Agitation and aggression in people living with dementia and mild cognitive impairment in shared-housing arrangements – validation of a German version of the Cohen-Mansfield Agitation Inventory-Short Form (CMAI-SF)

Nearly all people living with dementia (PlwD) will be affected by behavioral and psychological symptoms of dementia (BPSD) over the course of their disease [1]. BPSD include agitation and aggression but also other non-cognitive symptoms of dementia, such as depression, anxiety, delusions, or hallucinations. For PlwD, agitation in particular is a common complex of symptoms: prevalence rates range from 33 to 45% for PlwD in nursing homes, and the prevalence increases in more progredient illness states [2]. Occasionally, a higher prevalence (i.e. 85% for at least one agitation symptom) is noted [3]. Agitated and aggressive behaviors challenge both the PlwD and their caregivers [4, 5]. A recent meta-analysis showed that there are efficacious non-pharmacological interventions for agitation and aggression in PlwD [6]. However, BPSD still constitute a common cause of hospitalization for PlwD [7, 8], which in turn is associated with negative consequences, such as the risk of institutionalization and declines in mental, cognitive, and physical capabilities [9]. Furthermore, hospitalization can stabilize or worsen agitation and aggression, most likely by increasing unmet needs or distress [9]. In order to carry out valid and reliable research on agitation and aggression and its consequences, the sound scientific measurement of these constructs is a prerequisite.

Early efforts to describe and categorize different behaviors reflecting agitation and aggression were made by Cohen-Mansfield et al. [10], who described 29 behaviors in the Cohen-Mansfield Agitation Inventory (CMAI) each rated on a 7-point frequency scale [10, 11]. These behaviors were observed by professional caregivers in nursing homes and clustered into the three factors (I) “aggressive behaviors,” such as pushing, scratching, and grabbing, (II) “physically non-aggressive behaviors,” such as handlings things inappropriately, as well as (III) “verbally agitated behaviors,” such as complaining, negativism, and screaming [10]. Some later validation studies on older people with or without dementia in different countries were able to replicate the original three factor structure [12, 13], while some other studies with nursing home residents suggested the categorization of the 29 behaviors into four categories (an additional Factor IV called “hiding and hoarding” comprising the items “hiding” and “hoarding”) [14‐16].

Beyond that, Cohen-Mansfield et al. [17] developed the community version of the CMAI (CMAI-C), which is an expanded version of the CMAI including 36 items specifically adapted to day-care population. Using the CMAI-C, Cohen-Mansfield et al. [17] found that in semi-inpatient structures, such as day-care centers, agitation behaviors rated by family caregivers clustered into the three original categories for nursing home residents found by Cohen-Mansfield et al. [10], whereas agitation behaviors rated by staff clustered into the three categories: “physically non-aggressive behaviors” (corresponding with the original Factor II), “verbally agitated behaviors” (corresponding with the original Factor III), and “verbal aggressive behaviors” (behaviors that originally fell under Factor I). Thus, the concept was adapted to semi-inpatient and community-settings, and four possible behavioral categories were named: “verbally non-aggressive behaviors (VNAB),” “verbally aggressive behaviors (VAGB),” “physically non-aggressive behaviors (PNAB),” and “physically aggressive behaviors (PAGB).” By contrast, Weiner et al. [18] could not satisfactorily replicate the suggested factor structure in community-dwelling PlwD using the CMAI-C. Their results suggested that agitation behaviors in community-dwelling PlwD cluster unstable into the “verbally aggressive behavior,” “verbally agitated behavior,” and “hiding and hoarding” categories, whereby many physically aggressive behaviors do not occur or rarely occur in community-dwelling PlwD. The latter was also reported by Koss et al. [19]. As a consequence, Weiner et al. [18] suggested that the CMAI-C should be used instead as an overall measure and that subscale scoring does not seem applicable in community-dwelling PlwD.

