Causal Inference Matters To Nurses Assignment Paper
Since the early 1990s researchers have steadily built a broad evidence base for the association between nurse staffing and patient outcomes. However, the majority of the studies in the literature employ designs that are unable to robustly examine causal pathways to meaningful improvement in patient outcomes. A focus on causal inference is essential to moving the field of nursing research forward, and as part of the essential skill-set for all nurses as consumers of research. In this article, we aim to describe the importance of causal inference in nursing research and discuss study designs that are more likely to produce causal findings. We first review the conceptual framework supporting this discussion and then use selected examples from the literature, typifying three key study designs – cross-sectional, longitudinal, and randomized control trials (RCTs). The discussion will illustrate strengths and limitation of existing evidence, focusing on the causal pathway between nurse staffing and outcomes. The article conclusion considers implications for future research. Causal Inference Matters To Nurses Assignment Paper
ORDER YOUR PAPER HERE
Keywords: nurse staffing; causal inference; patient outcomes; outcomes research; nursing outcomes research; cross-sectional designs; longitudinal designs; randomized control trials; nurse organization; health policy
Citation: Costa, D. K., Yakusheva, O. (May 31, 2016) “Why Causal Inference Matters to Nurses: The Case of Nurse Staffing and Patient Outcomes” OJIN: The Online Journal of Issues in Nursing Vol. 21, No. 2, Manuscript 2.
Article below
Many hospital administrators and policy makers are interested in improving patient outcomes through exemplary nursing practice Many hospital administrators and policy makers are interested in improving patient outcomes through exemplary nursing practice, since the nursing profession is the largest healthcare workforce and nurses frequently interact with patients. Responding to the need for evidence to guide policy and practice, researchers have developed a robust evidence base of associations between nursing care and outcomes in the last two decades. Much of this evidence examines the relationship between nurse staffing and patient outcomes.
On a conceptual level, a positive contribution of nurse staffing to patient outcomes is intuitive and easily understood. Better staffing can allow nurses to perform better surveillance, spend more time with patients, and detect complications earlier, leading to favorable patient outcomes (Clarke & Aiken, 2003) Causal Inference Matters To Nurses Assignment Paper. Many correlational studies support the association between better nurse staffing and improved outcomes. While compelling, correlations between better nurse staffing and favorable patient outcomes do not address whether the observed improvement in outcomes is a direct result of increased nurse staffing, or the systematic differences between high-performing and low-performing organizations. For example, hospitals with better nurse staffing may also have better overall financial resources, facility characteristics, and more positive work environments, possibly leading to a misattribution of the resulting improvements in outcomes to nurse staffing. As a result, even after over two decades of research and a plethora of correlational studies, we remain unsure whether increasing nurse staffing will cause an improvement in patient outcomes.
In this selected review, we aim to discuss the importance of causal inference in nursing and outcomes research, while pointing out caveats so that nurses can be informed consumers of research. We also aim to describe areas of opportunity for examining causal pathways from nursing to patient outcomes, to further support progress in the field of nursing outcomes research.
Conceptual Framework
Nursing theory has undergone a fundamental transformation in the understanding of causal mechanisms linking nursing to outcomes, from the linear structure-process-outcome model (Donabedian, 1978) to its most general adaptation called the Quality Health Outcomes Model, (Mitchell, Ferketich, & Jennings, 1997). The advantage of the Quality Health Outcome model (QHOM) is that, while it incorporates the linear pathway from nursing structure to process to outcomes, it also acknowledges the complex and multi-directional relationships among the three elements (Mitchell et al., 1997). Although QHOM accurately reflects the reality of nursing care and patient outcomes as embedded in a complex inter-related system of causal pathways, empirical studies often take a more naïve view by interpreting an association between nurse staffing and outcomes as evidence of an underlying causal pathway, directly linking better nurse staffing to improved patient outcomes. We use the QHOM and the causal pathways described in the QHOM as a guide in our selected review.
