Bed-to-nurse ratios, provision of basic nursing care, and in-hospital and30-day mortality among acute stroke patients admitted to an intensivecare unit: Cross-sectional analysis of survey and administrative data

Sung-Hyun Cho a,*, Sung-Cheol Yun b

a Department of Nursing, Hanyang University, Seoul, Republic of Koreab Department of Preventive Medicine, College of Medicine, University of Ulsan, Seoul, Republic of Korea

International Journal of Nursing Studies 46 (2009) 1092–1101

A R T I C L E I N F O

Article history:

Received 29 August 2008

Received in revised form 5 December 2008

Accepted 1 February 2009

Keywords:

Basic care

In-hospital mortality

Intensive care unit

Nurse staffing

Stroke

30-Day mortality

A B S T R A C T

Background: The literature reports inconsistent evidence of the effects of nurse staffing on

mortality despite continuing examination of this association.

Objective: To examine differences in provision of basic nursing care and in-hospital and

30-day mortality by nurse staffing of ICUs and general wards among acute stroke patients

admitted to ICUs during hospitalization.

Design: A cross-sectional design that included survey and administrative data.

Settings and participants: The study included 6957 patients with hemorrhagic and

ischemic stroke who were admitted to ICUs of 185 Korean hospitals.

Methods: Nurse staffing of ICUs and general wards was graded based on the bed-to-nurse

ratios of each hospital. Provision of basic care was measured by whether five activities,

such as bathing and feeding assistance, were fully provided by ICU nursing staff without

delegation to patient families. Hospitals were categorized into low, middle, and high

mortality groups for in-hospital and 30-day mortality based on z-scores that indicated

standardized difference between observed and expected mortality after controlling for

patient characteristics.

Results: In 83.8% of hospitals, basic care was provided fully by ICU nursing staff. The

overall in-hospital and 30-day mortality rates were 21.9 and 25.4%, respectively. Hospitals

with higher ICU staffing were more likely to fully provide basic care. Better ICU and general

staffing tended to be associated with lower in-hospital and 30-day mortality. Compared

with in-hospital mortality, 30-day mortality had a more distinct increase as nurse staffing

became worse.

Conclusion: The findings provide evidence that nurse staffing may impact provision of

basic care and patient mortality and suggest the need for policies for providing adequate

nurse staffing.

� 2009 Elsevier Ltd. All rights reserved.

Contents lists available at ScienceDirect

International Journal of Nursing Studies

journal homepage: www.elsevier.com/ijns

* Corresponding author. Tel.: +82 2 2220 0798; fax: +82 2 2295 2074.

E-mail address: sunghcho@hanyang.ac.kr (S.-H. Cho).

0020-7489/$ – see front matter � 2009 Elsevier Ltd. All rights reserved.

doi:10.1016/j.ijnurstu.2009.02.001

What is already known about the topic?

� A

cute stroke patients have high risk for physiologicalchanges and thus, physiological assessment and inter-ventions provided by nurses are essential to decreasepatient mortality and disability.

S.-H. Cho, S.-C. Yun / International Journal of Nursing Studies 46 (2009) 1092–1101 1093

� T

he literature reports inconsistent evidence of the effectsof nurse staffing on hospital mortality despite continuingexamination of this association.

What this paper adds

� H

ospitals with higher ICU staffing were more likely tofully provide basic nursing care. � B etter ICU and general ward staffing tended to be

associated with lower in-hospital and 30-day mortality.

� C ompared with in-hospital mortality, 30-day mortality

had a more distinct increase as nurse staffing becameworse.

Stroke is the second leading cause of death in the world,accounted for 9.7% of all deaths in 2004, and is estimated toincrease to 12.1% of deaths worldwide by 2030 (WorldHealth Organization, 2008). In most developed countries,stroke is the third commonest cause of death (Sarti et al.,2000) and in Korea, is the second leading cause of deaths,accounting for 12.3% of total deaths in 2006 (KoreaNational Statistical Office, 2007). Stroke is also 1 of 10leading diseases in the global disease burden that increasesyears of life lost due to mortality and years of life lived withdisability (Lopez et al., 2006). The burden of stroke isexpected to rise, especially in low-income and middle-income countries (Strong et al., 2007).

The high mortality and morbidity of stroke indicatesthat nationally and internationally nurses encounterstroke patients quite frequently and have been takingcare of them across the spectrum of stroke care fromprevention to rehabilitation. In particular, acute strokepatients have high risk for physiological changes that canresult in adverse patient outcomes. Therefore, nurse-ledphysiological assessment (e.g., blood pressure, oxygensaturation, blood glucose, and body temperature) andsubsequent interventions are essential to decrease patientmortality, neurological impairment, and disability (Joneset al., 2007). The literature also reports that ‘skilled nursingcare’ is one of the characterizing features of stroke units,which are known to be more effective in reducing post-stroke morbidity and mortality than conventional wards(Langhorne and Pollock, 2002; Burton et al., 2009). Giventhat stroke patients require extensive and highly skillednursing care, the outcomes of stroke patients may be moresensitive to nursing care than those of other clinicalpopulations or overall patient groups.

To provide stroke patients with high quality nursingcare and prevent avoidable death and disability, a safe andadequate work environment for nurses should be main-tained in hospitals and care units. One aspect of thenursing work environment, nurse staffing (e.g., patient-to-nurse ratios and bed-to-nurse ratios), has been examinedas one of the critical factors that could influence nursingcare quality and patient outcomes. Despite accumulatingempirical research, however, the literature reports incon-sistent evidence of the association between nurse staffingand patient outcomes (Clarke, 2007). Examining strokepatients, whose outcomes are assumed to be particularlysensitive to nursing care, may elucidate the linkagebetween nurse staffing and patient outcomes.

1. Background

The quality of stroke care has been examined by usingpatient outcomes including pathophysiological para-meters, death, impairment, and activity limitation (Dun-can et al., 2000). Among those outcomes, mortality is themost frequently used outcome and can be further specifiedinto in-hospital death, 30-day and 3-month mortality, etc.The Agency for Healthcare Research and Quality (AHRQ)proposed ‘acute stroke mortality rate’ as one of 32inpatient quality indicators that measure hospital qualityof care using hospital administrative data (AHRQ, 2007).The acute stroke mortality rate was defined as the numberof deaths in the hospital out of all discharges of age 18years and older, and with the principal diagnosis code forstroke (AHRQ, 2008). In-hospital mortality, however, canlead to information bias because in-hospital deaths areinfluenced by the lengths of stay of hospitals, and alsomany stroke deaths occur after the initial acute hospita-lization (AHRQ, 2007). Therefore, data linkage to deathrecords to track deaths after discharge is recommended,and 30-day mortality, defined as deaths within 30 days ofhospital admission, can be a complementary measure toin-hospital mortality. The Organisation for Economic Co-operation and Development (OECD) also suggested ‘in-hospital mortality rate within 30 days of hospital admis-sion for stroke’ in the Health Care Quality Indicators Project(Armesto et al., 2007). This indicator, proposed as a proxyfor true 30-day fatality without requiring tracking patientsafter discharge should be lower, however, than true 30-daymortality (Mattke et al., 2006).

