HEALTH POLICY AND SYSTEMS

Impact of Nursing Unit Turnover on Patient Outcomes in Hospitals

Sung-Heui Bae, RN, MPH, PhD1 , Barbara Mark, RN, PhD, FAAN2 , & Bruce Fried, PhD3
1 Research Assistant Professor, School of Nursing, University at Buffalo, State University of New York, Buffalo, NY 2 Sarah Frances Russell Distinguished Professor, School of Nursing, University of North Carolina at Chapel Hill, NC 3 Associate Professor, Health Policy and Administration, School of Public Health, University of North Carolina at Chapel Hill, NC

Key words Nursing turnover, workgroup processes, patient safety, patient satisfaction Correspondence Dr. Sung-Heui Bae, School of Nursing, University at Buffalo, State University of New York, 211 Wende Hall, 3435 Main Street, Buffalo, NY 14214-3079. E-mail: sbae7@buffalo.edu Accepted: July 31, 2009 doi: 10.1111/j.1547-5069.2009.01319.x

Abstract
Purpose: The aim of this study was to examine how nursing unit turnover affects key workgroup processes and how these processes mediate the impact of nursing turnover on patient outcomes. Methods: A secondary data analysis was used to test the hypothesized model. This study used registered nurse and patient data from 268 nursing units at 141 hospitals collected as part of the Outcomes Research in Nursing Administration (ORNA II) project. Nursing units provided monthly nursing unit turnover rates for 6 consecutive months, and registered nurses completed questionnaires measuring workgroup processes (group cohesion, relational coordination, and workgroup learning). Patient outcome measures included unit-level average length of patient stay, patient falls, medication errors, and patient satisfaction scores. Results: Nursing units with moderate levels of turnover were likely to have lower levels of workgroup learning compared to those with no turnover (p<.01). Nursing units with low levels of turnover were likely to have fewer patient falls than nursing units with no turnover (p<.05). Additionally, workgroup cohesion and relational coordination had a positive impact on patient satisfaction (p<.01), and increased workgroup learning led to fewer occurrences of severe medication errors (p<.05). Conclusions: The ndings of this study provide specic information on the operational impact of turnover so as to better design, fund, and implement appropriate intervention strategies to prevent registered nurse exit from nursing units. Further investigation is needed to assess the turnover-outcomes relationship as well as the mediating effect of workgroup processes on this relationship. Clinical Relevance: Managing nursing unit turnover within appropriate levels at the nursing unit is critical to delivering high-quality patient care.

The adverse impact of nursing turnover on quality of patient care is a long-standing assumption, yet there is little understanding of the turnover-quality relationship or its underlying mechanisms. When turnover occurs, the remaining staff must adjust to newcomers, and turnover may affect the interaction and integration among staff members who remain (Price, 1977; Staw, 1980). Researchers have also suggested potential positive conse40

quences of turnover, such as introducing fresh ideas and keeping the organization from becoming stagnant (Staw). Most empirical research on nursing turnover has focused on a direct relationship between turnover and patient outcomes; the underlying mechanisms of the turnover-outcomes relationship have not been explored (Alexander, Bloom, & Nuchols, 1994; Castle & Engberg, 2005; Voluntary Hospital Association Inc., 2002).
Journal of Nursing Scholarship, 2010; 42:1, 4049. c 2009 Sigma Theta Tau International

Bae et al.

Impact of Nursing Unit Turnover

In order to understand the mechanisms by which nursing turnover is related to patient outcomes, it is necessary to explore the impact of nursing turnover on the nursing unit, which is the proximal context for individuals and a bounded interactive context created by nurses attributes, interactions, and responses (Kozlowski, Steve, & Bell, 2003). We applied a conceptual framework at the nursing unit level to examine the impact of nursing unit turnover on workgroup processes (workgroup cohesion, relational coordination, and workgroup learning) as well as on patient outcomes (patient satisfaction, average length of patient stay, patient falls, and medication errors).

Model of Turnover Consequences

The proposed model was formulated around the inputprocess-outcome (IPO) framework (McGrath, 1964). The framework provides a model to assess workgroup behavior and performance effectiveness, and most models of workgroup effectiveness incorporate it (Kozlowski et al., 2003). Below, we provide more details about the model.

esized that higher nursing turnover would lead to lower workgroup cohesion and lower relational coordination. Workgroup learning. refers to relatively permanent changes in the knowledge associated with the experience of an interdependent set of individuals (Kozlowski et al., 2003). Researchers have suggested that turnover impacts learning both positively and negatively. Individuals leaving a workplace cannot transfer personal experience and knowledge to those who remain; therefore, workplace history lessons are lost and a portion of workgroup memory disappears (Carley, 1992). On the other hand, learning requires both stability and change in the environment. Too much stability and unchanging behavior within a workgroup can lead to stagnation rather than cognitive growth (Hedberg, 1981). We hypothesized that, relative to nursing units with high or low levels of registered nurse (RN) turnover, nursing units with moderate levels of RN turnover would experience greater workgroup learning.

