The NOCIP (clinical trials.gov NCT01306695; PI: Luchsinger) is a RCT comparing an enhanced NYUCI intervention that includes a CHW case management component with a CHW case management intervention alone. The RCT is being conducted in a sample of 160 Hispanic family caregivers (80/arm) of persons with dementia mainly residing in the community of Northern Manhattan. The follow-up period is 6 months. Our study is approved by the Institutional Review Board of Columbia University Medical Center in New York City. Following consent, determination of eligibility and completion of baseline measures, the coordinator alerts the Data Coordinating Center (DCC) electronically either via encrypted email or data uploads to a secure server. Respondents are randomised to treatment or active placebo groups. The randomisation algorithm accommodates rolling enrolment, and the results are checked periodically for balance. All study personnel are fluent in English and Spanish. The study coordinator screens participants and gives eligibility information to the DCC at the Research Division of the Hebrew Home at Riverdale. The DCC provides the counsellor administering the NYUCI the identification numbers of participants randomised to the intervention. The study coordinator and the CHWs are blind to this randomisation process. Each participant is randomised to one of the two study CHWs, in order to maintain a balance of CHWs in the two study arms.
All study participants are caring for a family member (spouse, parent, sibling or other family member) with a clinical diagnosis of dementia. All people with dementia must be living at home when the caregiver enrols in the study.
The eligibility criteria for the study are the following:
- Ethnicity: caregiver must be Hispanic.
- Living arrangements: respondent is the spouse or is otherwise related to the care recipient and is a caregiver of the patient with dementia (although he/she does not have to live with the recipient).
- Care receiver must have a diagnosis of dementia.
- Care receiver must not have had a stroke with hemiparesis or any other motor impairment.
- Care receiver is not confined to a wheelchair.
- Care receiver does not suffer from Parkinson's disease.
- Care receiver does not suffer from any other disorder that severely limits movement.
- Caregiver does not have impaired speech.
- Caregiver is physically able to provide care.
- Caregiver does not have an exclusionary psychiatric diagnosis (depression with psychosis). Caregivers with clinical depression or other serious mental illness will be referred elsewhere for mental health treatment.
- Respondent will be in the area for next 7 months (vacation of <4 weeks is ok).
- The person with dementia or the caregiver has to be in contact with at least one relative or close friend living in the New York City metropolitan area.
- Hearing is sufficient to allow for communication.
Rationale for eligibility criteria of participants
NOCIP was originally planned for Hispanic spouse caregivers of persons with dementia, as in the original efficacy trials of the NYUCI. However, we found early on in our recruitment effort that most Hispanic caregivers in New York City were not spouses. Some spouses were informal and included common-law partners, and divorced spouses taking care of their ex-spouses. In addition, caregivers included adult children, nephews/nieces and siblings. Thus, following the CER principle of testing interventions in the “real world”, we modified our inclusion criteria to include any relative caregiver with authorisation from the funding agency. At the time of submission of this article, of 93 caregivers enrolled in the trial, most were women (87.1%) and included 37 wives/partners (39.7%), 41 daughters (44.0%), 2 sisters (2.1%) and 1 niece (1%).
In addition to targeting all relative caregivers compared to only spouse caregivers, our project has other important differences compared to the original studies of the NYUCI. The original NYUCI targeted caregivers of spouses with AD. Our study targets caregivers of persons with dementia of any type as long as the patient does not have a significant motor deficit (eg, hemiparesis from a stroke). This makes sense because it is increasingly accepted that dementia is more heterogeneous than previously thought34
and that boundaries between dementia subtypes (eg, AD vs vascular dementia) are arbitrary. The rationale for excluding persons with motor deficits is that these deficits represent an additional burden not targeted by the NYUCI.
The sampling frame for recruitment is self-identified Hispanic caregivers of persons with dementia in New York City. Our recruitment methods include:
- Promoting the study in the local memory disorders clinic.
- Promoting the study among physicians taking care of elderly persons in the Ambulatory Care Network of New York Presbyterian Hospital.
- Participation in health fairs and community talks.
- Promotion at caregiver support groups and senior centres.
- Mailing postcards promoting the study with assistance of a marketing company targeting households in Northern Manhattan with Hispanics aged 40 years and older.
