Secondary Logo

Journal Logo

Original Article

Do Hospital Bed Reduction and Multiple System Reform Affect Patient Mortality?

A Trend and Multilevel Analysis in New Zealand Over the Period 1988–2001

Davis, Peter PhD*; Lay-Yee, Roy MA*; Scott, Alastair PhD; Gauld, Robin PhD

Author Information
doi: 10.1097/MLR.0b013e3181468c92
  • Free


Internationally there has been a considerable change in the role of the hospital through the 1990s, with higher rates of admission, shorter periods of stay, and growing rates of outpatient and day care.1 An important strand in this change in role was a conscious restructuring of hospital workforce and redesign of work in inpatient settings across the developed world.2 Over this same period, many of these countries also underwent bouts of broader health reform.3 New Zealand, where the government pays for 80% of health care and public institutions dominate the health system, was no exception. The country undertook 4 sets of changes to the publicly-funded health system up to 2001 (see Fig. 1), including a succession of public hospital sector reorganizations.4 At the same time, in a related trend, the sector experienced a substantial reduction in the availability of inpatient beds.5

Reform phases of New Zealand’s publicly-funded health services.

The substantive interest in the New Zealand case is 4-fold. First, it was one of a group of countries with national health service-type systems that implemented a suite of market-oriented reforms from the late-1980s to the mid-1990s (the others being Italy, Spain, Sweden, and the United Kingdom).6 These reforms were typically intended to create a “market” for publicly-funded health services by instituting competitive tendering between government-purchasing agencies and service providers vying among one another to win contracts to provide public services, and also by transforming public hospitals into public corporations expected to function like private cost-conscious businesses. These were features of the second and third reform phases in New Zealand (see Fig. 1). Second, this suite of reforms probably went further and faster in New Zealand than anywhere else and were part of a broader reform thrust in economic and social policy. They also drew widespread popular and political opposition.6 Third, New Zealand simultaneously experienced both a substantial reduction in availability of public hospital beds and 4 separate structural reorganizations (Fig. 1).4 Fourth, even though many of these reform experiments were short lived, internationally, as Or has noted, “the lack of proper evaluation … is striking,”7 particularly with concerns about possible effects on access and quality.6

Given the strength and coherence of the reform program, and its powerfully managerial and efficiency objectives,4 3 key questions arise. First, how did the performance of the hospital system respond to bed reduction and the agenda of reform? Second, with the level of popular and political concern, how did patients fare in access and quality of care? Finally, in a reform period that simultaneously experienced both substantial bed reduction and multiple structural reorganizations of both the funding and administration of public hospitals, what effects, if any, did these exert on the outcome? In this context, there is a pressing need to assess whether major system adjustments, together with some versions of market reform, might adversely affect the quality of patient care.8

In pursuing these analytical goals, aside from the presentation of descriptive trend data, a multilevel statistical model will be used. There are 2 reasons for this. The technical reason for adopting such an approach is that it takes into account quite explicitly the fact that the discharge data are clustered by hospital.9 The second, more substantive, reason for adopting a multilevel modeling approach is that the technique permits the identification of explanatory effects at different levels of aggregation, including the hospital.10 Although earlier work has addressed the application of a multilevel model to these data focusing on the methodological issue of time-dependent clustering,11 the purpose of this article is to apply a multilevel model to the analysis of the possible influence of bed reductions and system reform on trends in patient outcomes.


In this analysis, we treat bed reduction and organizational restructuring as 2 separate, if related, processes of health system change. We view them as 2 dimensions to the broad thrust of health system reform that characterized New Zealand for over a decade. Indeed, this reform thrust should, in turn, be seen as part of a much wider series of changes that were introduced into the economy, beginning in the mid-1980s with the election of a reforming Labour administration, and then into social policy, with an incoming National (conservative) government.

