Behavioral manifestations of Alzheimer’s disease (AD) are emerging as an extremely important component of the illness and have been the target of many recent clinical trials.1 Several behavioral assessment instruments have been, and continue to be, widely used in AD research, including the Behavioral Pathology in Alzheimer’s Disease Rating Scale (BEHAVE-AD);2 the Neuropsychiatric Inventory (NPI);3 and the Behavioral Rating Scale for Dementia (BRSD) of the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD).4
The BRSD was designed to assess behavioral symptomatology in persons with dementia,4 but the degree to which such symptomatology is "normal" or "abnormal" has not been empirically established. Reisberg et al.5 compared BEHAVE-AD symptoms across 105 probable AD patients and a very small set of normal subjects (n=15), and although a considerable proportion (>20%) of the normal subjects demonstrated anxieties and phobias, affective disturbances, hallucinations, and activity disturbances, the authors did not characterize these as "normal" levels of prevalence.
Abnormal behaviors as assessed by the NPI were explored by Mega et al.6 across AD and normal elderly control subjects (NEC) and across three levels of AD severity, demonstrating limited prevalence of behavioral symptoms in control subjects. With respect to the 10 behavioral domains the NPI assesses, 88% of their AD sample showed behavioral symptoms, compared with roughly 12.5% of NEC.
The original 48-item BRSD4 has been revised to comprise 46 items.7 Patterson et al.8 found that NEC and subjects with AD were significantly different on BRSD total scores as computed via the 48-item algorithm, and the old and new scoring procedures do not differ enough to affect such comparisons. Mack et al.7 reported low (range —0.30 to 0.34), but statistically significant, correlations between Mini-Mental State Examination (MMSE)9 scores and 26 of the BRSD item frequency ratings, indicating that there may be some relationship between the frequency of these behaviors and AD severity. Neither prevalence of BRSD symptoms in persons with AD nor association of prevalence with MMSE score has yet been determined. The present study describes the prevalence of frequency-rated behavioral symptoms on the BRSD in persons with Alzheimer’s disease and in normal elderly control subjects.
To define an "abnormal" level of the specific behaviors assessed by the BRSD, we compared groups of NEC and well-characterized, community-dwelling AD patients on their BRSD responses. The two groups were similar in age and socioeconomic status. We first compared the total (46-item) BRSD scores7 for each group to determine both the range of total scores for the NEC and where the range of scores for AD patients fell relative to this "normal" range. Next, we took the 37 items that had frequency ratings and dichotomized the ratings as "present" or "absent" (defined below). We compared those behaviors rated as "present" across these groups at three time points over 1 year: baseline visit, 6 months, and 12 months. We sought to determine if the range of symptoms observed for AD patients was similar to the range for NEC, and if these differences were apparent over the course of our 1-year observation period.
In addition to comparing AD patients and NEC, we also investigated differences in item prevalence across AD severity levels. The AD group was divided into five MMSE- based strata, which reflect disease severity.10 Some studies have shown that there may not be a close relationship between cognitive decline and behavioral symptomatology.11 Mega et al.6 reported that some of the abnormal behaviors in the 10 NPI domains increased in prevalence, while others decreased, across mild, moderate, and severe AD. We anticipated that examining the prevalence of BRSD items across five MMSE-based AD severity strata might indicate where, in the cognitive-status continuum for AD, the highest probability of behavioral symptomatology on the BRSD might be expected.
The purposes of this study were to test the null hypotheses that 1) all 37 behavioral items would be "absent" in NEC; 2) the prevalence rates for all behavioral items would be significantly greater for AD patients than for NEC; and 3) the prevalence rates for the 37 behavioral items would be equivalent across the five AD severity levels we employed.
