Using Minimum Nurse Staffing Regulations to Measure the Relationship Between Nursing and Hospital Quality of Care

Abstract

This study tests whether changes in licensed nurse staffing led to changes in patient safety, using the natural experiment of 2004 California implementation of minimum staffing ratios. We calculated counts of six patient safety outcomes from California Patient Discharge Data from 2000 through 2006, using the Agency for Healthcare Research and Quality Patient Safety Indicators (PSI) software. For patients experiencing nonmortality-related PSIs, we measured mean lengths of stay. We estimated difference-in-difference equations of changes in PSIs using Poisson models and calculated the marginal impact of nurse staffing on outcomes from fixed-effect Poisson regressions. Licensed nurse staffing increased in the postregulation period, except for hospitals in the highest quartile of preregulation staffing. Growth in registered nurse staffing was associated with improvement for only one PSI and reduced length of stay for one PSI. Higher registered nurse staffing per patient day had a limited impact on adverse events in California hospitals.

A substantial evidence base supports the relationship between higher hospital nurse staffing levels and better patient outcomes (Aiken, Clarke, Sloane, Sochalski, & Silber, 2002; Kane & Shamliyan, 2007; Lang, Hodge, Olson, Romano, & Kravitz, 2004; Needleman, Buerhaus, Mattke, Stewart, & Zelevinsky, 2002). Although there is consensus that nursing is an important factor in ensuring quality of patient care (Institute of Medicine, 2011), very few studies have identified a causal relationship between nurse staffing levels and patient outcomes (Kane & Shamliyan, 2007). Most studies have been cross-sectional and thus subject to the possibility of reverse causation, leaving open the question of whether the observed staffing–quality relationship is associative or causal.

The gold standard for determining the effect of nurse staffing on outcomes would be the random assignment of staffing levels to different hospitals. Since this is not possible, we must rely on observational research to determine whether richer nurse-to-patient ratios lead to improvements in quality of care (Donaldson & Shapiro, 2010). Regulatory changes can provide an opportunity to examine whether there might be a causal relationship between nurse staffing and patient outcomes. Nurse staffing regulations have been established in 14 states plus the District of Columbia since 1999 (American Nurses Association, 2011). California is the only state to have fixed minimum licensed nurse-to-patient staffing ratios, which were implemented in 2004 after a lengthy implementation process (Coffman, Seago, & Spetz, 2002). Prior to the establishment of the regulation, there was significant variation in nurse staffing across California hospitals, with some hospitals staffing more richly than required and others with substantial short-falls (Coffman & Spetz, 1999; Kravitz et al., 2002; Spetz, 2011; Spetz, Seago, Coffman, Rosenoff, & O’Neil, 2000). Hospital nurse staffing increased after the announcement of the specific proposed ratios in early 2002 (Bolton et al., 2007; Chapman et al., 2009; Conway, Konetzka, Zhu, Volpp, & Sochalski, 2008; Cook, Gaynor, Stephens, & Taylor, 2012; Donaldson et al., 2005; Donaldson & Shapiro, 2010; Serratt, Harrington, Spetz, & Blegen, 2011; Spetz et al., 2009). Between 1999 and 2006, statewide mean registered nurse (RN) hours per patient day rose 16.2%, to a mean of 6.9 hours per patient day; increases in staffing ratios continued after 2004 because the ratios were phased in for medical–surgical, telemetry, and step-down units between 2004 and 2008 (Chapman et al., 2009). Hospital nurse staffing increased significantly more in California in the years after the legislation than in other states (Mark, Harless, Spetz, Reiter, & Pink, 2012; McHugh, Kelly, Sloane, & Aiken, 2011).

If higher nurse staffing ratios have a direct relationship with better quality of patient care, then we would expect larger improvements in quality of care in hospitals that had greater increases in nurse staffing ratios (Mark, Harless, McCue, & Xu, 2004). Four previous studies using longitudinal data have examined whether California’s minimum nurse-to-patient staffing regulations affected patient outcomes (Bolton et al., 2007; Cook et al., 2012; Donaldson et al., 2005; Mark et al., 2012). Three of these studies found no relationship between increased nurse staffing and patient outcomes (Bolton et al., 2007; Cook et al., 2012; Donaldson et al., 2005), but these examined only a limited set of patient outcomes. Two of the articles analyzed a convenience sample of California hospitals that participate in the Collaborative Alliance for Nursing Outcomes (CALNOC; Bolton et al., 2007; Donaldson et al., 2005). One longitudinal study of California’s regulations reported mixed effects, with improvement in the rate of failure-to-rescue after a postoperative complication, no change in rates of postoperative respiratory failure or sepsis, and a worsening of infections related to medical care (Mark et al., 2012). These equivocal findings demonstrate the need for additional analyses as well as assessment of the pathways through which increased RN-to-patient ratios might directly lead to improved patient safety.

New Contribution

This article reports findings from a comprehensive analysis of nurse staffing and quality of care in California from 2000 to 2006, using a statewide database that includes all nonfederal general acute care hospitals. This study builds on prior research in four ways. First, we examine more patient outcomes than in previous studies. The two studies from CALNOC had data on only patient falls and pressure ulcers (Bolton et al., 2007; Donaldson et al., 2005); Cook et al. (2012) analyzed pressure ulcer rates and mortality following a postoperative complication. Mark et al. (2012) considered mortality following a postoperative complication, postoperative respiratory failure, postoperative sepsis, and infections related to medical care. We consider six outcomes that are considered nurse-sensitive by organizations, such as the Nurse Quality Forum and the American Nurses Association (National Quality Forum, 2004). These outcomes are (a) failure to rescue, (b) decubitus ulcers, (c) selected infections due to medical care, (d) postoperative respiratory failure, (e) postoperative deep-vein thrombosis or pulmonary embolism (DVT/PE), and (f) sepsis.

