- Email: [email protected]
Predictive Validity of Neuropsychiatric Subgroups on Nursing Home Placement and Survival in Patients With Alzheimer Disease
Predictive Validity of Neuropsychiatric Subgroups on Nursing Home Placement and Survival in Patients With Alzheimer Disease Saw-Myo Tun, M.A., Daniel L. Murman, M.D., M.S., Heidi L. Long, M.S., Christopher C. Colenda, M.D., M.P.H., Alexander von Eye, Ph.D.
Objective: The aim of the study was to conceptualize neuropsychiatric symptoms in patients with Alzheimer disease as distinct symptom profiles with differential disease outcomes. Two outcomes of interest in the study were nursing home placement and survival. Method: Cluster analysis was used to categorize 122 patients with Alzheimer disease based on their neuropsychiatric symptoms as assessed by the Neuropsychiatric Inventory. Both the presence as well as the severity and frequency of symptoms were considered. After identification of the subgroups, the predictive validity of the categorization was tested on time to nursing home placement and time to death over a three-year period. Cox proportional hazard models were used to perform survival analysis. Important covariates such as severity of cognitive and functional impairments, comorbid medical conditions, presence of parkinsonism, and marital status were adjusted at baseline. Results: Based on the presence of neuropsychiatric symptoms, three subgroups were identified: minimally symptomatic, highly symptomatic, and affective/apathetic. Over a three-year period, the highly symptomatic group had an increased risk of nursing home placement. In addition, the rates of survival were significantly lower for the highly symptomatic and the affective/apathetic subgroups. Based on the severity and frequency of symptoms, two-cluster and four-cluster solutions were produced. The groupings based on severity and frequency of symptoms predicted significant differential outcomes in survival and nursing home placement. Conclusions: Neuropsychiatric subgroups were able to predict differential outcomes and identify those with an increased risk for a worse prognosis. The findings were discussed through their research and clinical implications. (Am J Geriatr Psychiatry 2007; 15:314–327) Key Words: Neuropsychiatric subgroups, nursing home placement, survival
Received October 23, 2005; revised May 22, 2006; accepted May 25, 2006. From Departments of Psychology (S-MT, AvE) and Neurology and Opthalomology (HLL) Michigan State University, East Lansing, MI; Department of Neurological Sciences (DLM) University of Nebraska Medical Center; and Dean of College of Medicine, Texas A&M (CCC) Health Science Center, College Station, TX. Send correspondence and reprint requests to Daniel L. Murman, Associate Professor, Department of Neurological Sciences, 982045 Nebraska Medical Center, Omaha, NE 68198-2045. e-mail: [email protected] © 2007 American Association for Geriatric Psychiatry
314
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al.
A
lzheimer disease (AD) is the leading cause of dementia in the United States, accounting for 60%–70% of the dementia cases.1 Of the significant number of older adults who develop AD, a sizable portion also goes on to develop neuropsychiatric symptoms in addition to the cognitive impairments. According to a recent estimate, neuropsychiatric disturbances are present in 75% of the individuals with dementia.2 Moreover, disturbances of this nature substantially increase the likelihood of a worse outcome in a number of domains, including a more rapid decline in activities of daily living, and an increase in cost of care and caregiver distress.3–7 Thus, a combination of a high prevalence rate and the debilitating nature of the psychologic disturbances had propelled researchers to search for an underlying pathophysiology of the symptoms and their clinical correlates. It has been suggested that such undertakings could be the first step toward the development of effective interventions for the condition. In recent years, however, concerns have been raised regarding the methodologies used in the study of neuropsychiatric symptoms. Traditionally, the study of neuropsychiatric symptoms entailed focusing exclusively on one symptom such as depression and examining its impacts on the patients (e.g., 6,8). However, such an approach disregards the high occurrence of comorbid neuropsychiatric conditions in patients with dementia.9,10 For example, it has been reported by Lyketsos and colleagues2 that 55% of patients with dementia have two or more neuropsychiatric disturbances. Thus, the previous approach of examining neuropsychiatric symptoms in isolation has a limited capacity for finding consistent patterns in outcome research and for offering clinically meaningful causal relationships.9 –11 To address this limitation in the study of neuropsychiatric symptoms, two alternative approaches had been offered. In one such strategy, the method of factor analysis was used to create meaningful groups of symptoms that are most likely to co-occur as assessed by a scale measuring neuropsychiatric symptoms such as BEHAVE-AD and the Neuropsychiatric Inventory. Despite variations in symptoms included in each factor, most studies reported three factors, with mood, psychosis, and hyperactive/frontal factors being the most consistent findings.9,11–13 Nevertheless, although this method addresses the problem of focusing too narrowly on one symptom, factor analysis does not
Am J Geriatr Psychiatry 15:4, April 2007
place patients into subgroups. Instead, it places the patients on a continuum of a continuous factor score. If subgroups are to be derived, then an arbitrary cutoff score must be defined for each factor. In addition, because the method of factor analysis allows for individual patients to be placed in a high-dimensional factor space, it does not necessarily create well-defined groups with distinct symptom profiles.10 Thus, the approach is not ideal for examining the differential outcomes of individuals based on their neuropsychiatric symptom profiles. Another approach that has been used is to create groups of individuals with differing symptom profiles using a latent class analysis method.10 Although solutions derived from latent class analysis correspond to that of factor analysis, the purpose of this approach is to identify meaningful groups of individuals rather than groups of variables.14 Hence, the latent class analysis approach avoids the limitation of the same individuals being placed under multiple categories of symptoms. This particular feature makes latent class analysis more suitable for use in neuropsychiatric outcome studies. Based on the results of their latent class analysis study10 and that of an epidemiologic study,15 Lyketsos and colleagues16 derived a classification system for categorizing neuropsychiatric systems in patients with AD. Although their effort toward the development of an empirically driven classification system should be applauded, their methodologies did not account for the severity and frequency of the symptoms. As observed by Meyers,17 the failure to account for severity and frequency of symptoms is a serious flaw in the study of neuropsychiatric disturbances in that a patient who occasionally experiences a mild symptom of agitation is deemed to have a similar degree of pathology as that of a patient endorsing persistent and severe agitation. Along the same line, such disparate individuals will be expected to have comparable clinical outcomes. Such line of reasoning is intrinsically problematic, and outcome research based on such line of conceptualization will have limited applicability.17 In the present study, an attempt was made to address the limitations of previous studies in the following ways: 1) psychiatric symptoms in patients with AD were conceptualized as clusters of symptoms that co-occur together rather than as disparate symptoms; 2) to maximize the predictive validity of
315
Neuropsychiatric Subgroups in AD groups of individuals on outcomes, individual patients were placed in nonoverlapping rather than in multiple groups; 3) the role of the severity and the frequency of neuropsychiatric symptoms were explored; and 4) predictive validity of subgroups was tested on nursing home placement and survival over a three-year period. To accomplish these aims, we used the method of cluster analysis. By using this flexible approach, the study was able to consider both the presence of symptoms, which is categorical in nature, and the frequency and the severity in symptoms, which are continuous. Also, cluster analysis has been suggested as a suitable method for looking at differences in profiles between individuals while avoiding the loss of critical information.18 On the other hand, von Eye and coauthors18 noted that the use of cluster analysis sometimes raises the concern of arbitrariness. To address such a concern, the authors outlined a series of decision-making steps to ensure that the choice of cluster analysis method is based on a well-justified position rather than a blind application of the method. Here, a few of the critical steps in the decisionmaking process are reported. First, there was a decision concerning whether there are disjoint or overlapping clusters. Given that one of the main objectives of the study was to identify nonoverlapping symptom profiles, a disjoint cluster approach appeared to be a natural choice. A second critical decision concerned the hierarchical versus nonhierarchical structures of the clusters. In this regard, an assumption of the study that the clustering of the patients into smaller groups will have an interpretable meaning led us to choose a hierarchical method. The third, and a fundamental decision of the study, related to the choice of base measures. Because the study was interested in examining the presence of neuropsychiatric symptoms, as well as the severity and frequency, the use of Pearson’s correlation appeared the most suitable. In the second segment of the study, the clinical use of the classification was tested. When classifying individuals into subgroups, the usefulness of the procedure depends not only on the identification of the distinct groups, but also on its ability to have predictive validity.14 Therefore, the study examined the differential impact of neuropsychiatric profiles on two domains of clinical interest, namely nursing home placement and survival.
316
METHOD Study Design The present study was a cohort study with crosssectional and longitudinal components. Specifically, data from the initial time point was used to classify patients with AD into groups based on the presence of neuropsychiatric symptoms and on the severity and the frequency of symptoms. After the classification, the groupings were used to predict the differential rates in time to nursing home placement and survival over a 3-year period.
