Mealtime Interactions between Nursing Home Staff and Residents with Dementia: A Behavioral Analysis of Language Characteristics

Background. Quality staff-resident communication is crucial to promote outcomes in nursing home residents with dementia requiring assistance during mealtimes. Better understanding of staff-resident language characteristics in mealtime interactions help promote effective communication, yet evidence is limited. This study aimed to examine factors associated with language characteristics in staff-resident mealtime interactions. Methods. This was a secondary analysis of 160 mealtime videos involving 36 staff and 27 residents with dementia (53 unique staff-resident dyads) in 9 nursing homes. We examined the associations of speaker (resident vs. staff), utterance quality (negative vs. positive), intervention (pre- vs. post-communication intervention), and resident dementia stage and comorbidities with expression length (number of words in each utterance) and addressing partner by name (whether staff or resident named their partner in each utterance), respectively. Results. Staff (utterance n = 2990, 99.1% positive, mean = 4.3 words per utterance) predominated conversations and had more positive, longer utterances than residents (utterance n = 890, 86.7% positive, mean = 2.6 words per utterance). As residents progressed from moderately-severe to severe dementia, both residents and staff produced shorter utterances (z=−2.66, p = .009). Staff (18%) named residents more often than residents (2.0%; z = 8.14, p < .0001) and when assisting residents with more severe dementia (z = 2.65, p = .008). Conclusions. Staff-resident communication was primarily positive, staff-initiated, and resident-oriented. Utterance quality and dementia stage were associated with staff-resident language characteristics. Staff play a critical role in mealtime care communication and should continue to initiate resident-oriented interactions using simple, short expressions to accommodate resident declining language abilities, particularly those with severe dementia. Staff should practice addressing residents by their names more frequently to promote individualized, targeted, person-centered mealtime care. Future work may further examine staff-resident language characteristics at word and other levels using more diverse samples.


Samples And Settings
In the parent trial, a convenience sample of 127 staff and 83 residents were recruited from 13 nursing homes (NHs) in Kansas, USA. NHs located within two hours from the research site and providing care for residents with dementia were recruited. In each NH, residents were eligible if they had 1) a dementia diagnosis, 2) staff-reported resistiveness to care, 3) long-stay status, 4) hearing capacity, and 5) a surrogate decision maker to provide informed consent were eligible (18). Staff were eligible if they 1) were older than 18 years old, 2) were permanent employees, 3) were able to communicate in English, and 4) delivered direct care for a resident participant for ≥ 2 times/week over the previous month.
Videos were recorded to capture staff-resident communication during daily care activities, including mealtimes, and were archived in the parent trial. The archived videos were screened for this study. Videos were eligible if they: 1) lasted for ≥ 1 minute, 2) captured mealtime interactions between one resident and one staff, and 3) captured utterances with adequate audio quality. A total of 1,748 videos were screened, from which, 1,588 videos were excluded due to lasting < 1 minute (n = 63), capturing other activities rather than mealtimes (n = 1,486), involving more than one staff and/or more than one resident (n = 34), and poor audio quality (n = 5). Thus, 160 videos were eligible for this study, of which, 110 were collected prior to the staff communication training (pre-intervention) and 50 were after the training (post-intervention) (18). The 160 videos lasted between 1-23.8 minutes (mean = 4.5, SD = 3.8). The 160 videos that were used in this study involved 27 residents and 36 staff (53 unique staff-resident dyads) in 9 NHs.
The nine NHs ranged from 43 to 163 beds in size (Median = 60 beds) and were distributed evenly on location (n = 4 rural, n = 5 urban), pro t status (n = 4 for-pro t, n = 5 non-for-pro t), and quality ratings (n = 4 rated1-3 stars, n = 5 rated 4-5 stars). Five NHs had one or more memory care units.
Resident characteristics (i.e., age, gender, race, ethnicity, dementia stage, physical comorbidities) were collected through medical records. Resident participants had a mean age of 85.6 years old (Table 1). All residents were White. The majority were female (63.0% vs. 37% male) and non-Hispanic (92.6% vs. 7.4% Hispanic). Residents had moderately severe (70.0%) or severe (30.0%) dementia as determined by reviewing Minimum Data Set (MDS) 3.0 using the Functional Assessment Staging in Alzheimer's Disease (FAST, total score ranges from 1, normal cognition, to 8, severe dementia) (19). Residents had moderate levels of physical comorbidities (mean = 27.1, range = 19-36) as evaluated by reviewing MDS 3.0 and clinical records using the Modi ed Cumulative Illness Rating Scale (total score ranges from 0 to 70 with higher scores indicating more comorbidities) (20).
Staff characteristics (i.e., age, gender, race, ethnicity, education, job title, years working as a caregiver, years working in current NH) were collected using self-report surveys. Staff participants had a mean age of 35.9 years old, worked as a caregiver for a mean length of 9.5 years, and worked at the current NH for a mean length of 4.0 years (  Four coders were trained by the rst author following a standard training and coding manual. After establishing inter-coder reliability, each of the four trained coders coded a subset of videos independently. Each utterance (a statement or question) was assigned a code. All utterances were coded as point events, where onset time (vs. offset time) of utterance was coded. Detailed coding process and conceptual and operational de nitions of all codes are described elsewhere (2,21,22). Coded data were exported from the Noldus Observer® to Excel worksheets.
Dependent Variables. For this study, two dependent variables-expression length and addressing partner by their name-representing word-level language characteristics were coded for each transcribed utterance of residents and staff.
Expression length, a continuous variable operationalized as the number of words produced in each utterance.
Addressing the partner by their name, a binary variable operationalized as whether staff or resident named their dyadic partner (i.e., resident or staff) in each utterance.
Independent Variables. For this study, the independent variables included speaker, utterance quality, and intervention, in addition to resident dementia stage and comorbidities.
Speaker, a binary variable operationalized as whether resident or staff produced each utterance.
Utterance quality, a binary variable operationalized as whether each utterance was coded as positive vs. negative in quality based on the re ned CUED.
Intervention, a binary variable operationalized as whether videos were collected before or after the dementia communication training was delivered to staff (pre-vs. post-intervention).