Therefore, in the structure of our analytical model, we treat reform phase and bed reduction as separate, but interrelated, system-wide trends shaping clinical and managerial practices at the hospital level—where we focus on length of stay and unplanned readmission—which, in turn, are likely to be important influences on mortality outcome. Ambulatory care is introduced as a system variable to represent the potential impact of changes in the nonhospital sector. For illustrative purposes only, we show 1 system-induced change—reduction in bed availability—and 1 hypothesized workload management response, length of stay, both graphed across 4 phases of reform (Fig. 2). As can be seen, the decline in bed availability started well before the specifically market phase of health reform, but plateaued before market reform was terminated, before declining again. The decline in length of stay followed a more consistent trend.

Impact of system-induced change.

Our model not only treats these 2 system trends as separate, if related, processes, but also makes no strong causal claims for it. Essentially, we are documenting a series of trends that point to a process of associated, and quite striking, changes in bed supply, in patient throughput and access, in clinical patterns of patient care, and in mortality outcomes. In essence, we are conducting a sophisticated, statistically enhanced, monitor of an uncontrolled reform experiment carried out in 1 country, albeit one of wider policy interest and generalizability to similar health systems.


Discharge data, comprising 7,682,497 records from all 34 New Zealand secondary and tertiary public hospitals, were accessed from the New Zealand Health Information Service (NZHIS) for the period 1988–2001 inclusive. After applying standard filtering conditions to ensure that definitions of variables were consistent and comparable over time,12 a total of 6,639,487 records were available for analysis.

For each record, NZHIS provided dates of admission and discharge, admission type (acute or arranged), age, ethnicity, domicile code, and diagnosis-related group (AN-DRG3.1), as well as 60-day postadmission mortality.13

Definition of Indicators

Throughput was taken to be the number of patient discharges in a year.

All discharges (or admissions) comprise day stay and inpatient cases; this excludes outpatient clinic visits that do not require admission to hospital, although they may be a precursor and/or follow up to an admission.

A day-stay admission was defined as one in which the admission and discharge dates were the same, regardless of intent; patients enter hospital for less than a day (eg, for a surgical procedure that does not require an overnight stay).

An inpatient admission was deemed to have occurred if the admission and discharge dates differed from each other; this involves the occupation of a hospital bed for a day or more.

The length of stay for each inpatient admission was defined as the discharge date minus the admission date, with subtraction of patient leave days.

The truncated length of stay was derived by assigning the 97th percentile value, calculated by diagnosis-related group and year, to those values greater than the 97th percentile value.

The average number of beds used in a year was calculated as the sum of all inpatient lengths of stay (truncated) divided by the number of days in the year.

An admission was defined as an emergency admission if the admission type was “acute” (AC) or “acute Accident Compensation Corporation covered” (ZC).13

A readmission was defined as having occurred if the patient returned to the same hospital, or was admitted to another hospital within 30 days of the original admission date, as long as the condition also belonged to the same major diagnostic category (MDC; derived from AN-DRG3.113).

An unplanned readmission was defined where the admission type was acute (AC), acute Accident Compensation Corporation covered (ZC) or “waiting list urgent” (WU).13

Use of hospital care, readmissions, and deaths are presented as (1) based on all admissions and (2) based on all patients, with only the last admission being considered in the case of multiple admissions in a given calendar year.

For the calculation of rates per 1000, the denominator was the New Zealand mean population for the year ended December 31; estimates of de facto population were provided for the years before 1991, with estimates of resident population provided thereafter.14

Maori are the indigenous people of New Zealand. To account for the underreporting of Maori patient ethnicity before 1996,15 ethnic affiliation for an admission was assigned as Maori where the same patient (identified by National Health Index number) had been recorded as Maori in a later admission.

NZDep96 deciles were derived from patient domicile codes as a measure of residential area deprivation (based on the proportions of people receiving welfare benefits, unemployed, on low income, with no access to a telephone or to a car, in single-parent families, with no educational qualifications, not living in their own home, and living in a crowded dwelling).16 The deprived group was defined as those patients domiciled in areas classified in NZDep96 deciles 9 and 10. We identified the deprived, Maori people, and the elderly (75 years and over) as potentially vulnerable social groups.

Ambulatory sensitive admissions are a subcategory of avoidable hospitalizations, comprising hospitalizations of people under 75 years of age from causes considered to be responsive to prophylactic or therapeutic interventions deliverable in ambulatory care settings.17

By linking hospital discharge and mortality data via patient NHI number, NZHIS was able to determine death within 60 days postadmission.