Unit-Level Patient Outcomes

Patient satisfaction. has been proposed by many as an indicator of nursing care quality (e.g., Ervin, 2006). Mrayyan (2006) dened patient satisfaction as the degree to which nursing care meets patients expectations of ideal care, including the art, technical quality, physical environment, availability, continuity, and efcacy of care. Greater cohesion among employees in customer service settings strengthens employee motivation to provide excellent service, which may lead to higher levels of customer (patient) satisfaction (Meterko, Mohr, & Young, 2004). Gittell (2002) found a strong relationship between relational coordination among care providers and patient satisfaction. Therefore, we hypothesized that lower nursing unit cohesion and lower levels of relational coordination would be associated with lower levels of patient satisfaction. Length of stay. (LOS) is often used as a measure of hospital efciency (Halter, 2006). Well-coordinated workgroups are expected to produce higher-quality outcomes and to do so more efciently (Gittell, 2002). Several researchers have concluded that improved nursephysician coordination and communication can reduce hospital LOS (e.g., Halter). In highly cohesive and coordinated nursing units, healthcare providers are able to better communicate information about patients and provide responsive care for patients clinical conditions. Therefore, we hypothesized that lower relational coordination would be related to longer LOS. Patient falls and medication errors. harm patient safety and cause hospitals to lose millions of dollars (Bates et al., 1997). A well-coordinated staff may
41

Workgroup Processes
Workgroup processes represent mechanisms that enable or inhibit the ability of team members to combine their capabilities and behavior (Kozlowski et al., 2003). Workgroup cohesion, relational coordination, and workgroup learning represent the domains of affective, behavioral, and cognitive workgroup processes, respectively (Kozlowski et al.). Workgroup cohesion is commonly dened as the overall attraction or bond among members of a group (Mullen & Cooper, 1994). Distinct from formal coordinating mechanisms such as rules and manuals, relational coordination is a spontaneous form of coordination, encompassing patterns of communication and relationships (Gittell, 2002). Turnover may negatively affect both workgroup cohesion and relational coordination. As increasing numbers of nurses leave the unit, those who remain may feel abandoned and question their own motivations for staying. It may trigger additional turnover, detachment, and a search for salient alternative memberships (Staw, 1980; Kovner, Brewer, Greene, & Fairchild, 2009; Brewer, Kovner, Greene, & Cheng, 2009). Similarly, when valued employees leave, communication ow and established relationships are disrupted (Price, 1977). In a nursing unit with frequent turnover, increased adjustment time is required for new staff, and the remaining nurses may need to be particularly careful when supervising new staff. Thus, as nursing unit turnover increases, relational coordination is not easily achieved. Therefore, we hypoth-

Impact of Nursing Unit Turnover

Bae et al.

recognize potential patient risks early and provide interventions to lower risks for patient falls. Coordination (e.g., openness and mutual understanding) between nurses and physicians and nurses and pharmacy personnel have been found to inuence appropriate drug selection and administration (Schmidt & Svarstad, 2002; Kopp, Erstad, Allen, Theodorou, & Priestley, 2006). Workgroup learning is also important for patient safety. Workgroups learn from their failures through assessing and changing work processes in response to past errors and thus perform better in the long run (Kaissi, 2006; Edmondson, 1999). However, when learning from errors does not occur, it not only discourages nurses from revealing and discussing them, but it can allow errors to remain uncorrected. We hypothesized that lower levels of relational coordination and workgroup learning would be associated with increased patient falls and medication errors.

vide better care than do nonteaching hospitals (Ayanian & Weissman, 2002). Evidence suggests that patient outcomes are positively associated with nurse education level, unit tenure, and care hours provided by RNs (Kane, Shamliyan, Mueller, Duval, & Witt, 2007). In order to control for patient severity, this study used patient age, perceived health status, and prior year hospitalizations.

Methods
Design and Sample
A secondary data analysis was conducted with data from the Outcomes Research in Nursing Administration Project II (ORNA II), a nonexperimental, longitudinal causal modeling study (Mark et al., 2007) that used the nursing unit as the unit of analysis. The ORNA II study was undertaken to investigate relationships among RN stafng adequacy, work environments, and organizational and patient outcomes. The ORNA II sample consisted of RNs and patients on 286 general and specialty medical-surgical nursing units from 143 randomly selected acute care hospitals throughout the United States from the Joint Commission accredited acute care facilities with at least 99 licensed beds. Nurse managers in each nursing unit provided data about nursing turnover, LOS, patient falls, and medication errors. All RNs in each nursing unit who had worked on that unit for at least 3 months were invited to participate in the study. Ten patients who were 18 years of age or older, able to speak English, and hospitalized on the unit for at least 48 hours were randomly selected from each participating unit to complete a patient satisfaction survey. ORNA II data collection began in 2003 and ended in 2004. Because of missing values for the selected variables, the nal dataset for the current study consisted of 268 nursing units from 141 hospitals.