- Mailing households on mailing lists of organisations of dementia caregivers in New York City.
Consideration of recruitment strategies
We considered several recruitment strategies. The easiest way to recruit for this study and our preferred approach would be to identify persons with dementia from administrative datasets with inpatient and outpatient information, including ICD-9 codes for dementia and information on emergency contacts and next of kin. However, this approach was not approved by the hospital centre because of several concerns. First, neither the person with dementia nor the caregiver had given consent to access their administrative data for research purposes. Second, the persons with dementia would not have the capacity to provide such consent and could not be consulted. Third, there would be no assurance that the listed next of kin or other caregiver contacted was actually aware of the diagnosis of dementia; this situation could cause harm if the investigators contacted family members who were unaware of the diagnosis of dementia. Thus, we have resorted to the strategies described in the methods section, including recruitment in the community (particularly senior centres and caregiver programmes), through outreach in medical services, including general medicine, geriatric practices, psychiatry practices and memory disorders clinics and through targeted talks to caregivers groups.
Summary of interventions
Both the NYUCI and the CHW intervention are carried out by bilingual personnel. The NYUCI is carried out by a counsellor with a Masters in Social Work (MSW) degree who has experience in dementia and caregiving issues. The CHW have at least a 2-year college degree (eg, health education), and are trained at a community-based organisation named Alianza Dominicana, Inc, following a curriculum for CHWs that has been previously established. All study visits and those for the interventions are carried out in the participants’ homes or place of preference. In the case of the intervention arm, in which participants receive both the CHW intervention and NYUCI, these interventions are carried out independently. There is no communication between the NYUCI counsellor and the CHW, and their visits do not coincide. The NYUCI is an active counselling intervention that targets specific issues related to caregiving for persons with dementia. The CHW intervention targets general well-being and provides passive information about resources for caregivers of persons with dementia.
Overview of the NYU caregiver intervention
The first component consists of two individual and four family counselling sessions that include relatives suggested by the caregiver. The content of these sessions is determined by the needs of each caregiver and other participating family members (eg, learning techniques for management of troublesome patient behaviour, and promoting communication among family members). These sessions last between 1 and 1.5 h. The second component of the intervention is participation in a support group to provide the caregiver with continuous emotional support and education. The third component of the treatment is ‘ad hoc’ counselling—the continuous availability of counsellors to caregivers and families to help them deal with crises and with the changing nature and severity of their relatives’ symptoms over the course of the disease. The emergence of new psychiatric and behavioural problems of patients, which are generally more stressful than the need for assistance with activities of daily living or physical limitations, often precipitate ad hoc calls from caregivers. Ad hoc counselling makes it possible for caregivers and families to determine the amount of contact they have with the counsellors beyond the scheduled structured sessions.
The NYUCI is being administered by a bilingual (English, Spanish) counsellor with an MSW degree and has experience with dementia.
Overview of the CHW intervention
The CHW intervention consists of two visits in month 1, followed by monthly visits until month 6. The main role of the CHW is to provide access to existing education and referral resources about dementia and caregiving. In addition, the CHW assesses other health and social issues and provides information on existing resources in New York City. The CHW carries a smartphone with real-time access to email, text, the internet and telephone. CHWs also provide participants with their telephone number and email address for ad-hoc contacts.
The CHWs are based at Alianza Dominicana. In summary, the CHWs provide existing written information from the New York City chapter of the Alzheimer's Association, medical resources related to dementia at local medical centres and the community of Northern Manhattan, senior centres, and support groups. This information is given in the first visit in a manual written for this study and reinforced in all subsequent visits and on an ad hoc basis.
During the second through fourth home visit the CHW deals with barriers and goal setting. On the basis of patient needs and preferences the CHW assists the caregivers in developing an individualised plan towards maintaining their health and well-being. The CHW orients the participant in the principles of self-management35
and engages the participant in a problem-solving process to: Set priorities for immediate problem resolution; set personal goals; develop a plan to accomplish those goals, and review results and revise the plan as needed. While the focus of the research intervention is on dementia education, the CHWs address these issues from an overall health and general well-being approach. This is consistent with the CHW model that takes a patient centred approach and not a narrow disease-specific focus.