Multilevel Analysis

We used the SAS Glimmix procedure to fit a sequential series of generalized linear mixed models, using the RSPL (Residual Subject-Specific Pseudo-Likelihood) method.18 Multilevel logistic regression was applied to inpatient admissions data from 28 general hospitals over a period of 14 years (1988–2001 inclusive) to analyze trends in 60-day postadmission mortality. We excluded 6 specialist hospitals that were different in character. For patients with multiple admissions in a given calendar year, only the last admission was included in the analysis. Patient admissions (level 1) were considered to be nested within hospitals (level 2).

Patient covariates were: age, gender, ethnicity (Maori, non-Maori), deprivation (deciles), diagnosis (10 MDC13 groupings: nervous system, ear nose and throat, respiratory system, circulatory system, digestive system, musculoskeletal, skin, reproductive, pregnancy/neonatal, other), and number of diagnoses (1, 2 or more).

Both time (years 1988–2001) and reform phase (Area Health Boards 1988–1992, Crown Health Enterprises 1993–1995, Health and Hospital Services 1996–1999, and District Health Boards 2000–2001) were incorporated as fixed effects. We took the time variable to be related to changes in hospital management, and the reform phase variable to represent system change. Hospital covariates were mean length of inpatient stay and percentage unplanned readmission, both calculated on an annual basis.

The percentage of ambulatory-sensitive admissions and the total number of inpatient beds available per 1000 population per year were included as fixed system variables.

Hospital identity was introduced as a random effect in our base intercept-only model and variance components monitored as covariates were gradually added to subsequent models. Individual variables were entered into the model in a step-wise fashion. Relative measures of both fixed and random effects of specific factors at the patient and hospital levels were calculated as the percentage of total variance explained by the linear predictor (fixed effects) and percentage of total variance attributable to the hospital level (random effect). For the full model (n = 2,977,606, excluding admissions with missing fixed effects), parameter estimates, and standard errors associated with fixed effects of time, patient and hospital covariates, ambulatory care, reform phase, and inpatient bed availability are presented.

By specifying that only the intercept in the model is random, we have assumed that the effect of each of the variables in the model is the same for each hospital.

Snijders and Bosker19 have proposed an extension of the definition of R2 suggested by McKelvey and Zavoina for a single-level logistic model, which allows the proportion of variance explained by a multilevel random intercept logistic model to be calculated. The procedure assumes that the dichotomous outcome Yij is determined by an underlying threshold model and conceives the variance of the underlying variable to be composed of the variance of the linear predictor σF2, the variance of the intercept τ02, and the level 1 residual variance. Because the scale of the latent variable is arbitrary, it is customary to set the level 1 residual variance equal to that for a standard logistic distribution (π2/3). The variance of the linear predictor is calculated as the variance of the estimated values for Yij when only the fixed terms of the model are used in the prediction calculation. Thus, the total variance of the underlying variable, Yij, can be written as:

The proportion of variance explained by the model (ie, by the linear predictor) is then:

and the proportion of variance residing at level 2 (ie, attributable to the random intercept) is:


Results are presented first as descriptive tabulations of trends, and second in a multivariate analytic framework.

Descriptive Trends

How Did the Performance of the Hospital System Respond?

In Table 1, information is presented on bed supply, patient throughput, and pattern of care in 34 public hospitals at 4 time points over the period of study, 1988–2001, reflecting the different reform phases. Although there was a gradual decline of about 15% in the number of beds available, there was a sharper drop of nearly a third in the number of beds actually used (reflecting a decline in occupancy rate from 67% to 53%). Despite this downward trend in the effective availability of hospital beds, the number of inpatient discharges increased by a quarter, and the number of discharges overall grew by over two-thirds. This striking growth in patient throughput—despite shrinking bed supply—was achieved by virtually halving average length of hospital stay and by increasing the number of day-stay admissions by almost 30%. However, the proportion of emergency admissions increased slightly and rates of readmission grew to about 1 in 10 of all discharges, amounting to an increase of nearly a quarter for inpatients over the period. A similar trend characterized unplanned readmissions for inpatients.

