Operationalisation of Mild Cognitive Impairment: A Graphical Approach

A new online tool allows mapping of the different classifications of mild cognitive impairment.

As an opportunity for identifying individuals at risk of developing dementia, MCI is an important concept. Yet lack of consensus criteria has lead to debate about the utility of MCI, resulting in calls for abandoning its diagnosis and adopting an alternative nosology [20][21][22]. Consensus conferences are now being held, even though MCI diagnoses are already used in clinical trials for prevention of Alzheimer disease [23]. The aim of this paper is to develop a framework for mapping the different classifications of MCI using retrospective information, assessing variations in defining criteria.

Creating a Framework for Mapping MCI
The first step to coding MCI is to determine the necessary criteria and thresholds for operationalisation of each definition. We compiled a comprehensive list of those classifications which represent different aspects and definitions of MCI. The necessary components for each were abstracted and formulated into a diagnostic algorithm.
The main problem encountered was that while some classifications have specific criteria for implementation (e.g., amnestic MCI [A-MCI] [13,14] and age-associated memory impairment [5]), others are vague descriptions that require interpretation as to the exact nature of the deficit (e.g., age-related cognitive decline [24]). Further complicating the problem is a lack of specification of screening tools and variability in: (1) the domain of impairment (memory versus nonmemory, single-versus multi-domain deficits); (2) cut-off scores; (3) acceptable restriction on activities of daily living; and (4) exclusion criteria.
Eighteen current definitions of early cognitive impairment were identified in a systematic review of the literature and mapped using a flow diagram as shown in Figure 1. Mapping is completed in two phases: following exclusion of all individuals with dementia, each classification is then operationalised independently. Each classification could be constructed from a subset of 15 different criteria, with memory impairment required as an essential feature in almost all classifications. Surprisingly, no two classification systems map on the same path. Each classification is operationalised in the population and has been previously applied to the Medical Research Council Cognitive Function and Ageing Study (MRC CFAS) [19]. Table  1 outlines how each set of criteria was operationalised in CFAS. To facilitate cross classification comparisons, criteria were consistent in terms of level of impairment unless cut-off thresholds were uniquely specified.
In reading the flow diagram, the classification arrived at depends on the direction of decision at each criteria (yes, criteria required are fulfilled as indicated by a green arrow, or no, criteria are not fulfilled as indicated by a red arrow). If an individual fails to meet the specified outcome for a given criterion they are excluded from further mapping.

Two Examples: Benign Senescent Forgetfulness and A-MCI
For example, following the flow diagram to arrive at a classification of benign senescent forgetfulness (BSF) takes just three steps: from the "START" box you move to the "Demented" box. If the individual does not have dementia (as indicated by the red arrow), you next move to the "Long-Term Memory Problem" box, and if the

Operationalisation of Mild Cognitive Impairment: A Graphical Approach Fiona E. Matthews * , Blossom C. M. Stephan, John Bond, Ian McKeith, Carol Brayne on behalf
individual shows a long-term memory problem with intact recent memory (as indicated by the green arrow), you then arrive at a classification of BSF. (See Figure 2 for the BSF-specific path.) In contrast, to arrive at a classification of A-MCI, seven steps are needed: (1) the individual is not demented (as indicated by the red arrow); (2) there is an objective memory impairment (one standard deviation below age-corrected norms, as indicated by the green arrow); (3) the individual or an informant complains of memory loss (as indicated by the green arrow); (4) there is no impairment in activities of daily living (as indicated by the red arrow); (5) there is no impairment of general cognitive functioning (as indicated by the red arrow); (6) the individual has no underlying medical or neurological condition (as indicated by the red arrow); and (7) there is no impairment in another non-memory domain (as indicated by the red arrow). (See Figure 3 for the A-MCI-specific path.)

