Editor's Choice | This Week in BMJ | Press releases
BMJ No 7116 Volume 315 Papers Saturday 1 November 1997
Impact of the NHS reforms on English hospital productivity: an analysis of the first three yearsNeil Soderlund, Ivan Csaba, Alastair Gray, Ruairidh Milne, James Raftery
AbstractObjectives: To evaluate the effect of purchaser mix, market competition, and trust status on hospital productivity within the NHS internal market. Methods: Hospital cost and activity data were taken from routinely collected data for acute NHS hospitals in England for 1991-2 to 1993-4. Cross sectional and longitudinal regression methods were used to estimate the effect of trust status, competition, and purchaser mix on average hospital costs per inpatient, after adjusting for outpatient activity levels, casemix, teaching activity, regional salary variation, hospital size, scale of activity, and scope of cases treated. Results: Real productivity gains were apparent across the study period for NHS hospitals on average. Casemix adjustment drastically improved cross sectional comparisons between hospitals. Gaining trust status and increasing host district purchaser share were associated with productivity increases after adjustment for casemix, regional salary differences, and hospital size and scope. Hospitals that became trusts during the study period were on average less productive at the beginning of the period than those that did not, and there were no significant productivity differences between trust waves at the end of the period in 1993-4. Market concentration was not associated with productivity differences. Conclusion: Further analysis is needed to determine whether overall and trust associated productivity gains are transient effects, one off shifts, or self perpetuating reorientations of organisational behaviour. Hospitals may have chosen to become trusts because they anticipated being able to increase productivity. Increases in the proportions of small purchasers were associated with increasing costs. Importantly, this study could not adjust for changes in the quality of care.
IntroductionThe NHS internal market aimed to make hospitals more efficient and responsive to local needs through various organisational changes. Yet their impact has hardly been evaluated. We examined changes in hospital productivity, defined as the changes in cost per inpatient episode, adjusted for several other factors, for the first three years of the NHS internal market, looking particularly at the impact of the reforms. Methods in briefTo make the results accessible to a broad range of readers we have kept mathematical formulas and economic jargon to a minimum. The full methods and sources of information are available on the BMJ's website (www.bmj.com), and more details of the methods and results are available from the authors on request. Here we outline the methods in brief. We used a set of cost function analyses: a cost function expresses the relation between what hospitals spend on the one hand and what they produce on the other, after adjustment for differences in the prices of inputs. Following Evans(1) we included several environmental factors, in addition to outputs and factor prices, which may have influenced productivity. These include elements of the NHS reforms. We hypothesised that NHS hospital costs per inpatient episode are a function of the levels of other outputs produced, the costs of hospital inputs, and a set of environmental constraints which reflect the hospital's internal structure and position in the "quasi-marketplace." One aspect of hospital output, quality, was not measurable, and our approach assumes that on average this is relatively constant across hospitals and over time. We used average hospital cost per inpatient in 1991-2 pounds sterling (total costs/numbers of inpatient episodes), including a notional capital charge, as the dependent variable in the model. The variables used to explain hospital cost variation, their derivation, postulated effect on costs, and year of collection are described in table A (www.bmj.com). The postulated relationship is: average cost per inpatient=a function of (average casemix, average long stay days per inpatient (the mean length of stay of episodes above the average for their healthcare resource group), % multiple episodes, outpatient attendances per inpatient, accident and emergency attendances per inpatient, day attendances per inpatient, student teaching units per inpatient, prices of capital items, wage prices, scale of activity, hospital size, degree of specialisation, trust status, competition from other hospitals, mix of purchasers). We postulated a simple linear relation between average costs per inpatient and the explanatory variables. To capture the differences both between providers and within providers over time we analysed all three years' data simultaneously. The model was estimated using two approaches. The first-the pooled cross sectional model-pooled all observations together and adjusted for other unmeasured effects that affect all providers, using dummy variables for the second and third study years. It examined the relation between absolute levels of the dependent and explanatory variables for each provider year with each treated as a unique data point. Thus there was no link between successive years for the same provider. The second approach-a fixed effects longitudinal model-examined the association between changes in explanatory variables from one year to the next and corresponding changes in the dependent variable. Additionally, overall hospital productivity by wave of trust status was assessed. Hospitals that had still not become trusts by the third year were labelled persistent directly managed units. This model was identical to the pooled cross sectional model except that the trust and year dummy variables were replaced by 11 dummy variables representing each trust wave for each of the study years (4 wavesx3 years-1). The adjusted costs per inpatient episode for each trust wave for each year were estimated. Only hospitals that had complete data for all three years were used for this analysis (510). The main sources of data were Hospital Episodes Statistics and Hospital Financial Returns. sas software was used for all computation.(2) Final sample sizes for the three years were 198, 219, and 221 hospitals. The unit of analysis was the NHS provider, which may have included one or more hospital sites. ResultsTable 1 gives summary statistics for the three years of data used, after excluding hospitals with large discrepancies between different sources of total recorded activity. Table 2 shows regression results for the pooled cross sectional model and the fixed effects longitudinal model. The coefficient represents the change in average costs per inpatient episode associated with a unit change in the respective explanatory variable. Confidence intervals (95%) for the slope coefficient are estimated as: slope |plus or minus 1.96 x standard error of slope.