Bed Supply, Patient Throughput, and Pattern of Care by Reform Phase, 1988–2001

How Did Patients Fare?

Table 2 presents data on the accessibility of hospital care, the pattern of admission for potentially vulnerable social groups, and outcomes for patients. It should be noted that the demographic pressure on a (declining) bed supply increased significantly over this period, as reflected in the growth of the population by nearly a fifth. Expressed as a rate per 1000, hospital admissions increased overall by 43% between 1988 and 2001, although only marginally—by 7.1%—for inpatient care. Assessing the impact of declining bed supply on groups that might be considered particularly vulnerable, it seems that their access to hospital care was not adversely affected over this period, with the rate per 100 admissions increasing by 52.7% and 0.5% for the elderly and Maori, respectively, though decreasing slightly by 5.5% for those living in areas of significant deprivation. The rate of ambulatory-sensitive admissions per 100 increased by 16.5% between 1988 and 2001. Finally, results are presented on patient outcomes. Over the period, crude (ie, unadjusted) rates of 60-day postadmission deaths per 100 admissions declined sharply overall but remained static for inpatients. After adjusting for age, however, a decrease of a quarter was evident in mortality rates for inpatients. Over the period of study, this equated to an average annual decline in age-adjusted, postadmission patient mortality of just under 2%.

Access to Hospital Care, Pattern of Admission, and Patient Outcomes by Reform Phase, 1988–2001

Table 3 provides a summary of average annual rates of 6 key supply, throughput, pattern of care, and outcome measures, by reform phase. Aside from inpatient discharge numbers, the first reform phase showed some of the strongest changes. Apart from discharge numbers, the final reform phase showed a sharp reduction in reform impact—increase in beds available, no decline in average length of stay, and no increase in the rate of unplanned readmissions.

Rates of Change Overall and by Reform Phase for Key Measures

Multilevel Analysis

What Effects Did Bed Reduction and Restructuring Have on the Outcome?

Table 4 presents results from the application of the multilevel model of the determination of 60-day postadmission mortality. Columns 2 and 3 give 2 measures of the extra explanatory value added at each stage. Column 4 presents coefficient estimates and standard errors for variables in the full model, whereas columns 5 and 6 give a more intuitive estimate of impact on patient outcome in terms of average change in the odds of death. The percentage of total variance explained by the linear predictor depends on the order of entry of the variables. This was as follows: “Intercept” (initial rate of patient mortality); year (overall linear trend in outcome); patient covariates (case-mix adjustment); hospital mean inpatient length of stay and percentage unplanned readmissions (measures of workload adjustment); percentage ambulatory-sensitive admissions; availability of inpatient beds; reform phase.

Multilevel Analysis of 60-Day Postadmission Death

There was confirmation of the overall linear trend shown in the descriptive data—an improvement in the odds of postadmission mortality of just under 2% per year (row 2, column 5). However, this accounted for a negligible percentage of total variance (column 2), although it was significant (column 4). Patient attributes accounted for over half all variation in mortality. The hospital-level workload adjustment measures both had a significant, and slightly deleterious, effect on patient outcomes. Neither of the system-level measures—ambulatory-sensitive admissions and availability of inpatients beds—had a significant impact. Reform phase did not adversely affect the improving trend in mortality, with both the market-oriented phases showing significant, positive effects. The proportion of total variance at the hospital level was never more than 1.70% and was in any case largely accounted for by case mix (patient covariates).



Over the period of study—1988–2001—the New Zealand public hospital system made very substantial inpatient bed reductions while maintaining both improvements in the quality of care (as measured by postadmission mortality) and high levels of throughput and access. The system also underwent 4 restructures, 2 involving experimentation with market-oriented reforms. The long-term improvement in patient outcomes was slowed slightly by aspects of hospital workload adjustment but not by reform phase.

