Nurse forecasting in Europe (RN4CAST): Rationale, design and methodology

Setting and sample

Twelve European countries (Belgium, England, Finland, Germany, Greece, Ireland, Norway, Poland, Spain, Sweden, Switzerland, and The Netherlands) were selected on the basis of research expertise, availability of patient discharge data from hospitals, geographic distribution, and duration of membership in the European Union. There are many similarities in healthcare in these selected countries, but also some striking differences, particularly in the size and structure of their healthcare workforces, the percentage of Gross Domestic Product spent on health care and the average length of acute hospital care stays [17]. The themes of similarity and difference within Europe create unique opportunities to study nursing workforce issues while learning common lessons across countries.

The setting for the RN4CAST-study focused on general acute hospitals (with at least 100 beds) that either have mixed age clienteles or treat adults only. This setting was chosen since general acute hospitals are the largest employers of nurses [17] and thus exert major influence on demand for nurses in most countries. In addition general acute hospitals represent, the largest share of national health expenditures [17] and are the sites of the largest proportion of medical errors leading to serious injury or death [18]. In each of the 12 countries (except Sweden) a study was conducted in at least 30 hospitals depending upon country size and number of hospitals. The selected hospitals represent either all of the relevant institutions in the country (Ireland, Norway) or are randomly selected, per country, from a registry of all general (non-specialized) hospitals. In Belgium, Germany, The Netherlands, Switzerland, England and Spain this selection was done at random within strata (geographical location within the countries, hospital size, and hospital type). In case randomly selected hospitals declined to participate a second or third wave of randomly selected hospitals were invited. In Belgium and Germany, hospitals (that are not selected at random) were also given the opportunity to participate on a voluntary basis. The impact of adding this group of voluntary participating hospitals will be assessed by means of sensitivity analyses. In Finland, Poland and Greece hospitals were selected via purposive sampling (i.e. geographical spread, hospital size, hospital type). Representativeness checks (i.e. hospital type and size) will be carried out in each country to assure the sample represents the population appropriately.

Within each hospital a minimum of 2 nursing units (1 general surgical and 1 general medical nursing unit) were randomly selected from a master list of nursing units. The study sample included only adult medical-surgical care nursing units since the science of linking different elements of nursing practice environment (including nurse staffing) to patient safety and clinical outcomes is best documented within this area [10–13, 15, 16, 19]. Specialized nursing units (e.g. intensive care & high dependency units, transplant care units, pediatric units, geriatric and long-term care nursing units) were excluded from the sampling frame. The minimum number of nursing units per hospital that were sampled varies between the country-specific protocols, ranging from 2 nursing units in Switzerland and Finland to all eligible nursing units in England (with a maximum of 10) and Norway. Six countries (Belgium, The Netherlands, Switzerland, Finland, Spain and Germany) sampled a variable number of nursing units based on hospital size (e.g. Belgium: 4 nursing units in hospitals with <500 beds; 6 nursing units in hospitals with 500 beds or more).

In each country all staff nurses (except nurses on sick leave, maternity leave or those who are on vacation) providing direct care to patients on the selected nursing units were included in the nurse survey. 'Nurses' are defined in each country as those meeting the European Union definition of trained and licensed nurses according to directive 2005/36/EC. In Sweden a different sampling design was used. Nurses were not approached through hospitals but via the Swedish Nursing Association (covering 85% of all nurses). Via the member register all registered nurses employed in hospitals and working in medical and surgical departments were selected. Nurses were asked to identify the hospital in which they work. This method has proved to be effective in previous research [10].

In five countries (Belgium, Poland, Greece, Finland, and Switzerland) all the selected hospitals were included in the patient survey. Whilst in other countries the patient survey was only conducted in a selection (Spain, Germany, and Ireland) or none of the hospitals (The Netherlands, Sweden, and Norway). A one-day census approach was used to select patients of the selected nursing units. All eligible patients (i.e. able to speak and understand the language of the questionnaire and to respond to the questions), present on the selected nursing units on the day of the census, were included in the study sample.

Instruments & measures

Drawing on previous experience of the 'International Hospital Outcome Study', wherever possible, existing instruments were used [20].

