Mark, Barbara A. PhD, RN, FAAN
Click on the links below to access all the ArticlePlus for this article.
Please note that ArticlePlus files may launch a viewer application outside of your web browser.
Nurses’ concerns about the adequacy of nurse staffing in acute care hospitals have been well documented in both the professional and in the lay press. 1,2 Yet, preliminary data from the most recent National Sample Survey of Registered Nurses 3 indicate that the number of registered nurses (RNs) increased from 2.6 million in 1996 to 2.7 million in 2000. However in 1996 1.51 million nurses worked full-time (of whom 60% worked in hospitals) in 2000 1.57 million nurses worked full-time (of whom 59% worked in hospitals). Further, the American Hospital Association (AHA) reports that from 1995 to 1999 (the most recent year for which data are available), the number of RNs increased from 1.08 million to 1.14 million. 4
Despite growth in the reported number of nurses, the “burden” of care for nurses has escalated. 5 Inpatient length of stay has declined substantially the average acuity of patients has risen. 6 Technological development expansion of knowledge allow more acutely ill patients to survive, further increasing the burden of care. 5 In its most recent staffing survey, the American Nurses Association (ANA) found that 74.1% of those surveyed report that their time had decreased for direct patient care, compared with only 9.7% who believed it had increased. 7. Further, theserespondents believed that quality declined because of inadequate staffing. In contrast a recent report of the California Healthcare Foundation, 8 reported that nurse staffing in so-called “best practices” California hospitals (those with reputations as “models for quality as good workplaces”) did not differ significantly from other California hospitals.
It seems intuitively obvious that patient acuity would be an important factor that affects nurses’ perceptions of the adequacy of staffing on their units. It is unlikely, however, that patient acuity alone adequately explains these perceptions. As part of the Outcomes Research in Nursing Administration Project (ORNA), we had the opportunity to complete a secondary analysis in which we examined various influences on nurses’ perception of staffing adequacy. Our purpose was to evaluate the extent to which selected hospital characteristics, nursing unit characteristics, nurse characteristics, and patient characteristics influenced nurses’ perceptions of staffing adequacy. Because perceptions are formed over time, we also were interested in whether perceptions of staffing adequacy were better explained by: 1) current conditions (ie, a cross-sectional model); 2) past conditions (ie, a delayed effects model); or 3) both current past conditions (ie, a dynamic model).
The theoretical framework for the ORNA project was based on structural contingency theory (SCT). SCT suggests that an organization’s performance depends on the fit between the organization’s context and how it is organized to accomplish its tasks. 9–11 Context is comprised of both the external and the internal environment of the nursing unit. To represent organizational context, we selected variables that described key hospital and nursing unit characteristics. 9 For hospital characteristics, we considered the hospital’s case mix index, its size, the number of “high tech” services offered, whether it was a teaching hospital, and the hospital’s pattern of admission volatility. Regarding nursing unit characteristics, we included the total number of nursing staff, the nursing skill mix, nursing work load, number of beds on the unit, availability of support services, patient technology, the educational level of the staff, and the unit’s pattern of admission volatility. Because the social information processing approach to task design suggests that individual characteristics affect perceptions, 12 we also measured two critical characteristics of nurses as individuals (experience and age). In addition, we included the average age of patients on the unit in our models. Our dependent variable was the average response on a nursing unit to a single question asking staff nurses about their perceptions of the adequacy of staffing on their units.
The ORNA project is a nonexperimental, longitudinal study conducted in 68 hospitals in the Southeastern United States, Texas, and the District of Columbia. After informed consent was obtained in all participating sites, 13 data on all variables were collected twice (referred to as Time 1 and Time 2) separated by an interval of 6 months.