Data analysis
Characteristics of facilities, staff, residents, and utterances were described using descriptive statistics (n/%, mean/SD). Poisson-link mixed-effects model was used to t expression length with speaker (resident vs. staff), utterance quality (negative vs. positive), and intervention (pre-vs. post-intervention) as xed effects. Poisson distribution is appropriate as the dependent variable, expression length, is a count of the number of occurrences during a de ned time interval (23). Logit-link mixed-effects model was used to t the binary measure of whether the speaker addressed the partner by their name with speaker, utterance quality, and intervention as xed effects. Further, resident dementia stage and comorbidities (log-transformed) were added to the models to examine the effects of resident characteristics on language and whether estimates of other xed effects change. All models included staff, resident, and staff-resident dyad as random effects and were t using the LMER package in R version 4.1.0 (24). The level of signi cance was set as .05.

Language characteristics
Staff (utterances n = 2990) spoke three times more often than residents (utterances n = 890, Table 2). Most utterances produced by staff (99.1%) and residents (86.7%) were positive. Staff produced longer expressions (mean = 4.30 words per utterance) than residents (mean = 2.64 words per utterance) in general, as well as in both positive and negative utterances (4.31 and 3.58 words in staff utterances vs. 2.53 and 3.37 words in resident utterances). Staff addressed residents by their name (18%) more often than residents naming staff (2%) in general, as well as in both positive and negative utterances (18% and 14.8% of staff utterances vs. 1.8% and 3.4% of resident utterances).

