Applied Ergonomics 86 (2020) 10310022013). This reflects an underlying principle of systems orientation where the person/team is an embedded component of a sociotechnical system (Dul et al., 2012). 1.1.The medication administration process Medication administration is ingrained in the nursing home work system, and specific tasks related to medication administration are often difficult to separate from other work processes in the daily care of the patients (Carayon et al., 2014; Jennings et al., 2011). The medication administration process (MAP) can be represented in six stages, from ordering, transcribing, dispensing, preparing, administering and observing (Carayon et al., 2014; Odberg et al., 2017). It involves different stakeholders performing different tasks using technologies such as electronic medication administration records (eMAR) while relating to organisational conditions, rules and guidelines within a physical environment. The major stakeholders in the current study are registered nurses (RN), medical doctors (MD), nurse assistants (NA), the patients and the pharmacists. The pharmacists were not a focus in the study, as they perform their role externally to the nursing homes in the current context, the only communication being via electronic or other correspondence. MAE’s may occur anywhere along the MAP, and errors committed in the first stages may affect the consecutive stages to cause potential ADE’s such as the patient falling critically ill due to receiving the wrong dosage or drug. Examples of other MAE’s may be failures of omission, mistaken patient identity or wrong route of administration. There are many reasons why MAE’s occur; some contributing factors are interruptions, poorly designed eMAR, lack of guidelines, high workload, poor interprofessional cooperation, lack of leadership and inadequate competence among staff (Al-Jumaili and Doucette, 2017; Andersson et al., 2018; Carayon et al., 2014; Dilles et al., 2011). There are also several measures designed to reduce the risk of MAE’s, such as double-checking of medications, updated checklists, the introduction of bar-coding technology, training in medication knowledge, teamwork training and various efforts to reduce interruptions. However, few in-terventions in themselves seem to have any significant effect in reducing MAE’s in nursing homes. In order for efforts to improve medication safety in nursing homes to be effective, they should be multifaceted as well as requiring a healthy management and strong leadership with adequate resources (Al-Jumaili and Doucette, 2017; A. Andersson et al., 2018; Carayon et al., 2014; Dilles et al., 2011). 1.2.Process modelling There is general acknowledgement in the human factors literature that to increase overall understanding of the health care system, we need to focus on processes within the work system rather than tasks (Wool-dridge et al., 2017). Care processes can be wound treatment, patient transport or medication administration. Other processes taking place simultaneously in the work system can be the introduction of new software, changing guidelines or renovations of the patient bathrooms. There are also the cognitive processes of the persons involved, that may explain or describe why and how some act in a certain way. These processes take place within the work system simultaneously. They interact and are part of what makes healthcare systems complex (Holden et al., 2013). There are numerous ways of modelling processes in healthcare, but some have proven more influential than others. Jun et al. (2009) found that flowchart and swimlane diagrams1 were most commonly used, while flowchart diagrams were more accessible. Likewise, system dia-grams of a similar type have proven useful when engaging in identifying risks in healthcare processes (Simsekler et al., 2018). Process modelling is one way to comprehend better how persons, technology, physical environment, organisational factors and care processes interact and work (Jun et al., 2009). It differs from more traditional flowcharts by including the stakeholders as separate headings in the top row, thereby portraying the persons performing different tasks in the MAP. According to Wooldridge et al. (2017), a SEIPS-based process modelling technique may provide a powerful tool in identifying barriers and facilitators in healthcare work processes. By integrating the work system into the process map, the resulting diagram presents a holistic representation of the activities involved in the process while also retaining a general overview of the dynamic in-teractions of the involved system elements. The process modelling in this study is based on Wooldridge et al. (2017) and incorporates the work system elements in the SEIPS model, systematically visualising the MAP. Since there are few studies that investigate medication administra-tion within nursing homes as a whole, the current study aimed to explore the dynamic interactions of stakeholders and work system elements in the MAP in a nursing home ward. The following research question Fig. 1.An adapted SEIPS model based on Carayon et al. (2006). 1Cross-functional process map to illustrate workflow and interrelated activ-ities, which visually distinguishes job sharing and responsibilities (Damelio, 2016). K.R. Odberg et al.