On the basis of the home assessment, CHW help with referrals to community-based resources that may be social or medically based, for example, accessing housing, public assistance, health insurance, immigration-related issues, day care (for children or elderly parents), services for domestic violence, etc, with the goal of eliminating immediate needs so the individual could make her/his health a priority. A strength of the community-based CHW model of service delivery is that through their training and exposure, CHWs are well aware of the existing programmes in the community and are able to play an active role in addressing these issues by serving as a point of contact for community-based resources.
For the purpose of this translation of the NYUCI, the primary outcomes are depressive symptoms, measured with the Geriatric Depression Scale (GDS)37
and burden, measured with the Zarit Caregiver Burden Scale (ZCBS).38
Secondary outcomes include caregiver health, measured with the caregiver physical health form39
Revised Memory and Problem Behavior Checklist,40
the Stokes Social Network scale41
and an assessment of the severity of patient dementia with the Global Deterioration Scale.42
Additionally, several Patient Reported Outcomes Measurement and Information System (PROMIS) (NIHPROMIS.org) measures are included measuring the domains of physical functioning, depression and fatigue. These outcome variables are collected at baseline and 6-month visits by a bilingual study coordinator who is blind to group assignment.
Rationale for outcome variables
We considered several measures of depression and stress as our primary outcomes. The selection of the GDS as the primary measure of depression was based on the use of this measure in the original study of the NYUCI, and findings from the literature related to differential item functioning (DIF) in measures of depression. Although there are few studies of DIF among Latino samples, our review43
of DIF in depression measures showed that many Center for Epidemiologic Studies Depression Scale (CES-D) items were biased for ethnically diverse groups. Thus, we are using the GDS, which although also containing items with DIF, have fewer such items, and less with somatic content. The latter have been found to be problematic with older individuals with comorbidities;44
such individuals will likely comprise a large part of the caregiver sample. Additionally, we are using the short-form depression measure from the National Institutes of Health PROMIS item bank45
as a secondary outcome. This measure has been found to be relatively DIF-free in the limited studies conducted. It has not been tested for DIF among Latino elderly, nonetheless, because of its primacy in future studies of depression, we will include it as an exploratory endpoint measure. The ZCBS was chosen because it has been shown to be a good measure of dementia caregiver burden and caregiver collapse.47
The ZCBS has been shown to improve in Hispanic communities in South Florida with the Resources for Enhancing Alzheimer's Caregiver Health (REACH), another intervention for dementia caregivers,48–50
and has also been used in studies of the NYUCI.22
Since an objective of the NYUCI and our CHW intervention is to alleviate caregiver burden we chose to include a measure of caregiver burden previously used in Hispanics as a co-primary outcome.
Statistical plan and data analysis
Sample size and power analysis Assumptions
The sample size calculations are based on the number of subjects needed to provide adequate power to test the primary hypothesis related to group differences in depression and burden at 6-month follow-up. The primary power calculations assume separate analyses of burden and depression measures; however, the use of MANOVA (multivariate analysis of variance) to perform a simultaneous test was also considered because it is generally more powerful, and makes use of more information. In addition, although full information likelihood estimation procedures will be used for the primary analyses, thereby allowing us to include participants who do not complete the follow-up assessment (on an intent-to-treat (ITT) basis), the power calculations include scenarios in which there is loss to follow-up as large as 20%. Based on the trial data extant, heterogeneous variances are not expected; however, this possibility was considered.
Effect sizes (Cohen's d51) for depression and burden
Studies of caregivers have used different depression measures; for example, the REACH study48
used the CES-D,52
whereas the NYUCI study used the GDS.37
. Both studies used the Zarit Burden Interview (ZBI).38
For the ZBI, we used the estimates from REACH because they included a sample of Hispanic caregivers; for the depression measure we used estimates from the GDS provided from the NYUCI study. Based on these studies, the following data were used for estimation: the baseline ZBI SDs in the NYU study were 9.46 and 10.86 in the treatment and enhanced care groups, respectively. In the Hispanic REACH sample, the estimates of the parameters are as follows:μT1
(treatment group mean and (SD)=16.9 (9.6); μT2
(6-month follow-up)=4.9 (9.1)μC1
(control group)=17.4 (9.9); μC2
=15.9 (9.9). The estimate used for α (reliability) was 0.85 and for r (correlation between times 1 and 2) between 0.50 and 0.70. The estimate used for σ (pooled SD) was ≈9.8. Because the SD for both the ZBI and the GDS in these studies was almost the same, we focused our power calculations on ZBI, realising that most apply equally to the GDS. A SAS macro was used for power calculations.
