Feasibility analysis of the design of a psychosocial intervention in dementia: Sensitivity analysis

Abstract

AbstractBackgroundA multicomponent staff training intervention was conducted to improve residential dementia care. The purpose of this work is to present the simulation of the effectiveness of the intervention design over time.MethodPilot study with mixed methods, two‐arm randomized controlled clinical outcomes in 7 care homes in Mexico City. Staff outcome measures were: Maslach Burnout Inventory (MBI); Approaches to Dementia Questionnaire (ADQ); Sense of Competence in Dementia Care Staff (SCIDS). Residents' outcome measures were: Quality of Life‐Alzheimer’s Disease scale (QoL‐AD) and Neuropsychiatric Inventory‐Nursing Home Version (NPI‐NH). Using the functions of the R package powerlmm, we performed computer simulations based on linear changes over a period of up to 24 weeks. The measurements were divided into three time points corresponding to the schedule of our study (baseline, post‐intervention and follow‐up). In the different trials, we specified different numbers of subjects and group sizes, including those in the present study, always considering an unbalanced cluster design. We also considered the dropout rate of participants, which in this study was almost 30% and focused on the post‐intervention period, with no additional dropouts at follow‐up. We conducted simulations to determine the performance of our longitudinal study design with fewer or no unanticipated dropouts.Result55 residents and 126 staff participants were recruited in seven care homes. We found that the highest power was achieved for MBI: Personal fulfillment and SCIDS: Sustaining personhood, which were the only variables with a significant effect for the Time × Treatment interaction in the model inference. Focusing on the first of these variables and assuming the specific values for each input parameter, we see that our study design with 7 care homes was potentially capable of detecting a significant effect size of the growth effects of the intervention. Thus, the estimated sample size for a power of 80% would be nearly 85 participants, which corresponds to the complete data (i.e., no dropouts), and about 138 participants for a power of 90%.ConclusionA higher level design that includes more care homes would be sufficient to achieve effective estimates above the 80% threshold while maintaining the lowest number of participants.