Participants The participants were drawn from the Cost of Health in Alzheimer disease Relative to Gained Effectiveness (CHARGE) study at Michigan State University.3,19 Participants who met the study criteria were recruited by mail from six neurology practices and three geriatric medicine clinics in Michigan. The inclusion criteria for the study were a clinical diagnosis of probable AD using the Alzheimer’s Disease and Related Disorders Association–National Institute of Neurological Communicative Disorder and Stroke (NINCDS-ADRDA) criteria and the availability of an informant who was in direct contact with the patient at least once a week. Given that different types of dementia may have distinctive neuropsychiatric profiles, only patients with the AD diagnosis were included to maintain the homogeneity of the sample. Of the 692 eligible participants, 128 patients (18%) responded and were subsequently included in the study. Ninety-seven percent of the participants in the study were white. The mean age of the participants was 76.2 years (standard deviation [SD]: 9.0) and the mean level of education was 12.9 years (SD: 3.1). Fifty-five percent of the participants were female. In many cases, the spouse served as an informant for the study (60%). In the remainder of the cases, daughters (30%), other relatives (7%), and friends of the patient (3%) were the informants. Twenty-two percent of the sample resided in a longterm care setting at baseline of which 41% was in an assisted living setting, 30% was in a foster care home, and another 18% was in a nursing home. The type of long-term care for three (11%) of the participants was
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al. not known. Also of note, two participants (2%) dropped out from the study at follow ups. Procedure After obtaining informed consent, trained examiners interviewed the patient and his or her informant separately for approximately 45 minutes. During the interview with the patient, he or she was asked to undergo a neurologic examination and an assessment of neuropsychologic functioning. The informant, meanwhile, provided the examiner with information on patients’ symptoms and comorbid medical conditions. For three years after the initial interview, the informant was contacted yearly by phone to complete surveys and for an interview. In these follow-up contacts with the informant, he or she reported the significant events in the past year such as hospitalization, institutionalization, or death. Measures The study used three questionnaires, three yearly interviews, a cognitive measure, and a neurologic rating scale. Of those, the Neuropsychiatric Inventory, the Cumulative Illness Rating Scale, the Dependence Scale, the Mini-Mental State Examination, and the Modified Colombia University Parkinson’s Disease Rating Scale were given at the initial time point. The interviews were given at each yearly follow up. Neuropsychiatric Inventory. The Neuropsychiatric Inventory (NPI) was used to assess the AD-related behavior symptoms that were present in the past month. This structural interview was developed by Cummings and colleagues20 to examine the presence, severity, and frequency of 10 commonly observed neuropsychiatric symptoms in patients with dementia: delusion, hallucination, agitation/aggression, depression, anxiety, euphoria/elation, apathy, disinhibition, irritability, and aberrant motor behavior. The NPI uses an informant to report on symptoms that they have observed in the patient in a specific time period. In a report by Cummings,21 the NPI was noted to have a strong interrater reliability with r ranging from 0.94 –1.0 and a test–retest reliability ranging from 0.79 – 0.86. Moreover, it was reported that the NPI has good construct and content validity and therefore is commonly used in dementia research.10,22 In the current study, the score indicat-
Am J Geriatr Psychiatry 15:4, April 2007
ing the presence of a neuropsychiatric symptom as well as the domain score that incorporates the severity and frequency of the symptom were used. In the presence of a neuropsychiatric disturbance, the domain score can range from 1–12. For example, if the informant endorses a symptom such as depression, he or she will rate the severity of that symptom, which ranges from 1– 4, and rate the frequency of the symptoms, which can range from 1–3. The domain score is the severity of symptom multiplied by the frequency of the symptom. Outcome Measures Follow-Up Interviews. At each yearly follow-up interview, the informant was asked whether the patient had been placed in a nursing home facility, which included traditional nursing home setting, an assisted living facility, and an adult foster care home, or had died. The interviewer recorded the date of such event, which was used to calculate the time to events over the follow-up period. Measures of Confounding Variables Mini-Mental State Examination. The Mini-Mental State Examination (MMSE) is a commonly used measure for determining the level of cognitive functioning in an individual. It has been reported that the MMSE is appropriate for use in examining the progression of cognitive decline and is able to capture a wide range of impairment in patients with AD.23 The scale on the MMSE ranges from 0 –30. Scores below 24 are taken as an indicator of impairment. In the present study, the score on the MMSE was used to adjust for cognitive impairment at baseline. In addition, a MMSE score of zero was used to exclude patients with very severe dementia, because at this stage of dementia, it is difficult to accurately assess for the presence of symptoms. Therefore, data from a total of six patients were excluded from the analyses, which reduced the original sample size of 128 to 122. Dependence Scale. The Dependence Scale is designed to measure the functional decline associated with progression of dementia as reported by an informant who is familiar with the patient’s daily activities.24 The measure contains 13 items, and the items are scored to determine the level of dependency ranging from 0 –5. Stern and colleagues24 re-
317
Neuropsychiatric Subgroups in AD ported the scale to have an intraclass correlation of 0.90, indicating strong interrater reliability and a good validity with other functional impairment scales. This measure was included to examine the patient’s functional impairment. Cumulative Illness Rating Scale. To control for the potentially confounding impacts of comorbid medial conditions on the outcome measures, the Cumulative Illness Rating Scale (CIRS) was used to evaluate the presence and the severity of comorbid medical conditions based on the functioning of the 13 organ systems.25 This measure has been used successfully to predict hospitalization and death.26 Modified Columbia University Parkinson’s Disease Rating Scale. This scale was used to quantify the presence of signs of parkinsonism that are commonly observed in patients with Parkinson disease and AD such as resting tremor, rigidity, bradykinesia, and posture abnormalities. The reliability of this scale is respectable, as has been previously reported.27 Severity of parkinsonism predicts the rate of functional decline and is correlated with some neuropsychiatric symptoms (e.g., psychosis). In addition, the presence of parkinsonism in patients with AD is a strong predictor of nursing home placement and survival.24 Thus, we used the total score on this scale to adjust for group differences in parkinsonism at baseline that may influence outcomes. Statistical Analyses Cluster Analysis. To account for both the presence and the severity of neuropsychiatric symptoms, two sets of cluster analyses were conducted. For both of the analyses, weighted scores were used to correct for high intercorrelations observed between some of the NPI domains. The Ward’s minimum sum of squares method with Pearson’s correlation as base measure was used. Because Ward’s method, like other types of hierarchical agglomerative methods, is sensitive to outliers, a nonexhaustive approach was used.28 Analyses on Outcome Measures. The predictive validity of the classification systems was tested in parallel using two outcome measures. A precedent for such a methodological approach has been set in previous studies.29 Severity of cognitive and functional impairments, parkinsonism, comorbid medical conditions, and/or marital status were adjusted at base-
318
line. The decision to adjust for a covariate was based on previous reports of a relationship with the outcome and on examination of cluster differences for a particular variable. For instance, if a particular variable was previously found to be a significant predictor of the outcome variable of interest, we examined whether there were significant differences between the clusters, and if so, the variable was included in the model. During the follow-up interviews, information was collected on whether the patient moved into longterm care or died during the follow up. The date of such events was recorded and used to calculate time to events during the 3-year follow up. Cox proportional hazard models were used to perform survival analysis because they allowed for inclusion of continuous and categorical variables.
RESULTS Presence of Neuropsychiatric Symptoms Based on the presence of neuropsychiatric symptoms, cluster analysis suggested a three-cluster solution. Cluster mean profiles are presented in Table 1. Cluster 1 (N ⫽ 38) had the lowest mean score on the NPI and thus was deemed to be “minimally symptomatic.” Cluster 2 (N ⫽53) had the highest mean score on the NPI, and the group as a whole appeared to be “highly symptomatic.” Cluster 3 (N ⫽ 31) had a moderate degree of symptoms, and an examination of the distribution of mean scores indicated that the group was “affective/apathetic” (see Figure 1). Nursing Home Placement. At the three-year time point, 11 of the 32 (34%) in the “minimally symptomatic” group, 27 of the 49 (55%) in the “highly symptomatic” group, and 11 of 27 (41%) in the “affective/ apathetic” group had been placed into nursing homes. On average, the number of days to nursing home placement was 914.90 days (SD: 321.85) for the “minimally symptomatic” group, 708.92 days (SD: 440.85) for the “highly symptomatic” group, and 808.26 days (SD: 399.90) for the “affective/apathetic” group. Figure 2 illustrates the observed survival curves in days to nursing home placement in three subgroups. The Cox proportional hazard model showed that
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al.
TABLE 1.
Cluster Mean Profiles for the Presence of Symptoms
Age (SD) MMSE score CIRS score EPS score Dependence score Total NPI score Delusions Hallucinations Agitation Depression Anxiety Elation Apathy Disinhibition Irritability Aberrant motor behavior
Cluster 1 (N ⴝ 38)
Cluster 2 (N ⴝ 53)
Cluster 3 (N ⴝ 31)
Fa
p
75.79 (9.14) 16.58 (7.60) 19.45 (3.36) 2.84 (2.70) 2.76 (1.03) 5.58 (6.83) 0.08 (0.27) 0.03 (0.16) 0.11 (0.31) 0.18 (0.39) 0.08 (0.27) 0.11 (0.31) 0.08 (0.27) 0.18 (0.39) 0.06 (0.23) 0.32 (0.47)
76.81 (8.42) 15.19 (8.28) 21.70 (3.81) 3.17 (3.89) 3.60 (1.10) 28.60 (14.29) 0.40 (0.49) 0.32 (0.47) 0.81 (0.40) 0.64 (0.48) 0.66 (0.41) 0.21 (0.42) 0.77 (0.42) 0.57 (0.50) 0.73 (0.45) 0.51 (0.51)
76.71 (8.92) 16.71 (6.37) 20.48 (3.74) 3.65 (3.18) 3.03 (1.05) 13.74 (9.78) 0.29 (0.46) 0.16 (0.37) 0.13 (0.34) 0.42 (0.50) 0.42 (0.50) 0.00 (0.00) 0.94 (0.25) 0.00 (0.00) 0.11 (0.32) 0.29 (0.46)
0.03 0.30 3.90 0.35 7.42 48.40 6.12 6.94 56.87 10.87 20.06 4.16 65.85 22.39 45.27 2.72
0.97 0.74 0.001 0.71 0.001 0.001 0.003 0.001 0.001 0.001 0.001 0.02 0.001 0.001 0.001 0.07
Fa – df ⫽ 2. SD: standard deviation; total NPI score: total score on the Neuropsychiatric Inventory; MMSE score: score on the Mini-Mental State Examination; CIRS score: score on the Cumulative Illness Rating Scale; EPS score: score on the Modified Columbia University Parkinson’s Disease Rating Scale; Dependence score: score on the Dependence Scale.
FIGURE 1.
Cluster Means of Neuropsychiatric Symptoms Based on the Presence of Symptoms
the overall model, which included the three cluster solutions based on the present of neuropsychiatric symptoms and the covariate of presence of parkinsonism, predicted time to nursing home placement (F⫽21.580; df⫽ 1, 107, p ⬍0.001). The choice of the
Am J Geriatr Psychiatry 15:4, April 2007
covariates to be included in the model was based on previous reports, correlations, and significant group differences in scores, and our aim was to keep the most parsimonious model. Of the covariates that were considered, presence of parkinsonism (2 ⫽
319
Neuropsychiatric Subgroups in AD 20.405; df⫽1, p ⬍0.001) was the only significant covariate. It should be noted that adding other covariates of functional impairment, MMSE, and severity of comorbid conditions did not produce an appreciable difference in results. After adjusting for the covariate, the subgroups, based on the presence of neuropsychiatric disturbances, significantly predicted nursing home placement at the three-year time point (Tarone-Ware test; 2 ⫽6.503; df⫽2, p ⬍0.05). However, pairwise comparisons did not indicate significant relationships (see Table 2 and Figure 2). Survival. At the three-year time point, 5 of 37 (14%) in the “minimally symptomatic” group, 20 of 53 (38%) in the “highly symptomatic” group, and 12 of 31 (39%) in the “affective/apathetic” group had died. The mean numbers of days alive for each group were as followed: 1060.16 days (SD: 132.19) for the “minimally symptomatic” group, 933.77 (SD: 272.45) days for the “highly symptomatic” group, and 969.23 days (SD: 227.28) for the “affective/apathetic” group. Figure 3 illustrates the observed survival curves for time to death in the three subgroups. Using the Cox proportional hazard model, the overall model with the three clusters and the presence of parkinsonism as a covariate significantly predicted death (F⫽34.371; df⫽1, 120, p ⬍0.001) at the 3-year time point. The presence of parkinsonism (2 ⫽ 28.242; df⫽1, p ⬍0.001) was a significant predictor of death. Inclusion of additional covariates did not contribute to a significant change in model. Thus, with
TABLE 2.
the presence of parkinsonism adjusted, the clusters based on the presence of symptoms remained a significant predictor of death over a three-year period (Tarone-Ware test; 2 ⫽7.280; df ⫽2, p ⬍0.05). In addition, pairwise comparisons indicated that “highly symptomatic” group was 3.6 times more likely to die than the “minimally symptomatic” group. Moreover, the “affective/apathetic” group was 3.2 times more likely to die than the “minimally symptomatic” group (see Table 2). Severity and Frequency of Neuropsychiatric Symptoms Based on the severity and frequency of symptoms, cluster analysis suggested two-cluster and four-cluster solutions. Characteristics of the subgroups for four-cluster solutions are presented in Table 3. In the two-cluster solution, cluster 1 (N⫽69) endorsed minimal symptoms, and cluster 2 (N⫽53) endorsed a high level of symptoms. In the four-cluster solution, cluster 1 (N⫽ 42) was noted to be “minimally symptomatic.” In cluster 2 (N⫽39), “affective symptoms” such as depression, anxiety, and apathy were noted. Cluster 3 (N⫽27) included individuals who were “predominantly apathetic.” Individuals in cluster 4 (N⫽14) were “highly symptomatic with psychotic features” (see Fig. 4). It should be noted that the mean profiles are based on the 0 –1 scale; however, the clusters were determined by the frequency and severity score.
