Balancing the NHS balanced scorecard!

Transcription

1 European Journal of Operational Research 185 (2008) Balancing the NHS balance scorecar! Brijesh Patel *, Thierry Chaussalet, Peter Millar Health an Social Care Moelling Group, University of Westminster, NG103 Copelan Builing, 115 New Cavenish Street, Lonon W1W 6UW, Unite Kingom Receive 1 April 2005; accepte 1 February 2006 Available online 13 October 2006 Abstract In the UK, the split between opposition an supporters views of the National Health Service (NHS) performance ratings system is growing. Objective argument an consensus woul be facilitate if a methoology was evelope which showe the cause an effect relationships between the components of the performance rating system. The NHS hospital trust performance ratings ata use in 2002 an 2003 were ownloae from the Department of Health performance rating website. Structural equation moelling was use to construct a causal-loop iagram showing the cause an effect relationships between the 16 common performance inicators in the two years. Scenario testing suggests that inicators of elaye transfer of care an of ata quality are compromise if emergency reamissions performance is improve. Ó 2006 Elsevier B.V. All rights reserve. Keywors: Health services; Balance scorecar; System thinking; Structural equation moelling; Performance management 1. Introuction The ageing populations an increase cost of meical care in most evelope countries have mae healthcare a very emaning task for governments worlwie. Hospital staff an managers are uner pressure an concerne for effective use of scarce resources an sustainable performance. To that effect, in the UK, the NHS has introuce a balance scorecar [1] (BSC) as a part of the framework for hospital trusts performance ratings. The purpose of the performance ratings framework is ebatable but is assume to contribute towars performance management in the NHS. * Corresponing author. aress: B.Patel41@wmin.ac.uk (B. Patel). Many more organisations, both public an private, have been focusing on long-term performance an the use of BSC as part of a performance management framework [2 5]. The BSC is a balance representation of performance of internal as well as external objectives. A typical BSC has financial, customer, internal processes, an learning an growth imensions. These imensions are interrelate with cause-an-effect relationships in a treelike fashion [6]. Cause-an-effect relationships amongst these imensions imply how fruits (financial measures) in the tree are relate to leaves (customers), trunk (processes) an roots (learning an growth). The NHS has been following a balance scorecar approach since The framework is rapily evolving [7 9]. Fig. 1 inicates that, prior to 2001, the performance ratings were base on a set of /$ - see front matter Ó 2006 Elsevier B.V. All rights reserve. oi: /j.ejor

2 906 B. Patel et al. / European Journal of Operational Research 185 (2008) Key Targets After 2001 Key Targets BSC CHI Review Before 2001 CHI Review Fig. 1. Change in performance rating framework for NHS hospitals in UK. key targets an reviews by the Commission for Health Improvement (CHI). Now, NHS hospital trusts are rate using BSC performance inicators in aition to the previous criteria. The current NHS balance scorecar is compose of three types of performance inicators: patient focus, clinical focus, an capability an capacity focus. More performance inicators (17) have been ae to the BSC since the first year of publishing performance ratings. Table 1 shows the performance inicators inclue in the NHS BSC for the years 2001/2002 an 2002/2003. Due to the complexity of the framework, hospital trusts fin it ifficult to interpret their performance ratings results to formulate an efficient strategy for performance improvement. The lack of literature (government or acaemic publications) on causal Table 1 Composition of the NHS balance scorecar for year 2001/2002 an 2002/2003 Inicators 2001/ /2003 Patient focus A&E emergency amission waits (4 hours) Better hospital foo Breast cancer treatment within a month Cancelle operations Day case booking Delaye transfers of care Nine month heart operation waits Outpatient A&E survey access & waiting Outpatient A&E survey better information, more choice Outpatient A&E survey builing relationships Outpatient A&E survey clean, comfortable, frienly place to be Outpatient A&E survey safe, high quality, co-orinate care Paeiatric outpatient i not atten rates Patient complaints proceure Privacy & ignity a Six month inpatient waits Thirteen week outpatient waits Total inpatient waits - % of plan Waiting times for Rapi Access Chest Pain Clinic Clinical Focus Clinical Negligence Deaths within 30 ays of a heart bypass operation a Deaths within 30 ays of selecte surgical proceures a reamission to hospital following ischarge reamission to hospital following ischarge for chilren reamission to hospital following treatment for a fracture hip reamission to hospital following treatment for a stroke Returning home following hospital treatment for fracture hip Returning home following hospital treatment for stroke Infection control proceures Methicillin Resistant Staphylococcus Aureus (MRSA) bacteraemia improvement score Thrombolysis treatment time Capacity an Capability Consultant appraisal Focus Data quality Fire, Health & Safety Information Governance Junior octors hours Sickness absence rate Staff opinion survey a Variables relate to special practices or high proportion of missing information.

