Montreal Cognitive Assessment

Abbreviation: MoCA

The Montreal Cognitive Assessment is a cognitive screening instrument that was designed by Ziad Nasreddine and presented in a journal article in 2005.¹⁾ The instrument serves to differentiate between different degrees of severity of a neurocognitive disorder in the sense of DSM-5. The Mini Mental State Examination (MMSE), which is otherwise frequently used for this purpose, does allow sufficiently good differentiation in the lower performance ranges, that is, in moderate and severe impairment. In the upper performance range, however, particularly when distinguishing a mild neurocognitive disorder (“mild cognitive impairment”, MCI)²⁾ from the neurocognitively healthy range, it measures relatively coarsely. With the MoCA, Nasreddine wanted to create an alternative to the MMSE specifically for this field of application. The measurement range of the MoCA essentially corresponds to that of the MMSE, that is, a very broad range between “extremely impaired” (0-5 points on the MoCA) and “normal” (26-30 points), and it is thus suited to the intended staging of mental performance over a very wide range. It is most often used for the rapid estimation of the cognitive performance of patients with incipient dementia or mild cognitive impairment. In the sense of staging it fulfills this purpose; for a differentiated view of individual areas of impairment, more complex instruments are needed. In 2000, just under 900 scientific studies appeared in which the MoCA was used (for comparison: for the MMSE there were about 1500). The trend in the use of the MoCA is rising.

The test material for the MoCA has always been available for download from the author's website. In the meantime, the site has developed into a professionally made portal that promotes the test and markets the associated training and the app. At present (April 2021), access to the test material is in principle possible only after registration and a one-hour training followed by certification. There are, however, exceptions for neuropsychologists. The test material is available in many different languages and versions; all translations and adaptations are certified personally by Nasreddine. The test material is available in two forms, as a paper version or as an app. The German paper version consists of a one-page test sheet and a four-page instruction for administration and scoring. The original version (which is by far the most frequently used in publications so far) is called version 7.1; parallel versions with different but comparable items are versions 7.2 and 7.3. There is now also a newer series of versions (8.1 to 8.3) with improved instructions and templates. There is only little information on the website about the comparability of the different versions. The app has been available for the iPad since 2017.

The copyright of the MoCA is held by Ziad Nasreddine. Administration, scoring, and interpretation are permitted only after prior training. For clinical purposes and for education, the test material can be used and reproduced without special permission. For commercial projects a specific permission is required, as it is for use in research projects. Details can be found on the website.

The administration is described in detail, as is the scoring, on four to five pages in the instructions for each version, which can be downloaded.

The number of correct answers is the test raw score. It ranges from 0 to 30.

In principle, subscores can also be calculated for the domains of memory (MoCA Memory Index Score, MoCA-MIS), executive functions (EIS), visual domain (VIS), language (LIS), attention (AIS), and orientation (OIS). Since no norms exist for these indices, however, they are not evaluated in TDB2Online.

Like the MMSE, the MoCA does not measure a normally distributed cognitive ability in a population, but grades basic cognitive abilities from a “normal state” down to a very low performance level. The upper end of the measurement range is arbitrarily fixed at a “healthy, unimpaired” level, the lower end at the loss of all cognitive abilities. In contrast to the MMSE, the MoCA includes more items of medium difficulty. Nevertheless, one must expect that the raw score distributions, at least in healthy younger persons, will not be normal but markedly negatively skewed.

For the MoCA we found no published raw score distributions from random samples. There are, however, two studies that have published percentile distributions in sufficient resolution, a large one with 5802 persons (Kenny et al., 2013)³⁾ and a considerably smaller one with 700 persons (Ojeda et al., 2016)⁴⁾. From these, the original distributions can be approximately back-calculated. Kenny et al. (2013)⁵⁾ give in Table 3 of their paper the raw scores for seven percentile points, eight age groups, and three education levels. From these data, raw score distributions were back-calculated, averaged with weighting across the education groups, and smoothed. The result is shown in Figure 1. The necessary information on the sample sizes of the education groups was found in another paper by the group (Kearney et al., 2011)⁶⁾.

