Research in Gerontological Nursing

Secondary Analysis: Multilevel Modeling 

Behavioral Symptoms of Dementia: A Dyadic Effect of Caregivers’ Stress Process?

Judy L.M. Campbell, PhD, ARNP; Meredeth A. Rowe, PhD, RN; Michael Marsiske, PhD

Abstract

This study used multilevel modeling to evaluate a newly revised model in which dementia caregivers’ stress process variables—perceived stress and emotional-behavioral responses—were posited as predictors of behavioral symptoms of dementia (BSD) within community-based dyads. Secondary analyses were conducted on data from a primary two-group (experimental and control) trial, in which experimental participants received a home monitoring system for managing nighttime activity in individuals with dementia. Models indicated that caregivers’ trajectories did not differ significantly between groups over time; however, the time-by-group interaction of BSD approached significance. Because BSD were not targeted, this indicated that the system may have indirectly lowered BSD. In addition, caregivers’ perceived stress and emotional-behavioral responses predicted BSD, on average (across all occasions) and from occasion to occasion, with higher levels associated with worse BSD. These limited results provide support for further research to investigate the nature of these relationships.

Abstract

This study used multilevel modeling to evaluate a newly revised model in which dementia caregivers’ stress process variables—perceived stress and emotional-behavioral responses—were posited as predictors of behavioral symptoms of dementia (BSD) within community-based dyads. Secondary analyses were conducted on data from a primary two-group (experimental and control) trial, in which experimental participants received a home monitoring system for managing nighttime activity in individuals with dementia. Models indicated that caregivers’ trajectories did not differ significantly between groups over time; however, the time-by-group interaction of BSD approached significance. Because BSD were not targeted, this indicated that the system may have indirectly lowered BSD. In addition, caregivers’ perceived stress and emotional-behavioral responses predicted BSD, on average (across all occasions) and from occasion to occasion, with higher levels associated with worse BSD. These limited results provide support for further research to investigate the nature of these relationships.

As the population continues to live longer, the incidence of dementia increases, with more than 13 million individuals expected to have dementia by the year 2050 (Hebert, Scherr, Bienias, Bennett, & Evans, 2003). The majority of individuals with dementia reside in the community (Czaja, Eisdorfer, & Schulz, 2000), with the assistance of informal caregivers (Riggs, 2001). Informal caregiving for individuals with dementia is considered uniquely challenging compared with other caregiving situations, with greater stress and more problems associated with care duties (Ory, Hoffman, Yee, Tennstedt, & Schulz, 1999).

Disruptive or disturbing behavioral symptoms of dementia (BSD) are understandably perceived as a problem, often representing the most difficult problem for those providing care; thus, in research, BSD have generally been conceptualized as an important predictor of stress for informal caregivers (Schulz & Martire, 2004). However, although the frequency and severity of BSD may certainly contribute to caregivers’ perception of these as a problem or stressor, caregivers’ own characteristics may also influence these perceptions. In addition, the relationship between BSD and caregivers’ stress process may be bi-directional; that is, caregivers’ stress may be as likely to influence BSD as the symptoms in individuals with dementia are to cause caregivers’ stress (Sink, Covinsky, Barnes, Newcomer, & Yaffe, 2006). In systematic reviews, perceived stress and emotional or psychological responses have been correlated with BSD. However, reviewers have also cited scarce longitudinal evidence and lack of rigor with a related inability to draw conclusions, particularly regarding the temporal or directional nature of these relationships (Ballard, Lowery, Powell, O’Brien, & James, 2000; Black & Almeida, 2004; Teri, 1997).

Because caregivers in community-based dementia care dyads play a predominant role in managing environmental stimuli and providing for needs of individuals with dementia, the intense role-related stress may impede their ability to provide the optimal caregiving environment for which they strive. For example, experts recommend that caregivers adjust their own approaches, affect, and demeanor to positively enhance the caregiving environment, meet the needs of individuals with dementia, and reduce BSD (Cohen-Mansfield, 2000; Hall, 1994; Kolanowski & Whall, 2000; Mittelman, Roth, Haley, & Zarit, 2004; Quayhagen et al., 2000; Smith, Hall, Gerdner, & Buckwalter, 2006). However, caregivers with excessive stress and negative emotional-behavioral responses may not have adequate psychoemotional resources to continually approach caregiving in the recommended manner. Negatively altered interactions or affect may be misinterpreted by individuals with dementia, and BSD may increase (Sink et al., 2006). Caregivers’ influence within dementia care dyads has been previously implicated when caregivers’ characteristics or caregiver-directed interventions have been linked to outcomes in individuals with dementia (de Vugt et al., 2004; McClendon, Smyth, & Neundorfer, 2004; Teri, McCurry, Logsdon, & Gibbons, 2005).

In this study, it is proposed that caregivers are influential within the dementia care dyad relationship. Individuals with dementia may react to cues from caregivers due to their cognitively limited perceptions. Thus, BSD may be conceptualized, at least in part, as a function of the caregivers’ stress process. It is proposed that caregivers’ perceived stress and emotional-behavioral responses to stress may help explain levels of BSD in care recipients.

Secondary Analysis of BSD as a Potential Consequence of Caregivers’ Stress Process

A monitoring system recently developed to assist caregivers in the management of nighttime awakenings in individuals with dementia was tested in a two-group experimental trial (referred to in this article as the primary study), using a repeated-measures design to collect information on caregiver sleep, burden, worry, depression, and mood within caregiving dyads longitudinally during a 1-year period (N = 53). The monitoring system was a newly developed technology that uses components similar to those of home security systems, a bed sensor, and bedside alerts for the caregivers to track nighttime activities in individuals with dementia. The overall aim of the primary study was to test the efficacy of the monitoring system on caregivers’ outcomes and on injuries or unattended exits in individuals with dementia. The system had no mechanism to directly influence BSD (Rowe, Lane, & Phipps, 2007; Rowe et al., 2009).

The study reported in this article is a secondary analysis of the data from the primary study to examine whether a relationship existed between caregivers’ stress process variables—perceived stress and emotional-behavioral responses—and simultaneously collected data on BSD. In addition, the longitudinal nature of the primary study allowed analysis of trajectories for change over time within, across, or between dyads, and analysis of experimental and control groups for a time-by-group interaction, for both caregivers’ perceived stress variables and BSD.

Theoretical Foundations

In previous work, the Stress-Health model (Schulz et al., 2004), regarding the effect of caregivers’ stress process on their own health, was adopted and extended to reflect the influence of caregivers within dementia care dyads. Details regarding the extension of the Stress-Health model have been previously reported (Campbell, 2009). Briefly, in the extended Consequences of Dementia Caregivers’ Stress Process model (Figure 1), the consequences of caregivers’ stress are viewed more broadly than the previously described effect on caregivers’ health alone and may include personal influences for the caregivers as well as dyadic influences on the individuals with dementia. In this model, BSD were conceptualized as an exemplar of a dyadic consequence of the caregivers’ stress process. To begin the process of testing the revised model, key concepts were identified. The caregivers’ perceived stress was considered a major predictor of dyadic effects on individuals with dementia, and it was thought that emotional-behavioral responses also had a role explaining the influence of caregivers’ stress within the dyad. These caregiver stress process variables were included as predictors of BSD in a reduced model, Dyadic Effect of Dementia Caregivers’ Stress: Behavioral Symptoms in Individuals with Dementia (Figure 2), for testing in this study.

The full Consequences of Dementia Caregivers’ Stress Process model. Adapted with permission from the Stress-Health model (Schulz & Martire, 2004).

Figure 1. The full Consequences of Dementia Caregivers’ Stress Process model. Adapted with permission from the Stress-Health model (Schulz & Martire, 2004).

The reduced Dyadic Effect of Dementia Caregivers’ Stress Process: Behavioral Symptoms in Individuals with Dementia model with key variables and relationships from the full model.

Figure 2. The reduced Dyadic Effect of Dementia Caregivers’ Stress Process: Behavioral Symptoms in Individuals with Dementia model with key variables and relationships from the full model.

For purposes of this secondary analysis, behavioral symptoms were conceptualized as a communication in response to stimulation from the environment, a situation, or the care relationship, specifically, the caregivers’ verbal and nonverbal communication of their own perceptions of stress. For example, individuals with dementia may be using BSD to try to communicate a need for attention or an intolerance of negative stimuli, including cues from the caregivers. To reflect multiple ways that individuals with dementia may be responding to the environment, including the care dyad relationship, a comprehensive measure of BSD was used. All noncognitive or neuropsychiatric behaviors reflecting perceptual, mood, behavioral, and vegetative responses were collectively considered BSD (Donaldson, Tarrier, & Burns, 1997, 1998; Finkel, Costa e Silva, Cohen, Miller, & Sartorius, 1996; Harwood, Barker, Ownby, & Duara, 2000; Smith & Buckwalter, 2005). The wide-ranging domains of the Neuropsychiatric Inventory Questionnaire (NPI-Q) (Kaufer et al., 2000) reflected this conceptualization of BSD, including an aggregate of symptom domains such as delusions/hallucinations, agitation/aggression, dysphoria/depression, anxiety, euphoria/elation, apathy/indifference, disinhibition, irritability/lability, aberrant motor behaviors, nighttime behavioral disturbances, and appetite/eating disturbances (Kaufer et al., 2000).

