New decision analytical models for management of intracranial aneurysms

Abstract

This thesis addresses decision analysis, cost-effectiveness models and the analysis of heterogeneity, applied to intracranial aneurysms and subarachnoid hemorrhage (SAH). Subarachnoid hemorrhage is a subset of stroke that usually occurs at relatively young age and has poor prognosis. Although, the risk of SAH can be reduced by screening for and preventive treatment of unruptured aneurysms, these strategies are not without risk to the patient. Regional variation of SAH incidence rates in Sweden was assessed using hospital admission data on 18,443 patients with SAH in 1987-2002. The incidence of SAH increased with age, was higher for women than for men, and higher in the north than in the south. The 28-day case fatality rate increased with age up to 80 years, and then leveled off. Both the SAH incidence and case fatality rates decreased over time. The cost-effectiveness of screening patients who survived an episode of SAH for new intracranial aneurysms was analyzed. In general, screening cannot be recommended, since it is not beneficial. However, screening was cost-effective in patients with an at least 4.5 times increased risk of both aneurysm formation and rupture. The cost-effectiveness of different treatment strategies for elderly patients with SAH was analyzed. Aneurysm treatment was more costly than conservative treatment, without providing additional health benefits, in patients admitted in poor condition more than ten days after SAH, and patients older than 80 years, admitted in poor condition more than four days after SAH. Cost-effectiveness was acceptable only for women aged 70-79 years, and men aged 70-74 years, admitted in good condition within four days. Occlusion of ruptured intracranial aneurysms instead of conservative treatment improves outcome in some elderly patients, but not in all, and costs will often be unacceptably high. The hypothesis of a constant, annual growth rate for all intracranial aneurysms was assessed in a simulation study and qualified as improbable. The actual growth process is probably irregular and discontinuous, consisting of periods with and without growth. Ignoring the effect of variation in disease progression within individuals over time in cost-effectiveness models will cause the benefits of screening strategies to be overrated. Survival datasets were simulated and unobserved heterogeneity (frailty) was added to determine the practical implications of using standard survival models instead of frailty models, in the presence of frailty. Standard models did not perform worse than frailty models in prediction tests. Ignoring frailty and the small degree of bias it may introduce in regression coefficients of covariates and survival parameters will have little practical implications in most settings. An extension of value-of-information methods concerning the expected value of perfect information (EVPI) is presented. The EVPI conceals separate contributions of underlying uncertainty in costs and effects to the expected monetary loss caused by all uncertainty combined. The EVPI is dissected in terms of its underlying cost and effect outcomes. It is demonstrated that for any EVPI there is an (infinite) range of decision problems with identical EVPIs differing only in terms of their contributions of expected costs and effects, the 'attributable EVPIs'.