Mediating Effects of Workgroup Processes

We hypothesized that workgroup processes mediate the nursing unit turnoverpatient outcomes relationship. Nursing turnover leads to changes in workgroup processes such as decreased member attraction to the nursing unit, ineffective coordination, and inaccurate communication. Such inefcient workgroup processes in turn negatively affect patient outcomes (e.g., patient satisfaction and LOS). On the other hand, compared to nursing units with very low or very high levels of turnover, units with moderate levels experience a balance between constancy and change (Hedberg, 1981). Workgroups can both learn from past mistakes and be creative and exible. This may help prevent patient falls and medication errors.

Nursing Unit, Hospital, Nurse, and Patient Characteristics

Nursing unit, hospital, nurse, and patient characteristics were included as control variables. Work complexity at the unit level has implications for workgroup processes (Argote, Insko, Yovetich, & Romero, 1995; Gittell, 2002) and it also contributes to work conditions that affect efciency and quality of care (Dunton, Gajewski, Taunton, & Moore, 2004). Unit size has been linked to nursing unit performance (Mark, Salyer, & Wan, 2003). In this study, characteristics of the hospital included hospital size, technological sophistication, and teaching status. Larger organizations with better support systems for patient care have been shown to increase the resources dedicated to improving quality of care and efciency (Kuhn, Hartz, Gottlieb, & Rimm, 1991). Advanced technological services also have been linked to quality of care (Kuhn et al.). For common conditions, teaching hospitals generally pro42

Data Collection
Data used for this study were collected over 6 consecutive months with data on input variables collected prior to process and outcome variables. In order to support the temporal ordering implied by the conceptual framework, we used data from selected months. The data gathered on nursing unit turnover in January and February were obtained prior to collection of the workgroup process data in March and patient outcomes (LOS, patient falls, and medication errors) data in April, May, and June. Similarly, the nursing unit turnover data from March and April were collected prior to obtaining data on workgroup cohesion in May and patient satisfaction data in June. This data collection time sequence allowed

Bae et al.

Impact of Nursing Unit Turnover

Table 1. Descriptive Statistics for Study Variables Variables Nursing unit turnover Nursing unit turnover % (JanFeb) Nursing unit turnover % (MarApr) Workgroup processes Workgroup cohesion Coordination with other healthcare providers Coordination with physicians and pharmacists Workgroup learning Patient outcomes Patient satisfaction Average length of patient stay Patient falls Medication errors Control variables Work complexity Unit size Hospital size Technological sophistication Teaching status Education level (proportion with BSN) Unit tenure (mo) RN hours (%) Patient age (yr) Health status Previous hospitalization Denition Means SD ICC(1) R wg

Crude turnover rates during January and February Crude turnover rates during March and April Nurse job satisfaction scale (Sauter et al., 1997) Relational coordination scale (Gittell, 2002) Error orientation questionnaire (Rybowiak et al., 1999) Patient satisfaction questionnaire (Bacon & Mark, 2009) Total number of patient days/patient discharges Total number of patient falls per 1,000 patient days Total number of medication errors per 1,000 patient days Work complexity (Salyer, 1996) Total number of nursing unit beds Total number of maintained beds Saidin index Ratio of medical and dental residents to the number of hospital maintained beds Proportion of nurses with a bachelors degree or higher Average months of nurses tenure on the unit Percentage of nursing care hours delivered by RNs Average age Patients perceived health status (ve categories) Hospitalization (yes/no)

4.29 4.58 4.38 3.64 3.70 3.79 3.43 4.51 4.03 0.77 3.84 33.59 346.55 4.62 0.13 0.37 74.39 61.87 56.76 3.46 0.53

6.47 6.43 0.45 0.20 0.22 0.32 0.22 1.06 2.36 1.31 0.49 11.46 188.13 1.82 0.25 0.194 32.63 14.37 7.53 0.44 0.21 0.13 0.09 0.13 0.08 0.07 0.83 0.99 0.97 0.84 0.93

0.15

0.72

Note. N=268. SD, standard deviation; ICC, intraclass correlation; R wg , interrater agreement.

us to examine causality in the turnover-process-outcome relationship.