Pairwise Compairsons Between Clusters Wald2
df
RR
p
0.525 0.137
2.824 0.136
1 1
1.691 1.147
0.09 0.71
0.589 0.388 0.606
3.110 1.099 1.600
1 1 1
1.803 1.474 1.834
0.08 0.29 0.21
1.285 1.171
6.506 4.774
1 1
3.614 3.226
0.01 0.03
0.988 0.925 2.240
3.334 2.629 14.564
1 1 1
2.685 2.521 9.396
0.07 0.11 0.001
Predictor Nursing home placement Presence of symptoms Cluster 1 versus cluster 2 Cluster 1 versus cluster 3 Frequency and severity of symptoms Cluster 1 versus cluster 2 Cluster 1 versus cluster 3 Cluster 1 versus cluster 4 Probability of death Presence of symptoms Cluster 1 versus cluster 2 Cluster 1 versus cluster 3 Frequency and severity of symptoms Cluster 1 versus cluster 2 Cluster 1 versus cluster 3 Cluster 1 versus cluster 4 Test used: Cox regression pairwise comparisons.
320
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al.
FIGURE 2.
Estimated Probability of Nursing Home Placement based on the Presence of Neuropsychiatric Symptoms
FIGURE 3. Estimated Probability of Death based on Presence of Symptoms
Am J Geriatr Psychiatry 15:4, April 2007
321
Neuropsychiatric Subgroups in AD
TABLE 3.
Characteristics of the Subgroups for Four-Cluster Solution Based on the Severity and Frequency of Neuropsychiatric Symptoms
Mean age (SD) Mean MMSE score Mean CIRS Mean EPS score Dependence score Total NPI score Delusions Hallucinations Agitation Depression Anxiety Elation Apathy Disinhibition Irritability Aberrant motor behavior
Cluster 1 (N ⴝ 42)
Cluster 2 (N ⴝ 39)
Cluster 3 (N ⴝ 27)
Cluster 4 (N ⴝ 14)
Fa
p
75.60 (9.44) 17.67 (7.13) 19.50 (3.37) 2.50 (2.18) 2.76 (0.98) 4.50 (4.75) 0.07 (0.26) 0.05 (0.22) 0.19 (0.40) 0.26 (0.45) 0.17 (0.38) 0.07 (0.26) 0.14 (0.35) 0.14 (0.35) 0.13 (0.34) 0.21 (0.42)
75.69 (8.34) 14.90 (8.38) 22.18 (3.83) 3.41 (4.54) 3.51 (1.12) 25.74 (7.49) 0.31 (0.47) 0.21 (0.41) 0.67 (0.48) 0.72 (0.46) 0.64 (0.49) 0.21 (0.41) 0.80 (0.41) 0.59 (0.50) 0.49 (0.51) 0.54 (0.51)
78.22 (7.98) 16.59 (6.82) 20.74 (3.45) 3.78 (3.26) 3.20 (1.18) 11.78 (7.23) 0.26 (0.45) 0.19 (0.40) 0.19 (0.40) 0.22 (0.42) 0.33 (0.48) 0.00 (0.00) 0.93 (0.27) 0.11 (0.32) 0.25 (0.44) 0.15 (0.36)
77.86 (9.11) 13.00 (7.46) 20.00 (4.11) 3.50 (2.50) 3.64 (1.01) 45.93 (9.68) 0.79 (0.43) 0.57 (0.51) 0.86 (0.36) 0.64 (0.50) 0.71 (0.47) 0.29 (0.47) 0.79 (0.43) 0.36 (0.50) 0.92 (0.28) 1.00 (0.00)
0.92 1.96 3.90 1.17 4.21 153.21 11.48 7.23 16.44 10.21 10.00 3.78 34.07 10.23 13.93 17.44
0.44 0.13 0.01 0.33 0.007 0.001 0.001 0.001 0.001 0.001 0.001 0.01 0.001 0.001 0.001 0.001
Fa ⫺ df ⫽ 3. SD: standard deviation; total NPI score: total score on the Neuropsychiatric Inventory; MMSE score: score on the Mini-Mental State Examination; CIRS score: score on the Cumulative Illness Rating Scale; EPS score: score on the Modified Columbia University Parkinson’s Disease Rating Scale; Dependence score: score on the Dependence Scale.
FIGURE 4.
Four-Cluster Means of Neuropsychiatric Symptoms based on the Severity and Frequency of Symptoms
An advantage of the four-cluster solution over the two-cluster model in this portion of the analyses was that it was able to provide finer differentiation. In other words, the four-cluster solution provided information on the roles of the “affective symptoms”
322
group (cluster 2) and the “predominantly apathetic” group (cluster 3) in addition to the low and high symptom groups seen with the two-cluster solutions. Thus, a decision was made to omit the two-cluster solution from further discussion.
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al. Nursing Home Placement. In Table 4, the number of individuals placed in nursing home after a threeyear period and the average time to nursing home are presented for each cluster. Survival curve for the four-cluster solution is presented in Figure 5. In the model, the covariate presence of parkinsonism was included. Inclusion of severity of comorbid medical conditions, marital status, and severity of functional impairment did not produce appreciable difference to the model. Results from the Cox proportional hazard model showed that the overall model, which included the four clusters, along with the presence of parkinsonism, was a significant predictor of nursing home placement (F ⫽18.936; df⫽ 1, 107, p ⬍0.001). In addition, the presence of parkinsonism (2 ⫽ 17.885; df ⫽1, p ⬍0.001) was a significant predictor. Similarly, the differential outcome of nursing home placement as predicted by the fourcluster solution was significant (Tarone-Ware test; 2 ⫽7.965; df⫽3, p ⬍0.05). Pairwise comparisons did not yield significant results (see Table 2). Survival. Table 4 shows the number of individuals in each subgroup who have died and the mean times to death. The survival curves are presented in Figure 6. The overall Cox proportional hazard model with the covariate of presence of parkinsonism was again significant (F ⫽29.151; df⫽1, 120, p ⬍0.001) in predicting mortality over a three-year period. The presence of parkinsonism was a significant covariate (2 ⫽23.787; df⫽1, p ⬍0.001). Moreover, cluster membership was a significant predictor of death (Tarone-Ware test; 2 ⫽19.337; df⫽3, p ⬍0 .001), indicating that the “highly symptomatic” group was
TABLE 4.
9.4% times more likely to die over a three-year period than the “minimally symptomatic” group (see Table 2).
CONCLUSIONS Overall, the current findings appear to provide evidence that AD patients with neuropsychiatric symptoms can be meaningfully grouped based on the presence of symptoms as well as the severity and frequency of symptoms. A decision was made to present the findings on presence and severity of symptoms in a parallel fashion rather than attempt to aggregate the findings because it could potentially lead to a loss of critical information. As mentioned previously, a precedent for such an approach had been set in past studies.29 The first major finding of the study was that based on the presence of neuropsychiatric symptoms, patients with AD can be categorized into three distinct neuropsychiatric profiles: minimally symptomatic, highly symptomatic, and affective/apathetic. Using the same methodology, categorization of patients with AD based on the severity and frequency of symptoms yielded two possible ways of grouping. With the first possibility, patients can be parsimoniously identified as minimally or highly symptomatic. However, with this strategy, limited information can be gained beyond the general impact of neuropsychiatric symptoms on the individual. With the second possible way of grouping, on the other hand, a finer differentiation can be achieved because four
Clusters Based on the Severity and Frequency of Neuropsychiatric Symptoms as Predictor of Nursing Home Placement and Mortality Over a Three-Year Period Nursing Home Placement
Two-cluster Cluster 1 Cluster 2 Four-cluster Cluster 1 Cluster 2 Cluster 3 Cluster 4
Mortality
Percent
Mean No. of Days to Death
SD
14 20
21 38
1,020.59 931.36
196.52 261.95
5 12 10 8
12 31 37 57
1,054.73 968.69 957.63 820.50
150.11 234.43 246.27 307.63
Percent
Mean No. of Days to Nursing Home
SD
No.
Died
22 26
37 53
862.27 736.39
367.96 430.36
68 53
11 19 12 7
31 54 52 50
932.72 754.71 711.13 690.57
324.14 416.91 421.98 475.49
41 39 27 14
No.
In Nursing Home
59 49 36 35 23 14
SD: standard deviation.
Am J Geriatr Psychiatry 15:4, April 2007
323
Neuropsychiatric Subgroups in AD
FIGURE 5. Estimated Probability of Nursing Home Placement based on the Severity and Frequency of Symptoms
FIGURE 6.
Estimated Probability of Death based on the Severity and Frequency of Neuropsychiatric Symptoms (four-cluster solution)
distinct symptom profiles were identified: minimally symptomatic, affective symptoms, predominantly apathetic, and highly symptomatic with psychotic features. In reviewing the various profiles that emerged based on the presence and severity of symptoms, it
324
was of note that a “minimally symptomatic” subgroup was found consistently across the different cluster solutions. This finding was reflective of the clinical picture in that there is a segment of patients with AD who have been noted to be relatively unaffected by neuropsychiatric disturbances.1 In addi-
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al. tion, being minimally symptomatic did not appear to be a function of severity of cognitive impairments in this study, because cognitive functioning was comparable across subgroups. A second notable subgroup we identified was the “predominantly apathetic” group. The emergence of apathy as a distinct neuropsychiatric profile was in support of past reports.30,31 Similarly, the present finding on the “affective symptoms” group was consistent with the literature. Specifically, previous attempts to classify neuropsychiatric symptoms had found the “affective factor” to be one of the most consistent findings across different classification methods.9,11–13 In addition to the primary objective of identifying neuropsychiatric symptoms as distinct symptom profiles, the aim also was to test the predictive validity of subgroups on outcome measures. The present findings indicated that the subgroups, regardless of whether the groups were based on the presence of symptoms or on the severity and frequency of symptoms, were able to predict nursing home placement and death over a three-year period. In particular, regardless of how the grouping was determined, individuals with a high level of symptomatology were at high risk for both nursing home placement and for death. In contrast, individuals with minimal symptoms consistently were at a lower risk for nursing home placement and death. Thus, the high symptom profile was a risk factor for both nursing home placement and survival. The finding on the relationship between neuropsychiatric symptoms and survival was consistent with other reports in the literature.32 Of note, a potential explanation for the increased risk of mortality associated with high symptom profile is that in recent studies, there were indications that certain atypical antipsychotic medications used for the treatment of psychotic symptoms in dementia may increase the risk of mortality.33 In addition, there have been previous reports of an increased risk of cerebrovascular incident with risperidone.34 Therefore, it is conceivable that an increase risk of mortality observed in the study for individuals with high symptom profile could partially be attributed to the use of antipsychotic drugs in these patients. In future studies, it would be of importance to consider drug use when examining the relationships between neuropsychiatric symptoms and mortality in AD.
