This is a pre-planned subgroup analysis of the PACE cluster randomized controlled trial (see data analyses plan submitted as an official deliverable to the European Commission in Additional file 1) [4, 5, 10]. This cluster-RCT was conducted in 78 nursing homes in Belgium, England, Finland, Italy, the Netherlands, Poland and Switzerland to compare PACE Steps to Success with usual care (2015–2017). This trial was registered at http://www.isrctn.com on July 30, 2015 (ISRCTN14741671). Randomization was performed at the nursing home level as the program involved the training of nursing home staff. After baseline measurement, randomization was stratified by country and median number of beds in a 1:1 ratio. Randomization was blinded and performed by independent statisticians. Because of the nature of the study, blinding of treatment was not possible for researchers or participants. More details about the PACE cluster-RCT have been published elsewhere [4, 5]. We reported this study following the CONSORT guidelines for randomized trials.
PACE Steps to Success was implemented over the course of 1 year, including 2 months for preparation, 6 months training for nursing home staff in the six steps, and 4 months consolidation. All countries had one or more country trainers. Each nursing home assigned one to six staff members as PACE coordinators. After being trained by two experienced trainers, the country trainers trained and supported the PACE coordinators who were in turn responsible for training and supporting fellow staff. The six PACE Steps included: 1) advance care planning with residents and families; 2) assessment, care planning, and review of resident needs and problems; 3) coordination of care via monthly multidisciplinary palliative care review meetings; 4) high-quality palliative care with a focus on pain and depression; 5) care in the last days of life and 6) care after death . The program included three dementia-specific elements: communication training in advanced dementia for the PACE coordinators, and two elements integrated into the training for all nursing home staff which emphasized dementia as a terminal illness (as part of Step 2) and offered symptom control strategies for residents with and without dementia (in Step 4) [4, 5].
Participating nursing homes
From a list of nursing homes, those located in a predefined country-specific geographical location were approached randomly by telephone or e-mail to invite them to participate in the study and to evaluate eligibility criteria using a standardized checklist. Inclusion criteria were the provision of on-site nursing care and personal assistance with activities of daily living and off-site medical care by general practitioners (GPs), having at least 30 beds, 15 or more residents having died in or outside the nursing home in the previous year to obtain sufficient power, consent to participation from management in writing before randomization, and agreement to allocate approximately 0.5 days per week for staff to act as PACE coordinators. We excluded nursing homes that had pilot-tested the program materials or used detailed palliative care guidelines/planning tools, the Gold Standards Framework and InterRAI-PC [4, 5].
Data collection and respondents
One contact person per nursing home identified all residents who had died in the previous 4 months. After-death structured questionnaires for each resident were sent to the staff member most involved in care (preferably a nurse), nursing home administrator and GP at baseline (month 0) and post-intervention (months 13 and 17). As sensitivity analyses showed no difference between program effects using the two post-intervention data, these combined post-intervention data were used in the primary analyses . In this subgroup analysis, we included residents for whom the presence and severity of dementia was determined, classified into three subgroups: advanced, non-advanced and without dementia. We deviated from our pre-planned subgroups (residents with and without dementia), so that we could better investigate the difference between residents with advanced and without dementia.
Measurements and outcomes
Nursing home staff and GP reported whether a resident “had dementia” or “was diagnosed with dementia”. Dementia was considered present if at least one indicated it was and not present when both indicated it was not or when one indicated this but the other neither returned the questionnaire nor answered the question. Dementia severity was determined using two highly-discriminatory staff-reported instruments, Cognitive Performance Scale (CPS) and Global Deterioration Scale (GDS); those with CPS scores of 5–6 and GDS stage 7 were classified as having advanced dementia, the others as non-advanced dementia. CPS classifies residents into six hierarchical cognitive performance categories, with higher scores indicating worse cognitive impairment . GDS stage 7 indicates very severe cognitive and functional deterioration .
Nursing home administrators reported a resident’s sex and age at time of death. Staff assessed functional status 1 month before death in terms of dependency level with eating, dressing and mobility using the Bedford Alzheimer Nursing Severity-Scale: categorized into ‘independent’, ‘needs assistance’, or ‘fully dependent’ .
Primary outcome was staff-reported comfort in the last week of life using the validated Comfort Assessment in Dying–End-of-Life in Dementia (CAD-EOLD) scale; see comprehensive description of outcomes in Additional file 2 [15, 16]. CAD-EOLD comprises four subscales: physical distress, dying symptoms, emotional distress and well-being. The CAD-EOLD total scores range between 14 and 42, with higher scores indicating better comfort. CAD-EOLD was found to have better psychometric properties and user-friendliness than other comfort measures in a mixed nursing home population, including residents with and without dementia [17,18,19]. Secondary outcome was staff-reported quality of care and dying in the last month of life measured using the validated Quality of Dying in Long Term Care (QOD-LTC) scale, comprising ‘personhood’, ‘preparatory tasks’ and ‘closure’ subscales . The QOD-LTC total scores range between 11 and 55, with higher scores indicating better quality of care and dying.
Linear mixed models were used to analyze continuous outcomes and account for the clustered nature of data, with staff, nursing home and country as random factors (only random intercepts) and group (intervention versus usual care), time (post-intervention combining data collected at months 13 and 17 versus baseline) and their interaction as fixed factors. We analyzed differential effects by calculating differences in mean change (post-intervention combining data collected at months 13 and 17 minus baseline) between the subgroups, both for the intervention and control groups (interaction group*time*dementia). For the differential effects, we present estimated differences (and 95% Confidence Intervals) in mean change between the subgroups. All hypothesis testing was two-sided. P-values and 95% Confidence Intervals were not adjusted for multiple testing. To address multiplicity concerns with Bonferroni correction, p-values should be compared against a 1% significance level to address multiplicity concerns examining dementia subgroups . In individual subgroups, we presented estimated mean scores and mean differences between groups post-intervention. All analyses were on an intention-to-treat and a complete-case basis, assuming data were missing at random. All statistical analyses were conducted using SAS 9.4 software (©SAS Institute Inc., USA).