Measures
Variable denitions are displayed in Table 1. In general, measurement of variables was straightforward. However, the approach to input, process, outcome, and some control variables requires additional explanation. Input variables. Crude turnover rates of RNs on each nursing unit constituted the main input variable. The formula for calculating turnover rates is a fraction, where the numerator is the total number of RNs who left a nursing unit during a given period and the denominator is the average number of RNs on staff in the unit over the same period. In order to test the hypotheses, two functional forms of turnover rates were used: a linear function and a dummy variable. We used a linear turnover term to test the model except for the nonlinear relationship between turnover and workgroup learning, which was tested by a turnover dummy variable. Because of a dearth of research available on the benecial levels of turnover on workgroup learning, we constructed the dummy variable using ve groups based on a sufcient sample size of

nursing units for each group. The ve groups were categorized as follows: zero (reference group), low, moderate, high, and very high. These categories were dened by rate ranges: 0.00% (119 units), greater than 0.00% to 3.30% (24 units), 3.31% to 4.50% (24 units), 4.51% to 7.50% (49 units), and greater than 7.50% (52 units). Process variables. Workgroup cohesion was measured using four items of a cohesion subscale from the Nurse Job Satisfaction Scale (Sauter et al., 1997) that assesses perceptions about how well nurses work together and get along, which meets the denition of workgroup cohesion and has been used in other studies (Chang, Hughes, & Mark, 2006). Principal axis factoring yielded a single-factor solution with all items loading greater than 0.50. The internal consistency reliability of the four items in the current study was 0.76. Relational coordination was measured by the Relational Coordination Scale (Gittell, 2002) encompassing four communication dimensions (frequent, timely, accurate, and problemsolving communication) and two relationship dimensions (shared goals and shared knowledge). Two variables were created: one to measure nurse perceptions about coordination with nine healthcare provider disciplines (attending MDs, house staff, physical therapists, respiratory
43

Impact of Nursing Unit Turnover

Bae et al.

therapists, laboratory technicians, case managers/social workers, pharmacists, radiologists, and dietary staff) and a second to measure coordination with physicians and pharmacists only. The latter variable was created because coordination with physicians and pharmacists was considered the critical factor in preventing medication errors. Principal axis factoring yielded a single-factor solution for both variables. Cronbachs alphas for relational coordination with other healthcare providers and for coordination with physicians and pharmacists were 0.95 and 0.87, respectively. In this study, the denition of workgroup learning was limited to learning from errors and failures associated with patient safety because we were interested in mistakes as an outcome. Workgroup learning was measured using a subscale (5 items) from the Error Orientation Questionnaire (Rybowiak, Garst, Frese, & Batinic, 1999) that measures the degree to which unit members actively think about and diagnose the source of errors. Principal axis factoring conrmed that the scale had only one factor, with all ve items having factor loadings greater than 0.50. Cronbachs alpha for this scale was 0.92. Outcome variables. Patient satisfaction includes satisfaction with overall courtesy, friendliness of the nursing staff, and promptness of nursing assistance (Bacon & Mark, 2009). Patients completed a 13-item Likert-type questionnaire with four response options. This scale had a Cronbachs alpha of 0.92. Average LOS was dened as the average number of inpatient days of care for patients (the total number of patient days in each unit divided by patient discharges) on the nursing unit. A patient fall was dened as an unplanned descent to the oor. Patient falls were measured by the total number of patient falls reported for each unit divided by the number of patient days. Because errors resulting in severe outcomes are less likely than other errors to go unreported, this study used a measure of medication error frequency that included only those errors that required increased nursing observation or medical intervention for patients. The rate of patient falls and medication errors were measured as the number of incidents per 1,000 patient days. Control variables. Measurement of control variables was straightforward. However, work complexity and technological sophistication need additional explanation. Work complexity was measured using a 7-item Likert-type scale developed to measure perceived environmental uncertainty (Salyer, 1996). Within a nursing workgroup, environmental uncertainty (e.g., types and volumes of patients) may provide more complexity in workgroup dynamics and work process among nurses. This scale measures work complexity in terms of the extent of frequent interruptions or unanticipated events. Cronbachs alpha for the scale in the current study was
44

0.85. Technological sophistication was measured by using the Saidin Index (Spetz & Baker, 1999), which is the weighted sum of the number of technologies and services available in the hospital. The index increases more with the addition of technologies that are relatively rare than with technologies that are more common.