In the pooled cross sectional model the most significant influence on costs was the casemix index variable. Episode inflation showed a significantly negative effect on costs. All remaining output variables showed significantly positive effects on cost. The positive coefficient on the inverse of patient episodes indicated decreasing costs per episode when patient numbers increase while capacity is held constant. This should not be interpreted as showing economies of scale, however. Specialisation (a narrower range of cases) and increased capacity (bed numbers) were associated with higher costs, whereas trust status and host purchaser share had negative effects, although these last two effects were not statistically significant at the 5% level. In the longitudinal model the effects of casemix, long stay days, wage prices, and specialisation were significantly reduced and did not contribute significantly to the model, probably because they changed little over the three years. The same phenomenon probably affected the market concentration index and the % of multiple episodes. In the longitudinal model, however, trust status and proportion of patients from the host district purchaser had a significant negative effect on average costs. The period dummy variables in both models indicate that overall productivity of the hospitals improved over the three years and that improvements between years 1 and 3 were significant at the 5% level. Tests for heteroskedasticity(3) and multicollinearity(4) showed that these were not significant problems for the models estimated.
Table 3 shows the results of the analysis where trust wave and period interaction terms replaced the trust and period dummy variables. Only the coefficients for the effects of interest are shown, together with 95% confidence intervals. At the beginning of the internal market, directly managed units about to become trusts-that is, second and third wave trusts-were significantly less productive than the persistent directly managed group. By 1993-4 there was no significant difference between each trust wave and the persistent directly managed units, although the latter were the least productive on average. For second and third wave trusts the steeper observed decline in costs was in the year immediately after becoming a trust compared with the other year for which they were studied. DiscussionExisting evaluations of the effect of reforms on hospital care tend to be politically polarised and poorly supported by data.(5) This discussion deals mainly with evaluating individual aspects of the NHS reforms, and the results implied by the coefficients on other terms are not dealt with in detail. The use of a patient level casemix adjustment is, however, unique in NHS hospital cost studies: its effect on results obtained in the cross sectional model was profound and suggests that previous studies that have compared hospitals without such adjustment(6-9) might be significantly biased. Although no prereform years were included in the analysis, the first year of the internal market (1991-2) was a steady state year, so it is reasonable to treat it as a control year. The two main data sources we used, the Hospital Episodes Statistics and the Hospital Financial Returns, are infrequently used for research. Apart from the KP70 reconciliation statistics to validate the Hospital Episode Statistics, no routine validation sources were available. Two sources of error may have influenced our results: random errors, and biases. Random errors in the Hospital Episodes Statistics would have been unlikely to affect substantially the results because of the large sample sizes (about 10 million episodes per year). Furthermore, providers had no incentives to inflate their recorded casemix since this was not being used for reimbursement. They may well have increased their apparent volume of activity, however, by discharging and readmitting patients or by reclassifying outpatients or day attenders as inpatients, and we have no way of detecting this. Errors in cost data may well occur in departmental costs because of different cost allocation mechanisms, and other studies have suggested discrepancies at this level.(10) This study used mainly total cost data, however, which is less susceptible to allocation errors. Furthermore, hospital accounts are audited annually, and the data we used were compiled and cleaned by the Audit Commission. The three main areas whereby the NHS internal market reforms might be expected to influence hospital productivity are trust status, and the managerial changes and incentives that that implies; competition between providers; and the establishment of small, discretionary purchasers in the form of fundholders.