We focus on one structural indicator of nursing quality from the National Database of Nursing Quality Indicators- nurse staffing (defined as nurse to patient ratio, nurse work load, proportion of registered nurses (RN), and the number of RN hours per patient day). We examine three key study designs – cross-sectional, longitudinal, and randomized control trials – and use selected studies with these designs. Each of these studies was published within the last 20 years and is indexed in PubMed to illustrate some of the limitations for determining causal inference Causal Inference Matters To Nurses Assignment Paper.
Cross-Sectional Study Designs
A variety of studies have used cross-sectional designs to examine the relationships between nurse staffing and patient outcomes. A variety of studies have used cross-sectional designs to examine the relationships between nurse staffing and patient outcomes. At the hospital- or unit-level, these studies find that hospitals or units that have better staffing, have better patient outcomes (Aiken et al., 2011; 2014; Aiken, Clarke, Sloane, Sochalski, & Silber, 2002; Harless & Mark, 2010; Kane, Shamliyan, Mueller, Duval, & Wilt, 2007; Lang, Hodge, Olson, Romano, & Kravitz; Needleman et al., 2011; Needleman, Buerhaus, Mattke, Stewart, & Zelevinsky, 2002; Rafferty et al., 2007; Sales et al., 2008; Silber et al., 2016). Two systematic reviews concluded that, as nurse to patient ratios improved, significant reductions were identified in the odds of nosocomial sepsis, cardiac arrest, respiratory failure, failure to rescue (Kane et al., 2007) as well as lower inpatient mortality rates, and shorter hospital stays were identified (Lang et al. 2004).
Several of these studies use sophisticated statistical analyses and multi-variable regression modeling to remove bias from influence of system-level factors and patient characteristics. For example, Aiken and colleagues, in a unique multi-hospital nurse survey study, were able to adjust for a large number of hospital-level factors while examining nurse staffing and patient outcomes (Aiken et al., 2011; 2002). Even after carefully adjusting for potential confounding effects, the researchers demonstrated a significant association between improved nurse to patient ratios and decreases in 30-day mortality and failure to rescue in adult surgical patients (Aiken et al., 2011; 2002).
even the most sophisticated cross-sectional study designs are unable to robustly model the complexity and multi-dimensionality of nurse staffing and outcomes However, even the most sophisticated cross-sectional study designs are unable to robustly model the complexity and multi-dimensionality of nurse staffing and outcomes, as described in the QHOM. Because cross-sectional data analyses effectively collapse data from a dynamic system of multi-directional causal relationships into a point-in-time slice of this complex system, their ability to inform policy and practice is limited (Imbens & Rubin, 2015). Adjusting for even the most extensive set of hospital characteristics does not guarantee that residual biases from unobserved or unmeasured variables are not confounding the results. For example, interprofessional teamwork can be difficult to measure and could continue to bias results in favor of finding an association of better nurse staffing with improved outcomes, even after adjusting for the easier to measure variables. Confounding in cross-sectional studies may also lead to understated results. For example, a failure to fully adjust for differences in patient complexity between tertiary and community hospitals could result in a finding of an inverse association of nurse staffing with patient outcomes if better-staffed tertiary hospitals serve a patient population at higher risk for negative outcomes. Yet another concern regarding the potential for understated results is the risk of so-called “suppression.” Suppression occurs when researchers examine several nursing characteristics (e.g., staffing, skill mix, education, experience, expertise) as part of the same analysis model.
Suppression occurs when researchers examine several nursing characteristics as part of the same analysis model. Suppression can be problematic because the variables (i.e. staffing, education, experience) may be correlated and may lie on each other’s causal pathway to patient outcomes (Maxwell & Cole, 2007). For example, the causal pathway from increased registered nurse staffing to patient outcomes could be mediated by increased likelihood that a nurse has a baccalaureate degree (versus non-baccalaureate prepared nurse), which itself is a known independent contributor to patient outcomes (Aiken, Clarke, Cheung, Sloane, & Silber, 2003) Causal Inference Matters To Nurses Assignment Paper. Including nurse staffing and education in one analytic model could, therefore, “control away” part of the causal pathway from nurse staffing to outcomes, resulting in suppression, or an understatement of the relationship between nurse staffing and outcomes.