The data described here were collected from community- dwelling persons with AD and from NEC subjects; both groups were participating in a multicenter study carried out by the Alzheimer’s Disease Cooperative Study12 to evaluate an array of assessment instruments, including the BRSD.10 We compared AD patients with NEC, and the MMSE-stratified AD subgroups with each other, on total scores at baseline and on each of the 37 frequency- rated behaviors at baseline, 6 months, and 12 months; the behaviors were dichotomized as present/absent as described below. The MMSE was administered to all participants at a screening visit. Informed consent was obtained for all participants (from caregivers for AD patients). Exclusion criteria for both populations included a recent (within 2 years) history of major psychiatric disorder, including depression.
t1 shows the mean age, educational level, MMSE score, and gender composition for the groups and subgroups. The NEC subjects were 64 normal elderly adults.10 The AD subjects were 242 community-dwelling individuals with NINCDS-ADRDA—based diagnosis of probable Alzheimer’s disease.13 AD subjects were required to discontinue all psychotropic medication (prescribed for behavior management) 4 weeks prior to the baseline visit, except for patients with MMSE score less than 10, for whom continuation on drug was permitted (n=11). Other drugs, including cholinesterase inhibitors (tacrine, physostigmine), were allowed for any patient on a stable dose for at least 4 weeks prior to baseline and/or for 1 week prior to the 6- and 12-month visits.10 Of the AD subjects, 35 (14.5%) reported use of cholinesterase inhibitors during the 12-month study.
The AD group was stratified by MMSE score into five levels:10 I: MMSE≥21; II: MMSE 16—20; III: MMSE 10— 15; IV: MMSE 5—9; V: MMSE 0—4.
The BRSD was administered by a clinician or technician to a caregiver (AD) or knowledgeable respondent (NEC) as a 48-item instrument, but was recoded to 46 items according to 1996 scoring rules,7 based on which the total weighted scores (maximum=167) were computed for total score analysis. The 46-item BRSD contains 37 items that specify behaviors rated with respect to their frequency, ranging from 0 (not in past month) to 4 (16 days or more in past month). Eight items (;ns9, 10, 12, 14, 15, 17, 26, 32) are framed relative to "before dementia began" and are rated 1 (yes) or 0 (no). One item (;ns46) inquires about behaviors not covered by the scale. (See t2 for list of frequency-rated items.) Our prevalence analyses were based on the responses for the 37 behavioral items at baseline, 6-month, and 12-month visits.
Responses to each of the 37 frequency-rated items were recoded as "present"=1 (3 or more days in past month) or "absent"=0 (0—2 days in past month). With a month- long observation window, we anticipated that we could adjust for any "yes" bias in responses, particularly for NEC, by rating up to 2 days in the past month as essentially "absent." Proportions of symptoms rated present per group (AD/NEC) were tabulated, indicating the prevalence of each BRSD symptom for each group.
Kruskal-Wallis testing was used to compare the groups (AD vs. NEC; AD by stratum) on MMSE and BRSD total scores at baseline. Chi-square tests, corrected for multiple comparisons by Holm adjustment,14 were used to determine the differences between prevalence rates for AD patients and NEC groups per BRSD item (2×2 contingency tables, present/absent by AD/NEC) at each of the three visits. We did not compare prevalence rates over time within groups. Chi-square or Fisher’s exact tests (Holm-adjusted) were then used to analyze the 5-by-2 contingency tables of presence/absence of each behavior by MMSE stratum (all at baseline). After Holm-adjustment, P-values less than 0.05 were deemed significant.
The data collection in the study from which this data set was obtained has been described in detail elsewhere.8,10 We report the same comparisons of MMSE as these previous reports, but the BRSD total and item scores reported here were calculated by an algorithm different from that used by Patterson et al.8
Three AD patients had no BRSD scores at baseline. The groups (AD vs. NEC) were not different in terms of demographic variables (all P>0.1), but as expected, AD patients had significantly lower MMSE scores (Z=—12.12, P<0.001) and significantly higher BRSD total scores (Z=—10.90, P<0.001) at the baseline visit than did NEC. BRSD total scores at baseline were significantly different across AD strata as well (BRSD total: χ2=32.45, df=4, P<0.001). t1 displays these scores.
