Sequential Relationships of Food Intake in Nursing Home Residents with Dementia: Behavioral Analyses of Videotaped Mealtime Observations

Abstract

Aims and Objectives.

This study examined the sequential relationships of food intake and the moderating role of the characteristics of intake and resident conditions.

Background.

Nursing home residents commonly experience insufficient food intake. While multilevel factors influence intake, evidence on sequential relationships is lacking.

Design.

The study was an observational study using secondary, behavioral analyses following the STROBE Statement.

Methods.

Videotaped observations (N=160) collected from a dementia communication trial during 2011-2014 were coded using the refined Cue Utilization and Engagement in Dementia Mealtime Video-coding Scheme during 2018-2019. The 160 videos involved 27 residents living with dementia and 36 staff in 9 nursing homes. Independent variables were the state (solid intake, liquid intake, no intake) of an intake episode occurring during mealtime (current episode), eating technique (resident-initiated, staff-facilitated) used in the next episode occurring after the current episode (subsequent episode), interval between adjacent episodes, and resident comorbidities and dementia stage. The dependent variable was the state of subsequent episode.

Results.

Successful liquid and solid intake increased odds of subsequent liquid and solid intake. Comorbidities were associated with decreased odds of subsequent liquid and solid intake for staff-facilitated episodes. When liquid intake occurred, staff-facilitation decreased odds of subsequent liquid intake; longer intervals between adjacent episodes increased odds of subsequent solid intake.

Conclusion.

Food intake was strongly and sequentially associated, and such temporal relationships were dependent on characteristics of the intake process and resident conditions.

Relevance to Clinical Practice.

The study findings supported that initiating successful intake facilitates continuity of successful intake during mealtime. Behavioral interventions tailored by comorbidities that modify characteristics of the food intake process may improve food intake.

INTRODUCTION

Nursing home (NH) residents living with dementia (residents) are at high risk of inadequate food intake (i.e., oral intake of solid and liquid food during mealtime) due to progressive cognitive and functional impairments and challenging mealtime behaviors (Keller et al., 2017; Mann et al., 2019). Inadequate food intake further leads to negative nutritional outcomes including dehydration, frailty, malnutrition, and weight loss which are fundamental to health and function, as well as subsequent health consequences including infection, decreased quality of life, morbidity, and mortality (Chao et al., 2021; Tamura et al., 2013). Ensuring adequate food intake is critical as a fundamental need for the increasing aging population with cognitive decline.

BACKGROUND

An intake episode (episode) is defined as the process of getting one bite of solid food or one drink of liquid food into the mouth (Liu, Batchelor, et al., 2020; Liu, Williams, et al., 2019). The intake process is a period of mealtime that starts at the beginning of the first intake episode and continues to the end of the last intake episode and may consist of one or more intake episodes. The definitions of and relationships among the key terms used in this study, including the intake process, intake episode, current episode, subsequent episode, adjacent episodes, and intake transition during mealtime are illustrated in Table 1. The intake process varies across meals and individuals, and is dynamic, complex, and interactive in nature, rather than static, unchanging, or changing in certain directions (Liu, Batchelor, et al., 2020; Liu, Tripp-Reimer, et al., 2020; Liu, Williams, et al., 2019). The dynamics of adjacent episodes that occur in time sequence during mealtime may influence the following episodes, all of which contribute to the changing flow of a specific meal. For example, characteristics of an intake episode (who initiated the episode, type of food being involved, duration of the episode, whether food was consumed) may influence characteristics of the next intake episode that occur afterwards, which further influence the following episodes within the same meal. Prior dementia care research has examined sequential relationships of staff behaviors with resident agitation (Gilmore-Bykovskyi et al., 2015) and aspiration (Gilmore-Bykovskyi & Rogus-Pulia, 2018) during mealtime, and sequential relationships of staff behaviors with resident agitation (Roth et al., 2002) and resistiveness to care (Belzil & Vézina, 2015) during hygienic care. However, little work has examined how each individual intake episode is associated with the next intake episode occurring within a meal (i.e., sequential relationships of food intake), and how characteristics of intake episodes and residents moderate these relationships.

Table 1.

Definitions of Key Terms Used in the Study

Term	Definition
Food intake	Oral intake of solid and liquid food during a meal.
Intake process	A period of mealtime that starts from the beginning of the first intake episode and continue to the end of the last intake episode and may consist of one or more intake episodes.
Intake episode	The process of getting one bite of solid food or one drink of liquid food into the mouth.
Current episode	An intake episode occurring during mealtime except the last episode.
Subsequent episode	The next episode occurring after the current episode during a meal.
Adjacent episodes	Two intake episodes that occur next to each other in adjacent time sequence during a meal (i.e., the current episode occurs first, followed by the subsequent episode as the next episode).
Intake Transition	The sequential flow between every two adjacent episodes during mealtime (i.e., from the current episode to the subsequent episode). Intake transition is the unit of analysis in the study.
Intake state	The outcome of each intake episode considering the type of food being involved in each intake episode, categorized as: intake of solid food, intake of liquid food, or no intake.
Current intake	The intake state of the current episode, categorized as: intake of solid food, intake of liquid food, or no intake.
Subsequent intake	The intake state of the subsequent episode, categorized as: intake of solid food, intake of liquid food, or no intake.