Data Analysis
The unit of analysis in this study was the nursing unit. Workgroup cohesion, relational coordination, workgroup learning, patient satisfaction, and work complexity were measured at the individual level and required aggregation to the unit level. Intraclass correlation coefcients [ICC(1)] and indices of interrater agreement (R wg ) were used to justify the aggregation of lower-level data to higher units of analysis. While the R wg is used in the event that observed group variances differed from some theoretically expected random variance, ICC(1) assesses how within-group variance contrasts with between-group variance. The common threshold for such justication is an R wg value equal to or greater than 0.70, and a larger ICC(1) is generally accepted as indicating a greater similarity among raters. Power was assessed using Cohens power tables. The realistically observed minimum effect size is any change in R-squared that ranges from 0.03 to 0.04. According to Cohen (1988), these effect sizes lie between small (0.02) and medium (0.15). The current analysis with a sample size of 268 units had sufcient power to capture the direct effects of turnover as well as the mediating effects of workgroup processes. The current study used both linear and count models, depending on the distribution of the process and outcome variables. Using the Breusch-Pagen and Hausman tests (Greene, 2003), ordinary least squares (OLS) estimators were compared to random and xed effects estimators. The specication tests strongly suggested that the clustering of nursing units within hospitals did not inuence the effect of nursing unit turnover on workgroup processes. Thus, a simple OLS model with robust standard errors was used for the workgroup process models. After the specication tests, OLS estimates were used for the patient satisfaction model. To account for the clustering of nursing units within hospitals, average LOS was estimated by using a random effects model. For patient falls and medication errors, a Poisson regression model with an adjustment for over dispersion was used.

Results
The means, standard deviations, and, where appropriate, ICC(1) and R wg for the study variables are presented

Bae et al.

Impact of Nursing Unit Turnover

Table 2. Effects of Nursing Unit Turnover on Workgroup Processes Relational coordination with other healthcare providers 0.003 (0.002) 0.008 (0.005) 0.085 (0.058) 0.183 (0.060) 0.058 (0.052) 0.062 (0.059) 0.145 (0.062) 0.000 (0.002) 0.000 (0.000) 0.023 (0.020) 0.122 (0.116) 0.315 (0.147) 0.001 (0.001) 0.002 (0.002) 4.562 (0.309) 0.107 2.96 0.106 (0.030) 0.001 (0.001) 0.000 (0.000) 0.018 (0.010) 0.082 (0.058) 0.071 (0.071) 0.000 (0.000) 0.001 (0.001) 4.005 (0.131) 0.145 4.69 0.130 (0.032) 0.000 (0.001) 0.000 (0.000) 0.008 (0.010) 0.144 (0.068) 0.007 (0.075) 0.000 (0.000) 0.001 (0.001) 4.147 (0.140) 0.149 4.10 0.121 (0.047) 0.001 (0.002) 0.000 (0.000) 0.019 (0.014) 0.040 (0.107) 0.089 (0.130) 0.001 (0.001) 0.001 (0.001) 4.091 (0.218) 0.125 3.02 Relational coordination with physicians and pharmacists 0.001 (0.002)

Workgroup cohesion Nursing unit turnover (JanFeb) Nursing unit turnover (MarApr) Turnover JanFeb=0% (reference group) 0<Turnover JanFeb3.3% 3.3<Turnover JanFeb4.5% 4.5<Turnover JanFeb7.5 7.5<Turnover JanFeb Control variables Work complexity Unit size Hospital size Technological sophistication Teaching status Education level Unit tenure RN hours Constant R-squared F value

Workgroup learning

Note. N=268. Signicant at .05; signicant at .01. Coefcient estimates presented and standard errors in parentheses.

in Table 1. Parameter estimates for each model are provided in Tables 2 and 3.

However, this study did not nd any signicant difference in workgroup learning between the reference group and other turnover groups. In addition, work complexity had a negative impact on all three workgroup processes.

Effects of Nursing Unit Turnover on Workgroup Processes

Table 2 presents the effects of nursing unit turnover on workgroup processes. The relationship between the workgroup process variables and nursing turnover was not signicant (workgroup cohesion: =0.008, p=.09; relational coordination with other healthcare providers: =0.003, p=.08). Our analysis suggested that nursing units with turnover rates between 3.31% and 4.50% were likely to have lower levels of workgroup learning, by 0.18 points, than nursing units with 0% turnover.

Effects of Workgroup Processes on Patient Outcomes

Table 3 summarizes the effects of workgroup processes on patient outcomes. Workgroup cohesion ( =0.091, p<.01) and relational coordination with other healthcare providers ( =0.159, p=.03) were signicantly associated with patient satisfaction in the separate models. When we included workgroup cohesion and relational coordination in a single model, only workgroup
45

Impact of Nursing Unit Turnover

Bae et al.

