the bottom of Table 14.

Mean costs (including the direct costs of intervention and health and social care system costs) for the 24-month period for carers in the START group were higher, although not significantly so, than for carers in the TAU group. The mean cost difference was £336 (95% CI –£223 to £895) for sample members on the EQ-5D and £303 (95% CI –£206 to £812) for sample members on the HADS-T measure (Table 15).

Cost-effectiveness ratio

Results from the net benefit regression using the two outcomes examined in the economic evaluation (QALYs and HADS-T score) over 1–24 months are summarised and the ICERs reported in Table 15. Similar to the analyses conducted over the 8-month time period, the costs and outcome differences are obtained after adjustment for baseline characteristics and are influenced by the sample with complete data for each outcome.

Over the 24 months, carers who received the START intervention had slightly, but not statistically significantly, higher total costs and better, although not statistically significant, differences in outcomes whether measured in terms of health-related QoL (QALY) or affective symptoms (HADS-T). We explored the decision-maker’s willingness to pay for gains in QoL and affective symptoms to assess whether or not these results imply that the START intervention is cost-effective. We computed the ICERs and plotted the associated CEACs, and examined the CIs around NMB.

Looking at QALY as the outcome, the mean cost per QALY gain was £11,200. The CEAC is shown in Figure 10. At the £20,000 per QALY threshold associated with NICE recommendations, the probability that the START intervention would be seen as cost-effective was 65%, and at the higher NICE threshold of £30,000 it was 75%.

FIGURE 10.

Cost-effectiveness acceptability curve: START intervention (manual-based coping strategy therapy) vs. TAU; health and social care perspective, with effectiveness measured in QALY gain over 24 months.

For the other outcome measure, the HADS-T measure of affective symptoms, mean cost per 1-point difference on the HADS-T was £199. The CEAC threshold for this outcome measure is shown in Figure 11. Assuming a willingness to pay of £500, the probability that the START intervention would be seen as cost-effective would be 78%.

FIGURE 11.

Cost-effectiveness acceptability curve: START intervention (manual-based coping strategy therapy) vs. TAU; health and social care perspective, with effectiveness measured in HADS gain over 24 months.

The discounted incremental costs and outcomes at 24 months, at rates of 0% and 6%, are shown in Appendix 1. Discounting the incremental cost and outcomes by 0% and 6% did not greatly alter the findings.

Sensitivity analysis

We followed the same approach as in the 8-month analysis to assess the robustness of the findings of the cost-effectiveness analysis at 24 months. In the first sensitivity analysis, we adjusted for significant differences on demographic and clinical predictors of missing values. This analysis did not alter the results and is summarised in Table 16. The mean ICER values from this analysis are £9767 per additional QALY and £125 per 1-point difference in HADS-T score.

Figure 12 shows the CEAC with QALY as the measure of outcome. The START intervention has an approximately 67% likelihood of being seen as cost-effective at the lower bound of NICE threshold of £20,000, rising to 75% at the £30,000 threshold.

FIGURE 12.

Cost-effectiveness acceptability curve for START with QALY as the outcome measure over 24 months.

Figure 13 shows the CEAC with HADS-T score as the measure of outcome. The START intervention has an approximately 89% likelihood of being seen as cost-effective at £500. The likelihood of cost-effectiveness is above 95% if the willingness to pay for a 1-point improvement in outcome is > £1000.

FIGURE 13.

Cost-effectiveness acceptability curve for START with HADS-T as the outcome measure over 24 months.

In the second sensitivity analysis adjusted for imbalances in baseline characteristics, the findings were similar to those from the primary analyses. The results are shown alongside those of the first sensitivity analysis in Table 16. The mean ICER values from this second sensitivity analysis are £8567 per additional QALY and £209 per 1-point difference in HADS-T score.

The CEAC with QALY as the outcome measure is shown in Figure 14. At a threshold of £20,000, the START intervention has a likelihood of being seen as cost-effective of 70%, rising to 80% at the £30,000 threshold.

FIGURE 14.

Sensitivity analysis of CEAC with QALY as the outcome measure over 24 months.

Figure 15 shows the CEAC with HADS-T score as the measure of outcome. This second sensitivity analysis adjusting for factors associated with imbalances in baseline characteristics slightly increased the likelihood that the START intervention would be seen as cost-effective. In this sensitivity analysis, the START intervention has an approximately 72% likelihood of being seen as cost-effective at £500 and remained above 82% if the willingness to pay for a 1-point improvement in outcome is over £1000.

FIGURE 15.

Sensitivity analysis of CEAC with HADS-T as the outcome measure over 24 months.

The CIs around NMB for these two sensitivity analyses are shown in Appendix 2.