The Social Ecological Model has been used to examine the multilevel factors at intrapersonal (resident), interpersonal (staff), environmental, and policy levels that are associated with food intake (Liu, Jao, et al., 2019; Liu et al., 2021). In this study, this model was used to guide at which level moderating variables were conceptualized (i.e., personal, environmental) in the examination of the potential moderation effects. Resident eating independence and fair/high (vs. low) social engagement, staff facilitation, and quality environmental stimulation are associated with increased food intake (Liu, Jao, et al., 2019; Liu et al., 2020; Liu, Williams, et al., 2019; Morrison-Koechl et al., 2021). Existing research into the multilevel factors that influence food intake has primarily focused on associative relationships using cross-sectional samples and fails to address the interactive and dynamic nature of the intake process. This study extended dementia mealtime care research by examining the sequential relationships of food intake using time-event data at the level of intake episodes, and how personal (i.e., resident characteristics and health conditions, resident and staff involvement in the intake process) and environmental (i.e., the type of food being consumed) factors moderate these relationships.

Videotaped observations and validated tools are critical for in-depth coding of characteristics of intake episodes and the time that each intake episode occurs, empowering the investigation of the complex and interactive intake process. The refined Cue Utilization and Engagement in Dementia (CUED) Mealtime Video-Coding Scheme was developed and validated to assess characteristics of intake episodes (Part I, focus of this study), staff person-centered and task-centered approaches, and resident positive, neutral, and challenging behaviors during mealtime (Liu, Batchelor, et al., 2020; Liu & Kim, 2021). The refined CUED has demonstrated ease of use, feasibility, adequate inter-coder reliability, and good construct and predictive validity. The use of the refined CUED and videotaped observations supports the analysis to examine sequential relationships of food intake.

Objectives

This study aimed to examine 1) the sequential relationships of food intake, and 2) the moderating effect of characteristics of intake episodes (i.e., whether staff or resident initiated/completed intake episodes, whether the episodes resulted in food intake, interval between adjacent episodes) and resident conditions (i.e., dementia stage, comorbidities) on the relationships. We hypothesized intake episodes are sequentially dependent, and characteristics of intake episodes and resident conditions moderate these sequential relationships. The role of resident comorbidities and dementia stage were examined to inform individualized, behavioral strategies tailored to these conditions to optimize intake.

METHODS

Design

This was a secondary analysis of archived videotaped observations collected from a randomized clinical trial during 2011-2014. The parent study evaluated the efficacy of a staff educational intervention to overcome communication barriers, improve person-centered communication, and decrease resident resistiveness to care (Williams et al., 2016). Rather than part of the parent study, this analysis used videos collected from the parent study and a newly developed behavioral coding scheme to address newly developed research questions focusing on the temporal relationships of food intake. The report of this observational research followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement (Supplementary File 1), a standardized guideline for reporting observational studies (Von Elm et al., 2014).

Sample and Setting

In the parent study, residents were eligible if they were: 1) diagnosed with dementia based on medical records, 2) long-stay residents, 3) resistive to care based on staff report, 4) able to hear staff communication, and 5) had a surrogate decision maker who provided informed consent. Staff were eligible if they were: 1) ≥18 years old, 2) English speaking, 3) permanent employees, and 4) provided direct care for a participating resident at least twice per week over the previous month. A total of 127 staff and 83 residents from 13 NHs in Kansas, USA were enrolled. The parent study used a randomized controlled trial design with cluster randomization in which paired NHs were assigned to either the intervention or wait-list control groups. The NHs assigned to wait-list control groups later crossed over and received the intervention. Morning care interactions between staff and residents during activities of daily living, including mealtime, were videotaped at 3 time points for intervention group (i.e., baseline, post intervention1-month, and post intervention 3-month) and at 5 time points for control group (i.e., baseline, pre-intervention 1-month, pre-intervention 3-month, post intervention1-month, and post intervention 3-month) during 2011-2014.

In this study, videotaped observations that were collected in the parent study were selected. Videos were included if they: 1) captured mealtime activities, 2) lasted ≥1 minute (to ensure adequate information on the intake process and dyadic interactions), 3) captured one-on-one interactions between one primary staff and one resident (to minimize interaction complexity), and 4) had adequate video/audio quality. Videos were excluded if the resident was taking medication rather than eating a meal, was being transferred to or from the dining location, or was present in the dining location but not eating a meal.

A total of 1748 videos were screened, among which 1588 were excluded due to not capturing mealtime activities (n=1486), lasting <1 minute (n=63), involving multiple staff and/or residents (n=34), and poor quality (e.g., too dark, n=5), leaving 160 eligible videos included for this study (Figure 1). Among the 160 videos, 110 were collected under usual care conditions (pre-intervention), and 50 were collected after staff communication training (post-intervention). The 160 videos captured segments of mealtime that involved dyadic interactions, rather than whole mealtime or certain periods of mealtime (e.g., beginning, middle, or end of mealtime).