Among factors influencing patient outcomes, nursestaffing has been an interest of nurses and nursingresearchers. However, the literature reports inconsistentlyon the effect of nurse staffing on patient outcomes acrossstudies (Clarke, 2007). Kane et al. (2007) concluded fromtheir systematic review and meta-analysis that increasedRN staffing was associated with lower hospital relatedmortality in intensive care units (ICUs). Two systematicreviews reported the literature offered minimal supportfor the staffing-outcome association, and the estimates ofnurse staffing effects were unreliable (Lang et al., 2004;Lankshear et al., 2005). A systematic review of ICU studiesreported a weak relationship between nursing resourcesand mortality (West et al., 2009). Other reviews on ICUstudies also reported no association of nurse staffing andhospital mortality and concluded that hospital mortalitymight not be sensitive enough to detect the consequencesof low nurse staffing in ICUs (Carmel and Rowan, 2001;Numata et al., 2006). However, most ICU studies selectedin these systematic reviews evaluated only in-hospitalmortality, excluding mortality after discharge (e.g., 30-daymortality). The inconsistent evidence and lack of con-sideration of post-discharge mortality suggest thatongoing research is needed to investigate the effect ofnurse staffing on 30-day mortality as well as in-hospitalmortality.

Staffing-outcomes research faces common methodolo-gical challenges, including staffing measurement, riskadjustment, and statistical modeling (Mark, 2006; Clarkeand Donaldson, 2008). In measuring nurse staffing, one

S.-H. Cho, S.-C. Yun / International Journal of Nursing Studies 46 (2009) 1092–11011094

issue is level of measurement—whether nurse staffingmeasured represents that of the whole organization orspecific targeted nursing units. Researchers in recentstudies prefer to use unit-specific staffing measuresbecause they are more valid and theoretically appropriatethan aggregated ones at the hospital level. Staffing-outcomes research also requires sound risk adjustmentto take into consideration differences in risks of patientoutcomes across hospitals. In developing statistical mod-eling, researchers have recommended that data on patientswho are clustered within organizations be treated byadvanced analytic methods (e.g., multilevel analysis).

Another possible contribution of nurses to strokepatient outcomes is found from studies that comparedstroke unit care with conventional care. Systematic reviewand other studies reported that case fatality and disabilitywere reduced in organized stroke units as compared withconventional wards (Stroke Unit Trialists’ Collaboration,1997; Kalra et al., 2000; Scottish Intercollegiate GuidelinesNetwork, 2002). Evans et al. (2001) suggested that thefavourable outcomes in stroke units were associated withbetter management process (e.g., more frequent patientmonitoring and early nutrition) and fewer secondarycomplications. Candelise et al. (2007) defined the strokeunit as a hospital ward with dedicated beds (at least 80%stroke admission) and with a dedicated stroke staff (i.e., atleast one physician and one nurse) who work exclusively inthe care of stroke patients. These study findings may reflectthat stroke units enable nurses to be dedicated to strokecare through providing frequent patient assessment,timely interventions, and coordinated care that conse-quently contribute to improving patient outcomes.

Despite the advantages of stroke units, stroke patientscan be admitted to various units including ICUs andgeneral wards (e.g., medical, neurological, and neurosur-gical) because stroke units are not always feasible. In anational study conducted by the Health Insurance Review& Assessment Service (HIRA) in Korea, out of total acutestroke patients who visited an emergency room, 43% weredirectly admitted to ICUs, 46% to general wards, and 7% tooperating rooms, whereas only 2% were admitted to strokeunits (HIRA, 2007). Furthermore, patients’ transfers withinthe hospital commonly occur between care units, such asfrom ICUs to general wards and general wards to ICUs. Thevariation in admission units and possible transfers suggestthat nurse staffing of general wards as well as ICUs shouldbe examined even when targeting patients exclusivelywho were admitted to an ICU during hospitalization.Consideration of both ICU and general ward staffingseparately is expected to minimize one methodologicalproblem of ICU studies: that care that patients receivedwithin and outside the critical care setting was not fullydistinguished or accounted for in examining the impact ofICU nurse staffing on in-hospital mortality (Numata et al.,2006).

The impact of nurse staffing can be also examined byusing the process approach in Donabedian’s structure-process-outcome model (Donabedian, 2003). Within thevariety in process measures of nursing care, whethernurses provide inpatients with basic nursing care (e.g.,feeding and toileting) completely without delegating it to

families can be influenced by nurse staffing. Familyinvolvement in inpatient care has been a traditionalcustom in Korea and possibly other Asian countries. Tzeng(2004) reported a long-standing Taiwanese custom thatfamily members or personal attendants hired by familiesprovide in-hospital care for patients, and nurses usuallyexpect to delegate some of their physical nursing activities(e.g., morning care) to family members and personalattendants. Although viewed positively as opportunitiesfor patient–family adaptation, family involvement mayincrease families’ caregiving burdens and interfere withtotal nursing care and infection control by nursingpersonnel (Tzeng, 2004; Cho and Kim, 2006). A policyresponse of the Korean government to family involvementin inpatient care was a financial incentive in which theNational Health Insurance (NHI) pays more inpatientnursing fees to hospitals with better nurse staffing. The aimof this policy was to prevent hospitals from delegating theprovision of nursing care to patient families or paidcaregivers, and ultimately deteriorating the quality ofnursing care (Ministry of Health and Welfare, 2001). Thisfinancial incentive was implemented first to general wards(e.g., medical-surgical units) in November 1999, andexpanded to neonatal ICUs in October 2007 and to adultICUs in July 2008.

