|Home | About | Journals | Submit | Contact Us | Français|
The Dementia Rating Scale (DRS) is a widely used measure of global cognition, with age- and education-corrected norms derived from a cross-sectional sample of adults participating in Mayo's Older Americans Normative Studies (MOANS). In recent years, however, studies have indicated that cross-sectional normative samples of older adults represent an admixture of individuals who are indeed cognitively normal (i.e., disease-free) and individuals with incipient neurodegenerative disease. Theoretically, the “contamination” of cross-sectional normative samples with cases of preclinical dementia can lead to underestimation of the test mean and overestimation of the variance, thus reducing the clinical utility of the norms. Robust norming, in which dementia cases are removed from the normative cohort through longitudinal follow-up, is an alternative approach to norm development. The current study presents a reappraisal of the original MOANS DRS norms, provides robust and expanded norms based on a sample of 894 adults age 55 and over, and critically evaluates the benefits of robust norming.
The Mattis Dementia Rating Scale (DRS) is a widely used measure of global cognition based on performance across five domain subtests (Attention, Initiation/Perseveration, Construction, Conceptualization, and Memory). Lucas and colleagues (1998) provided age- and education-corrected norms derived from a sample of 623 Caucasian adults participating in Mayo's Older Americans Normative Studies (MOANS), and these norms were subsequently adopted into the second edition of the test (Jurica, Leitten, & Mattis, 2001).
In the years since the publication of the original MOANS DRS norms, several additional studies have contributed to the DRS normative literature. Lichtenberg and colleagues published age- and education-stratified norms for older Caucasian and African American orthopedic inpatients in an urban medical setting (Bank, Yochim, MacNeill, & Lichtenberg, 2000; Lichtenberg, Ross, Youngblade, & Vangel, 1998), whereas Marcopulos, McLain, and Giuliano (1997; Marcopulos & McLain, 2003) published DRS norms for rural-dwelling Caucasian and African American elders with little to no formal education. In 2005, Rilling and colleagues (2005) published DRS norms derived from 307 participants in Mayo's Older African American Normative Studies (MOAANS). Preliminary norms for Spanish-speaking U.S. residents (Lyness, Hernandez, Chui, & Teng, 2006, 2007) have also recently become available, and DRS performance has been evaluated among American Indians from a Northern Plains tribe in Colorado (Jervis, Beals, Fickensher, & Arciniegas, 2007). More recently, Pedraza and colleagues (2007) provided normative change scores for the DRS in the form of reliable change indices calculated from over 1,000 MOANS and MOAANS participants.
Normative reference values for neuropsychological tests usually are obtained from cross-sectional samples of cognitively normal individuals. Cognitive “normality” is inferred based on clinical history or external criteria such as the absence of impaired scores on other cognitive measures, independence in instrumental activities of daily living, and/or statement of normal cognition obtained from the participant, informant, or primary care physician. Among older adults, however, cross-sectional normative samples represent an admixture of individuals who are indeed normal (i.e., disease-free) and individuals with incipient neurodegenerative disease (Saxton et al., 2004; Sliwinski, Lipton, Buschke, & Stewart, 1996). Those at a preclinical stage of neurodegenerative dementia may obtain baseline test scores that are within the realm of expectation when compared with age peers, despite representing a downward cognitive trajectory for the individual. The unintended consequence of including these individuals into conventional normative studies is that it leads to underestimation of the test mean and overestimation of the variance (De Santi et al., 2008; Ritchie, Frerichs, & Tuokko, 2007; Sliwinski et al., 1996).
Longitudinal robust norming is a method of establishing normative data that exclude individuals with preclinical dementia. This is accomplished through (a) longitudinal surveillance of all candidates considered at baseline for inclusion into a normative sample, (b) case ascertainment of individuals who change diagnosis away from normal at any time point during follow-up, and (c) exclusion from the normative sample of all cases diagnosed with dementia. Individuals lost to follow-up after their baseline assessment also may be excluded from robust normative samples, particularly if their baseline test scores contribute to a reduction in the test mean and overestimation of the variance (Holtzer et al., 2008).