Power for endpoint analyses
Although the primary analyses proposed a full information mixed model approach, to be conservative, the power calculations were also examined based on a two-group comparison of endpoint means (differences in means), with possible attrition. It was assumed that because of randomisation that there would be no need for baseline adjustment. The following assumptions were made: σ=9.8, α=0.05, R=0.85 (reliability), g=2 (groups). Assuming power of 0.80, with 80 per group (160 total) we would be able to detect a moderate effect size (Cohen's d=0.48)—equivalent to about 4.71 points on the ZBI or GDS—about a 5-point endpoint mean difference between groups. Sample size requirements were also examined for the detection of other endpoint differences: 4.0, 4.5 points and 5 points. Sample sizes to detect this range of effects are: 111, 88 and 71, respectively. Also examined were different scenarios regarding correlations between baseline and 6-month follow-up outcome measures. The following formula from Fleiss (pp. 4–5)53
was modified to include different scenarios related to correlations between the two waves of data:
adjusting for unreliability:
Assuming a sample size of 80 per group, and r (correlations between waves of data)=0.5, 0.6 and 0.7, the resulting estimates of effect sizes are δ=4.71, 4.21 and 3.65, thus demonstrating that a medium effect size (Cohen's d=0.37 to 0.48) could be detected with this sample size.
Power for longitudinal multivariate analyses
Assuming that the outcomes are correlated, power for MANCOVA was performed, taking into account possible baseline differences (using the observed means from the Hispanic REACH study) and adjusting for unreliability. We also modelled different correlations between the first and second waves of data.
and p is a proportion of the combined populations in either of the populations (p=0.5 for equal size in the intervention and control groups) (Cohen, pp. 490–493), and
(adjusted for unreliability and r). The non-centrality parameter (λ) is
(outcomes in MANOVA), and v=N−ky
−1 (N=total sample size). The effect sizes were obtained in an iterative procedure, based on the assumptions shown in .
Assumptions for power calculations and effect sizes for MANOVA
Power for MANOVA was also examined under several scenarios regarding the non-centrality parameter. The resulting λ's are shown in . The following assumptions were made: α=0.05, σ=9.8; reliability=
0.85.δ=5 point reduction in the intervention relative to the control group.
Power for multivariate analysis of variance examined under several scenarios regarding the non-centrality parameter
Power for comparing rates of change over time in response between groups
Power to detect a difference in slopes (β1A
) over the 6 months of the study was examined. The following formula provides an estimate of the required sample sizes54
, where xj
is time, measured as the duration between the first and jth occasion, j=0,1. The same assumptions as above were made. A smaller n (about 56 per group) was required. Although power is greater for evaluation of slopes over endpoint means, it is not recommended that sample sizes be less than 80, given that the power calculations in the REACH project also yielded requirements in the range of 80−100 per group.
In summary, across the methods (see ) Cohen's d ranged from 0.37 to 0.54 or between 4 and 5 points on the ZBI or GDS endpoint means—roughly equivalent to a 0.5 SD endpoint difference in means—or a moderate effect size. Conservatively, under the assumptions specified above, 80 subjects per group will provide power ≥0.80 to detect a 4−5 unit differential change in depression and burden, based on testing the time × group interaction, allowing for heterogeneous variances and serial correlations (). Even if the pooled variance is higher than assumed, medium effect sizes are still detectable. Thus, 80 subjects per arm will provide sufficient power to detect the hypothesised difference between the active control and the intervention arm of the study.
Summary of effect sizes for different approaches to power calculations (a,b)
Power for examining change over time in the outcomes of the the Zarit Caregiver Burden Scale (ZCBS) and the Geriatric Depression Scale (GDS).