Primary and secondary outcome of clinical effectiveness

This study was a pragmatic RCT of a multicomponent intervention. It is the first study to show that family carers of people with dementia referred to secondary or tertiary care benefit from a structured psychological intervention delivered by psychology graduates and supervised by a clinical psychologist. This change was significant in terms of both depression and anxiety symptoms and also in QoL and depression caseness. The intervention was effective 8 months after randomisation and continued to work 2 years later. Abusive behaviour reduced in both groups and there was no significant difference between them. There was an indication of possible reduction in the chances of being admitted to a care home but this was not significant.

Primary and secondary outcome of cost-effectiveness

We examined whether or not eight sessions of manual-based coping strategy therapy delivered over 8–14 weeks by supervised psychology graduates to family carers of people with dementia added to TAU was cost-effective compared with TAU alone. Over the 8-month evaluation period, the START intervention was found to have a high probability of being seen as cost-effective by reference to both primary outcome measures examined, i.e. improvements in carers’ affective symptoms and gains in carers’ health-related QoL. The sensitivity analyses considering alternative approaches to the analysis suggest a more cautious conclusion as to the cost-effectiveness of the START intervention.

Interpretation

The effect size in terms of the total mean affective symptoms was small, but previous evidence from studies in which researchers set out to calculate what a clinically important difference in the HADS would be suggests that treatment effects are in the range that is important to patients. 54 This study, based on consideration of an emotional function and mastery domain from a scale administered at the same times as the HADS, found the minimally important difference in the HADS-T score to be 1.61 and 1.68 for the former and latter domain, respectively.

Incidence of clinical depression increased in the control group but not in the intervention group, and the ORs indicate that, at follow-up at 8 months, those in the TAU group were four times more likely, and at 24 months were five times more likely, to have clinically significant depression, suggesting that the intervention is clinically important. In keeping with this, carer QoL improved.

We thought that in the long term this intervention may also delay care home admission of the person with dementia and, therefore, increase his or her QoL. The short-term follow-up over 8 months did not show this but at 2 years there was an indication that it may tend to delay care home admission and we will reconsider this effect over the following 5 years. Previous studies of information and advice to people with dementia and their families have suggested this can improve the patient’s QoL but not the carer’s psychological symptoms and an intervention which is targeted at the carer may not help the person with dementia to the same extent. 55

This may be the explanation for the lack of effect on care recipients’ QoL. However, QoL is a complex phenomenon and the measure we used is a global one, thus considering finance, housing and physical health in addition to well-being, and this may have made it less likely to change.

The cost-effectiveness finding is driven more by the outcome differences between the groups than by the cost difference; for example, at the 8-month time point, carers in the control group were four times more likely to have clinically significant depression than carers receiving the START intervention.

Our study was effective over time and acceptable to most participants, who made time and space for it despite the commitments of being a carer and, frequently, also being employed or being unwell themselves. There is little evidence of harm with withdrawal from the treatment being at a similar rate to withdrawal from TAU arm, although one carer said that he found the therapy too upsetting and three felt that it was not for them.

There are currently no interventions which have been shown to reduce elder abuse. 56,57 Our study was not powered to find a significant change in abuse and for ethical reasons we made clinicians aware of clinically significant abusive behaviour in the control group; thus, abused carers in this group were often offered clinical and social support as well as monitoring of the behaviour and, if felt appropriate, adult protection measures were taken. Thus, we think it is unlikely that there would be a significant difference between groups. We were pleased that abusive behaviour went down over time in both groups and previous research has not suggested that this is what would happen without intervention. 58 We do not know whether those people who acted in an abusive way were more likely to refuse to participate in the study (although the high rates of abusive behaviour even in relation to prevalence studies make this less likely)59 or to leave the study or if the care recipient was more likely to be admitted to a care home.

Study design

This was designed as a pragmatic study with broad inclusion criteria and followed closely the Medical Research Council guidance on the evaluation of a complex intervention. The clustering of therapists was accounted for by the analysis, and outcome measures were carefully considered for validity and to ensure the participants were not overburdened and an economic evaluation was conducted. The follow-up rates were satisfactory at > 80% and sensitivity analyses adjusting for missing outcome data showed similar results.

Sensitivity analyses

One of the strengths of the study is the sensitivity analyses, which showed the same results as for the primary analysis.

Bias

We took care to minimise bias in several ways. We used standardised measures validated for the population. Although randomisation was independent and follow-up raters were blinded to allocation to avoid bias in allocation or in rater assessment, the family carers inevitably knew which group they were in. Similarly, although the data were entered and cleaned while blind to intervention status, the statistical team were aware that the intervention group was the larger group.

