California Minimum Nurse Staffi ng Regulations and Research Design

of legislating higher nurse- patient ratios. One important reason is that permanent in- creases in nurse staffi ng levels that would be effected through minimum staffi ng legis- lation may have quite different effects than these transitory fl uctuations. Hospital stays average about fi ve days in length, and changing the amount of nursing care over all fi ve of these days is likely to impact care more than for just the fi rst one or two days. 3 More generally, the care received by any patient may react more to permanent rather than temporary staffi ng changes for a number of reasons. The effort levels of nurses may respond inversely to temporary surges in admissions, or hospitals may increase triage efforts and encourage less acutely ill patients to return for care at a later date when staffi ng ratios have normalized. Both mechanisms might lead to muted apparent effects of staffi ng levels on outcomes. Despite the interest in mandated increases in nurse staffi ng ratios as a means to improve the quality of healthcare, few studies have effectively addressed this question. Zhang and Grabowski 2004, Park and Stearns 2009, and Tong 2010 all study the effects of legislation aimed at increasing nurse staffi ng but the current study improves methodologically on each in several ways. 4 For example, Zhang and Grabowski re- gress changes in quality of care measures on changes in staffi ng levels among nursing homes surrounding passage of the national Nursing Home Reform Act NHRA in 1987, and fi nd no signifi cant effects. Unfortunately, for these relationships to refl ect the causal effect of staffi ng it must be the case that all of the time- series variation in staffi ng levels is exogenous, an assumption that is untested and hard to accept as plausible, especially given the other components of the NHRA for example, a man- dated reduction in the use of unnecessary restraints or drugs, etc.. Tong 2010 uses a research design similar to that employed in this paper, but focuses on mortality as the quality of care measure for nursing homes. Mortality is an awkward quality of care measure in this context since acutely ill patients are often transferred to a hospital so may not die in the nursing home. Moreover, in a recent national survey some 56 percent of nursing home residents had do- not- resuscitate advanced directives Center for Disease Control 2009, and so allowing a patient to die may not necessarily refl ect poor care in this setting.