The ORNA study began with 136 general medical-surgical nursing units in 68 randomly selected nonfederal, nonpsychiatric, not for profit, Joint Commission on Accreditation of Healthcare Organizations-accredited acute care hospitals with more than 150 beds. Due to attrition of two nursing units in one hospital before study implementation, we began the study with 134 nursing units. Further attrition during the first period of data collection reduced the sample to 127 units. Three additional units left the study during the second period of data collection and four hospitals provided data from only one nursing unit. Thus, the final sample size was 120 nursing units, 2 in each of 60 hospitals. On each unit, all RNs with more than 3 months’ experience were invited to participate. In addition, we selected a random sample of 10 patients on each unit who had been hospitalized a minimum of 48 hours, and who could respond to an English-language questionnaire. At Time 1, 90% of nursing units had more than 70% response rates (with 45% of units achieving 100% response rates). Staff nurse questionnaires were distributed (2,279), of which 1,749 were returned and 1,583 were useable (an initial response rate of 69.5%); of these nurses, 1,023 responded again at Time 2 (65% of those who responded at Time 1). At Time 1, response rates for patient data were greater than 80% on 80% of units; 1,346 patient questionnaires were returned, of which 1,231 were useable. During Time 2, during which 92% of units had better than 80% response rates, 1,235 patients responded.
Each hospital appointed a “study coordinator” who was responsible for the conduct of the study in that hospital. Study coordinators were provided 11/2 days of training by the research team to familiarize them with the purpose and goals of the study, to ensure conformity across multiple sites in approaches to data collection, and to assure consistency in the definition of key data elements, thus increasing reliability of data. 14
Hospital characteristics: Case mix index (CMI) was the Medicare case mix index assigned by the Healthcare Financing Administration. Hospital size was defined as the number of open and maintained beds. Technological complexity was the number of high technology services, out of 16 possible, offered by the hospital. Teaching status was defined based upon whether the hospital was a member of the Council of Teaching Hospitals.
Volatility of admissions was dummy coded and defined as follows: stable hospitals (the omitted category/reference group) were the hospitals in which admissions varied less than 5% in each of the 2 years prior to the beginning of the project. Growers were those where hospital admissions increased 5% or more in both years; hospitals were characterized as decliners if admissions decreased 5% or more in both years, and as unstable if the pattern of admissions displayed an increase greater than 5% in one year accompanied by a decrease greater than 5% in the other year. 15,16 These data were provided by the study coordinator, who retrieved the required information from appropriate hospital sources.
Nursing Unit Characteristics. Total nursing staff comprised all nursing staff (ie, RNs, licensed practical nurses [LPNs], unlicensed workers) who were assigned to the unit; float and temporary staff, clinical specialists, and nurse managers were not included in this total. Skill mix was defined as the proportion of the total nursing staff members who were registered nurses. Nursing work load was calculated by dividing the total number of inpatient days in the 6 months of the study by the number of full-time equivalent RNs on the unit. Unit size was defined as the number of beds on the nursing unit. Education was measured by the proportion of registered nurses on the unit who held a minimum of bachelors’ degrees. Unit admission volatility was measured analogously to hospital admission volatility, with admissions to the unit the key variable. Information about these variables was collected from nurse managers.
Availability of support services was evaluated with a 27-item checklist in which staff nurses indicated whether various support services were unavailable, inconsistently available, or consistently available (alpha = .85). 17,18 Many support services were included: laboratory specimen collection, patient transportation, computerized order entry, use of unit-dose medication system, and coordination of discharge planning. Higher scores indicate support services were consistently available. Data about patient technology, a measure of patients’ needs for nursing care (our proxy for patient acuity), were obtained from staff nurses using a 12-item scale with 5 response categories in which higher scores indicated more patients on the unit had complex nursing care needs (alpha = .76). We evaluated the factor structure of the scale with exploratory factor analysis. We used varimax rotation, maintained factors with eigenvalues greater than one, and summed over factors. Three factors, reflecting patient conditions that required comprehensive problem-solving by nurses, changed rapidly and unpredictably and required the use of technical equipment; explained 48% of the total variance in patient technology. Nursing units’ scores for availability of support services and patient technology were the averages of its staff nurses’ responses to the questionnaires. Because the nursing unit, not the individual nurse, was the level of analysis, before aggregation of individual-level measures, perceptual agreement was assessed with an index of within group agreement;19 values exceeded .80, assuring the appropriateness of aggregation.
