There are three general approaches to measuring everyday function: self-report, informant-report, and performance-based measures. Self-report has been shown to be problematic in individuals with cognitive impairment as evidenced by poor self and informant agreement, a gap that widens with dementia severity (DeBettignies, Mahurin, & Pirozzolo, 1990; Seltzer, Vasterling, Mathias, & Brennan, 2001). Alternatively, several performance-based measures of everyday functioning have been developed in which patients carry out specific, well-defined functional tasks under the direct observation of a trained rater. Some investigators have argued that this is the most valid and reliable method of assessing functional abilities (Giovannetti, Schmidt, Gallo, Sestito, & Libon, 2006). However, performance-based measures are most often administered under artificial conditions in which the individual is prompted to engage in a task and provided with all of the materials they need. Observed behavior under such contrived situations may differ greatly from what the individual does spontaneously in their real and familiar environment. Additionally, most performance-based scales are time consuming and require extensive equipment, rendering them impractical for routine use. More practical in many situations is obtaining ratings by someone who knows the patient well. Use of an informant or proxy to rate an individual's everyday functioning has been shown to be useful in differentiating individuals with dementia from healthy elders (DeBettignies et al., 1990; Isella et al., 2006; Jorm & Jacomb, 1989; Jorm & Korten, 1988; Kemp, Brodaty, Pond, & Luscombe, 2002; Seltzer et al., 2001) and in predicting who will go on to show further decline (Jorm, Christensen, Jacomb, Korten, & Mackinnon, 2001) or develop a dementia (Daly et al., 2000b; Harwood, Hope, & Jacoby, 1997). This method of functional assessment has the advantage of utilizing raters who are familiar with the individual's performance in real-world environments. Although an informant is not available for everyone, the approach is, in most situations, highly cost-effective and time efficient and easily implemented in both clinical settings and large-scale research studies.

A number of informant-based measures of everyday function currently exist. However, all suffer from a number of limitations, one of which has been an overreliance on global indices of functional status that lump potentially disparate functional abilities together. The proposition that there are distinguishable domains of cognitive function and that these can be measured relatively specifically by targeted neuropsychological tests is not controversial. That is, despite ongoing controversies regarding theoretical conceptualization of precise cognitive functions and how they are represented in the brain, it is clear that tests of episodic memory, for example, measure memory function, which is distinguishable from a set of expressive language functions that can also be measured by appropriate tests. However, very few instruments have been developed to specifically measure the everyday correlates of different neuropsychological domains. Those instruments that do exist focus on only one or a limited number of domains (Grace, Stout, & Malloy, 1999; Jorm & Korten, 1988; Williams, 1987).

The predominant model of everyday function in older adults has remained essentially unchanged for the last four decades. It broadly divides activities of daily living (ADLs) into low-level basic self-care behaviors and higher-level “instrumental” ADLs (IADL). Research has generally supported this hierarchical arrangement of functional skills by demonstrating that instrumental ADLs are affected earlier in the course of dementia (Kemp et al., 2002; Richardson, Nadler, & Malloy, 1995; Tomaszewski Farias, Mungas, & Jagust, 2005), whereas basic ADLs are preserved until relatively late in the course (Sclan & Reisber, 1992; Suurmeijer et al., 1994). Studies have also shown that instrumental ADLs with a strong cognitive component can reliably be distinguished from more basic ADLs (Fitzgerald, Smith, Martin, Freedman, & Wolinsky, 1993; Wolinsky & Johnson, 1991). Although the dominant model had divided ADLs into basic and instrumental ADLs, the critical underlying constructs may not reflect the distinction between whether an activity is “basic” or “instrumental” but rather may be based on the relevant abilities that support performance of the activities. It is likely that activities that would broadly be construed as instrumental or higher level functional activities could be further subdivided to reflect relevant underlying cognitive abilities. Thus, an alternative approach hypothesizes that different daily tasks vary in the degree to which they require specific cognitive abilities and that it will be useful to categorize functional tasks according to the underlying cognitive abilities that they require. Some tasks may require mostly episodic memory, for instance recalling items to be purchased at the store, whereas others may require mainly spatial abilities, for example, navigating the route to the store. If everyday function could be fractionated in this way, it would permit rationally based investigations of the relationships between specific types of neuropsychological deficit and specific types of functional impairments. The ability to link domains of daily function to particular domains of cognitive function could lead to a greatly improved understanding of daily function. For example, this might improve our ability to make meaningful predictions about which specific functional declines might result from specific cognitive impairments or about how the nature of functional change may vary across different clinical disorders that have different cognitive/neuropsychological profiles.