Am J Geriatr Psychiatry 15:4, April 2007
Another significant finding in the study was the role of apathy. In terms of its impact on nursing home placement, the predominantly apathetic group fell between the low and the high symptom profiles. This made intuitive sense in that the caregivers of the patients with AD endorsing apathy would find their apathetic behavior more tolerable and easy to care for than the high symptomatology. Therefore, there would be less of a need for placing the patient in a long-term care setting. An intriguing result, however, was that although caregivers may find apathy more tolerable, the presence of apathy appeared just as detrimental as the high symptom profile when it came to predicting mortality. In the three-cluster solution, the affective/apathetic and the highly symptomatic group had similarly low rates of survival. In the four-cluster solution, the predominantly apathetic group had a similar rate of survival as the predominantly affective group. Hence, it suggested that although the presence of apathy in patients with AD appears less destructive outwardly, it still has a detrimental impact on the patient’s survival and should not be overlooked as a target for treatment. Overall, it appeared that neuropsychiatric symptoms can be categorized based on the presence and severity of symptoms and that such grouping has predictive validity. In the present study, the threecluster solution produced by the presence of symptoms and the four-cluster solution based on the severity of symptoms yielded the most information. Specifically, it provided support for the emergence of apathy and affective subgroups as distinct neuropsychiatric profiles with differential outcomes in addition to the low and high symptom profiles. Moreover, it appeared that although the four-cluster solution based on the severity and frequency provided the “finest resolution,” the study indicated that the presence of symptoms also has predictive validity. Study’s Limitations There were a number of limitations in the study. First, it was possible that the presence and the severity of neuropsychiatric symptoms changed as the dementia progressed. Given that the symptoms were assessed at the onset of the study, the potential confounding influence of changes in neuropsychiatric symptoms over the three-year period could not be
325
Neuropsychiatric Subgroups in AD evaluated. Second, the NPI, used to assess neuropsychiatric symptoms in the study, was based on informant reports. Hence, it was conceivable that an informant bias in the reporting of the symptoms might have existed. However, when assessing the presence of symptoms in patients with AD with significant cognitive deficits, the use of an informant may be unavoidable. Third, although attempts were made to control for important covariates, it was not possible to control for all the potential covariates. For example, it would have been ideal to control for covariates such as caregiver burden and psychotropic drug use. Thus, it is conceivable that the effects of the covariates contributed to the findings. Fourth, the response rate for the study was low in that only 18% of the eligible sample responded to the request to participate in the study. Hence, generalizability of the study’s findings may be limited to populations similar to sample characteristics. Lastly, although the initial sample size of 122 patients with AD was respectable, the sample size became smaller as subgroups were identified. This was especially true for the four-cluster solution, which in one of its groups, contained a subsample of only 14 patients. Despite the statistically significant findings in the study even with a small sample size, it would be ideal to replicate these results using a larger sample. Directions for Future Studies Given the promising indications of neuropsychiatric symptom profiles having differential disease outcomes, further explorations are needed to test the extent of this predictive validity. For instance, it would be a useful endeavor to test the predictive validity in other outcome measures such as functional impairment, cognitive decline, and quality of
life. Along the same line, future studies could potentially test the usefulness of the methodology used for classifying neuropsychiatric disturbances in AD to other dementias such as dementia with Lewy bodies. Such application of the method to other dementias may provide an empiric validation for the clinical use of neuropsychiatric symptoms in differential diagnosis of dementias. Lastly, and perhaps most importantly, if the findings of neuropsychiatric symptom profiles having differential disease outcomes prove to be consistent across other outcome measures and dementias, an exploration into the potential neurobiologic underpinnings of the profiles appears warranted. For example, it is possible that the presence of certain underlying genetic characteristics or an involvement of particular brain structures in the disease process predispose an individual at a greater risk for a certain neuropsychiatric profile.1 By exploring such potentials, the field could move a step closer toward the development of more effective and targeted treatment options for neuropsychiatric symptoms and thereby may be able to alter the outcomes of patients with AD. Saw-Myo Tun’s work was supported, in part, by the Blue Cross Blue Shield of Michigan grant 885.SAP. Dr. Murman’s work was supported, in part, by the National Institute of Aging grant # K08-AG00864. Portions of this work have been accepted for presentation at the 58th Annual Scientific Meeting of the Gerontological Society of America, Orlando, FL, November 18 –22, 2005. This work was completed at Michigan State University College of Medicine, East Lansing, MI. Dr. Murman is on the speaker bureau for Pfizer, Inc. In addition, he receives research funding from Alzheimer’s Association.
References 1. Cummings JL: The Neuropsychiatry of Alzheimer’s Disease and Related Dementias. London, Martin Dunitz, 2003 2. Lyketsos CG, Lopez O, Jones B, et al: Prevalence of neuropsychiatric symptoms in dementia and mild cognitive impairment. JAMA 2003; 288:1475–1483 3. Murman DL, Chen Q, Powell MC, et al: The incremental direct costs associated with behavioral symptoms in AD. Neurology 2002; 59:1721–1729 4. Norton LE, Malloy PE, Salloway S: The impact of behavioral symptoms on activities of daily living in patients with dementia. Am J Geriatr Psychiatry 2001; 9:41–48 5. Kopetz S, Steele CD, Brandt J, et al: Characteristics and outcomes
326
of dementia residents in an assisted living facility. Int J Geriatr Psychiatry 2000; 15:586 –593 6. Lyketsos CG, Steele C, Baker L, et al: Major and minor depression in Alzheimer’s disease: prevalence and impact. J Neuropsychiatry Clin Neurosci 1997; 9:556 –561 7. Teri L: Behavior and caregiver burden: behavioral problems in patients with Alzheimer disease and its association with caregiver distress. Alzheimer Dis Assoc Disord 1997; 11:S35–S38 8. Payne JL, Sheppard JME, Steinberg M, et al: Incidence, prevalence, and outcomes of depression in residents of a long-term care facility with dementia. Int J Geriatr Psychiatry 2002; 17:247– 253
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al. 9. Frisoni GB, Rozzini L, Gozzetti A, et al: Behavioral syndromes in Alzheimer’s disease: description and correlates. Dement Geriatr Cogn Disord 1999; 10:130 –138 10. Lyketsos CG, Sheppard JME, Steinberg M, et al: Neuropsychiatric disturbance in Alzheimer’s disease clusters into three groups: the Cache County study. Int J Geriatr Psychiatry 2001; 16:1043–1053 11. Aalten P, de Vugt ME, Lousberg R, et al: Behavioral problems in dementia: a factor analysis of the Neuropsychiatric Inventory. Dement Geriatr Cogn Disord 2003; 15:99 –105 12. Hope T, Keene J, Fairburn C, et al: Behavior changes in dementia 2: are there behavioural syndromes? Int J Geriatr Psychiatry 1997; 12:1074 –1078 13. McShane R: What are the syndromes of behavioral and psychological symptoms of dementia? Int Psychogeriatr 2000; 12:147– 153 14. von Eye A, Bergman LR: Research strategies in developmental psychopathology: dimensional identity and the person-oriented approach. Dev Psychopathol 2003; 15:553–580 15. Lyketsos CG, Steinberg M, Tschanz JAT, et al: Mental and behavioral disturbances in dementia: findings from the Cache County Study on memory in aging. Am J Psychiatry 2000; 157:708 –714 16. Lyketsos CG, Breitner JCS, Rabins PV: An evidence-based proposal for the classification of neuropsychiatric disturbance in Alzheimer’s disease. Int J Geriatr Psychiatry 2001; 16:1037–1042 17. Meyers B: The evolving typology of neuropsychiatric complications of Alzheimer’s disease: the use of latent trait analysis. Int J Geriatr Psychiatry 2001; 16:1030 –1032 18. von Eye A, Mun EY, Indurkhya A: Typifying developmental trajectories: a decision making perspective. Psychol Sci 2004; 46: 65–98 19. Murman DL, Kuo SB, Powell MC, et al: The impact of parkinsonism on costs of care in patients with AD and dementia with Lewy bodies. Neurology 2003; 61:944 –949 20. Cummings JL, Mega M, Gray K, et al: The Neuropsychiatric Inventory: comprehensive assessment of psychopathology in dementia. Neurology 1994; 44:2308 –2314 21. Cummings JL: The Neuropsychiatric Inventory: assessing psychopathology in dementia patients. Neurology 1997; 48:S10 –S16
Am J Geriatr Psychiatry 15:4, April 2007
22. Schneider LS: Assessing outcomes in Alzheimer disease. Alzheimer Dis Assoc Disord 2001; 15:S8 –S18 23. Folstein MF, Folstein SE, McHugh PR: ‘Mini-Mental State’: a practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975; 12:189 –198 24. Stern Y, Tang M, Albert MS, et al: Predicting time to nursing home care and death in individuals with Alzheimer’s disease. JAMA 1997; 10:806 –812 25. Linn BS, Linn MW, Gurel L: Cumulative Illness Rating Scale. J Am Geriatr Soc 1968; 16:622–626 26. Miller MD, Paradis CF, Houck PR, et al: Rating chronic medical illness in burden in geropsychiatric practice and research: application of Cumulative Illness Rating Scale. Psychiatry Res 1992; 41:237–248 27. Richards M, Marder K, Bell K, et al: Interrater reliability of extrapyramidal signs in a group assessed for dementia. Arch Neurol 1991; 48:1147–1149 28. Everitt BS, Landau S, Leese M: Cluster Analysis, 4th ed. London, Arnold, 2001 29. Tubman JG, Vicary JR, von Eye A, et al: Qualitative changes in relationships between substance use and adjustment during adolescence. J Subst Abuse 1991; 3:405–414 30. Boyle PA, Mallowy PF: Treating apathy in Alzheimer’s disease. Dement Geriatr Cogn Disord 2003; 17:91–99 31. Boyle PA, Mallowy PF, Salloway S, et al: Executive function and apathy predict functional impairment in Alzheimer’s disease. Am J Geriatr Psychiatry 2003; 11:214 –221 32. Ganguli M, Dodge H, Mulsant BH: Rates and predictors of mortality in an aging, rural, community-based cohort. Arch Gen Psychiatry 2002; 59:1046 –1052 33. Schneider LS, Dagerman KS, Insel P: Risk of death with atypical antipsychotic drug treatment for dementia: meta-analysis of randomized placebo-controlled trial. JAMA 2005; 294:1934 – 1943 34. Brodaty H, Ames D, Snowdon J, et al: A randomized placebocontrolled trial for the treatment of aggression, agitation, and psychosis of dementia. J Clin Psychiatry 2003; 64:134 –143
327
Objective: The aim of the study was to conceptualize neuropsychiatric symptoms in patients with Alzheimer disease as distinct symptom profiles with differential disease outcomes. Two outcomes of interest in the study were nursing home placement and survival. Method: Cluster analysis was used to categorize 122 patients with Alzheimer disease based on their neuropsychiatric symptoms as assessed by the Neuropsychiatric Inventory. Both the presence as well as the severity and frequency of symptoms were considered. After identification of the subgroups, the predictive validity of the categorization was tested on time to nursing home placement and time to death over a three-year period. Cox proportional hazard models were used to perform survival analysis. Important covariates such as severity of cognitive and functional impairments, comorbid medical conditions, presence of parkinsonism, and marital status were adjusted at baseline. Results: Based on the presence of neuropsychiatric symptoms, three subgroups were identified: minimally symptomatic, highly symptomatic, and affective/apathetic. Over a three-year period, the highly symptomatic group had an increased risk of nursing home placement. In addition, the rates of survival were significantly lower for the highly symptomatic and the affective/apathetic subgroups. Based on the severity and frequency of symptoms, two-cluster and four-cluster solutions were produced. The groupings based on severity and frequency of symptoms predicted significant differential outcomes in survival and nursing home placement. Conclusions: Neuropsychiatric subgroups were able to predict differential outcomes and identify those with an increased risk for a worse prognosis. The findings were discussed through their research and clinical implications. (Am J Geriatr Psychiatry 2007; 15:314–327) Key Words: Neuropsychiatric subgroups, nursing home placement, survival
Received October 23, 2005; revised May 22, 2006; accepted May 25, 2006. From Departments of Psychology (S-MT, AvE) and Neurology and Opthalomology (HLL) Michigan State University, East Lansing, MI; Department of Neurological Sciences (DLM) University of Nebraska Medical Center; and Dean of College of Medicine, Texas A&M (CCC) Health Science Center, College Station, TX. Send correspondence and reprint requests to Daniel L. Murman, Associate Professor, Department of Neurological Sciences, 982045 Nebraska Medical Center, Omaha, NE 68198-2045. e-mail: [email protected] © 2007 American Association for Geriatric Psychiatry
314
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al.
