A W Nimmo, G Peterkin, D R Coid
SMJ 2003 49(1): 66-68
Department of Public Health, Grampian NHS Board, Summerfield House, 2 Eday Road, Aberdeen
Abstract
Information on the distribution of mortality and morbidity in general practice is scarce, and not easily accessible either by health authorities or individual general practices. Although the assessment of population mortality is a standard public health measure, colleagues in public health, information sciences and general practice rarely undertake such activity related to general practice populations. Mortality information can be used for various purposes. Examples are providing background data for clinical audits or alerting practitioners to quality issues in the healthcare industry or even suggesting environmental hazards. We measured the experience of mortality in all generally practices in Grampian in the years 1991 to 1999. All practices were notified of their actual and expected mortality over this period of time and asked for comment. Only three general practices had excess mortality experience in both the periods 1991 to 1995 and 1996 to 1999. Only a minority of practices commented on the results. It appears that a high presence of nursing homes in a practice might skew the results; this phenomenon will be central to further inquiry. uture methodology for recording death in general practice should take account of this, as well as providing an account of the qualitative aspects of patients’ need for a dignified, satisfactory death.
Introduction
For general practitioners to undertake reviews of effectiveness of clinical care, a variety of information may be needed. Amongst data that may be useful are descriptions of mortality of their patients, particularly in relation to other populations. Such populations may include neighbouring general practices, local government and health board areas. General practitioners might use mortality to stimulate clinical audits1,2,3 or to identify anomalies in the experience of death of their local community4,5. Identification of general practice based mortality should be a standard process, undertaken by population scientists and general practitioners. However, the technical processes are not necessarily straightforward. For example, patients may be mobile, and the sizes of denominator populations are difficult to estimate. Practices merge, which also cause the denominators to change. Furthermore, general practitioners, with a practice list of around 1,500 patients, would only expect to experience little more than one patient death a month. Small numbers of deaths make the confidence intervals on death rates relatively high, thereby making conclusions about comparisons between practices difficult. Notwithstanding these difficulties, we believed that we should attempt to overcome the technical problems. We wished to provide colleagues in general practice with more information about the circumstances of mortality in their local area, and those of their colleagues, and to enter a dialogue about future liaisons between information services, general practice and public health practitioners in Grampian.
Method
A variety of approaches are possible for analysing mortality in general practice6,7. Our strategy was to standardise for known causes of mortality in order to focus on the residual unexplained variation. This approach can be progressively refined to identify special sub-populations with their own mortality characteristics, which will be important in the future development of this work. The method chosen was based on deaths recorded on the Community Health Index (CHI). The records used enabled the identification of:
the deceased patient’s general practice
the patient’s age
sex of the patient
the deprivation category of the patient’s locality (derived from the postcode).
Comparable population data for each general practice could be derived from the CHI for living patients. This enabled the standardisation, for age, sex and deprivation of practice populations. The CHI data for Grampian is believed to be of good quality, and it has an overall inflation factor of approximately 4% above the Registrar General’s mid-year population estimate. At older ages the inflation factor is much smaller. Age groups were categorised to the "under fives", followed by 10-year age bands, and finally ages "85 and over". Deprivation was based on a local version of the Carstairs score, defined at the level of Census output areas rather than postcode sectors. This is a more effective way of identifying pockets of deprivation. The method used to standardize the score was the same as that used for postcode sectors data. The scores were then grouped into three category with the affluent group corresponding to negative scores less than -3, the deprived group with scores greater than +3, and the remainder assigned to a middle group. The indirect method of standardisation was used since it can cope with the problem of empty population cells, eg. a practice in an affluent area may have no patients in the deprived population category. The direct standardisation method is less effective in dealing with this problem. The period of study was 1991 to 1999. During this time there were a number of changes in the composition of practices in Grampian as a results of mergers or splits in general practices. This was handled by working at the greatest level of aggregation. Practices, which had split, were treated as whole practices throughout the period. Population samples were taken each year from the CHI, which enabled the population of merged practices to be constructed across the whole time period. Estimations of practice mortality were reported to practices, in the form of actual and expected numbers of deaths for 1991 to 1999. An estimation of statistical significance was provided. Expected deaths were calculated on the assumption of uniform mortality across Grampian practices with due allowance for variation in age, sex and deprivation categories. The Mantel-Haenszel method was used to test statistical significance. In view of the fact that multiple tests were being carried out (there were almost 100 practices involved) a conservative acceptance level was imposed on the tests, corresponding to p<0.001. This would ensure that there was only a 1-in –10 chance of any single one of the 100 practices being wrongly attributed to have extreme mortality when in fact it was normal. Practices were also supplied with anonymised information on mortality in a group of at least ten practices most similar in size to their own so that they could form an assessment of how they compared to other practices.