Based on the findings and methodological challenges instaffing-outcomes research, this study was conducted to(a) continue to investigate the relationship between nursestaffing and mortality to respond to inconsistent empiricalevidence; (b) evaluate both in-hospital and 30-daymortality because the literature has reported that manystroke deaths occur after discharge; (c) examine both ICUand general ward staffing separately to account forpossible patient transfers and care provided in generalwards; and (d) assess the provision of basic care in ICUs asthe process approach for evaluating ICU care quality inaddition to patient outcomes.

Fig. 1 illustrates the conceptual framework of this studythat was based on Donabedian’s structure-process-out-come model (Donabedian, 2003). Nurse staffing in ICUsand general wards (structure) was assumed to be related toprovision of basic care (process) and mortality (outcome).As represented by the dotted line, no direct linkage frombasic care to mortality was expected. The reason forexamining provision of basic care as a dependent variablewas because it was assumed to be sensitive to nursestaffing, rather than because basic care was an interveningvariable which would link nurse staffing to patientmortality. The conceptual framework also shows thatthe relationship between in-hospital and 30-day mortalitymay be influenced by discharge practices and lengths ofstay at hospitals (Rosenthal et al., 2000).

2. Methods

2.1. Data sources

The study was designed as a cross-sectional study thatused survey and administrative data. Information on thebed-to-nurse ratios of ICUs and their provision of basic careas of March 2003 was provided by an ICU survey that was

Fig. 1. Conceptual framework for nurse staffing, provision of basic care, and in-hospital and 30-day mortality.

S.-H. Cho, S.-C. Yun / International Journal of Nursing Studies 46 (2009) 1092–1101 1095

conducted by the HIRA in June 2003. Nurse staffing ofgeneral wards in the first quarter of 2003 was alsocollected from an administrative database of the HIRA.Patient demographic and clinical information wasobtained from medical claims data of the fourth quarterof 2002. Patients’ death dates required for calculating in-hospital and 30-day mortality were provided by the NHIenrollee database. Fuller description of these data sourcesis provided in previous work by Cho et al. (2008).

2.2. Sample

The study sample included acute stroke patients of age18 years or older who were admitted to an ICU duringhospitalization. Acute stoke was defined as either of twostroke subtypes – hemorrhagic or ischemic stroke – as theprimary diagnosis. Hemorrhagic stroke included subar-achnoid hemorrhage, which is coded as ‘I60’ in theInternational Classification of Diseases 10th Revision(ICD-10; World Health Organization, 2003), intracerebralhemorrhage (I61), and other nontraumatic intracranialhemorrhage (I62). Ischemic stroke indicated cerebralinfarction (I63). Patients having cancer in the secondarydiagnosis were excluded due to their higher risk of deaththan those with other secondary diagnoses. To obtainreliable estimates from the statistical analysis, hospitalsbelow the minimum patient volume per hospital, fewerthan 10 stroke patients, were then excluded. In the end, thestudy sample included 6957 patients from 185 hospitals(40 tertiary and 145 secondary). Tertiary hospitals refer tohospitals that provide the highest level of care in the

Table 1

Distribution of hospitals (N = 185) by bed-to-nurse ratiosa and staffing grades

ICU staffing General ward staffing

Grade 1 (<2.0) Grade 2 (<2.5) Grade 3 (<3.0)

Grade 1 (<0.50) 1 (0.5)

Grade 2 (<0.63) 1 (0.5) 2 (1.1)

Grade 3 (<0.77) 4 (2.2)

Grade 4 (<0.88)

Grade 5 (<1.00) 1 (0.5)

Grade 6 (<1.25)

Grade 7 (<1.50)

Grade 8 (<2.00)

Grade 9 (�2.00)

Total 2 (1.1) 7 (3.8)

Data are n (%) and empty cells indicate no case.a Indicates number of beds per registered nurse.

Korean healthcare system, and thus, official referrals madeby physicians in lower level hospitals are required forpatients to be admitted. Official physician referrals tosecondary hospitals, on the other hand, are not mandatory.

2.3. Measures

The bed-to-nurse ratio of ICUs was calculated bydividing the total number of ICU beds by the total numberof full-time equivalent registered nurses working in theICUs and thus, smaller numbers of beds per nurse indicatebetter staffing. Those bed-to-nurse ratios were thencategorized into nine staffing grades (Grades 1–9). Thereason for using staffing grades instead of the actual ratioswas to generate findings more appropriate for respondingto the primary concern of clinicians and policymakers—todetermine what staffing levels are adequate and safe, notjust whether there is a relationship between staffing andoutcomes (Clarke and Donaldson, 2008). Categorization ofthe staffing grades was based on the grading criteria of theNHI financial incentive that began to be implemented inadult ICUs in July 2008. As presented in Table 1, Grade 1(the best) is defined as when the number of beds per nurseis less than 0.50, and Grade 9 (the worst) is when the ratiois 2.00 or greater.

General ward staffing was also stratified into six grades,from Grade 1 (less than 2.0 beds per nurse) to Grade 6 (4.0or more beds per nurse) based on the criteria of the NHIfinancial incentive for general wards that had beenimplemented since November 1999 and also remainedeffective in 2002–2003. In the incentive system, there was

of ICUs and general wards.

Grade 4 (<3.5) Grade 5 (<4.0) Grade 6 (�4.0) Total

1 (0.5)

2 (1.1) 1 (0.5) 6 (3.2)

7 (3.8) 4 (2.2) 3 (1.6) 18 (9.7)

4 (2.2) 9 (4.9) 9 (4.9) 22 (11.9)

6 (3.2) 13 (7.0) 6 (3.2) 26 (14.1)

1 (0.5) 9 (4.9) 29 (15.7) 39 (21.1)

2 (1.1) 6 (3.2) 30 (16.2) 38 (20.5)

2 (1.1) 27 (14.6) 29 (15.7)

6 (3.2) 6 (3.2)

20 (10.8) 45 (24.3) 111 (60.0) 185 (100.0)

S.-H. Cho, S.-C. Yun / International Journal of Nursing Studies 46 (2009) 1092–11011096

one grade difference between tertiary and secondaryhospitals in the grading criteria. For example, the criterionfor Grade 1 in secondary hospitals is ‘less than 2.5 beds pernurse,’ which becomes the criterion for Grade 2 in tertiaryhospitals. In this study, however, the grading criteria wereunified into those for tertiary hospitals as shown in Table 1.

