Connected Language in Late Middle-Aged Adults at Risk for Alzheimer’s Disease

INTRODUCTION

A major goal of neuropsychological research in Alzheimer’s disease (AD) is early detection of cognitive decline. To date much of this research has focused on identifying individuals who meet criteria for mild cognitive impairment (MCI), because of the high risk of converting to dementia in this group [1, 2]. MCI diagnostic criteria established by a joint National Institute on Aging and Alzheimer’s Association consensus panel include the following: concern about cognitive decline by the person or significant other, preservation of independence in functional abilities, and impairment in one or more cognitive domains [2]. Although episodic memory impairment is often the main cognitive symptom of AD, considerable research evidence suggests that the neuropathological changes of AD develop years or decades before the onset of such cognitive symptoms [3–5]. Therefore the quest to find sensitive neuropsychological measures that are effective in detecting very early decline, that can be differentiated from typical changes associated with aging, has beenongoing.

In addition to episodic memory decline, declines in semantic memory have also been described early on the continuum of AD. Semantic memory refers to knowledge of concepts, objects, people, and words. Barbeau et al. [6] found that a group with MCI performed significantly worse than controls on several semantic memory tasks, including naming famous faces, cultural knowledge, and famous public events. The authors argue that the semantic deficits exist without temporal gradient (i.e., deficits exist no matter how old or recent the learning occurred), and thus may not be caused by episodic memory deficits. The authors also found that the semantic impairments were correlated with anterior, extra-hippocampal, temporal lobe brain structures as well as to the hippocampus. Didic et al. [7] describe semantic memory as being “context-free”; that is, the recall of previously learned facts, words, or names is dissociable from the original context in which it was learned, and is less dependent upon the hippocampus and more dependent upon anterior medial temporal structures which appear to be affected earlier in AD. Didic et al. [7] argue that even the more sensitive tests designed to assess episodic memory function, such as word list learning and recall, are “context-free” by nature and difficulties may be reflective of semantic, as well as episodic memory problems. Venneri et al. theorize that assessments targeted toward semantic memory may be more sensitive in distinguishing normal aging processes from preclinical AD; that is, normal aging typically results in a gradual decline in episodic memory, while semantic memory is usually spared, if not improved, in normal aging [8].

Semantic memory difficulties, within the context of everyday functioning, may be manifest by word retrieval difficulties and naming impairments, which are common complaints in both MCI and AD. Faber-Langendoen et al. [9] found that 36% of mild AD patients and 100% of severe AD patients had aphasia (i.e., impairment in some aspect of language function) concurrent with memory impairments.

In order to characterize word finding problems, clinicians and researchers often use confrontational naming tasks as a means of quantifying this difficulty. While declining performance on typical confrontational naming tasks is evident in AD [10], and confrontational naming tasks have been useful in staging more moderate to severe AD [11], such tests have not been useful in reliably detecting differences in preclinical AD or MCI [12, 13].

Word finding may also be evaluated by using verbal fluency tasks, in which patients are asked to name words of a category or words beginning with a specified letter, and are often used to assess problems with both executive function and language in AD [10]. Specifically, studies of verbal fluency in MCI have shown lower scores in those with MCI and AD compared to age-matched controls [14–16]. Recently, our group found differences in verbal fluency abilities between persons identified as “cognitively healthy” and those identified as ‘psychometric MCI’, a construct using robust norms across three longitudinal visits and analogous to early or ‘preclinical’ MCI, which may possibly identify very subtle cognitive dysfunction [17].