Figure 1: Back-calculated distribution shape of the MoCA raw scores in the study by Kenny et al. (2013) for eight age groups (group means)

It can be seen that the distributions become more skewed (more negatively skewed) the younger the persons are. This is quite simply because the younger examinees perform better and their scores move further toward the upper edge of the distribution. Overall, the distributions of the MoCA are more symmetrical than those of the MMSE, because the means do not come quite as close to the upper raw score limit.

Ojeda et al. (2016)⁷⁾ give, in their Table 3, scaled scores directly (mean 10, standard deviation 3, converted from the percentile values) for nine age groups. Considerably fewer persons fall into each age group than in the Kenny study, which is why the distributions turn out coarser. In Figure 2 they are shown in smoothed form. The age range in this study extends from 18 to 86 years, in Kenny et al. (2013) only from 50 upward. Essentially the same distribution shapes are seen, fairly symmetrical at higher ages and negatively skewed in younger persons.

Figure 2: Back-calculated distribution shape of the MoCA raw scores in the study by Ojeda et al. (2016) for nine age groups (group means)

Since the MoCA is essentially applied only to older persons and since we also found only few data for young persons in the norming studies presented below, we form the performance scores on the basis of the 60-64-year-olds, just as for the CERAD tests. It can be seen in both figures that the raw scores are skewed in this age group. In the performance score grid of the TDB2 test profile, a nonlinear mapping of the raw scores onto the performance scores will therefore result.

For the linearization of the scale we will therefore have to draw on these two studies. First, however, we want to look at all available norming studies individually, so that we can estimate how the norms of the studies by Kelly et al. (2013) and Ojeda et al. (2016) compare with all the others.

The original publication of the MoCA from 2005⁸⁾ contained data from two clinical groups (Alzheimer dementia, mild cognitive impairment) and a small group of healthy examinees (N=90). The age range of the examinees was not given; the mean age was 73 years with a standard deviation of 7 years. It was a simple convenience sample without further characterization (“recruited from the community”). The original publication contained no breakdown by age. Education was taken into account only in that one point was added to the raw score for persons with less than 13 years of schooling. The focus of the paper was on demonstrating a better ability of the MoCA to differentiate, compared with the MMSE, especially between mild cognitive impairment and cognitively healthy persons. A cutoff score of 26 (values below 26 indicate reduced performance) proved optimal for sensitivity and specificity in the paper. The problem of a dependence of the MoCA score on age and education, as is naturally to be expected and also showed up in later norming studies, was not recognized in the original publication. In follow-up studies, the level of the cutoff score was regularly called into question.

Since the norms of the original paper (only mean and standard deviation for the entire control sample) are not broken down by age, they are not considered for the meta-analysis of the norms. In Figure 3, however, they are included for comparison.

For the MoCA there are many norming studies, including in comparison with other internationally used test procedures. The fact that the instrument was translated into many languages within a short time and is available for clinical use at no further cost certainly contributed to this. Already in the first application studies by other authors, however, it became apparent that the cutoff scores published in the original paper could apply only to well-educated, above-average intelligent persons. This too made the need to create “own”, country- or culture-specific norms evident.

In March 2021 we searched the international literature for norming studies on the MoCA. We found 19 studies that examined examinees whose school education is roughly comparable to that in Germany. These come predominantly from countries of the Western world. Studies from countries in Africa, Asia, and South America were included only if the sample was comparable with regard to years of education. The studies found can be divided according to whether a representative sample was aimed at (either a random sample or at least a sample stratified on population data) or whether merely a sample that happened to be available was examined. The first group (representative samples) comprises 7 studies, which are discussed first.

In compiling the norms for TDB2Online, an attempt is always made to calculate population-representative norms stratified only by age. For the MoCA, norms that are also stratified by sex and/or by years of education were available in almost all studies. We always tried to average with weighting across these additional strata.