Basic nursing care included five activities: bathing,position change, urinary and bowel elimination, feedingassistance, and endotracheal suction. In the ICU survey, ICUmanagers were asked whether those activities wereprovided by nursing personnel or delegated to patientfamilies when they visited the patient in ICUs. Thisquestion was intended to assess the overall practice onprovision of basic care as managerial policy made at theICU or hospital level, rather than to examine basic careprovided to specific individual patients. ‘Full’ provision ofbasic care was defined as when all five activities wereprovided by ICU nursing staff and ‘partial’ as when at leastone activity was given by the patient’s family. Becausewhether delegation to family members had occurred ornot, rather than the extent of delegation (i.e., number ofactivities delegated), was the primary interest, provision ofbasic care was treated as a dichotomous (full vs. partial)variable.

In-hospital mortality was defined as death thatoccurred in the hospital or on the date of hospitaldischarge. The 30-day mortality was defined as deathwithin 30 days of hospital admission regardless of beinghospitalized or discharged. Patient characteristics used forrisk adjustment were age, gender, type of medical security(NHI or Medical Aid), and the primary and secondarydiagnoses. The secondary diagnoses that reflected patients’comorbid conditions were grouped into 19 diseasecategories based on ICD-10 codes.

2.4. Data analysis

Descriptive analyses were conducted on the distribu-tion of staffing grades, provision of basic care, andmortality rates by patient characteristics. To examinedifferences in mortality by nurse staffing, z-scores of in-hospital and 30-day mortality in each hospital werecomputed. These z-scores were derived from the standar-dized difference between observed and expected mortality(Normand et al., 1997) and thus, a positive value of a z-score means that the observed mortality of the hospital isgreater than the expected mortality. While acknowledgingdisadvantages in using observed-expected comparisons ofpatient outcomes (Shwartz et al., 1997), z-scores were usedbecause they were suitable to demonstrate the relation-ship between in-hospital and 30-day mortality at thehospital level, where hospital discharge practices takeplace.

The expected mortality of each hospital was obtainedthrough two steps. First, risk-adjusted probabilities of in-hospital and 30-day mortality in each patient wereestimated while controlling for patient characteristics(i.e., age, gender, type of medical security, primarydiagnosis, and secondary diagnosis). For this patient-levelmodeling, multilevel logistic regression analysis with theSAS GLIMMIX procedure (SAS Institute, 2006) was used.

This multilevel analysis assumed that patients admitted tothe same hospitals were clustered at their hospitals andwould receive a level of quality of care provided in thehospital. The second step was to compute the expectedmortality of each hospital by averaging the model-basedprobabilities of mortality for all patients within the samehospital.

The z-scores of in-hospital and 30-day mortality werethen grouped into three groups: low (z < �0.675), middle(�0.675 � z < 0.675), and high (z � 0.675). The cutoffpoints of ‘�0.675’ and ‘0.675’ represent the lowest 25%and the highest 25%, respectively, in the standardizednormal distribution curve. The differences in the distribu-tion of provision of basic care and mortality groups bystaffing grades were examined using chi-square tests.Finally, the relationships of nurse staffing to provision ofbasic care and mortality were assessed using logisticregression analyses. To contrast nurse staffing betweenlow and high mortality groups, the middle groups of in-hospital and 30-day mortality were excluded from theregression model.

3. Results

Table 1 presents the staffing grades of ICUs and generalwards. As for the ICU staffing, the greatest proportion of thetotal hospitals fell into Grade 6 (21.1%) and Grade 7(20.5%), whereas only one hospital met the highest staffingof Grade 1. About a fifth had the worst ICU staffing ofGrades 8 and 9. General ward staffing had a more skeweddistribution than ICU staffing did. There was no hospital inGrade 1 and the majority of hospitals fell into Grade 5(24.3%) and Grade 6 (60.0%). Some hospitals with a higherICU grade tended to also have a higher grade in generalwards. Only three hospitals had a worse ICU grade (Grades6–9) but a better general ward grade (Grades 1–4).However, the relatedness between ICU and general wardstaffing was not obvious because hospitals classified inGrades 5 and 6 in general wards had almost a full range ofICU grades from Grades 2–9.

In 83.8% of hospitals, all five activities of basic care wereprovided by ICU nursing staff (Table 2). In the rest of thehospitals, patient families were involved in at least one ofthe basic care activities when they visited their patient inICUs. Patient deaths were distinguished into three groupsas shown in Table 2: (A) in-hospital death within 30 days ofadmission; (B) in-hospital death after 30 days of admis-sion; and (C) death after discharge but within 30 days ofadmission. The overall in-hospital mortality rate was 21.9%(A + B) and the 30-day mortality rate was 25.4% (A + C). Themedians of total hospital stay and ICU stay were 18 daysand 6 days, respectively. These lengths of stay indicate thaton average, patients were cared for in the ICU for one thirdof their total hospital stay.

Distributions of patients and their mortality rates bycharacteristic are described in Table 3. The mean age was63 years and half of patients were 65 years or older. Fifty-two percent of the patients were female and 90% were NHIpatients. Regarding stroke subtypes as the primarydiagnosis, hemorrhagic strokes were a greater proportion(63.4%) than ischemic strokes. Within the hemorrhagic

Table 2

Descriptive analyses of provision of basic nursing carea, in-hospital and

30-day mortality, and length of stay.

Hospitals

(N = 185)

Patients

(N = 6957)

Provision of basic nursing

care, n (%)

Full 155 (83.8)

Partial 30 (16.2)

Death, n (%)

In-hospital death within

30 days of admission (A)

1337 (19.2)

In-hospital death after

30 days of admission (B)

189 (2.7)

Death after discharge within

30 days of admission (C)

429 (6.2)

Alive in hospital and within

30 days of admission

5002 (71.9)

Mortality rate, n (%)

In-hospital mortality (A + B) 1526 (21.9)

30-Day mortality (A + C) 1766 (25.4)

Length of stay, median

(mean � S.D.)

Total hospital days 18 (31.2 � 41.3)

ICU days 6 (10.9 � 15.6)

a Includes bathing, position change, urinary and bowel elimination,

feeding assistance, and endotracheal suction.

