Estimating the economic cost of nurse sensitive adverse events amongst patients in medical and surgical settings

Aims: To identify the costs associated with nurse sensitive adverse events and the impact of these events on patients’ length of stay. Design: Retrospective cohort study using administrative hospital data. Methods: Data were sourced from patient discharge information ( N = 5544) from six acute wards within three hospitals (July 2016– October 2017). A retrospective patient record review was undertaken by extracting data from the hospitals’ administrative systems on inpatient discharges, length of stay and diagnoses; eleven adverse events sensitive to nurse staffing were identified within the administrative system. A negative binomial


| INTRODUC TI ON
Internationally, there is an increased focus on ensuring patient safety and quality of care (Ehsani et al., 2006); this is driven, in part, by the substantial rate of adverse events in hospital settings (Rafter et al., 2017;Schwendimann et al., 2018;Thomas & Brennan, 2001;Vincent et al., 2001). These adverse events can result in serious health outcomes for patients and financial consequences for health systems (Ehsani et al., 2006;Kjellberg et al., 2017). The clinical implications and human burden associated with adverse events are well established (Needleman et al., 2006) with an economic analyses of the costs of these adverse events now emerging (Mittmann et al., 2012). These economic approaches include several retrospective cohort studies, employing regression analyses to estimate costs attributable to all adverse events.

| BACKG ROU N D
While some adverse events are unavoidable, the key challenge is preventable adverse events; that is, those events that occur in health care as a result of errors of commission or omission (Schwendimann et al., 2018). Estimates from international studies reveal that adverse events can occur in anywhere between 3%-22% of admissions (Rafter et al., 2017;Schwendimann et al., 2018), of which anywhere between 50% (de Vries et al., 2008) and 73% (Rafter et al., 2017) are considered preventable. One approach to determining the economic costs of adverse events is through the measurement of nurse sensitive adverse events, which, following Needleman et al., (2002) seminal work, are defined as patient outcomes that are potentially sensitive to nursing care.
One factor that can contribute to the risk of certain adverse events is nurse staffing, with low levels of nurse staffing associated with a number of adverse outcomes including mortality (Aiken et al., 2014;Driscoll et al., 2018;Griffiths et al., 2019;Kane et al., 2007;Needleman et al., ,,2006Needleman et al., ,, , 2011, missed care Kalisch, 2006;Recio-Saucedo et al., 2018), hospital-acquired pneumonia  and increased length of stay (LOS) (Duffield et al., 2011;Griffiths et al., 2018;Kalisch, 2006;Twigg et al., 2013). Without adequate staffing, nurses may not have the capacity to proactively minimize an adverse event that requires early detection and prompt intervention. However, as health services internationally are operating in an environment where resources are scarce and efficiencies are widely sought (Mittmann et al., 2012), nurse staffing is frequently the area of the health workforce that is reduced when savings are required (Pappas, 2008;Williams et al., 2017), and as a consequence, nurse sensitive adverse events can increase, which have significant human and economic costs.
A number of studies have explored the costs associated with an adverse event in a hospital setting (Kjellberg et al., 2017); however, there are a limited number of studies that have explored the relationship between nurse staffing and the costs associated with nurse sensitive adverse events in both medical and surgical patients. Those studies that have identified nurse sensitive adverse events predominantly align costs to medication errors, falls, pneumonia, urinary tract infections and pressure ulcers (Pappas, 2008;Tchouaket et al., 2017). Nurse sensitive adverse events are generally costed as a combined variable or individually with valid sample sizes ranging from 2495 (Pappas, 2008) to 2699 patients (Tchouaket et al., 2017). Overall prevalence of nurse sensitive adverse events were shown to range from 14.4% (surgical patients) to 21.5% (medical patients) with costs per case increasing by $1029 for medical patients and $903 for surgical patients who experienced an adverse event (Pappas, 2008). Previous studies have analysed costs at patient level within an individual hospital system with few studies identified that examined costs in a national system. Therefore, as Pappas (2008) highlights, there is a need to identify the costs associated with these adverse events. This will allow nurse leaders and policy makers to identify how the nursing resource in the provision of quality health care can impact on the economic outcomes associated with patient care (Pappas, 2008).
One approach to determining the economic costs of adverse events is through the measurement of adverse events that are sensitive to changes in nursing input, referred to as nurse sensitive outcomes. Needleman et al., (2002)  In addition, in recent years, a number of health systems have introduced a more systematic approach to determining nurse staffing levels and skill mix in hospital settings; the introduction of these systematic approaches has been associated with a need to identify the economic benefits of their introduction. However, with a few exceptions, studies in this area are limited and a recent review found an K E Y W O R D S adverse events, costs, economics, nurse staffing, nursing, nursing sensitive outcomes, patient outcomes absence of economic evaluations of staffing methodologies or tools (Griffiths et al., 2020).
In 2018, the Department of Health in Ireland published a policy document titled a Framework for Safe Nurse Staffing and Skill Mix in General and Specialist Medical and Surgical Care Settings in Adult Hospitals in Ireland (henceforth referred to as the Framework) (Department of Health, 2018). This document set out a number of recommendations, including the introduction of nursing hours per patient day to determine staffing levels. The introduction of the Framework was aligned to a programme of research which is examining the relationship between intentional changes to nurse staffing and outcomes, including the extent to which economic costings of adverse events related to patient outcomes change according to variations in nurse staffing. This includes examining the extent to which the costs in increasing nurse staffing are offset by a reduction in costs associated with a reduction in adverse events (see Drennan et al., 2018 for further details of the research). The first stage in this process, and the aim of this paper, is to identify costs associated with nurse sensitive adverse events and the impact of these events on patients' LOS.