Lu et al. (2011)⁹⁾ calculated norms for the MoCA within a large longitudinal study (China Cognition and Aging Study) in selected cities and rural zones of China. The study is based on a stratified cluster sample. All 6283 persons of the sample (age 65-103 years) were examined clinically, and persons with mild cognitive impairment, dementia, or other illnesses that can have an influence on cognitive variables were excluded. The MoCA was culturally adapted to the Chinese language and script in five places. The results are presented broken down for the four school education groups of 0 years, 1-5 years, 6-9 years, and 10 or more years. Methodologically, the study is one of the better ones in our overview. However, the total sample is not comparable in terms of education with samples from Central Europe. We might have included only the results of the two highest education groups in the meta-analysis of the norms (N=3663). Unfortunately, the standard errors reported in Table 2 of the paper yield such high standard deviations in all age groups, especially in the ranges 65-69 and 70-74, that comparability of even these education groups with the other studies is not given. The study was therefore not considered for the norming.

Freitas et al. (2011)¹⁰⁾ examined a stratified random sample of 650 persons over 25 years of age from Portugal. Care was taken to exclude persons with possible illness-related cognitive deficits. The paper presents the means and standard deviations of the MoCA raw scores as a function of education level (groups 1-4, 5-9, 10-12, and over 12 years) and age (groups 25-49, 50-64, and over 64). In the meta-analysis we included only the results of the three highest education groups (N=394).

The study by Kenny et al. (2013)¹¹⁾ already mentioned above is based on “The Irish Longitudinal Study on Ageing” (TILDA). Here a nationwide random sample of persons living at home was surveyed (N=5802). In this study, too, ill persons were excluded, although less strictly than in other studies. One exclusion criterion was based, for example, on the MMSE and excluded only persons with a raw score below 10. Persons with milder dementia could thus already enter the sample. TILDA works with complex models. Table 3 of the paper gives the model-estimated expected values for the 5th, 10th, 25th, 50th, 75th, 90th, and 95th percentiles as well as means and standard deviations for three education levels and 8 age levels. Another study by the research group (Kearney et al., 2011)¹²⁾ contains the necessary demographic data that make it possible to extrapolate the data within the age groups, with weighting, to the entire sample examined.

The study by Narazaki et al. (2013)¹³⁾ was directed at all residents of the Japanese city of Sasaguri who were over 64 years old and did not live in a care home, a total of 4979 persons. This study too is part of a larger cohort study. After contact in writing, 1977 persons participated validly in the MoCA. The persons were not examined clinically; exclusions because of dementia were made on the basis of self-reports. The results are reported in Table 1 of the paper in overlapping 10-year groups but can be linearly interpolated to 5-year groups.

In the study by Ojeda et al. (2016)¹⁴⁾, also already mentioned above, the authors examined 700 persons aged 18 to 86 years, predominantly from the Basque Country (60%) as well as from the city and surroundings of Valencia (21%); the rest are distributed across many provinces of Spain. The study is said to be representative of all of Spain with regard to demographic composition, even if this certainly does not apply to spatial origin. The study arose as a subproject of “Normacog”, a multicenter norming study of neuropsychological test instruments in Spain. Patients with dementia were excluded on the basis of a telephone interview. The results are available as means and standard deviations per age group; in addition, there are age-dependent percentile distributions. Thirteen percent of the participants had a school education in the range of 0 to 6 years, certainly predominantly those in the higher age groups. Since the results are not reported for the education groups individually, an exclusion was not possible here.

Borland et al.¹⁵⁾ reported in 2017 on a Swedish norming of the MoCA that ran as a follow-up of members of the “Malmö Diet and Cancer Study”, a prospective cohort study. As the initial sample (N=28,000), about 40% of the inhabitants of Malmö served in the 1990s. From this, 6103 persons were randomly selected in the years 2007 to 2012, of whom 860 were also tested with the MoCA in the later course of the investigations. Persons who achieved values below 24 on the MMSE or values below -1 sigma on a test of cognitive speed, or who reported cognitive deficits, were examined carefully clinically, with CT, laboratory tests (blood and cerebrospinal fluid), and with an extended neuropsychological test battery. Persons with any form of neurocognitive disorder were excluded from the norming study. As a result, the sample shrank from 860 to 758 persons. The results are reported in overlapping 10-year groups in the age range from 65 to 85. They can be interpolated to 5-year groups.