The level of aggregation of the study variables (hospital-level versus unit-level) may also have an effect on the study findings. Sales et al. (2008) compared hospital level and unit level analysis in approximately 120 Veterans Affairs Hospitals and found that the level of analysis did matter for findings (Sales et al., 2008). At the unit-level, there was a large difference in average nurse staffing for patients who had an intensive care unit (ICU) stay during their hospitalization compared to patients who were exclusively cared for on the medical-surgical floor. However, when examined at the hospital level, they found little difference between average nurse staffing for patients who had an ICU stay compared to patients who received care only on the medical-surgical floor. Additionally, patients who did not have an ICU stay experienced significantly lower mortality risk when cared for in better staffed units, whereas no association between staffing and mortality was identified for patients with an ICU stay. They concluded that aggregating nurse staffing to the hospital level, and effectively averaging patient and nurse staffing data at the hospital level, may produce biased estimates of the effects of nurse staffing. However, generally speaking, a higher level of aggregation (hospital-level) does not always exacerbate confounding from unobserved or unmeasured factors, nor does a finer level of aggregation (unit-level) necessarily alleviate it; the level at which the variables are measured only changes the type of variables most likely plaguing the results. For example, while unit-level comparisons of staffing and patient outcomes within one hospital might be less confounded by factors like availability of financial resources and hospital management practices, other factors such as unit culture or patient mix might play an inflated confounding role.
These types of confounding are virtually unavoidable in cross-sectional studies and could be responsible for a host of inconclusive findings of the link between nurse staffing and patient outcomes. Examples of these inconclusive results include: Needleman et al. (2002) reported an association between nurse staffing and mortality for surgical, but not medical patients; Van den Heede et al. (2009) found no significant association between nurse staffing and in-hospital mortality and failure-to-rescue; and a 2014 study by Kelly and colleagues did not demonstrate a significant association between hospital-level ICU staffing and mortality for mechanically ventilated adults (Kelly, Kutney-Lee, McHugh, Sloane, & Aiken, 2014).
Longitudinal study designs
Longitudinal studies that track changes in nurse staffing alongside changes in patient outcomes tend to be stronger in measuring causal effects Longitudinal studies that track changes in nurse staffing alongside changes in patient outcomes tend to be stronger in measuring causal effects (Imbens & Rubin, 2015) because they capture the dynamics of the associations of nurse staffing with patient outcomes. These analyses are relatively new in the literature.
The idea of the longitudinal approach is to track nurse staffing and patient outcomes within an organization over a period of time, thus removing confounding from differences in unmeasured system-level factors (e.g., financial resources, facility characteristics, work environment, patient acuity) that bias cross-sectional studies. An example is a longitudinal study by Weiss, Yakusheva, and Bobay (2011) that examined eight months of data on nurse staffing, 30-day readmissions, and emergency department use for general medical and surgical patients in 4 hospitals. Their results suggested that when unit-level nurse staffing was higher (i.e. fewer patients per nurse), rates of unplanned 30-day readmissions and emergency department (ED) use were lower (Weiss et al., 2011). Another example of a longitudinal study, but within a single center and single unit, was conducted in the United Kingdom (Tarnow-Mordi, Hau, Warden, & Shearer, 2000). In this 4-year study, researchers examined the relationship between nurse workload measured at the shift level within one adult ICU and its relationship with patient mortality. Their results suggested that ICU patients cared for by nurses with higher workloads are significantly more likely to die than those cared for by nurses that have lower workloads (Tarnow-Mordi et al., 2000).
Longitudinal studies are particularly robust when researchers are able to compare the longitudinal patterns of patient outcomes across several organizations or units measured concurrently, to test whether improvements in outcomes following an increase in nurse staffing are greater in organizations or units experiencing more drastic increases in nurse staffing. An example of this approach is a recent unit-level analysis over a 5-year period in 19 hospitals (Duffield et al., 2011). Duffield et al. (2011) found that when unit-level nurse staffing and workload were more favorable, incidence of patient falls and medication errors were lower.