A
lzheimer disease (AD) is the leading cause of dementia in the United States, accounting for 60%–70% of the dementia cases.1 Of the significant number of older adults who develop AD, a sizable portion also goes on to develop neuropsychiatric symptoms in addition to the cognitive impairments. According to a recent estimate, neuropsychiatric disturbances are present in 75% of the individuals with dementia.2 Moreover, disturbances of this nature substantially increase the likelihood of a worse outcome in a number of domains, including a more rapid decline in activities of daily living, and an increase in cost of care and caregiver distress.3–7 Thus, a combination of a high prevalence rate and the debilitating nature of the psychologic disturbances had propelled researchers to search for an underlying pathophysiology of the symptoms and their clinical correlates. It has been suggested that such undertakings could be the first step toward the development of effective interventions for the condition. In recent years, however, concerns have been raised regarding the methodologies used in the study of neuropsychiatric symptoms. Traditionally, the study of neuropsychiatric symptoms entailed focusing exclusively on one symptom such as depression and examining its impacts on the patients (e.g., 6,8). However, such an approach disregards the high occurrence of comorbid neuropsychiatric conditions in patients with dementia.9,10 For example, it has been reported by Lyketsos and colleagues2 that 55% of patients with dementia have two or more neuropsychiatric disturbances. Thus, the previous approach of examining neuropsychiatric symptoms in isolation has a limited capacity for finding consistent patterns in outcome research and for offering clinically meaningful causal relationships.9 –11 To address this limitation in the study of neuropsychiatric symptoms, two alternative approaches had been offered. In one such strategy, the method of factor analysis was used to create meaningful groups of symptoms that are most likely to co-occur as assessed by a scale measuring neuropsychiatric symptoms such as BEHAVE-AD and the Neuropsychiatric Inventory. Despite variations in symptoms included in each factor, most studies reported three factors, with mood, psychosis, and hyperactive/frontal factors being the most consistent findings.9,11–13 Nevertheless, although this method addresses the problem of focusing too narrowly on one symptom, factor analysis does not
Am J Geriatr Psychiatry 15:4, April 2007
place patients into subgroups. Instead, it places the patients on a continuum of a continuous factor score. If subgroups are to be derived, then an arbitrary cutoff score must be defined for each factor. In addition, because the method of factor analysis allows for individual patients to be placed in a high-dimensional factor space, it does not necessarily create well-defined groups with distinct symptom profiles.10 Thus, the approach is not ideal for examining the differential outcomes of individuals based on their neuropsychiatric symptom profiles. Another approach that has been used is to create groups of individuals with differing symptom profiles using a latent class analysis method.10 Although solutions derived from latent class analysis correspond to that of factor analysis, the purpose of this approach is to identify meaningful groups of individuals rather than groups of variables.14 Hence, the latent class analysis approach avoids the limitation of the same individuals being placed under multiple categories of symptoms. This particular feature makes latent class analysis more suitable for use in neuropsychiatric outcome studies. Based on the results of their latent class analysis study10 and that of an epidemiologic study,15 Lyketsos and colleagues16 derived a classification system for categorizing neuropsychiatric systems in patients with AD. Although their effort toward the development of an empirically driven classification system should be applauded, their methodologies did not account for the severity and frequency of the symptoms. As observed by Meyers,17 the failure to account for severity and frequency of symptoms is a serious flaw in the study of neuropsychiatric disturbances in that a patient who occasionally experiences a mild symptom of agitation is deemed to have a similar degree of pathology as that of a patient endorsing persistent and severe agitation. Along the same line, such disparate individuals will be expected to have comparable clinical outcomes. Such line of reasoning is intrinsically problematic, and outcome research based on such line of conceptualization will have limited applicability.17 In the present study, an attempt was made to address the limitations of previous studies in the following ways: 1) psychiatric symptoms in patients with AD were conceptualized as clusters of symptoms that co-occur together rather than as disparate symptoms; 2) to maximize the predictive validity of
315
Neuropsychiatric Subgroups in AD groups of individuals on outcomes, individual patients were placed in nonoverlapping rather than in multiple groups; 3) the role of the severity and the frequency of neuropsychiatric symptoms were explored; and 4) predictive validity of subgroups was tested on nursing home placement and survival over a three-year period. To accomplish these aims, we used the method of cluster analysis. By using this flexible approach, the study was able to consider both the presence of symptoms, which is categorical in nature, and the frequency and the severity in symptoms, which are continuous. Also, cluster analysis has been suggested as a suitable method for looking at differences in profiles between individuals while avoiding the loss of critical information.18 On the other hand, von Eye and coauthors18 noted that the use of cluster analysis sometimes raises the concern of arbitrariness. To address such a concern, the authors outlined a series of decision-making steps to ensure that the choice of cluster analysis method is based on a well-justified position rather than a blind application of the method. Here, a few of the critical steps in the decisionmaking process are reported. First, there was a decision concerning whether there are disjoint or overlapping clusters. Given that one of the main objectives of the study was to identify nonoverlapping symptom profiles, a disjoint cluster approach appeared to be a natural choice. A second critical decision concerned the hierarchical versus nonhierarchical structures of the clusters. In this regard, an assumption of the study that the clustering of the patients into smaller groups will have an interpretable meaning led us to choose a hierarchical method. The third, and a fundamental decision of the study, related to the choice of base measures. Because the study was interested in examining the presence of neuropsychiatric symptoms, as well as the severity and frequency, the use of Pearson’s correlation appeared the most suitable. In the second segment of the study, the clinical use of the classification was tested. When classifying individuals into subgroups, the usefulness of the procedure depends not only on the identification of the distinct groups, but also on its ability to have predictive validity.14 Therefore, the study examined the differential impact of neuropsychiatric profiles on two domains of clinical interest, namely nursing home placement and survival.
316
METHOD Study Design The present study was a cohort study with crosssectional and longitudinal components. Specifically, data from the initial time point was used to classify patients with AD into groups based on the presence of neuropsychiatric symptoms and on the severity and the frequency of symptoms. After the classification, the groupings were used to predict the differential rates in time to nursing home placement and survival over a 3-year period.