Second, we explore one pathway through which nurse staffing may affect patient outcomes—by allowing nurses more time to perform surveillance and intervene quickly when complications arise. We do this by estimating whether changes in nurse staffing are associated with the average length of hospital stay for patients experiencing complications. Improved surveillance for complications has been posited as leading to earlier intervention, earlier discharges, and fewer readmissions, because patients’ complications are identified earlier (Dresser, 2012). The relationship between staffing levels and length of stay (LOS) may help inform our understanding of how staffing affects patient safety, even though we cannot measure whether the ratios directly increased the surveillance time of nurses.

Third, our analysis uses a very flexible approach that allows each type of nursing personnel (RNs, licensed vocational nurses, unlicensed nursing assistants, and temporary “registry” RNs) to have independent and interactive effects on patient outcomes. We adjust nurse staffing per patient day using nursing intensity weights (NIWs), which are designed to account for the intensity of nursing care needed by patients, based on their diagnoses. Finally, we include every year of data available from 1999 through 2006, in our analysis, to capture both the initial and later effects of the regulations on nursing and patient safety in California.

Conceptual Framework

Minimum nurse-to-patient ratio regulations were supported with the expectation that they would lead to improved quality of patient care (American Nurses Association, 2011). This prospect can be derived from Donabedian’s structure–process–outcomes model (Donabedian, 1980). Nurse staffing is one aspect of the structure of care delivery that affects the process of patient care, and the process of care directly affects patient outcomes. Other key structural components may include hospital size, service mix, profit status, system affiliation, payer mix, and whether the hospital is engaged in teaching health professionals.

One key component of patient care as provided by nurses is surveillance, which involves identifying emerging or potential problems for patients, and intervening to correct or prevent adverse events (Dresser, 2012; Fagin, 2001). Prior research has highlighted the importance of surveillance in identifying emergent complications and preventing their progression or subsequent mortality (Dresser, 2012; Mitchell & Shortell, 1997). The surveillance process demands that nurses have sufficient time and opportunity to directly observe patients, as well as coordinate and deliver interventions. If nurse staffing is low, relative to patient needs, the process of surveillance may be compromised. Moreover, high nursing workloads could impinge on nurses’ ability to provide all needed care, leading to deleterious effects on quality (Kalisch, Landstrom, & Williams, 2009; Kalisch & Lee, 2010; Kalisch, Tschannen, & Lee, 2011, 2012). Nurses who face a heavy patient burden might be more prone to errors. In addition, increased workload may inhibit a nurse’s ability to engage in discharge planning, as well as patient and family teaching, which are important aspects of ensuring good outcomes (Weiss, Yakusheva, & Bobay, 2011).

Thus, increasing nurse staffing might enhance nurse surveillance, reduce missed care and errors, and improve patient and family teaching, leading to improvements in quality of care. In addition, vigilant surveillance may lead to more timely intervention for patients who develop complications, and thus greater staffing should be linked to shorter hospital stays for patients experiencing adverse events.

Method

Data Sources and Variables

We examined staffing and outcomes data in an unbalanced panel of 278 California general acute care hospitals that reported data to the California Office of Statewide Health Planning and Development (OSHPD) between 1999 and 2006. We excluded hospitals that had a mean adjusted daily census in their medical–surgical cost center of fewer than 12 patients, and we excluded reporting periods shorter than 274 days. Patient-level data from OSHPD’s Patient Discharge Data were used to measure counts of adverse patient events and average LOS for patients experiencing adverse events. Data on hospital characteristics and staffing levels were obtained from OSHPD’s Hospital Annual Financial Disclosure Reports and the American Hospital Association’s Annual Survey of Hospitals. NIWs were used to account for differences in patients’ needs for nursing care.

Patient outcomes.

Using the Patient Safety Indicators software (PSI, version 3.2a) developed by the Agency for Healthcare Research and Quality (AHRQ), we calculated hospital-level patient outcomes (Elixhauser, Pancholi, & Clancy, 2005). We focused on six indicators that may represent preventable patient safety events and were identified as sensitive to nursing care in previous research and by organizations such as the National Quality Forum and the American Nurses Association: failure to rescue (PSI 4), decubitus ulcer (PSI 3), selected infections due to medical care (PSI 7), postoperative respiratory failure (PSI 11), postoperative DVT/PE (PSI 12), and sepsis (PSI 13; Savitz, Jones, & Bernard, 2005).

To compute hospital-level PSIs, we used the OSHPD Patient Discharge Data. The OSHPD data allow reporting of up to 30 diagnoses (including e-codes) and 21 procedures. We used the OSHPD-defined present-on-admission indicators in our computations. From our base data sets, we removed hospital transfers, outpatient discharges, and patients whose LOS was unknown, zero, or greater than 365 days. The total observations available to produce the PSIs were 26,684,752. We removed from analysis those patients at hospitals that did not report fiscal data to OSHPD between 2000 and 2006 and patients dropped from analysis by the AHRQ software due to data inconsistencies. We matched patient discharges to the fiscal years for which hospitals report staffing data (described below) to ensure that the analysis of outcomes corresponded to the intervals at which we observe changes in staffing. Our final analysis data set included 1,645 hospital-fiscal years.

From the base data set of patient observations, we also calculated mean LOS for patients experiencing each PSI, excluding all patients who died because death truncates LOS (we thus do not compute mean LOS for failure to rescue). If richer nurse staffing levels contribute to more vigilant surveillance, then emergent complications may be identified earlier and treatment can ensue more rapidly. This should lead to shorter LOS for those who do not die.

Nurse staffing.

