Level of cognitive impairment required for diagnosis of MCI?

In the most recent recommendations given for MCI diagnostic criteria (Albert et al., Reference Albert2011), impairment can be present in one or more cognitive domains (see Table 4). The following description of impairment is given: “There should be evidence of lower performance that is greater than would be expected for the patient's age and educational background. If repeated assessments are available, then a decline in performance should be evident over time” (p. 271). One critique of this description has to do with a lack of standardization. Specifically, what constitutes “lower performance?” Since it is not explicitly stated, this question is open to interpretation. Moreover, reference is given to a “decline in performance over time” but no specifications as to what constitutes decline. Does this mean one, five, or ten percentage points in relation to age- and education-matched peers? Once again, this determination is open to interpretation. Another critique of this definition is that race was not mentioned along with age and education despite the use of inappropriate norms being associated with dementia misdiagnosis in some minority groups (Manly et al., Reference Manly, Byrd, Touradji, Sanchez and Stern2004). However, normative data corrected for race is often not available. Therefore, this issue will not be discussed in detail other than to urge clinicians to use measures with normative data corrected for race whenever possible.

Later in the Albert et al. (Reference Albert2011) paper, it is noted: “Cognitive testing is optimal for objectively assessing the degree of cognitive impairment for an individual. Scores on cognitive tests for individuals with MCI are typically 1 to 1.5 SD below the mean for their age and education matched peers on culturally appropriate normative data (i.e. for the impaired domain(s), when available). It is emphasized that these ranges are guidelines and not cutoff scores” (p. 273). The inclusion of this guideline is a clear improvement over previous diagnostic criteria and a step toward standardization. However, the difference between 1 and 1.5 SD below the mean is a wide range. In terms of percentiles, this places the range between the 7^th and 16^th percentile meaning the patient would have to score worse than either 84% or 93% of the population after correcting for age and education. Consequently, even for clinicians attempting to use a more standardized process by following the guidelines, differing applications of the same criterion are possible.

Another issue with this description of impairment is the failure to take into account premorbid intelligence (i.e. estimated intelligence before changes in cognition started to appear). This need to evaluate current cognition in relation to premorbid intelligence is highlighted by the guidelines for the evaluation of dementia and age-related cognitive decline (American Psychological Association, 1998). Although these guidelines are described as aspirational and not enforceable, guideline III-8 (i.e. when measuring cognitive changes in individuals, psychologists attempt to estimate premorbid abilities) is clear in relation to the subject. Without an estimate of premorbid intelligence, determining if a change in cognition is present is difficult. For instance, an older adult with an estimated premorbid intelligence quotient (IQ) falling into the 16^th percentile would be demonstrating no decline in relation to peers using a 1.0 SD cutoff. However, the same older adult would be considered impaired using the recommendations from Albert et al. (Reference Albert2011). This situation would be reversed for an older adult with an estimated premorbid IQ falling at the 84^th percentile. For this patient, performance on cognitive measures falling at the 50^th percentile represents a fairly substantial decline in cognition, and performance falling into the 16^th percentile represents a change of two SD. Without taking into account pre-morbid intelligence, the same impairment threshold would be used when, in reality, the first patient is exhibiting no changes in cognition while the second patient is exhibiting drastic changes in cognition.

Ideally, obtaining an estimate of premorbid IQ would be as easy as reading a report of an intellectual assessment that was administered to the patient sometime in the past. In reality, these kinds of records are almost never available. Instead, one basic and fairly accurate method is to evaluate a skill that is resistant to the changes in cognition associated with MCI and dementia. This is most easily accomplished by evaluating reading ability because reading ability does not significantly differ in relation to peers from childhood to adulthood even in older adults diagnosed with mild dementia (McGurn et al., Reference McGurn, Starr, Topfer, Partie, Whiteman and Lemmon2004). However, minor changes in reading have been noted early in the course of dementia, and these changes increase as the disease progresses (O’Carroll et al., Reference O’Carroll, Prentice, Murray, Van Beck, Ebmeier and Goodwin1995; Cockburn et al., Reference Cockburn, Keene, Hope and Smith2000). Consequently, for patients with moderate to severe dementia, reading ability cannot be relied upon to yield an accurate estimate of premorbid intelligence.