Purpose and Aims

The overall purpose of this secondary analysis was to test the proposed reduced model, Dyadic Effect of Dementia Caregivers’ Stress: Behavioral Symptoms in Individuals with Dementia (Figure 2). Patterns of change over time in the model’s variables and relationships among the variables were examined using multilevel modeling. Two specific aims and hypotheses were included.

Aim 1

The first aim sought to describe patterns of change over time in model variables, including perceived stress, emotional-behavioral responses, and BSD, with consideration of modeled change over time within dyads (Level 1) and across dyads (Level 2), in general and according to experimental group, investigating for a time-by-group interaction.

Aim 2

The second aim was to investigate the relationships between caregivers’ perceived stress, caregivers’ emotional-behavioral responses, and BSD over time, both within and across dyads. More specifically, the following research questions were investigated:

  • Question 1: Are average BSD (across all data points; not time varying) higher in dyads when levels of caregivers’ stress process variables are higher?
  • Question 2: On occasions (data points) when the population’s estimated time-varying levels of caregivers’ current stress process variables are, on average, higher than usual, are care recipients’ time-varying BSD higher, on average, as well?
  • Question 3: Given an occasion-to-occasion relationship in the population between caregivers’ current stress process variables and BSD in care recipients, are there differences among dyads in these relationships?

Hypotheses

It was hypothesized that both perceived stress and emotional-behavioral responses in caregivers would directly predict BSD, on average across all occasions and in current occasion-to-occasion analyses, and that these relationships would hold across dyads. It was further hypothesized that the system might moderate caregivers’ stress, and experimental dementia care dyads could have improvement (or less deterioration) over time on measures of caregivers’ stress process variables compared with controls. Because the system did not have any function that would directly influence BSD, it was hypothesized that system-related reductions in caregivers’ stress process variables could indirectly result in differences in BSD between groups; thus, such group differences might provide some validation for the directional hypothesis in the model (Figure 2).

Method

This secondary analysis used data collected during the 12-month primary study: at baseline and at months 2, 3, 4, 5, 6, 8, 10, and 12. Measures reflecting caregivers’ perceived stress/burden, depressive symptoms, and mood/affect, as well as care recipients’ BSD, were gathered simultaneously at each data point in the primary study (Rowe et al., 2006, 2010).

Primary Study Method

The primary study method has been reported elsewhere and is summarized in this article (Rowe et al., 2006, 2010). A convenience purposive sample of community-dwelling dementia care dyads was recruited from three areas in northern and central Florida, chosen to augment over-sampling of minorities. Dyads were recruited through advertisements, support groups, and dementia-appropriate clinics or organizations. Interested caregivers made contact with researchers. To be eligible for the study, care recipients were required to have a medical diagnosis of dementia and an English-speaking primary caregiver responsible for care, particularly at night. The caregivers needed to have concerns about nighttime activity of the individuals with dementia and could not have sleep conditions, medications, or cognitive or functional limitations that would limit response to system alerts (Rowe et al., 2006, 2010).

Data points were not exactly equidistant for study participants. Progression to the Month 2 data point for those in the experimental group followed a brief reliability phase; this was necessary to verify that sensors were manipulated appropriately to suit the dyad, the system was functioning properly, and caregivers had mastered operation of the system. Due to scheduling issues, subsequent months were not always exactly 1 month apart (Rowe et al., 2006).

Approval was obtained from the university’s Institutional Review Board, and caregivers gave informed consent for themselves and the individuals with dementia in their care; the individuals with dementia indicated their assent for participation. Caregivers were prescreened by telephone for preliminary inclusion criteria, and those who qualified were screened for cognition problems during the initial home visit using the Mini-Mental State Examination (MMSE) (Folstein, Folstein, & McHugh, 1975). One caregiver scored too low on the MMSE and did not enter the study.

To reduce participant burden, instruments were entered into a user-friendly laptop interface by a technical consultant, allowing caregivers to quickly complete the instruments, and researchers carried back-up paper copies in case a computer failed. The MMSE was completed for individuals with dementia early in the study to affirm the diagnosis of dementia; this was the only instrument that directly involved the individuals with dementia and was deferred if administration seemed inappropriate given the situation. Datasets never included personal or identifying information, and electronic storage and communications were accomplished through secure servers. Data outliers were checked for accuracy prior to analyses. In addition to Institutional Review Board oversight, Data and Safety Monitoring Board reviews for adverse events were conducted but did not trigger changes (Rowe et al., 2006, 2010).

The first four dyads recruited were used in a preliminary reliability study and were thus automatically recruited into the experimental group, since they already had the system in place. Subsequent experimental (with monitoring system) and control group (without system) assignment was random, allocated by a nonresearch staff member, for 45 of the dyads recruited. Two caregivers indicated willingness to participate only in the control group, and two voluntarily accepted control assignment due to incompatibility of the system with sleeping arrangements. The resulting groups were numerically and demographically similar. The primary reasons for withdrawal from the primary study were medical illness, death, or institutional placement (Rowe et al., 2006, 2010).

Secondary Analysis Method

Early in the primary study, the Institutional Review Board approved changes to allow addition of an instrument to measure BSD. Participants previously enrolled provided consent again, and BSD were assessed at later data points; thus, only 37 participants had baseline BSD data. Approximately 70% of the dyads completed the study. Institutional Review Board approval was obtained after study completion for the secondary analyses. Of the 53 dyads in the primary study, 4 had only baseline data and were excluded from the secondary analysis, which focused on change trajectories over time.

In the secondary analysis’ sample, the majority of caregivers were Caucasian (78%) and women (82%), with most being either wives (43%) or daughters (37%), and were generally well educated (33% were college graduates or had postcollege education). Regarding the individuals with dementia, mean age was 80 and the majority were men (54%); the majority had Alzheimer’s type dementia (79%) with moderate cognitive impairment (MMSE mean score = 14.67). There were no significant differences in demographic data according to experimental or control group.

Instruments

Instruments were available from the primary study that represented the model’s predictor variables related to caregivers’ stress. A short version of the Zarit Burden Interview (Bedard et al., 2001) corresponded conceptually with perceived stress (B.G. Knight, Silverstein, McCallum, & Fox, 2000). Two instruments provided information related to the concept of emotional-behavioral responses: the Center for Epidemiological Studies-Depression (CES-D) (Radloff, 1977) and the Positive and Negative Affect Schedule (PANAS) (Watson, Clark, & Tellegen, 1988).

Zarit Burden Interview. The Zarit Burden Interview (Zarit, Orr, & Zarit, 1985; Zarit, Reever, & Bach-Peterson, 1980) is considered valid and reliable, and the 22-item version is the instrument most often used to measure burden in dementia caregiving research. The short Zarit Burden Interview (Bedard et al., 2001) was validated using 413 caregivers with factor analysis, change scores, and item-total correlations to reduce the items to 12. Caregivers self-report aspects of burden on a scale ranging from 0 (never) to 4 (nearly always), with a potential range of 0 to 48 (Bedard et al., 2001). The Cronbach’s alpha coefficient was 0.89 for the sample in this secondary analysis.

CES-D. On the CES-D (Radloff, 1977), caregivers ranked the frequency of their depressive symptoms using a 4-point scale, with higher scores indicating worse symptoms (maximum = 60). The validity and reliability of the CES-D is widely supported; it has been used in dementia caregivers, and a score of ≥16 has been established as clinically significant (Beekman et al., 1997; Gallicchio, Siddiqi, Langenberg, & Baumgarten, 2002; Hooker, Manoogian-O’Dell, Monahan, Frazier, & Shifren, 2000; R.G. Knight, Williams, McGee, & Olaman, 1997; Radloff, 1977). Four items were originally worded to assess positive affect, with the intent of breaking tendencies toward negative responses (reverse-scored) (Radloff, 1977). This practice was later viewed as a violation of assessing the construct using positive symptoms; only the 16 negative-item factor was validated (Schroevers, Sanderman, van Sonderen, & Ranchor, 2000). In the primary study, positive items were changed to reflect the opposite negative affect, presenting a less confusing instrument. For example, feeling “not as good” replaced “just as good” as others. The Cronbach’s alpha coefficient was 0.92 in this secondary analysis.