Expression Length
The model revealed a signi cant main effect of speaker (z = 21.67, p < .0001) and utterance quality (z = 2.00, p = .046; Table 3), indicating that staff generally produced longer expressions than residents, and positive utterances were longer than negative utterances. The main effect of intervention was not signi cant (z=-1.10, p = .27). The three-way interaction between speaker, utterance quality, and intervention was signi cant (z = 4.74, p < .0001), which was driven by the signi cant interaction between speaker and utterance quality pre-intervention only (Fig. 1). Particularly, staff positive utterances were longer than their negative utterances (z = 2.80, p = .005), whereas resident positive utterances were shorter than their negative utterances pre-intervention (z=-6.20, p < .0001). The interaction between speaker and utterance quality was not signi cant post-intervention (z=-1.36, p = .17).
After adding resident comorbidities and dementia stage, the main effects of speaker (z = 18.68, p < .0001) and utterance quality (z = 2.20, p = .03), and the three-way interaction between speaker, utterance quality, and intervention (z=-3.82, p < .001) remained signi cant. The effect of resident dementia stage was signi cant (z=-2.66, p = .008), suggesting that as resident dementia stage progresses from moderately severe into severe, both residents and staff are likely to produce shorter utterances. The effect of resident comorbidities was not signi cant (z = 1.75, p = .08). Addressing The Partner By Their Name The model revealed a signi cant main effect of speaker (z = 8.14, p < .0001), indicating staff addressed residents by their name more often than residents addressing staff by their name during mealtimes ( Table 4

Discussion
This study described staff-resident language characteristics (i.e., expression length, addressing the partner by their name) during mealtime interactions and examined their associations with speaker, utterance quality, staff reception of dementia communication intervention, and resident characteristics (i.e., dementia stage, comorbidities). While staff predominated dyadic conversations, residents were also The study showed that staff and resident expression length was associated with resident dementia stage. This is consistent with prior work that residents with neurodegenerative deterioration produce more frequent, easier words and shorter, simpler sentences (10,12). Residents with dementia often experience communicational and discourse breakdowns due to cognitive and linguistic impairments, and are unable to adjust their expressions depending on the communicative contexts (26). They are less likely to understand what information and which level of detail is appropriate to convey to their staff partners, and are only able to produce short, simple, and possibly repeated words due to their word retrieval di culties and conversational ine ciency. Further, residents demonstrate progressive declines at both basic and complex levels of language (e.g., word, phrases, sentences, grammar) as their dementia stage progresses, such as di culties with naming and verbal uency, reduced phrase length, impaired phrase repetition, and reduced sentence generation and construction (27). For example, residents with severe dementia may only be able to speak approximately a half-dozen intelligible different words or fewer, or repeatedly use a single intelligible word over a day, a conversation, or a care interaction (19).
The study showed that staff expressions became shorter and simpler to accommodate residents' decline in understanding and mastery of language at the stage of severe (vs. moderately severe) dementia. Staff as cognitively intact individuals were able to adjust expressions based on their partners' needs in social interactions, such as using shorter, simpler phrases in communicating with residents with dementia (26). This is consistent with prior reports that conversational supports targeting care activities and the resident partner such as using repeated, continuing verbal cues are useful and effective strategies in managing mealtime challenges (28) and improving eating performance (29,30) in residents. A recent review also suggested that respect of resident care needs and communication ability and the use of a exible and adapted communication approach matching resident language ability are important factors associated with communication improvement between nursing staff and people with dementia (9). Therefore, besides accommodating residents' declining linguistic abilities, staff should be aware of residents' remaining capabilities and strengths in communication and provide linguistically stimulating environments that can facilitate implicit and effortless learning among residents during social events (e.g., mealtimes) to assist with their linguistic abilities.
Staff named residents more often as resident progresses from moderately severe to severe dementia stage. Addressing residents by their name during dyadic communication is a critical strategy of personcentered care to acknowledge resident identity, show respect, and establish emotional/personal connection, as well as to engage residents in activities (31). While addressing residents by their name in dyadic communication has been a highly recommended, simple, resident-centered care strategy (32), our ndings indicated less than 20% of staff utterances called their resident partner's name during mealtimes, indicating urgent needs for improvement of the use of this strategy in practice. One possible explanation why addressing the partner by name was not associated with utterance quality (i.e., quality of verbal communication) might be due to this low number of utterances involving calling a partner by name. Prior work has also reported mixed ndings on the associations between addressing resident by their name and communication (33,34), future research needs to examine their associations in larger, diverse samples.
Staff intervention to avoid elderspeak was not associated with staff-resident language characteristics in this study, possibly because the dementia communication training tested in the parent trial focused on reducing elderspeak (i.e., babytalk to older adults) by staff when communicating with residents during care activities in general, not necessarily focusing on other communication approaches or activities during mealtimes (18). A similar analysis based on communication studies that speci cally focused on improving mealtime interactions might yield different ndings and could be considered for future work. In addition, resident comorbidities were not associated with staff-resident language characteristics in this study. However, ndings were partially consistent with prior work that reported mixed ndings on the associations between communication and resident comorbidities (i.e., negative and no associations) (35). Future examination of the effect of comorbidities is needed.
The sample in the parent clinical trial focused on residents with staff-reported resistiveness to care during daily activities, which is a population that may require additional attention in dyadic communication.
Resident resistive behaviors are considered a way to communicate their needs, preferences, and wants and maybe the only way of communication for residents who cannot verbalize or have lost their language ability. Recent work showed that the use of person-centered verbal cues was associated with increased food intake among residents who were compliant, and were associated with decreased intake among those with resistive behaviors during mealtimes (8). Residents showing resistiveness usually indicate dissatisfaction with the provided care or food, and may require additional support from staff beyond simple cues and calling their names (8). While confronting restiveness to care from residents, staff reported experiences of discomfort as well as re ections on their own attitudes and behaviors as well as approaches that may help them manage and eventually reduce their discomfort (36). Meanwhile, staff reported the use of strategies, including reconceptualizing and understanding the meaning and underlying reasons for resistiveness to care, stepping back for a while to reduce the tension, accepting resistiveness to care as a way to communicate needs rather than disrupting mealtimes, and providing continuous support or reapproaching the resident at a later time as appropriate (36). This study did not consider the role of resistiveness to care because all residents had staff-reported resistiveness to care, and future research may consider examining the impact of resistive to care on language characteristics.