| Aims
The aims of this study were to identify the costs associated with nurse sensitive adverse events and the impact of these events on patients' LOS.

| Design
A retrospective patient record review was undertaken by extracting data from the hospitals' administrative systems.

| Participants
Data were collected from six acute adult wards within three Irish hospitals from July 2016 to October 2017 for 5544 admitted patients; these patients had a minimum stay of 24 h, and all were adult patients (aged 18 years and older). Hospitals in Ireland are classified into four generic types ranging from Model 1 (community hospital with subacute inpatient beds), Model 2 (inpatient care for low-risk medical patients), Model 3 (larger district hospital that admits acute medical and surgical patients) and Model 4 (large university teaching hospitals) (Acute Medicine Programme Working Group 2010). The hospitals in this study included a Model 2 hospital (109 beds), a Model 3 hospital (235 beds) and a Model 4 hospital (670 beds); in total, there are 38 hospitals in Ireland in Bands 2 to 4. The wards in each of the study hospitals were enrolled in a pilot study implementing a safe nurse staffing framework in medical, surgical and specialist settings in Ireland (Drennan et al., 2018). One of the first phases of this research was to estimate the costs associated with nurse sensitive adverse events.

| Data collection
In line with previous studies that estimate cost of total adverse events (Ehsani et al., 2006;Hoonhout et al., 2009;Mittmann et al., 2012), a retrospective patient record review was undertaken by extracting data from the Hospital In-Patient Enquiry (HIPE) system. The HIPE system is the method used in Ireland for collecting data on inpatient discharges and includes data on LOS and diagnoses within the hospital setting (O'Loughlin et al., 2005). Diagnoses are captured by ICD-10 codes and assigned diagnostic-related group (DRG) codes; the latter codes group cases that are clinically similar and that consume similar amounts of health care resources. In this study, ICD-10 codes are employed to determine the presence of an adverse outcome sensitive to nursing, and DRG codes are employed to estimate prices as described below.

| Nurse sensitive adverse events assessment
Based on the work of Needleman et al., (2002), 11 outcomes sensitive to nurse staffing were identified within the HIPE system: hospitalacquired urinary tract infections, pressure ulcers, hospital-acquired pneumonia, deep venous thrombosis, upper gastrointestinal bleeding, CNS complications, hospital-acquired sepsis, shock/cardiac arrest, wound infection, pulmonary failure and physiological/metabolic derangement: herein referred to as nurse sensitive adverse events. These outcomes were identified from the hospital discharge database using the algorithm developed by Needleman et al., (2002) in the United States. The US algorithm used the US ICD-9 coding system whereas Ireland uses the ICD-10; this is the same system used in Australia and New Zealand where McCloskey and Diers (2005) mapped the original ICD-9 codes in Needleman's study to the ICD-10 codes and these were used in this study (see Drennan et al., 2018 for further details).

