Purpose: To demonstrate analytical methods, developed during recent evaluations of treatments for psoriasis and chronic pain, which allow optimum treatment sequences to be identified for particular chronic diseases.

Background: Often, when treating chronic disease, a sequence of treatments will be ‘trialled' until a patient responds. Cost-effectiveness analyses that consider individual treatments as mutually exclusive comparators may not correctly identify the treatment that should be trialled first and which further treatments should be trialled. The analysis should consider all the available treatment sequences as comparators. However, it may be impractical to directly compare all sequences; with the 7 available psoriasis treatments there are 13,699 possible sequences. More tractable analytical methods are required.

Methods: The optimum ordering is identified by estimating the expected net-benefit per unit time for each treatment - accounting for the proportion of patients who do not respond. As there will be attrition as we proceed along a treatment sequence, we maximise the total expected net benefit by trialling the treatments in order of decreasing expected net-benefit per unit time. If a treatment has an expected net-benefit per unit time less than that for supportive care, its use is not cost-effective.

Results: The optimum sequence varies as a function of threshold willingness-to-pay for a QALY. For the treatment of psoriasis in patients at high risk of hospitalisation the systematic therapies, methotrexate, ciclosporin and Fumaderm, were included in the sequence irregardless of threshold. For the sub-group of patients with severe disease, intermittent etanercept entered the sequence at a threshold above £20,000; efaluzimab and continuous etanercept entered the sequence at a threshold above £25,000. The optimum sequences identified by the proposed methodology were found to be identical to the optimum sequences identified by a ‘conventional' incremental analysis including all treatment sequences as comparators. The analysis of treatments for chronic pain showed that oxycodone CR, morphine CR and Fentanyl would all be included in a treatment sequence at a threshold above £10,000.

Conclusion: The proposed methodology is convenient for the identification of optimum treatment sequences. The probability that a given treatment will be cost-effective at any point in the sequence can also be estimated. The methodology is valid for treatments that only provide benefit while they receive treatment and do not alter the progression of the disease.