Purpose and Scope
The focus of this model project is on the clinical progression of Alzheimer's Disease (AD) as measured by the cognitive portion of the Alzheimer's Disease Assessment Scale (ADAS-cog). We intend that this model will be sufficient to generate realistic patient-level ADAS-cog scores over time. We envision that this model will be an aid in planning clinical trials in AD, in interpreting the results of such trials, and as a tool in its own right to answer basic research questions about the phenomenology of the disease.
About the current models
- General: There are currently two distinct models being developed in parallel. Both are hierarchical models with study-level and patient-level random effects, describing ADAS-cog scores as function of treatment, dose level (if applicable), and time on treatment. See the files in /doc for more detailed mathematical descriptions.
- "cfbmodel" describes ADAS-cog change from baseline scores. An advantage of this model is that it utilizes both observed sample means and observed sample variances in the fitting process. A disadvantage is that the predictive distribution for the model extends beyond the known boundaries for the ADAS-cog (0 -- 70).
- "rawmodel" describes absolute ADAS-cog scores. An advantage of this model is that the predictive distribution for the model is constrained between zero and seventy, i.e. the natural constraints of the instrument are respected. A disadvantage is that observed sample variances are not currently leveraged in the model fitting process.
- Data: Data are from summary statistics from multiple published clinical trials, as well as individual patient data from the ADNI study. Both "cfbmodel" and "rawmodel" employ a covariance structure that weights meta data and individual data in a theoretically coherent manner.
- Implementation: The model is fitted using Bayesian methodology with diffuse priors. Model fitting computations are carried out using WinBUGS 1.4.3.
Planned enhancements
- The current "rawmodel", when fit to publicly available data only, does not reliably estimate the relationship between baseline MMSE and slope of natural progression. More publicly available individual data are need to improve this estimation. Correct characterization of this relationship is important for the purpose of clinical trial simulation, in order to assess the likely effect of enrolling more/less severe populations, and to anticipate likely differences between subset analyses based on baseline severity.