Participants The participants were drawn from the Cost of Health in Alzheimer disease Relative to Gained Effectiveness (CHARGE) study at Michigan State University.3,19 Participants who met the study criteria were recruited by mail from six neurology practices and three geriatric medicine clinics in Michigan. The inclusion criteria for the study were a clinical diagnosis of probable AD using the Alzheimer’s Disease and Related Disorders Association–National Institute of Neurological Communicative Disorder and Stroke (NINCDS-ADRDA) criteria and the availability of an informant who was in direct contact with the patient at least once a week. Given that different types of dementia may have distinctive neuropsychiatric profiles, only patients with the AD diagnosis were included to maintain the homogeneity of the sample. Of the 692 eligible participants, 128 patients (18%) responded and were subsequently included in the study. Ninety-seven percent of the participants in the study were white. The mean age of the participants was 76.2 years (standard deviation [SD]: 9.0) and the mean level of education was 12.9 years (SD: 3.1). Fifty-five percent of the participants were female. In many cases, the spouse served as an informant for the study (60%). In the remainder of the cases, daughters (30%), other relatives (7%), and friends of the patient (3%) were the informants. Twenty-two percent of the sample resided in a longterm care setting at baseline of which 41% was in an assisted living setting, 30% was in a foster care home, and another 18% was in a nursing home. The type of long-term care for three (11%) of the participants was
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al. not known. Also of note, two participants (2%) dropped out from the study at follow ups. Procedure After obtaining informed consent, trained examiners interviewed the patient and his or her informant separately for approximately 45 minutes. During the interview with the patient, he or she was asked to undergo a neurologic examination and an assessment of neuropsychologic functioning. The informant, meanwhile, provided the examiner with information on patients’ symptoms and comorbid medical conditions. For three years after the initial interview, the informant was contacted yearly by phone to complete surveys and for an interview. In these follow-up contacts with the informant, he or she reported the significant events in the past year such as hospitalization, institutionalization, or death. Measures The study used three questionnaires, three yearly interviews, a cognitive measure, and a neurologic rating scale. Of those, the Neuropsychiatric Inventory, the Cumulative Illness Rating Scale, the Dependence Scale, the Mini-Mental State Examination, and the Modified Colombia University Parkinson’s Disease Rating Scale were given at the initial time point. The interviews were given at each yearly follow up. Neuropsychiatric Inventory. The Neuropsychiatric Inventory (NPI) was used to assess the AD-related behavior symptoms that were present in the past month. This structural interview was developed by Cummings and colleagues20 to examine the presence, severity, and frequency of 10 commonly observed neuropsychiatric symptoms in patients with dementia: delusion, hallucination, agitation/aggression, depression, anxiety, euphoria/elation, apathy, disinhibition, irritability, and aberrant motor behavior. The NPI uses an informant to report on symptoms that they have observed in the patient in a specific time period. In a report by Cummings,21 the NPI was noted to have a strong interrater reliability with r ranging from 0.94 –1.0 and a test–retest reliability ranging from 0.79 – 0.86. Moreover, it was reported that the NPI has good construct and content validity and therefore is commonly used in dementia research.10,22 In the current study, the score indicat-
Am J Geriatr Psychiatry 15:4, April 2007
ing the presence of a neuropsychiatric symptom as well as the domain score that incorporates the severity and frequency of the symptom were used. In the presence of a neuropsychiatric disturbance, the domain score can range from 1–12. For example, if the informant endorses a symptom such as depression, he or she will rate the severity of that symptom, which ranges from 1– 4, and rate the frequency of the symptoms, which can range from 1–3. The domain score is the severity of symptom multiplied by the frequency of the symptom. Outcome Measures Follow-Up Interviews. At each yearly follow-up interview, the informant was asked whether the patient had been placed in a nursing home facility, which included traditional nursing home setting, an assisted living facility, and an adult foster care home, or had died. The interviewer recorded the date of such event, which was used to calculate the time to events over the follow-up period. Measures of Confounding Variables Mini-Mental State Examination. The Mini-Mental State Examination (MMSE) is a commonly used measure for determining the level of cognitive functioning in an individual. It has been reported that the MMSE is appropriate for use in examining the progression of cognitive decline and is able to capture a wide range of impairment in patients with AD.23 The scale on the MMSE ranges from 0 –30. Scores below 24 are taken as an indicator of impairment. In the present study, the score on the MMSE was used to adjust for cognitive impairment at baseline. In addition, a MMSE score of zero was used to exclude patients with very severe dementia, because at this stage of dementia, it is difficult to accurately assess for the presence of symptoms. Therefore, data from a total of six patients were excluded from the analyses, which reduced the original sample size of 128 to 122. Dependence Scale. The Dependence Scale is designed to measure the functional decline associated with progression of dementia as reported by an informant who is familiar with the patient’s daily activities.24 The measure contains 13 items, and the items are scored to determine the level of dependency ranging from 0 –5. Stern and colleagues24 re-
317
Neuropsychiatric Subgroups in AD ported the scale to have an intraclass correlation of 0.90, indicating strong interrater reliability and a good validity with other functional impairment scales. This measure was included to examine the patient’s functional impairment. Cumulative Illness Rating Scale. To control for the potentially confounding impacts of comorbid medial conditions on the outcome measures, the Cumulative Illness Rating Scale (CIRS) was used to evaluate the presence and the severity of comorbid medical conditions based on the functioning of the 13 organ systems.25 This measure has been used successfully to predict hospitalization and death.26 Modified Columbia University Parkinson’s Disease Rating Scale. This scale was used to quantify the presence of signs of parkinsonism that are commonly observed in patients with Parkinson disease and AD such as resting tremor, rigidity, bradykinesia, and posture abnormalities. The reliability of this scale is respectable, as has been previously reported.27 Severity of parkinsonism predicts the rate of functional decline and is correlated with some neuropsychiatric symptoms (e.g., psychosis). In addition, the presence of parkinsonism in patients with AD is a strong predictor of nursing home placement and survival.24 Thus, we used the total score on this scale to adjust for group differences in parkinsonism at baseline that may influence outcomes. Statistical Analyses Cluster Analysis. To account for both the presence and the severity of neuropsychiatric symptoms, two sets of cluster analyses were conducted. For both of the analyses, weighted scores were used to correct for high intercorrelations observed between some of the NPI domains. The Ward’s minimum sum of squares method with Pearson’s correlation as base measure was used. Because Ward’s method, like other types of hierarchical agglomerative methods, is sensitive to outliers, a nonexhaustive approach was used.28 Analyses on Outcome Measures. The predictive validity of the classification systems was tested in parallel using two outcome measures. A precedent for such a methodological approach has been set in previous studies.29 Severity of cognitive and functional impairments, parkinsonism, comorbid medical conditions, and/or marital status were adjusted at base-
318
line. The decision to adjust for a covariate was based on previous reports of a relationship with the outcome and on examination of cluster differences for a particular variable. For instance, if a particular variable was previously found to be a significant predictor of the outcome variable of interest, we examined whether there were significant differences between the clusters, and if so, the variable was included in the model. During the follow-up interviews, information was collected on whether the patient moved into longterm care or died during the follow up. The date of such events was recorded and used to calculate time to events during the 3-year follow up. Cox proportional hazard models were used to perform survival analysis because they allowed for inclusion of continuous and categorical variables.
RESULTS Presence of Neuropsychiatric Symptoms Based on the presence of neuropsychiatric symptoms, cluster analysis suggested a three-cluster solution. Cluster mean profiles are presented in Table 1. Cluster 1 (N ⫽ 38) had the lowest mean score on the NPI and thus was deemed to be “minimally symptomatic.” Cluster 2 (N ⫽53) had the highest mean score on the NPI, and the group as a whole appeared to be “highly symptomatic.” Cluster 3 (N ⫽ 31) had a moderate degree of symptoms, and an examination of the distribution of mean scores indicated that the group was “affective/apathetic” (see Figure 1). Nursing Home Placement. At the three-year time point, 11 of the 32 (34%) in the “minimally symptomatic” group, 27 of the 49 (55%) in the “highly symptomatic” group, and 11 of 27 (41%) in the “affective/ apathetic” group had been placed into nursing homes. On average, the number of days to nursing home placement was 914.90 days (SD: 321.85) for the “minimally symptomatic” group, 708.92 days (SD: 440.85) for the “highly symptomatic” group, and 808.26 days (SD: 399.90) for the “affective/apathetic” group. Figure 2 illustrates the observed survival curves in days to nursing home placement in three subgroups. The Cox proportional hazard model showed that
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al.
TABLE 1.
Cluster Mean Profiles for the Presence of Symptoms
Age (SD) MMSE score CIRS score EPS score Dependence score Total NPI score Delusions Hallucinations Agitation Depression Anxiety Elation Apathy Disinhibition Irritability Aberrant motor behavior
Cluster 1 (N ⴝ 38)
Cluster 2 (N ⴝ 53)
Cluster 3 (N ⴝ 31)
Fa
p
75.79 (9.14) 16.58 (7.60) 19.45 (3.36) 2.84 (2.70) 2.76 (1.03) 5.58 (6.83) 0.08 (0.27) 0.03 (0.16) 0.11 (0.31) 0.18 (0.39) 0.08 (0.27) 0.11 (0.31) 0.08 (0.27) 0.18 (0.39) 0.06 (0.23) 0.32 (0.47)
76.81 (8.42) 15.19 (8.28) 21.70 (3.81) 3.17 (3.89) 3.60 (1.10) 28.60 (14.29) 0.40 (0.49) 0.32 (0.47) 0.81 (0.40) 0.64 (0.48) 0.66 (0.41) 0.21 (0.42) 0.77 (0.42) 0.57 (0.50) 0.73 (0.45) 0.51 (0.51)
76.71 (8.92) 16.71 (6.37) 20.48 (3.74) 3.65 (3.18) 3.03 (1.05) 13.74 (9.78) 0.29 (0.46) 0.16 (0.37) 0.13 (0.34) 0.42 (0.50) 0.42 (0.50) 0.00 (0.00) 0.94 (0.25) 0.00 (0.00) 0.11 (0.32) 0.29 (0.46)
0.03 0.30 3.90 0.35 7.42 48.40 6.12 6.94 56.87 10.87 20.06 4.16 65.85 22.39 45.27 2.72
0.97 0.74 0.001 0.71 0.001 0.001 0.003 0.001 0.001 0.001 0.001 0.02 0.001 0.001 0.001 0.07
Fa – df ⫽ 2. SD: standard deviation; total NPI score: total score on the Neuropsychiatric Inventory; MMSE score: score on the Mini-Mental State Examination; CIRS score: score on the Cumulative Illness Rating Scale; EPS score: score on the Modified Columbia University Parkinson’s Disease Rating Scale; Dependence score: score on the Dependence Scale.
FIGURE 1.
Cluster Means of Neuropsychiatric Symptoms Based on the Presence of Symptoms
the overall model, which included the three cluster solutions based on the present of neuropsychiatric symptoms and the covariate of presence of parkinsonism, predicted time to nursing home placement (F⫽21.580; df⫽ 1, 107, p ⬍0.001). The choice of the
Am J Geriatr Psychiatry 15:4, April 2007
covariates to be included in the model was based on previous reports, correlations, and significant group differences in scores, and our aim was to keep the most parsimonious model. Of the covariates that were considered, presence of parkinsonism (2 ⫽
319
Neuropsychiatric Subgroups in AD 20.405; df⫽1, p ⬍0.001) was the only significant covariate. It should be noted that adding other covariates of functional impairment, MMSE, and severity of comorbid conditions did not produce an appreciable difference in results. After adjusting for the covariate, the subgroups, based on the presence of neuropsychiatric disturbances, significantly predicted nursing home placement at the three-year time point (Tarone-Ware test; 2 ⫽6.503; df⫽2, p ⬍0.05). However, pairwise comparisons did not indicate significant relationships (see Table 2 and Figure 2). Survival. At the three-year time point, 5 of 37 (14%) in the “minimally symptomatic” group, 20 of 53 (38%) in the “highly symptomatic” group, and 12 of 31 (39%) in the “affective/apathetic” group had died. The mean numbers of days alive for each group were as followed: 1060.16 days (SD: 132.19) for the “minimally symptomatic” group, 933.77 (SD: 272.45) days for the “highly symptomatic” group, and 969.23 days (SD: 227.28) for the “affective/apathetic” group. Figure 3 illustrates the observed survival curves for time to death in the three subgroups. Using the Cox proportional hazard model, the overall model with the three clusters and the presence of parkinsonism as a covariate significantly predicted death (F⫽34.371; df⫽1, 120, p ⬍0.001) at the 3-year time point. The presence of parkinsonism (2 ⫽ 28.242; df⫽1, p ⬍0.001) was a significant predictor of death. Inclusion of additional covariates did not contribute to a significant change in model. Thus, with
TABLE 2.
the presence of parkinsonism adjusted, the clusters based on the presence of symptoms remained a significant predictor of death over a three-year period (Tarone-Ware test; 2 ⫽7.280; df ⫽2, p ⬍0.05). In addition, pairwise comparisons indicated that “highly symptomatic” group was 3.6 times more likely to die than the “minimally symptomatic” group. Moreover, the “affective/apathetic” group was 3.2 times more likely to die than the “minimally symptomatic” group (see Table 2). Severity and Frequency of Neuropsychiatric Symptoms Based on the severity and frequency of symptoms, cluster analysis suggested two-cluster and four-cluster solutions. Characteristics of the subgroups for four-cluster solutions are presented in Table 3. In the two-cluster solution, cluster 1 (N⫽69) endorsed minimal symptoms, and cluster 2 (N⫽53) endorsed a high level of symptoms. In the four-cluster solution, cluster 1 (N⫽ 42) was noted to be “minimally symptomatic.” In cluster 2 (N⫽39), “affective symptoms” such as depression, anxiety, and apathy were noted. Cluster 3 (N⫽27) included individuals who were “predominantly apathetic.” Individuals in cluster 4 (N⫽14) were “highly symptomatic with psychotic features” (see Fig. 4). It should be noted that the mean profiles are based on the 0 –1 scale; however, the clusters were determined by the frequency and severity score.