The next study comes from the Czech Republic (Kopecek et al., 2017)¹⁶⁾. This study too is part of a larger longitudinal study. The authors collected a stratified sample of persons over 59 years of age in different regions of the Czech Republic. Persons with illnesses that can affect cognitive functions were excluded. Of the 568 persons recruited in this way, in a second step all were excluded who showed values below -2 sigma in two or more of 5 procedures. This further reduced the sample to 540 persons. Unfortunately, the results are given only very coarsely in the two age groups 60-74 and >74 and were linearly interpolated and extrapolated to 5-year groups for presentation in TDB2Online.

In all further studies, the study authors either did not take any great pains to plan the origin of the sample and adapt it to population data, or there are other reasons why the MoCA norms are not representative of an identifiable population. We discuss the publications again in chronological order.

The study by Rossetti et al. (2011)¹⁷⁾ is often mentioned in the literature, probably because it was one of the first independent studies of the MoCA on a large sample. It is based on a sample from the Dallas Heart Study, a multiethnic longitudinal study that aimed to find risk factors for cardiovascular disease. More than 50% African Americans were deliberately included in it, because their risks for cardiovascular disease were to be examined in particular. The oversampling of African Americans was not corrected in the norming study for the MoCA (in other studies of the group it was). Because of its composition, which is not representative of the population, we did not include this study among the norm sources. In 2017 a further paper by the group appeared (Rossetti et al., 2017)¹⁸⁾, in which MoCA norms were calculated only for the African Americans of the 2011 study. This norm source, too, is not considered for the TDB2 norms.

Gluhm et al. (2013)¹⁹⁾ published a norming study for the MoCA and MMSE on a relatively small sample aged between 20 and 89 years from the area around San Diego. The mean years of education of the sample were 15.4, thus very high, even taking into account that the age range of the sample is very wide. The paper prints means and standard deviations in 10-year groups.

Conti et al. (2015)²⁰⁾ examined 254 relatives of patients of a memory clinic and a senior center in Bologna, aged between 60 and 80 years. The examinees had to have an MMSE score of at least 24 and normal memory performance. The results are reported in four age groups. On average, the examinees had about 10 years of education, which in this age range is probably below the population average.

From Israel (Oren et al., 2015)²¹⁾ there is a very small convenience sample of initially 54 examinees between 60 and 82 years. On the basis of 5 further tests, 11 persons were excluded because they had achieved a value below minus 1.5 sigma in one of the tests. The MoCA means and standard deviations are reported in two age groups.

From Italy there is a further study (Santangelo et al., 2015)²²⁾ of 415 persons aged between 21 and 95 years, predominantly from Naples, Milan, and Siena, who had no previous neurological or psychiatric illnesses and whose MMSE was below a value that is not specified more precisely (reference to another paper, in which, however, it is not stated more precisely either). Table 3 of the paper gives M, s, and N of the MoCA scores for 7 age groups in 10-year groups between 20-29 and over 79. There is a problem with the results of this paper: over the entire age range, the MoCA means lie far below those of all other studies. The difference is greatest in the younger age groups. According to the sample description, in particular according to the years of education given, this paper should not differ greatly from others. Even if one considers only the highest education group among the 20-29-year-olds, those with more than 13 years, their mean (25.7) is still far below the means of the other studies. We searched for a correction of the data in subsequent volumes of the journal but found nothing. The values of the paper were not considered for the norms.

Malek-Ahmadi et al. (2015)²³⁾ published a smaller study with 205 participants in a longevity study from the area of Phoenix, Arizona, USA, aged between 70 and 99 years. Persons with an MMSE value below 26 were excluded; dementias and other neurological diseases were also exclusion criteria. The sample is thus highly selected. The average duration of education is not given but can be roughly presumed to be about 14.5 years from Table 2 of the paper. The MoCA means and standard deviations are reported in three 10-year age groups.