The retest reliability of the CMAI-C (r = 0.83) [19] and the CMAI (r = 0.85) [13] was reported to be acceptable in PlwD. Studies exploring construct validity including factor structure of the long form [10, 12‐16] and the CMAI-C [17, 18] in different settings are displayed in Table 1.

Recently, an observation tool (CMAI-O) for the original 29-item CMAI was also developed by Griffiths et al. [23] to rate behaviors on a 4-point frequency scale specifically by independent trained observers. The CMAI-O showed adequate internal consistency (α = 61) and convergent as well as discriminant validity with other instruments, while the factor structure was not analyzed.

Cohen-Mansfield et al. also developed a short form (CMAI-SF) based on the factor structure of the CMAI, which includes 14 items, each rated on a 5-point instead of a 7-point frequency scale [11, 22]. It was first described in a study of nursing home residents in which an interrater reliability of 81.8–92.3% was observed [22], and has been used in several studies in recent years [24‐30]. Unfortunately, there have been few studies examining construct validity including factor structure. In both nursing home residents in the United States of America (USA) and PlwD attending day care centers in Taiwan, the original three-factor structure consisting of “aggressive behavior”, “physically non-aggressive behavior”, and “verbally agitated behavior” was found (see Table 1).

Because the 14-item CMAI-SF [11, 22] has not been validated in German language, a validation study exploring the construct validity of the measure is scientifically necessary.

Whether the German short form of the instrument can replicate the number and content of the previously published factors remains to be explored. In addition, shared-housing arrangements (SHAs), i.e. innovative, homelike care environments [31], are classified as outpatient settings. However, the conditions are not the same as living in the family structure, but they are also more individual than in nursing homes. Thus, it is unclear whether the three factor structure that has been already found for the CMAI-SF in nursing home [20] and day care [21] can be replicated in the setting of SHAs or whether the four factor structure reported for the original CMAI in semi-inpatient settings such as day-care facilities [17] will be observed.

Therefore, the aim of this study is to explore the construct validity of a German version of the CMAI-SF in people with cognitive impairment in the outpatient setting of German shared-housing arrangements (SHAs).

In this first validation of the German version of the CMAI-SF, an unambiguous three-factor structure with acceptable internal consistency was confirmed with a PCA. Almost all item loadings were above 0.50, with the exception of two items loading above 0.46. The internal consistency of the total score form was high.

The three factor structure (“aggressive behavior”, “physically non-aggressive behavior”, and “verbally agitated behavior”) that has been already found for the CMAI-SF in nursing home [20] and day care samples [21] could be replicated in SHAs using the German version of the CMAI-SF. The first factor was largely consistent with the factor "aggressive behavior" found in the validation of the original 29-item CMAI by Cohen-Mansfield et al. [10] and Rabinowitz et al. [14] in nursing home samples, except that in these two previous studies, the item "strange noises" did not load on any factor, and in Cohen-Mansfield et al. [10], the item “other aggressive behaviors” could not be assigned to one factor because these behaviors appeared in less than 5% of the sample. The second factor in our study was largely consistent with the “verbally agitated behavior” factor from the original 29-item CMAI found by Cohen-Mansfield et al. [10] and Rabinowitz et al. [14], with the exception that in the present work, Item 6 "general restlessness, performing repetitive mannerisms, tapping, strange movements" also loaded on this factor. The content of this item includes behaviors that reflect physical agitation. However, the factor loading of this item constituted one of the two lowest loadings from the questionnaire (< 0.50); when we applied a varimax rotation, the ambiguous item had similar loadings on Factor 2 and the content-matching Factor 3. To avoid this ambiguity, we decided to use a promax rotation in the final version. The third factor was also largely consistent with the “physically non-aggressive behavior” factor found in the validation of the original 29-item CMAI by Cohen-Mansfield et al. [10] and Rabinowitz et al. [14] except that Item 6 "general restlessness, performing repetitive mannerisms, tapping, strange movements" did not load on this factor when a promax rotation was used. In our study, Item 13 “hiding things, hoarding things” also loaded on Factor 3 “physically non-aggressive behavior” just as it had in the day-care sample [17] and did not emerge as a fourth factor as it had in some nursing home validation samples [14‐16]. In summary, with a few exceptions, the factor structure closely resembled the structure found by Cohen-Mansfield et al. [10] and Rabinowitz et al. [14] in their nursing home samples using the original CMAI.