PANAS. This scale (Watson et al., 1988) was used for caregivers to rate relatively pure markers of affect (near zero loading on opposite factor) on a 5-point scale ranging from very slightly or not at all to extreme. The PANAS is considered internally consistent and valid (Crawford & Henry, 2004; Denollet & De Vries, 2006). In the primary study, the number of items was reduced to 10 (5 each for positive and negative) to decrease the burden of completing the tool in the sample of time-stressed dementia caregivers (Kelly & Rowe, 2006). Items were chosen for their greater variability in a sample of community-dwelling older adults, and items were compared to the full complement of choices for adequate reliability (Diehl, 2005). In this secondary analysis, only the negative items from the “reduced” PANAS were used to reflect the model’s concept of emotional-behavioral responses, including distressed, scared, irritable, nervous, and jittery. This scale was collected daily for 7 days at each data point and was averaged for a weekly value if at least 3 days were ranked. The Cronbach’s alpha coefficient was 0.96 for the instrument in this secondary analysis sample.

Conceptually, each of these measures provided a unique representation of emotional-behavioral responses: the CES-D ranked mood/affect retrospectively for a month, and the modified PANAS reflected current rankings for 7 days at each data point. Nonetheless, the CES-D and negative portion of the PANAS were moderately correlated (r = 0.57; p = 0.000 at baseline). Thus, although there was conceptual similarity in these measures in that they both represented emotional-behavioral responses, there was also unique contribution from each, both conceptually and statistically. The significant correlation of the measures allowed us to establish a composite score for emotional-behavioral responses by averaging z scores from both measures. This provided a more inclusive, comprehensive measure and reduced the number of variables requiring estimation in examining model relationships.

NPI-Q. The NPI-Q (Kaufer et al., 2000) was chosen to measure BSD (caregivers’ behavior-related distress rankings were available but were not used in this study). It is brief, requiring less time to complete (Forester & Oxman, 2003). The NPI-Q uses the primary caregiver as an informant, allowing a more lengthy assessment window (1-month retrospective) in a natural setting than direct observation provides. The original directed-interview NPI (Cummings, 1997; Cummings et al., 1994) has been established as reliable and valid, including test-retest reliability, interrater reliability, internal consistency, content, and convergent validity. It has been frequently cited in the research literature and is commonly used in clinical trials (Forester & Oxman, 2003; Lange, Hopp, & Kang, 2004). There is consensus that a score of ≥4 reflects clinical significance (Lyketsos, 2007; Lyketsos et al., 2002; Schneider et al., 2001; Steinberg et al., 2004).

Developed as a self-administration version of the NPI for easy caregiver completion, the NPI-Q demonstrated convergent validity with the NPI using 60 care dyads; correlation on behaviors ranged from 0.71 to 0.93 for various domains and 0.91 for total scores. Using simple descriptions derived from the NPI interviews, caregivers endorse whether symptom domains occur. For any domains endorsed, symptoms are ranked for severity on a scale ranging from 1 to 3, with a maximum range of 36. Frequency was omitted to improve brevity on the NPI-Q, since severity is more closely related to distress in the caregiver (Forester & Oxman, 2003; Lange et al., 2004). Although behavior domains on the NPI-Q may be considered independently or within established factors, it was the total score that was used in this study for a comprehensive, inclusive measure of BSD. The Cronbach’s alpha coefficient was 0.82 for the behavioral scale in this secondary analysis sample.

MMSE. The MMSE (Folstein et al., 1975) was used for inclusion of caregivers to rule out problems with cognition and was also used for care recipients. The MMSE was originally established as a measure of cognitive impairment, with higher scores (maximum = 30) reflecting better cognition. Despite some concerns, the MMSE remains widely accepted as a reliable and valid instrument for quick screening of cognitive impairment (Jones & Gallo, 2001). The Cronbach’s alpha coefficient was 0.82 in this secondary analysis sample.

Statistical Analysis

All statistical analyses were performed using SPSS version 14; the alpha was set at 0.05. Univariate, bivariate, and preliminary trend analyses preceded the multilevel analyses.

Missing Data. Due to a computer malfunction in the data collection interface, approximately 1% to 2% of the potential NPI-Q behavior items were considered systematically missing. When three of the items on the NPI-Q were endorsed, some participants were mistakenly directed to the screen that assessed their distress related to behaviors, skipping the option to rate the severity of BSD. These missing data influenced the sum scores for those participants, and because it occurred within the score, it could not be addressed in modeling, as is possible with dropouts or skipped appointments.

Consultations with an expert regarding missing data established that because the rate of missing data was low and the range of possible rankings narrow (1 to 3), complicated imputation techniques were considered inappropriate (J.L. Schafer, personal communication, April 12, 2007). Simple techniques commonly used were also not appropriate. Averaging across dyads at a particular data point did not take into account the nine correctly measured items in the within-person data at that point. In addition, averaging the person’s measures prior to and subsequent to missing data would have diminished the occasion-to-occasion change, and within-person trends over time were of interest in this study. Therefore, this missingness was remedied by calculating a mean of the items available for that person at that data point if at least nine items were scored and multiplying to return the value to scale (Fitzmaurice, Laird, & Ware, 2004; J.L. Schafer, personal communication, April 12, 2007; Schafer & Graham, 2002). This allowed use of the person’s data that was available without skewing results with data imputed from sources in conflict with study aims.

Univariate and Bivariate Statistics. For each model variable, including caregivers’ perceived stress and emotional-behavioral responses, and behavioral symptoms in individuals with dementia, univariate and bivariate statistics were assessed. Histograms were examined for normality. The outcome of random assignment to groups for each model variable was established using t tests. Pearson correlations were used to establish relationships between variables for the study period. Means and standard deviations were calculated on each model variable (caregivers’ perceived stress and emotional-behavioral responses, and behavioral symptoms in individuals with dementia) for the entire sample and according to experimental status, and were compared across all data points using t tests and graphs.

Preliminary Trend Analyses. Multilevel models for change over time must address the underlying change trends, with the simplest trend, linear, serving as the default. Preliminary inspection of individual within-dyad trends for model variables indicated substantial variability during the study period. The variation in direction and magnitude of change warranted within-dyad modeling and suggested that change models should address both linear and quadratic trends. Orthogonal quadratic change functions were thus computed to avoid multicollinearity between the linear and quadratic terms (using the residuals from a regression of the quadratic term on the linear term) to reflect the unique, independent contribution of quadratic change. As a result of the orthogonal computations, all intercepts reported in this article reflect the midpoint of the study.

Multilevel Modeling (MLM). The repeated measures, missing data points, and unequal time distances for data points made MLM a natural fit for these analyses. MLM techniques are versatile, and models can be estimated in a number of ways. Fixed effects represent population estimates, and random effects identify differences among the individuals in the aggregate. In this secondary analysis, measurements over time (Level 1) were nested within dyads (Level 2), with group status (experimental or control) representing an influential dyad-level factor (rather than a third level in the hierarchy). The use of repeated measures in longitudinal change studies may be considered a special case of MLM that can address clustering of data within persons (Cho, 2003), or in this case, within dyads. “Days from baseline” was used to accommodate unequal time distances; however, “month” of the study was used as the repeated term to structure the covariances and for data point comparisons. More information regarding the specific modeling process is outlined in Table 1.

Estimation of Parameters for Multilevel Analyses

Table 1: Estimation of Parameters for Multilevel Analyses

To address Aim 1, models were constructed to estimate population trajectories of change for all time-varying model variables, subsequently comparing average trajectories of the experimental and control groups. For Aim 2, predictor models estimated the within-dyad and population’s across-dyad (average) effect of caregivers’ stress process variables on BSD over time. Caregivers’ predictors were meaned for the average effect over all occasions and were centered for the current effect from occasion to occasion. (Table 2 provides more information on interpretation.) The model-building process for hypothesized relationships in Aim 2 used the accepted change model for BSD from Aim 1 but without consideration of the experimental group variable as a baseline for comparison. In these analyses, it was the relationship between caregivers’ variables and BSD, regardless of treatment, that was of concern in assessing time-varying model relationships.

Interpretation of Caregivers’ Stress Process Predictor Variables in Aim 2 Analyses

Table 2: Interpretation of Caregivers’ Stress Process Predictor Variables in Aim 2 Analyses

Results

Univariate and Bivariate Descriptives of Observed Raw Data

Sample means and standard deviations were calculated for all model variables across all data points (Table 3). The model variables were considered a family of related assessments; thus, the significance of the p value was adjusted using the Bonferroni method (three t tests, 0.05/3 = 0.017) to reflect the study alpha of 0.05. Accordingly, there were no significant differences according to experimental or control group at baseline for model variables. In addition, t tests describing experimental and control group differences at each data point are shown in Table 3, and observed raw scores for each of the model variables are graphed in Figure 3 to visually compare means for experimental and control dyads at each data point. Finally, all available data points for all dyads were aggregated, and a significant experimental group difference in means was found for each of the model variables (see last column in Table 3); however, these aggregate comparisons do not reflect change over time. Bivariate analysis showed that BSD were similarly correlated to perceived stress (r = 0.53) and emotional-behavioral responses (r = 0.48); for each of these caregiver variables, the positive correlation indicated that higher caregivers’ stress process variables were associated with worse BSD.