Nurse characteristics: Experience was measured in categories reflecting number of years of nursing experience (less than 1 year, 1–5 years, 6–10 years, 11–15 years, and more than 15 years). Age was the nurse’s age in years.
Patient Characteristics: We had information on age, a single demographic characteristic of patients.
Perceptions of Adequacy of Staffing: Staffing adequacy was the unit’s average score calculated with data obtained from staff nurse responses to a single question asking them to evaluate the adequacy of staffing on their units (ie, very much above average, somewhat above average, average, somewhat below average, very much below average). Although there may be some concern about the use of a single item to measure the dependent variable, Youngblut and Casper 20 report that global single item indicators can provide valid and reliable measures.
Analysis: Data were analyzed using descriptive statistics and bivariate analysis. Observations in our sample are clustered by nursing units (measured twice), which in turn are clustered in hospitals. Clustering of this kind violates the assumption of independence required for ordinary least squares regression (OLS), since errors for observations from the same nursing unit or hospital are likely to be correlated. OLS estimates of the regression coefficients are unbiased, but the OLS estimates of the standard errors of those coefficients may be biased. 21 To address this problem, we use robust standard errors, correcting possible correlation across nursing units and hospitals. 22 Hence, the normalized (beta) regression coefficient estimates we report are the same as in standard OLS regression, but the estimates of the standard errors are robust to violation of the assumption of independent observations by cluster, and also robust to unspecified sources of heteroscedasticity 23
Because of significant nonlinearity in the relationship between case mix index and the dependent variable, we added a quadratic term (ie, case mix squared, CMI2) to each of the models. 24 No other nonlinear relationships were detected. Model testing proceeded in an incremental fashion corresponding to the three models we proposed. By using independent and dependent variables measured simultaneously, Model 1 (the cross-sectional model) evaluates how well perceptions of staffing adequacy are predicted by current conditions alone. By using independent variables from Time 1 to predict staffing adequacy at Time 2, Model 2 tests the “delayed effects” model. Finally, in Model 3, the dynamic model, we incorporate both current and past conditions by adding the lagged value of the dependent variable (that is, perceptions of staffing adequacy measured at Time 1) as an independent variable to predict perceptions of staffing adequacy at Time 2. The dynamic model addresses concern about the possibility of unknown sources of omitted variable bias, particularly unmeasured historical influences. For example, consider the effect of a recent cut in staffing just before data collection at a hospital that had above average staffing before the cut. Nurses in such a unit might have lower perceptions of staffing adequacy than a similarly staffed unit that had not suffered a cut. Including the lagged value of the dependent variable as an independent variable can control such unmeasured sources of omitted variable bias. This approach also is preferred because it often is more useful for policy analysis: Suppose Model 1 or Model 2 (ie, models without the lagged value of the dependent variable) indicates that change in one of the independent variables should lead to a change in perceptions of staffing adequacy. If there are unmeasured historical influences omitted from the regression model, then a change in that variable may not, in fact, lead to the predicted change in staffing adequacy. Thus, Model 3, with the lagged value of the dependent variable, may be superior for policy analysis because it controls such unmeasured factors and because it indicates the extent to which perceptions of staffing adequacy are not wholly determined by contemporaneous conditions.
Table 1 illustrates the mean and standard deviations of all continuous variables at both Time 1 and Time 2. Between Time 1 and Time 2, work load decreased significantly (paired t119 = 2.83, P = .006), due primarily to a decrease in the number of patient days, rather than to an increase in the number of RNs. Case mix index increased significantly (t119 = −4.26, P = .000); as did patient age (t119 = −2.97, P = .004). Nurses’ level of experience also increased (t119 = −7.3, P = .000), reflecting less experienced nurses’ greater tendency to turnover. 25
At time 1, perceptions of better staffing adequacy were associated with the number of high technology services (r = .216, P = .018), with the hospital’s case mix index (.206, P = .024), and better nursing skill mix (.204, P = .025). Perceptions of less adequate staffing were associated with work load (-.221, P = .015), and with larger unit size (-.231, P = .011). At Time 2, perceptions of better staffing adequacy were again significantly associated with the number of high technology services (.278, P = .002), although the number of beds on the unit (-.309, P = .001), and patient technology, a measure of patients’ needs for nursing care (-.198, P = .030) were associated with perceptions of poorer staffing adequacy.