Trust status
The Radical Statistics Group has suggested that some of the apparent efficiency gains in the internal market may be due to one off disposal of fixed assets or so called "episode inflation" within a single admission.(5) The latter is included in the model, and does indeed appear to inflate productivity gains, but the trust effect persists even after adjustment for multiple episodes. Some of the cost decreases associated with trusts may have been due to disposal of capital stock. In an unpublished analysis, however, we found significant trust related decreases in several non-capital costs.
Analyses of the effect of trust status by trust wave (table 3) sheds
more light on the timing of productivity changes. For second and third
wave trusts the largest gain in productivity was in the year of gaining
trust st
Competition
Purchaser shares
Conclusion
Comparisons of performance between hospitals should take
casemix into account, as failure to do so could significantly bias
results Gaining independent trust status was associated with
significant productivity gains for NHS hospitals, although some of the
effect may have been due to self selection; and at the end of the study
period productivity differences between trust waves were
non-significant Competition between hospitals had no significant effect on
productivity during the first three years of the internal market Hospitals that contract with many smaller purchasers other
than their host district are more costly, other factors being equal
Computing facilities for processing national inpatient data
sets were provided by the National Casemix Office, Winchester. Thanks
to Christopher Bliss, Anthony Zwi, Nick Hicks, Ali McGuire, and two
anonymous referees.
Funding: National Casemix Office and the Oxford and East Anglia
Regional Health Authority.
Conflict of interest: None.
For Health Policy,
Department of Public Health and Primary
Care, Centre for Sociolegal
Studies, National Casemix Office,
Correspondence to: Dr Neil Soderlund, Centre
References
1 Evans RG. Behavioural cost-functions for hospitals.
Can J Econ 1971;4:198-215.
2 SAS Institute. SAS: statistical analysis
system. Cary, NC: SAS Institute, 1994.
3 White H. A heteroskedasticity-consistent covariance matrix
estimator and a direct test for heteroskedasticity.
Econometrica 1980;48:817-38.
4 Belsey DA, Kuh E, Welsch RE. Regression
diagnostics. New York: John Wiley, 1980.
5 Radical Statistics Health Group. NHS "Indicators of
success|mK: what do they tell us? BMJ 1995;310:1045-50.
6 Propper C. Market structure and prices: the responses
of NHS hospitals to costs and competition. Bristol: Department
of Economics, University of Bristol, 1994.
7 Bartlett W, Le Grand J. The impact of the NHS reforms
on hospital costs: some preliminary findings. Bristol: School
for Advanced Urban Studies, University of Bristol, 1992.
8 Bartlett W, Le Grand J. The performance of trusts. In:
Robinson R, Le Grand J, eds. Evaluating the NHS reforms.
London: Kings Fund, 1994:54-73.
9 Smee CH. Self-governing trusts and GP fund holders: the
British experience. In: Saltman RB, Von Otter C, eds.
Implementing planned markets in health care: balancing social
and economic responsibility. Buckingham: Open University Press,
1995;177-208.
10 SÍ-derlund N, Milne R, Gray A, Raftery J. Differences in
hospital casemix, and the relationship between casemix and hospital
costs. J Public Health Med 1995;17:25-32.
11 Zwanziger J, Melnick G. The effects of hospital competition and
the Medicare PPOs programme on hospital cost behaviour in California.
J Health Econ 1988;7:301-20.
12 Miller P. Nearest and dearest? Or are GP fundholders
responding to price differentials. MSc thesis. University of
York, 1994.
13 Glennerster H, Matsaganis M. The UK health reforms: the
fundholding experiment. Health Policy 1993;23:179-91.
14 Culyer AJ, Posnett J. Hospital behaviour and competition. In:
Culyer AJ, Maynard A, Posnett J, eds. Competition in health
care: reforming the NHS. Houndmills: Macmillan Press, 1990.
15 Dranove D. Pricing by non-profit institutions. The case of
hospital cost-shifting. J Health Econ 1988;7:47-57.
16 Foster RW. Cost-shifting under cost reimbursement and
prospective payment. J Health Econ 1985;4:261-71.
17 Sloan FA, Becker ER. Cross-subsidies and payment for hospital
care. J Health Polit Policy Law 1984;8:660-85.
18 Staten M, Dunkelberg W, Umbeck J. Market share and the illusion
of power. Can Blue Cross force hospitals to discount? J Health
Econ 1987;6:43-58.
(Accepted 21 May 1997)
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||