The study is based on a pilot intervention that evaluated the results of an individualized face-to-face AMI intervention combined with a resident-managed oral health campaign in one rent-subsided older adult building in central Connecticut [44]. The intervention was successful in improving gingival health and reducing plaque, but the relative impact of each of the intervention components could not be evaluated because they were held simultaneously. The clinical trial protocol is designed to disaggregate and compare the individual and combined effects of the face-to-face intervention and the oral health campaign using a modified fractional factorial crossover design [45]. The design allows for disaggregating a pilot study that combined 2 interventions (AMI and oral health campaign) with positive results to enable an evaluation of which of the 2 components would have the best effect. Sequencing (systematic recombining), an element of the modified fractional factorial design (MFFD) is an efficient way of determining whether one combination or another has a better immediate and long-term effect. The MFFD design avoids the cost of multiple sites and control groups. The intervention activities are linked to the conceptual domains in the adapted Fishbein model (see Figure 1).

The primary study aims are to evaluate the 2 main components of the intervention, the AMI and the oral health campaign, against each other and in different sequences and to assess the mechanisms (norms, beliefs, attitudes, intentions, and practice) through which the intervention operates at each time point.

The study hypotheses include the following: (1) the face-to-face intervention component of the overall intervention (AMI) will produce better short-term clinical outcomes and changes in mediators than the oral health campaign component; (2) the oral health campaign component followed by the individual component will produce better midterm and long-term clinical outcomes than the individual component followed by the oral health campaign; (3) the 2 sequences will result in differences in changes in the mediating cognitive domains at midterm; (4) within the face-to-face component, exposure to more mediating domains will result in better clinical outcomes; and (5) both conditions will have an equivalent effect on the secondary outcome—oral health quality of life at end point.

The study is a group randomized controlled trial (GRCT) in which 6 study buildings of between 125 and 375 apartment buildings housing independently living adults aged 62 years and older and people with disabilities are matched by size and randomized to one of 2 conditions (AMI followed by campaign [AB] and campaign followed by AMI [BA]), 3 buildings in each condition. In condition AB, the face-to-face intervention (A) precedes the oral health campaign (B) and in condition BA, the sequence is reversed. Buildings are selected based on size and geographic distance from one another; paired based on population age, ethnicity, and gender characteristics; and randomized by the study’s biostatistician. The procedure is described in the study protocol, which can be found on the study website [46]. Prior network studies in similar buildings [47,48] indicate that there is little or no communication among residents across buildings, and residents do not move from building to another. Thus, the risk of contamination is low. However, those who do move from one study building or condition to another will be eliminated from the study. The intervention is administered in 3 cycles of 18 months each, 2 buildings per cycle, 1 from each condition. Measures are taken at baseline (T0), after the first intervention (T1), after the second intervention (T2), and 6 months post T2 (T3); T0-T1 measures assess the impact of one intervention against the other, T1-T2 measures assess the impact of both interventions in different sequence, and T2-T3 evaluates outcome sustainability (see Figure 2).

The 6 study buildings are located in low-income areas of 3 Connecticut towns. The ethnic and racial composition of the population in these buildings consists of approximately 40% African American/Caribbean; 45% Latinos (mainly Puerto Ricans); and 10% to 15% other residents, mainly of European American or South/South East Asian origin. The rate of people with disabilities ranges from 15% to over 40%. To participate, buildings must have a common space and private locations for surveys and clinical assessments. Memoranda of agreements are signed with the managements of all buildings to guarantee participation and exempt building management from responsibility for any adverse intervention consequences. The study population speaks English, Spanish, or both. All materials used for recruiting, consenting, survey administration, and implementation of both intervention approaches are translated from English to Spanish and back translated to ensure that meaning in the 2 languages is preserved. Recruitment and all interventions are conducted in both languages or either language based on the participant’s choice.

The study population includes 360 residents, approximately 60 from each building, totaling 180 in each condition. The expected attrition rate over 4 time points is 25%. The inclusion criteria include (1) being aged 18 years and older, (2) having permanent residence (6 months or more) in study buildings, (3) living without conservator, (4) judged competent to give informed consent by responding correctly to 5 simple questions about the study during the consent process, and (5) 2 or more natural teeth. The exclusion criteria include (1) being temporary or short-term building resident; (2) being under conservatorship, (3) cognitively unable to give informed consent or respond to 3 to 5 questions about the study, (4) being edentulous; (5) having a history of infective endocarditis, prosthetic cardiac valve replacement in past 6 months, or insertion of an arterial stent or myocardial infarction in past 6 weeks or being on dialysis. The study is approved annually by the University of Connecticut Health institutional review board (IRB) and by NIDCR and Rho consultants.