Compared with cross-sectional norms, longitudinal robust norms appear to have greater predictive value in identifying individuals who decline from normal to mild cognitive impairment (MCI) or Alzheimer's disease (AD). De Santi and colleagues (2008) created robust norms for a battery of neuropsychological tests derived from 113 individuals who retained a diagnosis of normal on at least three successive evaluations. These robust norms were then applied to the baseline test scores of an independent cohort of adults followed longitudinally, some of whom eventually declined from normal to MCI or AD. Compared with conventional norms derived from a cross-sectional sample of another 256 participants, robust norms identified a greater proportion of decliners as having cognitive impairment at baseline. Likewise, Holtzer and colleagues (2008) examined data from 1,326 older adults participating in the Einstein Aging Study and created robust norms from a subset of 307 individuals diagnosed as cognitively normal at baseline and during at least two annual follow-up visits. The robust norms identified a larger proportion of baseline cognitive impairment in incident dementia cases than a set of conventional norms.
Although better at predicting who will eventually decline from normal cognition to the early stages of AD, the advantage of robust norms appears to diminish once patients advance to a diagnosis of MCI or dementia. Specifically, De Santi and colleagues (2008) found that both conventional and robust norms performed equally well at confirming cognitive impairment among patients with known AD as well as in patients with MCI who progressed to AD.
Some studies have failed to identify any advantage to robust norms. Marcopulos and McLain (2003), for example, recalculated norms from a subsample of participants who had contributed to a conventional, cross-sectional normative sample 4 years earlier. Ninety-four of the original 131 participants were retested, and “robust” norms were derived from a subset of 81 individuals who demonstrated no significant decline when compared with their baseline test performance. The investigators observed that their revised norms did not differ appreciably from the norms calculated from the original conventional sample. They noted, however, that decline was defined psychometrically and no clinical or medical information was obtained to make a formal diagnosis of dementia. As such, it is unclear how many actual incident dementia cases were present in the original sample, the revised sample, or the excluded group of “decliners.” Ritchie and colleagues (2007) also found no significant difference in the accuracy of conventional versus robust norms derived from the Canadian Study of Health and Aging. This null result was attributed to an equal or lesser-than-expected incidence of dementia in their sample compared with the general population, resulting in minimal contamination in the normative cohort. The authors noted that robust norms may be most useful when the incidence of dementia in the normative sample exceeds that observed in the general population.
In aggregate, these studies suggest that longitudinal robust norms are at least equally useful as cross-sectional conventional norms in detecting cognitive impairment among older adults and may be more accurate than conventional norms in predicting preclinical dementia cases. Moreover, robust norms are superior to conventional norms if there is reason to suspect that the incidence of dementia in the normative cohort may be greater than expected in the general population.
With the exception of Marcopulos and McLain (2003), to our knowledge the DRS has not been included in any robust normative study. In light of the potential benefits of robust norms, the current study aims to retrospectively review the original MOANS DRS normative cohort, which was derived from a conventional, cross-sectional sample, and determine the proportion of preclinical dementia cases present in that sample. In addition, we update and improve upon the available normative data for the DRS by providing longitudinally robust norms derived from an expanded sample of Caucasian older adults.
A retrospective review of the Mayo Clinic Alzheimer's Disease Research Center (ADRC) database was conducted to identify community-dwelling, independently functioning adults who met the following criteria at the time of their baseline assessment: (a) age 55 or older, (b) normal cognition based on consensus diagnostic opinion, and (c) administration of the DRS. A consensus diagnosis was rendered at every encounter by a team consisting of at least one behavioral neurologist, clinical neuropsychologist, and nurse or family physician. Normal cognition at baseline was defined based on (a) self, informant and physician reports; (b) capacity to independently perform activities of daily living, based on informant report; (c) no active or uncontrolled central nervous system, systemic, or psychiatric condition that would adversely affect cognition, based on physician report; and (d) no use of psychoactive medications in amounts that would be expected to compromise cognition or for reasons indicating a primary neurologic or psychiatric illness. Importantly, performance on the DRS did not contribute to the baseline diagnosis.
