Research setting

We have published the details of the HTA-SADD trial method and results elsewhere. ^{Reference Banerjee, Hellier, Dewey, Romeo, Ballard and Baldwin9} To summarise, the trial involved people with probable or possible dementia of the Alzheimer's type according to NINCDS-ADRDA (National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association) criteria ^{Reference McKhann, Drachman, Folstein, Katzman, Price and Stadlan10} and depression (defined as a Cornell Scale for Depression in Dementia (CSDD) ^{Reference Alexopoulos, Abrams, Young and Shamoian11} score of over 7) referred to old age psychiatric services in nine English sites: Birmingham, Cambridge, Leicester, Liverpool, Manchester, Newcastle, North London, Southampton and South London. Ethical approval was obtained from the North West 7 (Greater Manchester) Ethics Committee.

Sample size

A sample size of 507 was calculated to provide 90% power to detect a two-point CSDD difference (s.d. = 5; standardised effect-size (SES) = 0.4) for 13-week sertraline/placebo and mirtazapine/placebo comparisons, and 86% power at 39 weeks. This was revised during the trial because of slower than forecasted recruitment to 339. ^{Reference Banerjee, Hellier, Dewey, Romeo, Ballard and Baldwin9}

Randomisation

After baseline assessment and consent, participants were randomised to three groups: sertraline, mirtazapine and placebo (trial registration: ISRCTN88882979 and EudraCT 2006-000105-38). All participants also received normal clinical care. The target doses were 150 mg sertraline or 45 mg mirtazapine daily with titration over 8 weeks. Thereafter it was open to clinicians to adjust the dose. The Clinical Trials Unit at the Institute of Psychiatry independently undertook treatment allocation.

Economic evaluation

The primary economic evaluation was a cost-effectiveness analysis comparing differences in treatment costs for patients receiving sertraline, mirtazapine or placebo with differences in effectiveness as measured by the primary outcome, total CSDD score ^{Reference Alexopoulos, Abrams, Young and Shamoian11} over two time periods: 0–13 weeks and 0–39 weeks. The secondary analysis was a cost-utility analysis using quality-adjusted life years (QALYs) computed from the Euro-Qual (EQ-5D) ¹² and societal weights (obtained from data from a sample of the general public to aggregate items into an overall EQ-5D score) ^{Reference Dolan, Gudex, Kind and Williams13} over those same periods. Both the primary and secondary economic evaluations were undertaken from two perspectives: (a) health and social care agencies; and (b) health and social care agencies and unpaid carers.

Resource use

Resource-use data for each person were collected over a retrospective period of 6 months before randomisation. At 13 weeks, follow-up data were collected retrospectively for a 3-month period and at 39 weeks for a retrospective period of 6 months. Services and support received by the study participants were recorded on a resource-use questionnaire adapted from the Client Service Receipt Inventory (CSRI), ^{Reference Beecham, Knapp and Thornicroft14} including in-patient stays, out-patient attendances, day hospital treatment, visits to social clubs, meals at lunch clubs, day care visits and hours spent in contact with community-based professionals. The study also collected data on volunteer support, befriending and telephone care-line support, and also on unpaid support provided by friends and relatives. Contacts made with voluntary workers and support provided by friends and relatives were also measured in hours of care support time. The prescribed daily doses for the medications were calculated from the trial medication log, and prescribing periods were weighted to the changing dose regime.

Unit costs

All unit costs were estimated at 2009/2010 prices and were collected from sources in the public domain (Table 1). Costs per unit of measurement for each type of service (such as per in-patient day, per appointment, per attendance, per visit or per contact with health and community-based professionals including voluntary services) were taken from a widely used compendium. ^{Reference Curtis15} The National Health Service Schedule of Reference Costs was used to estimate the cost of out-patient attendances. ¹⁶ The unit cost of medication was obtained from the British National Formulary. ¹⁷

We collected information from each study participant's main carer on time they spent providing unpaid care and support, and time spent by friends or relatives regularly providing help for the trial participant. We asked respondents to estimate the hours of unpaid care and support from all such sources in an average or typical week. Opportunity costs were attached to these hours using, first, an estimate of replacement cost (the unit cost of a paid local authority home care worker ^{Reference Curtis15} ) and, second, the cost of lost employment (gross hourly wage of a carer in paid employment; zero for a carer not in paid employment).

Cost estimation

Data on resource use from the CSRI and medication prescribed from the medication logs were combined with unit cost data to estimate total costs for each participant in the trial. Three main categories of costs were analysed: medication costs, aggregated health and social care costs (primary care and hospital-based visits and community-based contacts) and (opportunity) cost of time spent caregiving by relatives and friends. Costs were categorised in this way to facilitate comparison of costs alongside measures of effectiveness from the various study perspectives. The costs of services and support used by patients were derived by combining medication, health and social care resource utilisation data with unit costs. Costs were calculated for the periods 0–13 weeks and 0–39 weeks.