Figure 1.

Selection of Eligible Mealtime Videotaped Observations from the Parent Study

The 160 videos involved 36 staff and 27 residents (53 unique staff-resident dyads) in 9 NHs (Table 2). Staff participants had a mean age of 35.9 years (range=19-79), worked as a caregiver for a mean length of 9.5 years (range=.25-31), and worked at the current NH for a mean length of 4.0 years (range=.1-13). Most staff were female (81%), non-Hispanic (75%), White (75%), and Certified Nursing Assistants (86%). Most staff had completed or were attending college (72%). All residents were White, with a mean age of 85.6 years (range=64-104). Most residents were female (63%) and non-Hispanic (92.6%). Residents had moderately severe (70%) or severe (30%) dementia as measured by the Functional Assessment Staging in Alzheimer’s Disease Scale (total score ranges from 1, normal cognition, to 8, very severe dementia) (Sclan & Reisberg, 1992). Residents had moderate levels of comorbidities (range=19-36) as measured by the Modified Cumulative Illness Rating Scale (total score ranges from 0 to 70, higher score indicating more comorbidities) (Knoefel & Patrick, 2003).

Table 2.

Participant Characteristics

Continuous variables	N	Mean	SD	Range
Staff age (year)	36	35.86	12.38	19 - 79
Staff years worked as a caregiver	36	9.54	8.61	0.25 - 31
Staff years worked in current facility	36	4.03	3.69	0.1 - 13
Resident age (year)	26	85.6	8.64	64 - 104
Resident comorbidities1	24	27.1	5.27	19 - 36
Categorical variables	N	%
Staff gender
Male	7	19.4
Female	29	80.6
Staff race
White	27	75
African American	9	25
Staff ethnicity
Non-Hispanic	27	75
Hispanic	9	25
Staff education
High School	10	27.8
College	26	72.2
Staff job title
Certified Nursing Assistants	30	85.7
Registered Nurse	2	5.7
Licensed Practical Nurse	3	8.6
Resident gender
Male	10	37.0
Female	17	63.0
Resident race
White	27	100
Resident ethnicity
Non-Hispanic	25	92.6
Hispanic	2	7.4
Resident dementia stage
Moderately severe dementia	14	70
Severe dementia	6	30

Note. The 160 videos involved N=36 staff and N=27 residents (N=53 unique staff-resident dyads) in 9 nursing homes. The counts in some categories do not add up to total due to missing values.

¹Residents level of comorbidities as measured by the Modified Cumulative Illness Rating Scale (total score ranges from 0 to 70, higher score indicating more comorbidities).

Data Collection

Part I of the refined CUED was used to code four characteristics of each intake episode: 1) the duration based on the starting time point (when the staff or resident began to move food from a tray or container to the mouth area using a utensil or hand) and ending time point (when the utensil or hand was taken away from the mouth area whether food entered the mouth or not), 2) eating technique indicating the person who initiated/completed each episode (resident-completed, staff-facilitated), 3) the type of food being consumed (solid food including regular solid food and texture-modified food such as pureed, liquid food including regular drinks and texture-modified liquid such as thickened liquid), and 4) the outcome (intake, no intake) (Supplementary Table 1). Videotaped observations were each coded second-by-second by one of four trained coders during 2018-2019 using Noldus Observer^® 14.0 (Noldus Information Technology Inc., Leesburg, VA, USA). Noldus Observer^® 14.0 is a software for behavioral research that supports coding and visualizing behaviors on a timeline, unravels the sequence of events, and integrates different data modalities into a complete set. The coders were trained by the first author through coding gold standard videos following a standard training and coding manual. Inter-coder reliability of Part I of the refined CUED was established among the four trained coders using 22 randomly selected videos of the sample (Cohen’s Kappa range=.93-.99, 95% CI=.92-.99, ± 1s tolerance) (Liu, Batchelor, et al., 2020).

Ethical Considerations

Ethical approvals were obtained through Institutional Review Boards (IRBs) of universities where the parent study and this secondary analysis (IRB ID#: 201208797, approved on November 12^th, 2018) were conducted. In the parent study, eligible staff and resident participants were screened and informed of study procedures. Eligible staff participants were enrolled with written informed consent. All resident participants who were eligible for the parent study had moderate-to-severe dementia and were considered unable to make own informed decisions. Therefore, written consent from residents’ surrogate decision makers as well as written assent (or oral content when residents were unable to write/sign) from residents were obtained to enroll residents. In addition, oral assent from staff and residents were obtained before each video recording session.

Data analysis

Coded data representing characteristics of intake episodes were exported from Noldus Observer^® to Excel worksheets and then to R 4.0.5 (R Core Team, 2021). Mixed Multinomial Logit Models were fitted using the R package mclogit to examine the sequential relationships of food intake. The multinomial logit model provides a quantitative relationship between independent variables and the categorical dependent variable with three or more categories. The mixed model approach accounts for the correlation and clustering effects within dyads.