Health economics

Each carer recruited to the study was scheduled to have eight sessions with the therapist but some carers had fewer sessions. Although this was taken into account in calculating the costs of the intervention, the impact that different numbers of sessions might have had on carers’ outcomes was not the focus of this study.

The evaluation was conducted from a health and social care perspective, and concentrated on outcomes experienced by carers. We did not, therefore, measure the costs of treatment and care services used by the individuals with dementia who were being supported by these carers, nor did we attach monetary values to the time spent by carers in providing support to their relatives.

Sample size for the study design was calculated on the basis of the power required to demonstrate differences in one of the effectiveness measures and not on the basis of costs or cost-effectiveness. Although it would have been preferable for the study also to have been powered on an economic variable, this would have required a considerable increase in sample size given that economic measures tend to be highly skewed. In turn this would have had implications for both the research budget and ethics, as it would have been necessary to recruit participants beyond the point at which clinical dominance has been determined. We used CEACs to represents the uncertainty in the estimation of the ICER.

Generalisability

We recruited a sample of carers with varying demographic and clinical characteristics from a range of services. People were only excluded if they had dementia or lived too far away for the therapy to be delivered. Further evidence of external validity is the demographic similarity between those who consented and those who did not. Participants were recruited from three mental health trusts and from a neurology centre for rare dementias and were from urban, suburban and rural areas. The trust differed in geographical location (Essex and London) and in referrers (psychiatrists, neurologists, Admiral Nurses and other mental health professionals).

Participants were in the main children (44%) or partners (42%) of the person with dementia but were also in-laws, nieces and nephews, grandchildren, siblings and friends. The participants were of either sex (31% male), throughout the adult age range (18–89 years) and from a range of ethnic (around three-quarters white UK) and socioeconomic backgrounds. We had thought that those carers in paid work might be unable to access the intervention, but this did not appear to be the case, and carers were working full-time or part-time, or were retired or not currently working.

The carers’ scores on the outcome measures and burden at baseline ranged widely. Most of the people with dementia had relatively mild dementia at baseline, with the mean clinical dementia scale rating being 1.5–1.6, with 1 meaning mild and 2 meaning moderate dementia. Thus, we think that the intervention is generally acceptable to and the results are generally applicable to the family carers around the time people present to secondary care with dementia in the UK. The levels of anxiety and depressive symptoms, case-level anxiety and depression, neuropsychiatric symptoms and carer abusive behaviour, however, were slightly higher than in a recent cohort study of newly referred people with dementia and so those with more problems may have been more likely to consent to the study. 9,60

We did, however, inform the clinical teams about abusive behaviour of the carers in the TAU group when there was no intervention in place and, thus, may have improved the outcome for the TAU.

Structured psychological treatment of depression in other groups

A recent structured intervention in family carers of people with stroke with care competencies taught and assessed by nurses did not improve their mood, QoL or burden, or improve patient outcome. 15 This is further evidence that knowledge by itself is not efficacious.

Previous work has found that a RCT of telephone-administered psychotherapy for depressive symptoms worked over the 16 weeks it was administered, but was no longer significantly efficacious at 1 year. 68 Similarly, a RCT of a ‘coping with depression’ psychoeducational group programme, with booster session up to 4 months after the intervention had finished, was effective in reducing depression at 6 months but not at 12 months. 69

In contrast, a RCT of CBT as an adjuvant therapy for treatment-resistant depression found that it was effective at the 6- and 12-month follow-up. 70 Ten therapist-delivered sessions of CBT over the internet were effective in reducing depression at 4 and 8 months (equivalent to our short-term measures). 71 CBT treatment for health anxiety was effective at 6 months, 1 and 2 years with no significant increase in costs for anxiety and depression, but was ineffective in increasing health-related QoL. 72

We are unsure what the difference is between the therapies that continue to be efficacious for affective symptoms and those that do not, but our intervention adds to the evidence that therapies can continue to be effective over years.

Comparison with other health economics studies

There is little previous evidence on the cost-effectiveness of psychosocial interventions for carers of people with dementia. A recent review found some evidence that such interventions could lead to greater improvements in outcome and also generate cost savings. 20 However, only one of the studies covered by that review employed a similar therapeutic approach to the START intervention. 21 It examined the cost-effectiveness of a modular multicomponent intervention delivered in carers’ homes, with three sessions by telephone, supplemented by five group sessions (five or six carers in each) delivered by telephone. Focusing on hours of care-giving, the authors found a significant difference over the 6-month study period, with carers in the intervention group having more time to dedicate to activities unrelated to caregiving, which has potentially positive impacts on emotional well-being and QoL.