Recruitment takes place during the first 5 to 6 months of each 18-month intervention period in each cycle. Recruiters are bilingual and of diverse ethnic and racial backgrounds, matching the backgrounds and languages of most participants. Steps in recruitment include initial discussions with building management, tenant associations, or other internal administrative bodies; 2 formal presentations to building residents, followed by presence on site several times a week; and informal engagement with residents and hosting of informal events, gatherings, and refreshments in community rooms and on building floors throughout the recruitment period. Public events sponsored by the study are conducted in communal spaces that can accommodate individuals with disabilities or in wheelchairs.

Participants found to be eligible are contacted within 2 to 3 days for an appointment to obtain informed consent and to conduct the clinical assessment. At the appointment, they are rescreened, informed about the study, and written consent is obtained. Consent forms are read to participants in English or Spanish, regardless of reading ability. Participants who are unable to read give verbal consent, which is recorded by the consenter and witnessed by a second member of the study staff. The signatures of both are recorded on the form with the participant’s name. Ineligible participants are invited to join an oral health campaign committee in their building.

Once consented, participants in the study are clinically assessed after which they are scheduled for a survey within the next 2 weeks. Retention is encouraged through continuous staff presence in the intervention buildings, phone calls and face-to-face reminders, home visits, casual encounters, flyers, face-to-face communication, and study-hosted social events such as bingo and ice cream socials throughout the study. These methods have been found to be useful for retention in prior studies on other health topics with similar populations in subsidized senior housing [29,44,49]. After completion of these steps, participants are engaged in scheduled one-on-one sessions (AMI) or for participation in oral health fairs, depending on the building. One month after having completed each of these interventions, participants complete a survey and clinical assessment (T1) and enter into the second phase of intervention. This is followed by a third survey and clinical assessment (T2) 1 month after completion of the second intervention. Furthermore, 5 to 6 months after the third evaluation point, participants receive a final clinical assessment and repeat the oral quality of life scale (T3).

The 2 interventions are conducted orally and through demonstrations to accommodate those with limited literacy. Participation does not require the ability to read.

Sample size calculations were based on the primary outcomes of continuous measures of GI and PS using estimates of effect and variation from a pilot grant. The MFFD design has 2 sequences (AB and BA) and 2 periods (1 and 2). Power was based on the first period, which conceptually can be considered a 2-arm parallel design with cluster randomization at the building level. Assuming a typical intraclass correlation in the range of 0.01 or 0.02, a design with 6 clusters of n of 56 for each condition would have effective sample sizes of approximately 160 to 220, respectively [53]. With n of 153 per group, our study has more than 95% power to detect a priori identified clinically meaningful mean differences.

The study evaluation design includes intervention moderators, mediators, process variables, and outcome measures. All alpha coefficients are from the study’s baseline data.

All survey data are collected using the Questionnaire Development System (QDS) [69] software and face-to-face administration. Once the interviewer or other field team member realizes a potential data entry error, immediately after the interview, he or she will notify the data manager to double check the entered responses and will correct any error in QDS Data Warehouse. The data manager may also identify key variable entry errors during the data management process. All changes are automatically documented and tracked for quality control. The data are uploaded into QDS Warehouse in batches and cleaned and amalgamated into a data analysis database for each cycle. In addition, 3 separate databases collect all participant tracking and intervention and clinical assessment data. These data are converted to a spreadsheet and merged into the master data analysis survey database by cycle. Data files for the 4 time points are integrated first for T0-T1, then for T0-T1-T2, and finally for T0-T3 to allow for longitudinal short, intermediate, and long-term analyses. AMI intervention data quality is checked by reviewing the first 5 files and an additional 5 files chosen randomly per cycle in hard copy and audio files in English and Spanish. Survey quality checks are conducted every 6 months to review missing data or errors in entry. Records are kept of any changes in data analysis files related to routine cleaning, and additional records are kept for variable recodes, new variable construction, and outliers during the analysis of the baseline data and subsequent time points.

A comprehensive statistical plan addresses the modified fractional factorial crossover design. The usual inspection for outliers and influential data points is conducted along with summary statistics and evaluations of distributions of the data. In the case of nonnormal data, we use standard transformations (eg, log transformations) or explore alternatives (eg, nonparametric approaches or other distributions such as Poisson) or create categorical or nominal variables.