3 B. Patel et al. / European Journal of Operational Research 185 (2008) effect relationships between ifferent types of inicators poses ifficulty in conceptualising ynamics an operations. Disagreements an oubts raise about the utility of NHS performance ratings [10 12] are the manifestation of this ifficulty. These isagreements an oubts are similar to the funamental questions highlighte by Akkermans an van Oorschot while introucing BSC in their recent case stuy of the use of system ynamics in BSC evelopment [13]. They emphasise that the relevance of the balance scorecar evelope, an the processes use in eveloping it, shoul be teste rather than assume. They also expresse concern whether all inicators shoul work in the same irection or counteract. These same issues might be behin the isagreements an oubts about the utility of the NHS balance scorecar. This paper proposes that a causal-loop iagram (CLD) can provie a holistic view of the system an help reuce the split between ifferent views about the NHS performance ratings framework. Proviing information about the connections an interactions between ifferent performance inicators woul increase awareness of performance ratings an the formulation of performance improvement strategies. A CLD is a way of presenting moels of system thinking base on cause-an-effect relationships between ifferent system components or aspects. System thinking explains the ynamics of a system uner stuy an overcomes limitations of linear moelling by also accounting for feeback effects. In orer to ientify cause-an-effect relationships amongst performance inicators in the NHS BSC, we apply structural equation moelling [14,15] (SEM) to NHS Hospital Trusts performance ratings ata for 2001/2002 an 2002/2003. SEM has been use successfully for valiating causal effect assumptions in social sciences, an encompasses an extens regression, econometric, an factor analysis methos. Bollen s [15] milestone work in SEM literature has built parallels an clarifie ifferences between SEM an other traitional an establishe methos. A resulting significant SEM moel is use to construct a CLD, on the basis of which we investigate possible trae-offs (compromises) within the NHS BSC. 2. Methoology Investigating cause-an-effect relationships between inicators requires a structure an logical approach: ata selection an pre-processing, moel investigation, an moel representation. Fig. 2 outlines the framework aressing the objective of the paper with these methoological processes an outputs Data selection an pre-processing We use NHS Performance Ratings ata for the years 2001/2002 an 2002/2003, publishe by the Department of Health in Microsoft Ò Excel Ò atasheets ( performanceratings, last accesse February 8, 2005). For each year worksheets escribe the inicators an list the names of the trusts by bans accoring to their performance. Despite the limitations of the ata, two years of performance ratings ata are sufficient to satisfy the time-lag assumption [16] relate to causality an justify the use of the moification inex (MI) for moel investigation in SEM. The MI provies possible moel parameters that woul improve moel fit an significance [17]. Table 1 shows that the two years have ifferent sets of performance inicators. Nineteen inicators are common to both years (seven patient focus, seven clinical focus, an five capability an capacity focus inicators). Three inicators Deaths within 30 Days of a Heart Bypass Operation, Deaths within 30 ays of Selecte Surgical Proceures, an Privacy an Dignity have a high proportion Processes Output Data Selection an Pre-processing Moel investigation Moel Representation Causal-loop iagram Scenario Testing Fig. 2. Illustrate methoology.