Eligible participants with a diagnosis of normal cognition at baseline included 1,175 adults. Of these, 101 individuals were lost to follow-up after their baseline assessment and constitute the “single encounter” group. Another 180 individuals progressed to a consensus diagnosis other than normal at a follow-up encounter and constitute the “preclinical dementia” group. Diagnoses at follow-up in this group included MCI (n = 119), AD (n = 39), vascular/mixed dementia (n = 7), dementia with Lewy bodies (n = 3), and other conditions (n = 12). The remaining 894 adults received a consensus diagnosis of normal cognition at every follow-up encounter and constitute the “robust normal” sample (Fig. 1). APOE genotype, an established risk factor for AD, was available on 99% of all participants (n = 1,165).
In addition, all 623 participants in the Lucas and colleagues (1998) normative cohort were specifically reviewed to determine how many remained cognitively normal during subsequent encounters, progressed to dementia, or were lost to follow-up after their baseline testing. Four-hundred and ninety adults (78.5%) out of 623 returned for at least one follow-up visit (range = 1–16). Of these, 334 retained a consensus diagnosis of normal across all visits, 62 progressed to MCI, and 53 progressed to AD. The remaining 41 participants did not retain a primary diagnosis of normal cognition due to other medical conditions (e.g., intracerebral hemorrhage, dementia with Lewy bodies, vascular dementia, Parkinson's disease, posterior cortical atrophy). Overall, 76 (15.5%) out of 490 participants with follow-up encounters (or 12.2% out of all 623 participants) eventually received a primary diagnosis of dementia. Participants from this original cohort who remained cognitively normal during all follow-up visits (n = 334) are included in the current, expanded normative set. That is, the current robust sample of 894 adults includes robust normal individuals (n = 334) from the Lucas and colleagues article, plus a new sample of 560 subjects recruited during the intervening years.
All data were obtained in full compliance with study protocols approved by the Mayo Clinic Institutional Review Board.
The DRS is a 144-point instrument used in clinical and research settings for the detection, differential diagnosis, and staging of dementia (Jurica et al., 2001). It measures attention, orientation, word fluency, motor initiation and perseveration, visuospatial construction, conceptualization, and memory, which are organized into five domain subtests (Attention, Initiation/Perseveration, Construction, Conceptualization, and Memory). Test items are arranged hierarchically, and full credit is given to a section if the individual answers the initial items correctly. Administration time usually takes 20–40 min.
Group differences for continuous variables were examined using ANOVA with post hoc Tukey's honestly significant difference tests. Group differences for categorical variables were examined with Pearson's chi-squared. Given the unique association of education and DRS scores found in the previous Mayo normative study (Lucas et al., 1998) and other samples (Marcopulos et al., 1997; Schmidt et al., 1994; Smith et al., 1994), hierarchical linear regression was used to explore this relationship in the robust normal group. Age was entered first into the hierarchical model, followed by education.
The methodology used to develop the current norms was similar to that used in previous Mayo normative studies (e.g., Ivnik et al., 1990; Lucas et al., 2005). Specifically, an a priori decision was made to include a minimum of 70 individuals within each age group to increase the likelihood of stable estimates. Individuals at the extreme ends of the age distribution were combined until the minimum sample size of 70 was reached. Overlapping midpoint age intervals were used to maximize the available information, with midpoints selected at 3-year intervals and a range around each midpoint of ±5 years (Pauker, 1988). The distribution of raw scores within each midpoint age group was then normalized by assigning standard scores based on actual percentile ranks. This strategy provides a normative reference for adults surrounding a particular age point. For additional details regarding this methodology, see Ivnik and colleagues (1990).
Demographic characteristics and mean DRS total scores for all 1,175 participants in the updated sample are presented in Table 1. The three groups differed on age, F(2, 1172) = 9.46, p < .001, and education, F(2, 1172) = 4.83, p < .01. Specifically, robust normal participants were marginally younger than single encounter (p = .05) and significantly younger than preclinical dementia (p < .001) participants. Robust normal participants had a significantly higher level of education than single encounter participants (p < .05), but not when compared with preclinical dementia participants. There was no significant difference between the groups on sex distribution, χ2(2) = .57, p = .75. Robust normals had the lowest proportion of APOE ε4 allele carriers (23.6%); however, this was not significantly different when compared with the proportion of APOE ε4 carriers among single encounter, χ2(1) = 1.85, p = .17, or preclinical dementia, χ2(1) = 3.63, p = .06 groups. When considering only the robust normal sample, 11 adults (1.2%) were APOE ε4/4 homozygotes.