Larouche et al. (2016)²⁴⁾ collected, in the Canadian province of Québec, in 12 institutions, data from healthy examinees who had also taken part in testing with the MoCA within the framework of studies with other research questions. In total, the data of 1019 persons aged between 41 and 98 years entered the norming study. All persons had also been tested for cognitive deficits with various instruments. The mean years of education were 14.4, clearly above the expected value for Québec, as the authors write. Simple means per age group were not tabulated. The entire analysis is based on a multiple regression analysis whose equation is published. In the electronic supplement to the paper there is also an Excel program with which the expected values of the MoCA can be calculated in the 5-year groups customary in TDB2. Sex and years of education must also be entered. In the article, the authors mention a population statistic of the province of Quebec for the mean years of education per birth year. We inserted these values into the formula and thereby obtained a quite good estimate of the population expected values. They should not lie far from those of a stratified sample, which can also be seen in Figure x.

Konstantopoulos et al. (2016)²⁵⁾ examined a convenience sample of 710 persons aged between 20 and 85 years from 6 larger cities in Greece and from 2 cities in the Greek part of Cyprus. The authors placed great emphasis on excluding ill persons with possible cognitive impairments and cognitively less gifted persons (those with a value below minus 1.5 SD in one of three tests). The results are reported in 10-year groups and three education groups and were averaged with weighting across the three education groups.

Thomann et al. (2018)²⁶⁾ examined 283 persons aged between 65 and 91 years who had shown interest in participating in studies of a dementia outpatient clinic in Basel. Dementia and other illnesses were carefully excluded. Means are available in four 5-year groups. The mean years of education, at 13.6, are relatively high for this age group.

The study by Sachs et al. (2020)²⁷⁾ is similarly structured. Here a total of 6600 examinees in 32 Alzheimer Disease Centers in the USA were examined with, among other instruments, the MoCA. There is no indication of any claim to representativeness; the mean years of education are 16, and persons with dementia were excluded. Mean MoCA scores are available for 10-year age groups.

An earlier study from the Alzheimer Disease Centers in the USA (Weintraub et al., 2018)²⁸⁾, in which the MoCA was likewise normed as one of several procedures, was not considered by us for the TDB2 norms. In this study the mean years of education of the persons were even above 16 years, and the MoCA values were accordingly even higher than those of Sachs et al. (2020).

Figure 3 shows the means of the studies discussed so far across the course of age in five-year periods from 50 to 89 years (green the random samples, orange the convenience samples). Even though some of the studies also examined younger and older (only a single study) age groups, we restrict ourselves to the age groups from 50 to 89, because only here are reasonable case numbers available. Figure 3 also shows the mean values across the studies as thick lines, with those from random samples in green and those from convenience samples in orange. The number of cases varies greatly across the studies; for example, in the 70–74 age group, it ranges from 1,627 in Sachs et al. (2020) to 7 in Oren et al. (2015). We did not calculate the mean values on a weighted basis, because otherwise individual studies would have been given too much weight.

Figure 3: Means of the MoCA raw scores across the course of age in six studies with random samples (green) and eight studies with convenience samples (orange), together with their simple average

The figure shows that the individual norms differ from one another, in part very strongly, more so in older age groups than in younger ones. The studies with random samples tend to cluster in the lower, those with convenience samples in the upper performance range. This is to be expected and can also be observed with other test procedures. In connection with the MMSE it was already discussed earlier in the TDB2Online wiki. On average, the studies with convenience samples lie about one raw point above those with random samples in the age range from 50 to 69, and two or more raw points above them in the age range from 70 years. The raw scores of the original paper by Nasreddine et al. (2005) are extremely high at 27 points for the age range between 65 and 79, which can be seen nicely in the figure.

In the section Distributional properties of the raw scores we saw that there are percentile distributions in only two studies on which a linearization of the scale can be built. In the age group 60-64, on which we will base the performance scores, both studies, Kenny et al. (2013) and Ojeda et al. (2016), lie in performance level almost exactly at the average of all studies with randomized samples. Thus, as far as the mean level of the studies is concerned, nothing speaks against basing the calculation of the performance scores on these two studies alone. In the further course of age, however, both studies showed poorer mean performance than the average of all studies with random samples. For the age-normed scores we will therefore use a correction that takes this into account.

Figure 4 shows the performance scores for MoCA raw scores from 10 to 30, based on an percentile-to-normal transformation of the percentile values of the studies by Kenny et al. (2013) in blue and Ojeda et al. (2016) in green. Both curves run almost congruently. In the end regions, where they diverge somewhat, the larger study (Kenny) will determine the transformation in the weighted averaging.