Advantages and Disadvantages
The flow diagram details this process for each of the 18 different classification systems, allowing for reproduction across studies. Each concept has been detailed in turn with a criteria-specific graph and the appropriate references on the CFAS Web site (http://www.cfas. ac.uk/mciprogram/). The clinical utility of the concept of mild cognitive impairment depends on the validity of the diagnosis and the ability to predict higher rates of progression to dementia. The advantage of retrospective definition of these concepts is that they are not adjusted by current knowledge or changing criteria, an unfortunate downside of consensus criteria being that they can be influenced by changing knowledge over time. The disadvantage of mapping with retrospective information is that it removes clinical experience from the definition, and the information measured from all individuals must have enough scope to encompass the entire original definition.
A fundamental problem with a multiple system approach is the failure in consistency of classification. Prevalence estimates for each classification have been previously calculated in CFAS using the pathways  shown in Figure 1, and were found to be highly variable (range 0.1%-42%) [19]. Some of this difference results from the fact that not all criteria explain pathological ageing, but rather "normal" ageing. Although the distinction between those definitions associated with normal age-related change and those with pathological ageing is not apparent from prevalence estimates alone, it is seen with lower conversion to dementia in those groups defined by non-pathological classifications. Furthermore, the same individual could be classified as impaired on one system and normal on another, even within criteria that are supposedly investigating abnormal change. This makes the interpretation and comparison of results across studies very difficult, where not only the populations but additionally the criteria chosen to estimate MCI are different.
At first glance, a solution to the complexity of the diagram appears simple: reduce all classification

Criteria Operationalisation
Dementia AGECAT organic symptom level ≥3. This corresponds to a diagnosis of dementia as defined by DSM-III-R [25]. Long-term memory intact/short-term memory impaired CAMCOG remote memory score below the 16th centile score and recent memory score greater than the 16th centile score. Subjective memory complaint Self or informant report. Combined score created from three questions including: (1) Have you had any difficulty with your memory? (self report); (2) Have you tended to forget things recently? (self report); and (3) Has he/she had any difficulty with his/her memory? (informant report). Responses dichotomised into "noncomplainers" or "complainers" (positive response to one or more questions).

Exclusions
Unique to each classification. Could include any combination of the following: history of heart attack, chest pain, angina, stroke, Parkinson disease, intermittent claudication, emotional problems, diabetes mellitus, asthma, arthritis, meningitis, head injury, thyroid problems, pernicious anaemia, depression, anxiety, chronic bronchitis, and high blood pressure. General cognitive decline Mini Mental Examination (MMSE) score ≤21. Gradual decline (present for at least six months) One or more positive responses to the following memory questions, with decline reported as being gradual: (1) Do you have to make more effort to remember things than you used to? What sort of things?; (2) When did you notice this beginning?; (3) Did it come on suddenly?; (4) Would you say there has been a deterioration of memory over a period of more than two years?; and (5) Did these problems with memory begin rapidly or gradually? Minor errors in orientation CAMCOG orientation subtest score below the 16th centile score. Clinical problem (i.e., depression or anxiety) Depression and anxiety both defined as AGECAT symptom level ≥3. Other (non-memory) cognitive impairment Below threshold on one or more of the following subtests of the CAMCOG: orientation, language, attention/ calculation, praxis, abstract thinking, and perception. Impaired activities of daily living Modified Townsend Disability Scale, with an additional three items. Impairment defined as requiring help at least several times per week with washing, cooking, and dressing, or as being housebound. Mental status questionnaire Maximum score of 10 derived from the following questions: (1)  systems to a single concept through the amalgamation of all defining criteria, particularly the measurement of objective memory and the medical (and disability) exclusion criteria. However, this solution assumes that within these criteria there is one that is the best for identification of at-risk individuals. Furthermore, definitions, particularly those of objective cognitive impairment, depend on arbitrary and varying thresholds, frequently with no reference to specific values, methods, or screening measures. In retrospective studies, the mapping of these thresholds will primarily be constrained by study design, though the use of different thresholds can be used to determine the most optimal threshold value to accurately distinguish those individuals at high risk of dementia from those with low dementia risk.

Conclusion
It is time to re-examine the concept of MCI. The diagnostic disparity and the lack of consistency in case definition calls into question what exactly is being captured in each classification. This is a fundamental weakness of research on MCI, as highlighted by the complicated nature of Figure 1. Using this flow diagram, MCI systems can be mapped in other population datasets to investigate: (1) what are the best boundaries for impairment; (2) which tests are most sensitive for measuring each criteria; (3) which criteria, if any, can adequately predict individuals at risk of developing dementia; and (4) would adopting multiple systems across different populations (specialist clinic versus population based) and age groups be more appropriate? It is hoped that graphical operationalisation of the criteria will aid in diagnostic consistency and assist in the visualisation of the current problem, with the aim of formulating a gold standard definition for both research and clinical practice.