As expected, mean total DRS scores differed among the three groups, F(2, 1172) = 16.30, p < .001, with robust normals obtaining a significantly higher mean DRS score compared with single encounter (p < .001) and preclinical dementia (p < .001) participants. Consistent with prior studies on the effects of robust norming, the conventional (full) sample of 1,175 participants (M = 135.8, SD = 5.8) underestimated the DRS mean and overestimated the variance when compared with the robust subsample. Moreover, this effect was also evident when comparing the current robust subsample with the mean DRS score (M = 134.7, SD = 6.8) calculated from the original subsample of 623 participants.
Frequency distributions of demographic data and study encounters for the robust normative sample are presented in Table 2. Approximately 89% of the sample had at least two follow-up visits and almost half (45.3%) had a minimum of five follow-up visits during which a consensus diagnosis of normal was retained. Among preclinical dementia cases, the median number of follow-up visits was four and the mode was six (range = 1–8). Single encounter participants were lost to follow-up for a variety reasons, but no formal attempt has been made to ascertain the specific factors contributing to their study withdrawal.
Table 3 shows the association between demographic variables and DRS scores in robust normal participants. As expected, age and education were significantly associated with the majority of DRS subtest and total scores. Sex was associated only with the Construction subtest, accounting for less than 1% of variance (r2 = .008). To determine if this association was sufficiently relevant to stratify the Construction subtest in our normative tables, we then analyzed the mean Construction scores between men and women across 5-year age bands. Mean Construction scores were significantly different for men and women only between the ages of 80 and 84, with women obtaining on average 0.12 points more than men. Across all ages, the mean Construction subtest score for men was 5.73 and women 5.81. Given the rather trivial relationship between sex and Construction subtest scores in our sample, and the lack of association to any other DRS subtest, it was not considered for the normative analyses.
Normative data for DRS subtest and total scores are presented in Tables 44–1313.. Robust age-corrected scaled scores are presented in the leftmost column of each table, with corresponding percentile ranks in the rightmost column. To use these norms, first select the table corresponding to the patient's age at the time of test administration. Then search for the patient's raw score and refer across to the corresponding scaled score (M = 10, SD = 3) and percentile rank. The age range and sample size used to create norms for each group are provided beneath each table.
Results from the hierarchical linear regression model showed education to contribute unique variance to DRS total scores, F(2, 891) = 108.9, p < .001, with an increment from r2 = .12 in the age-only model to r2 = .20 in the age and education model. To aid clinicians and investigators using the DRS, robust age- and education-corrected scaled scores can be derived for the total raw score using the equation presented in Table 14. Scaling constraints due to marked deviation from normality precluded education adjustments for each DRS subtest.
Finally, a preliminary post hoc analysis was performed of the incremental diagnostic validity and clinical utility of these robust norms in the detection of “baseline” cognitive impairment in preclinical dementia cases. Scaled scores were obtained for baseline DRS total scores from 49 preclinical dementia cases (39 AD, 10 with other dementia diagnoses) using robust and original norms. These scores were compared with the baseline scaled scores from 49 cognitively normal adults selected from the robust sample. To minimize the likelihood of spurious findings, 3 separate random samples (with replacement) of 49 cognitively normal adults were drawn. Cognitive impairment was defined as a scaled score lower than 7. Across the three comparisons, use of the original norms resulted in sensitivity of 14.3%, whereas the new robust norms resulted in sensitivity of 24.5%. Specificity using original norms ranged from 91.8% to 95.9%, and using robust norms ranged from 89.8% to 93.9%. Positive predictive values ranged from 63.6% to 77.8% using the original norms and 70.6% to 80.0% using robust norms, an average increase of 3.2%. Negative predictive values ranged from 51.7% to 52.8% using original norms and 54.3% to 55.4% using robust norms, an average increase of 2.4%.
