This study is part of the larger IMPAKT in nursing homes study, an IKT study [26] (Fig. 1).

Trial design

The study was conducted as a pragmatic two-arm cluster randomized controlled trial (RCT).

Setting and participants

The responsibility for LTC in Norway is placed at the municipal governmental and geographical level. Around 10% of beds in nursing homes are used for short-term purposes, such as post-acute care and rehabilitation, leaving the greater part for long-term beds. Both types of beds offer 24/7 services on a needs-based admission, and when it is not possible to provide the level of care needed by home-based nursing.

This study took place in a Norwegian urban-suburban municipality serving a general population of 292 482 [27]. Twenty-one public nursing home facilities in the municipality were administered in one nursing home organization and all were eligible for inclusion. In total, these nursing homes had 1466 beds, and the size of the NH facilities ranged from 30–107 beds (mean 70 beds) at the commencement of the study. There were no for-profit facilities in the municipality, and private non-profit facilities were excluded from the trial as they had less affiliation to the municipal nursing home provider.

Intervention group

The intervention was formed as a KT capacity program consisting of an educational component and a facilitation-upon implementation component.

The clinical innovation

The decision to choose NEWS2 as the clinical innovation was undertaken in the IKT partnership. Practice Development Nurses (PDNs) hold a particular responsibility for KT and professional development in the organization and were considered key stakeholders when identifying a clinical area of collective relevance for the KT project [28]. Researchers organized a 60-min workshop, in an already scheduled meeting of PDNs. To prepare for the workshop, each PDN was asked to collect ‘clinical uncertainties’ in their respective nursing homes, as care staff had experienced over the last month. In the workshop, one common list of uncertainties was developed, then ranked according to the number of times the same area of uncertainty appeared. Researchers performed literature searches to determine which of them had sufficient level of evidence to pursue as a knowledge-gap. The highest ranked clinical uncertainty was how to assess clinical deterioration in the residents, using NEWS2. This priority was supported by nursing home physicians and the top management of the organization.

Although NEWS2 is a validated tool within acute care settings, the working group behind NEWS2, the UK Royal College of Physicians, suggest that it be adapted for use outside the hospital setting [29]. These adaptations should consider factors such as clinical competencies and patient socio-demographics. In our study, the necessary adaptations were made during a workshop that included the Head of physicians in the nursing home organization and his team. This process was thoroughly documented, and for transparency, the documentation has been translated and included in Additional file 1.

Implementation strategy

The KT capacity program consisted of two distinct parts. Part I was an educational KT capacity program tailored to the KT needs of care staff [28] and lasting for the academic spring semester in 2019. During the program, participants from each facility developed local knowledge-to-action (K2A) plans for the implementation of NEWS2. Part II was a facilitation-upon-implementation period, where the PDN in each NH applied their KT competence, in locally developed K2A plans. Part II commenced when Part I ended, from June 2019 until March 31, 2020. More details about the intervention is available in Table 1, and in the GREET checklist (Additional file 2).

Table 1 Description of the educational component of the IMPAKT interventionControl group

The control group continued with care as usual, without any interference from the IMPAKT project. Yet, several national and local initiatives overlapped with our clinical innovation that encouraged the use of NEWS2 in community settings. For instance, the Norwegian patient safety program published a national resource called “Early detection and fast response in somatic health deterioration”, where the NEWS2 tool was available online for all care settings in Norway [22]. At least two facilities in the control group participated in a local learning network, established to facilitate the use of NEWS2. In line with the pragmatic design of our trial, we made no attempt to mitigate other initiatives that could influence the adoption of NEWS in the control group.

Outcomes

We assessed the effect of the implementation strategy based on the rate of documented use of NEWS2 at the resident level. Secondary outcomes included the use of NEWS2 in clinical situations with clear indications for NEWS2 assessment, defined as when residents acquired infections or were transferred to acute care.

Sample size

A statistician calculated the a priori sample size based on an expectation of 10% improvement in the use of NEWS2 in the intervention group compared to the control group. To have a power of 80% to detect this difference between groups at a 5% level of significance, accounting for correlation between outcomes within the same clusters, we calculated that we needed minimum 470 residents or 7 nursing homes per arm. For more information about power calculations see Additional file 3.

Enrollment and randomization

The nursing home organization committed to participation in the overall IMPAKT study funding was secured. Two facilities were considered unfit to participate by the Director of the NH organization and were excluded prior to randomization.

This study is a cluster randomized RCT. The rationale for cluster randomization is related to the practical challenge of randomizing the intervention across the nursing home population. Randomizing at the nursing home level, rather than using simple random sampling, is a more feasible approach often used in the healthcare setting.

Enrolled nursing homes were paired, based on a list of disidentified facilities, to match on size (number of beds) and type of beds (long-term beds or short-term beds). Subsequently, the NH pairs were randomly assigned 1:1 to either intervention or control using the random number generator in IBM SPSS statistics. This ensured concealment of the allocation. All residents residing in the nursing homes during the study period were included in this study.

Data collection

Individual resident-level registry data was extracted from electronic patient journals. A Structured Query Language (SQL) syntax was developed and validated with the contractor of the electronic patient journal system (DIPS ASA), based on its codebook. The query was run by the IT department in the municipality.

Blinding

Blinding was not possible among care providers, residents or researchers. Outcome assessors were blinded.

Statistical methods

Firstly, a generalized additive model was used to smooth the outcome in presented graphs. Linear multilevel growth model analysis was used to assess the impact of the KT intervention on the use of NEWS2, which was measured by NEWS2 assessments per patient month. The statistical models accounted for the correlation between outcomes within the same cluster (NH id) and for repeated measurements among residents (resident id) over time. All analysis were done in the R statistical environment using the package “lme4” [36, 37].

The likelihood ratio test was used to examine best model fit for both fixed and random effects using a maximum likelihood estimator. For the fixed effects, likelihood ratio test was used to compare nested models to determine if the fixed effects being tested significantly improved the model fit. Specifically, we tested a three-way interaction effect (time*allocation*period) between the use of NEWS2 by time (per month), allocation to either intervention or control group, and period (baseline/clinical intervention/follow-up). In addition, we tested a four-way interaction to see if the secondary outcomes moderated the effect of the intervention on the use of NEWS2 (time*allocation*period*referral- or infection rates). The significance level was set at 0.05.

For the random effects, the higher order variance components were tested against a simpler model excluding higher-order interactions. In our study this meant assessing fit based on random intercept by resident id and nursing home id (a three-level model). A model with three-level random intercept and a fixed effect three-way interaction had the best model fit.

The intraclass correlation coefficient (ICC) was used to indicate the proportion of the total variance that could be explained by the group-level clustering. An ICC value greater than 0.05 or 5% is considered to indicate a meaningful variation at the group level. In this study the ICC value was 0.22 indicating a multilevel factor structure between nursing homes that should be accounted for.

The effect size measures will be expressed as the (unstandardized) mean difference (Md) and standardized mean difference (d). We calculated standardized mean difference for multilevel growth models (dGMA-raw) based on recommendations by Feingold [28]. This effect size can be interpreted similar to the Cohen’s d heuristic where an effect size of 0.2 can be considered a small effect, 0.5 represents a medium effect and 0.8 a large effect [38].