Predictors of severity of behavioral disturbance among community-dwelling elderly individuals with Alzheimer's disease: A 6-year follow-up study


Correspondence address: TakashiAsadaMD Division of Old Age Psychiatry, Musashi Hospital, National Center of Neurology and Psychiatry, 4-1-1 Ogawahigashicho, Kodaira City, Tokyo, 187-0031, Japan. Email:


Abstract To clarify which baseline factors, including apolipoprotein E (ApoE) genotyping and caregiver characteristics, predict the future severity of behavioral disturbance among community-dwelling elderly with Alzheimer's disease (AD), we conducted a longitudinal study for up to 6 years. Fourteen kinds of behavioral disturbance were evaluated. Analyses using data from 62 subjects with at least three annual assessments revealed that the baseline severity of behavioral disturbance was the strongest predictor of future severity, followed by baseline stage of dementia or use of support services. However, ApoE ɛ4 had only a marginal effect at most. The behavioral disturbances examined in the present study appear to be innately determined phenomena rather than merely representing stages of AD.


Behavioral disturbances and psychoses are prominent features of Alzheimer's disease (AD).1 Several studies have shown that these problems are the major predictors of caregiver burden and need for institutional placement.2

A range of cross-sectional studies have found that some clinical factors are associated with these problems.3–5 Possible association between apolipoprotein E (ApoE) ɛ4 gene and these problems has been also examined.6–11 However, these reports are also cross-sectional in study design. Recently, some studies have reported the natural course of these problems.12–18

It is now necessary to determine whether behavioral disturbances are innately determined phenomena which characterize a subgroup of AD patients, or phenomena that merely represent the stage of the illness. In contrast, behavioral disturbances seem to be influenced by the personal characteristics and situational circumstances of the caregiver. Thus, the extent to which caregiver factors contribute to these problems is another important question.

As part of a longitudinal study of community-dwelling AD patients conducted for up to 6 years in Yamanashi Prefecture, Japan, we intended to investigate these issues. Using the data on behavioral disturbances obtained from the present study, we carried out analysis to determine which factors predict the future severity of the disturbances. As putative predictors, we adopted ApoE genotyping and caregiver factors including use of support services, as well as the patients' clinical characteristics and demographics.


The subjects and methods of the present study have been described in detail elsewhere.18 Briefly, between February and December 1992, 146 candidate participants were recruited from outpatients of our clinic at Yamanishi Medical University, voluntary patients, and patients identified by formal service providers. All participants underwent the following examinations: activities of daily living (ADL) measurement using the Barthel index, manual neurological examination, mental examination using Hasegawa's Dementia Scale Revised (HDSR),19 staging of dementia using the Clinical Dementia Rating (CDR),20 routine laboratory tests including brain computed tomography (CT) and/or magnetic resonance imaging (MRI), and an interview. The interview elicited demographic data including years of education, information about their medical and psychiatric histories and medications currently being used, and the duration of illness.

As a result of the screening, 107 subjects meeting the following criteria participated in the study: suffering from probable AD based on NINCDS-ADRDA21 criteria, living in a private residence with responsible caregivers, not having a current or past diagnosis of psychiatric illnesses other than AD.

The frequencies of the behavioral disturbances observed by primary caregiver were assessed using the Troublesome Behavior Scale (TBS),22 which was designed to quantify specific observable behavior usually associated with dementia. The reliability and validity of the TBS have been established. Each of 14 questions assess the frequency of verbal and physical activity that is likely to create a burden for caregivers in the preceding month (from 0 = never, to 4 = once everyday and more). The 14 items of the TBS are false accusation, ill-natured denial and/or distortion, hiding and/or losing things, interfering with a happy home circle, being restless and/or noisy at night, physical and/or verbal abuse, repetition and/or clinging, wandering, pica, rummaging, making the dwelling dirty, crying and/or screaming, dangerous behavior, and quarreling with others.

We employed confirmatory factor analysis to estimate values for the latent variables of the scale, which are sums of direct and indirect contribution of the observed variables. The analysis revealed three structural variables: factor 1 (agitation), factor 2 (hyperactivity), and factor 3 (miscellaneous). Each of the items and their estimated loadings with scoring methods have been presented in a previous report.18

Using clinical screening questions, we also asked the caregivers to evaluate whether the visual, auditory and speech functions of the patients were preserved or impaired using simple yes/no questions. We studied the interrater reliability of the caregiver's evaluation by calculating a measure of agreement (κ), from 52 randomly selected pairs of primary caregivers and other family members. The values for vision, hearing, and speech were 0.95, 0.89, and 0.92, respectively, which confirmed the interrater reliabilities.