A major limitation of longitudinal designs is that they are still subject to confounding and suppression effects A major limitation of longitudinal designs is that they are still subject to confounding and suppression effects from time-varying covariates, although, arguably, to a lesser degree than cross-sectional studies. For example, changes in quality-improvement practices or seasonal variation in patient outcomes could overstate or understate the effect of changes in nurse staffing on patient outcomes. Additionally, the statistical power might be limited if nurse staffing remains relatively stable during the study period. However, the fact that these types of studies tend to produce the most consistent evidence of a positive impact of nurse staffing on many patient outcomes is reassuring.
Randomized Controlled Trials (RCTs)
The “gold standard” of research evidence to establish the causal effect of nurse staffing on patient outcomes is a RCT [however] clinical equipoise does not exist to ethically move forward with a clinical trial to randomly assign patients to higher levels of nurse staffing. The “gold standard” of research evidence to establish the causal effect of nurse staffing on patient outcomes is a RCT. An example of an RCT in nursing outcomes research would include increasing nurse staffing (i.e. fewer patients per nurse) in a randomly selected subset of study hospitals (or units) and then, comparing patients’ outcomes pre- and post-intervention to hospitals (or units) where staffing was constant. Implementing this study design in a large representative sample of organizations would allow researchers to accurately measure the causal effect of a change in nurse staffing on patient outcomes. However, due to financial, logistical, and ethical barriers to randomization of patients to different levels of nurse staffing, no randomized controlled trials of nurse staffing and patient outcomes currently exist in the literature. Most importantly, given the robust evidence of an association between nurse staffing and outcomes, clinical equipoise does not exist to ethically move forward with a clinical trial to randomly assign patients to higher levels of nurse staffing.
The studies closest to an interventional design in nursing outcomes literature are analyses of the California mandated nurse staffing ratios and its association with patient outcomes (Aiken et al., 2010; Bolton et al., 2007; McHugh, Kelly, Sloane, & Aiken, 2011; Spetz et al., 2009). These studies were conducted after Assembly Bill 394 was signed into legislature in 2004 in California. The study design is technically a “natural experiment” because the “intervention,” or increased nurse staffing, did not occur in a controlled study setting. Natural experiments tend to be not as robust as true RTCs; for example, the intervention in this case was limited to one state and was not randomized Causal Inference Matters To Nurses Assignment Paper.
While mandated nurse-to-patient ratios did increase the number of nursing hours per patient day in California (Spetz et al., 2009), this did not uniformly translate to improved patient outcomes. A report by the California Nursing Outcomes Coalition (Donaldson, et al., 2005) documented no statistically significant change in patient safety and quality outcomes, such as decreased falls or the prevalence of pressure ulcers (Hertel, 2012). Two other studies did not demonstrate an association between mandated nurse-to-patient ratios in California and fewer pressure ulcers (Bolton et al., 2007), lower rates of failure to rescue, pressure ulcers, or deep vein thrombosis (Spetz et al., 2009). However, none of these analyses included a control group. In other words, adjustment for patient outcomes in other states was not performed, which could have biased the findings; patient outcomes could have been improving or worsening for reasons unrelated to California’s mandate. This point is critical because the timing of the California minimum nurse staffing mandate overlapped with the introduction of Deficit Reduction Act of 2005 that introduced sweeping changes to the Medicare and Medicaid programs, significantly limiting healthcare access for some of the most vulnerable segments of the population (Rosenbaum & Markus, 2006).