4 908 B. Patel et al. / European Journal of Operational Research 185 (2008) of missing information or relate to special practice, hence, are not applicable to all hospital trusts. Thus this stuy is base on the 16 common performance inicators for which ata is available. Data pre-processing involve efinition of missing values for selecte performance inicators an conversion of all percentage variables into fractions of one to minimise scale ifference Moel investigation The following general path iagram provies a conceptual structure for the moel of causal relationships investigate. The moel tests the causal effect of 2001/2002 inicators (x j ) on 2002/2003 inicators (y i ). x j γ ij The above moel can be specifie in terms of equations as y i ¼ c ij x j þ f i ; for i ¼ 1;...; m an j ¼ 1;...; n: Here c ij is the coefficient suggesting the presence an measuring the weight of the causal relationship between the variables x j an y i. The variable f i represents the exogenous unobserve influence on the variable y i. For i = j, x j an y i are observations of the same inicator for 2001/2002 an 2002/2003 respectively. We use AMOS (Analysis of Moment Structures), a software a-in package for SPSS, to examine the above structural equation moel. The final significant moel was the result of forwar backwar iterations; at each iteration we inclue parameters suggeste by the MI an unselecte non-significant parameters that improve moel fit. We stoppe this process when no more parameters suggeste by the MI foun significant an improve moel fit. Details of the proceure can be foun in technical report [18]. The MI prouce by AMOS is as escribe by Jöreskog an Sörbom [19] an explaine by Sörbom [17] Moel representation The significant structural equation moel provie information about causal relationships amongst the selecte performance inicators use y i ζ i in the NHS BSC. Each significant causal relationship ientifie by this moel was translate into either a positive or a negative causal effect in a CLD. 3. Results 3.1. The Causal-loops The Causal-loop iagram in Fig. 3 is the common minimum significant structure erive from the 16 common performance inicators of the NHS BSC. As more information can be mae available, other performance inicators can prove to be linke with this structure using the same methoology. Clinical Negligence, a performance inicator of Clinical Focus in the NHS BSC, was not foun significantly relate with other inicators inclue in the stuy (see Fig. 3) an therefore is shown in Fig. 3 but omitte in later figures. There are 12 reinforcing loops (R) an five balancing loops (B). Reinforcing loops inicate a structure with exponential growth or ecay, also known as snowball effect. Balancing loops, also known as swinging-weights, inicate that the structure has an in-built counter effect, which controls growth or ecay. In a CLD, reinforcing loops have no or an even number of negative effect links. For any cause an effect variables linke with negative link, the effect will have opposite results than the cause variable. In reinforcing loops though, the effect variables linke with only every o numbere negative link are in contraictory position to that of the other variables in the loop. This feature is pivotal to scenario testing. On the other han, negative effect links in the balancing loops stabilise such contraictory positions with counter effects at the next iteration in the loop. In Fig. 3, Reamissions, efine as emergency reamission to hospital within 28 ays, is a critical inicator since it is part of all reinforcing loops an all balancing loops except one. Reamissions has the largest number of links (outwar an inwar together) to other inicators. Its position in this erive network of causal relationships means that improving emergency reamission inuces further improvement at the next iteration (year) an egraation in emergency reamission inuces further egraation. The percentage of planne target achieve for the total number of patients waiting for an inpatient appointment (Total

