Joint seminar with the School of Public Health.
Terry Therneau
Mayo Clinic
Room: HLTH LHS-1621
Challenges in modeling cognitive aging
In this talk I will discuss key issues in modeling cognitive aging for both direct measures of cognition (e.g. memory tests and activities of daily living) and assessment of underlying processes via biomarkers (e.g. amyloid PET or serum assays). The fundamental challenge is that the timeline of the underlying processes will often span 20 or more years in a given person, while the median observation time for any given participant is normally no more than 4-6 years even for long term studies. Additional challenges include the presence of substantial between subject variation in the age of onset of the processess. For amyloid, for instance, APOE e4+ shifts onset by about 6 years, remaining per subject variation is about +- 9. There is also variation in the availability of measurements since different cogitive tests and blood samples are colletected at different times in different studies. Modeling and inference is further challenged by a low signal-to-noise ratio for many markers. Further challenges arise from informative censoring. The goal is to understand how multiple markers relate to any given underlying pathology, and to study how multiple underlying pathologies may interact. One subject may exhibit pathology A earlier than others while others may exhibit pathology B earlier.
Several different modeling approaches have been proposed in the literature, but often focus on only one or two of the above. A successful modeling approach in this milleau needs to effectively integrate as much information as possible, across patients, ages, and markers. Examples we have found useful are multi-marker nonlinear mixed effects models and hidden Markov models; each has both strengths and weakness, and both are computationally challenging.