Results
A sample of the feedback of results to individual practices is shown Table I. This shows the actual and expected numbers of deaths for 23 practices with an average list size of between 1000 and 4000 patients, for the years 1991 to 1999. In this example practice "P" had a significant deviation from expected mortality (higher than expected). Some of the overall results are shown in Table II where the practices with the highest and lowest numbers of patients have been listed. Five practices were found to have mortality significantly different from expected in both the periods 1991 to 1995 and 1996 to 1999. Three of these practices having high mortality and the other two are low.
The close correlation between observed and expected deaths is essentially due to variation in practice size, and what is really of interest is when the divergence between these values is statistically significant. Practices with significantly greater than expected numbers of deaths are shown as triangles, those with fewer than expected are shown as circles, while the remainder are represented by diamond symbols.
Three practices responsed, following notification of results, and their comments were varied. One with higher than expected mortality said that the practice served a number of nursing homes in the practice area and specialised in palliative care. The implication of this was that a high proportion of their elderly patients was frail. Such patients would probably have short life expectancy and therefore contribute to a higher rate of death. Analysis of this apparent phenomenon will be the subject of a subsequent report.
Practitioners from one merged practice wished to know whether any difference existed between the practices prior to the merger. This was indeed found to be the case as only one of the component practices had high mortality before the merger. After the merger the mortality in the practice has continued to be high, but not to a level that was statistically significant. The likelihood of a statistical test being "significant" depends on the absolute number of deaths recorded. These were relatively small because of the short period of the merger’s existence. It is also more difficult to detect significant deviation in mortality in small practices than in large ones due to the smaller number of deaths involved in any given period.8
Discussion
The thrust of this work has been to provide more information for practitioners about the experience of mortality of their patients. At this time doubts remain about the accuracy of the information and the way in which it might be used. However, the constant review of such data will lead to its improvement and its usefulness to practitioners. There are two main sources of mortality data in Scotland – from the General Register Office for Scotland (GROS) and from the CHI. GROS mortality records form a richer source of data since they contain extra information on cause of death. However it was extremely difficult to link this data to individual general practices.2,7 Information about the doctor signing the death certificate was often restricted to a surname and a town of practice. The field that records details of the patient’s own General Practitioner was usually blank, except in very recent years. Even where information was available it was textual, so that names and addresses would have to be manually matched to a standard list in order to determine the patient’s practice.
Preliminary work on this study revealed that there are some systematic differences between the deaths recorded on the CHI and those recorded by GROS. For example, institutionalised hospital patients are deleted from the CHI after they have been in hospital for a period of two years, and they do not have routine access to a general practitioner. When such patients die they will be recorded on the GROS system but not on CHI. Another problem occurs with neonatal deaths where the child may have died before they were ever registered with a practice. Members of the armed forces have their own primary care system. When deaths occur amongst service personnel, they are not be recorded on the CHI. Such deaths cannot be assigned to particular practices.
Practice death registers have been advocated as a way of improving information on deaths in general practice4,6, but they are not yet widely available. In the absence of effective registers, our attempt to produce a quantitative, analytical model of death in practices exposes some of the difficulties of methodology. The medical profession often regards death as a failure of treatment rather than an inevitable consequence of living. It is in this context that we have approached this project with circumspection. Nonetheless it is hoped that appropriate examination of these data may assist in the pursuit of the clinical governance agenda in general practice. A way forward may be to develop models that measure the quality of peri-mortal experience. This might be based on information from carers, professionals and relatives with some measurable factors of social, physical and psychological wellbeing. Death, where possible, should be pain-free, in a setting agreed to be the best possible and should be a dignified experience. If sudden and unexpected, the medical response should be timely, with appropriate support for the bereaved. A model we might copy could be the "APGAR" score used at birth to measure the health and viability of the new-born at the start of life. An essentially qualitative analysis would assist practices in measuring performance in several areas and would provide useful information in the development of improved quality of care in partnership with relatives and carers.
REFERENCES
1 Holden J, O’Donnel S, Brindley J, Miles L. Analysis of 1263 deaths in four general practices. Br J Gen Pract 1998;48:1409-1412.
2 Khunti K. A method of creating a death register for general practice. Br Med J 1996;312:952.
3 Webb R, Esmail A. An analysis of practice-level mortality data to inform a health needs assessment. Br J Gen Pract 2002;52:296-299.
4 Bhopal R. Death registers in general practice would be a means of preventing malpractice and murder. Br Med J 2000;320:1272.
5 Knox EG. An epidemic pattern of murder. J Publ Health Med 2001;24(1):34- 37.
6 Mohammed MA, Cheng KK, Rouse A, Marshall T. Bristol, Shipman, and clinical governance: Shewhart’s forgotten lessons. Lancet 2001:357:463- 467.
7 Coldwells A, Fraser F, Tavendale P, Crooks G, Peterkin G. Using death certicates to identify malpractice might be difficult. Br Med J 2001;322:303.
8 Frankel S, Sterne J, Davey Smith G. Mortality variations as a measure of general practitioner performance: implications of the Shipman case. Br Med J 2000;320:489.