Risk adjustment for hospital use using social security data: cross sectional small area analysis
BMJ 2002; 324 doi: https://doi.org/10.1136/bmj.324.7334.390 (Published 16 February 2002) Cite this as: BMJ 2002;324:390All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
Carr-Hill et al present a model and formula using both Family Credit
and Income Support data. We know from other work that both these variables
are probably correlated with use of in-patient services and so, as the
authors say, their appearance in the model seems "intuitively
appropriate". However they are also probably correlated with each other
which is presumably the reason that Family Credit has a negative
coefficient in the model, which is not intuitively appropriate at all. Any
attempt to include two highly correlated variables in the same model
usually results in one of them having a negative sign even though it is
positively correlated with the variable of interest. The solution is to
only include only one of the variables or to construct a single variable
which combines the two variables using principal components or some
similar method.
Competing interests: No competing interests
Editor –Carr Hill et al. propose new socio-economic indicators for
allocating public funding to hospital services, using socioeconomic and
demographic variables that can be updated between census1. Although they
used spatial techniques to reflect the influence of health care supply on
usage, we think, however, that they overlooked two critical aspects of
space : clustering and contextual factors of health care use.
As we have shown, it is incorrect to treat data from contiguous
spatial entities (such as wards) as independent observations, because this
overlooks the fact that they have similar levels of unobserved socio-
economic factors, environmental risks, and health status 2. Overlooking
such spatial clustering in the dependent variable (resources use) or
independent variables (socio-economic or health status) will lead to over-
estimation of parameter precision. It might also bias the results if
location can act as a confounding factor, as illustrated by some
epidemiological studies 3. Both over-precision and bias will leads to
incorrect conclusions , hence, unfair resources allocation. Modern
spatial techniques would help tackle such spatial clustering, for example
by the way of simultaneous auto-regressive models.
Moreover, aggregating the data at ward level conflates different
levels of analysis such as the household, ward, and district levels.
Being a elderly person living alone may increase health care needs at the
household level, but this relation could also be much stronger in rural
areas where the impact of isolation is harder to temper than in urban
centres. Conversely, urban centres may have poorer environmental
standards, making health care use more likely. In a multilevel analysis,
Jones and Duncan concluded that “In general, irrespective of individual
characteristics, places with a low income or a high deprivation suffer the
worst health […]” 4. The allocation of resources would be made fairer by
identifying such contextual factors that, for given individual risk
factors, decrease or increase the association between socio-economic or
health status and health care use. The authors have used dummy variables
to control for between-boards differences. This entails ignoring health or
socio-economic differences between boards. This is a pity, as the authors
have themselves previously shown that up to 44% of the variance of health
care use is due to inter-district variation 5. It is certainly a
significant improvement to allocate resources with socio-economic
indicators that are more up-to-date. But it is a mistake to downplay the
role of ecology in health, particularly if this were to lead to resource
redistribution from metropolitan boards to rural ones.
References
1.Carr-Hill RA, Jamison JQ, O'Reilly D, Stevenson MR, Reid J,
Merriman B. Risk adjustment for hospital use using social security data:
cross sectional small area analysis. BMJ 2002;324:390.
2.Lorant V, Thomas I, Deliège D, Tonglet R. Deprivation and mortality
: implication of spatial autocorrelation . Social Science and Medicine
2001;53:1711-9.
3.Clayton DG, Bernardinelli L, Montomoli C. Spatial correlation in
ecological analysis. Int J Epidemiol 1993;22:1193-202.
4.Jones K,.Duncan C. Individuals and their ecologies : analysing the
geography of chronic illness within a multilevel modelling framework.
Health and Place 1995;1:27-40.
5.Smith P, Sheldon TA, Carr Hill RA, Martin S, Peacock S, Hardman G.
Allocating resources to health authorities: results and policy
implications of small area analysis of use of inpatient services. BMJ
1994;309:1050-4.