Assessment of isolated language functions, however, might not capture the magnitude of problems experienced in everyday communication contexts. For evaluation that more closely approximates language demands of everyday life, researchers have turned to assessment of spontaneous, connected speech, both as a diagnostic resource as well as a research tool toward understanding the subtle deficits reported by patients in their natural settings. Connected language is elicited by having an individual produce spoken or written language in response to a particular question or stimulus, including picture description tasks [18–21], familiar story re-telling [22] (e.g., “Cinderella” story), wordless picture book descriptions [23], and general autobiographical interviews [24]. Language is then analyzed using measures such as efficiency (e.g., idea density), conciseness (e.g., how many words required to express an idea), complexity (e.g., syntax complexity), and speech fluency (e.g., the number of repetitions, circumlocutions, filled pauses). Analysis of connected language has revealed subtle differences years or decades before diagnosis of AD [20, 26]. These language differences were a salient finding in the Nun Study [25], in which idea density in early life writing samples was an accurate predictor of later-life progression to AD. Links between content of language samples and preclinical AD also were reported by Cuetos et al. [20], who examined spontaneous language in a group of 19 carriers of the E280A mutation in the Presenilin 1 gene at the preclinical stage of AD (mean age = 43.2). Carriers showed deficits in the production of semantic categories in a picture description task during the pre-clinical phase of disease. There were no differences between carriers and non-carriers in average sentence length or total number of sentences, which implies that clinical examination of language productivity would not necessarily detect group differences. More recently, Ahmed et al. [27] examined longitudinal profiles of 15 adults (mean age 71.2) with autopsy-confirmed AD and found decreasing linear trends in syntactic complexity and semantic and lexical content across stages of disease, including the MCI stage of disease. Given evidence of subtle language changes in MCI, it was of interest to see if connected language differences could be a cognitive marker in very early declines that are prior to a diagnosis of MCI. Specifically, we hypothesized that participants who were cognitively healthy would differ on connected language outcomes from a subset with psychometric evidence of aMCI (i.e., the subset did not have a clinical diagnosis of MCI, but did meet statistical criteria for early decline as described in the methods). While evidence for early language change exists based on retrospective analysis [26–28], to our knowledge no studies have examined connected speech and language prospectively in an at-risk cohort.

In the current study, we examined connected language samples in a group of adults (ages 40–65 years at baseline) who were at risk for AD, defined as “psychometric MCI” (pMCI). The psychometric MCI construct refers to the presence of memory declines across at least two of three longitudinal study visits based on internal norms (described further in the Methods section). We hypothesized that adults who were clinically cognitively intact but met psychometric criteria of MCI (see Koscik et al. [29]) would have lower scores on measures of content, syntactic complexity, and speech fluency than age- and education-matched adults defined as cognitively healthy.

MATERIALS AND METHODS

RESULTS

The average age of the participants included in these analyses was 63.1 years (SD = 6.3); 22% were female; and the average WRAT reading standard score was 110.6 (7.4). The CH and pMCI groups differed significantly in terms of APOE-4, in that the CH group had a higher percentage of participants with at least one E4 allele; family history status was not significantly different. Table 2 shows that the demographic characteristics of the pMCI groups were very similar, as would be expected based on the matched sampling strategy that was used. Table 2 also describes the two samples regarding neuropsychological performance, and is reflective of the defining nature of the psychometric aMCI criteria [10, 29]. Specifically, the verbal and learning scores were significantly different between the two groups, but the executive function scores were not. Both letter fluency and category fluency, which were not part of the psychometric aMCI diagnostic criteria, yielded worse scores for the PMCI group than the CH group (p = 0.0005 and p < 0.0001, respectively); the BNT was not significantly different between groups.

Connected language results are presented in Table 3. To facilitate review of all language outcomes in a single graph, the mean(se) of the corresponding z-scores across several language outcome measures are shown in Fig. 1 by cognitive status groups. The CH and PMCI groups differed significantly on two content measures: pMCI subjects produced fewer unique words (p = 0.028) and fewer semantic units (p = 0.018) than the CH group. Marginal differences were found in Propositional Idea Density (p = 0.09) and in the Moving Average Type-Token Ratio (MATTR) (p = 0.06). The CH and PMCI groups did not differ significantly on Maze Index or filled pauses/utterance, nor on any of the syntactic complexity measures.

DISCUSSION

In this study, spontaneous language samples were examined from two groups of late-middle-aged participants drawn from a cohort enriched for family history of AD: those who were evidencing psychometric memory decline based on internal robust norms (pMCI) and those who were classified as cognitively healthy. Psychometric memory decline was identified in participants who showed episodic memory scores that were more than 1.5 SD below robust norms on at least 2 of 3 study visits. We hypothesized that the pMCI group would perform worse on language variables characterizing semantic content, syntactic complexity, and speech fluency, than those who were cognitively healthy.

The pMCI group performed worse on both phonemic (letter) and semantic (category) fluency, two measures that were not included in the psychometric criteria for determining the groups. The confrontational naming task (BNT) was not significantly different between groups, consistent with previous MCI literature [10].

