OBJECTIVES: Results from cost-effectiveness (CE) models have been shown to vary with structural assumptions. We compare structural assumptions in models estimating the effectiveness of hypothetical disease-modifying treatments (DMTs) for Alzheimer’s disease (AD).
METHODS: A targeted literature search of the Medline database from 2010 to present identified disease models that include hypothetical DMTs. Structural assumptions were compared relating to model design, disease progression, institutionalization, mortality, and DMT efficacy.
RESULTS: Ten published models were reviewed: 4 patient-level simulations, 5 cohort-level Markov models, and 1 using average MMSE decline rates. Nine models included predementia health states, and one model included the diagnostic process. Events in simulations included conversion to dementia (two studies), institutionalization (two studies), and treatment discontinuation (two studies). Health states in the Markov models differed in how pre- and post-dementia states were subdivided by severity and site of care. Simulations used results from regression models that included measures of cognition, function, behavior dependence, and, in one model, biomarkers to represent disease progression. Markov models used transitions between disease stages measured by cognition and/or function to represent disease progression. Institutionalization rates presented in Markov models as incidence (one study) or prevalence (one study) by disease severity. Mortality varied by severity as an additive (two studies) or multiplicative (three studies) factor but only varied by age and gender in two studies. DMT efficacy was included as slowing and/or halting disease progression for a given period or changing the time to dementia. Structural sensitivity analysis was not included in most models; results were mixed in three studies that estimated the impact of alternative mortality assumptions.
CONCLUSIONS: Models that estimated the effectiveness of hypothetical DMTs for AD varied in key structural assumptions that may impact estimated CE ratios. Results from models for AD should be validated against observational data to increase credibility of the results.
