Ideally, one state in the model will have a probability close to 1 for each individual, while other states will have probabilities near 0, indicating that a subject’s cognitive profile is known with near certainty. As long as models are correctly maybe specified, this near certainty will indeed be obtained given a sufficient amount of testing [7]. Once classification is completed, subjects who share a cognitive profile can be aggregated, and observed rates of conversion to AD between the resulting subgroups can be compared. In our study, we used NP data collected by the Alzheimer’s Disease Neuroimaging Initiative (ADNI) [9] to evaluate the usefulness of poset models to identify specific cognitive phenotypes associated with conversion from MCI to AD [7,8,10,11].

We hypothesized that poset modeling would generate interpretable and sufficiently detailed cognitive phenotypes with clearly differentiated rates of conversion from MCI to AD. Given its established importance in progression risk for AD, Apolipoprotein E (APOE) e4 status was also taken into account [12,13]. Materials and methods Study sample MCI subjects enrolled in the ADNI (n = 389) were included in the classification analysis if their scores on the selected NP battery were available at both baseline and at 24 months. The sample was 64.5% male and 93.1% Caucasian. About 3.8% were African American, 2.8% were Asian, and 0.3% were American Indian or Alaskan Native. Mean age was 74.8 years (SD = 7.5) and mean length of education was 15.7 years (SD = 3.0).

Modeling and classification approach We used neuropsychologist expert opinion (JJ and HYT) to map the relationship between selected Carfilzomib ADNI NP measures (ADAS delayed recall and word recognition subscales and number cancellation; auditory verbal learning test (AVLT) Trial 6 and List B; Boston naming test; category fluency; trail making Test A and Test B, and Wechsler adult intelligence scale-revised (WAIS-R) digit symbol substitution) and the cognitive functions required to perform them (episodic memory at four different levels, word fluency, cognitive flexibility, perceptual motor speed, and attention). See Table ?Table11 for the listing of these specifications. Measures were selected based on the types of functions they tested and retained based on statistical criteria such as discriminatory properties and correspondence with model fit. Sunitinib buy Given the reliance on expert opinion, data-analytic validation is important. Statistical details on how this validation was performed are provided in Additional file 1 (Appendix).