The OSHPD Hospital Annual Financial Disclosure Reports provide total annual productive hours for 10 categories of hospital employees, in each type of patient care unit or cost center. We measured nurse staffing as a ratio of the number of productive hours (nonvacation, non–sick leave hours) per patient day (HPPD) in medical–surgical units. We focus on medical–surgical units because these account for the majority of inpatient days, and most patients spend at least some portion of their hospital stay in these units. Prior to the implementation of the staffing regulations, analyses indicated that the regulations would have the greatest impact in medical–surgical departments (Kravitz et al., 2002; Spetz, 2001). Minimum staffing ratio regulations existed for intensive and critical care units since the late 1970s and were not changed by the 1999 law. Thus, we treat medical–surgical staffing as a proxy for hospital-wide staffing.

The HPPD metric should be highly correlated with nurse-to-patient ratios but cannot be directly matched to nurse-to-patient ratios. The nurse-to-patient ratio, as specified in California’s regulations, is a real-time measure of the number of patients assigned to each licensed nurse who has a direct patient care assignment. A greater annual number of productive hours per patient day, as measured using the OSHPD data, will suggest that a hospital is more likely to meet minimum staffing regulations but does not tell us whether a hospital needs to increase nurse staffing to comply with the regulations for every shift on every day.

While a risk-adjustment strategy based on severity of illness is built into the AHRQ PSI software, it does not account for the intensity of nursing care needed by patients. We thus made an adjustment for the nursing care provided to patients by incorporating NIWs into our computations (Lichtig, Knauf, & Milholland, 1999; Needleman, Buerhaus, Mattke, Stewart, & Zelevinsky, 2001). NIWs were developed for the New York State Department of Health by the New York State Nurses Association in 1985 to adjust diagnosis-based payments for the amount of nursing care required (Knauf, Ballard, Mossman, & Lichtig, 2006). They were updated most recently in 2007. The NIW index equals 1.0 for a hospital with average levels of nursing care; values greater than 1.0 indicate higher intensity, and values lower than 1.0 indicate lower intensity than average. For each diagnosis, there are separate NIWs for intensive care and acute care patient days. The California OSHPD data do not indicate the numbers of days spent by each patient in acute versus intensive care. Thus, following the method used by Mark and Harless (2011), we used HCUP data from Maryland (which designates acute and intensive care days) to calculate overall mean NIWs for each diagnosis-related group. These NIWs were linked to the OSHPD patient discharge data, and we constructed an index to adjust HPPD.

In our regression analysis, we included separate NIW-adjusted HPPD variables for each of the RNs, licensed vocational nurses (LVNs), and unlicensed aides and orderlies (Aides). In a supplementary analysis, we further separated nonregistry RNs (regular employees) and registry RNs (temporary staff from a placement agency) and obtained qualitatively similar results. In the California regulations, LVNs may comprise up to half of licensed nurses. We also included a full set of interacted HPPD variables; for each staff type, we included the squared value of its HPPD and the product of its HPPD with each other staffing type. These interactions account for potential complementary and substitutive relationships between nurse types.

Hospital characteristics.

Prior research found that hospital structural characteristics have significant effects on staffing and/or quality of care. Hospital size and payer mix have been associated with patient outcomes in many studies (Kovner & Gergen, 1998; Lichtig et al., 1999; Mark et al., 2004; Needleman et al., 2002; Volpp & Buckley, 2004). We measure size as the natural logarithm of the number of available medical–surgical beds and payer mix as the percentage of inpatient days for patients insured by Medicare and the percentage of inpatient days for patients insured by Medicaid.

To control for time-invariant characteristics of hospitals, we include hospital-level fixed effects in all multivariate analyses. Hence, we do not include as explanatory variables hospital characteristics that do not change over time, such as urban/rural location or teaching status, or variables with little variation over time, such as profit status and system affiliation. We also include fixed effects for time.

Data Analysis

We sorted hospitals into quartiles based on their preregulation licensed nurse staffing levels (RNs and LVNs) in medical–surgical units. We did not use NIWs to define quartiles because California’s staffing regulations do not account for nursing intensity. We defined the preregulation period as prior to 2002 because California released draft regulations in January 2002, and previous research has found that there was little or no change in staffing before that year (Bolton et al., 2007; Donaldson et al., 2005; Spetz et al., 2009). The preregulation period is compared with a transitional period that precedes the regulations taking effect (2002–2003), an initial regulatory period when the medical–surgical ratios were 1 nurse per 6 patients, and a final regulatory period when the medical–surgical ratios were enriched to 1 nurse per 5 patients. We tested whether there were differences in staffing changes across quartiles of preregulation staffing levels.

Next, we estimated multivariate difference-in-difference regression models to examine the effects of changes in nurse staffing on changes in PSIs, using a fixed-effects Poisson model. The dependent variables are the counts of observed adverse outcomes. To control for differential risk, we use the count of expected adverse outcomes for each PSI as produced by AHRQ software using a risk-adjustment model. The differences in outcomes are compared both across quartiles and for each of the transitional, initial, and final regulation periods, with the preregulation period as the omitted period. These equations include fixed effects for hospitals, which control for hospital-specific characteristics that do not change over time (including the quartile to which the hospital is assigned). The equations also include fixed effects for each time period, which are measured as variables that represent the proportion of days in a reporting period that falls into the time period (preregulation, transitional, initial, and final). Finally, we add a full set of quartile–time period interactions. The equation is

$$E\left[y_{i,t}\mid {\mathrm{\alpha }}_i,{\mathrm{\tau }}_t,q_i^k,x_{i,t}\right]={\mathrm{\alpha }}_i\exp \left[{\mathrm{\tau }}_t{\mathrm{\delta }}_0+\left({\mathrm{\tau }}_t\times q_i^k\right){\mathrm{\delta }}_1+x_{i,t}^{\mathrm{\text{'}}}{\mathrm{\delta }}_2\right],$$

where $y_{i,t}$ is the number of adverse incidents at hospital $i$ in reporting period $t$, ${\mathrm{\alpha }}_i$ is the hospital fixed effect, ${\mathrm{\tau }}_t$ represents our dummy-like variables for regulatory periods, $q_i^k$ represents the set of dummy variables for preregulation staffing quartile $(k=\mathrm{1,2},3)$, and $x_{i,t}$ represents other explanatory variables; the vector ${\mathrm{\delta }}_1$ represents the nine difference-in-difference parameters (three quartiles and three time periods). Quartile 4 is treated as the baseline group from which the differences are measured. Other explanatory variables were percent Medicare inpatient days, percent Medicaid inpatient days, and the natural log of the number of beds in the medical–surgical cost center.