Measures such as the national adult reading test (NART: Nelson, Reference Nelson1982) provide a good example of how to best assess reading to gain an estimate of premorbid intelligence. The NART contains 50 low-frequency/irregular English words (e.g. ache and thyme), and the patient is instructed to read aloud each word with the number of words correctly pronounced comprising a total score. Normative data correcting for age, gender, and education are available (Kiely et al., Reference Kiely, Luszcz, Piguet, Christensen, Bennett and Anstey2011). Other techniques such as using a statistical model designed to estimate premorbid intelligence from demographic variables are also available (Crawford et al., Reference Crawford, Millar and Milne2001). Although reading tests such as the NART have been shown to yield the most accurate estimates of premorbid IQ, measures combining reading ability, best performance, and demographic variables are needed and have the potential to be more accurate than a test of reading ability used in isolation (Griffin et al., Reference Griffin, Mindt, Rankin, Ritchie and Scott2002).

Turning again to the patients example, it can be demonstrated how a lack of standardization of current MCI criteria affects the diagnostic process – even when following NIA/AA recommendations (i.e. 1 to 1.5 SD below the mean). Mary exhibited general cognitive functioning falling at the 45^th percentile and immediate memory falling at the 20^th percentile. These scores are clearly above both recommended cutoffs. However, she exhibited delayed memory falling at the 15^th percentile, and this score lies in the gray area between 1 and 1.5 SD below the mean. Does this mean that impairment is present and a diagnosis of aMCI applicable? Unfortunately, the answer to this question is not clear. It is also difficult to ascertain the significance of Mary's scores because an estimate of premorbid intelligence was not obtained. Is Mary's premorbid intelligence average, below average, or above average? This information would certainly give a better perspective about her current scores. A judgment could be made by taking into consideration occupation and education. However, Mary wed her husband and became a homemaker shortly after graduating from high school. In this instance, enough information is not present for even educated clinical judgments to be made.

What extent of limitations to daily functioning is allowed in MCI?

As highlighted earlier, diagnostic criteria related to daily functioning in MCI are constructed in a manner that assumes delineation between virtually intact functional abilities for MCI and impaired functional abilities for dementia. However, the results of prior research do not support this delineation for a number of reasons (Goldberg et al., Reference Goldberg2010). First, it has been shown that although basic ADLs (e.g. eating, dressing, bathing) remain preserved in people with mild cognitive impairments, higher-order IADLs (e.g. financial and medication management, using public transportation, interacting with technology) are often subtly impaired (Winblad et al., Reference Winblad2004; Okonkwo et al., Reference Okonkwo2008; Triebel et al., Reference Triebel2009). Second, medical and psychiatric conditions that are reliably associated with cognitive changes are, in turn, also reliably associated with changes in everyday functioning (Goldberg et al., Reference Goldberg, Ragland, Torrey, Gold, Bigelow and Weinberger1990; Heaton et al., Reference Heaton1994). Third, given that cognitive decline is associated with changes in everyday functioning, the cognitive impairments characteristic of MCI should affect functional skills to some level (Goldberg et al., 2010). In sum, despite the dearth of research related to the topic, the view of preserved daily functioning in MCI appears problematic and further research is clearly needed.

In addition to issues with the current diagnostic criteria, the evaluation of daily functioning in MCI is flawed because standardized measures sensitive to minor functional changes are not commonly utilized (Goldberg et al., Reference Goldberg2010). At present, the most common method for garnering information about the everyday functioning of a patient is to interview a close relative. However, the reliability of this method suffers because of halo effects, memory lapses, and limited knowledge that can lead to an overestimation of the functional capacity of MCI patients (Loewenstein et al., Reference Loewenstein2001). One way to increase the utility of functional evaluations would be to introduce the use of performance-based measures into standard neuropsychological test batteries. Recent research has demonstrated that patients with MCI show significantly compromised functional abilities when evaluated with performance-based measures, and this trend extends to patients identified as having intact functioning by family members (Goldberg et al., 2010). Thus, it appears that simply asking the family about daily functioning is not a reliable method for determining functional abilities; subsequently, a quantitative representation of functional skills is needed for the purpose of diagnostic clarification.

Cognitive complaint as a criterion for MCI?

The evolution of subjective memory complaints as a diagnostic indicator for MCI is highlighted in Table 1. For the original Petersen criteria, the subjective memory complaint needed to come from the patient. Thus, a diagnosis relied on the patient's level of awareness and/or willingness to acknowledge memory changes. This requirement was changed in MCI-R criteria to include complaints by family members about functioning in other cognitive domains. For the most current MCI diagnostic recommendations (i.e. NIA/AA), reports can originate from the patient, the patient's physician, or other informants.

