Introduction

Empirical investigation into the temporal ordering of Alzheimer’s disease (AD) signs has largely focused on cerebrospinal fluid and structural magnetic resonance imaging biomarkers [1,2]. Behavioral symptoms, including cognitive and functional decline, appear to follow these physiological changes in many patients [1].

Many investigators have assumed that cognitive changes precede functional changes prior to AD diagnosis [1]. Indeed, the concept of mild cognitive impairment (MCI) was proposed to describe early cognitive impairment prior to the manifestation of frank functional loss sufficient to warrant a diagnosis of dementia [3]. These MCI criteria required that functional abilities be “essentially normal.” Despite this criterion, numerous patients labeled with MCI exhibit some degree of functional impairment [4–6]. In recognition of these observations, the most recent recommendations of a National Institute on Aging and Alzheimer’s Association work group explicitly allow for functional problems in the diagnosis of MCI due to AD. These new criteria require only that patients “generally maintain their independence of function in daily life, with minimal aids or assistance,” thus blurring the temporal ordering of cognitive and functional decline in older adults at risk for dementia [7].

There have been few direct investigations of the temporal ordering of cognitive and functional declines prior to AD. Amieva and colleagues reported descriptively that declines in semantic memory and conceptual formation appeared to precede declines in activities of daily living in a sample of 350 AD converters [8]. In contrast, a recent community-based longitudinal study reported that baseline difficulties with certain informant-reported instrumental activities of daily living among cognitively intact individuals predicted MCI two years later [9]. Further, it has been suggested that a structured clinical interview focusing on functional changes is capable of diagnosing AD prior to the inflection point of cognitive decline [10]. Indeed, standardized neuropsychological tests have been criticized in their ability to predict future functional declines [11]. Thus, it remains possible that functional changes precede measurable cognitive changes, or that there is reciprocal causality between cognition and function prior to dementia onset.

The temporal ordering of cognitive and functional declines following AD diagnosis has largely been ignored. We recently demonstrated strong coupling of cognitive and functional declines both between and within patients diagnosed with probable AD [12]. However, whether there are leading and lagging relationships within those parallel trajectories remains unclear.

The purpose of this study was to investigate the temporal ordering of cognitive and functional declines separately in older adults with or without AD. In both samples, cognition was measured with neuropsychological tests, and function was measured using a self-report questionnaire completed by either the individual (in the sample of non-demented older adults) or an informant (in the sample of AD patients). We used autoregressive cross-lagged panel analysis to compare the predictive abilities of the cognitive variables for function versus functional variables for cognition. This approach allows for an exploration of reciprocal causality between cognition and function.

Methods

Data from two longitudinal studies were used in the present analyses. A sample of initially non-demented older adults was drawn from the Washington Heights/Hamilton Heights Inwood Columbia Aging Project (WHICAP) [13,14]. A sample of prevalent Alzheimer’s disease (AD) cases was drawn from The Predictors Study, a multicenter study of AD progression [15].

Statistical analysis

Descriptive statistics were computed in SPSS version 19 (IBM Corp., Amonk, NY). Bivariate autoregressive cross-lagged panel analyses were conducted in Mplus version 7 (Muthén & Muthén, Los Angeles, CA) using maximum likelihood estimation. Missing data were managed with full information maximum likelihood (FIML), which uses all available data for parameter estimation.

In the cross-lagged models, each variable at each occasion was regressed onto the same variable at the previous occasion (autoregressive paths) as well as the other variable of interest at the previous occasion (cross-lagged paths). In addition, occasion-specific correlations between the two variables of interest were estimated. A schematic of the general model is shown in Figure 1. This framework allows for an exploration of reciprocal causality between cognitive and functional variables.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Schematic of the autoregressive cross-lagged model.

Four time points were analyzed in the WHICAP models. Twelve time points were analyzed in the Predictors model. Refer to the text for the covariates included in each model (not shown).

https://doi.org/10.1371/journal.pone.0073645.g001

In the current study, three cross-lagged models were estimated. The first two examined relationships between function and one of the two cognitive abilities in WHICAP (i.e., memory or language) over an average of five years. In addition, these models were re-run on the subset of 612 individuals who converted to dementia during the study period in order to more specifically examine temporal ordering prior to confirmed dementia onset. The third model examined the relationship between function and global cognitive status in The Predictors Study over 5.5 years. In the subset of 265 participants recruited in 1998, a supplementary model was run in which a time-varying covariate reflecting illness burden was added to the model to confirm that results were not dramatically altered by the consideration of comorbid health conditions. Due to this smaller sample size used in this analysis, only 2.5 years of data were evaluated. Model fit was evaluated with the root mean square error of approximation (RMSEA<0.08), standardized root mean square residual (SRMR<0.05), and the comparative fit index (CFI>.95).