Table 3. Effects of Workgroup Processes on Patient Outcomes Patient satisfaction Workgroup cohesion Relational coordination with other healthcare providers Relational coordination with physicians and pharmacists Workgroup learning Control variables Work complexity Unit size Hospital size Technological sophistication Teaching status Education level Unit tenure RN hours Patient age Health status Previous hospitalization Constant R-squared F value/Wald chi-squared 0.026 (0.113) 0.068 (0.029) 0.002 (0.001) 0.000 (0.000) 0.012 (0.010) 0.035 (0.061) 0.183 (0.073) 0.001 (0.000) 0.001 (0.001) 0.006 (0.002) 0.077 (0.031) 0.028 (0.070) 2.221 (0.356) 0.237 5.94 0.102 (0.132) 0.010 (0.006) 0.002 (0.000) 0.063 (0.054) 0.340 (0.334) 0.241 (0.347) 0.002 (0.002) 0.006 (0.005) 0.001 (0.009) 0.476 (0.145) 0.027 (0.309) 5.130 (1.610) 40.08 0.048 (0.069) 0.004 (0.003) 0.000 (0.000) 0.009 (0.023) 0.378 (0.163) 0.155 (0.170) 0.000 (0.001) 0.002 (0.002) 0.005 (0.005) 0.228 (0.076) 0.035 (0.165) 2.228 (0.842) 0.079 (0.028) 0.115 (0.071)

Average length of patient stay

Patient falls

Medication errors

0.486 (0.320)

0.068 (0.187) 0.873 (0.411) 0.581 (0.295) 0.133 (0.160) 0.008 (0.006) 0.001 (0.001) 0.045 (0.055) 0.336 (0.416) 1.239 (0.430) 0.003 (0.002) 0.006 (0.005) 0.011 (0.010) 0.105 (0.174) 0.443 (0.373) 1.115 (1.909)

Note. N=268. Signicant at .05; signicant at .01. Coefcient estimates presented and standard errors in parentheses.

cohesion was signicantly related to patient satisfaction. After assessing the relationships among cohesion, coordination, and patient satisfaction, we found that relational coordination had a positive indirect impact on patient satisfaction through workgroup cohesion. The random effects model for LOS revealed that relational coordination with other healthcare providers was not associated with average LOS. Relational coordination with physicians and pharmacists ( =0.873, p=.03) and workgroup learning ( =0.581, p=.03) were signicantly associated with medication errors. But these workgroup process variables were not associated with patient falls. In the model of medication errors, a 1-point increase in workgroup learning led to a 44% decrease in medication errors. Contrary to our expectation, the relationship between relational coordination with physicians and pharmacists and medi46

cation errors was positive. A possible explanation for this unexpected nding is that, in nursing units with higher levels of relational coordination, serious medication errors might be more likely to be reported compared to units with lower levels of coordination. In terms of control variables, increased work complexity was associated with decreased patient satisfaction. Higher levels of nurse education were related to lower levels of medication errors and decreased patient satisfaction. Involvement in teaching was associated with lower rates of patient falls.

Mediating Effects of Workgroup Processes

This study predicted that workgroup processes would mediate the effects of nursing unit turnover on patient

Bae et al.

Impact of Nursing Unit Turnover

outcomes. These mediation hypotheses required the testing of three equations: (a) the effects of nursing unit turnover on patient outcomes, (b) the effects of nursing unit turnover on workgroup processes, and (c) the combined effects of nursing unit turnover and workgroup processes on patient outcomes. To show mediation, all of these effects must be signicant, and the signicance of the associations between nursing turnover and patient outcomes must be reduced by adding workgroup processes to the model (Baron & Kenny, 1986). However, nursing unit turnover was only related to patient falls ( =0.297, p=.02). The results suggested that nursing units with turnover rates between 0% and 3.3% of turnover were likely to have lower levels of patient falls, by a 24% decrease, than nursing units with 0% turnover. Thus, it was not possible to test mediating effects of workgroup processes on the turnover-outcomes relationship.

Discussion
The most important nding of this study was the adverse impact of turnover on workgroup learning. Compared to nursing units with 0% turnover, units with between 3.31% and 4.5% turnover during January and February experienced lower levels of workgroup learning. It is important to take into consideration the consequences associated with decreased quality resulting from a lack of workgroup effectiveness and learning. As we mentioned, previous turnover research does not account for the underlying mechanism of the turnoveroutcomes relationship, focusing instead on the direct relationship (Alexander et al., 1994; Castle & Engberg, 2005; Voluntary Hospital Association Inc., 2002). Our ndings provide empirical evidence elucidating one process by which turnover negatively affects patient outcomes. Further turnover research needs to focus on workgroup processes as a consequence of turnover and an underlying mechanism of the turnover-outcomes relationship. The study ndings also support the need to increase workgroup cohesion and coordination to improve patient satisfaction. This nding is consistent with ndings from previous research and emphasizes the importance for patient satisfaction of positive affective and effective coordination among group members (Gittell, 2002; Meterko et al., 2004). Similarly, this study found that nursing units scoring higher on workgroup learning had fewer medication errors. This nding provides empirical evidence for current research regarding patient safety such as developing and sustaining nursing unit processes where nurses are encouraged to discuss and learn from their errors. Workgroup learning should receive further attention in research and practice to prevent medication errors.