The strongest evidence of the effect of nurse staffing on patient outcomes, and one most closely approximating a RCT, compared nurse-to-patient ratios in California with similar ratios in states without the minimum staffing mandate (Aiken et al., 2010). The researchers used Pennsylvania and New Jersey as the two states without nurse staffing legislation, and examined the effect of the California mandate while accounting for trends in nurse staffing and patient outcomes in the unaffected states. After collecting primary survey data from 22,336 hospital staff nurses in California, Pennsylvania, and New Jersey in 2006 and linking it to state hospital discharge databases of patient outcomes, the researchers found that the California’s mandated nurse staffing ratios led to significant reduction in patient mortality. Using predicted probabilities, they concluded that if New Jersey and Pennsylvania were to have the same nurse-to-patient ratios as California, there would be approximately 14% fewer deaths in New Jersey and 11% fewer deaths in Pennsylvania (Aiken et al., 2010).
Conclusion and Future Research
Overall, it is clear that cross-sectional studies provide support for a strong association of nurse staffing and better patient outcomes. Longitudinal analyses (ones that track changes in nursing and patient outcomes overtime), and study designs closest to a randomized controlled trials (Aiken et al., 2010) are more likely to produce evidence of causal relationship between nurse staffing and patient outcomes, as compared to cross-sectional analyses. Longitudinal and RCT designs are able to reduce confounding from the multi-directionality of the causal pathways, so that the findings can be interpreted as evidence of the underlying linear structure-process-outcome causal pathway. However, longitudinal studies and RCTs are currently limited in the literature.
Future research should focus on strengthening the evidence base of a causal relationship between nurse staffing and patient outcomes. Future research should focus on strengthening the evidence base of a causal relationship between nurse staffing and patient outcomes. While a RCT of nurse staffing and patient outcomes may not be feasible or ethical, practical approaches to strengthening causal inference still exist. These approaches include carefully studying policy changes and their impact on patient outcomes, like California staffing mandates or the Institute of Medicine recommendation to increase nurse education. Increasing use of electronic health records across American hospitals also provides researchers with a more readily accessible source of longitudinal data for generating causal evidence of the relationship between nursing and patient outcomes. Researchers should continue to use these approaches to strengthen the evidence of a causal pathway between nurse staffing and outcomes.
ORDER YOUR PAPER HERE
Combining causal analyses with large, representative data sets of hospitals is the important next step Nonetheless, it is important that an increased focus on causality does not come at the expense of generalizability. Now, existing evidence is primarily derived from either large generalizable cross-sectional studies with limited potential for causal inference, or from causal longitudinal and policy evaluation studies with limited generalizability. Combining causal analyses with large, representative data sets of hospitals is the important next step in informing policy and practice to identify causal pathways of how nurse staffing contributes to meaningful improvements in patient outcomes.
References
Aiken, L. H., Cimiotti, J. P., Sloane, D. M., Smith, H. L., Flynn, L., & Neff, D. F. (2011). Effects of nurse staffing and nurse education on patient deaths in hospitals with different nurse work environments. Medical Care, 49(12), 1047-1053. doi:10.1097/MLR.0b013e3182330b6e
Aiken, L. H., Clarke, S. P., Cheung, R. B., Sloane, D. M., & Silber, J. H. (2003). Educational levels of hospital nurses and surgical patient mortality. JAMA: The Journal Of The American Medical Association, 290(12), 1617-1623.
Aiken, L. H., Clarke, S. P., Sloane, D. M., Sochalski, J., & Silber, J. H. (2002). Hospital nurse staffing and patient mortality, nurse burnout, and job dissatisfaction. JAMA: The Journal Of The American Medical Association, 288(16), 1987-1993.
Aiken, L. H., Sloane, D. M., Bruyneel, L., Van den Heede, K., Griffiths, P., Busse, R., Sermeus, W. (2014). Nurse staffing and education and hospital mortality in nine European countries: A retrospective observational study. The Lancet, 383(9931), 1824-1830. doi:10.1016/S0140-6736(13)62631-8
Aiken, L. H., Sloane, D. M., Cimiotti, J. P., Clarke, S. P., Flynn, L., Seago, J. A., Smith, H. L. (2010). Implications of the California nurse staffing mandate for other states. Health Services Research, 45(4), 904-921. doi:10.1111/j.1475-6773.2010.0114.x. Causal Inference Matters To Nurses Assignment Paper