The purpose of this study was twofold: First, to retrospectively review the original MOANS DRS normative sample and determine how many cases of preclinical dementia were present; and second, to update the available DRS norms with an expanded, longitudinally robust sample. Of the 490 adults in the original sample who returned for additional evaluations, 62 of them obtained a follow-up diagnosis of MCI and another 53 eventually progressed to AD. Considering the 1,074 adults in the expanded sample who had normal cognition at baseline and at least one follow-up visit, approximately 6% constituted cases of preclinical dementia. This rate is generally consistent with other robust normative studies in suggesting that cross-sectional norms may contain a non-trivial number of individuals at a preclinical stage of neurodegenerative dementia. For instance, Holtzer and colleagues (2008) reported a 4.6% rate of preclinical dementia cases, Saxton and colleagues (2004) reported a 10.4% rate of preclinical AD cases, and De Santi and colleagues (2008) reported a 16.1% rate of preclinical AD cases. Moreover, it is now clear that our original norms slightly underestimated the overall DRS mean and overestimated the variance when compared with a robust sample of cognitively normal adults.
There are several strengths to the present investigation. First, by revisiting the original normative sample and conducting a critical reappraisal, we have confidence that the robust norms provided in this article represent an accurate depiction of DRS performance among cognitively normal (i.e., disease-free) older adults. Second, by excluding from the robust sample not only individuals who have progressed to dementia but also those with MCI or a single encounter, we have minimized the likelihood that those with any incipient neurodegenerative process would be included. We are aware that not all individuals with MCI will progress inexorably toward dementia and many will revert back to normal. However, our conservative approach ensures that this normative cohort is as free of neurodegenerative disease as theoretically possible. Third, we considered the role of APOE in the selection of the normative sample. The proportion of APOE ε4 carriers was not significantly different between the robust and preclinical dementia samples, although there was a trend toward more ε4 carriers in the preclinical dementia group. Only 11 participants in the robust sample were ε4 homozygotes, and their mean DRS score of 136.4 (SD = 5.6) was similar to the mean DRS score for the entire robust sample. For this reason, we opted to keep these 11 individuals in the normative group. Fourth, the current study draws from a relatively large sample size and includes a substantial number of adults 80 years of age and older.
Finally, a preliminary validation of the diagnostic and clinical utility of these updated, robust norms was performed. When considering the DRS scores of preclinical dementia cases “at baseline,” the robust norms result in an almost twofold increase in sensitivity with relatively minimal loss of specificity when compared with the original norms. Moreover, in this preliminary analysis, the use of the updated robust norms contributes to net gains in positive (3.2%) and negative predictive values (2.4%). Although these gains are relatively modest, the reader is urged to remember that these are comparisons of baseline, preclinical DRS scores (i.e., prior to a clinical diagnosis of dementia). The modest magnitude of these gains suggests that in well-characterized, large normative cohorts in which the incidence of dementia does not exceed substantially that found in the general population, there is likely to be minimal contamination of test scores. As such, the extraordinary efforts and funding required to follow large cohorts longitudinally may not be practical for the sole purpose of deriving robust norms. Rather, the development of such norms may be most feasible in centers where longitudinal neuropsychological data are collected in service of other primary aims.
As noted in the original DRS normative study, the validity of all norms depends on the extent to which the test-taker shares similarities with the normative sample. The robust MOANS sample in the current study is comprised of Caucasian adults predominantly living in economically stable areas in the Midwest region of the USA. The underrepresentation of participants with limited education warrants caution when applying these norms to those with fewer than 8 years of formal education. DRS normative data for African American older adults are also available (Rilling et al., 2005).
Over the past two decades, investigators at the Mayo Clinic have provided normative data for a variety of neuropsychological tests under the auspices of the MOANS and MOAANS projects. It is hoped that the robust norms for the DRS presented in this article will be beneficial to clinicians and investigators in their evaluation of older adults.
This work was supported by the National Institutes of Health (NS054722 to OP, AG00831 to JAL, GES, NRG, RCP, RJI, AG016574 to JAL, GES, NRG, RCP, RJI, AG006786 to RCP, RJI); and the State of Florida Alzheimer's Disease Initiative (FL-2J-02 to JAL, NRG).
We are grateful for the assistance of staff members in the Psychological Assessment Laboratory at Mayo Clinic Alzheimer's Disease Research Center (ADRC) and the Memory Disorders Clinic at Mayo Clinic, Florida. We also want to thank Roger A. Mueller and Matthew R. Miller for their invaluable database assistance.