| Validity
The identification of costs associated with nurse sensitive adverse events at patient and unit level, as in this study, has been reported as having greater validity than measuring costs from large administrative data sets that do not distinguish from those events that are not nurse sensitive (Pappas, 2008). The analysis of the HIPE data allows costs to be estimated at patient level.

| Pricing methodology
Data from HIPE are not linked to a financial system that generates costs/prices per patient; however, the Health Pricing Office (HPO, 2019) has published a price list which provides the prospective description of prices to be paid for providing services for admitted patients by DRG. These prices are based on retrospective cost and activity data, and adjustments are made for funding policies, reflecting expected expenditure and available budgets. The primary function of the HPO price list is to inform budget allocations for acute public hospitals in Ireland (where the payer is also the provider). We employed this list to estimate the price of episodes of care captured in the data set (using assigned DRGs) and to compare LOS to national averages for those DRGs.
The national price list (HPO, 2019) is generated through activitybased funding (ABF) methods using a relative measure of resource consumption called weighted units; these are set so that the average resource consumption across all cases is one. Cases that consume resources more or less than average resource consumption have a weighted unit greater or less than one. To estimate weighted units, the Health Pricing Office estimate relative values (RVs) associated with the diagnosis. There are separate RVs for day cases and inpatients and furthermore, there are separate RVs for inpatients that have a typical LOS, referred to as inliers, and for high and low outliers; that is, LOSs that are substantially above or below the inlier range. The monetary value of any case is the total weighted unit multiplied by the base price associated with one weighted unit (for further information on methodology employed see HPO (2019)). For example, for a patient assigned DRG IO3B (hip replacement, minor complexity), the price list reveals the national average LOS is 7.2 days (range 2−20 days); the RV associated with this DRG is 2.083 which applies to LOS ranging from 2-20 (referred to as inlier relative value) and price is set at €10,038 (weighted unit base price is approximately €4819). LOS above 20 is associated with 0.098 RV per additional day (there is no lower RV for this DRG). Therefore, if a patient who is assigned DRG 103B has a LOS of 23 days, they have three high outlier days. Their total weighted unit then is the RV for the inlier period (2.083) plus the RV for the high outlier days (0.098 * 3 days), giving a price of €11,456 for the episode. This translates to €498 daily average (total price/LOS). This methodology is employed to estimate total and daily average prices for all patients enrolled in the study.

| Data analysis
In line with previous studies (Brown et al., 2002;Ehsani et al., 2006;Hoonhout et al., 2009;Kjellberg et al., 2017;Rafter et al., 2017;Vincent et al., 2001), to estimate the financial impact of nurse sensitive adverse events, we determined their impact on LOS, specifically on LOS beyond the national average for that DRG termed 'additional LOS'. We hypothesized that the presence of nurse sensitive adverse event(s) will increase LOS, to which we can assign a price to quantify the economic impact of nurse sensitive adverse events. LOS is treated as a count variable (nonnegative integer). As there was overdispersion in the count data (where the mean of the variable is smaller than the variance of the variable; Jones, 2007), we employ a negative binomial model (over the Poisson model) (Deb et al., 2017). This count model is a generalization of the Poisson regression model because it has the same mean structure as a Poisson regression and it has an extra parameter (a gamma-distributed error term) to model the overdispersion (Cameron & Trivedi, 2005). Maximum likelihood estimates are used, wherein iterations are generated until the change in log likelihood is sufficiently small. Within the model, the outcome (y) depends on a set of explanatory variables (x), and it is assumed the probability of the event occurring (λ) = exp(xβ).
The exponential function is employed to ensure that the intensity of the process so that the mean number of events, given x, is positive (Jones, 2007). Within the model the coefficients can be tested using a conventional t test; the conditional mean is a function of the explanatory variables, and alpha estimates the degree of overdispersion in the data. The variance is a quadratic function of the mean, and the approach has been applied extensively in health studies (Jones, 2007). The function form can be presented pected value of the outcome variable y i for subject i, x i are the independent variables with corresponding regression coefficients β n and i is the disturbance term (see Cameron and Trivedi (1998)).
In this study, the additional LOS is the dependent variable (y); the explanatory variables (x) are the number of nurse sensitive adverse events, patients' age and gender, admission type and case complexity. Incidence rate ratios and the average marginal effects are estimated. Data were analysed using Stata Version 14 (StataCorp, 2015).

| E THIC AL CONS IDER ATIONS
The research ethics committees of the three hospitals in which the research took place granted ethical approval.