In this study, the dependent variable was the intake state of the subsequent episode, a variable with three categories (i.e., intake of solid food, intake of liquid food, no intake). The intake transition between adjacent episodes was the sequential flow from the current episode to the subsequent episode during mealtime. The five independent variables were 1) the intake state (i.e., intake of solid food, intake of liquid food, no intake) of the current episode (i.e., any intake episode occurring during mealtime except the last episode), 2) eating technique (i.e., resident-initiated, staff-facilitated) used in the subsequent episode (i.e., the next episode occurring after the current episode), 3) the interval between adjacent episodes (the time period between the beginning of the current episode and the beginning of the subsequent episode), 4) resident comorbidities, and 5) resident dementia stage. All intake transitions observed in each video were included in the model. For example, an observation with a total of 4 intake episodes resulted in 3 intake transitions, all of which were analyzed (Figure 2). Dyads were fit as a random effect to account for within dyad correlation due to repeating dyads.

Figure 2.

Illustration of relationships among intake process, intake episode, current episode, subsequent episode, adjacent episodes, and intake transition during mealtime using a meal with four intake episodes (three intake transitions) as an example

Two-way interactions for all five independent variables were examined. Based on the mixed multinomial logit model, the probabilities that the intake state of the subsequent episode (subsequent intake) is solid food, liquid food, or no intake can be estimated based on the values of independent variables. This model helps understand the dynamic relationships of adjacent episodes and how modifications of certain characteristics of intake episodes (e.g., eating technique, interval between episodes) influence the probabilities of subsequent intake.

Covariates included resident characteristics (i.e., age, gender), staff characteristics (i.e., age, gender, race, ethnicity, education, years of caregiving experience), and treatment time points (i.e., pre- and post- intervention). The intervention tested in the parent study focused on the importance of communication, especially reduction of elderspeak (babytalk), rather than mealtime-specific dyadic communication or strategies to modify food intake. To increase the sample size of the sofisticated models used in the study and generate more valid estimates, we used videos collected from both pre-intervention and post-intervention, and controlled for the treatment time point in the analysis. In Table 4, the effects of treatment time points were not statistically significant on either liquid food intake (p=0.639) or solid food intake (p=0.348), which further validated the combination of the pre-intervention and post-intervention data. Resident race and ethnicity were not included as covariates due to limited variability (i.e., all residents were white and only two residents were Hispanic). Missing data was not imputed. All continuous variables were centered by subtracting their means. The interval between adjacent episodes (one of the independent variables) was log transformed for a more symmetrical distribution. Level of significance was set at α=0.05.

Table 4.