Figure 4: Percentile-to-normal transformation of the percentile values into IQ-scaled standard scores for two samples

see the detailed test documentation

The MoCA and MMSE measure similar cognitive dimensions and correlate moderately to highly with each other, depending on the composition of the sample. For practice, a table that can convert a MoCA score into an MMSE score (and vice versa) would be helpful. For such purposes there is the method of “equating” or “linking” of two test procedures,²⁹⁾ mostly used in the form of equipercentile linking. In this, measured values of the two procedures are equated that have the same percentile rank within their respective distributions.

In the literature (as of March 2021) we found nine papers that dealt with the conversion of MoCA scores into MMSE scores. They are summarized with the most important data in Table 1.

Table 1: Studies with equipercentile data of the MoCA and MMSE

—-
HC: Healthy Control; AD: Alzheimer Dementia; MCI: Mild Cognitive Impairment; PD Parkinson Disease

The nine studies cover a wide range from healthy examinees to patients with severe impairments. Different diagnoses from the spectrum of age-related diseases are represented. Most studies used the R program “equate”³⁹⁾ and smoothed the raw score distributions with a log-linear procedure before the equipercentile linking. Whether the education correction proposed in the original publication of the MoCA (adding one raw point for persons with less than 12 years of schooling) was used is not reported in some of the papers.

Figure 6: Equivalence curves of the MoCA and MMSE in nine individual studies and in the mean course

We averaged the conversion tables published in the nine papers without weighting and then smoothed them with a third-order moving average. Figure 6 shows the course of the transformation curves of the nine studies and, in addition - dashed thick black - the course of the mean curve. The individual studies lie relatively close together. For a given MoCA or MMSE raw score, the values in the other procedure differ by more than three raw scores only in the outer ranges. Averaging across the 9 papers therefore probably yields very reasonable values. It also becomes clear that in the upper performance range the MMSE does not differentiate well: the MoCA scores from 24 to 30 are matched by only three MMSE scores (28-30). Values below 10 MoCA points occur only rarely even in these large samples: here the graphs of the individual studies diverge. In some studies these values were also merely extrapolated. This need not disturb too much: below 14 MMSE or 7 MoCA points, cognitive abilities lose their significance for the clinical picture of dementia diseases, and other behavioral characteristics become more important.

In the following Table 2, the equivalence values averaged across the nine studies are listed for both conversion directions. In most papers they were available only for the direction “MoCA to MMSE”, so they were averaged in that way here. For the direction “MMSE to MoCA”, the MoCA values for integer MMSE values were linearly interpolated between the neighboring values. In the table the equivalence values are listed with two decimal places. These values are intended for studies with larger patient numbers and can increase accuracy there. For conversions in individual cases, the equivalence value must be rounded to the nearest whole number:

Table 2: Equivalence values for the conversion of MoCA into MMSE scores and vice versa

As with other test procedures, the 19 norming studies yield very different norms. The countries of origin of the studies say little about the level of the norms. Even among studies from the same country there can be large differences. Wherever possible, TDB2 does not rely on a single norming study. If further studies are available, they can be included in the creation of the norms, with recognizably bias-free samples being weighted more heavily.

As with other test procedures, studies with random samples generally yield lower norms than those with convenience samples. This has to do with the selection of examinees in the convenience samples: mostly they are persons from the environment of university institutes, in particular also those who are themselves interested in having their cognitive performance measured. These examinees show higher intellectual abilities than a random sample. In older examinees, however, there is also a bias in random samples: persons enter the sample who have manifest or even only incipient dementia. The older the examinees, the greater the corresponding bias. All 19 studies formulated corresponding exclusion criteria, although to differing degrees and often not described very systematically. The effect that selections by diagnoses, or also by later diagnoses in the longitudinal course, have was shown by us on the basis of data from the US ADAMS longitudinal study⁴⁰⁾ in the discussion section of the wiki entry on the Mini Mental State Examination (as a single test). The data are also discussed there in more detail.