Our results suggested that the pMCI group expressed fewer semantic units and fewer unique words than the cognitively healthy group, which is a similar finding to that seen in studies with persons with MCI and AD [20, 28]. Recently, Berisha et al. [28] found that former President Ronald Reagan’s use of unique words declined over time, prior to his diagnosis of AD. Similarly, while the pMCI group in our sample exhibited verbal learning and memory declines across longitudinal visits that were significant according to robust norms, they were not diagnosed with clinical MCI because they were not sufficiently impaired according to general test norms. Therefore, the pMCI group did not meet formal diagnostic criteria for MCI and are categorized as having a psychometric condition; these findings may suggest and reinforce the idea that language symptoms may be present antecedent to meeting formal diagnostic criteria for the clinical condition (aMCI). The semantic unit findings may represent subtle differences in semantic processing, particularly within the context of the demands of producing spontaneous speech, and may be predictive of larger language changes over time. That the Semantic Units measure was not significantly correlated with either verbal fluency tasks, but was correlated with the Rey Auditory Verbal Learning Test, suggests that it may tap into additional language or other cognitive processes, and may contribute unique knowledge regarding preclinical language changes.

The differences in semantic units between the two groups could support the hypothesis that semantic memory declines may be reflective of the underlying early neuropathology in the transentorhinal, mediotemporal networks, while the hippocampus may still be spared [8]. Further, when the task is embedded within spontaneous speech, a more qualitative analysis affords an informative characterization of problems within a functional context. The task itself is relatively quick and noninvasive for patients to complete, and may be informative not only for very early diagnosis, but also for disease course monitoring in both pharmacological and nonpharmacological clinical trials.

Secondary analyses were conducted to adjust for the higher percentage of participants with APOE-4 in the cognitively healthy group. Since APOE-4 is considered to be a risk factor for cognitive decline even in normally aging populations [46], we were interested in adjusting for this possible confound in our control group. This adjustment of controlling for APOE-4 strengthened the results of the Semantic Units and unique words findings, and also revealed that the pMCI group’s language samples had lower propositional idea density than the CH group. While several studies have shown that lower propositional density in early life predicted future AD risk [25, 47], and one small study found that young APOE-4 carriers had lower propositional density than noncarriers [48], the effect of APOE-4 status on language abilities in aging and in AD risk is not well understood. Whether APOE-4 status exerts an effect on future language declines will be addressed in longitudinal follow-up studies on this cohort.

No differences were detected in syntactic complexity; this may be due to the possibility that syntactic complexity remains preserved at preclinical or early phases, as evidenced previously [49]. However, Le et al. [50] analyzed several novels of British writers over time and found some decreases in the use of passive voice, consistent with other previous findings [51]. Therefore it is possible that our samples of the picture description task are simply too short to detect syntactic differences. The same can be said for the lack of findings with respect to fillers, revisions, and repetitions in this group. It is also possible that syntactic differences will be more detectable upon longitudinal analysis of change within individuals.

The greatest limitations of this study include the small sample size and not adjusting for multiple comparisons. We powered our study to detect plausible effect sizes using an alpha of 0.05; adjustments for multiple comparisons would have greatly reduced our statistical power, and increasing the sample size was beyond our resources at the time of analysis. An additional limitation includes the fact that several measures that have been shown to identify impairment in AD (e.g., error monitoring (the proportion of revised errors to total errors), timing aspects such as performance deviations per minute or response-time to word-finding delays) were not analyzed, as we based our measures on those that have been examined in MCI or preclinical AD. Nonetheless, the addition of such measures would add clarity to the limited research available in this population. Future studies will include factor analyses of the many discourse measures examined in previous AD literature in order to more concisely characterize aspects of language and to reduce the amount of multiple comparisons.

The strengths of this study include utilizing connected language analysis measures in a unique population: a cohort of clinically healthy individuals who are at-risk for AD. Thus far, connected language analysis has proven to be useful in detecting deficits at early stages of autopsy-confirmed AD [19] and in preclinical stages of disease in carriers of the E280A mutation [20]. An additional strength of this study is examining groups that are based upon internally-derived norms [29], which allowed us to identify a “preclinical” MCI group. Further examination of connected language using larger sample sizes of the pMCI and CH groups longitudinally may yield important findings regarding improving the ability to detect early language deficits, aiding in differential diagnosis, and providing direction for non-pharmacological communication interventions.

In conclusion, these results suggest that subtle yet detectable changes in connected language may be present during preclinical stages of cognitive decline. For relatively little investment of patient time and effort, language tasks can yield rich information not only about semantic storage and retrieval, but also about the complex cognitive processes involved in producing spontaneous, connected language. Further, these analyses may help to provide tools for differential diagnosis among variants of cognitive decline, such as AD versus Primary Progressive Aphasia [18], or between dementia and other co-morbid conditions such as depression [52]. Future research will examine changes over time, to better understand how such changes correspond to or predict clinical outcomes, document disease progression, and aid in determining strategies and interventions for improved communication between patients and caregivers.