Third, we estimated multivariate fixed-effects Poisson models in which each PSI was a dependent variable to measure the marginal effect of additional nurse staffing on patient safety. In these equations, we explicitly controlled for NIW-adjusted HPPD in the medical–surgical cost center for each of RNs, LVNs, and Aides, as well as their squares and interactions. We included as control variables percent Medicare inpatient days, percent Medicaid inpatient days, the natural log of the number of beds in the medical–surgical cost center, and fixed effects for time. Using the coefficients from these models, we computed the percent change in adverse incidents associated with a one NIW-adjusted HPPD increase in RN staffing at the 25th percentile, 50th percentile, and 75th percentile of NIW-adjusted RN staffing in medical–surgical cost centers.

Finally, to explore the potential role of increased nurse staffing on surveillance, we estimated linear regression equations to examine the mean length of patient stay. Explanatory variables were NIW-adjusted HPPD for each type of nurse (RN, registry RN, LVN, and Aide), the squares and interactions of these, the natural log of the number of beds in the medical–surgical cost center, percent Medicaid inpatient days, and percent Medicare inpatient days. We computed the marginal effects on LOS of 1-hour increases in nonregistry RN HPPD at the 25th, 50th, and 75th percentiles of staffing.

For all multivariate models, we computed standard errors that are robust to hetero-skedasticity. All models were estimated using Stata 11 (STATA Corp, College Station, TX). $P<.05$ was considered statistically significant.

Results

Nurse Staffing Changes

California’s minimum nurse staffing regulations led to staffing changes for most California hospitals. However, some hospitals had staffing ratios above the mandated level prior to 2002—at least for some or most shifts—and may not have needed to increase staffing. Table 1 presents mean nurse HPPD and mean changes in HPPD for the preregulation, transition, initial, and final regulation periods for hospitals from the lowest (Quartile 1) to highest (Quartile 4) level of preregulation licensed nurse HPPD. Furthermore, Table 1 provides difference-in-difference calculations of changes in staffing relative to Quartile 4. The top panel shows that the total licensed nurse staffing (including registry RNs) for hospitals in Quartiles 1, 2, and 3 increased significantly after the preregulation period $(p<.001)$, and the difference from the preregulation period increased each year. By the final period, mean HPPD increased by 1.91 HPPD for Quartile 1, 1.93 HPPD for Quartile 2, and 1.71 HPPD for Quartile 3 $(p<.001)$. Mean HPPD did not significantly increase for hospitals in Quartile 4 during the transition period but rose thereafter to be 0.76 HPPD greater in the final period $(p<.001)$. The growth in HPPD among hospitals in Quartiles 1, 2, and 3 were significantly larger than the HPPD growth in Quartile 4 $(p<.001)$.

Table 1.

Mean Nurse HPPD and Changes by Regulatory Period and Preregulation Licensed Nurse Staffing Quartile

Staffing measure	Period	Quartile 1	Quartile 2	Quartile 3	Quartile 4
Licensed nurse HPPDa	Preregulation	4.22	4.92	5.67	6.85
	Transition	4.73	5.48	6.04	6.80
	Postinitialb	5.75	6.30	6.97	7.28
	Postfinal	6.10	6.84	7.29	7.66
Change in HPPD from preregulation period	Transition	0.57***	0.57***	0.46***	−0.10
	Initial	1.57***	1.39***	1.38***	0.39*
	Final	1.91***	1.93***	1.71***	0.76***
Difference-in-difference relative to Quartile 4	Transition	0.66***	0.66***	0.56***
	Initial	1.18***	1.00***	1.00***
	Final	1.15***	1.17***	0.95***
Nonregistry RN HPPD	Preregulation	3.26	3.92	4.25	5.16
	Transition	3.30	4.17	4.26	5.02
	Initial	4.15	4.77	4.91	5.48
	Final	4.43	5.30	5.39	5.95
Change in HPPD from preregulation period	Transition	0.05	0.22	0.05	−0.24
	Initial	0.89***	0.83***	0.72***	0.23
	Final	1.17***	1.34***	1.19***	0.72**
Difference-in-difference relative to Quartile 4	Transition	0.28	0.46**	0.29
	Initial	0.66**	0.60**	0.49*
	Final	0.45	0.62*	0.47
Registry RN HPPD	Preregulation	0.40	0.37	0.58	0.51
	Transition	0.72	0.69	0.95	0.83
	Initial	0.73	0.81	1.15	0.90
	Final	0.85	0.81	1.12	0.88
Change in HPPD from preregulation period	Transition	0.37***	0.38***	0.40***	0.37***
	Initial	0.38***	0.50***	0.61***	0.44***
	Final	0.48***	0.49***	0.56***	0.41***
Difference-in-difference relative to Quartile 4	Transition	0.00	0.01	0.03
	Initial	−0.06	0.06	0.17
	Final	0.07	0.08	0.15
LVN HPPD	Preregulation	0.56	0.63	0.84	1.17
	Transition	0.71	0.61	0.84	0.94
	Initial	0.87	0.72	0.90	0.90
	Final	0.82	0.73	0.78	0.83
Change in HPPD from preregulation period	Transition	0.15*	−0.03	0.01	−0.23***
	Initial	0.31***	0.06	0.06	−0.28**
	Final	0.26**	0.10	−0.04	−0.37***
Difference-in-difference relative to Quartile 4	Transition	0.37***	0.19**	0.24**
	Initial	0.58***	0.34**	0.34**
	Final	0.63***	0.47***	0.33**
Overall patient-to-nurse ratio assuming 26.4 HPPD	Preregulation	6.35:1	5.44:1	4.70:1	3.98:1
	Transition	5.75	5.02	4.51	4.02
	Initial	4.74	4.34	3.89	3.75
	Final	4.62	3.99	3.77	3.55
Number of hospitals		57	57	57	57