S.-H. Cho, S.-C. Yun / International Journal of Nursing Studies 46 (2009) 1092–1101 1097

strokes, the highest proportion was intracerebral hemor-rhage, followed by subarachnoid hemorrhage. In thesecondary diagnoses as the comorbid condition, 53% werediseases of the circulatory system, and 20% were diseases

Table 3

Descriptive analyses of the distribution of patients (N = 6957) and in-hospital a

Patients, n (%)

Age (years)

18–39 353 (5.1)

40–64 3129 (45.0)

65+ 3475 (49.9)

Gender

Male 3372 (48.5)

Female 3585 (51.5)

Type of medical security

National Health Insurance 6265 (90.1)

Medical Aid 692 (9.9)

Primary diagnosis (stroke subtype)

Hemorrhagic stroke 4409 (63.4)

Subarachnoid hemorrhage 1526 (21.9)

Intracerebral hemorrhage 2557 (36.8)

Other nontraumatic intracranial hemorrhage 326 (4.7)

Ischemic stroke

Cerebral infarction 2548 (36.6)

Secondary diagnosis

Cerebrovascular disease 1840 (26.4)

Hypertensive disease 1451 (20.9)

Ischemic heart disease 207 (3.0)

Other disease of circulatory system 220 (3.2)

Disease of nervous system 1389 (20.0)

Disease of respiratory system 514 (7.4)

Endocrine, nutritional and metabolic disease 313 (4.5)

Others (12 categories) 1023 (14.7)

of the nervous system. Older patients aged 65 years orolder had higher in-hospital and 30-day mortality ratesand a 5.5% greater difference between in-hospital and 30-day mortality rates as compared to the younger age groups.NHI patients had higher mortality rates than those withMedical Aid. Hemorrhagic strokes had higher in-hospitaland 30-day mortality rates than ischemic strokes. Intra-cerebral hemorrhage had the highest rates of in-hospital(25.2%) and 30-day (27.9%) mortality.

Given those differences in mortality rates by patientcharacteristics, the expected mortality at each hospital wasestimated using two multilevel logistic regression modelsfor each in-hospital and 30-day mortality measure thatcontrolled for those patient characteristics. All five patientcharacteristics were significantly associated with both in-hospital and 30-day mortality (results not shown). Good-ness of fit of the two models was evaluated usinggeneralized chi-square statistics and found to be adequate;the fit statistics of generalized chi-square/degree offreedom in the in-hospital and 30-day mortality modelswere 0.96 and 0.97, respectively.

z-Scores of in-hospital and 30-day mortality, whichwere the standardized difference between observed andexpected mortality, were then computed and categorizedinto three groups: low, middle, and high. The proportionsof the three mortality groups are presented in Table 4.While the proportions of 25% in the low, 50% in the middle,and 25% in the high group would be expected according tothe normal distribution, the low and high in-hospitalmortality groups accounted for 31% and 32%, respectively.

nd 30-day mortality rates by patient characteristics.

In-hospital mortality (%) 30-Day mortality (%)

16.7 18.1

19.4 20.8

24.7 30.2

21.8 24.7

22.1 26.1

22.3 25.9

18.6 21.1

23.5 26.4

21.8 24.8

25.2 27.9

18.7 22.7

19.2 23.5

27.9 30.9

14.7 18.1

28.5 32.4

28.2 31.4

15.3 18.7

27.2 32.7

20.4 25.9

25.5 28.3

Table 4

Distribution of hospitals (N = 185) by low, middle, and high mortality groups based on z-scoresa of in-hospital and 30-day mortality.

In-hospital mortality 30-Day mortality Total

Low (z < �0.675) Middle (�0.675 � z < 0.675) High (z � 0.675)

Low (z < �0.675) 41 (22.2) 17 (9.2) 0 (0.0) 58 (31.4)

Middle (�0.675 � z <0.675) 12 (6.5) 43 (23.2) 12 (6.5) 67 (36.2)

High (z � 0.675) 0 (0.0) 20 (10.8) 40 (21.6) 60 (32.4)

Total 53 (28.6) 80 (43.2) 52 (28.1) 185 (100.0)

Data are n (%).a Indicates the standardized difference between observed and expected mortality. The expected mortality of each hospital was estimated using

multilevel logistic regression analysis with the SAS GLIMMIX procedure, while controlling for patient characteristics (i.e., age, gender, type of medical

security, primary diagnosis, and secondary diagnosis).

S.-H. Cho, S.-C. Yun / International Journal of Nursing Studies 46 (2009) 1092–11011098

The 30-day mortality groups were distributed more closelyto the normal distribution. The low and high groups for 30-day mortality occupied 29 and 28%, respectively. Aninteresting finding was that the three groups for in-hospital and 30-day mortality did not always coincide. Forexample, 60 hospitals were located in the high in-hospitalmortality group but one third of those 60 hospitals were inthe middle group for 30-day mortality. However, themajority of hospitals were categorized into the samegroups for in-hospital and 30-day mortality (i.e., low–low,middle–middle, and high–high).

Table 5 presents the comparisons of provision of basiccare and in-hospital and 30-day mortality groups by ICU andgeneral ward staffing. Due to the small numbers of hospitalswith high staffing grades, the ICU and general ward gradeswere collapsed into four and three groups, respectively,which made the statistical comparisons simple and stable.The proportions of hospitals that fully provided basic carediffered by ICU staffing (P = 0.038). Ninety-two percent ofhospitals with Grades 1–3 provided basic care completely,whereas 69% of those with Grades 8 and 9 did. Thedistribution of in-hospital mortality groups differed statis-tically by ICU staffing (P = 0.039) without difference bygeneral ward staffing. However, hospitals in ICU staffingGrades 4 and 5, not those in Grades 1–3, had the greatestproportion (44%) of the low in-hospital mortality group andthe lowest percentage (17%) of the high group. Theproportion of 30-day mortality groups were distributeddifferently by both ICU (P = 0.045) and general ward

Table 5

Differences in provision of basic care and in-hospital and 30-day mortality gro

Basic nursing care In-hospital mor

Full Partial Low

ICU staffing

Grades 1–3 (<0.77) 23 (92.0) 2 (8.0) 9 (36.0)

Grades 4 and 5 (<1.00) 43 (89.6) 5 (10.4) 21 (43.8)

Grades 6 and 7 (<1.50) 65 (84.4) 12 (15.6) 17 (22.1)

Grades 8 and 9 (�1.50) 24 (68.6) 11 (31.4) 11 (31.4)

P 0.038

General ward staffing

Grades 1–4 (<3.5) 10 (34.5)

Grade 5 (<4.0) 20 (44.4)

Grade 6 (�4.0) 28 (25.2)

P

Data are n (%).

(P = 0.005) staffing. As ICU staffing became worse, theproportion of the low 30-day mortality group decreasedstepwise, with that of the high group increasing. Similarly,as general ward staffing became worse, the proportion of thehigh 30-day mortality group also increased stepwise.However, hospitals with Grade 5 had a greater percentage(44%) of the low group than those of Grades 1–4.

