Table 2.

Summary of included studies

	Model (reference)	Intervention and setting	Main outcome(s)	Modelling approach/ framework, time horizon and cycle	Data sources
					Disease onset, progression and mortality data	Intervention effectiveness	Utilities/outcomes measure	Costs
Primary prevention	Zhang, Kivipelto [26]	Hypothetical intervention reducing risk of AD onset in Sweden	QALYs	Purpose-built Markov model with 3 health states. 20-year time horizon, 1 year cycles	CAIDE population based study on risk factors of 1409 individuals [44]	Hypothetical intervention	Swedish studies (EQ5D) [41, 57]	Swedish National Board of Health and Welfare
	Tsiachristas & Smith [32]	Preventative treatment with B-vitamin supplement for people aged 60 and over with elevated levels of tHcy in the UK	Life-Years; QALYs	Stochastic probabilistic decision tree; lifetime horizon.	Disease progression not modelled. Disease onset based on prevalence data; mortality from life tables.	Effectiveness of intervention based on a systematic review in lieu of randomised controlled trials [37]	General population EQ5D survey [36]	Taken from a UK study [58]
Secondary prevention	McMahon [28]	Functional neuroimaging vs. standard work-up of patients for AD diagnosis at specialised AD clinics in the US	QALYs	Markov model based on a previously published study [40]; 6-week cycles, 18-month time horizon.	Progression within AD and AD mortality from CERAD study [40]. Non-AD mortality from CDC.	Screening effectiveness from US-based study [59]	Utility weights obtained from the Neumann et al. [40]	Primary data from hospital databases; existing literature
	Silverman, Gambhir [31]	PET vs. standard diagnostic methods in clinical diagnosis of AD in the US	Number of accurate diagnoses	Purpose-built decision-tree, unspecified time horizon	Adapted from a wide range of published data	Results of PET screening reported in the study	Not used - CEA	Defined by Medicare reimbursement rates
	Weimer and Sager [30]	Early detection and treatment of AD patients in a US (Wisconsin) setting. Two treatments considered.	MMSE score change	Monte Carlo model. Lifetime horizon, 1 year cycles	Adapted from a range of published data and estimates. Data from CVD risk study on 5000 people was used to estimate hazard ratio for death.	A range of published data and estimates	Adapted from [40]	A range of published data and estimates
	Dixon, Ferdinand [35]	One-off screen of 75 year olds in England and Wales	Number of additional diagnoses	Static decision model with lifetime time horizon	Not provided	Results of screening based on MMSE (assumed 89% sensitivity, 95.5% specificity)	Not used – CBA	A range of published data and estimates
	Saito, Nakamoto [27]	Community based dementia screening in a US setting	Dementia diagnosis through MMSE	Purpose-built Markov model with 6-state 10-year time horizon, 1 year cycles	Adapted from [46, 48] which investigated 61 and 1145 patients, respectively	Results of screening program reported in study	Not used - CEA	Adapted from a Canadian study [60]
Tertiary prevention	McDonnell, Redekop [33]	A hypothetical intervention which slows cognitive decline in AD patients in the Netherlands	MMSE score change, care setting, mortality	Two regression-based simulation models – one modelling MMSE score, another- care setting and mortality. 10-year time horizon, 6 month cycles	Calculated from a Dutch study [38] with 7528 participants.	Hypothetical intervention	Not used – CEA	From Dutch national data, agencies/ ministries
	Martikainen, Valtonen [29]	Cognitive-behavioural family intervention to delay admission to nursing home in Finland	Time to nursing home admission	Markov model. Adapted from [40]. Model has 4 states, 5-year time horizon, 1 year cycles	Adapted from the original US-based model (with minor adjustments) – based on longitudinal study with 1145 patients [40]	Based on a US study of 206 subjects [61]	From the original US-based model	From national datasets; some resource utilisation based on expert panel
	Mirsaeedi-Farahani, Halpern [34]	Deep-brain stimulation therapy for slowing memory loss in AD patients compared to standard treatment	QALYs	Purpose-built Markov model with 5 states, 5-year horizon, 1 year cycles	Adapted from Neumann et al. [46] and Spackman et al. [47]	Actual success rate of deep brain stimulation is unknown, so was varied from 0 to 100%	A range of published data	Costs obtained from [62]