Note. HPPD = hours per patient day; RN = registered nurse; LVN = licensed vocational nurse.

^a.Licensed nurses include RNs, registry RNs, and LVNs. This category excludes aides and orderlies.

^b.The two postregulation periods are the initial regulation at 1:6 Med-Surg (January 2004–March 2005) and final regulation at 1:5 Med-Surg (March 2005–December 2006).

^*p < .05.

^**p < .01.

^***p < .001.

Table 1 also indicates the differences in the mix of staffing types as staffing levels increased. We find that the most staffing growth occurred for nonregistry RNs and some decline occurred in LVN staffing between 1999 and 2006. Mean HPPD of nonregistry RNs increased significantly from the preregulation period in the initial and final regulation periods for hospitals in Quartiles 1, 2, and 3 $(p<.001)$. It also increased for hospitals in Quartile 4 in the final regulatory period. The change in nonregistry RN HPPD was significantly greater for hospitals in Quartiles 1, 2, and 3 as compared with Quartile 4 in the initial period and, for Quartile 2, in the transition and final period as well. HPPD for registry RNs increased significantly $(p<.001)$ in each of the periods and for each of the quartiles, though this change was of a similar size across all four quartiles. LVN HPPD were greater after the preregulation period for hospitals in Quartile 1 ($p<.05$), but they declined for hospitals in Quartile 4 ($p<.01$).

While mean annual HPPD cannot directly measure whether a hospital was in compliance with California’s staffing regulation at all times, it can be used to estimate annual average patient-to-nurse ratios. The bottom panel of Table 1 shows this calculation assuming that an average patient day represents 26.4 hours of care (Unruh, Fottler, & Talbott, 2003). Since many patients are admitted in the morning and discharged in the afternoon, the standard 24-hour patient day is too low. Imputed patient-to-nurse ratios decreased over time, as expected. The ratio was 6.35:1 for hospitals in Quartile 1 and 3.98:1 for hospitals in Quartile 4 in the preregulation period. The regulatory mandated staffing ratio in the final period was 5:1. By the final period, the difference in the imputed ratio was smaller across quartiles, ranging from 4.62:1 (Quartile 1) to 3.55:1 (Quartile 4).

Multivariate Analyses of Nurse Staffing and Patient Outcomes

A difference-in-difference approach was used to examine changes in patient outcomes. Poisson regressions were estimated to compare changes in patient outcomes across quartiles of preregulation staffing. Table 2 presents summary statistics for the variables used in these regressions, and Table 3 provides estimates of the percentage change, relative to the preregulation period, in the incidence of each of the PSIs for Quartile 4 hospitals (the parameter estimates for the variables Transition, Initial, and Final), as well as estimates of the difference in percentage changes by time period in Quartiles 1, 2, and 3. For hospitals in Quartile 4, there was significantly greater incidence after the preregulation period for four of the PSIs: pressure ulcers (PSI 3, $p<.05$), infections related to medical care (PSI 7, $p<.01$, except in the final period), postoperative PE/DVT (PSI 12, $p<.001$), and postoperative sepsis (PSI 13, $p<.01$). Incidence of mortality following a complication (PSI 4) was 16.7% lower in the final period than in the preregulation period ($p<.05$).

Table 2.

Variable Means (Standard Deviations), by Preregulation Licensed Nurse Staffing Quartile

Variable	Quartile 1	Quartile 2	Quartile 3	Quartile 4
Pressure ulcers: Observed incidents	5.2 (8.0)	7.2 (8.4)	7.4 (13.2)	7.2 (8.4)
Pressure ulcers: Risk-adjusted percent	0.26 (0.33)	0.27 (0.30)	0.24 (0.31)	0.25 (0.23)
Failure-to-rescue: Observed incidents	1.7 (2.4)	2.6 (2.6)	3.3 (4.7)	3.5 (4.9)
Failure-to-rescue: Risk-adjusted percent	27.87 (27.79)	27.27 (21.58)	32.80 (30.44)	24.87 (25.01)
Infections: Observed incidents	4.7 (5.9)	7.6 (7.9)	7.7 (10.5)	9.9 (13.1)
Infections: Risk-adjusted percent	0.10 (0.10)	0.13 (0.10)	0.12 (0.10)	0.14 (0.13)
Respiratory failure: Observed incidents	6.9 (8.0)	8.6 (8.1)	11.5 (19.1)	12.4 (21.3)
Respiratory failure: Risk-adjusted percent	1.28 (1.21)	1.01 (0.75)	1.08 (1.51)	0.89 (0.64)
DVT/PE: Observed incidents	4.6 (5.5)	10.2 (13.6)	9.5 (17.9)	15.2 (23.8)
DVT/PE: Risk-adjusted percent	0.26 (0.24)	0.34 (0.27)	0.29 (0.25)	0.38 (0.34)
Sepsis: Observed incidents	1.1 (1.9)	1.6 (2.1)	2.0 (3.3)	2.1 (2.9)
Sepsis: Risk-adjusted percent	0.73 (1.22)	0.69 (0.99)	0.71 (1.13)	0.70 (1.16)
Nurse intensity weight (NIW)	3.11 (0.33)	3.20 (0.34)	3.20 (0.38)	3.25 (0.43)
NlW-adjusted nonregistry RN HPPD	3.70 (1.16)	4.32 (1.15)	4.50 (1.18)	5.14 (1.39)
NlW-adjusted registry RN HPPD	0.64 (0.75)	0.61 (0.57)	0.86 (0.79)	0.72 (0.68)
NlW-adjusted LVN HPPD	0.72 (0.63)	0.65 (0.51)	0.85 (0.75)	1.00 (0.86)
NIW-adjusted aide HPPD	2.61 (0.97)	2.58 (1.00)	2.22 (1.06)	2.09 (1.06)
Percent Medicare IPD	30.9 (12.0)	29.0 (10.3)	29.0 (11.9)	33.0 (11.6)
Percent Medicaid IPD	15.6 (12.0)	14.8 (9.9)	14.8 (11.9)	13.7 (9.5)
Number of medical-surgical unit beds	83.5 (53.6)	94.8 (52.3)	106.1 (91.1)	112.5 (83.8)
Number of observations	409	410	407	419
Number of hospitals	57	57	57	57