Our study has several limitations. First, the proposed model assumed a lagged impact of turnover on processes and outcomes, which implies that turnover affects relational coordination a few months after turnover occurs. The 2-month period of time for collection of turnover data may not have been sufcient to assess true variation in and levels of turnover. Also, there is no way of knowing whether those 2 months represent typical patterns of nursing turnover. Alternatively, it is possible that the effect of turnover on workgroup processes is contemporaneous rather than lagged, that high levels of nursing turnover undermine workgroup cohesion at the same time that turnover occurs, and this contemporaneous effect was not explored in this study. Future studies should employ a longitudinal design with repeated measures of all variables. This would lead to knowledge regarding the period of time in which nursing turnover is likely to result in less effective workgroup processes. Another limitation stems from missing variables that might affect turnover, workgroup process, and patient outcomes. For example, managers support and supervision have been found to inuence workgroup processes and organizational effectiveness (Kozlowski et al., 2003) and also are associated with nurse intent to stay (Kovner et al., 2006). Further research on turnover needs to account for the role of managers in the model. This study used patient age, previous hospitalization, and perceived health status to control patient acuity. In future research, patient acuity needs to be controlled by a more comprehensive method. Additionally, future turnover researchers may consider the use of a moderator, a concept distinct from a mediator. The fundamental assumption of the IPO framework is that workgroup processes are the underlying mechanisms mediating the impact of nursing unit turnover on patient outcomes. Moderators affect the direction and strength of the turnover-outcomes relationship and could be used to explore the turnover-outcomes relationship and provide insight into the characteristics of the most at-risk nursing units.

Conclusions
The current instability in the nursing workforce implies adverse impacts on the continuity and quality of patient care. Research to examine and to better articulate the processes and outcomes associated with nursing turnover will be crucial if healthcare organizations are to meet these challenges under shortage conditions. The results of this study should encourage further research focused on how nursing unit turnover affects workgroup processes and patient outcomes. Future work
47

Impact of Nursing Unit Turnover

Bae et al.

related to the impact of nursing turnover on various outcomes may provide frontline nurse managers with practical information needed to address the challenges of turnover.

Acknowledgments
This work was supported by grant 5R01NR003149 from the National Institute of Nursing Research and was partially funded by the Graduate School at the University of North Carolina at Chapel Hill through a Dissertation Completion Research Fellowship.

Clinical Resources r Nursing Shortage Fact

Sheet, American Association of Colleges of Nursing. http://www.aacn.nche. edu/Media/FactSheets/NursingShortage.htm State Nursing Workforce Reports, American Association of Colleges of Nursing. http://www.aacn. nche.edu/Media/NsgWrkFrcReps.htm

References
Alexander, J.A., Bloom, J.R., & Nuchols, B.A. (1994). Nursing turnover and hospital efciency: An organization-level analysis. Industrial Relations, 33(4), 505. Argote, L., Insko, C.A., Yovetich, N., & Romero, A.A. (1995). Group learning curves: The effects of turnover and task complexity on group performance. Journal of Applied Social Psychology, 25(6), 512529. Ayanian, J.Z., & Weissman, J.S. (2002). Teaching hospitals and quality of care: A review of the literature. Milbank Quarterly, 80(3), 569593. Bacon, C.T., & Mark, B. (2009). Organizational effects on patient satisfaction in hospital medical-surgical units. Journal of Nursing Administration, 39(5), 220227. Baron, R.M., & Kenny, D.A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 11731182. Bates, D.W., Spell, N., Cullen, D.J., Burdick, E., Laird, N., Petersen, L.A., et al. (1997). The costs of adverse drug events in hospitalized patients. Adverse drug events prevention study group. JAMA, 277 (4), 307311. Brewer, C.S., Kovner, C.T., Greene, W., & Cheng, Y. (2009). Predictors of RNs intent to work and work decisions 1 year later in a U.S. national sample. International Journal of Nursing Studies, 46(7), 940956. Carley, K. (1992). Organizational learning and personnel turnover. Organization Science, 3(1), 2046. Castle, N., & Engberg, J. (2005). Staff turnover and quality of care in nursing homes. Medical Care, 43(6), 616626.
48

