P3‐090: Change in the Connectivity of Multidimensional Outcomes in a Clinical Trial of Rosiglitazone for Alzheimer's Disease

Arnold Mitnitski,An Zeng,Michael Gold,Kenneth Rockwood

DOI: https://doi.org/10.1016/j.jalz.2011.05.1530

2011-01-01

Abstract:Understanding disease progression in dementia trials relies chiefly on multidimensional measures, such as the Alzheimer's Disease Assessment Scale - Cognitive subscale (ADAS-Cog), Neuropsychiatric Inventory (NPI) and the quality of life measure EQ-5D. Typically, mean changes in each measure are compared statistically. Recent work has suggested that change in the degree of connectivity associated between items which make up these scores can reflect underlying disease processes. We evaluated change in connectivity in patients treated with rosiglitazone, which does not impact disease progression. This is a secondary analysis of data from a 6-month, parallel-group, placebo-controlled, randomized clinical trial. Patients with mild-moderate AD (n= 639; mean age (SD) = 72.5 (8.3); 63.2% women) were randomly assigned to rosiglitazone 2mg (n=166; 133 (80.1%) of 166 completed 6 months), rosiglitazone 8mg (n=165; 127 (77.0%) of 165 completed), or placebo (n=166; 132 (79.5%) of 166 completed). Connectivity graphs were used to represent relationships between the 26 items making up the ADAS-Cog, NPI and EQ-5D. Two items are connected if their co-occurrence cannot be explained by chance; otherwise, they are not related (null hypothesis, H0). When the pair-wise mutual information (PMI) measure between 2 items is significantly different from zero, H0 is rejected. Significant relationships are graphed as lines between two items. The degree of connectivity is represented by the number of connections between items. At baseline, the 26-point connectivity graphs (11 ADAS-Cog + 10 NPI + 5 EQ-5D) showed a mean connectivity level of 141. By 6 months, this had declined to 110 (p< 0.001). A staging effect was evident. For people with mild AD, the mean connectivity level at baseline was 60, compared to 30 in moderate dementia. By 6 months, the number of connections had diminished to 37 and 21 respectively (P< 0.001). Connectivity graphs offer a novel means of quantifying the relationships between the many domains typically evaluated in dementia trials. By summarizing a high degree of dimensionality in a single number, while graphing different patterns of connectivity, they can facilitate understanding of differential response patterns seen in dementia clinical trials. Whether disease-modifying drugs would stabilize connectivity patterns is of interest.