The Mixed Multinomial Logit Models to Predict Intake Transitions

Characteristics	Intake of Liquid Food				Intake of Solid Food
	b (SE)	OR	95% CI for OR	p	b (SE)	OR	95% CI for OR	p
Interval between adjacent episodes	−.34 (.37)	.72	.34, 1.49	.370	−.69 (.37)	.50	.25, 1.03	.059
Eating Technique (staff-facilitated vs. resident-initiated)	−1.74 (.62)	.18	.05, .58	.005	−.41 (.61)	.67	.20, 2.19	.501
State of current episode (intake of liquid food vs. no intake)	2.67 (.65)	14.40	4.03, 51.45	<.001	1.24 (.67)	3.45	.93, 12.72	.063
State of current episode (intake of solid food vs. no intake)	1.57 (.60)	4.80	1.49, 15.43	.009	1.44 (.60)	4.23	1.31, 13.65	.016
Eating technique (staff-facilitated vs. resident-initiated) * Resident comorbidities	−.16 (.06)	.86	.75, .97	.016	−.35 (.07)	.70	.62, .80	<.001
Eating technique (staff-facilitated vs. resident-initiated) * Resident dementia stage (severe vs. moderately severe)	.27 (.57)	1.32	.43, 3.98	.628	.98 (.57)	2.68	.87, 8.21	.086
Eating technique (staff-facilitated vs. resident-initiated) * State of current episode (intake of liquid food vs. no intake)	−1.43 (.67)	.24	.07, .88	.032	−.58 (.70)	.56	.14, 2.21	.407
Eating technique (staff-facilitated vs. resident-initiated) * State of current episode (intake of solid food vs. no intake)	.64 (.70)	1.89	.48, 7.43	.361	−.87 (.71)	.42	.11, 1.68	.221
Eating technique (staff-facilitated vs. resident-initiated) * Interval between adjacent episodes	−.43 (.33)	.65	.34, 1.24	.194	.25 (.33)	1.29	.68, 2.45	.441
Interval between adjacent episodes * Resident comorbidities	.03 (.05)	1.03	.94, 1.13	.452	−.01 (.04)	.99	.91, 1.07	.723
Interval between adjacent episodes * Resident dementia stage (severe vs. moderately severe)	.09 (.35)	1.10	.55, 2.20	.789	.57 (.33)	1.77	.92, 3.40	.087
Interval between adjacent episodes * State of current episode (intake of liquid food vs. no intake)	.10 (0.39)	1.10	.51, 2.37	.800	1.05 (.38)	2.85	1.37, 5.95	.005
Interval between adjacent episodes * State of current episode (intake of solid food vs. no intake)	.36 (.37)	1.44	.70, 3.00	.325	.37 (.34)	1.44	.73, 2.83	.288
Resident comorbidities * State of current episode (intake of liquid food vs. no intake)	−.004 (.08)	1.00	.84, 1.18	.960	.06 (.08)	1.06	.91, 1.23	.481
Resident comorbidities * State of current episode (intake of solid food vs. no intake)	.01 (.09)	1.01	.85, 1.21	.896	.12 (.08)	1.13	.96, 1.32	.139
Resident dementia stage (severe vs. moderately severe) * State of current episode (intake of liquid food vs. no intake)	−.50 (.68)	.61	.16, 2.30	.465	.06 (.64)	1.06	.30, 3.75	.924
Resident dementia stage (severe vs. moderately severe) * State of current episode (intake of solid food vs. no intake)	−1.37 (.71)	.25	.06, 1.02	.053	−.59(.67)	.55	.15, 2.06	.377
Staff gender (male vs. female)	.39 (.64)	1.48	.42, 5.22	.545	.13 (.61)	.88	.27, 2.92	.838
Staff race (African American vs. White)	.77 (.58)	2.16	.69, 6.71	.185	.94 (.56)	2.57	.85, 7.74	.093
Staff ethnicity (Hispanic vs. non-Hispanic)	−.25 (.68)	.78	.20, 2.97	.712	1.19 (.63)	3.27	.96, 11.14	.058
Staff age, year	.00 (.02)	1.00	0.95, 1.05	.991	−.02 (.02)	.98	.934, 1.03	.490
Staff education (high school vs. college)	1.02 (.49)	2.76	1.05, 7.25	.039	.13 (.47)	1.14	.45, 2.86	.788
Staff years of direct caregiving	.01 (.08)	1.01	.86, 1.17	.949	.07 (.08)	1.07	.92, 1.25	.358
Resident age, year	−.01 (.05)	.99	.90, 1.08	.776	−.02 (.04)	.98	.90, 1.06	.627
Resident gender (female vs. male)	−.88(.55)	.41	.14, 1.20	.105	−.56 (.50)	.57	.22, 1.53	.265
Resident comorbidities	.10 (.09)	1.11	.93, 1.32	.250	.12 (.09)	1.13	.95, 1.34	.175
Resident dementia stage (severe vs. moderately severe)	−0.26 (.86)	.77	.14, 4.12	.761	−1.09 (.84)	.34	.07, 1.74	.193
Treatment time point (post- vs. pre- intervention)	−.14 (.29)	.87	.49, 1.54	.639	−.27 (.28)	.77	.44, 1.34	.348
Intercept	.69 (.74)	1.99	.46, 8.65	.354	0.80 (.72)	2.22	.54, 9.09	.269

Note. b = model coefficients, SE=standard error. OR= Odds Ratios. 95% CI= 95% Confidence Interval for OR. Among the 1531 transitions of intake state, there were 318 transitions with missing data on resident age (71 transitions involving 1 resident), comorbidities (218 transitions involving 3 residents), dementia stage (318 transitions involving 7 residents). Missing was not imputed in the analysis, therefore, a total of N=1213 transitions of intake state with complete data (involving N=30 staff and N=19 residents) were used in the models.

RESULTS

Characteristics of Videos, Intake Episodes, and Intake Transitions

The 160 videos had a mean duration of 4.6 minutes (range=1.03-23.8, Table 3). A total of 1691 intake episodes were coded, with a mean duration of 4.7 seconds (SD=3.45). Half of the intake episodes were initiated by residents (49.9%), and the other half were facilitated by staff (50.1%). The 1691 intake episodes resulted in a total of 1531 intake transitions between episodes (1531 intake transitions=1681 intake episodes-160 videos) with a mean interval of 22.76 seconds (range=1.37-240.4). Among the 1531 intake transitions, 46.6% transitioned from intake of solid food to intake of solid food (26.9%), intake of liquid food (14.0%), or no intake (5.7%); 38.9% transitioned from intake of liquid food to intake of solid food (13.9%), intake of liquid food (21.4%), or no intake (3.6%); and 14.5% transitioned from no intake to intake of solid food (5.7%), intake of liquid food (4.1%), or no intake (4.6%). Of the 1531 intake transitions, 65% occurred under usual care conditions (pre-intervention), and 35% occurred after staff communication training (post-intervention).

Table 3.

Characteristics of Videos, Intake Episodes, and Intake Transitions

Continuous variables	N	Mean	SD	Range
Duration of videos*, minute	160	4.57	3.89	1.03 - 23.8
Duration of intake episodes#, second	1691	4.70	3.45	0.33 - 36.4
Interval between adjacent episodes&, second	1531	22.76	21.57	1.37 - 240.4
Categorical variables	N	%
Eating technique used in the current episodes&
Resident-initiated	764	49.9
Staff-facilitated	767	50.1
Intake Transitions&
From intake of solid food to	714	46.6
Intake of solid food	412	26.9
Intake of liquid food	215	14.0
No intake	87	5.7
From intake of liquid food to	595	38.9
Intake of solid food	213	13.9
Intake of liquid food	327	21.4
No intake	55	3.6
From no intake to	222	14.5
Intake of solid food	88	5.7
Intake of liquid food	63	4.1
No intake	71	4.6
Treatment time points &
pre- intervention	995	65.0
post- intervention	536	35.0

Note. The N=160 videos involved a total of N = 1691 intake episodes and N = 1531 transitions of intake states.