Table 1 Unit cost for 2009–2010

Service	Unit cost (£)	Source
In-patient (bed days)	299	Reference Curtis15 , Reference Banerjee, Murray, Foley, Atkins L Schneider and Mann26
Day hospital (attendance)	50–205	Reference Curtis 15 , 16 , Reference Banerjee, Murray, Foley, Atkins L Schneider and Mann 26
Out-patient (appointment)	21–165	16
Accident and emergency (attendance)	37–97	Reference Curtis15
General practitioner (per surgery consultation)	28	Reference Curtis15
Geriatrician (min)	1.83	Reference Curtis15
Nurse (min) a	0.43–0.52	Reference Curtis15
Occupational therapist (min)	0.65	Reference Curtis15
Community psychiatrist (min)	1.83	Reference Curtis15
Counsellor (min)	0.57	Reference Curtis15
Psychologist (min)	1.20	Reference Curtis15
Chiropodist (contact)	0.37	Reference Curtis15
Social worker (min)	0.67	Reference Curtis15
Care manager (min)	0.82	Reference Curtis15
Home care worker/care attendant (min)	0.35	Reference Curtis15
Sitting scheme (min)	0.45	Reference Curtis15
Self-help group (min)	0.57	Reference Curtis15
Meals on wheels (meal)	4.8	b
Dentist (min)	2.90	16
Optician (min)	0.48	c
Day care (day)	42–66	Reference Curtis15
Lunch club (meal)	7	d
Social club (session)	5	e

a Practice nurse, district nurse health visitor, community psychiatric nurse, cardiac nurse, incontinence nurse.

b http://www.ic.nhs.uk/webfiles/publications/009_Social_Care/pss0910expfinal/pss0910updateOct2011/Personal_Social_Services_Expenditure_Report_2009_10.pdf

c New calculation: there is a recommended fee payable to ophthalmic medical practitioners who administer sight tests, although optometrists undertake most tests. Optometrist salaries vary depending on practice setting (private or hospital or combination of the two). Typical salaries in private practice based on salary data collected June 2009 (http://www.prospects.ac.uk/optometrist_salary.htm) ranged from £19 500 to £28 000. In hospital settings, optometrist salaries are usually covered by the Agenda for Change pay scale. Average salary for private practice was used. Cost per hour was estimated based on 41 weeks per annum, 38 hours per week.

d http://cash-online.org.uk/content/1/6/3/; uprated using the Consumer Price Index (CPI). ^{Reference Curtis15}

e Cost of adult social club at 2004/05 uprated using the pay and prices inflator. ^{Reference Curtis15}

Statistical analysis

An ‘intention-to-treat’ analysis was carried out to preserve the unbiased distribution of factors in the groups produced by randomisation. Missing resource-use data were singly imputed. If there was no report on the use of a particular resource, we assumed that it was not used. If participants reported on a resource but not the quantity used, we imputed this amount from within treatment-group means for participants with data for that item at the same assessment point.

Health and social care costs for 0–13 weeks and 0–39 weeks (and health/social care and costs of unpaid carer costs for the parallel analysis from the broader perspective for the same time periods) were regressed in turn on treatment allocation, baseline cost, baseline CSDD and centre. To mitigate the effects of skewness, non-parametric bootstrapping methods were used to estimate 95% confidence intervals (CIs) for mean costs. Where the bias-corrected 95% CIs of between-group change scores excluded zero, they could be judged to be significant at P = 0.05 or lower.

Estimates of bootstrapped mean cost and effectiveness were used to estimate an incremental cost-effectiveness ratio (ICER) for each analysis. The ICER for each replication was calculated as:

$({\text{cost}}_b-{\text{cost}}_a)∕({\text{effect}}_b-{\text{effect}}_a),$

which summarises the cost difference between two treatments per incremental difference in the outcome (CSDD and EQ-5D in turn). The EQ-5D was measured directly from patients – as recommended by National Institute for Health and Clinical Excellence (NICE) guidelines ¹⁸ – and weighted by a valuation of changes in quality of life reported from UK population data. ^{Reference Dolan, Gudex, Kind and Williams13} Health effects were then expressed in terms of QALYs. The ratio statistic compared the treatments in terms of observed differences in costs and effects, regardless of whether those differences were statistically significant.

Uncertainty about the cost and effectiveness estimates was addressed by plotting cost-effectiveness acceptability curves (CEAC). A CEAC was used to assess trade-offs between costs and outcomes, showing the likelihood of each of the two medications in turn being seen as cost-effective relative to the other or relative to placebo, given different (implicit monetary) values placed on incremental outcome improvements. In this net-benefit approach, monetary values of incremental effects and incremental costs for each case are combined, and the net benefit derived as:

$\mathrm{NB}=\lambda \times ({\text{effect}}_b-{\text{effect}}_a)-({\text{cost}}_b-{\text{cost}}_a),$

where: a is the comparator, b is drug treatment, NB is net benefit, and λ is willingness to pay for a unit improvement in CSDD-depression severity score (primary evaluation) or an additional QALY (secondary evaluation). The impact on costs given uncertainty around the value attached to informal care inputs was assessed in one-dimensional sensitivity analysis. All analyses were completed in Stata (version 11) and SPSS 17 on Windows.