Note. DVT = deep-vein thrombosis; PE = pulmonary embolism; RN = registered nurse; HPPD = hours per patient day; LVN = licensed vocational nurse; IPD = inpatient day.

Table 3.

Difference-in-Difference Fixed-Effects Poisson Model Estimates of Percentage Change in Incidence of PSIs in Hospitals in the Three Lowest Preregulation Staffing Quartiles Compared to Hospitals in the Highest Preregulation Staffing Quartile^a

	Pressure ulcer	Failure to rescue	Infections	Respiratory railure	DVT/PE	Sepsis
Transition	1 1.4 (10.8)	0.8 (9.5)	27.9** (8.9)	0.2 (1 1.9)	34.4*** (7.3)	1 1.3 (1 l.l)
Initial	38.8** (13.3)	−2.7 (7.8)	20.5** (7.5)	−7.3 (1 1.3)	5 1.5*** (7.5)	60.7*** (15.8)
Final	28.6* (12.7)	−16.7* (6.9)	6.3 (12.1)	−8.6 (13.6)	54.0*** (8.6)	40.9** (14.0)
Quartile 1 × Transition	8.3 (13.5)	−8.6 (12.9)	−4.7(10.8)	0.2 (13.4)	−8.4 (9.5)	−24.3 (13.7)
Quartile 1 × Initial	−21.9 (12.1)	−25.0 (13.4)	3.1 (15.6)	−2.9 (15.1)	−9.5 (10.5)	−36.2* (17.2)
Quartile 1 × Final	−23.2 (13.6)	−24.3* (12.1)	14.7(18.0)	9.1 (17.7)	−17.5* (8.1)	10.8 (21.8)
Quartile 2 × Transition	7.2 (14.3)	−18.5* (9.0)	1.6 (9.6)	−0.5 (12.1)	−10.2 (9.6)	4.5 (15.3)
Quartile 2 × Initial	−6.3 (14.3)	14.9 (14.3)	7.9 (13.1)	9.5 (14.5)	−4.4 (13.6)	−25.4 (13.3)
Quartile 2 × Final	0.7(16.6)	6.9 (13.7)	37.1 (21.6)	1 1.3 (17.6)	0.3 (13.5)	21.2 (20.1)
Quartile 3 × Transition	−8.6 (20.3)	−18.5* (9.2)	−0.4 (8.9)	−2.2 (12.4)	−1.3 (9.3)	19.6 (18.5)
Quartile 3 × Initial	−24.7(17.1)	−28.1 **(8.8)	−3.5 (8.9)	0.8 (13.3)	−8.5 (10.1)	−17.7(13.6)
Quartile 3 × Final	−1 3.4 (20.7)	−7.9 (1 1.2)	1.5 (14.6)	−6.7(14.2)	−8.1 (7.4)	10.0 (14.6)
Number of observations	1,607	1,207	1,614	1,628	1,600	1,463
Number of hospitals	223	199	224	226	222	203

Note. PSI = patient safety indicator; DVT = deep-vein thrombosis; PE = pulmonary embolism; OSHPD = California Office of Statewide Health Planning and Development.

^a.Heteroskedasticity-robust standard errors are reported in parentheses. We incorporated the differing hospital reporting period dates to match reporting periods and the regulatory periods. The DD Poisson regression contains dummy-like variables, defined as the proportion of days of the OSHPD reporting period falling in a regulatory period. Other control variables include the percent Medicare inpatient days, percent Medicaid inpatient days, and the natural log of the number of beds in the medical–surgical cost center.

^*p < .05.

^**p < .01.

^***p < .001.

There were few differences across quartiles of preregulation staffing in the incidence of adverse events. For hospitals in Quartile 1, which increased nurse staffing significantly after the preregulation period, there was a significantly greater improvement in the incidence of mortality following a complication by the final postregulation period $(p<.05)$. There also was less of an increase for PE/DVT in the final period $(p<.05)$ and for sepsis in the initial period $(p<.05)$ as compared to Quartile 4, where the estimated percentage changes are all positive. For hospitals in Quartile 2, the only difference (as compared with Quartile 4) was in greater improvement in mortality following a complication during the transition period ($p<.05$). Hospitals in Quartile 3 also had a greater reduction in the incidence of mortality following complications during the transition $(p<.05)$ and initial $(p<.01)$ periods.

