Observation versus coiling for unruptured intracranial aneurysms in octogenarians: A decision-analytic modeling study

Management of unruptured intracranial aneurysms in patients aged 80 years and older remains uncertain because treatment morbidity, rupture risk, post-rupture outcome, and competing mortality interact. We constructed a competing-risk Markov decision model comparing observation with endovascular coiling in Korean patients at index ages 80, 85, and 90 years; microsurgical clipping and stent-assisted coiling or flow diversion served as additional scenarios. The primary outcome was the undiscounted lifetime cumulative incidence of aneurysm- or treatment-attributable events—treatment morbidity, treatment death, rupture death, or rupture poor outcome—counted as a first-event composite in which all four components carry equal weight regardless of severity, duration, or reversibility. Periprocedural coiling morbidity and mortality were anchored to a U.S. National Inpatient Sample analysis from 2001–2008, and residual post-treatment rupture risk was fixed at 10% of the natural-history rate. Under this original endpoint, median differences favored observation in every age-90 panel and in all male panels at age ≥ 85, and favored coiling in selected age-80 panels and selected age-85 female panels; the probability that coiling was favored ranged from 0.000 to 0.984. A 24-panel, 3%-discounted severity-weighted quality-adjusted analysis was added, together with structural sensitivities for treatment-morbidity recovery, residual rupture risk, and coiling morbidity. Across 30 endpoint-direction comparisons, with each Ma panel mapped to its two legacy marginals, 21 were concordant; all nine discordant comparisons shifted from observation toward coiling, and eight had 95% quality-adjusted uncertainty intervals spanning zero. Median-direction changes under the executed structural sensitivities were confined to modified Rankin Scale-based panels with reference quality-adjusted differences near zero and uncertainty intervals spanning zero. Modeled treatment preference was therefore conditional on endpoint definition and key structural assumptions, including historical treatment-risk estimates and uncalibrated residual post-treatment rupture risk; these model outputs do not establish treatment recommendations for individual patients.