Analytical approach:
A Markov model was developed to assess the clinical and economic impact of the two strategies. The time horizon of the analysis was five years and the authors stated that both a societal perspective and that of the health care payer were adopted.
Effectiveness data:
The clinical data, which were transition probabilities, were mainly derived from a previous study by the same authors. This study followed Taiwanese patients with mild or moderate AD receiving or not receiving donepezil. However, no other details on this key source were given. The key clinical outcome was the relative risk of disease progression with or without donepezil.
Monetary benefit and utility valuations:
The utility valuations were based on a previous US study which used the Health Utility Index Mark II with a sample of 528 AD caregivers of patients cared for at home.
Measure of benefit:
Quality-adjusted life-years (QALYs) were chosen as the summary benefit measure and they were discounted at an annual rate of 3%.
Cost data:
The three main cost categories considered were medical expenses paid by the National Health Insurance, out-of-pocket co-payments, and unpaid informal care. Direct medical costs included the costs of AD care, co-morbidity treatments, and acquisition fee of donepezil. The productivity costs included caregiver time. All costs were presented as macro-categories. They, and related data on resource consumption, were derived from published studies involving Asian and Taiwanese patients. Caregiver time was valued using two approaches, the replacement method and the labour cost method, which provided high and low estimates of costs. The cost of donepezil was based on the reimbursement rate per tablet. All costs were in US dollars ($) and a 3% annual discount rate was applied to future costs. The price year was not clearly reported but the cost data were collected in 1999.
Analysis of uncertainty:
A deterministic univariate sensitivity analysis was performed on all model inputs and a tornado diagram was generated. Influential model inputs were further investigated in a probabilistic sensitivity analysis. The types of distributions were stated.