| Characteristics of the sample
In total, data were collected on 5544 inpatients; 50% of patients were male and the average age was 62.4 years (SD 19.56) (see Table 1). The majority of patients were admitted as emergency cases (vs. elective) (83%). With regard to complexity, over a third of patients in the sample were classified as having major complexity, 11% intermediate and 52% minor (this classification was determined from the DRG assigned). The mean LOS was 9.9 days (see Table 1); 10% of patients had LOS beyond the national average for their DRG.
Applying the HPO methodology, as discussed above, to the DRGs and LOS data extracted for the sample, we estimated the price per admission for patients in the study. The average price per admission was €8544 (SD €13,318). The average daily price per patient included in the study, based on their individual LOSs, was €1456 (SD €1515).
Sixteen per cent of the sample (n = 897) had at least one nurse sensitive adverse event during their episode of care. Of these, 76% had one event; 20% two events; and 4% had three or more nurse sensitive adverse events. Amongst those who had nurse sensitive adverse events, the average number was 1.3 (SD 0.56). Metabolic derangement was the most frequently occurring event (36.7%, n = 330), followed by hospitalacquired pneumonia (24.9%, n = 223) and urinary tract infection (22.6%, n = 203). Those who had at least one nurse sensitive adverse event were predominately female (52%), had an average age of 71.8 years, 94% were classified as an emergency admission and 71% were assigned a DRG with an intermediate complexity classification level. In addition, amongst those patients who experienced at least one nurse sensitive adverse event, the mean LOS was 18.32 days (see Table 1); 19% of patients had LOS beyond the national average for their DRG, which was 4.37 days on average.  For ease of interpretation, the results are presented as incident rate ratios (see Table 2): each nurse sensitive adverse event was associated with a 25% increase in additional LOS (p =.001), holding all else constant. To understand the model better in terms of LOS, the marginal effects (see Table 2

| Economic impact
Using the results of the negative binomial regression, whereby each nurse sensitive adverse event increased the LOS beyond national average by 0.48 days and applying the estimated daily average price of inpatient stay per patient (€1456), we estimated the economic TA B L E 1 Summary statistics of sample.

Full sample (n = 5544)
Nurse sensitive AE (n = 897) impact associated with each nurse sensitive outcome to be €694; this accumulated to €0.8 million for the study population.