For period 1, the parallel-arm phase, standard approaches for a parallel 2-arm randomized study are conducted for the first period of the sequence, across all buildings. This analysis allows direct comparison of the AMI versus the campaign approach for clinical outcomes (study hypothesis 1). For periods 1 and 2, assessing sequence of interventions, study hypothesis 2, which sequence gives better clinical outcomes, are addressed using repeated-measures models. Depending on whether the outcome is continuous or dichotomous, we will use general linear mixed models or general estimating equations to fit a model with intervention and period effects using the MIXED procedure in SAS. Each set of measures from the same person (eg, gingival outcomes) is treated as a correlated cluster of data. An auto-correlation structure of 1 is likely to be appropriate. Customized contrasts can be constructed to make comparisons between time points. Time-varying covariates can be used in these models, and some missing data (under missing at random or missing completely at random assumptions where applicable) are allowable.

For baseline and health status variables, as this study randomizes at the level of site and individuals are not randomized, we will stratify some results (or adjust in models) for demographics such as age, gender, and marital status and individual characteristics such as health status, ADLs, depressive symptoms (CES-D), and oral health status (self-assessment). Any variable that shows potential for confounding or effect modification will be accounted for appropriately in all stages of the analysis.

For intervention variables, as all subjects are exposed to the same interventions, we will unpack the intervention into (1) dosage (percentage of talking points covered for all focal points —domains—addressed in the intervention and (2) exposure to focal point messages during campaigns. Study hypothesis 5 regarding dosage of interventions will be assessed by creating an ordinal scale that measures an individual’s exposure to talking points through participation in the AMI and exposure to messages via campaigns, thus creating a measure of dosage. We will then assess if dosage predicts better outcomes. This can be done directly (eg, compare mean gingival scores across dosage levels with analysis of variance) or dosage can be used as a predictor variable in the statistical models and causal pathway analyses. Practice to mastery (a 4-point Likert scale) and focal points (count variable) can be analyzed using similar techniques and approaches.

To address the secondary outcome of GOHAI, we will conduct a set of secondary analyses with GOHAI as a continuous outcome. It will also be incorporated into our statistical/mediator models as a predictor to see if it acts as an effect modifier.

To address cognitive mediators, the following variables will be incorporated into the causal pathway analysis to address example hypotheses 3 and 4: knowledge, oral health beliefs/norms, social support, oral hygiene behaviors, fears/worries, practice to mastery, and behavioral intentions. We will use longitudinal mediation to compare the two conditions (AMI followed by oral health campaign against oral health BA) in terms of whether they differ significantly in the amount of change that they engender in the GI and PS, indirectly through the mediators measured in our model.

We are interested in examining whether there is a change in our mediators over time, and if so, whether this change significantly leads to a change in behaviors, as assessed by the GI and PS. Recent methodological studies actually show that the analysis of change is more appropriately captured by specifying change (differences) in self-efficacy and our other mediators and behavioral outcomes between adjacent time points (ie, T2-T1 and T3-T2) as latent difference scores [70-73], and explicitly modeling these change scores to represent dynamic change, that is, the impact of change in the mediators on change in the behavioral outcomes.

Hypothesis #3 will be tested using latent change score analysis specifying AMI and campaign as predictors of change in behavioral beliefs (ΔBB), OH social support (ΔSS), locus of control beliefs (ΔCB), self-efficacy (ΔSE), fears/worries (ΔF/W), social norms (ΔSN),practice to mastery (ΔP2M), and behavioral intentions (ΔBI) from baseline to first and second follow-up.

To test hypothesis 4, we will conduct dynamic mediation using latent change (difference) score mediation analysis. The analyses will be conducted in Mplus 7, where change (difference) scores representing differences in each mediator and outcome variable from baseline to final follow-up that is, T2-T1 (ΔT₂₁) and T3-T2 (ΔT₃₂) will be generated. It is the difference scores of the mediators and outcome variables that are modeled. Generally speaking, causal paths specifying latent (true) change, represented by change in knowledge, oral health beliefs/norms, oral health social support, oral hygiene behaviors, fears/worries, practice to mastery skills, and behavioral intentions in going from T1 to T2 and T2 to T3 will be specified, to predict change in GI and PS at T3 (ie, ΔGingival index₃₂, ΔPlaque score₃₂). To assess the impact of the change in self-efficacy (ΔSE), on changes in GI scores (ΔGingival index), direct and indirect paths from change in self-efficacy from baseline to first follow-up (ΔSE₂₁) that predict the change in GI at the final follow-up (ΔGingival index₃₂) will be specified and tested. Select direct and indirect paths from change scores of the other mediator variables to change scores of behavioral outcomes will also be specified in a similar fashion, and all these relationships will be tested simultaneously [74].