For older adults who are socially isolated and do not see their families or physicians except on rare occasions, the opportunities for others to notice changes in cognition are limited. This can also cause the family to disregard or not recognize signs of cognitive decline or to give more credence to extraordinary but low-frequency events (i.e. overreliance on salient data). Furthermore, many older adults receive medical attention in a primary care setting. However, the length of the average primary care visit does not leave much time for a nurse or physician to notice mild changes in the cognition of an older adult who is seen only a few times a year, even if that often (Flocke et al., Reference Flocke, Frank and Wenger2001). In fact, 50%–66% of all patients of dementia are missed in primary care, and patients with mild to moderate dementia make up the largest percentage of missed patients (Boustani et al., Reference Boustani, Peterson, Hanson, Harris and Lohr2003). Combined with the fact that detecting the mild changes in cognition associated with MCI is more challenging than detecting the changes associated with mild to moderate dementia, it is not difficult to see why relying on the judgment of family members or physicians may be problematic. Furthermore, although these issues have been highlighted in regard to cognitive complaints, it is not difficult to see how similar issues can influence the evaluation of a patient in relation to all of the individual MCI criteria.

Issues with the standardization of this criterion are also present and using terms such as “subjective” or “concern” means the definition of these terms will vary from person to person. In a review, Abdulrab and Heun (Reference Abdulrab and Heun2008) identified 44 studies that contained a clear reference to subjective memory impairment. Of these 44 studies, at least five different methods were used to assess for a cognitive complaint. In response, Abdulrab and Heun (Reference Abdulrab and Heun2008) called for a standardized set of criteria to be validated for use in future research. In addition, the practicality of collecting self- and informant-reports is questionable and up to 27%–34% of older adults without dementia or depression report memory impairments (Toblansky et al., 1995; Jonker et al., Reference Jonker, Launer, Hooijer and Lindeboom1996; Riedel-Heller et al., Reference Riedel-Heller, Matschinger, Schork and Angermeyer1999). In the general older adult population, these complaints have been noted to increase with age from 43% of adults aged 65–74 years to 88% of adults over 85 years (Larrabee and Crook, Reference Larrabee and Crook1994). Thus, cognitive complaints are likely more a function of normal aging than of objective impairment.

In addition to the previously mentioned issues, the incremental validity added to the diagnostic process by a cognitive complaint is questionable. In a review by Reid and Maclullich (Reference Reid and Maclullich2006), the ability of subjective memory complaints to predict objective deficits in performance on tasks of memory was examined. In this review, both cross-sectional and longitudinal studies examining the association between subjective memory complaints and actual memory impairment were selected for inclusion. Of the six cross-sectional studies discussed in the review, two used a single question to assess for subjective memory complaints. Positive associations were reported in these two studies. However, Reid and Maclullich (Reference Reid and Maclullich2006) noted that little can be inferred from these associations because a single question was used to assess for subjective memory complaints, and a comprehensive neuropsychological battery was not utilized in either study. In the other four cross-sectional studies, comprehensive measures were utilized, but weak associations were reported in two studies while no associations were reported in the remaining two studies. Consequently, associations between subjective memory complaints and actual cognitive impairments cannot be adequately examined using the cross-sectional studies available in the literature. Using available longitudinal studies, Reid and Maclullich noted that subjective memory complaints can be used to predict future impairment but not necessarily current impairment. Although more comprehensive measures of cognition were used in these studies, the evaluation and definition of subjective memory impairments varied considerably from study to study. Moreover, the majority of studies utilized measures of subjective memory impairment that have not been adequately scrutinized or validated.

The association between depression and subjective memory complaints was not explored in any of the six cross-sectional studies discussed in the Reid and Maclullich (Reference Reid and Maclullich2006) review. For the longitudinal studies “variable assessment of depressive and anxiety and personality variables” (p. 478) were noted. Clear associations were noted in the studies examining subjective memory complaints and depression, anxiety, and personality variables. In fact, these variables were more strongly related to the reporting of memory impairment than was performance-based cognitive functioning. This demonstrates that variables other than cognitive impairment can influence human judgment and calls into question the extent that psychological variables are affecting reports of memory impairments over and above actual cognitive impairment. Consequently, the value of including a cognitive complaint as a diagnostic indicator for MCI needs to be questioned.