5 B. Patel et al. / European Journal of Operational Research 185 (2008) Clinical Negligence Breast cancer treatment within a month 13 wks Outpatient Waits B5 / 2 2 / B5 Junior Doctor s Hrs Reamissions for Chilren R4 0 6 mths Inpatient Waits Reamissions for Stroke B4 1 Delaye Transfer of Care B3 R8 B2 Total Inpatient Waits - %of Plan Reamissions for Fracture hip Information Governance Reamissions R9 Staff Satisfaction Survey R6 R7 Sickness Absence Rate R2 R3 Cancelle Operations B1 Data Quality R B Reinforcing loop Balancing loop R5 Fig. 3. Causal-loop iagram for the NHS balance scorecar. Inpatient Waits % of Plan) an the amount of time waste through absences as a percentage of staff time available for the irectly employe NHS staff (Sickness Absence Rate) are the secon most connecte inicators. These three performance inicators, one of clinical focus ( Reamissions), one of patient focus (Total Inpatient Waits % of Plan) an one of capability & capacity focus (Sickness Absence Rate) have links to an from all three types of inicators. The efinitions of Reamissions an Sickness Absence Rate are intuitive. The thir inicator, Total Inpatient Waits % of Plan, is efine as a measure of eviation from the planne targets. On a given year, a trust may serve more or less than its planne patient intake, from the total number of patients who are waiting for an appointment. The purpose of this performance inicator is to reuce waiting times for patients. There are two loops in Fig. 3, labelle as 2/B5 an B5/2, which are balancing or reinforcing epening on the ominance of the causal link/effect from the Total Inpatient Waits % of Plan to the Reamissions for Stroke Scenario testing We investigate the effects of intervention on Reamissions, the most linke inicator. For the purpose of analysis, the six possible scenarios can be groupe in two sets of three scenarios, epening on whether the balancing loop B3 in Fig. 3 is assume or not to be ormant. In either case similar results are obtaine, therefore only the three scenarios where B3 is ormant are consiere. The links to Reamissions for Stroke have ifferent polarity (see Fig. 3), therefore their combine effect on this inicator cannot be ascertaine without further assumptions regaring their ominance. In Scenario A (Fig. 4), we assume that the negative link from Total Inpatient Waits % of Plan is non-ominant, i.e. the links which it affects are ominate by other competing effects, an are able to ascertain all effects in the CLD. However, when we assume this link is ominant, more assumptions are require to ascertain the combine effect of the links to Reamissions for Chilren. Thus, in Scenario B (Fig. 5) an Scenario C (Fig. 6), the link from Total Inpatient

6 910 B. Patel et al. / European Journal of Operational Research 185 (2008) Clinical Negligence Breast cancer treatment within a month Delaye Transfer of Care 13 wks Outpatient Waits B5 / 2 2 / B5 Junior Doctor s Hrs Reamissions for Chilren B3 R8 R4 0 6 mths Inpatient Waits B2 Reamissions for Stroke B4 Total Inpatient Waits - %of Plan 1 Non-Dominant link Reamissions for Fracture hip Information Governance Reamissions R9 R7 Staff Satisfaction Survey R6 Sickness Absence Rate R2 R3 Cancelle Operations B1 Data Quality R B Reinforcing loop Balancing loop Dormant loop R5 Fig. 4. Scenario A. The negative link from the Total Inpatient Waits % of Plan to the Reamissions for Stroke is assume non-ominant. Waits % of Plan an from Reamissions for Stroke is assume non-ominant respectively. The states of inicators resulting from these three scenarios are liste in Table 2. In all three scenarios, the percentage of patients whose transfer from hospital was elaye (Delaye Transfer of Care) will be compromise ue to the irect negative link bringing the effect of improvement in Reamissions to Delaye Transfer of Care. On the other sie, the negative link from Reamissions will compromise the Data Quality inicator ue to the effect of the negative link in the reinforcing loop. N.B. In the long term, this effect will be ilute because