Findings by Experimental Group for the Sample’s Observed Raw Data Across Study

Table 3: Findings by Experimental Group for the Sample’s Observed Raw Data Across Study

Observed raw scores for each variable compared by occasion and according to experimental group (means and standard error bars).Note. CES-D = Center for Epidemiological Studies-Depression; CON = control; EXP = experimental; NPI-Q = Neuropsychiatric Inventory-Questionnaire; PANAS = Positive and Negative Affect Schedule.

Figure 3. Observed raw scores for each variable compared by occasion and according to experimental group (means and standard error bars).Note. CES-D = Center for Epidemiological Studies-Depression; CON = control; EXP = experimental; NPI-Q = Neuropsychiatric Inventory-Questionnaire; PANAS = Positive and Negative Affect Schedule.

Aim 1: Modeling of Change over Time for Model Variables

Caregivers’ Stress Process Variables. To compare change over time in experimental and control caregivers’ variables, models with experimental group added were evaluated for improvement over previous models. For perceived stress, adding the experimental group did improve the model fit criteria, and modeled perceptions of stress were significantly higher in the control group, worsening over the study period. However, the groups’ differing modeled trajectories (similar to time × group interaction) were not significant. For emotional-behavioral responses, adding the experimental group did not improve the model fit criteria; this meant that comparing the groups’ modeled trajectories over time was inappropriate. Therefore, experimental and control groups’ trajectories did not differ over time for either caregiver variable as hypothesized in Aim 1.

BSD. Modeling BSD over time provided more information and is thus reported in more detail. First, the “unconditional means model,” without consideration of time, indicated that 35% of the unexplained variance was within individuals with dementia (17.29 + 32.27 = 49.56; 17.29/49.56 = 0.35), with the majority of variance across all individuals with dementia. Second, neither linear nor quadratic trends over time described the entire sample adequately; thus, the default (linear) form of change over time was used in modeling. Although the criteria for model fit were improved, linear change over time across dyads was nonsignificant, indicating that the linear trend did not represent all participants well. Finally, adding the experimental group did improve the model fit. The population’s estimated mean for BSD including all data points was 9.084 (at midstudy; p = 0.00); experimental group differences approached significance, with control dyads averaging 3.19 points higher (at midstudy; p = 0.056). The differing change trajectories for experimental versus control dyads as modeled linearly over time also approached significance (p = 0.075), with control dyads worsening, increasing approximately 0.06 per day from baseline (p = 0.082), and with experimental dyads stable (changing −0.003, not significant). Because larger numbers of parameters are more difficult to estimate, assessment of random effects, or within-dyad differences in slopes from population estimates, were not possible; the Level 1 within-dyad effect over time needed to be removed to allow the computer to settle on a solution for the estimates. The remaining random variability for BSD did support the use of predictors in further modeling.

Aim 2: Hypothesized Relationships

Adding both average and current levels of caregivers’ perceived stress and emotional-behavioral responses as predictors improved model fit criteria significantly and reduced unexplained variance, indicating that both caregiver variables were associated with BSD. For the first two research questions, estimation of fixed (population) effects included average (not time-varying) levels of predictors across dyads and current (time-varying from occasion to occasion) levels of predictors within dyads (Table 2). Estimation of random effects (within-dyad differences from population estimates) in the last research question included only current levels of predictors, or occasion-to-occasion change. Modeling these caregiver stress process variables as predictor effects for BSD explained 33% of the previously unexplained variance across participants for perceived stress and 27% for the emotional-behavioral responses. In addition, again when modeled separately, perceived stress and emotional-behavioral responses explained 13% and 6% of the remaining variance within individuals with dementia, respectively. Model-derived estimations of effects for the three research questions associated with Aim 2 are discussed in the following section.

Question 1: Are average BSD (across all data points; not time varying) higher in dyads when levels of caregivers’ stress process variables are higher? Both caregivers’ perceived stress and emotional-behavioral responses had a relationship with BSD in this study. At midstudy, for each point that the caregiver’s average levels of perceived stress varied from the estimated population’s typical, BSD differed by 0.42 (p = 0.00), on average. In addition, when average levels of caregivers’ emotional-behavioral responses differed by 1 point from the estimated population’s typical, BSD varied by 3.57 (p = 0.00), on average. When controlling for these average caregiver effects, perceived stress was substantially instrumental in explaining BSD, such that both the intercept and effects of time were nonsignificant or negligible. On the other hand, when controlling for the caregiver’s average emotional-behavioral effects, even though the effect of time was nonsignificant, the intercept remained significant (10.67, standard error = 0.73, p = 0.00). Thus, controlling for emotional-behavioral effects did not fully explain levels of BSD at midstudy.

Question 2: On occasions (data points) when the population’s estimated time-varying levels of caregivers’ current stress process variables are, on average, higher than usual, are care recipients’ time-varying BSD higher, on average, as well? For each point that the population’s estimated time-varying levels of current caregivers’ perceived stress varied from the typical, levels of BSD changed by 0.33 (p = 0.00), on average. In addition, for each point that the population’s estimated time-varying levels of current caregivers’ emotional-behavioral responses varied from usual, BSD differed by 2.14 (p = 0.00), on average. Thus, both caregivers’ perceived stress and emotional-behavioral responses “traveled together” with changes in BSD, either increasing or decreasing simultaneously. Although the magnitude of the emotional-behavioral effects may seem high compared with that of perceived stress, the emotional-behavioral responses variable was based on z scores with a range constrained between 1 and −1; thus, these values reflect a maximum possible change in BSD.

Question 3: Given an occasion-to-occasion relationship in the population between caregivers’ current stress process variables and BSD in care recipients, are there differences among dyads in these relationships? Models including caregivers’ stress process variables did explain further within-dyad variance in comparison to the baseline model. However, when random (within-dyad) time-varying effects of perceived stress and emotional-behavioral response were investigated, no significant differences were identified. This indicates that these estimated time-varying relationship patterns generally held across dyads.

For Aims 1 and 2, the fixed (population) estimates are shown in Table 4. Random effects (individual dyad differences) are omitted since they did not improve the models significantly. More detail about the modeling process is available from the corresponding author.

Estimated Parameters for Each Variable Modeled Over Time and by Experimental Status, then with Caregivers’ Stress Process Variables as Predictor Effects on BSD Modeled

Table 4: Estimated Parameters for Each Variable Modeled Over Time and by Experimental Status, then with Caregivers’ Stress Process Variables as Predictor Effects on BSD Modeled

Discussion

Support of Proposed Theoretical Model

In general, there is evidence to support the relationships posited in the model between caregivers’ stress process and BSD. Modeling the effect of caregivers’ stress process variables on BSD revealed that BSD increased when caregiver stress process variables were, on average, higher than usual (current levels). In addition, in those dyads characterized by higher caregivers’ stress measurements, BSD were more severe (average levels) at midstudy.

In modeling change trajectories, analyses indicated that BSD worsened over time in the control group, and this effect approached significance. Particularly after the midpoint of the study, BSD were increasingly worse in control dyads, whereas BSD in experimental dyads stabilized across time. Importantly, this occurred even though the modest improvements in caregivers’ stress process variables over time were not significant. The primary study’s intervention had the potential to influence caregivers’ stress process variables but did not have properties to directly influence BSD. The lack of significant group differences in caregivers’ variables is counterintuitive when attempting to reconcile results to the hypotheses in this study; it is unclear how BSD changes could approach significance over time without corresponding significant changes in caregivers’ stress process. It is possible that changes in disease occurred differently in the two groups or that individuals with dementia in experimental dyads received more sleep related to caregivers’ system alerts of their arising and subsequent encouragement to return to bed. However, another possible explanation is that small improvements in caregivers’ stress (not significant) may have substantial impact within the dyad. In general, this evidence provides limited support for the directionality of the relationships in the model, from caregivers’ stress to BSD.

In this study, perceived stress seemed to have a greater impact on BSD than did emotional-behavioral responses, as evidenced by the intercept remaining significant when controlling for emotional-behavioral responses. In the proposed model, level of emotional-behavioral responses was thought to explain why perceived stress would influence BSD. It is not clear why emotional-behavioral responses did not have greater impact in this study.