Competing interests: No competing interests
Rural areas may need more health care resources in England too
EDITOR - Carr-Hill et al have shown that the use of a risk adjustment
formula in Northern Ireland that incorporates direct measures of poverty
at small area level would move resources from urban to rural areas1. Their
inclusion of social security data suggests one way of overcoming the
limitations of using indirect census based proxies to assess need for
health care. Another is to use existing epidemiological evidence to derive
direct estimates of morbidity in different areas. We have modelled the
impact of applying a morbidity-based capitation methodology to the
allocation of resources for inpatient care for coronary heart disease
(CHD)2. We also find that rural areas, particularly those with older
demographic profiles, would stand to gain most from the introduction of a
direct needs-based approach to resource allocation.
We derived morbidity-based allocations for inpatient CHD use for 34
primary care organisations (PCOs) serving 3.54 million patients in 7
health authorities in contrasting locations in England. Age, sex and class
adjusted prevalence rates of severe (Grade 2) angina and myocardial
infarction (MI) recorded in the Health Survey for England were attributed
to PCO populations. An age cost curve was established by dividing
historical HRG reference cost expenditure on CHD in the total sample by
the number of people in each age cohort who, on the basis of our
epidemiological estimates, would be expected to have symptoms of severe
angina and/or MI. This was then applied to our estimates at a local level
in order to determine a CHD clinical programme budget for each PCO.
Percentage differences between morbidity-based allocations and
indicative allocations based on the current Hospital and Community
Services (HCHS) formula were compared against Townsend's Material
Deprivation scores, the percentage of population aged 65+ and the DETR
Index of Multiple Deprivation Access domain scores. We found that a
morbidity-based capitation methodology resulted in a significant shift of
hospital resources for CHD away from PCOs serving deprived areas (r=0.845;
p<_0.001 towards="towards" pcos="pcos" serving="serving" populations="populations" with="with" older="older" demographic="demographic" profiles="profiles" r="0.847;" p0.001="p0.001" and="and" rural="rural" p0.001.="p0.001." p="p"/> There is a growing concern that the needs of rural populations are
not adequately reflected in the formulae used to make funding allocations
to the NHS3,4. The move to PCO allocations is likely to magnify such bias.
Serious consideration should thus be given to the fact that, using two
very different approaches to capturing 'direct' need, there appears to be
case for transferring resources from urban to rural areas.
Sheena Asthana, Principal Lecturer, Department of Social Policy,
University of Plymouth, Drake Circus, Plymouth, PL4 8AA,
sasthana@plymouth.ac.uk
Alex Gibson, Senior Lecturer, Department of Geography, University of
Exeter, Amory Building, Rennes Drive, Exeter, EX6 4PN,
a.gibson@exeter.ac.uk
Graham Moon, Professor, Institute for the Geography of Health, University
of Portsmouth, Milldam, Burnaby Rd, Portsmouth, PO1 3AS,
graham.moon@portsmouth.ac.uk
John Dicker, Associate Director, Information Management and Technology,
Iechyd Morgannwg Health, 41 High Street, Swansea, SA1 1LT,
john.dicker@morgannwg-ha.wales.nhs.uk.
Philip Brigham, Senior Research Fellow, Department of Social Policy,
University of Plymouth, Drake Circus, Plymouth, PL4 8AA,
pbrigham@plymouth.ac.uk
1 Carr-Hill, RA., Jamison, J.Q., O'Rielly, D., Stevenson, M.R., Reid,
J., Merriman, B. Risk adjustment for hospital use using social security
data: cross sectional small area analysis. BMJ 2002;324:390-2.
2 This research has been funded by the Economic and Social Research
Council's Health Variations Programme (Reference L128251031)
3 White, C. Who gets what, where - and why? The NHS allocation system in
England is failing rural and disadvantaged areas. Rural Health Forum and
University of St Andrews, 2001.
4 Asthana, S., Brigham, P. Gibson, A. Health Resource Allocation in
England: What Case can be Made for Rurality? Rural Health Allocations
Forum and University of Plymouth, 2002.
No competing interest
Competing interests: No competing interests