The findings of the Reid and Maclullich (Reference Reid and Maclullich2006) review largely support an earlier review of the literature on the association of memory complaints and impaired cognition. In the earlier review, a weak association was noted between subjective memory complaints and objective cognitive impairment (Jonker et al., Reference Jonker, Geerlings and Schmand2000). Subjective memory complaints were, however, predictive of future decline. Noted in both the Reid and Maclullich (Reference Reid and Maclullich2006) review and the Jonker, Geerlings, and Schmand (Reference Jonker, Geerlings and Schmand2000) review, memory complaints had a consistently stronger association with depression than with cognitive impairment. Memory complaints were also associated with age, gender, and education.

To conclude, the issues with using a cognitive complaint as a diagnostic indicator are well demonstrated by the case of Mary. Recall that Mary reports her cognition and memory to be “just fine.” Consequently, using original Petersen criteria, she would not be a candidate for a MCI diagnosis. Contradictions are found in the reports of her family: her daughter reports her mother's memory has declined more than would be expected of someone her age, but her son reports an opposing opinion. Between the two family members, whose judgment is to be deemed as more valid? If a cognitive complaint was not part of the diagnostic criteria, this question would not require consideration.

Effect #1: Inconsistent prevalence and incidence estimates

Earlier it was noted that one effect of the lack of standardization of diagnostic criteria could be that the samples used in MCI research may, in fact, be representative of quite different populations. This critique is well supported by current research on the prevalence and incidence of MCI in the general population. From 1995–2009, nine studies on MCI incidence rates were conducted in North America, South America, and Europe. Although each of these nine studies used similar diagnostic criteria, a lack of standardization led to the application of these criteria being varied from study to study. As noted in a review by Luck et al. (Reference Luck, Luppa, Briel and Riedel-Heller2010), MCI diagnostic criteria were modified in four of the nine studies. In three studies, a subjective report of cognitive impairment was not required, and in one study a subjective report of cognitive impairment and preserved daily functioning were not required (The definition of daily functioning used in this study was similar to the most recent NIA/AA recommendations.). Impairment thresholds also varied among studies, and performance <1.5 SD below the mean, <1.0 SD below the mean, and less than the 10^th percentile in relation to age-, education-, and sometimes gender- and race-corrected scores were all used (Luck et al., Reference Luck, Luppa, Briel and Riedel-Heller2010).

Given how a lack of standardization has led to different applications of MCI diagnostic criteria in incidence studies for both original Petersen and MCI-R criteria, it is not surprising that the range of estimates of MCI incidence is quite large. For any MCI type, estimates range from 8.5–76.8 per 1,000 person-year (Larrieu et al., Reference Larrieu2002; Busse et al., Reference Busse, Bischkopf, Riedel-Heller and Angermeyer2003; Lopez et al., Reference Lopez2003; Solfrizzi et al, Reference Solfrizzi2004; Tervo et al., Reference Tervo2004; Verghese et al., Reference Verghese2006; Caracciolo et al., Reference Caracciolo2008; Manly et al., Reference Manly, Tang, Schupf, Stern, Vonsattel and Mayeux2008; Ravaglia et al., Reference Ravaglia2008; Chaves et al., Reference Chaves, Camozzato, Godinho, Piazenski and Kaye2009). Estimates for aMCI range from 8.5–40.6 per 1,000 person-year while estimates of non-aMCI range from 28–36.3 per 1,000 person-year. Higher estimates were noted in studies that did not require a subjective report of cognitive impairment. As can be seen from these estimates, the incidence of MCI in the population is not well understood and reflects a need for the standardization of the diagnostic criteria. Age, education, and vascular risk factors were consistently associated with increased incidence of MCI. No studies on the incidence of MCI using NIA/AA criteria have yet been conducted.

Several studies have been conducted comparing the prevalence of MCI in the population according to both MCI and MCI-R criteria. As with the incidence studies discussed above, the interpretation of the criteria vary from study to study and highlight the need for standardized criteria. For instance, in a sample of 3,327 older adults over the age of 75 years who were receiving care from a general practitioner, the prevalence of MCI was 15.4% using Petersen criteria and 25.2% using MCI-R criteria (Luck et al., Reference Luck2007). The SIDAM was used to evaluate cognitive impairment, and impairment was considered present if performance fell more than 1 SD below the mean using age- and education-corrected scores. The SIDAM-ADL scale was used to evaluate daily functioning, and patients with one or no impairments on the 14 SIDAM-ADL items were considered functionally unimpaired. Significant risk factors included age, vascular disease, depression, and APOE4 genotype.