Findings in Relation to Past Research

As in previous studies, BSD were a substantial issue in the care dyads in both groups in this study, remaining above the accepted level of clinical significance (>4) (Lyketsos, 2007) throughout the study. Caregivers in this study also had consistently high levels of perceived stress on the short Zarit Burden Interview (Bedard et al., 2001; O’Rourke & Tuokko, 2003), and CES-D scores often approached 16, the cut-off point for clinical depression risk (Radloff, 1977). In contrast, the weekly mean of scores on negative affect were <1 on a scale of 0 to 5, possibly because these items were rated on arising before caregiving duties for the day were well under way.

Each of the caregivers’ stress process variables accounted for approximately one third of the variance in BSD that was unexplained in the baseline change model. These findings are in agreement with previous works that have proposed caregiver influence. In a population-based study of more than 5,000 dementia care dyads, caregivers’ depression and burden were among predictors of increased BSD (Sink et al., 2006), and one review highlighted caregivers’ influence on BSD (Dunkin & Anderson-Hanley, 1998). In a prospective study of 96 dementia care dyads, caregivers’ role stress (negative attitude toward the care recipient) predicted worse social behaviors in individuals with dementia, and caregiver factors explained 32% of variance in care recipients’ quality of life (Burgener & Twigg, 2002).

The relationship between the two caregiver variables, perceived stress and emotional-behavioral responses, is embedded within the Stress-Health model, which emerged from previous theory and research and has extensive support from literature reviews (Goode, Haley, Roth, & Ford, 1998; Pearlin, Mullan, Semple, & Skaff, 1990; Pinquart & Sorensen, 2003a, 2003b, 2007; Schulz et al., 2001; Schulz & Martire, 2004; Vitaliano, Zhang, & Scanlan, 2003). Several researchers have presented evidence that may explain why caregivers with a more intense stress process likely care for recipients with increased BSD. For example, previous quality of relationship (i.e., how “communal” or reciprocal) predicted both depressive symptoms in caregivers and frequency of harmful treatment of the care recipient (Williamson & Shaffer, 2001). In addition, caregivers with higher strain and distress were also higher in “expressed emotion,” or criticism toward the care recipients (Tarrier et al., 2002). Longitudinally, caregivers’ high expressed emotion was predictive of increased negative behaviors over time (Vitaliano, Young, Russo, Romano, & Magana-Amato, 1993). A small body of research proposes that caregivers’ coping may have influence on individuals with dementia. Included in this body of research are studies regarding:

  • Caregivers’ management strategies and their relationship with BSD (de Vugt et al., 2004).
  • Caregivers’ cognitive decline and related increased BSD, possibly linked to the inability of caregivers to provide an optimum care environment (de Vugt et al., 2006).
  • Caregivers’ ineffective coping and related low survival rates in individuals with dementia (McClendon et al., 2004).
This evidence indirectly connects the intensity of the stress process to BSD.

Strengths, Limitations, and Directions for Future Research

The use of MLM allowed change within dyads to be considered in analyses, with the potential of more precise estimates than is seen in traditional analyses. In MLM, dyads may “borrow strength” from the population when there is higher occasion-to-occasion variability within the dyad’s measurements than exists in the sample. In addition, missing data points are accounted for efficiently. This emphasis on precision (over bias toward the sample) is the primary motivation to use MLM for model estimation (J.L. Schafer, personal communication, April 12, 2007; Singer & Willett, 2003).

On the other hand, the secondary nature of this study meant a lack of control over data. Better statistical support of the reliability of the modified PANAS measure would improve confidence in that data from the primary study. Although statistical analyses that established adequate reliability of the modified PANAS were reported in personal communication, the statistical data themselves were not available for reasons beyond the authors’ control. In addition, covariates were limited to those available in the primary study. An example of a covariate that was not available in the primary study data is that of the quality of the pre-care relationship. Covariates such as gender, relationship, and race were equal in the study’s groups; however, direct modeling of covariates could improve future estimates. Finally, although measures collected in the primary study can be considered appropriate for stress process variables, it is not known whether they are the best measures. These issues may need to be clarified in future research.

Related to the outcome measure, this study used a comprehensive measure of BSD, which included behavioral, psychological, vegetative, and other symptoms. It would be useful for future analyses to include subscales of the behavioral domains. This might allow more clarity regarding the types of responses in individuals with dementia affected by their caregivers’ stress process. In addition, related to the predictors, the combined emotional-behavioral responses measured included current data from the PANAS recorded in the mornings during a weeklong data point. Although combining this with CES-D data was considered a more comprehensive, relevant measure, the retrospective CES-D data alone might have been more similar in nature to the other variables’ measures. Future research may need to compare various measures.

Some researchers note concern about using caregivers’ proxy reports for BSD. Others acknowledge that using caregivers as raters may be most logical, noting that caregivers are intimately familiar with BSD exhibited and can report a wider time frame than direct observation can capture, and that caregiver reports have been correlated to direct observation (Cotter, Burgio, Roth, Gerstle, & Richardson, 2008; Davis, Buckwalter, & Burgio, 1997). The Cohen-Mansfield Agitation Inventory, a common caregiver-report measure of BSD, was significantly correlated to formal caregivers’ direct observations of BSD (Cohen-Mansfield & Libin, 2004). These findings support the use of caregivers’ reports in community-based dyads. Nonetheless, future research with more objective measures should be considered. For example, videotaped observation of BSD or actigraphy to measure wandering or repetitive movements could enhance measurement of BSD. Likewise, heart rate monitoring might substantiate subjective reports of perceived stress. Although objective measures may strengthen the conclusions of the study, the use of MLM accommodated for minor variations in measurements over time, somewhat diminishing effects of self- and caregiver report.

Finally, the model’s relationships were assessed with dyadic data collected simultaneously, and the relationships may be bidirectional. In these analyses, limited support existed for directionality proposed in the model used, from caregivers’ stress to BSD. However, simultaneous assessment of model pathways across data points within dyads using more sophisticated multilevel mediation techniques may clarify these relationships.

Conclusion

Further research may allow BSD to emerge as a viable consequence of caregivers’ stress process, leading to improved intervention strategies that target BSD through relieving caregivers’ stress, interrupting the cycle of negative dyadic effects, and improving more distal outcomes (such as institutionalization) for these vulnerable dyads. Such research has practical, protocol, societal, financial, and policy implications for dementia care. More important, development of a “dyadic” approach, acknowledging the importance of attention to primary caregivers’ needs, may directly affect the quality of life for millions of community-based dementia care dyads.