Although the broad features of system change in the New Zealand case have been described before,4–8 the detailed impact of this period of health restructuring has not been established to date. Furthermore, although this pattern of hospital bed reduction and associated decline in length of stay has been a well-established trend in developed countries,20 the effects of such change on access and quality have not been determined, except for subnational systems, for limited periods of time, and for specific patient conditions.21

Interpretations and Issues

Other studies have established both a marked change in the pattern of hospital practice across the developed world and the maintenance of access and quality of care in the hospital sector despite an apparent diminution in supply.21 This study confirms these previous findings to a degree and does so for an entire health system over an extensive period of time and under circumstances of multiple reform. In addition, this investigation has had the advantage of being able to apply patient-level data to the analysis and introduce a powerful patient outcome measure (ie, 60-day postadmission mortality for all patient conditions).

A natural expectation from studies of system restructuring was that the speed and scale of documented declines in hospital inpatient bed capacity would adversely affect both access and quality.2,8 A complicating factor was the coincidence with bed reductions of both multiple restructures and market-oriented reforms, for which there is evidence of a disruptive impact on core infrastructure in the New Zealand case.8,22 The fact that such adverse outcomes have not eventuated in any clear-cut way could be due to one or more of the following factors:

  1. An existing position of excess capacity. New Zealand’s level of provision of hospital beds was indeed at the high end for the Organization for Economic Cooperation and Development (OECD).3
  2. The substitution of other inputs for reductions in beds (eg, more staff). There seems to be little evidence for this. Overall, staffing ratios remained stable and funding was at times severely constrained.23
  3. Diversion of patients to private sector. The private sector is largely a niche provider catering for elective surgery. Although it is difficult to obtain published data on admissions to the private sector, the information we have suggests that there was little change over this period (NZHIS; Data on file, personal communication).
  4. Rapid adjustment of management and professional practices consistent with trends in other countries,1 including reduced length of stay and increases in readmission rates and in use of day stay. Tables 1–3 provide supportive evidence.
  5. A long-term trend toward improved patient outcomes consistent with the advances in clinical practice and increased longevity. The data indicate steady improvement in age-adjusted postadmission mortality for the period of over a decade of study.

Strengths and Limitations

A considerable strength of the current study is its comprehensiveness and time span; aside from a “niche” private sector (see below), it addresses the impact of bed reduction and hospital sector change for an entire health system over a period stretching for more than a decade. Other studies have generally assessed these trends at the subnational level, usually investigating a single hospital, a city or other clustering, a system, or a regional grouping, and generally over short periods, and frequently for specific patient conditions.

Another advantage of this investigation is that data at the patient level were available in the New Zealand hospital information system. Furthermore, individual patients can be uniquely identified and, through this identifier, linkages can be made to the country’s mortality register, thus permitting the routine calculation of postadmission death rates. Furthermore, rates can be calculated separately for admission and patient denominators. It is also the case that the New Zealand experience provides perhaps the best opportunity to detect the impact of system change, given the combination of far-reaching public sector change, multiple health reconfigurations, market reform, and a program of major bed reduction.

There are a number of weaknesses in this study. First, data on the private hospital sector are not available. However, it should be noted that private hospitals generally cater for a “niche” within the wider sector, tending largely to concentrate on the long-term care of the elderly and on low-risk, routine elective surgery. Second, although patient-level information provides the foundation of the analysis, it would have been ideal to have had some measure of severity or acuity for patient admissions. This study has instead relied on patient age, case mix, and number of diagnoses, as means of approximating clinical complexity. Third, limited information was available on the supply side of hospital care, such as number and mix of staff, equipment, and financial resources, etc. Although the number of beds in use can be treated to some extent as a proxy for a broader range of hospital supply variables, changes in these other features—such as the number and mix of staff—could have made a significant difference to the outcomes of the analysis.


New Zealand was one of a number of national health service-type systems that experimented with market reforms in the 1990s. It also undertook substantial bed reduction and other restructures. By means of a number of compensatory mechanisms, the hospital sector maintained, and even increased, throughput. Patient access—particularly for identified “vulnerable” groups—did not diminish and postadmission mortality steadily improved, although this rate of improvement was slowed by features of workload adjustment but not by reform phase. These findings suggest that, other things being equal, national public hospital systems can maintain high levels of performance and patient responsiveness while undergoing drastic organizational change.