Presence/Absence of Symptoms
t2 shows the endorsement rates of the 37 behavioral items, per group, at baseline and 12-month visits; as described below, the 6-month visit data did not differ from the 12-month data, and so they are not shown. At least 10% of NEC endorsed items ;ns11, 13, and 20 at baseline, items ;ns13 and 20 at 6 months, and item ;ns11 at 1 year. We also saw that between 5% and 10% endorsed items ;ns1, 2, 7, 19, and 30 at baseline; items ;ns3, 11, and 21 at 6 months; and items ;ns13, 19, and 20 at 1 year.
Although we hypothesized that the prevalence rates for the NEC would be zero for the behavioral items, we implicitly hypothesized that all, or most, items would be endorsed for subjects with AD. In fact, we found that many of the items were endorsed for 10% or fewer AD patients at each visit, including items ;ns6, 8, 16, 22, 23, 28, 34, 35, 37, 38, 39, 44, and 45 at baseline; ;ns6, 8, 16, 18, 23, 28, 34, 37, 38, 39, 40, and 44 at month 6; and ;ns6, 8, 16, 18, 23, 28, 37, 38, 39, 40, and 44 at month 12. These items, as listed in t2 for baseline and 12-month visits, were endorsed for 10% or fewer of the AD patients on at least two of three consecutive visits, except for items ;ns22, 35, and 45, which were all endorsed for less than 10% of patients at baseline and for more than 10% at the next two visits.
Prevalence Rates for AD vs. NEC
We saw some consistency over the three visits as to which items were endorsed significantly more often for AD patients than for NEC subjects (P<0.05 after Holm adjustment). At all visits, endorsement rates were significantly higher among AD patients than NEC for items ;ns2, 3, 5, 11, 19, 20, 21, 24, 25, 30, 31, 33, 41, 42, and 43. Additionally, we saw that at both 6-month and 1-year visits, endorsement rates for AD patients were significantly higher than for NEC on items ;ns1 and 27. Items ;ns4 and 29 were endorsed significantly more for AD at 12 months only, and item ;ns36 was endorsed more only at 6 months. Although the endorsement rates were always higher for the AD group than for NEC, these differences did not reach significance consistently, or at all, for 22 of 37 items analyzed.
Prevalence Across AD Severity
F1 contains the item endorsement rates for the behavioral items across MMSE-based subgroups of the AD subjects at baseline. Holm-adjusted figures indicate significant differences across MMSE-based domains for fewer than half of the analyzed items: ;ns2, 13, 21, 23, 24, 27, 28, 30, 34, 40, 41, 44, and 45.
As can be seen in F1, the endorsement rates of items ;ns41, 44, and 45 do not increase linearly over MMSE strata, but seem to have rather more complex trends. Endorsement rates for items ;ns23, 27, and 29 tend to plateau across strata 1—3 or 4 and then increase, whereas those for items ;ns2, 21, 24, and 40 seem to increase in a more linear way across MMSE strata.
The 10 most prevalent behaviors at baseline in each AD severity group are presented in t3. We can see that the majority of these are among the top 10 across all strata (;ns2, 3, 11, 19, 20, 25, 30) and that increasingly more subjects exhibit most of these behaviors as AD severity increases. Notable exceptions are item ;ns11 (tiredness), which appears to have an essentially stable endorsement rate across severity groups, and item ;ns30 (repetitive behavior), the only item with a decreasing trend in endorsement across groups.
Although prevalence rates were uniformly higher for AD subjects, not all of the BRSD items were endorsed for significantly more AD than NEC subjects. At baseline, 35% of the BRSD items analyzed were endorsed for less than 10% of AD subjects. AD patients with MMSE scores of 10 or lower were allowed to continue on any psychotropic medications they were currently on, but since only 4.5% of the sample met this criterion, this was unlikely to have affected the outcome.
Although prevalence rates were higher for AD patients than for NEC, these differences failed to reach significance for a majority of items. The "normal limits" of prevalence for these items may be greater than for those where the prevalence for AD patients and NEC differs significantly.
AD by MMSE-Based Severity Subgroup
We considered two possible outcomes of the comparisons of prevalence across severity groups at baseline. The first was that endorsement rates would be higher for individuals with worse cognitive status as representative of their higher impairment. This was observed for physical signs of anxiety, crying, agitation, irritability, uncooperativeness, restlessness, misidentification of things, and beliefs that TV characters are real and that people are in the house (items ;ns2, 5, 19, 20, 21, 24, 35, 40, and 41).