of its relationship with Cancelle Operations in loop B1. The resulting negative effect from Data Quality improves Reamissions for Fracture Hip. Specific to each scenario, whether Reamissions for Stroke will be compromise or not epens on the en result of the negative effect from Total Inpatient Waits % of Plan, the positive effect from Reamissions for Fracture Hip, an the irect positive effect from Reamissions. Reamissions for Stroke is compromise in Scenario B (Fig. 5) an Scenario C (Fig. 6), but not in Scenario A(Fig. 4). Consequently, base on the position of Reamissions for Stroke, the inicator of the percentage of junior octors complying in full with the New Deal on junior octors hours (Junior Doctors Hours) an that of the percentage of patients seen within 13 weeks of GP written referral (13 weeks Outpatient Waits) will be compromise if Reamissions for Stroke is not compromise, which is the case in Scenario A. Reamissions for Chilren is compromise in Scenario A an C but not in Scenario B. The presence of ifferent non-ominant links ifferentiates each scenario an the state of the concerne performance inicators. Compromise inicators are highlighte with grey-shae rectangles in Figs The scenario comparison in Table 2 provies a clearer picture regaring possible compromises. Table 2 lists the performance inicators that have been participating in the various loops an their

7 B. Patel et al. / European Journal of Operational Research 185 (2008) Clinical Negligence Breast cancer treatment within a month Delaye Transfer of Care 13 wks Outpatient Waits B5 / 2 2 / B5 Junior Doctor s Hrs Reamissions for Chilren B3 R8 R4 Non-Dominant link 0 6 mths Inpatient Waits B2 Reamissions for Stroke B4 Total Inpatient Waits - %of Plan 1 Reamissions for Fracture hip Non-Dominant links Information Governance Reamissions R9 R7 Staff Satisfaction Survey R6 Sickness Absence Rate R2 R3 Cancelle Operations B1 Data Quality R B Reinforcing loop Balancing loop Dormant loop R5 Fig. 5. Scenario B. The positive link from Reamissions for Fracture Hip to Reamissions for Stroke, the positive link from emergency reamission to Reamissions for Stroke an the negative link from 6 months Inpatient Waits to Reamissions for Chilren are assume non-ominant. propose status uner ifferent scenarios. It suggests that compare to overall emergency reamission, Data Quality an Delaye Transfer of Care inicators are compromise in all scenarios. Junior Doctors Hours an 13 weeks Outpatient Waits inicators are compromise in Scenario A. Reamissions for Stroke is compromise in Scenario B an C; an Reamissions for Chilren is compromise in Scenario A an C. In all scenarios, the loop B3 is assume ormant, i.e. its effect is ignore. If the loop B3 is not ormant then each scenario coul be extene to have sub-scenarios. It will also reuce the effectiveness of the intervention brought for emergency reamission conitions, an the contrast between com- promise an non-compromise performance inicators. 4. Discussion We applie structural equation moelling to ientify the relationships between 16 common performance inicators in the NHS Balance Scorecars (BSC) for 2001/2002 an 2002/2003 an erive a causal-loop iagram (CLD) to show how the inicators interact. Scenario testing inicates that the NHS BSC has conflicting subsets of inicators. As far as the Reamissions inicator is concerne, reucing Reamissions in NHS hospitals impacts negatively on Delaye Transfer of Care an Data Quality. These averse effects woul have to be compensate by other intervention(s), which in turn coul increase Reamissions. Usually CLDs are create following qualitative an formative stuies involving stakeholers an expert opinion. Data is collecte to valiate the moel. Commercially, BSCs are use to foster sustainable growth an evelopment. In these moels the four imensions use learning an growth, internal processes, customers an financial have establishe cause-an-effect relationships. However, although the imensions in the NHS BSC are meaningful, their inter-relationships are unclear.