^*Number of videos (N=160).

^#Number of intake episodes (N=1691).

^&Number of transitions of intake states (N=1531). Duration of videos = the time period from the beginning to the end of a video. Duration of intake episodes = the time period from the beginning to the end of the same intake episode in a video. Interval between adjacent episodes = The time period between the beginning of the current episode and the beginning of the subsequent episode. Intake Transitions = Transitions of Intake states between adjacent episodes

Sequential Relationships of Characteristics of Intake Episodes and Resident Conditions on Intake Transitions

Intake states were sequentially related (Table 4). Successful intake of liquid food was associated with increased odds of subsequent intake of liquid food (OR=14.4, 95% CI=4.03, 51.45) and solid food with marginal significance (OR=3.45, 95% CI=.93, 12.72). Successful intake of solid food was associated with increased odds of subsequent intake of liquid food (OR=4.8, 95% CI=1.49, 15.43) and solid food (OR=4.23, 95% CI=1.31, 13.65).

Interaction effects were significant for eating technique by comorbidities, eating technique by intake state of the current episode (current intake), and interval between episodes by intake state of the current episode. Increased comorbidities were associated with decreased odds of subsequent intake of liquid (vs. no intake; OR=.86, 95% CI=.75,.97) and solid food (vs. no intake; OR=.70, 95% CI=.62,.80) when the subsequent episode was staff-facilitated (vs. resident-initiated). Figure 3 shows the interaction effect for eating technique by comorbidities for subsequent intake of solid food: when the subsequent episode was staff-facilitated, increased comorbidities were associated with decreased probability of successful intake of solid food. However, such relationship was opposite for the resident-initiated subsequent episode: increased comorbidities was associated with slightly increased probability of successful intake of solid food.

Figure 3.

Illustration of interaction effects for resident comorbidities (measured by the Modified Cumulative Illness Rating Scale) by eating technique (staff-facilitated vs. resident-initiated)

When there was an intake of liquid food (vs. no intake), staff-facilitation (vs. resident-initiation) decreased odds of subsequent intake of liquid food (OR=.24, 95% CI=.07, .88) and longer intervals between adjacent episodes increased odds of subsequent intake of solid food (OR=2.85, 95% CI=1.37, 5.95). Figure 4 shows the interaction effect for interval between episodes by current intake: increased intervals between adjacent episodes were associated with increased probability of food intake when the type of food was altered (e.g., from current intake of liquid food to subsequent intake of solid food).

Figure 4.

Illustration of interaction effects for the interval between adjacent intake episodes [log(interval)] by the state of current intake episode

DISCUSSION

This is the first study that examines sequential relationships of intake episodes as well as the moderating effects of characteristics of intake episodes and resident conditions using the validated CUED coding scheme and videotaped observations. Findings support the hypotheses that food intake is sequentially dependent, and the sequential relationship is moderated by characteristics of intake episodes (i.e., intake state of the current episode, whether staff or resident initiated/completed the next intake episode, interval between adjacent episodes), as well as resident comorbidities.

Characteristics of intake episodes

Successful intake of either liquid or solid food (vs. no intake) increased the odds and probability of successful subsequent intake of both liquid and solid food. Additionally, when there was a successful intake of liquid food, the odds of subsequent intake of liquid decreased when staff facilitated the intake (vs. residents initiated); and longer intervals between intake episodes increased the odds of successful intake of solid food. Furthermore, when the type of food changed between adjacent episodes (liquid→solid, solid→liquid), longer intervals increased the probability of successful subsequent intake. Conversely, when the type of food being consumed stayed the same (solid→solid, liquid→liquid), increased intervals decreased the probability of successful subsequent intake.

This study focused on the characteristics of the intake process, including whether residents are positively engaged in initiating each food episode, whether residents are given adequate time to chew/swallow and be ready for the next episode, and whether residents are given alternate types of food. These characteristics are important targets to achieve optimal mealtime care because they represent easy-to-modifiable factors that may influence the frequency and amount of food intake during mealtime in long-term care residents. Recent work has identified multifaceted factors (i.e., resident initiating and completing food intake with support of staff continuous engagement, staff offering liquid food, quality environmental stimulation with social interaction) that influence resident eating performance (Liu et al., 2017; Liu, Williams, et al., 2019; Palese et al., 2019) and food intake (Abdelhamid et al., 2016; Liu, Jao, et al., 2019; Wen Liu et al., 2020). However, research was lacking on how food intake is sequentially related as well as the role of these characteristics on the sequential relationships of food intake. NH residents, especially those with some ability to eat by themselves, are frequently provided with full assistance during mealtime (Liu, Tripp-Reimer, et al., 2020; Liu, Williams, et al., 2019).