References

  • Ballard, C., Lowery, K., Powell, I., O’Brien, J. & James, I. (2000). Impact of behavioral and psychological symptoms of dementia on caregivers. International Psychogeriatrics, 12(Suppl. S1), 93–105. doi:10.1017/S1041610200006840 [CrossRef]
  • Bedard, M., Molloy, D.W., Squire, L., Dubois, S., Lever, J.A. & O’Donnell, M. (2001). The Zarit Burden Interview: A new short version and screening version. The Gerontologist, 41, 652–657. doi:10.1093/geront/41.5.652 [CrossRef]
  • Beekman, A.T., Deeg, D.J., Van Limbeek, J., Braam, A.W., De Vries, M.Z. & Van Tilburg, W. (1997). Criterion validity of the Center for Epidemiologic Studies Depression scale (CES-D): Results from a community-based sample of older subjects in the Netherlands. Psychological Medicine, 27, 231–235. doi:10.1017/S0033291796003510 [CrossRef]
  • Black, W. & Almeida, O.P. (2004). A systematic review of the association between the behavioral and psychological symptoms of dementia and burden of care. International Psychogeriatrics, 16, 295–315. doi:10.1017/S1041610204000468 [CrossRef]
  • Burgener, S. & Twigg, P. (2002). Relationships among caregiver factors and quality of life in care recipients with irreversible dementia. Alzheimer Disease and Associated Disorders, 16, 88–102. doi:10.1097/00002093-200204000-00006 [CrossRef]
  • Campbell, J. (2009). A model of consequences of dementia caregivers’ stress process: Influence on behaviors in dementia and caregivers’ behavior-related reactions. Research and Theory for Nursing Practice, 23, 181–202. doi:10.1891/1541-6577.23.3.181 [CrossRef]
  • Cho, S.H. (2003). Using multilevel analysis in patient and organizational outcomes research. Nursing Research, 52, 61–65. doi:10.1097/00006199-200301000-00010 [CrossRef]
  • Cohen-Mansfield, J. (2000). Theoretical frameworks for behavioral problems in dementia. Alzheimer’s Care Quarterly, 1(4), 8–21.
  • Cohen-Mansfield, J. & Libin, A. (2004). Assessment of agitation in elderly patients with dementia: Correlations between informant rating and direct observation. International Journal of Geriatric Psychiatry, 19, 881–891. doi:10.1002/gps.1171 [CrossRef]
  • Cotter, E.M., Burgio, L.D., Roth, D.L., Gerstle, J. & Richardson, P. (2008). Comparison of caregiver and occupational therapist ratings of dementia patients’ performance of activities of daily living. Journal of Applied Gerontology, 27, 215–225. doi:10.1177/0733464807310681 [CrossRef]
  • Crawford, J.R. & Henry, J.D. (2004). The Positive and Negative Affect Schedule (PANAS): Construct validity, measurement properties and normative data in a large non-clinical sample. British Journal of Clinical Psychology, 43(Pt. 3), 245–265. doi:10.1348/0144665031752934 [CrossRef]
  • Cummings, J.L. (1997). The Neuropsychiatric Inventory: Assessing psychopathology in dementia patients. Neurology, 48(Suppl. 6), S10–S16.
  • Cummings, J.L., Mega, M., Gray, K., Rosenberg-Thompson, S., Carusi, D.A. & Gornbein, J. (1994). The Neuropsychiatric Inventory: Comprehensive assessment of psychopathology in dementia. Neurology, 44, 2308–2314.
  • Czaja, S.J., Eisdorfer, C. & Schulz, R. (2000). Future directions in caregiving: Implications for intervention research. In Schulz, R. (Ed.), Handbook on dementia caregiving: Evidence-based interventions for family caregivers (pp. 283–319). New York: Springer.
  • Davis, L.L., Buckwalter, K. & Burgio, L.D. (1997). Measuring problem behaviors in dementia: Developing a methodological agenda. Advances in Nursing Science, 20(1), 40–55.
  • de Vugt, M.E., Jolles, J., van Osch, L., Stevens, F., Aalten, P. & Lousberg, R. et al. (2006). Cognitive functioning in spousal caregivers of dementia patients: Findings from the prospective MAASBED study. Age and Ageing, 35, 160–166. doi:10.1093/ageing/afj044 [CrossRef]
  • de Vugt, M.E., Stevens, F., Aalten, P., Lousberg, R., Jaspers, N. & Winkens, I. et al. (2004). Do caregiver management strategies influence patient behaviour in dementia?International Journal of Geriatric Psychiatry, 19, 85–92. doi:10.1002/gps.1044 [CrossRef]
  • Denollet, J. & De Vries, J. (2006). Positive and negative affect within the realm of depression, stress and fatigue: The two-factor distress model of the Global Mood Scale (GMS). Journal of Affective Disorders, 91, 171–180. doi:10.1016/j.jad.2005.12.044 [CrossRef]
  • Diehl, M. (2005). [Selection of reliable subset of items on the PANAS based on variability in older adults.] Unpublished raw data.
  • Donaldson, C., Tarrier, N. & Burns, A. (1997). The impact of the symptoms of dementia on caregivers. British Journal of Psychiatry, 170, 62–68. doi:10.1192/bjp.170.1.62 [CrossRef]
  • Donaldson, C., Tarrier, N. & Burns, A. (1998). Determinants of carer stress in Alzheimer’s disease. International Journal of Geriatric Psychiatry, 13, 248–256. doi:10.1002/(SICI)1099-1166(199804)13:4<248::AID-GPS770>3.0.CO;2-0 [CrossRef]
  • Dunkin, J.J. & Anderson-Hanley, C. (1998). Dementia caregiver burden: A review of the literature and guidelines for assessment and intervention. Neurology, 51(Suppl. 1), S53–S60.
  • Finkel, S.I., Costa e Silva, J., Cohen, G., Miller, S. & Sartorius, N. (1996). Behavioral and psychological signs and symptoms of dementia: A consensus statement on current knowledge and implications for research and treatment. International Psychogeriatrics, 8(Suppl. 3), 497–500. doi:10.1017/S1041610297003943 [CrossRef]
  • Fitzmaurice, G.M., Laird, N.M. & Ware, J.H. (2004). Applied longitudinal analysis. Hoboken, NJ: Wiley and Sons.
  • Folstein, M.F., Folstein, S.E. & McHugh, P.R. (1975). “Mini-mental state.” A practical method for grading the cognitive state of patients for the clinician. Journal of Psychiatric Research, 12, 189–198. doi:10.1016/0022-3956(75)90026-6 [CrossRef]
  • Forester, B.P. & Oxman, T.E. (2003). Measures to assess the noncognitive symptoms of dementia in the primary care setting. Primary Care Companion to the Journal of Clinical Psychiatry, 5, 158–163. doi:10.4088/PCC.v05n0403 [CrossRef]
  • Gallicchio, L., Siddiqi, N., Langenberg, P. & Baumgarten, M. (2002). Gender differences in burden and depression among informal caregivers of demented elders in the community. International Journal of Geriatric Psychiatry, 17, 154–163. doi:10.1002/gps.538 [CrossRef]
  • Goode, K.T., Haley, W.E., Roth, D.L. & Ford, G.R. (1998). Predicting longitudinal changes in caregiver physical and mental health: A stress process model. Health Psychology, 17, 190–198. doi:10.1037/0278-6133.17.2.190 [CrossRef]
  • Hall, G.R. (1994). Caring for people with Alzheimer’s disease using the conceptual model of progressively lowered stress threshold in the clinical setting. Nursing Clinics of North America, 29, 129–141.
  • Harwood, D.G., Barker, W.W., Ownby, R.L. & Duara, R. (2000). Relationship of behavioral and psychological symptoms to cognitive impairment and functional status in Alzheimer’s disease. International Journal of Geriatric Psychiatry, 15, 393–400. doi:10.1002/(SICI)1099-1166(200005)15:5<393::AID-GPS120>3.0.CO;2-O [CrossRef]
  • Hebert, L.E., Scherr, P.A., Bienias, J.L., Bennett, D.A. & Evans, D.A. (2003). Alzheimer disease in the U.S. population: Prevalence estimates using the 2000 census. Archives of Neurology, 60, 1119–1122. doi:10.1001/archneur.60.8.1119 [CrossRef]
  • Hooker, K., Manoogian-O’Dell, M., Monahan, D.J., Frazier, L.D. & Shifren, K. (2000). Does type of disease matter? Gender differences among Alzheimer’s and Parkinson’s disease spouse caregivers. The Gerontologist, 40, 568–573. doi:10.1093/geront/40.5.568 [CrossRef]
  • Jones, R.N. & Gallo, J.J. (2001). Education bias in the Mini-Mental State Examination. International Psychogeriatrics, 13, 299–310. doi:10.1017/S1041610201007694 [CrossRef]
  • Kaufer, D.I., Cummings, J.L., Ketchel, P., Smith, V., MacMillan, A. & Shelley, T. et al. (2000). Validation of the NPI-Q, a brief clinical form of the Neuropsychiatric Inventory. The Journal of Neuropsychiatry and Clinical Neurosciences, 12, 233–239. doi:10.1176/appi.neuropsych.12.2.233 [CrossRef]
  • Kelly, A. & Rowe, M. (2006). Use of night home monitoring system to improve caregiver well-being [Abstract]. The Gerontologist, 46(Special Issue 1), 547.
  • Knight, B.G., Silverstein, M., McCallum, T.J. & Fox, L.S. (2000). A sociocultural stress and coping model for mental health outcomes among African American caregivers in southern California. Journals of Gerontology. Series B, Psychological Sciences and Social Sciences, 55, P142–P150. doi:10.1093/geronb/55.3.P142 [CrossRef]
  • Knight, R.G., Williams, S., McGee, R. & Olaman, S. (1997). Psychometric properties of the Centre for Epidemiologic Studies Depression Scale (CES-D) in a sample of women in middle life. Behaviour Research and Therapy, 35, 373–380. doi:10.1016/S0005-7967(96)00107-6 [CrossRef]
  • Kolanowski, A.M. & Whall, A.L. (2000). Toward holistic theory-based intervention for dementia behavior. Holistic Nursing Practice, 14(2), 67–76.
