A significant correlation between general measures of dementia, such as the DRS or the MMSE, has been consistently reported, suggesting that error rates, and ultimately DLPFC pathology, might simply be predicted by general levels of dementia. We found that the mean antisaccade error rate of AD patients, 55%, was relatively low compared with previously reported antisaccade error rates of 50–80%. Although this study was not strictly comparable to previous studies, the comparison reveals that the exclusion of more severely demented patients may have resulted in lower mean error rates relative to previous studies, which did include severely demented patients.
We were unable to replicate the previously reported correlations between error rates and MMSE scores within the AD group, likely for several possible reasons. First, the relationship between MMSE and antisaccade error rates in previous studies may have been driven by the more severely demented patients who consistently perform poorly on the antisaccade task, and were excluded for our study. As discussed above, this suggests that antisaccade error rates, and potentially frontal neuropathology, may not reflect overall dementia severity during mild stages of AD. Second, the heterogeneous nature of AD renders the MMSE an unreliable metric for dementia severity. For instance, lower MMSE scores might reflect domain-specific impairments in language or memory, which are heavily weighted in the MMSE, while executive functions remained preserved, or at least are not well captured by the MMSE. The DRS is more weighted for dorsolateral frontal functions but the smaller sample size may have been insufficient to detect correlation. Both possibilities are not mutually exclusive and could contribute to the differences between this study's findings and previous investigations.
We considered the possibility that the group differences were attributable to failure to maintain task instructions over task blocks (Welch 1947
). As noted in the methods, subjects were required to point to the correct location where they were supposed to look for three consecutive trials prior to the start of the first block. Instructions were reinforced between blocks. Although loss of task set cannot be ruled out as contributing to our findings, we do not consider this to be explanatory as the patients appeared able to maintain task set for the 72-sec duration of the block, as indicated by their ability to switch instructional set between pro- and antisaccade blocks, even though they were error prone. The low rates of prosaccade errors (3.5% for AD versus 1.9% for NC), although significantly different, also suggests that the AD patients were able to follow the instructions. To conclusively rule out task set maintenance problems, future studies should verify task set instructions before and after each block. Augmenting fixation cues with task set information, further reducing the set maintenance element of the task, could be used as a manipulation check to evaluate set maintenance effects.