Finally, the relationships of nurse staffing to provisionof basic care and mortality were further examined usinglogistic regression analyses. Table 6 shows the odds ratios(ORs) from the analyses. Compared with hospitals inGrades 8 and 9, those in Grades 1–3 (OR, 5.27; 95% CI, 1.05–26.4) and Grades 4 and 5 (OR, 3.94; 95% CI, 1.22–12.7) weremore likely to provide basic care fully to their patients (P

for trend = 0.007). Hospitals in Grades 4 and 5 for ICUstaffing (OR, 0.26; 95% CI, 0.09–0.80) and Grade 5 forgeneral ward staffing (OR, 0.33; 95% CI, 0.13–0.80) hadstatistically less likelihood of having high in-hospitalmortality than the reference groups. As for 30-daymortality, better ICU and general ward staffing gradeswere associated with a lower probability of being classifiedinto the high mortality group. Compared with the ICUstaffing of Grades 8 and 9, being in Grades 1–3 (OR, 0.13;95% CI, 0.03–0.61) or Grades 4 and 5 (OR, 0.23; 95% CI,0.07–0.78) was related to decreases in the odds of havinghigh 30-day mortality. Hospitals with Grades 1–4 or Grade5 for general ward staffing also had a decreasing tendencyto have high 30-day mortality than those with Grade 6 (P

for trend = 0.001).

ups by ICU and general ward staffing of hospitals (N = 185).

tality 30-Day mortality

Middle High Low Middle High

9 (36.0) 7 (28.0) 11 (44.0) 11 (44.0) 3 (12.0)

19 (39.6) 8 (16.7) 18 (37.5) 21 (43.8) 9 (18.8)

31 (40.3) 29 (37.7) 17 (22.1) 35 (45.5) 25 (32.5)

8 (22.9) 16 (45.7) 7 (20.0) 13 (37.1) 15 (42.9)

0.039 0.045

12 (41.4) 7 (24.1) 10 (34.5) 16 (55.2) 3 (10.3)

15 (33.3) 10 (22.2) 20 (44.4) 16 (35.6) 9 (20.0)

40 (36.0) 43 (38.7) 23 (20.7) 48 (43.2) 40 (36.0)

0.107 0.005

Table 6

Odds ratios (ORs) of ICU and general ward staffing of hospitalsa in provision of basic care and in-hospital and 30-day mortality groups.

Basic nursing care (N = 185)

(event = full vs. partial provision)

In-hospital mortality (n = 118)

(event = high vs. low group)

30-Day mortality (n = 105)

(event = high vs. low group)

OR (95% CI) P OR (95% CI) P OR (95% CI) P

ICU staffing

Grades 1–3 (<0.77) 5.27 (1.05–26.4) 0.043 0.53 (0.15–1.87) 0.327 0.13 (0.03–0.61) 0.010

Grades 4 and 5 (<1.00) 3.94 (1.22–12.7) 0.022 0.26 (0.09–0.80) 0.019 0.23 (0.07–0.78) 0.018

Grades 6 and 7 (<1.50) 2.48 (0.97–6.37) 0.059 1.17 (0.44–3.10) 0.748 0.69 (0.23–2.04) 0.498

Grades 8 and 9 (�1.50) 1.0 1.0 1.0

P for trend 0.007 0.030 0.001

General ward staffing

Grades 1–4 (< 3.5) 0.46 (0.16–1.34) 0.153 0.17 (0.04–0.69) 0.013

Grade 5 (<4.0) 0.33 (0.13–0.80) 0.014 0.26 (0.10–0.66) 0.005

Grade 6 (�4.0) 1.0 1.0

P for trend 0.031 0.001

a Hospitals in the middle in-hospital (n = 67) and 30-day (n = 80) mortality groups were excluded from the regression analysis to contrast high versus low

mortality groups.

S.-H. Cho, S.-C. Yun / International Journal of Nursing Studies 46 (2009) 1092–1101 1099

4. Discussion

This study poses at least four limitations in theinterpretation and generalization of the study findings.First, nurse staffing of ICUs and general wards wasaggregated at the hospital level and thus, the averagedstaffing grades used in this study would not be equal to thereal staffing of the units where individual patients werecared for. Second, risk adjustment undertaken in thecalculation of z-scores might not be complete, becauselimited patient characteristics were controlled for due tolack of clinical information in the administrative databasesused in this study. However, the homogenous sample of thisstudy, including only patients who were admitted to an ICU,would reduce much variation in severity of illness amongstroke patients, as compared to including all stroke patients.Third, differences in the criteria for ICU admission mightexist across hospitals and influence inclusion and exclusionof patients in the study sample. Lack of controls for hospitalcharacteristics would also pose selection and referral bias,although the study sample was designed to include hospitalsas homogenous as possible by targeting only secondary andtertiary hospitals meeting the minimum patient volume.

A fourth limitation is that including ICU patientsexclusively, on the other hand, has led to a differentdistribution of stroke subtypes and their mortality rates inboth international and national statistics. In this study,hemorrhagic stroke was more common (63.4%) thanischemic stroke, whereas ischemic stroke is the majority(65–80%) in Korea and other countries (Feigin et al., 2003;Kitamura et al., 2006; HIRA, 2007). The overall in-hospitalmortality rate of 21.9% (23.5% of which was hemorrhagicand 19.2% ischemic) was also higher than that of a nationalstudy that reported 11.1% in-hospital mortality (22.2% ofwhich was hemorrhagic and 5.2% ischemic) (HIRA, 2007).The greater proportion of hemorrhagic strokes that hadhigher mortality is assumed to raise the overall mortalityrate in this study. The higher mortality rates, especially forischemic stroke patients, would also be attributed to thehigher level of severity of illness among ICU patients.

We evaluated the provision of basic nursing care as theprocess indicator of ICU care quality. Five basic careactivities were not fully provided by ICU nursing staff in16.2% of hospitals. This finding reflects that familycaregiving occurred even in ICUs and is still valid as asocial norm in Korean society. Partial provision of basiccare may not be understood by international nurses whobelieve total holistic care should be provided by nursingstaff to patients, more particularly to ICU patients. Theassociation of better staffing with a greater tendency tofully provide basic care found in this study, however,suggests that involuntary involvement of patient familiesin basic care can be reduced by improving ICU staffing. Thispositive relationship also supports the NHI financialincentive that aims to prevent hospitals from delegatingnursing care to patient families, through paying morenursing fees to better-staffed hospitals. Not relying onfamily caregiving is also expected to improve nursingmorale and the care standard because family involvementmay make basic care ‘‘missed nursing care,’’ and leadnurses to become accustomed to omit basic care andbelieve ‘‘it’s not my job’’ (Kalisch, 2006).