The other possible outcome was that endorsement rates would be lower for individuals with worse cognitive status because they would be unable to perform or exhibit the behaviors. Notably, endorsement rates on repetitiveness (item ;ns30) were significantly higher for subjects with higher MMSE scores, and endorsement rates on this item alone tended to decrease as MMSE scores decreased.
For the majority of items, however, we observed no significant differences in endorsement across MMSE- based domains for BRSD items. In fact, we do not see evidence strongly in support of either of these alternative hypotheses; we have not found evidence that cognitive status corresponds strictly to behavioral symptomatology in this analysis. Lower MMSE scores were not uniformly associated with either higher or lower endorsement rates; nor did we see BRSD items that reflect particular functional domains, such as speech or physical movement, endorsed less and less as MMSE scores dropped off. Endorsement rates for the behavioral items were, in some cases, very low across AD strata; in other cases the endorsement rates increased in parallel with decreasing MMSE (except for item ;ns30, repetitiveness), or were very low or absent for higher MMSE but relatively high for subjects with lower MMSE.
From a purely descriptive perspective, we also saw that on some items endorsement was highest for middle strata (hopelessness, feels life is not worth living, social withdrawal; items ;ns4, 8, and 31) or that it was lowest for middle strata and highest for lowest/highest MMSE scores (sad appearance, excessive physical complaints, sudden changes in emotion, visual hallucinations; items ;ns3, 16, 18, and 45). However, t3 shows that the most frequently endorsed items tend to be relatively homogeneous across the AD levels. This implies that there may be a core set of behaviors that are likely to be observed over the course of AD (physical signs of anxiety, sad appearance, tiredness, agitation, irritability, uncooperativeness, restlessness, purposeless behavior, and repetitive behavior; items ;ns2, 3, 11, 19, 20, 21, 24, 25, and 30).
The majority of these nine symptoms appear to be more prevalent as dementia severity increases cross-sectionally; we saw, however, that over 12 months endorsement rates were essentially stable over all subjects (see t2), although there was some decrease for tiredness (item ;ns11), and some increase in uncooperativeness (item ;ns21).
Although there is analytic support for BRSD subscores,4,7 this set of nine behaviors includes at least one symptom each from the Depression, Behavioral Dysregulation, Inertia, and Irritability/Aggression domains, as well as two symptoms (sad appearance and repetitiveness, items ;ns3 and 30) not in any subscore. Analysis of all BRSD items, rather than total or subscores, may therefore be important to fully describe behavioral disturbance in persons with AD in both research and clinical applications.
Mack et al.7 describe a "short form" of the BRSD that includes items ;ns3, 4, 8, 19, 20, 22, 24, 25, 27, 30, 41, 43, and 45 (see t2), plus four other BRSD items not analyzed here (items ;ns9, 10, 12, and 14: loss of enjoyment, loss of initiative, change in sleep pattern, change in appetite). The short form comprises the 16 BRSD items that met one or more of the following criteria: high loading on a BRSD subscale factor; high average frequency; high proportion of ratings of 1—4; high number of ratings of 3 or 4 (high frequency); or high correlation with MMSE.7 The subset of items derived from t3 and the BRSD short-form subset have considerable overlap (items ;ns3, 19, 20, 24, 25, and 30; sad appearance, agitation, irritability, restlessness, purposeless behavior, repetitive behavior). Given the different criteria for the establishment of these two BRSD subsets, their similarity supports the possibility that there may be a core set of behavioral symptoms in AD.
One of the purposes of this analysis was to determine if normal elderly control subjects endorsed any of the items on the BRSD, and if so, whether or not these endorsement rates were uniformly and significantly lower than those for AD. We defined endorsement as frequency of at least 3 days in the past month. We found 3 of 37 items with endorsement rates for NEC near or above 10% (tiredness, sleep problems, easily irritated); 14/37 items were endorsed for 3% to 15% of NEC at some point over our study (see t2). Mega et al.6 reported a 12.5% prevalence rate for any one (or two, in one case) of 10 possible noncognitive behaviors in their normal group. Although the least frequent rating allowed by Mega et al. (less than once per week) was classified as "absent" in the present study, our reported prevalence rates for NEC are similar.