Multi-level behavioral interventions including education, environmental modifications, feeding assistance, and oral supplementation have been commonly reported to address mealtime challenging behaviors, eating function, dehydration, food intake, and malnutrition in people with dementia; however, evidence of the impact of these interventions has been limited and inconsistent (Borders et al., 2020; Bruno et al., 2021; Liu et al., 2014; Liu et al., 2015). Findings of this study reinforce the need for feasible, easy-to-implement, and innovative interventions focusing on initiating and maintaining successful intake flow as early as possible, supporting independence by providing continuous staff engagement, and allowing adequate time between intake episodes during mealtime. Such reinforcement may require a multilevel approach including supportive systems and team environment, and further facilitate the transition of the long-term care philosophy, practice, and policy from task-centered to person-centered, which are highly recommended for people with dementia (Caspar et al., 2020). These behavioral interventions, rather than the current care practice of providing full assistance by staff to residents and rushing residents during mealtime, have potential to ensure adequate intake.

Resident comorbidities and dementia stage

This study demonstrates that supporting independent eating is moderated by resident comorbidities. When an intake episode is facilitated by staff, increased comorbidities decrease the odds of successful subsequent intake of both liquid and solid food (vs. no intake). While prior research indicates that more comorbidities are associated with decreased eating performance (Liu et al., 2017), and that decreased eating performance reduces intake (Liu, Jao, et al., 2019), findings of this study adds to the literature by demonstrating the role of comorbidities in directly moderating the sequential relationships of food intake. Findings indicate the need for individualized and tailored behavioral strategies in mealtime care by resident comorbidities. While fewer comorbidities were associated with increased odds for successful intake, staff should focus on assisting residents with higher comorbidities, especially when staffing for mealtime care is limited.

Consistent independence in food intake was not significantly moderated by resident dementia stage. This finding is not consistent with prior reports that more severe cognitive impairment and more severe dementia stage are associated with decrease in eating performance (Liu et al., 2016; Liu et al., 2017) and pace of food intake (Liu, Jao, et al., 2019). However, a limitation of this study is lack of variation in dementia stage of resident participants, all of whom were diagnosed with moderately severe or severe dementia. Future research is needed on the role of dementia stage on food intake using larger diverse samples. Information on the role of resident conditions is important to guide the use of individualized and tailored behavioral strategies to improve intake.

Directions for Future Research

This study primarily focused on characteristics of intake episodes and resident conditions, and future research may examine additional characteristics at resident-, caregiver-, and facility-levels. For example, while solid and liquid food types were considered in this study, data on diagnosis of dysphagia, related types of special diet, taste of food, and food texture were not collected in the parent study and thus not included in this analysis. Future research may collect additional data on these resident-level and food characteristics. Additionally, facility characteristics including availabilities, types, and ratios of staffing for mealtime care such as food service, nutrition, speech-language, and direct care, should be considered in future research. This study focused on odds of food intake, and future research may examine sequential relationships of the amount of food intake and moderating factors. This study provided evidence of temporal (rather than causal) relationships of food intake. Therefore, findings of the study may not imply the causality of food intake state transitions. Findings of this study facilitate further research in which causal relationships between characteristics of intake state transitions and resident health conditions (intervention target) and food intake outcomes should be examined.

Mealtime is a complex, dynamic, and interactive process, especially when dyadic interactions are involved, and therefore requires tailored (rather than standardized) care based on individual’s preferences and needs (Li et al., 2021). A Markov Chain model, a useful approach to model a stochastic or random process of categorical variables (i.e., intake states) with sequential dependence, can be constructed to estimate probabilities of intake transitions based on findings of this study (Brémaud, 1999). In practice, the Markov Chain model can be used to evaluate and compare alternative behavioral strategies in relation to clinically relevant outcomes. Building upon this study, this model helps visualize and understand the dynamic flow of the entire intake process and how modifications of certain characteristics of intake episodes (e.g., eating technique, interval between intake episodes) influence the odds of food intake in subsequent episodes as well as the flow of the intake process. For a given behavioral strategy (e.g., staff facilitates all intake episodes or only after an unsuccessful intake episode), the corresponding transition probabilities can be used to evaluate long-run performance parameters such as the percentage of intake episodes of successful solid or liquid intake compared to no intake, or finite horizon metrics such as the average number of episodes of successful solid or liquid intake and no intake over the first 10 or 20 episodes. Such information is valuable to compare and select feasible and effective behavioral strategies that can be used by staff to modify the flow and dynamics of a meal to improve food intake.

The refined CUED has also been validated to assess staff person-centered and task-centered approaches and resident positive, neutral, and challenging behaviors from both verbal and nonverbal perspectives. Using rich data coded from videotaped observations, we will further examine the sequential relationships between staff approaches and resident behaviors, as well as the sequential relationships between food intake and staff approaches and resident behaviors. Through these investigations, we expect to identify specific person-centered and task-centered approaches that may promote resident positive, neutral, and/or challenging behaviors, which may further lead to increased or decreased food intake. This information is important to guide the development and implementation of innovative behavioral interventions (e.g., staff training on person-centered, individualized mealtime care) to improve mealtime care quality and individual outcomes.