However, all the main caregivers were asked about the nature of the relationship they had with dependents, the number of cohabiting family members, and the presence or absence of official and self-help services used (e.g. day and respite cares, training course for caregivers). To determine their psychological health, they were given the 30-item General Health Questionnaires (GHQ30)23 to complete.

Thereafter, for as long as a subject lived in his/her own home with family members, we made annual evaluations using the abovementioned scales. Between April and July 1994, we asked the subjects who were still participating to provide blood samples for ApoE genotyping.24 Informed consent was obtained from each subject, or when necessary the legal guardian.

Principal components analysis with varimax rotation was performed to obtain summary measures for the patient clinical variables at entry. This analysis identified three factors corresponding to stage of dementia (CDR: factor loading = 0.91, Barthel: 0.69, HDSR: 0.85), functions related to communication (speech: 0.79, hearing: 0.51, vision: 0.73), and behavioral disturbances (agitation: 0.96, hyperactivity: 0.65, miscellaneous: 0.87). This component structure accounted for 74% of the variance. Scores for these three factors were constructed for all subjects. Besides these scores, we employed demographics (age at onset, sex, years of education), number of ApoE ɛ4 alleles and use of psychotropics as explanatory variables in the multiple regression analysis described below. In addition, the following caregiver characteristics were employed: the relationship to the dependent (spouse or others), the number of cohabiting family members (>3 or ≤3), GHQ30 score, and support services used (2: both of social and self-help, 1: one of them, 0: none).

Unlike cognitive function and ADL, which decline over the course of AD, there appears to be a tendency for behavioral disturbance to occur in a relatively stage-specific sequence. Therefore, the maximum value (Max) and minimum value (Min) for each behavioral factor between the second and the last annual observations of each subject were used as the indices of the overall severity of behavioral disturbances during the course. In order to clarify which patient characteristics and caregiver variables at the entry were predictive of the two indices, we performed the stepwise procedure of multiple regression analysis (SAS) using only the data for those subjects who had undergone at least three annual assessments. We assumed that the above patient variables would play a fundamental role in the manifestation of the behavioral disturbances, and the caregiver variables would modify them. Therefore, we performed preliminary analyses using only the patient variables: stage of dementia, functions related to communication, behavioral disturbances, demographics (age at onset, sex, education), number of ɛ4 alleles, and use of psychotropics. The caregiver variables were then entered into the models which included all the patient variables that had contributed significantly to the indices in the preliminary analyses.


The final subjects of the present study were 62 patients who underwent at least three assessments (mean: 5.2, range: 3–7). There are two reasons why we used only the data from these patients. First, the significance of the ApoE gene for the development of AD was first reported after we had started the present study, and we began ApoE genotyping in 1995. Thus, for patients who had died or become institutionalized before 1995, blood samples for ApoE genotyping were not available. Second, with the exception of the initial results, we used two indices (Max and Min) to represent the overall severity of behavioral disturbance during the observation period.

The baseline characteristics of the subjects and their main caregivers were as follows: subjects: age, 77.5 ± 10.4 years; male sex, 30%; years of education, 9.7 ± 2.3 years; CDR score (sum of boxes), 9.3 ± 3.2; HDSR score, 15.4 ± 6.8; Barthel index score, 89.5 ± 15.8; presence of communication impairment, 41%; psychotropics user, 15%; ApoE ɛ4 carrier, 61%. Caregivers: age, 55.7 ± 11.9 years; male sex, 15%; spouse caregiver, 27%; number of family members, 3.4 ± 1.5; GHQ30 score, 10.5 ± 8.1; support services user, 39%.

At the end of 6-year follow up, the characteristics of the final subjects were as follows: CDR score (sum of boxes), 16.5 ± 2.5; HDSR score, 7.5 ± 6.3; Barthel index score, 38.1 ± 22.7; presence of communication impairment, 58%.

Table 1 shows the results of the multiple regression analyses which clarified predictive factors of the overall severity of behavioral disturbances. For both the indices (Max and Min), the baseline severity of the behavioral disturbances was the strongest predictor, followed by the baseline stage of dementia or use of support services.