  • Lange, R.T., Hopp, G.A. & Kang, N. (2004). Psychometric properties and factor structure of the Neuropsychiatric Inventory Nursing Home Version in an elderly neuropsychiatric population. International Journal of Geriatric Psychiatry, 19, 440–448. doi:10.1002/gps.1108 [CrossRef]
  • Lyketsos, C.G. (2007). Neuropsychiatric symptoms (behavioral and psychological symptoms of dementia) and the development of dementia treatments. International Psychogeriatrics, 19, 409–420. doi:10.1017/S104161020700484X [CrossRef]
  • Lyketsos, C.G., Lopez, O., Jones, B., Fitzpatrick, A.L., Breitner, J. & DeKosky, S. (2002). Prevalence of neuropsychiatric symptoms in dementia and mild cognitive impairment: Results from the Cardiovascular Health Study. Journal of the American Medical Association, 288, 1475–1483. doi:10.1001/jama.288.12.1475 [CrossRef]
  • McClendon, M.J., Smyth, K.A. & Neundorfer, M.M. (2004). Survival of persons with Alzheimer’s disease: Caregiver coping matters. The Gerontologist, 44, 508–519. doi:10.1093/geront/44.4.508 [CrossRef]
  • Mittelman, M.S., Roth, D.L., Haley, W.E. & Zarit, S.H. (2004). Effects of a caregiver intervention on negative caregiver appraisals of behavior problems in patients with Alzheimer’s disease: Results of a randomized trial. Journals of Gerontology. Series B, Psychological Sciences and Social Sciences, 59, P27–P34. doi:10.1093/geronb/59.1.P27 [CrossRef]
  • O’Rourke, N. & Tuokko, H.A. (2003). Psychometric properties of an abridged version of the Zarit Burden Interview within a representative Canadian caregiver sample. The Gerontologist, 43, 121–127. doi:10.1093/geront/43.1.121 [CrossRef]
  • Ory, M.G., Hoffman, R.R. III. , Yee, J.L., Tennstedt, S. & Schulz, R. (1999). Prevalence and impact of caregiving: A detailed comparison between dementia and nondementia caregivers. The Gerontologist, 39, 177–185. doi:10.1093/geront/39.2.177 [CrossRef]
  • Pearlin, L.I., Mullan, J.T., Semple, S.J. & Skaff, M.M. (1990). Caregiving and the stress process: An overview of concepts and their measures. The Gerontologist, 30, 583–594. doi:10.1093/geront/30.5.583 [CrossRef]
  • Pinquart, M. & Sorensen, S. (2003a). Associations of stressors and uplifts of caregiving with caregiver burden and depressive mood: A meta-analysis. Journals of Gerontology. Series B, Psychological Sciences and Social Sciences, 58, P112–P128. doi:10.1093/geronb/58.2.P112 [CrossRef]
  • Pinquart, M. & Sorensen, S. (2003b). Differences between caregivers and noncaregivers in psychological health and physical health: A meta-analysis. Psychology and Aging, 18, 250–267. doi:10.1037/0882-7974.18.2.250 [CrossRef]
  • Pinquart, M. & Sorensen, S. (2007). Correlates of physical health of informal caregivers: A meta-analysis. Journals of Gerontology. Series B, Psychological Sciences and Social Sciences, 62, P126–P137.
  • Quayhagen, M.P., Quayhagen, M., Corbeil, R.R., Hendrix, R.C., Jackson, J.E. & Snyder, L. et al. (2000). Coping with dementia: Evaluation of four nonpharmacologic interventions. International Psychogeriatrics, 12, 249–265. doi:10.1017/S1041610200006360 [CrossRef]
  • Radloff, L.S. (1977). The CES-D scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, 385–401. doi:10.1177/014662167700100306 [CrossRef]
  • Riggs, J.A. (2001). The health and long-term care policy challenges of Alzheimer’s disease. Aging & Mental Health, 5(Suppl. 1), S138–S145.
  • Rowe, M., Lane, S. & Phipps, C. (2007). CareWatch: A home monitoring system for use in homes of persons with cognitive impairment. Topics in Geriatric Rehabilitation, 23, 3–8.
  • Rowe, M., Phipps, C., Horne, C., Kelley, A., Lawson, A. & Lehman, B. et al. (2006). CareWatch study: Introduction, methods, and impact on caregiver worry [Abstract]. The Gerontologist, 46(Special Issue 1), 547.
  • Rowe, M.A., Ahn, H., Benito, A.P., Stone, H., Wilson, A. & Kairalla, J. (2010). Injuries and unattended home exits in persons with dementia: A 12-month prospective study. American Journal of Alzheimer’s Disease and Other Dementias, 25, 27–31. doi:10.1177/1533317508323138 [CrossRef]
  • Rowe, M.A., Kelly, A., Horne, C., Lane, S., Campbell, J. & Lehman, B. et al. (2009). Reducing dangerous nighttime events in persons with dementia using a nighttime monitoring system. Alzheimer’s & Dementia, 5, 419–426. doi:10.1016/j.jalz.2008.08.005 [CrossRef]
  • Schafer, J.L. & Graham, J.W. (2002). Missing data: Our view of the state of the art. Psychological Methods, 7, 147–177. doi:10.1037/1082-989X.7.2.147 [CrossRef]
  • Schneider, L.S., Tariot, P.N., Lyketsos, C.G., Dagerman, K.S., Davis, K.L. & Davis, S. et al. (2001). National Institute of Mental Health Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE): Alzheimer disease trial methodology. American Journal of Geriatric Psychiatry, 9, 346–360.
  • Schroevers, M.J., Sanderman, R., van Sonderen, E. & Ranchor, A.V. (2000). The evaluation of the Center for Epidemiologic Studies Depression (CES-D) scale: Depressed and positive affect in cancer patients and healthy reference subjects. Quality of Life Research, 9, 1015–1029. doi:10.1023/A:1016673003237 [CrossRef]
  • Schulz, R., Beach, S.R., Lind, B., Martire, L.M., Zdaniuk, B. & Hirsch, C. et al. (2001). Involvement in caregiving and adjustment to death of a spouse: Findings from the caregiver health effects study. Journal of the American Medical Association, 285, 3123–3129. doi:10.1001/jama.285.24.3123 [CrossRef]
  • Schulz, R., Belle, S.H., Czaja, S.J., McGinnis, K.A., Stevens, A. & Zhang, S. (2004). Long-term care placement of dementia patients and caregiver health and well-being. Journal of the American Medical Association, 292, 961–967. doi:10.1001/jama.292.8.961 [CrossRef]
  • Schulz, R. & Martire, L.M. (2004). Family caregiving of persons with dementia: Prevalence, health effects, and support strategies. American Journal of Geriatric Psychiatry, 12, 240–249. doi:10.1176/appi.ajgp.12.3.240 [CrossRef]
  • Singer, J.D. & Willett, J.B. (2003). Applied longitudinal data analysis: Modeling change and event occurrence. New York: Oxford University Press.
  • Sink, K.M., Covinsky, K.E., Barnes, D.E., Newcomer, R.J. & Yaffe, K. (2006). Caregiver characteristics are associated with neuropsychiatric symptoms of dementia. Journal of the American Geriatrics Society, 54, 796–803. doi:10.1111/j.1532-5415.2006.00697.x [CrossRef]
  • Smith, M. & Buckwalter, K. (2005). Behaviors associated with dementia. American Journal of Nursing, 105(7), 40–52. doi:10.1097/00000446-200507000-00028 [CrossRef]
  • Smith, M., Hall, G.R., Gerdner, L. & Buckwalter, K.C. (2006). Application of the progressively lowered stress threshold model across the continuum of care. Nursing Clinics of North America, 41, 57–81. doi:10.1016/j.cnur.2005.09.006 [CrossRef]
  • Steinberg, M., Tschanz, J.T., Corcoran, C., Steffens, D.C., Norton, M.C. & Lyketsos, C.G. et al. (2004). The persistence of neuropsychiatric symptoms in dementia: The Cache County study. International Journal of Geriatric Psychiatry, 19, 19–26. doi:10.1002/gps.1025 [CrossRef]
  • Tarrier, N., Barrowclough, C., Ward, J., Donaldson, C., Burns, A. & Gregg, L. (2002). Expressed emotion and attributions in the carers of patients with Alzheimer’s disease: The effect on carer burden. Journal of Abnormal Psychology, 111, 340–349. doi:10.1037/0021-843X.111.2.340 [CrossRef]
  • Teri, L. (1997). Behavior and caregiver burden: Behavioral problems in patients with Alzheimer disease and its association with caregiver distress. Alzheimer Disease and Associated Disorders, 11(Suppl. 4), S35–S38.
  • Teri, L., McCurry, S.M., Logsdon, R. & Gibbons, L.E. (2005). Training community consultants to help family members improve dementia care: A randomized controlled trial. The Gerontologist, 45, 802–811. doi:10.1093/geront/45.6.802 [CrossRef]
  • Vitaliano, P.P., Young, H.M., Russo, J., Romano, J. & Magana-Amato, A. (1993). Does expressed emotion in spouses predict subsequent problems among care recipients with Alzheimer’s disease?Journal of Gerontology, 48, P202–P209.
  • Vitaliano, P.P., Zhang, J. & Scanlan, J.M. (2003). Is caregiving hazardous to one’s physical health? A meta-analysis. Psychological Bulletin, 129, 946–972. doi:10.1037/0033-2909.129.6.946 [CrossRef]
  • Watson, D., Clark, L.A. & Tellegen, A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology, 54, 1063–1070. doi:10.1037/0022-3514.54.6.1063 [CrossRef]
  • Williamson, G.M. & Shaffer, D.R. (2001). Relationship quality and potentially harmful behaviors by spousal caregivers: How we were then, how we are now. The Family Relationships in Late Life Project. Psychology and Aging, 16, 217–226. doi:10.1037/0882-7974.16.2.217 [CrossRef]
  • Zarit, S.H., Orr, N.K. & Zarit, J.M. (1985). The hidden victims of Alzheimer’s disease: Families under stress. New York: New York University Press.
  • Zarit, S.H., Reever, K.E. & Bach-Peterson, J. (1980). Relatives of the impaired elderly: Correlates of feelings of burden. The Gerontologist, 20, 649–655. doi:10.1093/geront/20.6.649 [CrossRef]