Table 4 presents the marginal effects of 1-hour increases in NIW-adjusted RN HPPD, measured as the percent change in the number of adverse incidents. None of the estimated changes are statistically significant at conventional levels $(p<.05)$. However, the patterns of change are consistent with those presented in Table 3. An additional hour of NIW-adjusted RN HPPD was associated with approximately 2.4% fewer incidents of mortality following a postoperative complication at the 25th percentile of HPPD, 2.0% fewer at the median, and 1.7% fewer at the 75th percentile. For all other PSIs, an increase in HPPD is estimated to increase the incidence of adverse events, although these estimates are not statistically significant. The results were similar when we separated the nonregistry and registry RNs.

Table 4.

Marginal Effect of a 1-Hour Increase in NIW-Adjusted RN HPPD, Percentage Change in Incidence of PSIs^a

RN HPPD Percentile Value	Pressure ulcer	Failure to rescue	Infections	Respiratory failure	DVT/PE	Sepsis
25th	1.0 (4.5)	−2.4 (3.9)	4.7 (3.4)	1.4 (2.7)	3.2 (3.3)	2.8 (5.9)
50th	0.7 (3.4)	−2.0 (3.0)	3.6 (2.5)	1.1 (2.1)	3.1 (2.6)	3.2 (4.5)
75th	0.4 (2.5)	−1.7 (2.5)	2.4 (2.2)	0.7 (1.8)	2.9 (2.1)	3.6 (3.5)

Note. NIW = nursing intensity weight; RN = registered nurses; HPPD = hours per patient day; PSI = patient safety indicator; DVT = deep-vein thrombosis; PE = pulmonary embolism.

^a.Estimates are obtained from fixed-effects Poisson regression models that include variables measuring NIW-adjusted HPPD in the medical–surgical cost center for three staffing measures (RNs, LVNs, and Aides), their squares, and interactions. Also included as controls are the variables percent Medicare inpatient days, percent Medicaid inpatient days, and the natural log of the number of beds in the medical–surgical cost center, and fixed effects for time. The table provides the estimates for the percent change in adverse incidents associated with a one NIW-adjusted HPPD increase in RN staffing at the 25th percentile (4.12), 50th percentile (5.00), and 75th percentile (5.98) of NIW-adjusted RN staffing in medical–surgical cost centers.

In Table 5, we present the marginal effects of a 1-hour increase in NIW-adjusted RN HPPD as the percent change in mean LOS for patients experiencing the adverse event. The mean LOS for patients experiencing selected infections due to medical care decreased significantly, with a larger decline found among hospitals with RN HPPD at the 25th percentile (−10.0%, $p<.001$) and at the median (−7.0%, $p<.001$). The LOS also decreased with the addition of nurses for pressure ulcers and postoperative respiratory failure, but these changes are not statistically significant. The LOS rises for postoperative sepsis, and the relationship is mixed for PE/DVT (all not statistically significant). Complete regression results are available from the corresponding author on request.

Table 5.

Marginal Effect of a 1-Hour Increase in NIW-Adjusted RN HPPD, Percentage Change in Average LOS for Patients Experiencing the PSI^a

RN HPPD percentile value	Pressure ulcer	Infections	Respiratory failure	DVT/PE	Sepsis
25th	−3.4 (3.1)	−10.0*** (2.7)	−2.9 (2.7)	2.6 (3.0)	3.6 (6.1)
50th	−2.8 (2.3)	−7.0*** (2.1)	−1.8 (2.1)	0.9 (2.4)	3.8 (4.9)
75th	−2.1 (1.7)	−3.5 (1.8)	−0.5 (1.8)	−0.9 (2.0)	4.1 (4.1)

Note. NIW = nursing intensity weight; RN = registered nurses; HPPD = hours per patient day; LOS = length of stay; PSI = patient safety indicator; DVT = deep-vein thrombosis; PE = pulmonary embolism.

^a.Estimates are obtained from fixed-effects linear regression models that include variables measuring NIW-adjusted HPPD in the medical–surgical cost center for three staffing measures (RNs, LVNs, and Aides), their squares, and interactions. Also included as controls are the variables percent Medicare inpatient days, percent Medicaid inpatient days, and the natural log of the number of beds in the medical–surgical cost center, and fixed effects for time. The table provides the estimates for the percent change in average LOS for patients experiencing the PSI associated with a one NIW-adjusted HPPD increase in RN staffing at the 25th percentile (4.12), 50th percentile (5.00), and 75th percentile (5.98) of NIW-adjusted RN staffing in medical–surgical cost centers.

^***p < .001.

Discussion

This analysis found strong evidence that nurse staffing in California acute care hospitals increased after mandated minimum staffing ratios. Total licensed nurse HPPD increased across all quartiles of preregulation staffing, with the largest increases occurring among hospitals in the lower quartiles of preregulation staffing. There were differences in the timing and composition of these staffing increases. Hospitals in the top quartile did not increase their licensed nurse HPPD during the transition period, and later increases focused on greater RN HPPD while cutting LVN HPPD. Hospitals in the bottom three quartiles also displayed smaller changes in staffing during the transition period than in the initial and final periods. Hospitals in the lowest quartile of preregulation staffing did not have greater growth in RN HPPD in the final period than did hospitals in the top quartile; rather, the overall increase in licensed nurse HPPD came from a larger increase in LVN HPPD among the lowest quartile hospitals.

We found overall trends of increasing incidence of four PSIs and a decline in mortality following a postoperative complication; these patterns are consistent with national trends (Downey et al., 2012). Although there were significantly greater increases in licensed nurse HPPD for the bottom three preregulation staffing quartiles relative to the top quartile, changes in patient outcomes were mixed. As has been found in other recent research, we measured a significantly greater decrease in mortality following complication among hospitals in the lower three quartiles of preregulation staffing (Mark et al., 2012). These hospitals had larger growth in overall licensed nurse HPPD following passage of the staffing, even though there were differences in the mix of RNs and LVNs in the staffing changes.