Our approach to the analyses is guided by our own experience,56
and recent reviews of the relative advantages of constrained full information likelihood treatment of the outcome versus inclusion of the baseline value as a covariate in ANCOVA,57
and methods of estimation (generalised estimating equations (GEE) vs mixed random effects).58
The primary proposed analyses will use mixed random effects models, and a Full Information Maximum Likelihood (FIML) approach, with sensitivity analyses using GEE. The change from pretreatment to post-treatment values of continuous outcomes will be modelled as functions of baseline values, treatment and the interaction of baseline and treatment. Such an approach allows for possible group heterogeneity in residual variances and serial correlations that may require modelling. Based on prior analytic experience with the outcome variables, it is not expected that transformations will be necessary.
Prior to analyses, baseline values of all variables from each arm will be examined; however, no p values will be provided, and covariates are not proposed for inclusion in the main analyses of treatment effects. Depending on the severity of missing data, other modelling techniques may be used. Examination of baseline differences on key variables between completers and those lost-to-follow-up will be conducted to inform about the nature of the missing data. The ITT analyses performed using SAS PROC MIXED will permit all individuals with at least one observation to be included. Other methods of examining missing data (eg, propensity scores and multiple imputation) will be considered in sensitivity analyses.
Depending on the observed correlation between the dependent variables, MANOVA or MANCOVA (multivariate analysis of covariance) will be performed in sensitivity analyses. A significant interaction term for one of the groups would indicate that the effect of one of the treatments is different for ZBI and GDS; in that case, two treatment effects will be estimated for each outcome. If the interaction is not significant, a model with only main effects for the outcomes and treatment will be fit and the treatment effect (common for ZBI and GDS will be estimated from this model. In addition to significance testing, we will estimate the treatment effects with 95% CI. The general hypothesis is that, controlling for covariates (if needed), the vector of means will differ over time between groups. Or we can adjust each vector of means for prescore level, and test the hypothesis of equality of means for the groups using Wilks’ λ or Hotelling's T2
. More powerful methods such as a risk score test 59
may be used, depending upon whether all endpoints are affected equally or not. Bartlett's test of sphericity will inform about the degree of intercorrelation among the outcome measures in order to determine suitability of the basic MANOVA model. Using collinearity diagnostics and examination of correlations, the final covariate set will be selected. It is anticipated that there will be: ky
=2 non-redundant outcomes (depression and burden) kc
=1 exogenous baseline covariate kx
=1 dummy variable (NYU intervention). Depending on the results of the analyses of bias due to attrition or selection, other covariates may be included.
Possible attrition bias will be addressed using information from the baseline assessment. Completers and dropouts will be compared with respect to sociodemographics, baseline ZBI and GDS and other covariates. A logit model of attrition will be developed. If attrition is significantly related to one or more baseline characteristics, the predicted values from this model can be used as a covariate to adjust for differential attrition. Depending on the degree of bias, another approach is to perform propensity score analyses, in which the treatment groups are combined, and a logistic regression predicting original group membership from covariates performed. The resulting probabilities are then arrayed in quintiles, and the subjects within each quintile randomly assigned to new groups. The analyses will be re-run with the new group designations in order to determine if the effects were similar in the new analyses with groups equalised.
Under the assumption that the missing data are either Missing Completely at Random (MCAR) or Missing at Random (MAR), we will use the above-described maximum-likelihood approach to estimate treatment effects including the baseline data for these subjects in the analysis, in conjunction with the covariate to adjust for attrition bias (if necessary). Scales will be prorated for missing data, using individual imputation algorithms developed by the measurement statisticians at the DCC. Missing data are only replaced for those who are missing less than 50% of items.
Because the analysis is based on ITT, an attempt will be made to obtain post-treatment data from all participants randomised, regardless of level of attendance, thus minimising loss to 6-month follow-up. Because most programmes do not retain all participants nor do all receive the same ‘dose’ of the intervention, inclusion of participants who received only part of their targeted programme is more reflective of the real-world impact. Secondary analyses will be conducted to investigate the impact of differential participation, stratifying the participants in the treatment conditions based on their degree of participation and examining differences between strata on the outcome measures at follow-up. However, it is acknowledged that the sample sizes are small for such analyses.