Chang, Y.K., Hughes, L.D., & Mark, B. (2006), Fitting in or standing out: Nursing workgroup diversity and unit-level outcomes. Nursing Research, 55(6), 373380. Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: L. Erlbaum Associates. Dunton, N., Gajewski, B., Taunton, R.L., & Moore, J. (2004). Nurse stafng and patient falls on acute care hospital units. Nursing Outlook, 52(1), 5359. Edmondson, A.C. (1999). Psychological safety and learning behavior in work teams. Administrative Science Quarterly, 44(2), 350383. Ervin, N.E. (2006). Does patient satisfaction contribute to nursing care quality? Journal of Nursing Administration, 36(3), 126130. Gittell, J.H. (2002). Coordinating mechanisms in care provider groups: Relational coordination as a mediator and input uncertainty as a moderator of performance effects. Management Science, 48(11), 14081426. Greene, W.H. (2003). Econometric analysis. Upper Saddle River, NJ: Prentice Hall. Halter, C.W. (2006). Inuence of hospital context on cost and length of stay following cardiac catheterization. Nursing Economic , 24(5), 246252. Hedberg, B. (1981). How organizations learn and unlearn. In P. Nystrom & W. Starbuck (Eds.), Handbook of organizational design (pp. 327). Oxford, UK: Oxford University Press. Kaissi, A. (2006). An organizational approach to understanding patient safety and medical errors. Health Care Manager, 25(4), 292305. Kane, R.L., Shamliyan, T., Mueller, C., Duval, S., & Witt, T. (2007). Nursing stafng and quality of patient care. AHRQ publication 07-E005. Rockville, MD: Agency for Healthcare Research and Quality. Kopp, B.J., Erstad, B.L., Allen, M.E., Theodorou, A.A., & Priestley, G. (2006). Medication errors and adverse drug events in an intensive care unit: Direct observation approach for detection. Critical Care Medicine, 34(2), 415425. Kovner, C.T., Brewer, C.S., Greene, W., & Fairchild, S. (2009). Understanding new registered nurses intent to stay at their jobs. Nursing Economics, 27 (2), 8198. Kozlowski, S.W.J., Steve, W.J., & Bell, B.S. (2003). Work groups and teams in organization. In W.C. Borman, D.R. Ilgen, & R.J. Klimoski (Eds.), Comprehensive handbook of psychology: Industrial and organizational psychology. New York: Wiley. Kuhn, E.M., Hartz, A.J., Gottlieb, M.S., & Rimm, A.A. (1991). The relationship of hospital characteristics and the results of peer review in six large states. Medical Care, 29(10), 10281038. Mark, B.A., Hughes, L.C., Belyear, M., Chang, Y., Hofmann, D., Jones, C. B., et al. (2007). Does safety climate moderate the inuence of stafng adequacy and work conditions on nursing injuries? Journal of Safety Research, 38(4), 431 446.

Bae et al.

Impact of Nursing Unit Turnover

Mark, B.A., Salyer, J., & Wan, T.T.H. (2003). Professional nursing practice: Impact on organizational and patient outcomes. Journal of Nursing Administration, 33(4), 224234. McGrath, J.E. (1964). Social psychology: A brief introduction. New York: Holt, Rinehart, & Winston. Meterko, M., Mohr, D.C., & Young, G.J. (2004). Teamwork culture and patient satisfaction in hospitals. Medical Care, 42(5), 492498. Mrayyan, M.T. (2006). Jordanian nurses job satisfaction, patients satisfaction and quality of nursing care. International Nursing Review, 53, 224230. Mullen, B., & Cooper, C.L. (1994). The relationship between group cohesiveness and performance: An integration. Psychosocial Bulletin, 115, 210227. Price, J.L. (1977). The study of turnover. Ames, IA: Iowa State University Press. Rybowiak, V., Garst, H., Frese, M., & Batinic, B. (1999). Error orientation questionnaire (EOQ): Reliability, validity, and different language equivalence. Journal of Organizational Behavior, 20(4), 527547. Salyer, J. (1996). Development and psychometric evaluation

of an instrument to measure staff nurses perception of uncertainty in the hospital environment. Journal of Nursing Measurement , 4(1), 3348. Sauter, M., Boyle, D., Wallace, D., Andrews, J., Johnson, M., Bates, M., et al. (1997). Psychometric evaluation of the organizational job satisfaction scale. Journal of Nursing Measurement , 5(1), 5369. Schmidt, I.K., & Svarstad, B.L. (2002). Nurse-physician communication and quality of drug use in Swedish nursing homes. Social Science & Medicine, 54(12), 17671777. Spetz, J., & Baker, L.C. (1999). Managed care, hospital competition, and the availability of technology: Implications for health care costs. San Francisco, CA: Public Policy Institute of California. Staw, B.M. (1980). The consequences of turnover. Journal of Occupational Behaviour, 1(4), 253273. Voluntary Hospital Association Inc. (2002). The business case for work force stability. No. 2002 VHA Research Series (7). Irving, TX. Retrieved November 9, 2009, from http://www.studergroup.com/tools/selection andretention/ TheBusinessCaseforWorkforceStability.pdf

49

Copyright of Journal of Nursing Scholarship is the property of Blackwell Publishing Limited and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.