If, in the present study, the endorsement rates for NEC had been all zero, then a "normal" total weighted score on the BRSD could have been taken as zero. However, the range of scores for NEC was at its widest (0— 53 out of a possible 167) at baseline, and at its narrowest (0—34) at 6 months (maximum at 12 months was 37). Although both the mean and the maximum BRSD total scores for the NEC were considerably smaller than those of the AD patients, there was some overlap (see t1). The implication is that AD patients with total weighted BRSD scores at or below about 50 may, in fact, be "normal" in terms of their behavioral symptomatology. Therefore, it may not be appropriate to label AD patients "behaviorally disturbed" if their total BRSD scores fall within the range demonstrated by these NEC. Further, in clinical trials where the objective is to normalize a patient’s total BRSD score, BRSD scores upon enrollment would need to be outside of the range observed for the normal elderly control subjects here. Endpoints would then be scores within this normal range, rather than at zero—which could have implications for power calculations.
An alternative approach is to define the limit of the "normal" range as the upper bound on the 95% confidence interval for the mean of the baseline total score (6.77, corresponding to 7 points, based on the NEC data presented here). This establishes a "normal" range as scores from 0 to 7, which will include the mean NEC BRSD total score 95% of the time—even though this approach does not address the actual distribution of the scores of normal subjects, namely, that they can reach upwards of 50 points.
Prevalence rates for NEC on any of the 37 items reported here would also represent "normal limits." Similarly, changes over time or during a study that leave the item and total scores within the ranges observed for NEC would be characterized as "normal fluctuation" and would not be attributed to treatment effects or even disease progression.
There were several symptoms (restlessness, purposeless activity, repetitiveness; items ;ns24, 25, and 30) where NEC endorsement was extremely low while AD endorsement was upwards of 40%; these were three of the six items with the highest AD endorsement rates. These items may be particularly informative with respect to differentiating AD and NEC; the plots in F1 suggest the need for closer study of the relationship between the responses to these items and MMSE score or the course of AD.
Although these three symptoms were among the most prevalent in this AD patient sample, ranging in endorsement rates from 44.2% to 65.7% (t3), they were present only in about 38% of the AD sample reported by Mega et al.6 We also found slightly lower baseline prevalence rates than did Mega and colleagues for hallucinations (8.4%—9.6% vs. 10% in their sample) and delusions (3.8%—17.5% vs. 22% in their sample). However, the BRSD provides separate ratings for different kinds of hallucination and delusion; if instead there were only one item apiece, "any hallucination" or "any delusion," our prevalence rates might be higher than the individual item—based rates we observed.
The difference in reported prevalence rates across the Mega et al. study and the present one may also be due, at least in part, to the makeup of our respective subject groups: whereas the Mega et al.6 sample focused on consecutive outpatients visiting a single clinic for assessment of dementia, our sample included nearly five times as many subjects who were enrolled in a multisite clinical study.
A second purpose of this study was to use an item- level approach to investigate the relationship between cognitive functioning and behavioral symptomatology in community-dwelling AD patients. We found that physical signs of anxiety, restlessness, misidentification of things, and thinking people on TV are present (items ;ns2, 24, 35, and 40) had fairly linear relationships with MMSE (F1). However, many symptoms, such as hopelessness, feeling life isn’t worth living, physical complaints, and delusions that one’s spouse is unfaithful or an imposter (items ;ns4, 8, 16, 37, and 39) did not appear to be associated with MMSE.
We plan to follow up this analysis by focusing on the subset of BRSD items that is maximally descriptive of the total BRSD score, with the best power to discriminate AD patients from NEC. In its current form, the BRSD includes a complex set of behavioral symptoms, and we have attempted to establish "normal limits" for both the set (total score) and the component items. Item-level analyses of the BRSD may be an important source of information; prevalence rates of these items, as well as change over time, in normal elders must be taken into account in long-term studies of behavioral symptoms in AD.