Data analysis

Preliminary analyses of the raw country-specific datasets including descriptive work to identify out-of-range values for variables, conflicting results, missing values and possible data entry errors will be performed. A cleaned version of the different data sources will be organized in an interrelated multilevel meta-database. This multi-country database contains information at the hospital level (survey general management), the individual nurse level (data resulting from the nurse survey) and the patient level (data from the hospital discharge dataset and the patient survey). The cross-country patient outcomes database will not contain all the original data but rather a selection of original variables (e.g. age, sex) and deduced variables (e.g. patient outcomes, diagnostic information, co-morbidities). Patient outcomes and co-morbidities will be based on published work like that of Aiken et al. [10, 34], the Charlson Index [41] and the work of Silber et al. [42, 43] Cross-mapping will be carried out for the codes used in the different algorithms because the use of different local coding languages and grouping systems (e.g. ICD-9 vs. ICD-10; APR-DRG vs. AP-DRG). This transformation process of internationally available algorithms to the context of local databases has already been done with success by several project partners [13, 14].

Two major types of analyses will be done using the multi-country survey and outcomes data. The first will involve descriptive and comparative analyses of variables reflecting commonalities and differences, policy implications and the strengths and weaknesses of the nurse workforce across countries. In these analyses individual level nurse data will be used and aggregated at the country level.

The second type of analysis involves the modeling of relationships between core independent and dependent hospital variables within and across countries. Independent measures will include staffing and work environment variables. Dependent measures will be both indicators of experiences of nurses (e.g. job satisfaction, burnout, intention to leave the job) on the job and patient outcomes variables. Regression models that estimate average differences in continuous outcome variables or differences in the odds of various negative events for nurses and patients will be fitted. The clustering of nurses and patients within hospitals (and, in the case of cross-national analysis, of hospitals within countries) will require the use of multilevel modeling strategies [44]. A five staged approach will be employed. In a first set of analysis the factor structure of the PES-NWI and the risk-adjustment procedures will be explored. Multilevel factor analytic techniques [45] will be used to confirm the structure of the subscales from the PES-NWI to be used as indicators of nurse practice environments. The within- and between-hospital variability of the nursing practice environment components will be studied. This is necessary prior to potential aggregation of some measures to the nursing or hospital level. Logistic regression models will be used with the patient discharge data (patient demographics, co-morbidities, diagnostic categories) to derive propensity scores, based on factors affecting the likelihood of mortality and failure-to-rescue, which will serve as case mix adjusters in further analyses. In a second step, a two-level model (in each country separately) will be applied to study the relationship between hospital characteristics (aggregate measures obtained from the nurse survey such as nursing workload, nursing practice environment) and outcome measures (e.g. nurse assessed quality of care, retention, job satisfaction) obtained at the level of the individual nurse via the survey. In a third step, a two-level model (in each country separately) will be applied to study the relationship between hospital characteristics (aggregate measures obtained from the nurse survey such as nursing workload, nursing practice environment) and patient outcomes (data on the patient level obtained from the hospital discharge datasets or the patient survey) within each country. In a fourth step, the analyses of steps 2 & 3 will be performed across countries thereby introducing a third level (i.e. country: characteristics of the country) into the model allowing cross-country analyses.

Whenever missing data are present, statistical imputation methods will be used to provide alternative analyses (as a sensitivity analysis) to the approaches where cases with missing data are omitted.

Appraisal of current nurse forecasting models and policy synthesis

An extensive literature review was undertaken to gather information allowing the team to evaluate and appraise the current nurse workforce projection models and forecasts. In addition to searching the traditional databases like CINAHL, Embase and Medline, teams in each country contacted institutions and stakeholders in each country to obtain published and unpublished data and reports about forecasting models. Based on the results of the literature review, the researchers will appraise and evaluate the currently employed forecasting and planning models.