Table 2 provides the regression coefficients, levels of significance, F-tests, and variance explained for Models 1 through 3. In Model 1 (the cross-sectional model), the hospital characteristics of CMI (β = .980, P = .011) and CMI2 (β = −0.852, P = .028) significantly predicted nurses’ perceptions of staffing adequacy. The opposing signs on these variables indicate that the effect of an increase on staffing perceptions depends on the level of case mix intensity. At lower levels of CMI, a rise in the index improves perceptions of staffing adequacy, but each successive unit increase results in a smaller increase in perceptions of staffing adequacy. For this model, the improvement in perceived staffing adequacy per unit increase in the CMI is positive, but diminishing, up until CMI equals 1.688. At that point, further increases in CMI lead to decreases in perceptions of staffing adequacy. Perceptions of staffing adequacy were positively associated with the hospital being a grower (β = .140, P = .021). Two unit characteristics: unit size (β = −0.489, P = .003) and patient technology (our proxy for acuity) (β = −0.341, P = .000) were significantly associated with perceptions of less adequate staffing. The consistency of support services’ availability (β = 0.191, P = .032) was associated with perceptions of better staffing adequacy. Neither the nurse nor the patient characteristics were significantly associated with perceptions of staffing adequacy. This model explained 34.8% of the variance in staffing adequacy.
Model 2 (the delayed effects model) has the same pattern of signs for the CMI (positive) and CMI2 (negative) coefficients, but in Model 2, only the coefficient for CMI is significantly different from zero (β = 0.769, P = .024). Similar to Model 1, the hospital’s being defined as a “grower” improved nurses’ perceptions of staffing adequacy (β = .153, P = .025). Consistent with the Time 1 model, the number of beds on the unit (β = −.542, P = .000) and patient technology (β = −.216, P = .028) significantly reduced perceptions of the adequacy of nurse staffing. In Model 2, the total number of staff (β = 0.323, P = .039) and skill mix (β = .340, P = .034) had a statistically significant effect on perceptions of staffing adequacy; note, however, that the coefficients for these two variables were similar to the coefficients in Model 1, in which the coefficients were not significant. None of the nurse characteristics or patient age was a significant predictor of staffing adequacy. Model 2 explained 32.5% of the variance.
Model 3, the dynamic model, which includes the lagged value of the dependent variable, explained 51.2% of the variance. In this model, the hospital’s being characterized as a “grower” (β = .101, P =.025) contributed significantly to better perceptions of staffing adequacy. The nursing unit characteristics of unit size (β = −.331, P = .026), and patient technology (β = −.293, P = .001) were associated with perceptions of poorer staffing adequacy. In addition, the Time 1 lagged variable of staffing adequacy was strongly associated with Time 2 perceptions of staffing adequacy (β =.471, P = .000). Again in Model 3, the coefficients for CMI and CMI2 have opposites signs (β = .859, P = .018; β = −.794, P = .038, respectively). In this model, the inflection point (ie, the point at which further increases in the CMI result in perceptions of worse staffing was 1.69), compared with the average CMI at Time 2 of 1.42.