Pairwise Compairsons Between Clusters Wald2
df
RR
p
0.525 0.137
2.824 0.136
1 1
1.691 1.147
0.09 0.71
0.589 0.388 0.606
3.110 1.099 1.600
1 1 1
1.803 1.474 1.834
0.08 0.29 0.21
1.285 1.171
6.506 4.774
1 1
3.614 3.226
0.01 0.03
0.988 0.925 2.240
3.334 2.629 14.564
1 1 1
2.685 2.521 9.396
0.07 0.11 0.001
Predictor Nursing home placement Presence of symptoms Cluster 1 versus cluster 2 Cluster 1 versus cluster 3 Frequency and severity of symptoms Cluster 1 versus cluster 2 Cluster 1 versus cluster 3 Cluster 1 versus cluster 4 Probability of death Presence of symptoms Cluster 1 versus cluster 2 Cluster 1 versus cluster 3 Frequency and severity of symptoms Cluster 1 versus cluster 2 Cluster 1 versus cluster 3 Cluster 1 versus cluster 4 Test used: Cox regression pairwise comparisons.
320
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al.
FIGURE 2.
Estimated Probability of Nursing Home Placement based on the Presence of Neuropsychiatric Symptoms
FIGURE 3. Estimated Probability of Death based on Presence of Symptoms
Am J Geriatr Psychiatry 15:4, April 2007
321
Neuropsychiatric Subgroups in AD
TABLE 3.
Characteristics of the Subgroups for Four-Cluster Solution Based on the Severity and Frequency of Neuropsychiatric Symptoms
Mean age (SD) Mean MMSE score Mean CIRS Mean EPS score Dependence score Total NPI score Delusions Hallucinations Agitation Depression Anxiety Elation Apathy Disinhibition Irritability Aberrant motor behavior
Cluster 1 (N ⴝ 42)
Cluster 2 (N ⴝ 39)
Cluster 3 (N ⴝ 27)
Cluster 4 (N ⴝ 14)
Fa
p
75.60 (9.44) 17.67 (7.13) 19.50 (3.37) 2.50 (2.18) 2.76 (0.98) 4.50 (4.75) 0.07 (0.26) 0.05 (0.22) 0.19 (0.40) 0.26 (0.45) 0.17 (0.38) 0.07 (0.26) 0.14 (0.35) 0.14 (0.35) 0.13 (0.34) 0.21 (0.42)
75.69 (8.34) 14.90 (8.38) 22.18 (3.83) 3.41 (4.54) 3.51 (1.12) 25.74 (7.49) 0.31 (0.47) 0.21 (0.41) 0.67 (0.48) 0.72 (0.46) 0.64 (0.49) 0.21 (0.41) 0.80 (0.41) 0.59 (0.50) 0.49 (0.51) 0.54 (0.51)
78.22 (7.98) 16.59 (6.82) 20.74 (3.45) 3.78 (3.26) 3.20 (1.18) 11.78 (7.23) 0.26 (0.45) 0.19 (0.40) 0.19 (0.40) 0.22 (0.42) 0.33 (0.48) 0.00 (0.00) 0.93 (0.27) 0.11 (0.32) 0.25 (0.44) 0.15 (0.36)
77.86 (9.11) 13.00 (7.46) 20.00 (4.11) 3.50 (2.50) 3.64 (1.01) 45.93 (9.68) 0.79 (0.43) 0.57 (0.51) 0.86 (0.36) 0.64 (0.50) 0.71 (0.47) 0.29 (0.47) 0.79 (0.43) 0.36 (0.50) 0.92 (0.28) 1.00 (0.00)
0.92 1.96 3.90 1.17 4.21 153.21 11.48 7.23 16.44 10.21 10.00 3.78 34.07 10.23 13.93 17.44
0.44 0.13 0.01 0.33 0.007 0.001 0.001 0.001 0.001 0.001 0.001 0.01 0.001 0.001 0.001 0.001
Fa ⫺ df ⫽ 3. SD: standard deviation; total NPI score: total score on the Neuropsychiatric Inventory; MMSE score: score on the Mini-Mental State Examination; CIRS score: score on the Cumulative Illness Rating Scale; EPS score: score on the Modified Columbia University Parkinson’s Disease Rating Scale; Dependence score: score on the Dependence Scale.
FIGURE 4.
Four-Cluster Means of Neuropsychiatric Symptoms based on the Severity and Frequency of Symptoms
An advantage of the four-cluster solution over the two-cluster model in this portion of the analyses was that it was able to provide finer differentiation. In other words, the four-cluster solution provided information on the roles of the “affective symptoms”
322
group (cluster 2) and the “predominantly apathetic” group (cluster 3) in addition to the low and high symptom groups seen with the two-cluster solutions. Thus, a decision was made to omit the two-cluster solution from further discussion.
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al. Nursing Home Placement. In Table 4, the number of individuals placed in nursing home after a threeyear period and the average time to nursing home are presented for each cluster. Survival curve for the four-cluster solution is presented in Figure 5. In the model, the covariate presence of parkinsonism was included. Inclusion of severity of comorbid medical conditions, marital status, and severity of functional impairment did not produce appreciable difference to the model. Results from the Cox proportional hazard model showed that the overall model, which included the four clusters, along with the presence of parkinsonism, was a significant predictor of nursing home placement (F ⫽18.936; df⫽ 1, 107, p ⬍0.001). In addition, the presence of parkinsonism (2 ⫽ 17.885; df ⫽1, p ⬍0.001) was a significant predictor. Similarly, the differential outcome of nursing home placement as predicted by the fourcluster solution was significant (Tarone-Ware test; 2 ⫽7.965; df⫽3, p ⬍0.05). Pairwise comparisons did not yield significant results (see Table 2). Survival. Table 4 shows the number of individuals in each subgroup who have died and the mean times to death. The survival curves are presented in Figure 6. The overall Cox proportional hazard model with the covariate of presence of parkinsonism was again significant (F ⫽29.151; df⫽1, 120, p ⬍0.001) in predicting mortality over a three-year period. The presence of parkinsonism was a significant covariate (2 ⫽23.787; df⫽1, p ⬍0.001). Moreover, cluster membership was a significant predictor of death (Tarone-Ware test; 2 ⫽19.337; df⫽3, p ⬍0 .001), indicating that the “highly symptomatic” group was
TABLE 4.
9.4% times more likely to die over a three-year period than the “minimally symptomatic” group (see Table 2).
CONCLUSIONS Overall, the current findings appear to provide evidence that AD patients with neuropsychiatric symptoms can be meaningfully grouped based on the presence of symptoms as well as the severity and frequency of symptoms. A decision was made to present the findings on presence and severity of symptoms in a parallel fashion rather than attempt to aggregate the findings because it could potentially lead to a loss of critical information. As mentioned previously, a precedent for such an approach had been set in past studies.29 The first major finding of the study was that based on the presence of neuropsychiatric symptoms, patients with AD can be categorized into three distinct neuropsychiatric profiles: minimally symptomatic, highly symptomatic, and affective/apathetic. Using the same methodology, categorization of patients with AD based on the severity and frequency of symptoms yielded two possible ways of grouping. With the first possibility, patients can be parsimoniously identified as minimally or highly symptomatic. However, with this strategy, limited information can be gained beyond the general impact of neuropsychiatric symptoms on the individual. With the second possible way of grouping, on the other hand, a finer differentiation can be achieved because four
Clusters Based on the Severity and Frequency of Neuropsychiatric Symptoms as Predictor of Nursing Home Placement and Mortality Over a Three-Year Period Nursing Home Placement
Two-cluster Cluster 1 Cluster 2 Four-cluster Cluster 1 Cluster 2 Cluster 3 Cluster 4
Mortality
Percent
Mean No. of Days to Death
SD
14 20
21 38
1,020.59 931.36
196.52 261.95
5 12 10 8
12 31 37 57
1,054.73 968.69 957.63 820.50
150.11 234.43 246.27 307.63
Percent
Mean No. of Days to Nursing Home
SD
No.
Died
22 26
37 53
862.27 736.39
367.96 430.36
68 53
11 19 12 7
31 54 52 50
932.72 754.71 711.13 690.57
324.14 416.91 421.98 475.49
41 39 27 14
No.
In Nursing Home
59 49 36 35 23 14
SD: standard deviation.
Am J Geriatr Psychiatry 15:4, April 2007
323
Neuropsychiatric Subgroups in AD
FIGURE 5. Estimated Probability of Nursing Home Placement based on the Severity and Frequency of Symptoms
FIGURE 6.
Estimated Probability of Death based on the Severity and Frequency of Neuropsychiatric Symptoms (four-cluster solution)
distinct symptom profiles were identified: minimally symptomatic, affective symptoms, predominantly apathetic, and highly symptomatic with psychotic features. In reviewing the various profiles that emerged based on the presence and severity of symptoms, it
324
was of note that a “minimally symptomatic” subgroup was found consistently across the different cluster solutions. This finding was reflective of the clinical picture in that there is a segment of patients with AD who have been noted to be relatively unaffected by neuropsychiatric disturbances.1 In addi-
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al. tion, being minimally symptomatic did not appear to be a function of severity of cognitive impairments in this study, because cognitive functioning was comparable across subgroups. A second notable subgroup we identified was the “predominantly apathetic” group. The emergence of apathy as a distinct neuropsychiatric profile was in support of past reports.30,31 Similarly, the present finding on the “affective symptoms” group was consistent with the literature. Specifically, previous attempts to classify neuropsychiatric symptoms had found the “affective factor” to be one of the most consistent findings across different classification methods.9,11–13 In addition to the primary objective of identifying neuropsychiatric symptoms as distinct symptom profiles, the aim also was to test the predictive validity of subgroups on outcome measures. The present findings indicated that the subgroups, regardless of whether the groups were based on the presence of symptoms or on the severity and frequency of symptoms, were able to predict nursing home placement and death over a three-year period. In particular, regardless of how the grouping was determined, individuals with a high level of symptomatology were at high risk for both nursing home placement and for death. In contrast, individuals with minimal symptoms consistently were at a lower risk for nursing home placement and death. Thus, the high symptom profile was a risk factor for both nursing home placement and survival. The finding on the relationship between neuropsychiatric symptoms and survival was consistent with other reports in the literature.32 Of note, a potential explanation for the increased risk of mortality associated with high symptom profile is that in recent studies, there were indications that certain atypical antipsychotic medications used for the treatment of psychotic symptoms in dementia may increase the risk of mortality.33 In addition, there have been previous reports of an increased risk of cerebrovascular incident with risperidone.34 Therefore, it is conceivable that an increase risk of mortality observed in the study for individuals with high symptom profile could partially be attributed to the use of antipsychotic drugs in these patients. In future studies, it would be of importance to consider drug use when examining the relationships between neuropsychiatric symptoms and mortality in AD.