Two additional limitations of previously developed measures of everyday function include poor sensitivity to mild functional impairment and to change over time. Most previous functional instruments were developed to assess functional impairments that occur in the midst of a frank dementia, with the focus often being on functional impairments that occur within the moderate and severe stages of disease. With the emerging emphasis on identifying the prodromal stages of dementia [e.g., a state often referred to as Mild Cognitive Impairment (MCI)] in anticipation of disease altering treatments, it is important to be able to detect the very mild functional changes that occur before a dementia can be diagnosed. Finally, increased sensitivity to subtle differences in function will likely result in a more accurate assessment of change over time. This will have the benefit of improving our ability to characterize the patterns of change in function over time and to monitor change in response to treatment.

To address the limitations of existing functional instruments, we have developed a new functional instrument called Everyday Cognition (ECog). The goal was to create a psychometrically rigorous instrument to assess the functional abilities of older adults across a wide range of ability, spanning normal aging through mild to moderate dementia. Particular emphasis was placed on assessing those functional changes that may occur very early in the course of an incipient degenerative disease, for example during the syndrome of MCI. Development of the ECog was guided by an underlying conceptual model that suggests (1) everyday functioning is a multidimensional rather than a uni-dimensional construct and, (2) different domains of everyday function can be measured by identifying functional tasks that rely, to large extent, on particular cognitive abilities. Thus, an a priori goal was to develop a multidimensional instrument capable of measuring impairment and change in domains of everyday/real-world functioning relevant to specific neuropsychological domains: Everyday Memory, Everyday Language, Everyday Semantic Knowledge, Everyday Visuospatial abilities, and three everyday executive domains including Everyday Planning, Everyday Organization, and Everyday Divided Attention. These functional domains were identified because they correspond to well-accepted domains of cognitive functioning and are important in the evaluation of different types of dementia in older adults.

In this paper we describe the development and initial validation of the ECog. Confirmatory factor analysis (CFA) was used to evaluate the construct validity of the ECog, and to determine whether the factor structure of the instrument supports the proposed individual subscales or domains. We expected that items would be highly intercorrelated but that within-domain correlations would be stronger than cross-domain correlations. Thus, individual items would be influenced by a nonspecific factor representing overall level of functional impairment and by an independent domain-specific factor corresponding to the functional domains measured by the ECog. Next, associations between the ECog and established measures of everyday function and disease/cognitive impairment severity were evaluated (convergent validity). Finally, we examined how different clinical groups (healthy older adults, older adults with MCI, and those with dementia) performed on each of the scales of the ECog (external validity). The incremental validity of the domain specific ECog factors in discriminating the clinical groups was also examined.

Although we hypothesized that everyday function is a multidimensional construct, it was also anticipated that the different functional domains would be inherently intercorrelated so that all domains would be influenced by, or represented in, a nonspecific factor. It has been a long held belief that intellectual or cognitive abilities can be represented, in part, by a general, nonspecific factor (the `g' factor), but that there is also remaining variance that can be parceled into more specific domains. Traditional approaches to neuropsychological assessment have generally used measures that include both nonspecific and domain-specific components of variation. While acknowledging the intercorrelations between these domains, this approach generally does not explicitly utilize models to separate specific and nonspecific contributions to the test scores. We used an alternate strategy in this study (bifactor model), that is, we explicitly and independently modeled domain specific and nonspecific contributions to everyday function. This approach is particularly relevant to examining the utility of forming subscales (Reise, Morizot, & Hays, 2007) and has been used to investigate the psychometric properties of other instruments (Chen, West, & Sousa, 2006; Patrick, Hicks, Nichol, & Krueger, 2007; Reise et al., 2007; Stockdale, Gridley, Ware Balogh, & Holtgraves, 2002).

Thus, confirmatory factor analysis, using a bifactor approach was used to examine the latent factor structure of the ECog to determine if there was support for our hypothesized multifactorial model. We first examined the fit of a simple one-factor model, which represented everyday function as a unitary construct, and this model did not adequately fit the data. As such, we then evaluated a variety of different multidimensional models. Subsequent models included a general, nonspecific factor, along with various domain-specific factors that were uncorrelated with the general factor. In this way, the domain-specific factors account for unique variance not explained by the general factor.

