This multi-center feasibility study with nurse-led interventions used a quasi-experimental research design. Here we partially report on our trial outlined in the protocol published previously [9]. The process evaluation conducted with qualitative methods is reported separately.

We conducted this study at three Swiss nursing homes in the cantons of Zürich and Thurgau. A convenience sample of 164 patients were observed for 147 days over a total study duration of 425 days or until they were censored. People with dementia were recruited for this study if they lived in one of the three nursing homes and had a dementia diagnosis in their records or symptoms of dementia as evidenced by the minimum data set. The minimum data set items: comatose status, short-term memory, ability to make decisions, making self understood and eating performance were assessed [10, 11] with the nursing homes to determine people with dementias’ eligibility for the study.

The intervention was aimed at frontline nursing home staff, including registered nurses and healthcare assistants, interns, and those in other nursing roles. To participate, frontline staff had to be 18 years or older and had to have been employed at the respective nursing home for at least a year prior to the study. Furthermore, they had to be employed at least a part-time equivalent of 30% and adequate spoken and written language skills in order to participate in the study. The previously published study protocol presented the sample-size and statistical power considerations for our event history and multilevel analyses [9].

The pain frequency, pain intensity, pain episode duration, and pain-free intervals—as measured or derived by the “Beurteilung von Schmerz bei Demenz” (BESD; the German version of the Pain Assessment in Advanced Dementia Scale [PAINAD]) scores—were assessed by the participating frontline staff at least once daily, and over 24 h at suspected pain events, during the three 49-day periods [12]. Like with most observational scales for pain assessment in dementia also with BESD, there is no true certainty if the observed individual is in pain or not [13]. BESD demonstrated high levels of inter-rater and retest reliability and was developed for use with patients with advanced dementia [14,15,16]. It covers five categories to observe (breathing, negative vocalization, body language and consolability) with three manifestations each (e.g. none = 0 to severe = 2) [12]. BESD scores are calculated based on observations during care and range from 0 to 10, with a cut-off at 2 points indicating mild pain [17].

Data were collected by trained frontline staff during all shifts when pain was suspected or at least once a day. They filed their observations during a baseline period (T0) and two follow-up periods (T1 and T2) these are illustrated regarding absolute time in Table 1.

As mentioned above, BESD scores were assessed by the participating frontline staff at least once daily and at suspected pain events. In addition, the sociodemographic data of the people with dementia enrolled in this study were obtained via the minimum data set. Since each observation period spanned 49 days data for 147 days was analyzed.

The two-part intervention was delivered for 49 days after baseline and again after the first follow-up. After the first intervention cycle and before T1, there was an observation-free period of 60 days. There was also an observation-free period of 119 days between the second intervention cycle and T2. Therefore, for each nursing home, the total study duration amounted to 430 days.

The nurse-led intervention was divided into two parts: (1) an on-site training workshop for frontline staff, and (2) on-site case studies and individual coaching with the frontline staff. Intervention elements are detailed in Table 2. The two part intervention was designed by the research team around best practice recommendations for the advanced nursing process [18,19,20] without user or patient involvement.

The two-hour training workshop was conducted on-site by external clinical nurse specialists trained in geriatric and palliative care and aimed to train frontline staff in systematic pain assessment. The on-site coaching was designed to guide the frontline staff in systematic pain management, including systematic charting. The case studies were related to bedside coaching and communication training, to support interdisciplinary communication about pain and symptom management in people with dementia. The on-site team took fifteen to sixty minutes to discuss pain and symptom management and their observations during caregiving with the intervention nurse. Before suggesting prescription changes to the team, the intervention nurse conferred with an on-call geriatrician during the coaching intervention.

The Zürich cantonal ethics committee and the ethics committee of Eastern Switzerland (2016-0001) considered this study at registration a quality improvement project without the need for informed consent. Nonetheless, participating people with dementia and their legal representatives were asked to provide written informed consent. The study was registered with the German registry of clinical trials (DRKS00009726).

We fitted mixed-effect growth curve models with Stata 15.1, using spline regression with five knots to estimate the adjusted course of pain frequency, pain intensity, and pain-free interval duration. Because the data were derived from repeated measures, had multiple levels, and had nested factors (e.g., people with dementia nested in departments, nested in nursing homes), a simpler model may not produce unbiased and precise estimates [22]. The data were checked for linearity, additivity, and normality, and no issues were found. Missing data stemmed from attrition (i.e. deaths) which the models where adjusted for. We adjusted the models for nursing homes, age (centered at the mean), sex, number of previous pain events, and deaths during the study period. To assess the course of pain frequency by nursing home, we included interaction terms based on the spline knots, with the nursing home variable as a fixed effect. In addition, patient-specific random effects were included to capture potentially correlated observations for the same person with dementia (i.e., clustering due to repeated measurement). Similar models, in which splines were replaced with period-specific dummy variables, were estimated for both the baseline and follow-up periods.