Am J Geriatr Psychiatry 15:4, April 2007
Another significant finding in the study was the role of apathy. In terms of its impact on nursing home placement, the predominantly apathetic group fell between the low and the high symptom profiles. This made intuitive sense in that the caregivers of the patients with AD endorsing apathy would find their apathetic behavior more tolerable and easy to care for than the high symptomatology. Therefore, there would be less of a need for placing the patient in a long-term care setting. An intriguing result, however, was that although caregivers may find apathy more tolerable, the presence of apathy appeared just as detrimental as the high symptom profile when it came to predicting mortality. In the three-cluster solution, the affective/apathetic and the highly symptomatic group had similarly low rates of survival. In the four-cluster solution, the predominantly apathetic group had a similar rate of survival as the predominantly affective group. Hence, it suggested that although the presence of apathy in patients with AD appears less destructive outwardly, it still has a detrimental impact on the patient’s survival and should not be overlooked as a target for treatment. Overall, it appeared that neuropsychiatric symptoms can be categorized based on the presence and severity of symptoms and that such grouping has predictive validity. In the present study, the threecluster solution produced by the presence of symptoms and the four-cluster solution based on the severity of symptoms yielded the most information. Specifically, it provided support for the emergence of apathy and affective subgroups as distinct neuropsychiatric profiles with differential outcomes in addition to the low and high symptom profiles. Moreover, it appeared that although the four-cluster solution based on the severity and frequency provided the “finest resolution,” the study indicated that the presence of symptoms also has predictive validity. Study’s Limitations There were a number of limitations in the study. First, it was possible that the presence and the severity of neuropsychiatric symptoms changed as the dementia progressed. Given that the symptoms were assessed at the onset of the study, the potential confounding influence of changes in neuropsychiatric symptoms over the three-year period could not be
325
Neuropsychiatric Subgroups in AD evaluated. Second, the NPI, used to assess neuropsychiatric symptoms in the study, was based on informant reports. Hence, it was conceivable that an informant bias in the reporting of the symptoms might have existed. However, when assessing the presence of symptoms in patients with AD with significant cognitive deficits, the use of an informant may be unavoidable. Third, although attempts were made to control for important covariates, it was not possible to control for all the potential covariates. For example, it would have been ideal to control for covariates such as caregiver burden and psychotropic drug use. Thus, it is conceivable that the effects of the covariates contributed to the findings. Fourth, the response rate for the study was low in that only 18% of the eligible sample responded to the request to participate in the study. Hence, generalizability of the study’s findings may be limited to populations similar to sample characteristics. Lastly, although the initial sample size of 122 patients with AD was respectable, the sample size became smaller as subgroups were identified. This was especially true for the four-cluster solution, which in one of its groups, contained a subsample of only 14 patients. Despite the statistically significant findings in the study even with a small sample size, it would be ideal to replicate these results using a larger sample. Directions for Future Studies Given the promising indications of neuropsychiatric symptom profiles having differential disease outcomes, further explorations are needed to test the extent of this predictive validity. For instance, it would be a useful endeavor to test the predictive validity in other outcome measures such as functional impairment, cognitive decline, and quality of
life. Along the same line, future studies could potentially test the usefulness of the methodology used for classifying neuropsychiatric disturbances in AD to other dementias such as dementia with Lewy bodies. Such application of the method to other dementias may provide an empiric validation for the clinical use of neuropsychiatric symptoms in differential diagnosis of dementias. Lastly, and perhaps most importantly, if the findings of neuropsychiatric symptom profiles having differential disease outcomes prove to be consistent across other outcome measures and dementias, an exploration into the potential neurobiologic underpinnings of the profiles appears warranted. For example, it is possible that the presence of certain underlying genetic characteristics or an involvement of particular brain structures in the disease process predispose an individual at a greater risk for a certain neuropsychiatric profile.1 By exploring such potentials, the field could move a step closer toward the development of more effective and targeted treatment options for neuropsychiatric symptoms and thereby may be able to alter the outcomes of patients with AD. Saw-Myo Tun’s work was supported, in part, by the Blue Cross Blue Shield of Michigan grant 885.SAP. Dr. Murman’s work was supported, in part, by the National Institute of Aging grant # K08-AG00864. Portions of this work have been accepted for presentation at the 58th Annual Scientific Meeting of the Gerontological Society of America, Orlando, FL, November 18 –22, 2005. This work was completed at Michigan State University College of Medicine, East Lansing, MI. Dr. Murman is on the speaker bureau for Pfizer, Inc. In addition, he receives research funding from Alzheimer’s Association.
References 1. Cummings JL: The Neuropsychiatry of Alzheimer’s Disease and Related Dementias. London, Martin Dunitz, 2003 2. Lyketsos CG, Lopez O, Jones B, et al: Prevalence of neuropsychiatric symptoms in dementia and mild cognitive impairment. JAMA 2003; 288:1475–1483 3. Murman DL, Chen Q, Powell MC, et al: The incremental direct costs associated with behavioral symptoms in AD. Neurology 2002; 59:1721–1729 4. Norton LE, Malloy PE, Salloway S: The impact of behavioral symptoms on activities of daily living in patients with dementia. Am J Geriatr Psychiatry 2001; 9:41–48 5. Kopetz S, Steele CD, Brandt J, et al: Characteristics and outcomes
326
of dementia residents in an assisted living facility. Int J Geriatr Psychiatry 2000; 15:586 –593 6. Lyketsos CG, Steele C, Baker L, et al: Major and minor depression in Alzheimer’s disease: prevalence and impact. J Neuropsychiatry Clin Neurosci 1997; 9:556 –561 7. Teri L: Behavior and caregiver burden: behavioral problems in patients with Alzheimer disease and its association with caregiver distress. Alzheimer Dis Assoc Disord 1997; 11:S35–S38 8. Payne JL, Sheppard JME, Steinberg M, et al: Incidence, prevalence, and outcomes of depression in residents of a long-term care facility with dementia. Int J Geriatr Psychiatry 2002; 17:247– 253
Am J Geriatr Psychiatry 15:4, April 2007
Tun et al. 9. Frisoni GB, Rozzini L, Gozzetti A, et al: Behavioral syndromes in Alzheimer’s disease: description and correlates. Dement Geriatr Cogn Disord 1999; 10:130 –138 10. Lyketsos CG, Sheppard JME, Steinberg M, et al: Neuropsychiatric disturbance in Alzheimer’s disease clusters into three groups: the Cache County study. Int J Geriatr Psychiatry 2001; 16:1043–1053 11. Aalten P, de Vugt ME, Lousberg R, et al: Behavioral problems in dementia: a factor analysis of the Neuropsychiatric Inventory. Dement Geriatr Cogn Disord 2003; 15:99 –105 12. Hope T, Keene J, Fairburn C, et al: Behavior changes in dementia 2: are there behavioural syndromes? Int J Geriatr Psychiatry 1997; 12:1074 –1078 13. McShane R: What are the syndromes of behavioral and psychological symptoms of dementia? Int Psychogeriatr 2000; 12:147– 153 14. von Eye A, Bergman LR: Research strategies in developmental psychopathology: dimensional identity and the person-oriented approach. Dev Psychopathol 2003; 15:553–580 15. Lyketsos CG, Steinberg M, Tschanz JAT, et al: Mental and behavioral disturbances in dementia: findings from the Cache County Study on memory in aging. Am J Psychiatry 2000; 157:708 –714 16. Lyketsos CG, Breitner JCS, Rabins PV: An evidence-based proposal for the classification of neuropsychiatric disturbance in Alzheimer’s disease. Int J Geriatr Psychiatry 2001; 16:1037–1042 17. Meyers B: The evolving typology of neuropsychiatric complications of Alzheimer’s disease: the use of latent trait analysis. Int J Geriatr Psychiatry 2001; 16:1030 –1032 18. von Eye A, Mun EY, Indurkhya A: Typifying developmental trajectories: a decision making perspective. Psychol Sci 2004; 46: 65–98 19. Murman DL, Kuo SB, Powell MC, et al: The impact of parkinsonism on costs of care in patients with AD and dementia with Lewy bodies. Neurology 2003; 61:944 –949 20. Cummings JL, Mega M, Gray K, et al: The Neuropsychiatric Inventory: comprehensive assessment of psychopathology in dementia. Neurology 1994; 44:2308 –2314 21. Cummings JL: The Neuropsychiatric Inventory: assessing psychopathology in dementia patients. Neurology 1997; 48:S10 –S16
Am J Geriatr Psychiatry 15:4, April 2007
22. Schneider LS: Assessing outcomes in Alzheimer disease. Alzheimer Dis Assoc Disord 2001; 15:S8 –S18 23. Folstein MF, Folstein SE, McHugh PR: ‘Mini-Mental State’: a practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975; 12:189 –198 24. Stern Y, Tang M, Albert MS, et al: Predicting time to nursing home care and death in individuals with Alzheimer’s disease. JAMA 1997; 10:806 –812 25. Linn BS, Linn MW, Gurel L: Cumulative Illness Rating Scale. J Am Geriatr Soc 1968; 16:622–626 26. Miller MD, Paradis CF, Houck PR, et al: Rating chronic medical illness in burden in geropsychiatric practice and research: application of Cumulative Illness Rating Scale. Psychiatry Res 1992; 41:237–248 27. Richards M, Marder K, Bell K, et al: Interrater reliability of extrapyramidal signs in a group assessed for dementia. Arch Neurol 1991; 48:1147–1149 28. Everitt BS, Landau S, Leese M: Cluster Analysis, 4th ed. London, Arnold, 2001 29. Tubman JG, Vicary JR, von Eye A, et al: Qualitative changes in relationships between substance use and adjustment during adolescence. J Subst Abuse 1991; 3:405–414 30. Boyle PA, Mallowy PF: Treating apathy in Alzheimer’s disease. Dement Geriatr Cogn Disord 2003; 17:91–99 31. Boyle PA, Mallowy PF, Salloway S, et al: Executive function and apathy predict functional impairment in Alzheimer’s disease. Am J Geriatr Psychiatry 2003; 11:214 –221 32. Ganguli M, Dodge H, Mulsant BH: Rates and predictors of mortality in an aging, rural, community-based cohort. Arch Gen Psychiatry 2002; 59:1046 –1052 33. Schneider LS, Dagerman KS, Insel P: Risk of death with atypical antipsychotic drug treatment for dementia: meta-analysis of randomized placebo-controlled trial. JAMA 2005; 294:1934 – 1943 34. Brodaty H, Ames D, Snowdon J, et al: A randomized placebocontrolled trial for the treatment of aggression, agitation, and psychosis of dementia. J Clin Psychiatry 2003; 64:134 –143
327