Across all three models, the hospital characteristics of being defined a “grower” and having a more complex mix of patients had consistent positive relationships to perceptions of staffing adequacy. In addition, having a more complex case mix has a positive relation to perceptions of staffing adequacy, but only up to a point. After this point, perceptions of the adequacy of staffing decrease. Across models, the two nursing unit characteristics of the number of beds and patient technology had consistent negative effects on perceptions of staffing adequacy. Availability of support services was only significant in Model 1 and measures of staffing (ie, total staff and skill mix) were only significant in Model 2. Neither the nurse characteristics (experience or age) nor patient age was significant in any of the four models. The lagged value of the dependent variable in Model 3 was significant.
Nurses believe unquestionably that there is a staffing crisis and that poor staffing is coupled with decreased quality of care. 1,2 It also is clear that we are in the midst of a worsening shortage of nurses. 26 According to the American Hospital Association, 126,000 nurses are needed to fill currently vacant positions. 4 Further, 89% of hospitals surveyed perceive that there are more job openings than RNs available to fill those jobs. 4 Frequently suggested policies to address the shortage are directed at increasing the overall supply of nurses through innovations in regulation and licensing, media campaigns to highlight the worth of nursing, reform in nursing education, and improvement of the organization of nursing services in hospitals. The outcomes of these policy changes, if successful in increasing the aggregate supply of RNs, are at best long-term and do not address nurses’ current concerns.
Our findings suggest several avenues for nurse executives and nurse managers to pursue. First, across all models, and with other variables controlled, the relationship of the number of beds on the unit to perceptions of staffing adequacy was consistent and negative in direction: larger unit size was associated with perceptions of less adequate staffing. When we categorized units into quartiles based on the number of beds on the unit, perceptions of staffing adequacy for units in the lowest quartile (ie, those with less than 27 beds, and which, on average, had 21 beds) were significantly higher (significance of Scheffe test P < .007) than those in the highest quartile (ie, those with more than 40 beds, and which, on average, had 51 beds). Although perceptions of staffing adequacy decreased in a linear fashion based on unit size, only the comparison between the smallest and largest units was significant. Although this suggests that on average units with fewer than 27 beds may be optimal, additional research is needed before firm conclusions can be reached. Such research should take into account additional factors, such as units’ architectural and spatial design, occupancy rates, use of computerized information systems, ergonomic and other human factors, and engineering components. These factors may affect how nurses perform their work; thus, they may have an impact on their perceptions of the adequacy of staffing on their units.
Second, higher levels of patient technology (our proxy for patient acuity) also was associated, across all models (and with other variables controlled) with perceptions of less adequate staffing. Although this finding seems intuitively evident (nurses on units with sicker patients thought staffing was less adequate), and though it is impossible for managers to alter patient acuity, there may be strategies to address this issue. Although not included in the analysis described here, we have reported that higher patient acuity is associated with lower levels of nursing satisfaction. 16 Taken together, these findings suggest that comprehensive new employee unit-based orientation and continuing education programs, strengthened support systems for new staff through preceptor programs, continued opportunities for staff nurses to interact with others in their cohort, and provision of individualized stress management programs, may help nurses cope with the increasingly serious illnesses experienced by patients. Therefore, they may be useful in altering negative perceptions. Other interventions, (eg, computerized order entry, the use of automated medication dispensing devices, mechanical devices to assist with patient lifting, and use of hand-held personal data assistants) may streamline the provision of physical care and thus reduce some of the burden on staff. Perhaps these interventions may allow nurses more time to comfort patients and families (a critical aspect of professional nursing care, and one that is frequently left undone). 27 A fruitful area for future research would be to investigate the relationships among patient acuity, perception of staffing adequacy, nursing satisfaction, and turnover. If causal relationships can be documented, it would be important to interrupt a potentially downward spiraling process in which high levels of patient acuity lead to perceptions of poor staffing, which in turn lead to dissatisfaction and turnover, generating further negative impacts on remaining nurses’ perceptions of the adequacy of staffing on their units.