Limitation
The video sample captured primarily segments of meals (vs. full meals) and 1:1 (vs. 1:2, 2:1, or other complex) interactions. Future research may examine staff-resident language characteristics using fullmeal observations that capture varied dynamic complexity of mealtime interactions. Videotaped observations collected from pre-and post-intervention were used in this study. While measures (e.g., the photographer and videotaping were present at the dining area but no video was taped to improve familiarity with participants) have been taken to minimize the effect of videotaping on staff-resident interactions, staff may perform differently from usual care due to social desirability. This study primarily focused on the analysis of word-level language characteristics and future work may expand the analysis to other levels of language (e.g., syntactic or/and discourse level) using more diverse samples. Staff participants were primarily direct care providers and resident participants had exclusively moderatelysevere to severe dementia with staff-reported resistiveness to care in NHs. Therefore, ndings have limited generalizability to other staff-resident populations in other care settings.

Conclusion
Quality dyadic communication is crucial to promote care quality as well as resident behaviors, function, hydration, and nutrition during mealtimes. This study provided preliminary evidence on the associations of staff-resident language characteristics with utterance quality and resident dementia stage. Residents with dementia have limited language ability which in uences the complexity and length of conversations they can initiate and understand. Our ndings indicate there is room to increase the use of personcentered care strategies including simple, short expressions and addressing residents by name during mealtime care practice of people with dementia. Future work needs to examine staff-resident language characteristics at the word and other levels using larger, diverse samples such as full-meal observations.

Declarations
Ethics approval and consent to participate: The parent trial and this study were approved by Institutional Review Boards at the University of Kansas Medical Center and the University of Iowa.

Consent for publication: Not applicable
Availability of data and materials: Video observations are identi able data and will not be open to public due to privacy/ethical reasons. Non-identi able data including coded data from videos that support the ndings of this study are available from the corresponding author upon reasonable request.
Competing interests: No con ict of interest has been declared by the authors. Authors' contributions: WL and SOY contributed to study conceptualization, data analysis plans, interpretation of ndings, and writing and revision of the manuscript. WL contributed to video screening and coding, and exporting/managing data for analysis. SOY contributed to data cleaning and analysis. YLJ and AP contributed to revision of the manuscript. All authors meet the criteria for authorship and have approved the nal draft submitted. All those entitled to authorship are listed as authors.

Figure 1
Expression length as a function of Speaker, Intervention and Utterance Quality.
Note: Error bars indicate the standard errors of the mean.