However, the assignment of Item 6 remained unsettled because the ambiguous results from the varimax rotation also showed that Item 6 loaded on the content-matching Factor 3 ("physically non-aggressive behavior"), but the results from the promax rotation indicated that Item 6 loaded on Factor 2 (“verbally agitated behavior”). Therefore, future research should further investigate the results of this exploratory factor analysis. In fact, Sun et al. [21] in their validation of a Chinese version of the CMAI-SF also found that item 6 loaded on the "verbally agitated behavior" factor – they explained this by considering that people living with dementia express their demands or unmet needs in part through agitation, tapping, and strange movements, which could therefore be classified as "verbally aggressive."

The four-factor structure found in day-care centers by Cohen-Mansfield et al. [17] did not appear in the present work, which is plausible considering that people in SHAs (i.e. our sample) live in the care environment 24/7, thus closely resembling life in a nursing home. Indeed, SHAs are seen as an alternative to a nursing home, when living at home is no longer possible [50]. In addition, even if a high degree of autonomy and self-determination is guaranteed to the residents in SHAs, the residents live in a care environment with professional care and not at home [31, 50, 51]. This resemblance to a nursing home is further supported by the fact that we detected a factor structure in our sample, whereas it was not possible for Weiner et al. [18] to detect a factor structure for community-dwelling PlwD because several of the CMAI items had a frequency below 5% in this subgroup. Nonetheless, using the CMAI-SF, Sun et al. [21] also found the original three-factor structure in people with dementia in day care settings, so it could also be concluded that the CMAI-SF in particular has a three-factor structure both in the nursing home and in semi-inpatient settings such as day care. Further research is needed to sufficiently clarify this issue.

Internal consistency, measured with Cronbach’s α, was good for the total CMAI-SF and acceptable for the three factors. All items increased the internal consistencies of the CMAI-SF total score and the three factor scores: For all items, Cronbach’s α “if item deleted” was below Cronbach’s α for the complete scale or the factors. All the CMAI-SF items showed moderate to high discriminatory power, whereas item difficulty was rather low. These results suggest that people with MCI and mild to moderate dementia in German SHAs had relatively low agitation levels. However, all 14 items were relevant in the sense that each behavior was described in at least 5% of the PlwD by the proxy raters.

With respect to construct validity, the convergent validity of the CMAI-SF was confirmed. As hypothesized, there were significant and high correlations between the CMAI-SF total score and the NPI-NH agitation item as well as the NPI-NH “agitation & restless behavior” factor [33]. Furthermore, QoL decreased significantly as agitation/aggression increased, as there were low to moderate significant negative correlations between the CMAI-SF total score and QUALIDEM total and subscores. The discriminant validity of the CMAI-SF was confirmed, as there was a low but significant positive correlation between the CMAI-SF total score and the NPI-NH apathy item (as hypothesized on the basis of Griffiths et al., 2020) [23].

The present study is the first study to validate a German version of the CMAI-SF, the 14-item short form of the CMAI. The results show that the CMAI-SF is a time-efficient, reliable, and valid scale for assessing agitation in people with MCI and dementia in SHAs. In our exploratory factor analysis, we identified three factors, which were similar to the factors “aggressive behavior,” “verbally agitated behavior,” and “physically non-aggressive behavior” already found in nursing home samples for the original CMAI and the CMAI-SF and in day care samples for the CMAI-SF. Thus, the findings provide preliminary evidence that the CMAI-SF can be used instead of the long-form CMAI to reduce time, costs, and the burden on proxy raters, when investigating agitation and aggression in people with dementia and MCI in future trials. However, further research is needed to confirm these results.