Table 1. Table 1.  Predictors of the overall severity of the behavioral disturbances. A higher Troublesome Behavior Scale (TBS) score, earlier stage of dementia and no use of support services at the study entry were predictive of more severe symptomatology for ‘agitation’ and ‘miscellaneous’. Only higher TBS score was predictive of more severe symptomatology for ‘hyperactivity’
Factor 1 (Agitation)
MaximumR2 = 0.38 MinimumR2 = 0.35
ItemParameter estimateP valueItemParameter estimateP value
Intercept− 0.270.35Intercept− 0.59< 0.01
First TBS0.27< 0.011st TBS0.18< 0.01
Stage0.01< 0.05Stage0.01< 0.05
Services− 0.42< 0.05Services− 0.24< 0.05
Factor 2 (Hyperactivity)
MaximumR2 = 0.26 MinimumR2 = 0.30
ItemParameter estimateP valueItemParameter estimateP value
Intercept0.150.19Intercept− 0.86< 0.01
First TBS0.28< 0.011st TBS0.25< 0.01
Factor 3 (Miscellaneous)
MaximumR2 = 0.36 MinimumR2 = 0.30
ItemParameter estimateP valueItemParameter estimateP value
Intercept− 0.430.17Intercept− 0.36< 0.05
First TBS0.25< 0.011st TBS0.14< 0.01
Services− 0.42< 0.05Stage0.01< 0.01
Stage0.01< 0.05Services− 0.23< 0.05

With respect to the ApoE ɛ4 allele, the analysis revealed only a tendency (parameter estimate 0.21, P = 0.08) to suggest an association between the allele and the maximum value of ‘miscellaneous’.


Using the longitudinal data for up to 6 years, we determined the predictive factors of the behavioral disturbances during the course of AD. As putative predictors, we employed ApoE genotyping and the situation of caregivers, as well as the clinical and demographic characteristics of the patients.

The main finding of this study was that the baseline severity of behavioral disturbances was the strongest predictor of the overall severity. In addition, as we have already reported, the correlation coefficients between the Max and the Min showed higher values.18 Therefore, there should be a subgroup of patients in whom the disturbances occur continuously or tend to recur. Thus, we should bear in mind that patients who manifest more behavioral symptomatology at a point will do so for several years.

Even if the disturbances occur in a stage-specific sequence, most of the disturbances examined here can be manifested by those who retain relatively better function. Thus, it is likely that subjects at a milder stage (better function) at the baseline manifest more symptomatology during subsequent observations.

Taking these findings together, the behavioral disturbances examined appear to be innately determined phenomena rather than merely representing stages of AD.

In contrast, use of support services at entry predicted less symptomatology with respect to ‘agitation’ and ‘miscellaneous’. This result suggests that day or respite care itself might have an effect in ameliorating the disturbances. A more likely explanation is that caregivers who use these services, especially as training courses and self-help networks, might obtain practical knowledge which is useful for coping with the disturbances.

Based on cross-sectional data, several study groups have examined the relationship between the ɛ4 allele and psychopathology in patients with AD.6–11 Of these studies, Ramachandran et al. found a threefold increase in depression and psychosis among ɛ4 allele carriers6 and Murphy et al. found more psychopathology among the carriers.7 However, the remaining four studies did not reveal such results. As far as behavioral disturbances examined in the present study are concerned, we conclude that the allele has a marginal effect, if any, on some forms of disturbance.

When interpreting these results, we must be aware of possible selection bias in the present study. To evaluate the longitudinal course of the behavioral disturbances, we used only the data from those who had undergone at least three annual assessments. However, if behaviorally disturbed patients were less likely to continue participating in the follow ups, the results of the present study would have been systematically biased by the absence of the patients with more severe behavioral disturbances.

To examine these possibilities, we compared the 62 final subjects and the 45 who had dropped out with regard to the following items: age at onset of AD, sex, behavioral symptomatology (TBS score), global dementia severity (sum of CDR boxes) and cognitive function (HDSR score). As a result, age at onset and percentage of male patients were higher for the drop-outs than for the final subjects (P < 0.05). No significant difference was found between the groups for any of the other items. Thus, we must be aware that age at onset and sex might have biased the present results to some extent.

A major limitation of the present study is that we did not evaluate psychiatric features such as depressive and psychotic symptoms. Although current behavior rating scales in widespread use include items related to these features,25–27 no Japanese versions of these scales were available when we commenced the present study. However, with the omission of psychiatric features, the items of the TBS are similar with those of the widespread scales. Therefore, most of the observable behaviors manifested as verbal and physical activities are covered by the TBS. Finally, the present results must be considered important for those who have to deal with elderly patients with AD in a clinical setting.