Limitations

Videos used in this study captured part of the meals rather than whole meals and one-on-one interactions between NH staff and residents with moderate to severe dementia and resistiveness to care. Therefore, the study findings on sequential relationships of food intake may not be generalized to mealtime interactions that include more than one resident or staff (e.g., 2:1, 1:2, 2:2 interactions), residents with other diagnoses (e.g., mild dementia, stroke), or other care settings (e.g., home/community settings).

CONCLUSION

This study provided preliminary evidence that strong sequential relationships of food intake existed and were moderated by characteristics of intake episodes and resident comorbidities. Initiating successful intake as early as possible increases the chance of the continuity of successful food intake flow during mealtime. Tailored behavioral strategies by individual comorbidities that modify characteristics of intake episodes have potential to improve food intake.

RELEVANCE TO CLINICAL PRACTICE

Despite the increased prevalence, risks, and consequences of inadequate food intake, residents are not always provided with optimal mealtime care (Hammar et al., 2016; Lea et al., 2017). Direct care staff, who remain an unsupported and unprepared workforce that is critical to quality care (Booi et al., 2021), report a need for training on the use of resident - and relationship - centered, individualized care approaches integrating management of fluctuating mealtime challenging behaviors to achieve optimal care while minimizing risks of inadequate food intake through building relational and emotional connections (Douglas et al., 2021; Li et al., 2021). Effective, multi-level interventions that optimize mealtime care quality warrant fundamental components, including supportive organizational systems, foundational and interprofessional person-centered cultures, and adequate staff training (Brunner et al., 2021; Fazio et al., 2020; Villar et al., 2021). These components are critical to reconceptualize challenging behaviors as indicators of individual needs for behavioral support and facilitate the transition of care philosophy to person-centeredness and should be considered.

Adequate mealtime assistance required on average 35 to 42 minutes of staff time, compared with 5 to 9 minutes per resident per meal staff spent in usual mealtime care (Simmons et al., 2008). While the average length of the video sample used in this study (4.6±3.9 minutes) was considered inadequate assistance time, it is comparable to the actual time staff spent in mealtime care (5-9 minutes) and may to some extent represent usual NH care practice. This information emphasizes the potential generalizability of the study findings on current NH mealtime care practice through increased understanding of the sequential relationships of food intake and moderating factors, which will empower direct care staff in using feasible behavioral strategies to improve intake.

First, the knowledge that successful intake is strongly sequentially related emphasizes the importance of initiating successful intake as early as possible and maintaining the flow of successful intake for as long as possible, which increases the potential of the continuity of successful intake throughout mealtime. While staffing may be limited during mealtime care, it is critical that staff identify residents who need assistance for initiation only vs. continuing assistance, actively engage residents to initiate mealtime with tailored assistance, and maintain assistance quality to those residents in need of continuing care to ensure the flow of successful intake. As the flow of successful intake is initiated and maintained, there is increasing chance that more successful intake may follow subsequently, which will contribute to the continuity of successful intake.

Second, the knowledge that the sequential relationships were moderated by characteristics of intake episodes emphasizes the use of common, feasible behavioral strategies. These strategies include encouraging resident involvement in eating, alternating solid and liquid food choices, offering liquid food frequently, and allowing adequate time between bites or drinks. For example, whether an intake episode resulted in intake of liquid or solid food or not, residents should always be encouraged to initiate and complete the next intake episode which increases the probability of food intake. Engaging residents in eating and offering liquid food when residents struggle with solid food were associated with resident eating independence and pace of food intake (Liu, Jao, et al., 2019; Liu, Williams, et al., 2019). Staff reported the use of individualized assistance and modification (e.g., temperature, smell, and types) of food, which were the most straightforward stimuli during mealtime, to support resident mealtime independence (Liu, Tripp-Reimer, et al., 2020). Findings of this study further add to the literature of dementia mealtime care research and demonstrate that these feasible, easy to use strategies may facilitate the successful food intake flow during a specific mealtime.

Third, the study findings add to the literature and indicate the need of context-based behavioral strategies to better manage mealtime care and improve food intake. For example, following a successful intake of liquid food, the probability of a subsequent intake of liquid food is higher when resident (rather than staff) initiates it. Therefore, residents should be highly encouraged to initiate each intake episode and intake attempt of liquid food as appropriate, especially when hydration is priority, rather than being completely assisted without considering residents’ functional ability and level of engagement.

Impact Statement.

What does this paper contribute to the wider global clinical community?

• Nursing home residents commonly experience insufficient food intake. While multilevel factors influence intake, evidence on sequential relationships is lacking.

• This study showed that food intake was strongly and sequentially associated, and such temporal relationships were dependent on characteristics of the intake process and resident conditions.

• Tailored behavioral strategies that facilitate successful initiation of mealtime and modifies the characteristics of the food intake process has promise to improve food intake outcomes in people with dementia.