III. California Minimum Nurse Staffi ng Regulations and Research Design

This paper assesses the causal impact of a law requiring a minimum threshold of nurse staff per patient on the quality of healthcare provided by nurs- ing homes in California. On July 22, 1999, then Governor Gray Davis responded to growing concerns about quality of care in nursing homes by signing Assembly Bill AB 1107, a law that required all skilled nursing facilities nursing homes to provide 3. Average length of stay in hospitals was 4.8 days in 2006 according to Health, United States, 2009, www .cdc .gov nchs data hus hus09 .pdf102 4. Konetzka, Stearns and Park 2008 study the impact of RN staffi ng on the prevalence of pressure sores and urinary tract infections in nursing homes, using the adoption of Medicare Prospective Payment System as an instrument for changes in RN staffi ng. The exclusion restriction that the adoption of Medicare PPS affects outcomes only through changes in staffi ng, however, is a strong assumption that is untested in the paper. a minimum of 3.2 HPRD of direct nursing care. Medi- Cal California’s Medicaid program reimbursement rates increased via a wage pass- through to partially com- pensate fi rms for increased labor costs. The law was silent on the skill- mix of this care, so fi rms could meet the requirement by increasing their total hours worked by the combination of three types of nurses: registered nurses RNs, the most skilled and highest paid, licensed vocational nurses LVNs, frequently referred to as licensed practical nurses, or LPNs, in other states, or nurse aides NAs, the least skilled, also called nurses’ aides. AB 1107 supplemented the Federal standards established by the 1987 NHRA studied by Zhang and Grabowski 2004 and replaced a previous state standard that required nursing homes to provide three nurse hours per resident day HPRD, but allowed double counting of RN and LVN hours. 5 While in 1999 nearly all facilities were in compliance with the previous standard, nearly three- quarters of all fi rms had staffi ng levels below the 3.2 HPRD threshold required by the new law as of January 1, 2000 Matsudaira forthcoming. 6 Facility compliance with the staffi ng law was monitored through regular visits by state auditors that included checks of staffi ng records and patient registers, and a range of penalties ranging from monetary fi nes up to a loss of license were established for noncompliance. 7 The structure of AB 1107 lends itself to a transparent research design to study its effects on quality of care outcomes, and to explore the relationship between nurse staffi ng and patient outcomes. The crux of the research design employed here is to compare changes in quality of care outcomes for fi rms with varying degrees of expo- sure to the minimum staffi ng law. Let the variable GAP i represent the staffi ng increase that would be necessary for a facility to just comply with the law, based on their aver- age staffi ng level in the years 1997 and 1998 HPRD 9798 i —the two years prior to adoption of the AB1107. Thus, for a fi rm employing 2.7 HPRD the average for fi rms not in compliance prior to passage, GAP i = 0.5 and for a fi rm employing 3.3 HPRD, GAP i = 0. 8 If facilities comply with the law’s mandates, then we should expect to see staffi ng increases that are proportional to GAP i —fi rms with staffi ng levels below 3.2 HPRD should increase their staffi ng, with greater increases for fi rms initially fur- ther below the threshold. On the other hand, fi rms with staffi ng levels already above the threshold face no legislatively induced pressure to hire more nurses. So although we may observe increases in nurse staffi ng levels among these higher staffed facilities due to common shocks to all nursing homes, these should be more muted. To the ex- tent that more nursing hours per resident causes higher quality of care, then we should observe similar patterns in the changes in quality of care measures across fi rms with different initial staffi ng levels. To be more precise, let y itc represent some nursing home outcome like the number of registered nurse hours employed or the fraction of residents with pressure sores in 5. Among other provisions, the NHRA required that 1 all homes have licensed nurses RNs or LVNs on duty 24 hours per day, 2 all homes have an RN on duty at least 8 hours every day, and 3 homes with 60 or more beds have a full- time RN employed as a director of nursing DON. 6. Facilities were notifi ed, however, that enforcement of the new standard would begin in April of 2000. A separate law, AB 394, addressed hospitals but was not implemented until 2004. 7. Unfortunately, there is little information about how vigorously the law was enforced though as shown later fi rms did comply, albeit imperfectly, with the law. 8. More succinctly where a facility’s average hours is measured before passage of the law with no double counting of LVN or RN hours over the years 1997 and 1998. facility i in county c and year t, and defi ne ∆y ic ≡ y ic 2004 – y ic 1999 , or the change in the outcome four years after the policy was enacted. 9 I estimate the impact of the mini- mum nursing staffi ng legislation through variations of the following equation 1 ∆y ic = β + θGAP i + γ c + πHPRD 9798 i + ν ic where HPRD 9798 i represents the facility’s initial staffi ng level and y c represents a vector of 31 “county” fi xed effects. 10 11 The parameter of interest is θ which measures the effect of the staffi ng legislation on y itc . I estimate three permutations of this equation, variably imposing the restrictions that y c and or π are equal to zero. Since the depen- dent variable is the change in the outcome, the model implicitly controls for time- invariant facility characteristics that might affect either staffi ng levels or quality of care, such as permanent components of technology or patient mix, or location- specifi c factors such as proximity to nurse training centers. The inclusion of county- level fi xed effects y c accounts for county- level changes in determinants of ∆y ic such as changes in county supply and demand conditions in both the labor and “output” mar- kets local population demographics and economic conditions, changes in Medicaid or Medicare reimbursement regimes, market concentration ratios, and so on. I also test the sensitivity of the results to inclusion of the initial staffi ng level HPRD 9798 i . This controls for other changes in the determinants of outcomes that may be correlated with initial staffi ng levels. One potential source of this type of confound- ing is the switch to a prospective pay system for Medicare, which some researchers have suggested led to reductions in RN staffi ng. The ability to make such reductions may be correlated with staffi ng levels, since low staff facilities may be constrained by federal standards for licensed nurses created by the NHRA. Another possibility, how- ever, is that inclusion of HPRD 9798 i may pick up the effects of competitive pressures to increase staffi ng for fi rms initially above the threshold. If the law causes initially low- staffed fi rms to increase their staffi ng levels, fi rms already in compliance may be induced to increase their staffi ng levels to maintain a competitive edge as it is an indi- cator of quality that is easy for potential customers to observe. If so, θ might underes- timate the full impact of the law. Because the model above identifi es the impact of the law by comparing changes in outcomes across fi rms with differing initial staffi ng levels, similar to a standard differ- ence in differences model, the key identifi cation assumption is that in the absence of the minimum staffi ng law there would be no difference in the changes in the outcome related to GAP i . I test this “common trends” assumption by estimating Equation 1 for key outcome variables in the preperiod 1996–99, and verifying that the estimates of θ are statistically indistinguishable from 0. While not necessarily a source of bias for estimating the causal effect of the law, a potential concern in making inferences about the impact of staffi ng on quality of care outcomes involves behavioral responses to the 9. I have estimated models with various difference lengths, but present these longer term results only for brevity of presentation. All results are qualitatively similar, with the exception that staffi ng impacts of the law are more muted over a shorter time horizon as can be seen in Figure 1. 10. Some small counties are grouped together with contiguous larger counties. A list of the county groupings with the number of facilities in each county and the fraction that were out of compliance before the minimum staffi ng legislation was passed appears in Appendix Table 1. 11. I omit “d” superscripts on the coeffi cients for simplicity, but all coeffi cients may differ by the difference length. law by both nursing homes and patients. For example, nursing homes might substitute away from other inputs into quality of care as they increase their nurse employment and patient sorting to nursing homes might change as well. To the extent that patient sorting by unobserved acuity changes in such a way that more ill patients become more likely to sort into nursing homes with low initial staffi ng levels and thus bigger increases in staffi ng after the law becomes effective, this might lead to no apparent effect of staffi ng on patient outcomes. In the results section below, I show that there appears to be no effect of the law on observable measures of either facility- level fac- tor substitution or patient level sorting. While the potential for unobserved responses exists, the scant prior literature on this suggests the magnitudes of such a response are likely to be very small. While it might seem natural to use this setup to study the causal effect of total staff- ing all nurse hours on patient outcomes through an instrumental variables design— for example instrumenting the change in total nurse staffi ng with GAP i —I eschew such an approach here. In this context it is very likely that the causal effect of a change in staffi ng will depend heavily on the mechanism that induces such a change, violating the so- called “stable unit treatment- value assumption” SUTVA discussed by Rubin 1986. Put simply, one can easily imagine that a different minimum staffi ng law that required increases in higher- skilled nurses RNs would have quite different effects on patient outcomes than the California legislation. While the California legislation is similar to legislation adopted or proposed in many other states and so the results here have broad applicability, it may be inappropriate to use the results to infer the causal effects of different types of staffi ng increases. Keeping these concerns in mind, the two- stage least squares estimates can be easily inferred from the “fi rst- stage” esti- mates of the effects on nurse staffi ng and the “reduced form” estimates of the effect of the law on patient outcomes presented below.

IV. Data and Descriptive Overview