Analytical approach:
A Markov model was developed to simulate the ongoing probabilities of not adhering, partly adhering, or fully adhering to treatment, and of relapse and hospital admission, for a cohort of multi-episode patients. The focus of the evaluation was the first treatment choice; up to four sequential choices of therapy were modelled. The time horizon was five years. The authors stated that the perspective was that of the direct medical payer, which was the Swedish health care system.
Effectiveness data:
The main treatment effectiveness data were the risk ratio of relapse, the level of adherence, side-effects (extrapyramidal symptoms, tardive dyskinesia, weight gain and diabetes), and treatment discontinuation. The risk ratio of relapse was from a published mixed-treatment comparison for some therapies, and from several published studies for the other therapies. Adherence rates by treatment and side-effect rates were from observational data. Treatment switch rates were derived from a large observational study.
Monetary benefit and utility valuations:
The utility values for the model health states were from a UK study that used the time trade-off technique on 49 patients with stable schizophrenia and 75 lay people.
Measure of benefit:
The measure of benefit was quality-adjusted life-years (QALYs) gained. Benefits were discounted at an annual rate of 3%.
Cost data:
The resource use associated with a relapse was from a six-month comparative UK study. The additional resource use for side-effects was from National Institute for Health and Care Excellence (NICE) guidelines. In-patient resource use was Swedish National data. Drug costs were from the Swedish Pharmaceutical Benefits Agency (TLV). All other unit costs were from regional Swedish sources. All costs were reported in Swedish kronor (SEK). The price year was 2009. The costs were discounted at an annual rate of 3%.
Analysis of uncertainty:
Several one-way sensitivity analyses were conducted, by varying the key parameters by ±25%. Probabilistic sensitivity analysis was conducted by varying every model parameter simultaneously to assess the overall uncertainty in the cost-effectiveness estimate. Cost-effectiveness planes were used to present the results.