Intervention description {11a}
The intervention, i.e., the training of nursing, social work, administrative, and other staff on R-REM is conducted in three separate sessions: (1) recognition and risk factors, (2) management, and (3) implementation of guidelines. The trainer who administers the three modules is an experienced doctoral-level, adult education professional who participated in the development and modification of the modules; she has extensive experience in staff training. Modules are provided to facilities randomized to the intervention after the completion of baseline data collection and to those randomized to usual care after completion of 12-month follow-up data collection.
Description of module 1: recognizing R-REM
Module 1 covers the extent of R-REM which includes evidence; risk factors associated with the victims, perpetrators, and environment; and the role of cognitive impairment. Different forms of mistreatment are covered, including physical, psychological, sexual, and theft. This module is delivered in the form of an experiential half-hour in-service training, plus pre- and post-tests designed to be conducted at the ALRs.
Description of module 2: management of R-REM
A.
Introduction and review of previous session and pretest
B.
Film on management of elder mistreatment
C.
A presentation of the SEARCH (Support, Evaluate, Act, Report, Care Plan, Help to Avoid) approach to R-REM management; review, lessons learned; post-test.
The 25-min film for this module was designed by the research team and directed and produced by the New York University Department of Media Production. It was narrated by distinguished journalist, Charles Osgood, and includes a discussion of what constitutes putative evidence of serious abuse, such as bruises, cuts, or more serious injuries (broken bones or cracked ribs). Mistreatment such as verbal aggression and threats, sexual harassment, and missing belongings are discussed. Three skits by professional actors are presented: skit 1, most obvious form of elder mistreatment: physical assault; skit 2, less obvious form of elder mistreatment: verbal insult; and skit 3, subtle form of elder mistreatment: psychological abuse, e.g., wandering uninvited into another’s room and rummaging through another resident’s property.
Each skit is followed by an example of a poor staff response to the event as well as a better practice, and by commentaries by leading multidisciplinary experts in elder abuse, representing different perspectives: psycho-social, medical, nursing, administrator/legal.
The final component of the video is a review of nine steps to manage and curb R-REM.
Description of module 3: implementation of best practices related to R-REM
1.
Introduction and review of previous session;
2.
Presentation of implementation methods and forms (The R-REM Behavior Recognition and Documentation Sheet -- BRDS); discussion of methods for completion;
3.
Presentation of filmed vignettes for practice in the completion of the BRDS;
4.
Review practice sheets and lessons learned;
5.
Review of implementation guidelines.
The focus of this session is on the intervention fidelity and implementation measures, including implementation of reporting guidelines. The training includes video vignettes that are rated and reviewed to confirm skills. Ways to enhance positive group relationships and the use of community to counteract individual acts of mistreatment are discussed, addressing the question, how can staff work together to structure the social and physical environment to mitigate R-REM? The importance and rules for reporting R-REM are reviewed.
Sample size {14}
Power calculations are provided for the primary distal outcome: falls, including accidents and injuries, also the outcome requiring the largest sample size. Upstate, it is expected that an average of 50 residents per site will be selected for a total of 300. Downstate, the facilities are larger, and it is expected that the average size will be 125 residents or 750 total. The proposed sample size is 6 facilities and 525 residents per arm.
Using the canonical link (Logit), the generalized linear model (GLIMMIX):η = log(pijk/(1 − pijk)) = β0 + β1Xijk + FUjk, where FUjk is a random effect associated with facility and unit.
The formula for the sample size per group using the method of Diggle is:
\( {m}^{\ast }={\left({z}_{\alpha}\sqrt{2\overline{P}\overline{Q}}+{z}_{\beta}\sqrt{p_0{q}_0+{p}_1{q}_1}\right)}^2\left(1+\left(n-1\right)\rho \right)/\left(n{\left({p}_1-{p}_0\right)}^2\right) \); and using the GEE method:
\( {m}^{\ast }=\frac{{\left({Z}_{1-\alpha /2}+{Z}_{1-\beta}\right)}^2\left({\pi}_1{p}_0\left(1-{p}_0\right)+{\pi}_0{p}_1\left(1-{p}_1\right)\right)\left(1+\left(n-1\right)\rho \right)}{2n{\pi}_0{\pi}_1{\left({p}_0-{p}_1\right)}^2} \). The formula below is adjusted with variance inflation factor (
Vif) and reliability (
Rel):
m = Vifm*/
Rel and
$$ {V}_{\mathrm{if}}=\left(1+\left({n}_e-1\right){\rho}_e\right)\left(1+\left({n}_s-1\right){\rho}_s{n}_e{p}_e/\left(1+\left({n}_e-1\right){\rho}_e\right)\right) $$
where
ns = 5 is the average number of units in the facility,
ρs = 0.015 is the ICC for facility,
ne = 17 is the average number of residents in the unit,
ρe = 0.03 is the ICC for unit, and the reliability is
Rel = 0.95.