All of the multifactorial models fit the data better than the one-factor model, supporting the notion that everyday function should be thought of as a multidimensional construct. In the simplest multifactorial model, which included an everyday memory and a nonmemory factor (in addition to the global factor), one of the fit indices (RMSEA) still suggested an inadequate fit. Expanding the model to include domain-specific factors associated with Everyday Memory, Language, Visuospatial, and Executive function improved the fit indices such that they were all within acceptable ranges. Further dividing the Everyday Executive factor into Planning, Organization and Divided Attention further improved model fit. However, we did not see a similar pattern of improved fit when separating Everyday Language and Everyday Semantic Knowledge into separate factors (the fit of the seven and eight factor models was almost identical). In addition, these two factors, unlike the others were highly correlated. Based on these results we chose to retain the model represented by one general factor and six domain-specific factors (Everyday Memory, Everyday Language, Everyday Visuospatial Function, Everyday Planning, Everyday Organization, and Everyday Divided attention). Thus, the final model we used was a modified version of our a priori model in that we collapsed the Language and Semantic Knowledge factors.

To provide support of convergent validity the relationships between the ECog and traditional measures of everyday function were evaluated. As expected, there was a strong relationship between the ECog general factor and the two established measures of global functional status (the CDR and BRDS). However, the domain-specific factors of the ECog were correlated much more modestly with these existing instruments. Thus, as expected, nonspecific functional impairment accounts for most of the correlation with these global measures; the domain specific ECog components are not strongly correlated with the external measures independent of the nonspecific ECog component. Such findings suggest that the domain-specific factors are measuring something not captured by these traditional instruments.

Next, we examined whether there were clinical group differences on each of the ECog factors. We focused on three groups of older adults: those who are cognitively normal, those with MCI, and individuals with a clinical diagnosis of dementia. Results showed large group differences on the general factor of the ECog. This indicates that there are significant differences in the overall levels of everyday function across these clinical groups, with the normal person showing the least degree of change relative to their baseline, the MCI group showing an intermediate level of functional impairment, and the demented group showing the greatest degree of functional impairment. This was so even though MCI had been diagnosed using standard criteria that excluded cases with clinically significant functional impairment. Thus, this is an important finding in that there are few existing instruments to assess everyday function that are sensitive to the relatively subtle changes that occur in the transition from normal function to MCI and dementia.

When all of the factor scores were entered together into a discriminant function analysis, again the general factor discriminated all three of the groups; however, other specific domains added incremental discriminative power. Specifically, Everyday Memory added to the discrimination of Normal from MCI, whereas Everyday Language helped discriminate dementia from MCI. Such findings provide evidence of incremental validity of both Everyday Memory and Everyday Language. These findings are conceptually consistent with the progression of pathology and neuropsychological impairment that occurs with Alzheimer's disease. That is, the syndrome of MCI in our sample (and in the literature) is most often associated with memory impairment, which is in keeping with the notion that early memory decline is a harbinger of AD secondary to the early involvement of the medial temporal lobe structures. As the disease progresses to include greater cortical involvement, other neuropsychological domains become involved. For example, it is well known that AD pathology typically progresses from the hippocampus to involve temporal lobe cortical regions quite early in the course of the disease (Braak & Braak, 1991), resulting in early language changes. Thus, the fact that functional changes associated with memory discriminate MCI from normal older adults, and language-related functional changes discriminate dementia from MCI is consistent with the pathological and neuropsychological progression of early AD. The fact that other domain-specific factors did not add discriminative power may reflect the fact that our dementia group was rather mildly impaired.

The syndrome of MCI is known to be heterogeneous; therefore, we further examined whether different subtypes of MCI showed different ECog factor profiles. For these analyses, because of the sample sizes, we examined the difference between two groups, amnestic MCI and multiple domain MCI. The amnestic MCI group had predominant memory impairment, whereas the multiple domain MCI group was comprised of individuals who typically had mild neuropsychological impairments on measures of memory and at least one other neuropsychological domain. We found that while the amnestic MCI and the multiple domain MCI groups did not differ in terms of Everyday Memory, the multiple domain MCI group showed greater functional impairment in most of the non-memory functional domains including Everyday Language, Everyday Visuoperception, and Everyday Planning (with statistical trends indicating more impairment in Everyday Organization and Divided Attention). Such findings support the association between impairment in neuropsychological domains and impairment in domain-specific functional domains. Interestingly, the amnestic MCI group showed more impairment on the general, nonspecific factor of the ECog than the multiple domain MCI group. This may seem counterintuitive, but it is important to remember that the domain-specific ECog factors are independent of the general factor. It suggests that the multidomain cases had milder, but more diffuse impairment.