| DISCUSS ION
This is the first study in Ireland to report the costs associated with nurse sensitive adverse events and builds on previous research in this area through the use of routinely collected data and by considering a wider range of nurse sensitive adverse events than previously reported (Pappas, 2008;Tchouaket et al., 2017). The occurrence of a nurse sensitive adverse event was 16%; this is higher than adverse events previously reported by Tchouaket et al., (2017) in Canada (6.2%), similar to rates reported for surgical patients (14.4%) but lower than that reported for medical patients in the United States (21.5%) (Pappas, 2008). However, it is of note that this study measured a wider range TA B L E 2 Negative binomial regression: impact of nurse sensitive adverse event on length of stay. The z value follows a standard normal distribution which is used to test against a two-sided alternative hypothesis that the Coef. is not equal to zero. The probability that a particular z test statistic is as extreme as, or more so, than what has been observed under the null hypothesis is defined by p > z. b Likelihood ratio test of alpha =0. This is the likelihood ratio chi-square test that the dispersion parameter alpha is equal to zero.
c Chibar2(01) indicates that the distribution on the likelihood ratio test statistics is a 50:50 mixture of a chi-square with no degrees of freedom and a chi-square with 1 degree of freedom (rather than the usual chi-square with 2 degree of freedom) (StataCorp, 2015).
of nurse sensitive adverse events compared with previous research; in particular, when compared with Pappas (2008) and Tchouaket et al., (2017) whose research measured five nurse sensitive adverse events.
As would be expected, and similar to other studies, patients who experienced a nurse sensitive adverse event while in hospital had a longer LOS when compared with patients without (Ehsani et al., 2006).
The descriptive statistics reveals higher additional LOS amongst those who had at least one nurse sensitive adverse event compared with the general sample (2.60 compared with 4.37). This is at the lower end of Tchouaket et al., (2017) estimates that adverse events added 4.0 to 12.3 days to LOS, depending on the adverse event experienced by the patient. Additionally, the regression reveals that each nurse sensitive adverse event increases average LOS beyond the national average by 0.48 days. In addition, we estimated the average cost associated with each nurse sensitive adverse event to be €694; this was lower than the costs identified in the United States, where the cost per adverse event per case increased by €1590 in medical patients and €1396 in surgical patients (2008 USD converted to 2019 EUR) (Pappas, 2008).
A previous study in Ireland identified a total of 247 adverse events in 211 admissions, and assigned an average cost for adverse events of €5550, accumulating to over €194 million annually (Rafter et al., 2017).
The study presented here advances on this methodology. Rather than applying an average cost for events, we take into consideration the patients' diagnosis and compare their length of to stay to what is typical for that diagnosis and assign values using the recently published Admitted Patient Price List (HPO, 2019). Therefore, applying more precise estimates, as advocated by Drummond et al., (2015 (Connors, 2018).
Following the international financial crisis of 2008, there was pressure on national public health care services to make cost savings either through staff reductions or replacing qualified staff with unregistered roles. This resulted in the reduction of the nursing workforce in Ireland between 2008 and 2014 where there was a 10.7% decrease in nurses at staff nurse grade (Williams & Thomas, 2017). Simultaneously, health care costs are rising, inhibiting investments in new initiatives, even those aimed at improving patient safety (Kjellberg et al., 2017). This has direct consequences for front line personnel such as nurses, whose workload has increased due to increasing patient complexity and dependency (Duffield et al., 2011. Nursing staff play a central role in ensuring patient safety and patient surveillance. In light of cost reduction efforts, resulting shortages and ensuing pressures, these staff have to increase their workloads and provide efficiencies. While nursing costs represent at least 50% of most hospitals' expenses, this cohort of staff drive overall hospital quality and safety and are an important component of hospitals' infrastructure, as well as an effective intervention for achieving operational efficiency and success (Coster et al., 2018;Pappas, 2008). Knowing the costs of preventable adverse events from an economic perspective is valuable, and estimates can be used to inform decisions regarding designing and investing in patient safety initiatives, health policy and priority setting (Hoonhout et al., 2009;Mittmann et al., 2012). However, the evidence that does exist in this area is both limited and mixed (Griffiths, Ball, Drennan, et al., 2016;Twigg et al., 2015). This study estimates the economic costs associated with potentially avoidable adverse events in Ireland, using outcomes that Needleman et al., (2002) (Kjellberg et al., 2017;Nielsen et al., 2013;Rasmussen et al., 2014); therefore, initiatives that aim to prevent such events have the potential to generate significant cost savings to the health system. For example, increasing nurse staffing levels and improving nurse-to-patient ratios have been found to reduce patient mortality (Aiken et al., 2014;Driscoll et al., 2018;Griffiths et al., , 2019Rothberg et al., 2005) and generate financial savings (Needleman et al., 2006;Newbold, 2008;Rothberg et al., 2005;Twigg et al., 2013). Studies, such this one, provide information on health care costs which can be used alongside initiatives preventing adverse events thus providing insights into the mechanisms for improving safety, patient satisfaction and perhaps generating cost savings (Kjellberg et al., 2017 Therefore, while we acknowledge the value assigned to nurse sensitive adverse events in this analysis is a proxy for the true cost, and may be an underestimation, it is the best estimate available.

| CON CLUS ION
This study estimates the economic cost of nurse sensitive adverse events in the acute care setting throughout Ireland; however, the approach used will also be applicable to researchers and policy makers internationally. These estimates quantify, in monetary terms, the value of potentially avoidable adverse events, which are substantial at €91.3 million annually. The estimates provided here can be used to inform and prepare the way for future economic evaluations of nurse staffing initiatives that aim to improve patient care and safety.
Such analyses could inform future budgetary and resource allocation decisions; thereby providing effective information to support financial, managerial and policy functions in the health service while maintaining and improving patient outcomes. While not all nurse sensitive adverse events are preventable, the identification of these events in secondary data and the impact that they have on patient outcomes and costs can facilitate the identification of those that can be avoided and used in the measurement of the association between nurse staffing levels and proportion of adverse events that occur in health care settings.

ACK N OWLED G EM ENTS
We

CO N FLI C T O F I NTE R E S T
Jonathan Drennan is a member of the Editorial Board of the Journal of Advanced Nursing.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/jan.14860.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available on request due to privacy/ethical restrictions.