In all the models, growth in admissions during the prior 2 years was positively associated with nurses’ perception of staffing adequacy. Further, nursing units in “grower” hospitals were the only units to report more beds at Time 2 than at Time 1, an increase of just over 1 bed, from 31.5 beds to 32.7 beds. Nursing units in “declining” hospitals reported the lowest proportion of RNs to total nursing staff. Further, although “unstable” hospitals and “growers” were almost identical in unit size (approximately 32 beds), “growers” had more staff compared with “unstable” hospitals (36.8 FTEs vs 29.7 FTEs). In addition, nursing unit skill mix of “growers” increased from 55% at Time 1 to 59% at Time 2. A likely explanation for this finding is that a key organizational requirement allowing the institution to support such growth included provision of ample nursing staff to care for patients. However, some caution is warranted, because only four hospitals were growers.
Less consistent were our findings concerned with the relationships among total staff, skill mix, and perception of staffing adequacy, which were significant only in Model 2 (ie, the delayed effects model). This finding supports our conclusion that nurses’ perception of the adequacy of staff on their units depends on more than just the number and mix of personnel.
We failed to find relationships between nurses’ perception of staffing adequacy and two critical variables (availability of support services, which was significant in only one model, and work load, which was not significant in any of the models). Rather than conclude that these relationships do not exist, however, it is more likely that the failure to find relationships may be due to problems in measurement. Although our questionnaire to measure the availability of support services exhibited good reliability (alpha = .85), it is possible that the designated support services (examples include venipuncture/ blood specimen collection, patient transport, scheduling post-discharge appointments, unit dose medications, feeding patients) did not include those with the most direct impact on nurses’ work and their perceptions of staffing adequacy. Further study would be necessary to identify those services nurses believe are most critical.
A concern for measurement also extends to our operationalization of work load, which we based on concepts of production efficiency in which higher efficiency is suggested by more units of output (patient days) per unit of input (RNs). The measure does not take into account the uncertainty and unpredictability of patient care on the unit level, nor does it address differences in perceptions across shifts. For example, Norrish 28 found that nurses reported patient turnover rates of 40% to 50% in a single shift. Lawrenz 29 found similar unpredictability with the number of admissions, discharges, and transfers averaging from 25% to 70% of midnight census. Since our work load measure was derived from patient days (which is formulated from midnight census), the measure probably underestimates nursing work load substantially. Unfortunately, until there is a comprehensive, reliable and well-validated measure of nursing work load that can be used across multiple settings, such standardized, but limited measurement approaches may be necessary.
Models 1 and 2 (the cross-sectional and delayed effects models) explain only 34.8% and 32.5% of the variance in nurses’ perceptions of staffing adequacy, suggesting that a considerable portion of the variance remains unexplained. Model 3, the dynamic model that includes the lagged value of the dependent variable, therefore, is our preferred specification. The large beta weight (.471) associated with the lagged perceptions variable in Model 3 indicates that nurses’ perceptions of staffing adequacy at Time 1 had a strong impact on their perceptions of staffing adequacy at Time 2. These findings suggest that administrators who choose to make changes in the areas we have described earlier should realize that changes made currently not only affect current perceptions, but in the future also. Thus, sufficient time must be allowed for changes to take effect.
Norrish 1 points out that nurse staffing levels are constantly moving targets. Unceasing fluctuations in patient volume, both at the hospital level and at the unit level, create unpredictability in nursing work load. When volume increases, more workers must be found, agency and other temporary workers may be hired, and/or overtime may be offered or mandated. Conversely, when volume decreases, staff may be subject to random days off, with or without loss of pay. These conditions may influence nurses’ perceptions of the adequacy of staffing, independent of their actual work load, the actual number of staff available, or the skill mix of the staff. Because accurate predictions of work load may be nearly unattainable in the chaotic environment of the current acute care hospital, it may be that nurses’ perceptions of the adequacy of staffing suffer even in those hospitals in which administrators are engaged various activities to provide the best possible staffing. In these instances, continuing to develop and to maintain open avenues of communication and enhancing employees’ trust (both in the organization and through relationships with its administrators) may be critical complements to other, more concrete strategies, in addressing negative perceptions about staffing adequacy.
© 2002 Lippincott Williams & Wilkins, Inc.