8 912 B. Patel et al. / European Journal of Operational Research 185 (2008) Clinical Negligence Breast cancer treatment within a month Delaye Transfer of Care 13 wks Outpatient Waits B5 / 2 2 / B5 Junior Doctor s Hrs Reamissions for Chilren B3 R8 R4 Non-Dominant link 0 6 mths Inpatient Waits B2 Reamissions for Stroke B4 Total Inpatient Waits - %of Plan 1 Reamissions for Fracture hip Non-Dominant links Information Governance Reamissions R9 R7 Staff Satisfaction Survey R6 Sickness Absence Rate R2 R3 Cancelle Operations B1 Data Quality R B Reinforcing loop Balancing loop Dormant loop R5 Fig. 6. Scenario C. The positive link from Reamissions for Fracture Hip to Reamissions for Stroke, the positive link from Reamissions to Reamissions for Stroke an the negative link from Reamissions for Stroke to Reamissions for Chilren are assume non-ominant. Table 2 Tabulate patterns base on Scenarios A, B, an C

9 B. Patel et al. / European Journal of Operational Research 185 (2008) Ientifying the statistically significant relationships between the inicators in the NHS BSC woul provie the basis for unerstaning the ynamics of a basket of performance inicators. This objective forms the basis of this novel approach to reconstruct the relationships between inicators of the NHS BSC. Using structure equation moelling we etermine the statistical relationships (pvalue < 0.05) between 16 of the common inicators in the scorecar. Using casual-loop iagrams we foun that the key inicator was emergency reamission. The central role of emergency reamission within the current set of inicators suggests that the starting point to aress the recurring problems in NHS hospitals is to focus on the clinical an social causes of emergency reamissions. A wor of caution shoul be expresse, however, because our CLD was base only on two years Performance Data for all Acute NHS Trusts in Englan. Three performance inicators were exclue as they ha insufficient ata. If the ata ha been available, incluing these inicators might have ae more links an loops. However, the structure presente here woul remain vali as the existing relationships woul not have change. Thus causal-loop iagrams coul be use as a reference framework, to buil a consensus for policy actions that woul lea to a solution for current problems. In this stuy, the purpose of the reconstructe CLD is to unerstan the impact of the basket of BSC inicators, rather than to valiate each relationship separately. However in some cases, practical explanations can be provie. For instance the relationship between Reamissions an Delaye Transfer of Care + Length of Stay Higher-level Inicators Use Reamission Inicators Not Use for the NHS Balance Scorecar Fig. 7. Connections to lower-level inicator(s). Delaye Transfer of Care coul be explaine by introucing an intermeiate inicator measuring length of stay (Fig. 7). The relationships between length of stay an emergency reamission have been investigate by Leylan [20] an Kossovsky et al. [21]. Their finings imply that elaye transfers of care results in increase length of stay, so there is a positive correlation between the corresponing inicators. The NHS star rating system has receive numerous criticisms [22 26] aime at both iniviual components an at the overall scheme. The approach evelope here provies a basis for answering questions about how an what works in the existing healthcare system. 5. Conclusion Managing performance in large organisations such as the NHS consists in orchestrating continuously a vast number of inicators. Frameworks such as BSC are useful strategic tools that link various performance inicators to performance management activities/processes of the organisation. However, their success is etermine by the knowlege of relationships between inicators an how these relationships aress long-term performance goals. Using historic NHS performance inicators we have ientifie a list of relationships that can be further investigate for in-epth knowlege. This knowlege can be acquire from the etaile interrelationships between ifferent aspect of the present healthcare units an the whole system. Scenario analysis results suggest that current performance improvement may not be sustainable ue to a conflict of inicators in the NHS BSC. This raises questions about the long-term effect of current policies in the UK, which aim