A policy synthesis will combine the literature review on nurse forecasting models with the results of the data analysis about the impact of different aspects of the nursing work environment on patient outcomes and nurses' job experiences. The likely impact of a range of policy instruments (e.g. an increase in nursing education program, investment in recruitment methods, and investments in the nursing practice environment) on patient outcomes and nurse retention and ultimately staffing levels and patient outcomes, will, for instance, be estimated from country-specific data. Sensitivity analyses considering the impact of altering the underlying assumptions in various ways will be used to frame the estimates generated. This will result in several types of country-specific scenarios for the nurse workforce in each country in the coming decades including the simulation of what happens when different conditions in the nursing workforce are changed. A synthesis document will be created and indicating how similar or different the conclusions of the data analyses are across countries, whether or not it is possible to identify clear Europe-wide conclusions from the work in RN4CAST.

Ethical issues

The project has been granted financial support from the European Commission. Depending on national legislation, the study protocol was approved by either central ethical committees (e.g. nation or university) or local ethical committees (e.g. hospitals). Proof of the ethical approvals has been submitted to the editorial board of this journal for verification. The consortium has developed strict criteria (included in the project proposal and additional internal documents) regarding the sampling of nurses and patients, the storage, flows and access of the data to safeguard the security, privacy and confidentiality.

Policy and scientific impact of the project

The study will make a strong significant scientific contribution by shifting the main focus of nursing workforce planning from rather simple projections in demand and supply of labour to impact on patient safety and quality. The innovative nature of the RN4CAST project is the consideration of new factors in workforce planning such as the work environment recruiting and retention of nurses within the profession and the link between nursing adequacy and patient safety and quality. The study will generate the necessary scientific basis to underpin informed policy decisions on health systems and more effective and efficient strategies of nursing workforce planning. The studies of Needleman et al. [47] and Rothberg et al. [48] show, based on the data collected in the United States, how the research findings within this field of research have become sophisticated enough to be reanalyzed and extrapolated to provide policy guidance. Based on the available data, one proposes that mandating minimum nurse-to-patient ratios in hospitals at 4 or 5 patients per nurse across the United States would yield a cost-benefit ratio (in terms of cost per life saved) superior to that for percutaneous coronary interventions for acute myocardial infarction and routine Pap smears to screen for cervical cancer [48]. The second economic analysis, the extrapolation from results of a major national study of nearly 800 hospitals, suggests that moving all United States general hospitals to a skill mix of highly educated and somewhat more expensive personnel to the 75% percentile nationally would not only save lives, but could actually reduce expenses for the health care system as a whole by lowering complications and shortening length of stay [47]. This latter analysis suggests that the policy approaches increasing skill mix rather than the numbers of personnel would be more effective in improving quality in American hospitals.

Also in Europe, it is illustrated that the potential quality gain in acute hospitals can be substantial. It was shown, for example, that if Belgian acute hospitals would manage to shift the prevalence of "failure-to-rescue" (deaths of patients with complications), to the performance level of the current 20th percentile, the overall in-hospital mortality in Belgian hospitals would decrease with 3.95% in the group of medical patients and with 6.50% in the group of surgical patients [49]. Similar effects were shown in decreasing the number of complications such as hospital acquired infections, pressure sores, wound infections and others.

The study will allow countries to learn from the experience of other health systems and their sustainability, taking into account the importance of national contexts and population characteristics. Focus will be on how nurse staffing and organizational aspects of health systems impact on patient outcomes. The study can serve as an evaluative lens on the nurse workforce impacts of recent national policy initiatives in health care, such as the recent adoption of prospective payment and Diagnostic Related Groups in Germany and the slowing of investments in the National Health Service in England.

Stakeholders Engagement

Simultaneously to the research activities, the project entails dissemination and stakeholder activities toward achieving the study objectives. An impact assessment preceded the establishment of a stakeholder panel representing patient, nursing and healthcare organizations at the European level. The main role of the international panel is to raise awareness of the project and to support the research team in formulating policy recommendations based on the scientific results. Next to the international stakeholder panel, in each partnering country national stakeholder committees are formed to further gain support for the project.

Next to the large-scale European part of the RN4CAST project, three International Cooperating Partner Countries of the European Union (Botswana, China and South Africa) participate in the project consortium. All three countries will at least perform a pilot study to provide a broader international perspective on the study results.