Results

Discussion

The main finding was that cognitive score more consistently predicted subsequent functional ability than vice versa in both non-demented older adults and patients with Alzheimer’s disease (AD). Among non-demented older adults, composites scores of memory or language reliably predicted self-reported functional ability at every subsequent visit, whereas self-reported functional ability only predicted subsequent cognitive scores two-thirds of the time. In addition, standardized regression coefficients were larger for the prediction of functional ability by cognitive domain. Restricting the sample to incident dementia cases, cognitive score predicted subsequent functional ability at two of the three follow-up visits, while functional scores did not predict subsequent cognitive scores at any visit. Among patients with prevalent AD, cognitive score predicted informant-rated functional ability at each subsequent visit, whereas functional ability only predicted subsequent cognition at three out of 11 follow-up visits. Further, the magnitude of association between cognitive score and subsequent functional ability was larger than that between functional score and subsequent cognition for every visit. These findings were independent of chronic illness burden and are most consistent with the view that cognitive declines precede functional declines.

These results validate the use of neuropsychological tests for functional prognosis among older adults with and without AD. It is notable that a previous study reporting poor prognostic ability of neuropsychological tests did not examine cognitive abilities assessed in the present study: word list learning, basic language ability, and verbal abstraction [11]. Together, these results support the importance of assessing memory and language with tests that require greater executive demands to detect the earliest deficits of AD [26].

Results provide minimal evidence for reciprocal causality between cognition and function and clarify our earlier demonstration of tight coupling of cognitive and functional declines over the course of AD by showing that decrements in the ability to perform tasks of daily living temporally follow and are predicted by decrements in cognition [12]. Thus, it appears likely that functional impairment in AD is a direct result of cognitive impairment.

Importantly, our finding that cognitive declines preceded functional declines in the sample of initially non-demented older adults was replicated, and even more pronounced, in the subset of participants who went on to develop dementia. The interaction between cognitive and functional decline appears to begin prior to formal dementia diagnosis. It has been suggested that the Clinical Dementia Rating (CDR), a comprehensive clinical interview focusing on real-world functioning, is sufficient to diagnose dementia prior to the inflection point of cognitive decline [10]. However, it should be noted that the CDR is much more than a functional scale. Not only does it assess activities of daily living, it includes a more wide-ranging clinical interview with both patient and knowledgable informant. It also involves direct interrogation of cognitive abilities (e.g., recall of a name and address). Such a comprehensive clinical interview supplemented by overt cognitive testing represents the gold standard in clinical diagnosis and prognosis. Our results show that in the absence of such thorough clinical assessment, standardized neuropsychological testing is also capable of predicting later changes in functioning.

There were minimal differences in the observed temporal associations in the two populations studied (a community-based study of initially non-demented older adults and a clinic-based study of older adults with prevalent AD). In the initially non-demented sample, associations between cognition and subsequent function were somewhat stronger for later time points, perhaps because these variables became more coupled as cognitive and functional abilities declined. Indeed, the magnitudes of association between the cognitive variables and subsequent functioning at these later time points were similar to those seen between global cognitive status and subsequent functioning in the prevalent AD sample.

Amieva and colleagues (2008) have descriptively shown that declines in cognitive performance appear to precede declines in activities of daily living prior to AD diagnosis [8]. The current findings extend this observation by showing that cognitive scores significantly and independently predicted functional scores at subsequent visits. While the magnitudes of the associations between cognition and subsequent function were similar at later visits, language scores at the initial study visit appeared to be a stronger predictor of functioning at the second visit than memory scores. These results may be consistent with those of Amieva et al. (2008), who reported that verbal fluency and verbal abstraction, both of which were included in our language composite, were the first tests to evidence declines prior to AD conversion in their French sample. Alternatively, function could have been more sensitive to declines in language than memory if both domains declined similarly in the present study.

A main strength of this study is its large numbers of older adults both prior to and following dementia onset. Another strength is the use of panel analysis, which allowed for a direct examination of the temporal ordering of cognitive and functional changes. While cognitive scores were limited to a single measure of global cognitive status in the AD sample, we were able to replicate our results with composite scores of memory and language tests in the sample of initially non-demented older adults. A limitation of analyses on the WHICAP sample was the lack of an informant-based measure of function, as the presence of a reliable informant was not an inclusion criterion for this community-based study. It should be noted that the same pattern of results (i.e., cognitive declines precede and predict functional declines) was found in the Predictors sample, in which function was measured via informant report.

In conclusion, the present study supports the view that cognitive declines precede and cause functional declines in late life. In addition, standardized neuropsychological tests represent a useful marker of functional prognosis during all stages of dementia. While the current study cannot address the temporal ordering of biomarkers as described by Jack et al. (2010, 2013) [1,2], our findings support the hypothesis that cognitive changes precede functional changes prior to AD diagnosis as depicted in that model.