A unique approach of this study was to examine 30-daymortality as well as in-hospital mortality. As deaths ofstroke patients are known to often occur after discharge, aquarter of the total 30-day mortality occurred afterdischarge in this study. This is similar to the finding thatin-hospital deaths within 30 days of admission repre-sented about 80% of all 30-day deaths (Mattke et al., 2006).Regarding the relationship between in-hospital and 30-day mortality, the overall 30-day mortality was 3.5%higher than in-hospital mortality, and had a distributioncloser to the normal distribution than in-hospital mortalitydid. In addition, in-hospital and 30-day mortality of eachhospital were not always classified into the same group ofthe three mortality groups (i.e., low, middle, and high).Countries and hospitals that have shorter lengths ofhospital stay are expected to have a greater proportionof deaths after discharge and consequently, a greaterdifference between in-hospital and 30-day mortality.

S.-H. Cho, S.-C. Yun / International Journal of Nursing Studies 46 (2009) 1092–11011100

Although both in-hospital and 30-day mortality dif-fered by staffing grades, a more linear increase was foundin 30-day mortality as staffing grades became worse. Thedistinct relationship of 30-day mortality to nurse staffingmay suggest that 30-day mortality is more sensitive tonursing care quality and nurse staffing than in-hospitalmortality is. It also raises the issue of whether in-hospitalmortality includes the time window appropriate forevaluating the true effect of nurse staffing on patientmortality thoroughly, at least of patients with acute stroke.In-depth understanding of patients’ specific conditions,their resource utilization (e.g., length of stay), and themechanism of nursing impact on patient outcomes shouldprecede developing and selecting nursing sensitive qualityindicators.

Finally, both staffing grades of ICUs and general wardswere examined independently in this study. The finding thatICU patients stayed at general wards for two thirds of theirtotal hospital stay supports the need for considering bothstaffing levels of ICUs and general wards. Although in-hospital mortality did not differ by general ward staffing inthe chi-square test, favourable general ward staffing had atendency to lower in-hospital and 30-day mortality, which issimilar tothe association between ICU staffing and mortality.Quantifying and comparing the separatecontributions of ICUand general ward staffing to mortality would be the nextquestion in future research that includes more detailedinformation on care units and patient care. Future researchthat uses a more advanced study design and analyticalapproach is also needed to examine the independent andinterdependent or integrated impact of nurse staffing acrossthe care units within the institution on patient outcomes.

5. Conclusions

Better nurse staffing was associated with fuller provi-sion of basic nursing care and lower mortality among ICUpatients with stroke. This study provides empiricalevidence that nurse staffing may impact patient outcomes.The findings also suggest that organizational and nationalpolicies are needed to improve nurse staffing particularlyin countries where the contribution of nursing care topatient outcomes is undervalued. Also, more thorough andvarious explorations of patients, their diseases, andnursing interventions are essential to determine the effectsof nursing care on patient outcomes.

Conflict of interest. None.

Funding. This work was supported by the Korea Research

Foundation Grant funded by the Korean Government

(MOEHRD) (KRF-2006-331-E00406).

Ethical approval. The data used in this study were collected by

the Health Insurance Review & Assessment Service (HIRA).

Under the HIRA, the Central Evaluation Committee, which is

composed of representatives of consumers, physicians and

hospitals, and the government, decided to conduct this pro-

ject and the Minister of Health and Welfare approved it. The

first author was able to access the databases as a guest

researcher of the HIRA.

References

Agency for Healthcare Research and Quality, 2007. Guide to InpatientQuality Indicators: Quality of Care in Hospitals—Volume, Mortality,and Utilization (Version 3.1). [Online]. Agency for HealthcareResearch and Quality. Available at: http://www.qualityindicator-s.ahrq.gov/downloads/iqi/iqi_guide_v31.pdf. [accessed 5 June 2007].

Agency for Healthcare Research and Quality, 2008. Inpatient QualityIndicators: Technical Specifications (Version 3.2). [Online]. Agencyfor Healthcare Research and Quality. Available at: http://www.qua-lityindicators.ahrq.gov/downloads/iqi/iqi_technical_specs_v32a.pdf.[accessed 1 July 2008].

Armesto, S.G., Lapetra, M.L.G., Wei, L., Kelley, E., the Members of the HCQIExpert Group, 2007. Health Care Quality Indicators Project 2006 DataCollection Update Report. [Online]. Organisation for Economic Co-operation and Development. Available at: http://www.oecd.org/dataoecd/57/22/39447928.pdf. [accessed 11 August 2008].

Burton, C.R., Fisher, A., Green, T.L., 2009. The organisational context ofnursing care in stroke units: a case study approach. InternationalJournal of Nursing Studies 46, 86–95.

Candelise, L., Gattinoni, M., Bersano, A., Micieli, G., Sterzi, R., Morabito, A.,on the behalf of the PROSIT Study Group, 2007. Stroke-unit care foracute stroke patients: an observational follow-up study. Lancet 369,299–305.

Carmel, S., Rowan, K., 2001. Variation in intensive care unit outcomes: asearch for the evidence on organizational factors. Current Opinion inCritical Care 7, 284–296.

Cho, S.H., Hwang, J.H., Kim, J., 2008. Nurse staffing and patient mortality inintensive care units. Nursing Research 57, 322–330.

Cho, S.H., Kim, H.R., 2006. Family and paid caregivers of hospitalizedpatients in Korea. Journal of Clinical Nursing 15, 946–953.

Clarke, S.P., 2007. Registered nurse staffing and patient outcomes in acutecare: looking back, pushing forward. Medical Care 45, 1126–1128.

Clarke, S.P., Donaldson, N.E., 2008. Nurse staffing and patient care qualityand safety. In: Hughes, R.G. (Ed.), Patient Safety and Quality: AnEvidence-Based Handbook for Nurses. Agency for HealthcareResearch and Quality, Rockville, pp. 1–25.

Donabedian, A., 2003. An Introduction to Quality Assurance in HealthCare. Oxford University Press, New York.

Duncan, P.W., Jorgensen, H.S., Wade, D.T., 2000. Outcome measures inacute stroke trials: a systematic review and some recommendationsto improve practice. Stroke 31, 1429–1438.