The following table assumes
α = 0.05, power (1 − β) = 80%,
R = 0.95, and
Vif = 2.5 (with n
e = 17, ICC
Unit = 0.03,
Ns = 5, ICC
Facility = 0.015, the fall rate at follow-up:
p0 = 33% and
p1 = 20.75%). Two scenarios for
ρ (the average correlation of the outcomes over waves) were posited:
Group | Fall rate P(y = 1) | Diggle method | GEE method |
Baseline rate (%) | 6 months (%) | 12 months (%) | Combined 6 months and 12 months (%) | M (ρ = 0.5) | M (ρ = 0.6) | M (ρ = 0.5) | M (ρ = 0.6) |
Inter-vention | 37 | 22 | 19.5 | 20.75 | 404 | 431 | 399 | 425 |
Usual care | 37 | 34 | 32 | 33 | 404 | 431 | 399 | 425 |
As shown, with 525 per group, power is adequate for intent-to-treat (ITT) analyses, including all respondents. Because randomization is at the facility rather than individual level, there is the potential for imbalance on baseline variables and missing covariate data. Assuming attrition of 15% at wave 1, it is estimated that 445 per group will have at least baseline and one additional wave of data included, using the EM missing data algorithm; thus, power is adequate to detect the posited difference in fall reduction of about 12% even with attrition.
Although we did not include power calculations for the process level 2 and level 3 evaluation outcomes in the protocol, a brief summary of these analyses is provided. As shown, the sample size requirement for these outcomes is less than that of the primary distal outcome of falls, including injuries for which the power calculations were provided in the study protocol.
With respect to the level 2 knowledge outcome, in an earlier study [
7] of 270 to 340 staff, we performed a paired
t test using a mixed model analyses adjusting for clustering within facilities and covariates as needed. The effect size across primary and sensitivity analyses (which were highly significant;
p < 0.001) was between − 0.696 and − 0.964, indicating that a minimally detectable effect size was a one-point or less improvement on a 10-item knowledge test. In the current study, we calculated power for the knowledge test under reasonable scenarios observed in previous studies. For
ρ = 0.60 (the correlation between pre and post-tests), a small effect size (Cohen’s
d = 0.20, 0.24 points assuming
σ = 1.2, or 0.30 points assuming
σ = 1.5) is detectable. As shown, a very small effect size (less than one point on a 10-point knowledge test) is detectable, given the anticipated sample size of staff.
With respect to the level 3 process evaluation outcome of increased reporting, in a previous study [
7] in another setting, it was observed that the intervention group reported significantly more incidents after implementation of the intervention than the usual care group. A Poisson regression (generalized multivariate linear model; GML) with a log link will be used to model the incident event counts in the current study. Our previous analyses showed an annual prevalence of R-REM of 25%. Assuming that the usual care and intervention group start at about the same level, power was calculated to detect clinically important differential rates of R-REM reporting. The assumptions were as follows:
α = 0.05, 1-β = 0.80,
R2 = 0.16 (adjusted for multivariate covariates),
Re (reliability) = 0.95, and variance inflation factor (
VIF) = 1.18. The results show that given the proposed sample size, it is possible to detect differential R-REM reporting incidence rates of
λ0 = 0.25 in the usual care group and
λ1 (intervention group) = 0.40 for the GML method and 0.385 for sensitivity analyses using the exact score method. An incidence rate of (
λ0) = 0.25 translates to
π0 = 1−e
−λ0 = 22.1% (the proportion reporting R-REM in the usual care group), and (
λ1) = 0.40 translates to
π1 = 1−e
−λ1 = 33.0% (the proportion of R-REM reports among the intervention group). With a total
N = 800 (400 patients per group), power is 80% to detect differences in reporting as small as 10 to 11% with a two-sided test (
α = 0.05), adjusted for multivariate covariates, unreliability, and clustering. This difference is smaller than that observed in previous studies.