at giving increasing autonomy to top performing hospitals. The finings in this research are inferences about an average acute NHS hospital trust. The stuy is using the maximum possible number of inicators that coul have been use. The use of subset of inicators is a limitation of ata availability. The Department of Health can easily incorporate all possible inicators by recalculating inicator values with calibrate an consistent criteria from their existing ata silos. The network of causal relationships in this stuy is base on ientifie linear relationships between performance inicators using structural equation moelling. Non-linear relationships coul provie more etaile characteristics

10 914 B. Patel et al. / European Journal of Operational Research 185 (2008) of the relationships between inicators an coul be investigate in further research. Further research will look for approaches to incorporate the changing composition of the BSC an observe characteristics over a longer perio. References [1] R.S. Kaplan, D.P. Norton, The balance scorecar measures that rive performance, Harvar Business Review 70 (1) (1992) [2] K. Castañea-Ménez, K. Mangan, A.M. Lavery, The role an application of the balance scorecar in healthcare quality management, Journal for Healthcare Quality 20 (1) (1998) [3] M.A. Fitzpatrick, Let s bring balance to health care, Nursing Management 33 (3) (2002) [4] J.R. Griffith, J.A. Alexaner, R.C. Jelinek, Measuring comparative hospital performance, Journal of Healthcare Management 47 (1) (2002) [5] J. Solano et al., Integration of systemic quality an the balance scorecar, Information Systems Management 20 (1) (2003) [6] R.S. Kaplan, D.P. Norton, The Balance Scorecar: Translating Strategy into Action, Harvar Business School Press, Boston, MA, [7] DoH, NHS Performance Ratings: Acute Trusts 2000/01, Leaflet, Department of Health, Lonon, UK, 2001, 23p. [8] DoH, NHS Performance Ratings an Inicators 2002, Department of Health, UK, Available from: <www. performance.oh.gov.uk/performanceratings/2002/tech_inex_ trusts.html>. [9] DoH, NHS Performance Ratings: Introuction, Department of Health, UK, July, Available from: < [10] J. Carvel, Four of the Top Hospitals Fall Out of Running, The Guarian, Lonon, 2003, July 16. [11] Society-Guarian, NHS Star Ratings: Reaction in Quotes, The Guarian, Lonon, 2003, July 16. [12] T. Shifrin, New NHS Measures Don t go Far Enough, The Guarian, Lonon, 2004, July 2. [13] H.A. Akkermans, K.E. van Oorschot, Relevance assume: A case stuy of balance scorecar evelopment using system ynamics, Journal of Operational Research Society 56 (8) (2005) [14] R.H. Hoyle, Structural Equation Moeling: Concepts, Issues an Applications, Sage Publications, Thousan Oaks, California, [15] K.A. Bollen, Structural Equations with Latent VariablesWiley Series in Probability an Mathematical Statistics, Wiley, New York, Chichester, [16] G.W. Bohrnstet, D. Knoke, Statistics for Social Data Analysis, thir e., F.E. Peacock Publishers Inc., Itasca, IL, [17] D. Sörbom, Moel moification, Psychometrika 54 (3) (1989) [18] B. Patel, Forwar backwar stepwise selection strategy for moel investigation, Technical Report, Health an Socialcare Moelling Group, University of Westminster, Lonon, 2005, 17pp. [19] K.G. Jöreskog, D. Sörbom, LISREL 7: A Guie to the Program an Applications, secon e., SPSS Inc., Chicago, IL, [20] A.H. Leylan, Examining the relationship between length of stay an reamission rates for selecte iagnoses in Scottish hospitals, IMA Journal of Mathematics Applie in Meicine an Biology 12 (3 4) (1995) [21] M.P. Kossovsky et al., Unplanne reamissions of patients with congestive heart failure: Do they reflect in-hospital quality of care or patient characteristics? American Journal of Meicine 109 (5) (2000) [22] R.M. Barker, M.S. Pearce, M. Irving, Star wars, NHS style, BMJ 329 (7457) (2004) [23] A. Gullan, Health professionals question star ratings for NHS, BMJ 325 (7358) (2002) 236. [24] Z. Kmietowicz, Star rating system fails to reuce variation, BMJ 327 (7408) (2003) 184. [25] R. Mannion, H. Davies, M. Marshall, Impact of star performance ratings in English acute hospital trusts, Journal of Health Services Research an Policy 10 (1) (2005) [26] I. Snelling, Do star ratings really reflect hospital performance? Journal of Health Organization an Management 17 (3) (2003)