Evans, A., Perez, I., Harraf, F., Melbourn, A., Steadman, J., Donaldson, N.,Kalra, L., 2001. Can differences in management processes explaindifferent outcomes between stroke unit and stroke-team care?Lancet 358, 1586–1592.

Feigin, V.L., Lawes, C.M.M., Bennett, D.A., Anderson, C.S., 2003. Strokeepidemiology: a review of population-based studies of incidence,prevalence, and case-fatality in the late 20th century. Lancet Neurol-ogy 2, 43–53.

Health Insurance Review & Assessment Service, 2007. Evaluation ofAppropriateness of Stroke Care. Health Insurance Review & Assess-ment Service, Seoul.

Jones, S.P., Leathley, M.J., McAdam, J.J., Watkins, C.L., 2007. Physiologicalmonitoring in acute stroke: a literature review. Journal of AdvancedNursing 60, 577–594.

Kalisch, B.J., 2006. Missed nursing care: a qualitative study. Journal ofNursing Care Quality 21, 306–313.

Kalra, L., Evans, A., Perez, I., Knapp, M., Donaldson, N., Swift, C.G., 2000.Alternative strategies for stroke care: a prospective randomizedcontrolled trial. Lancet 356, 894–899.

Kane, R.L., Shamliyan, T.A., Mueller, C., Duval, S., Wilt, T.J., 2007. Theassociation of registered nurse staffing levels and patient outcomes:systematic review and meta-analysis. Medical Care 45, 1195–1204.

Kitamura, A., Nakagawa, Y., Sato, M., Iso, H., Sato, S., Imano, H., Kiyama, M.,Okada, T., Okada, H., Iida, M., Shimamoto, T., 2006. Proportions ofstroke subtypes among men and women�40 years of age in an urbanJapanese city in 1992, 1997, and 2002. Stroke 37, 1374–1378.

Korea National Statistical Office, 2007. 2006 Annual Report on the Causeof Death Statistics. Korea National Statistical Office, Daejeon.

Lang, T.A., Hodge, M., Olson, V., Romano, P.S., Kravitz, R.L., 2004. Nurse-patient ratios: a systematic review on the effects of nurse staffing onpatient, nurse employee, and hospital outcomes. Journal of NursingAdministration 34, 326–337.

Langhorne, P., Pollock, A., 2002. What are the components of effectivestroke unit care? Age and Ageing 31, 365–371.

Lankshear, A.J., Sheldon, T.A., Maynard, A., 2005. Nurse staffing andhealthcare outcomes: a systematic review of the internationalresearch evidence. Advances in Nursing Science 28, 163–174.

S.-H. Cho, S.-C. Yun / International Journal of Nursing Studies 46 (2009) 1092–1101 1101

Lopez, A.D., Mathers, C.D., Ezzati, M., Jamison, D.T., Murray, C.J.L., 2006.Global and regional burden of disease and risk factors, 2001: sys-tematic analysis of population health data. Lancet 367, 1747–1757.

Mark, B.A., 2006. Methodological issues in nurse staffing research. Wes-tern Journal of Nursing Research 28, 694–709.

Mattke, S., Kelley, E., Scherer, P., Hurst, J., Lapetra, M.L.G., the HCQI ExpertGroup Members, 2006. Health Care Quality Indicators Project InitialIndicators Report. [Online]. Organisation for Economic Co-operationand Development. Available at: http://www.oecd.org/dataoecd/1/34/36262514.pdf. [accessed 26 April 2007].

Ministry of Health and Welfare, 2001. Ministry of Health and WelfareNotification No. 2001-43. Gazette, 8 August, 8 pp.

Normand, S.L.T., Glickman, M.E., Gatsonis, C.A., 1997. Statistical methodsfor profiling providers of medical care: issues and applications.Journal of the American Statistical Association 92, 803–814.

Numata, Y., Schulzer, M., van der Wal, R., Globerman, J., Semeniuk, P.,Balka, E., FitzGerald, J.M., 2006. Nurse staffing levels and hospitalmortality in critical care settings: literature review and meta-analy-sis. Journal of Advanced Nursing 55, 435–448.

Rosenthal, G.E., Baker, D.W., Norris, D.G., Way, L.E., Harper, D.L., Snow, R.J.,2000. Relationships between in-hospital and 30-day standardizedhospital mortality: implications for profiling hospitals. Health Ser-vices Research 34, 1449–1468.

Sarti, C., Rastenyte, D., Cepaitis, Z., Tuomilehto, J., 2000. Internationaltrends in mortality from stroke, 1968 to 1994. Stroke 31, 1588–1601.

SAS Institute, 2006. The GLIMMIX Procedure, June 2006. [Online]. SASInstitute. Available at: http://support.sas.com/rnd/app/papers/glim-mix.pdf. [accessed 21 February 2007].

Scottish Intercollegiate Guidelines Network, 2002. Management ofPatients with Stroke. [Online]. Scottish Intercollegiate GuidelinesNetwork. Available at: http://www.sign.ac.uk/pdf/sign64.pdf.[accessed 5 June 2007].

Shwartz, M., Ash, A.S., Iezzoni, L.I., 1997. Comparing outcomes acrossproviders. In: Iezzoni, L.I. (Ed.), Risk Adjustment for MeasuringHealthcare Outcomes. 2nd ed. Health Administration Press,Chicago, pp. 471–516.

Stroke Unit Trialists’ Collaboration, 1997. How do stroke units improvepatient outcomes? A collaborative systematic review of the rando-mized trials. Stroke 28, 2139–2144.

Strong, K., Mathers, C., Bonita, R., 2007. Preventing stroke: saving livesaround the world. Lancet Neurology 6, 182–187.

Tzeng, H.M., 2004. Roles of nurse aides and family members inacute patient care in Taiwan. Journal of Nursing Care Quality 19,169–175.

West, E., Mays, N., Rafferty, A.M., Rowan, K., Sanderson, C., 2009. Nursingresources and patient outcomes in intensive care: a systematic reviewof the literature. International Journal of Nursing Studies 46, 993–1011.

World Health Organization, 2003. International Classification of Dis-eases 10th Revision (Version for 2003). [Online]. World HealthOrganization. Available at: http://www.who.int/classifications/apps/icd/icd10online2003/fr-icd.htm. [accessed 3 August 2006].

World Health Organization, 2008. World Health Statistics 2008. [Online].World Health Organization. Available at: http://www.who.int/who-sis/whostat/EN_WHS08_Full.pdf. [accessed 1 July 2008].