Further evidence of the domain-specificity of these scales might also be obtained by comparing the ECog profiles of various diagnostic groups known to have different cognitive profiles. For example, we hypothesize that although the general, nonspecific functional factor may show similar degrees of impairment across different dementia types who are at similar disease stages, there will be domain-specific differences such that AD is associated with prominent Everyday Memory impairments, frontotemporal dementia is associated with prominent impairments in everyday executive domains such as Everyday Planning, and syndromes such as Primary Progressive Aphasia or Semantic Dementia will be associated with a prominent Everyday Language impairment. Thus, it is possible that the pattern of functional impairment, like the pattern of neuropsychological impairment, will aid in differential diagnosis of these disorders. Our current sample included only small numbers of non-AD dementia types and therefore precluded this type of analysis but this is an area of ongoing investigation at our Center.

Although we hypothesize that to at least some extent the specific everyday cognitive domains will relate to their neuropsychological counterparts, this remains an important empirical question that we are also pursuing. It is likely that there will be complex relationships between neuropsychological functions and everyday cognition. For example, we suspect that there are a variety of different scenarios in which different neuropsychological impairment(s) could lead to similar functional deficits. For example, it may be that neuropsychological deficits in memory lead to problems in Everyday Memory but that deficits in executive functions can also lead to similar changes in Everyday Memory (although in the later case functional changes may also occur across a wider range of functional domains). Importantly this instrument will serve as a tool to systematically collect information on how a person is functioning in different cognitive domains of everyday function, and to test theoretically driven hypotheses about how specific neuropsychological impairments affect specific areas of everyday function.

All of the ECog scales had very low correlations with participants' level of education. This is in contrast to the typically strong association between neuropsychological test scores and education, suggesting that particularly for individuals with very low or very high levels of education, assessment of everyday function using the ECog may represent an indicator of an incipient dementia that is less confounded than cognitive testing by education and related demographic characteristics. In designing the response options for the questionnaire items, we specifically chose to obtain ratings of a person's current level of functioning compared to their own baseline because we wanted to measure new or acquired functional changes, rather than preexisting or lifelong difficulties. This type of response option has been used successfully with other informant-based measures of cognitive and functional change (i.e., IQCODE) and has shown similar low relationships with demographic variables (Del-Ser et al., 1997; Farias et al., 2004; Morales et al., 1997).

We explicitly modeled nonspecific and domain-specific dimensions of independent function in this study. This has conceptual and methodological advantages in that it separates these sources of variation, but has a relative practical disadvantage in terms of the complexity of calculating factor scores. This is because item loadings are weighted so that the general factor is uncorrelated with specific factors, and as a result, computer scoring is required for practical applications. A different approach would use correlated factors defined by the items contributing to the domain-specific factors in this study. This approach would not include a general factor, and nonspecific variance would be included in the factor scores for the specific domains. This second approach is more commonly used in neuropsychology, where scores from different domains are known to be correlated, and domain-specific effects are inferred from relative peaks and valleys across profiles. An advantage of the second approach is that use of a simple summary score for each ECog domain (by summing items and dividing by the number of items completed) would yield domain scores that would closely approximate scores from a confirmatory factor analysis, and thus this approach can be implemented in applied settings without computer scoring. These two approaches are closely related conceptually; factors in the second approach essentially correspond to each domain-specific factor from the first approach being added to the general factor. We have used the second approach in a previous publication (Farias et al., 2006). Ultimately, either approach could be used depending on the specific needs and resources.

There are likely other important dimensions of everyday function, not included in the ECog. To this end, the ECog is not exhaustive in terms of its assessment of all possible important domains. For example, social judgment and self regulation behaviors are not explicitly measured. However, there are a number of other informant-based ratings scales that assess these frontally mediated behavioral syndromes (i.e., FrSBe; Grace et al., 1999).

There are limitations to relying on the reports of informants because they can be subject to the effects of systematic bias. For example, informant characteristics such as mood or degree of caregiver burden (Jorm et al., 1994; Teri, 1997) can affect ratings. Informant report has, however, been shown to reliably differentiate demented from nondemented individuals and such information can be useful in predicting who will go on to develop further changes (Daly et al., 2000b; Monnot, Brosey, & Ross, 2005). These previous findings, along with some of the results in this paper, provide evidence that informants can reliably judge the functional and cognitive abilities of patients. Informants may not be as accurate in rating the everyday cognitive abilities of individuals with only mild changes, and thus there may be a threshold level of functional change that informants can accurately observed. However, the present study demonstrates that informant ratings of individuals with only mild cognitive impairment but not demented, differ both from the informant ratings of cognitively normal elders and those with dementia.