This study would not have been possible without the efforts of other members of the project team—Suzanne Gower, Patrick Graham, Mary Finlayson, Monir Hossain, Chris Errington—and the participating hospitals.


1. McKee M, Healy J. The changing role of the hospital in Europe: causes and consequences. Clin Med. 2001;1:299–304.
2. Sochalski J, Aiken LH, Fagin CM. Hospital restructuring in the United States, Canada, and Western Europe: an outcomes research agenda. Med Care. 1997;35(10 Suppl):OS13–OS25.
3. Docteur E, Oxley H. Health Care Systems: Lessons From the Reform Experience. Paris, France: OECD; 2003.
4. Ashton T, Mays N, Devlin N. Continuity through change: the rhetoric and reality of health reform in New Zealand. Soc Sci Med. 2005;61:253–262.
5. Ministry of Health. The Health and Independence Report. Wellington, New Zealand: Ministry of Health; 2001.
6. Laugesen M. Why some market reforms lack legitimacy in health care. J Health Polit Policy Law. 2005;30:1065–1100.
7. Or Z. Improving the Performance of Health Care Systems: From Measures to Action (A Review of Experiences in Four OECD Countries). Paris, France: OECD; 2002.
8. McCloskey BA, Diers DK. Effects of New Zealand’s health reengineering on nursing and patient outcomes. Med Care. 2005;43:1140–1146.
9. Austin PC, Tu J, Alter DA. Comparing hierarchical modelling with traditional logistic regression analysis among patients hospitalized with acute myocardial infarction: should we be analyzing cardiovascular outcomes data differently? Am Heart J. 2003;145:27–35.
10. Chang WC, Midodzi WK, Westerhout CM, et al. Are international differences in the outcomes of acute coronary syndromes apparent or real? A multilevel analysis. J Epidemiol Community Health. 2005;59:427–433.
11. Hossain M, Graham P, Gower S, et al. Hierarchical generalised linear models with time-dependent clustering: assessing the effect of health sector reform on patient outcomes in New Zealand. Health Serv Outcomes Res Methodol. 2003;4:169–186.
12. Ministry of Health. Hospital Throughput 2000/2001: For DHBs and Their Hospitals. Wellington, New Zealand: Ministry of Health; 2002.
13. New Zealand Health Information Service. Data Dictionary National Minimum Dataset 1999/2000. Wellington, New Zealand: Ministry of Health; 1999.
14. Statistics New Zealand. Demographic Trends 2002. Wellington, New Zealand: Statistics New Zealand; 2003.
15. Te Roopu Rangahau Hauora a Eru Pomare. Counting for nothing: understanding the issues in monitoring the disparities in health. Soc Policy J New Zealand. 2000;14:1–16.
16. Crampton P, Salmond C, Sutton F. NZDep96: Index of Deprivation. Report No. 8. Wellington, New Zealand: Health Services Research Centre; 1997.
17. Ministry of Health. Our Health, Our Future—Hauora Pakari, Koiora Roa: The Health of New Zealanders 1999. Wellington, New Zealand: Ministry of Health; 1999.
18. SAS Institute Inc. SAS/STAT 9.1 User’s Guide. Cary, NC: SAS Institute Inc; 2004.
19. Snijders TAB, Bosker RJ. Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. London, England: Sage Publications; 1999.
20. Hensher M, Edwards N, Stokes R. The hospital of the future: international trends in the provision and utilisation of hospital care. BMJ. 1999;319:845–848.
21. Brownell MD, Roos N, Burchill C. Monitoring the impact of hospital downsizing on access to care and quality of care. Med Care. 1999;37(6 Suppl):JS135–JS150.
22. Gauld R. One step forward, one step back: restructuring, evolving policy and information technology and management in the New Zealand health sector. Gov Inf Q. 2004;21:125–142.
23. Gauld R. Revolving Doors: New Zealand’s Health Reforms. Wellington, New Zealand: Institute of Policy Studies; 2001.

health system reform; patient outcomes; multilevel analysis

© 2007 Lippincott Williams & Wilkins, Inc.