There was little evidence of significantly different changes in the other PSIs across preregulation staffing quartiles. Declines in pressure ulcer incidence (PSI 3) among the lower three quartiles were not statistically significant but suggest that there were lower postregulation rates as compared with the top quartile. This result is consistent with previous studies that found that California’s nurse staffing regulations did not lead to improvements in rates of decubitus ulcers (Bolton et al., 2007; Cook et al., 2012; Donaldson et al., 2005). There also was a pattern of lower incidence of PE/DVT in the lower three quartiles as compared with the top quartile, although this was significant only for Quartile 1 hospitals in the final period. There were mixed changes for postoperative sepsis and respiratory failure. Lower-quartile hospitals had a pattern of greater incidence of infections due to medical care, but these differences were not statistically significant. In general, the largest decreases in PSIs relative to Quartile 4 occur in Quartile 1, which also is the quartile in which we observed the largest increase in licensed nurse staffing.

The mixed difference-in-difference findings are consistent with results of models that directly predict incidence of adverse events as a function of nurse staffing. Higher ratios of NIW-adjusted RN HPPD are associated with lower rates of mortality following complication and higher rates of all other outcomes, but none of these differences are statistically significant. For infections related to medical care, the relationship between nurse staffing and average LOS is negative and statistically significant. The relationship also is negative for pressure ulcers and postoperative respiratory failure, although not statistically significant. This may indicate that when more hours of RN care are available per patient, nurses are better able to detect and address complications, which is consistent with the surveillance role of nursing care. Improved surveillance could be associated with both greater measured rates of adverse events, because they are identified more accurately, and shorter LOS, because intervention is more timely. The improvements in mortality rates following postoperative complications also are consistent with the possibility that increased nurse staffing is linked to improved outcomes through the surveillance role of nursing.

Limitations

While this analysis has the advantage of examining an externally driven change in nurse staffing, the analysis may be confounded by other trends. During the 2000s, many hospitals established quality improvement programs in response to increased public attention to medical errors. There have been few systematic analyses of the impact of these other quality-improvement programs and how their efforts may have complemented or detracted from the impact of nurse staffing ratios. Ideally, we would include variables in the analysis that measure the implementation of various important quality-improvement programs in each hospital; however, such data do not exist and collecting such data would be extremely difficult. To the extent that these quality improvement efforts constitute a statewide trend, the yearly time controls in the multivariate analysis control for this; however, if there is variation in the implementation of these efforts across hospitals, the yearly variables do not control for this confounding factor.

Our dependence on NIWs to adjust for patient acuity is another potential limitation. Because we do not have data on the number of days each patient spent in intensive care versus general care units, we had to develop an estimate of the NIW for each diagnosis-related group using data from Maryland. If patterns of intensive versus general care differ in California, these NIWs may not accurately represent the need for nursing care among California patients. Nonetheless, we think that using NIWs is an improvement over relying on the Medicare case mix index, which was designed to measure intensity of overall resource use (Mark & Harless, 2011).

The PSI software does not measure adverse patient events perfectly. Recent research has found that the rates of adverse events computed from the PSI software are up to 10 times lower than the rates measured using an alternative method, but that although the PSI method has low sensitivity, it has high specificity (Classen et al., 2011). Thus, while our measure of adverse events does not capture all incidents, it also is not likely to be biased.

Because we were interested in the potential surveillance role of nurses in preventing, detecting, and intervening in complications, we limited our calculation of average LOS to patients who experienced complications. It is possible that increased nursing HPPD reduced LOS for all patients, and analysis of overall LOS would be of interest. Such an analysis would need to address heterogeneity in diagnoses, number of complicating diagnoses, and other factors that may affect LOS but are not easily controlled. In addition, the LOS for patients who do not experience complications tends to be quite short, and there is likely to be little marginal impact on LOS of enriched staffing across a general population of patients. Among patients who experience complications, however, there is the potential for significantly extended LOS if complications are allowed to progress and worsen. Thus, among this subset of patients, we expect the surveillance role to be more important.

Finally, our analysis only considers 2 years after the ratios were implemented; changes in nurse staffing might not have an immediate impact on patient safety and continued increases in staffing due to the phasing-in of the regulations may have been important.

Implications for Policy and Practice

California’s minimum nurse-to-patient staffing regulations were intended to improve the quality of patient care, but to date there is only mixed evidence that they achieved this goal. This analysis and other recent studies indicate that the role of nursing care in improving patient outcomes is multidimensional; multiple structural and process factors are important (Aiken et al., 2011; Needleman et al., 2009; Needleman et al., 2011).

Regulatory approaches to improving quality of patient care can have unintended consequences. California’s minimum staffing regulation has been linked to a more rapid wage growth than in other states (Mark, Harless, & Spetz, 2009), to declines in operating margins (Reiter, Harless, Pink, & Mark, 2012), and to a reduction in the amount of uncompensated care provided by some hospitals (Reiter, Harless, Pink, Spetz, & Mark, 2011). These “costs” may be worthwhile if the regulations achieved any benefits. We find evidence that increased nurse staffing was associated with lower rates of mortality following complications and also limited evidence that higher staffing is linked to shorter lengths of stay for patients experiencing some complications. This suggests that some benefit may arise from nurses having more time for surveillance of patients; however, this conclusion is tentative and requires more research.

The net effect of nurse staffing legislation remains unknown. California’s fixed minimum staffing ratio regulations have not been consistently linked to improvements in the quality of care, although it is possible that benefits to patient care will be found in the long term. The comparatively flexible regulations established in other states have yet to be evaluated, and in general, the full set of factors that facilitate improved nursing care is poorly understood. Thus, policymakers should tread cautiously as they consider new nurse staffing regulations and carefully weigh the potential of quality and work environment improvement against the possible costs of regulation.