Applied Ergonomics 86 (2020) 1031003guided the study: How can SEIPS based process modelling visualise barriers and facilitators in the medication administration work system of a nursing home ward? 2. Methodology 2.1.Design The study used a qualitative design (Morse, 2016). Data were collected using (1) observations of the medication administration pro-cess in a nursing home, supplemented with (2) interviews with health-care professionals in the nursing home. A deductive content analysis used combined data from the observations and interviews to inform the SEIPS-based process modelling. 2.2.Setting The study took place in a Norwegian nursing home ward with palliative patients in need of medical treatment. The palliative ward had six patient rooms, mostly occupied for the duration of the study. A small nurse station was centrally located with an inner office for the nurse manager. Computers, charts, documents, a printer, telephone, and a mobile medication trolley were housed in the nurse station. Two parallel corridors provided access to the patient rooms as well as a common living room with a small kitchenette. The medicine room was 60 m away in the opposite part of the building past another ward. Patients had a varying length of stay as some needed only short-term assistance (2–5 days); if their condition improved, they were discharged; if it worsened, they were transferred to other facilities. Other patients stayed for weeks or months. Most patients had extensive medical regimes including administration of oral medications, infusions, and patches at four reg-ular intervals each day. The staff consisted of 25 members – six full-time registered nurses (RN), two nurse assistants (NA) and an associated medical doctor (MD) in a 50% with two regular visits per week. The remaining staff members had either short-term engagements or held less than 50% positions at the ward. 2.3.Recruitment The nursing home ward is part of a network of development centres in Eastern Norway and was chosen due to the leaders expressing an intention to participate in relevant research. The main author contacted the manager of the nursing home in January 2016, and the manager of a palliative ward agreed to participate. All the staff members were briefed in a joint meeting and asked to take part in the study, to which they agreed. The staff members were informed that they might be asked to participate in interviews at a later stage. The inclusion criteria set for the interviews were that the participants had a regular position working 50% or more and that they had a role during the medication administration process. After three months of observations, the staff members fulfilling the inclusion criteria were asked to participate in the interviews. In all, 9 staff members, ranging from nurses2 with post-graduate education in palliative care, RN’s, a nurse manager, an MD, and NA’s were asked. All agreed to be inter-viewed, and they were again informed of confidentiality and of the possibility to withdraw. 2.4.Data collection The main author performed partial participant observations (Bour-geault et al. (2010) from April to October 2016. All staff with tasks related to medication management in the ward were observed during daytime shifts and evening shifts on weekdays. A few observations took place during night shifts. Most observations took place during daytime shifts in order to capture periods of high activity related to patient care and medication administration. While staff members performed tasks related to medication administration, the researcher took notes and paid attention to the various work system elements, such as the use of tech-nology or how noise and interruptions affected their work. After each day of observations, the notes were transcribed by the researcher. Halfway through the observation period, the staff members who met the inclusion criteria were interviewed. All observations (70 h) (See Appendix 1: Observation guide) and interviews followed guides based on the five elements of the SEIPS model: persons, physical environment, tasks, tools & technology and organisation. The interviews were semi- structured, lasting from 30 to 60 min, and were performed in a sepa-rate room in the nursing home facility (See Appendix 2: Interview guide). The goal of the interviews was to explore how the staff related to and experienced the medication administration process. The interviews were audio-recorded and transcribed partly by the researcher and by a professional transcription service. After the interviews were conducted, the remaining period (July to October) of observation narrowed in its focus, seeking to refine specific elements in the MAP pinpointed in the interviews. 2.5.Trustworthiness The use of a sole observer, an intensive care nurse by training, may cause bias, such as prior conditioning (Bourgeault et al., 2010). At the same time, familiarity with the setting and the medication administra-tion process may lead to insights otherwise missed. The researcher was aware that the interaction with the participants could influence or bias the surroundings and the data collected (O’Brien et al., 2014). To limit bias, special attention was given to how data was interpreted. By introducing individual interviews of the stakeholders at the mid-point of the data collection period, member checks of early interpretations of the observation data helped clarify and elaborate identified issues. In the remaining observation period, the researcher’s interpretations were refined and an accurate description of the medication administration process was arrived at. The research team, consisting of three profes-sional researchers with a diverse background in nursing and engineer-ing, had regular meetings during the data collection and analysis. They discussed key issues and performed iterative analytical reflections to ensure the reflexivity and trustworthiness of the findings. To ensure the dependability of the data, the observations were made by the same researcher throughout the six months, using an observation guide. Over time, the staff members grew used to a researcher being present and probably took less notice (Lincoln and Guba, 1985). Excerpts from the interviews and observation notes are included in the results section to increase the transferability of the findings. 2.6.Analysis and process map development The transcribed observation notes and interviews were analysed using deductive content analysis (Elo and Kyng€as, 2008). A catego-risation matrix based on the five elements of the SEIPS model (columns) and the six stages of the MAP (rows) formed the categories. The delin-eation between the different stages was mainly based on observations, together with descriptions of the MAP (Carayon et al., 2014). As activ-ities in the ward were seldom linear, the building of the process model involved interpreting where certain activities belonged. For example, activities in stage one (ordering) and stage two (transcribing) often took place simultaneously. The first part of the deductive content analysis involved the entire research team individually reading the entire data material to make sense of the data. Relevant parts of the material were marked, and the data were discussed to form an analytical approach to answer the research question. To identify relevant meaning units, the focus was on 2 Both registered nurses and nurses with post-graduate education are described as RN’s in the study. K.R. Odberg et al.