In a study of 883 older adults over the age of 60 years receiving care from a general practitioner, the prevalence of MCI was 16.6% using MCI-R criteria but fell to 3% when using original Petersen criteria (Artero et al., Reference Artero, Petersen, Touchon and Ritchi2006). A computerized neuropsychological evaluation assessing a number of cognitive domains was used to evaluate for cognitive impairment. Patients were considered impaired if performance fell more than 1.5 SD below age- and education-corrected scores on a memory or other cognitive task. Daily functioning was required to be “essentially normal” (p. 467). However, specifics about evaluation of daily functioning were limited to the following quote: “A validated ADL scale with finely graded hierarchical subscales was used to assess alterations in everyday functioning in collaboration with both patients and caregivers” (p. 466).

The two studies outlined above were chosen to highlight how poor standardization can lead to differing estimates using both Petersen and MCI-R criteria in a similar sample (i.e. patients of general practitioners). In the study by Luck et al. (Reference Luck2007), the authors judged cognitive impairment to be present if scores fell <1 SD below the mean using age- and education-corrected scores on a neuropsychological test. In the Artero and colleagues’ study, the authors did not judge impairment to be present unless age- and education-corrected scores were <1.5 SD below the mean. Evaluation of daily functioning was also not uniform between the two studies. In Luck and colleagues’ study, a patient's daily functioning was judged as impaired if impairments in two or more areas were displayed. However, daily functioning was required to be essentially normal by Artero and colleagues. Although comparisons between the prevalence estimates noted in these two studies are not appropriate due to the difference in ages of the samples (75+ vs. 60+), it can easily be seen how a lack of standardization would lead to different prevalence estimates using even an identical sample. No studies on the prevalence of MCI using NIA/AA criteria have yet been conducted.

Effect #2: Varying estimates of conversion from MCI to dementia

As with estimates of incidence and prevalence, the lack of standardization of MCI diagnostic criteria has contributed to estimates of the rates of conversion from MCI to dementia to vary widely in the research literature. In a study using original Petersen criteria, 32.2% of patients diagnosed with MCI progressed to dementia within two years (Amieva et al., Reference Amieva2004). In this study, general cognitive functioning was judged to be intact for patients with at least a primary school education scoring >25 on the MMSE. For patients with less education, general cognitive functioning was judged as intact if scores were >23 on the MMSE. For both education groups, domain specific impairment was judged as present if scores were <1.5 SD below the mean using age- and education-corrected scores on either a test of working memory or semantic fluency. No dependency for basic ADLs and slight dependency for higher-order IADLs were also required for an MCI diagnosis. In another study in which a differing interpretation of original Petersen criteria was used (i.e. MMSE scores <25 regardless of education for intact general cognitive functioning, 1.0 SD cutoff for domain specific impairment, and normal daily functioning), a conversion rate of 27% was noted over the course of ten years, and only 15% of patients progressed to dementia over the first two years (Ganguli et al., Reference Ganguli, Dodge, Shen and DeKosky2004).

MCI-R criteria have been noted as exhibiting better predictive validity of conversion to dementia (i.e. sensitivity of 95% and specificity of 66%) than original Peterson criteria (i.e. sensitivity of 5% and specificity of 91%) when the same impairment cutoffs are used for both groups (Artero et al., Reference Artero, Petersen, Touchon and Ritchi2006). However, issues are also present with MCI-R criteria as their application can lead to strikingly different conversion estimates. For instance, when using cutoffs of <1 and <1.5 SD below the mean on the SIDAM for both Petersen and MCI-R criteria, the best predictive validity (i.e. sensitivity of 74% and specificity of 73%) was observed when MCI-R criteria were coupled with a 1 SD cutoff (Busse et al., Reference Busse, Hensel, Guhne, Angermeyer and Riedel-Heller2006). When using a 1.5 SD cutoff, specificity increased to 86.1% but at the expense of a decrease in sensitivity to 40.4%. Overall, conversion rates varied by 45% depending on which diagnostic criteria were used and how they were applied. No studies of rates of conversion from MCI to dementia using NIA/AA criteria have yet been conducted. Of course, given that MCI and dementia lie on a continuum and that progression from normal cognition to impaired cognition is a progressive process, a certain level of arbitrariness exists for any definition of MCI and is well illustrated by the debate over a cognitive cutoff. However, when cutoffs are used that best predict conversion and, thereby, best differentiate between possible underlying causes of MCI (i.e. degenerative vs. metabolic) (see Table 3) this arbitrariness is mitigated by the potential to identify patients with MCI due to a primary progressive etiology. This issue of the arbitrariness of cognitive cutoffs also extends to conversion from MCI to AD, given that both lie on a continuum and that progression is a slow process. However this discussion is beyond the scope of this paper.