Estimation of Parameters for Multilevel Analyses

Estimation Process Component Statistical Technique Description
Model estimation Maximum likelihood Computer uses an iterative procedure to determine the most probable estimates, that is, examines within- and between-dyads variability and assigns more weight to sample means when participants have high within-dyads variability or missing data points.
Steps in estimation Unconditional means Model without predictors to establish variance for further modeling using intraclass correlations.
Unconditional growth or change Model with only time as a predictor to establish whether enough variance existed within dyads to model Level 2 factors.
Unstructured conditional model Modeled parameters assess for reduction in within-dyads variance, that is, experimental group in change models and caregivers’ stress variables in predictor models.
Structured models Model including factors or predictors and with the covariance structured.
Model comparisons (criteria used in model-building steps) −2 log likelihood Used to evaluate subsequent improvement of models compared to the baseline unconditional models.
Pseudo-R2 Assesses model’s ability to explain variance, similar to the use of R2 in traditional models. However, this value represents reduction of explainable variance; a fairly small change may mean a large pseudo-R2.
Structure of covariance Variance components Default in initial models, since unstructured models have a high number of parameters for estimation.
Repeated variable Final models structured with a repeated variable on a random statement (i.e., in this study, month was used).
Alternate covariate structures Final models run with several covariance structures (e.g., autoregressive), chosen based on expected longitudinal variances and correlations across elements.
Akaike information criterion (AIC) Criterion used to compare structured models with the default; this criterion balances complexity with parsimony in indicating the best model fit. The lowest AIC was considered the best fit of the data.

Interpretation of Caregivers’ Stress Process Predictor Variables in Aim 2 Analyses

Level of Analysis Variable Meaning Interpretation in Relation to Outcome of BSD
Fixed
Level 2: across or between dyads Average levels perceived stress and emotional-behavioral responses (predictors meaned) Establishes dyads’ mean levels of predictors. Dyads’ means were then considered across dyads for the population effect; each dyad’s contribution to the estimate was the same regardless of data point; a dyad’s means are compared to the population’s typical effect. What is the extent to which caregivers’ general levels of stress process variables affect BSD, when compared across dyads? For example, are those dyads that are characterized with higher than average caregivers’ stress also likely to have more severe BSD in their care recipient? This effect reflected how caregivers’ stress process influenced BSD in general, not changing over time.
Level 1: within dyads Current levels perceived stress and emotional-behavioral responses (predictors centered) Establishes dyads’ current deviation or variation from their own mean at each given data point. Dyad’s current deviations considered across dyads for the population effect at each data point; each dyad’s contributions to the estimate varied from one data point to another. What is the extent to which caregivers’ current levels of stress process variables within dyads affect BSD within dyads? For example, on occasions (data points) when caregivers’ stress process variables were higher than the estimated population’s typical, were care recipients’ BSD higher, on average, as well? This effect reflected how caregivers’ changing stress process influenced BSD over time.
Random
Level 1: within dyads Current levels perceived stress and emotional-behavioral responses (predictors centered) Establishes dyads’ current deviation or variation from their own mean at each given data point. Each dyad’s trajectory of current deviations from their own mean was compared to population’s typical trajectory. Given a relationship between caregivers’ stress process variables and BSD from occasion to occasion (over data points in study), do differences exist among dyads in these predictor-outcome relationships or do the relationships generally hold across all dyads?

Findings by Experimental Group for the Sample’s Observed Raw Data Across Study

Parameter Baseline Month 2 Month 3 Month 4 Month 5 Month 6 Month 8 Month 10 Month 12 All Months
Perceived stress (Short Zarit Burden Interview) (r = 0.53)a
  Mean (SD) 21.86 (8.83) 21.43 (7.70) 20.65 (8.97) 20.00 (8.15) 20.15 (9.04) 19.92 (8.88) 19.09 (8.61) 21.10 (9.55) 20.70 (10.48) 20.74 (8.62)
  Diff 5.14 5.44 5.16 3.94 4.32 3.76 5.60 6.87 7.40 5.13
  t test 2.11 2.53 2.06 1.55 1.52 1.31 1.91 2.08 1.92 5.81
  df 47 44 46 38 37 36 30 28 25 333.38
  p value 0.04 0.02 0.05 0.13 0.11 0.14 0.07 0.05 0.07 0.000
Emotional-behavioral responses (CES-D and NegPANAS combinedb) (r = 0.48)a
  Mean (SD) 0.102 (0.906) –0.011 (0.942) –0.086 (0.977) 0.058 (0.908) 0.054 (0.879) –0.048 (0.866) –0.147 (0.742) 0.069 (0.932) –0.021 (0.935) –0.002 (0.90)
  Diff 0.34 0.48 0.27 0.56 0.36 0.23 0.54 0.52 0.55 0.415
  t test 1.33 1.87 0.98 2.01 1.31 0.82 2.14 1.84 1.42 4.51
  df 47 42 47 38 38 37 23 29 31 354.47
  p value 0.19 0.07 0.33 0.06 0.20 0.42 0.04 0.08 0.17 0.000
Behavioral symptoms (NPI-Q behaviors)
  Mean (SD) 11.58 (7.12) 9.49 (7.17) 11.00 (7.53) 11.22 (7.14) 9.04 (6.80) 9.70 (7.00) 10.61 (6.98) 11.07 (7.95) 11.44 (8.44) 10.55 (7.28)
  Diff 2.92 3.34 4.60 2.35 0.063 1.89 2.56 2.89 3.14 3.01
  t test 1.18 1.36 2.12 0.98 0.03 0.82 1.04 1.53 1.65 3.72
  df 32 31 43 34 34 35 30 27 21 293.98
  p value 0.25 0.19 0.04 0.33 0.98 0.42 0.31 0.14 0.11 0.000

Estimated Parameters for Each Variable Modeled Over Time and by Experimental Status, then with Caregivers’ Stress Process Variables as Predictor Effects on BSD Modeled

Caregivers’ Perceived Stress Caregivers’ Emotional-Behavioral Responses BSD
Model Variables Estimate (SE) Estimate (SE) Estimate (SE)
Fixed effects (across population)
Intercept 18.56 (1.54) 0.0002 (0.12) 9.08 (1.19)
Linear time effect 0 (0) 0.00004 (0.00) 0 (0)
Quadratic time effect 0.00004 (0.00) 0.000005 (2.21) N/Aa
Treatment effect at midstudy (control) 5.03 (2.12) 3.19 (1.64)
Linear time (Treatment group change per day from baseline)
  Control 0.005 (0.00) N/Aa 0.006 (0.00)
  Experimental 0.0002 (0.00) N/Aa –0.003 (0.00)
Quadratic time (Treatment group change per day from baseline)
  Control 0.00003 (0.00) N/Aa
  Experimental 0 (0) N/Aa
Effects of perceived stress on BSDb
BSD intercept, controlling for perceived stress 1.86 (2.01)
Effect of average perceived stress at midstudy 0.42 (0.09)
Effect of current perceived stress from occasion to occasion 0.33 (0.08)
Effects of emotional-behavioral responses on BSDb
BSD intercept controlling for emotional-behavioral responses 10.67 (0.73)
Effect of average emotional-behavioral responses at mid-study 3.57 (0.91)
Effect of current emotional-behavioral responses from occasion to occasion 2.14 (0.61)
Authors

Dr. Campbell is Associate Professor, Remington College of Nursing, Lake Mary, Dr. Rowe is Professor, College of Nursing, and Dr. Marsiske is Associate Professor, Department of Clinical and Health Psychology, University of Florida, Gainesville, Florida. Dr. Rowe is also Nurse Scientist, James A. Haley Veterans’ Hospital, Tampa, Florida.

The authors disclose that they have no significant financial interests in any product or class of products discussed directly or indirectly in this activity. Judy L.M. Campbell was supported in this dissertation research by a Hartford Geriatric Nursing Initiative Building Academic Geriatric Nursing Capacity scholarship and a University of Florida Alumni Fellowship. The primary study was funded by the National Institute of Nursing Research (3R42NR004952-03S1, 5R42NR004952-03).

Address correspondence to Judy L.M. Campbell, PhD, ARNP, Associate Professor, Remington College of Nursing, 660 Century Point, Suite 1050, Lake Mary, FL 32746; e-mail: judy@mavweb.com.

Received: January 21, 2009
Accepted: May 13, 2010
Posted Online: September 30, 2010

10.3928/19404921-20100901-02

Sign up to receive

Journal E-contents